RESEARCH ARTICLE SUMMARY

ADVANCED IMAGING

Extended-resolution structuredillumination imaging of endocyticand cytoskeletal dynamicsDong Li Lin Shao Bi-Chang Chen Xi Zhang Mingshu Zhang Brian MosesDaniel E Milkie Jordan R Beach John A Hammer III Mithun PashamTomas Kirchhausen Michelle A Baird Michael W Davidson Pingyong Xu Eric Betzig

INTRODUCTION Various methods of super-resolution (SR) fluorescence microscopy havethe potential to follow the dynamic nanoscaleinteractions of specific macromolecular assem-blies in living cells However this potential isoften left unfulfilled either owing to themethodrsquos inability to follow these processesat the speeds dictated by nature or becausethey require intense light that can substan-tially perturb the very physiology one hopes tostudy An exception is structured illuminationmicroscopy (SIM) which can image live cellsfar faster and with orders of magnitude lesslight than required for other SR approachesHowever SIMrsquos resolution is usually limitedto only a twofold gain beyond conventionaloptical microscopes or ~100 nm with visiblelight

RATIONALE We endeavored to find ways toextend SIM to the sub-100-nm regime while re-taining to the greatest extent possible theadvantages that make it the preferred SRmeth-od for live-cell imaging Our first solution usedan ultrahigh numerical aperture (NA) lens andtotal internal reflection fluorescence (TIRF) toachieve 84-nm resolution at subsecond acqui-sition speeds over hundreds of time points inmultiple colorsnear thebasalplasmamembraneOur second exploited the spatially patternedactivation of a recently developed reversiblyphotoswitchable fluorescent protein to reach45- to 62-nm resolution also at subsecond ac-quisition over sim10 to 40 time points

RESULTS We used high-NA TIRF-SIM toimage the dynamic associations of cortical fila-

mentous actin with myosin IIA paxillin orclathrin as well as paxillin with vinculin andclathrin with transferrin receptors Thanks tothe combination of high spatial and temporalresolution we were able to measure the sizesof individual clathrin-coated pits through theirinitiation growth and internalizationWewere

also able to relate pit sizeto lifetime identify andcharacterize localized hotspots of pit generationand describe the interac-tion of actin with clathrinand its role in accelerat-

ing endocytosis With nonlinear SIM by use ofpatterned activation (PA NL-SIM) we moni-tored the remodeling of the actin cytoskeletonand the dynamics of caveolae at the cell sur-face By combining TIRF-SIM and PA NL-SIMfor two-color imaging we followed the dynamicassociation of actin with a-actinin in expand-ing filopodia and membrane ruffles and char-acterized shape changes in and the transportof early endosomes Last by combining PANL-SIMwith lattice light sheetmicroscopy weobserved in three dimensions and across theentire volume of whole cells the dynamics ofthe actin cytoskeleton the fusion and fissionof mitochondria and the trafficking of vesi-cles to and from the Golgi apparatus each ataxial resolution fivefold better than that ofconventional widefield microscopyIn addition through direct experimental

comparisons we demonstrated that the reso-lution for our methods is comparable with orbetter than other SR approaches yet allowed

us to image at far higher speeds andfor far longer durations To under-stand why this is so we developed adetailed theoretical model showingthat our methods transmit the infor-mation encoded in spatial frequenciesbeyond the diffraction limitwithmuchgreater strength than do other alter-natives and hence require far fewerphotons emitted from the specimenusing far less intense light

CONCLUSION High-NA TIRF-SIMand PA NL-SIM fill an unmet needfor minimally invasive tools to im-age live cells in the gap between the100-nm resolution traditionally as-sociated with SIM and the sub-60-nmregime of protein-specific structuralimaging served by single-moleculelocalization microscopy

RESEARCH

944 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

The list of author affiliations is available in the fullarticle onlineCorresponding author E-mail betzigejaneliahhmiorgCite this article as D Li et al Science 349 aab3500(2015) DOI 101126scienceaab3500

Two approaches for improved live-cell imaging at sub-100-nm resolution (Left) Association of corticalactin (purple) with clathrin-coated pits (green) the latter seen as rings (inset) at 84-nm resolution via acombination of total internal reflection fluorescence and structured illumination microscopy at ultrahighnumerical aperture (high-NA TIRF-SIM) (Right) Progression of resolution improvement across the actincytoskeleton of a COS-7 cell from conventional diffraction-limited TIRF (220-nm resolution) to TIRF-SIM(97-nm resolution) and nonlinear SIM based on the patterned activation of a reversibly photoswitchablefluorescent protein (PA NL-SIM 62 nm resolution) (Left and right represent single frames from time-lapsemovies over 91 and 30 frames respectively Scale bars 2 mm (left) 3 mm (right)

ON OUR WEB SITE

Read the full articleat httpdxdoiorg101126scienceaab3500

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om

RESEARCH ARTICLE

ADVANCED IMAGING

Extended-resolution structuredillumination imaging of endocytic andcytoskeletal dynamicsDong Li1 Lin Shao1 Bi-Chang Chen1 Xi Zhang23 Mingshu Zhang2 Brian Moses4

Daniel E Milkie4 Jordan R Beach5 John A Hammer III5 Mithun Pasham6

Tomas Kirchhausen6 Michelle A Baird57 Michael W Davidson7

Pingyong Xu2 Eric Betzig1dagger

Super-resolution fluorescence microscopy is distinct among nanoscale imaging tools inits ability to image protein dynamics in living cells Structured illumination microscopy(SIM) stands out in this regard because of its high speed and low illumination intensitiesbut typically offers only a twofold resolution gain We extended the resolution of live-cellSIM through two approaches ultrahigh numerical aperture SIM at 84-nanometer lateralresolution for more than 100 multicolor frames and nonlinear SIM with patternedactivation at 45- to 62-nanometer resolution for approximately 20 to 40 frames Weapplied these approaches to image dynamics near the plasma membrane of spatiallyresolved assemblies of clathrin and caveolin Rab5a in early endosomes and a-actininoften in relationship to cortical actin In addition we examined mitochondria actin andthe Golgi apparatus dynamics in three dimensions

Fluorescence microscopy continues to play akey role in elucidating structure and func-tion of living systems thanks to its ability toimage specific proteins with single-moleculesensitivity as well as its capacity to study

in vivo dynamics in a minimally invasive man-ner Its power has grownwith the introduction ofsuper-resolution (SR) techniques (1) that extendits diffraction-limited spatial resolution [~200 nmfor green fluorescent protein (GFP)] by as muchas an order of magnitude However although theSR imaging of fixed specimens the most com-mon modality offers the highest resolution itdoes so at the considerable risk of altering thevery ultrastructure it hopes to reveal because ofboth the fixation process itself (fig S1) (2) andthe extremely high density of fluorescent markersrequired to achieve such resolution (3) Further-more with the advent of genetically encodedmarkers for electron microscopy (EM) (4 5) the

continued preeminence of SR microscopy forprotein-specific structural imaging at the nano-scale is no longer assuredA different situation emerges for in vivo im-

aging in which EM is too destructive and fixa-tion is not involved However although thiswould appear to be the ideal niche of SR micros-copy SR techniques such as localization micros-copy (6 7) stimulated emission depletion (STED)microscopy (8 9) and reversible saturable op-tical fluorescence transitions (RESOLFT) micros-copy (10 11) place extraordinary demands on thephoton budget represented by the product of thenumber of fluorescentmolecules in the specimenand the number of photons each can emit beforebleaching irreversibly (fig S2) (12) They also re-quire specialized photoswitchable labels and ex-citation intensities of 103 to 108 Wcm2 whichare orders ofmagnitude greater than the 01Wcm2

under which life evolved (fig S3) As a resulttime-lapse measurements with these techniquesrarely consist of more than a few frames andphototoxic changes to cellular physiology can setin quickly even at the lower end of this range(movie S1) In addition typical SR acquisitionspeeds of ~1 s to several minutes per frame aretoo slow to follow processes that move fasterthan ~1 to 50 nms without introducing motion-induced artifacts (fig S4) whereas common res-olution metrics such as the Nyquist criterion forlabeling density (6 7) or the width of an isolatedfeature (8ndash11) tend to substantially overestimatethe true spatial resolution (figs S5 and S6)A notable exception is structured illumination

microscopy (SIM) which in vivo (13ndash15) canimage in multiple colors using conventional flu-

orescent labels as fast as 11 framess (13) at in-tensities of only 1 to 100 Wcm2 Its primarylimitation is that its resolution in vivo has beenlimited to ~100 nm for GFP or only twice beyondthe diffraction limit This has provided the mo-tivation for the development of other in vivondashcompatible SR methods but to date all sufferfrom substantial limitations as noted above

Extending SIM resolution viahigh-numerical-aperture optics

We extended the resolution of live-cell SIM bytwo independent means In the first we usedthe higher numerical aperture (NA) afforded bya commercially available 17-NA objective to im-age at 84-nm resolution (for GFP) (fig S7) Al-though the total internal reflection fluorescence(TIRF) condition at this NA confines observa-tions towithin ~50 to 200 nmof the basal plasmamembrane (fig S8) it also restricts the excitationto only a small fraction of the cellular volumefurther reducing phototoxicity eliminating out-of-focus background and leaving unaffected apotential reservoir of cytosolic target moleculesthat might be recruited to the plasmamembraneat later time points With this approach termedhigh-NA TIRF-SIM we could image dynamic as-sociations between proteins in a variety of sys-tems at sub-100-nm resolution often for 80 to100 time points including filamentous actin(mApple-F-tractin) and enhanced GFP (EGFP)ndashmyosin IIA (Fig 1 A and B Movie 1 and fig S9)mApple-f-Tractin and mEmerald-paxillin (Fig 1C and D and movie S2) mEmerald-paxillin andmTagRFP-vinculin (RFP red fluorescent protein)(Fig 1 E and FMovie 2 and fig S10) mEmerald-clathrin light chain b (CLTB) and Alexa 568ndashtagged transferrin (fig S11 and movie S3) andmEmerald-CLTB and mCherry-Lifeact (Fig 2E to G) (16) Acquisition times were typicallyless than 1 s per color per frame although thetime interval between frames was adjusted ineach case in order to match the dynamics of in-terest and to give the cell time to recover fromthe effects of the excitation applied during theacquisitionIn the case ofmEmerald-paxillin andmTagRFP-

vinculin we found that both proteins expandedinward for those focal adhesions near the periph-ery of an HFF-1 cell (Movie 2 and fig S12)mEmerald-paxillin dominated in the peripheral-facing end of these adhesions whereasmTagRFP-vinculin increased in concentration toward theinterior (Fig 1 E and F and fig S10C) In contrastfor adhesions located far away from the periph-ery mTagRFP-vinculin dominated (fig S10A)and there was little change in the distribution ofeither protein over time The resolution of ourdata was comparable with that demonstrated bymeans of live-cell localizationmicroscopy (fig S13)(6) However we could image in two colors atintensities of only 30 to 100 Wcm2 in a total ac-quisition time of 167 s versus a single coloracquired at 1 kWcm2 in 25 s in the localizationcase This represents a 90 to 97 decrease inexcitation intensity and a 15-fold increase inimaging speed

RESEARCH

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-1

1Janelia Research Campus Howard Hughes Medical InstituteAshburn VA 20147 USA 2Key Laboratory of RNA Biologyand Beijing Key Laboratory of Noncoding RNA Institute ofBiophysics Chinese Academy of Sciences Beijing 100101China 3College of Life Sciences Central China NormalUniversity Wuhan 430079 Hubei China 4ColemanTechnologies 5131 West Chester Pike Newtown Square PA19073 USA 5Cell Biology and Physiology Center NationalHeart Lung and Blood Institute National Institutes ofHealth Bethesda MD 20892 USA 6Department of CellBiology and Pediatrics Harvard Medical School and Programin Cellular and Molecular Medicine Boston ChildrenrsquosHospital Boston MA 02115 USA 7National High MagneticField Laboratory and Department of Biological ScienceFlorida State University Tallahassee FL 32310 USAPresent address Research Center for Applied Sciences AcademiaSinica Taipei 11529 Taiwan daggerCorresponding author E-mailbetzigejaneliahhmiorg

For clathrin (Fig 2) we saw that matureclathrin-coated pits (CCPs)were resolved as rings(fig S14 and movie S4) presumably because thedistal end of each resides outside the evanescentexcitation field When imaged at 37degC CCPs largeand stable enough to be resolved as a ring at one ormore time points grew to a ~152-nmmedian max-imum diameter (Fig 2B) (17) in BSC-1 cells ex-pressing EGFP-CLTA (Fig 2A and fig S14) withsimilar results inCOS-7 cells expressingmEmerald-CLTB (Fig 2F and fig S15B) and persisted foron the order of 1 min (Movie 3 and fig S15A)Linear regression revealed a positive correlationof 020 nms between maximum ring diameterand lifetime (Fig 2C) which is consistent withpreviously observed correlations between lifetimeand clathrin intensity or cargo size (18)In BSC-1 cells most CCPs were isolated at any

given time but the sites of their initiation did notfollow a Poisson distribution (fig S16) Insteadwhen the 1297 initiation events over the course ofMovie 3 were binned into 032- by 032-mm sitescorresponding to a mean density of 029 eventsper site 365 of all events occurred at sites ofone or more additional events compared with135 expected if they were Poisson distributedat this mean density Indeed at 36 of the sitesfive or more CCPs were generated sequentially(for example one marked by green arrows atdifferent time points in Fig 2D) which wouldotherwise be a very rare event (00053) assum-ing Poisson statistics Although such ldquohot spotsrdquohave been observed previously (19 20) in our case

we found that these consisted of single persistentsubdiffractive patches of clathrin fromwhichmul-tiple CCPs emerged (Movie 4) like bubblesIn COS-7 cells mEmerald-CLTB appeared as

both isolated rings and larger structures (Fig 2FMovie 5 figs S17 and S18 and movie S5) thelatter consisting of aggregates of rings (fig S19)They may be related to clathrin plaques made ofextended clathrin lattices of low curvature (20)also referred to elsewhere as flat clathrin lattices(21) Although they persisted far longer than iso-lated CCPs individual rings would occasionallydetach from these aggregates (fig S19) In no in-stance did we observe large homogenous patchesof clathrin as wemight expect for the flat clathrinlattices common in EM images (22)The role of actin in clathrin-mediated endocy-

tosis inmammalian cells remains an area of somedebate (17 23ndash26) Our two-color imaging ofCOS-7 cells by means of high-NA TIRF-SIM re-vealed that all aggregates of mEmerald-CLTBrings were associated with mCherry-Lifeact overat least part of their areas at all times (fig S19)In contrast approximately equal populations ofindividual CCPs completed endocytosis eitherwith (Fig 2E and fig S20 A and B) or without(fig S20 C and D) recruitment of Lifeact in thefinal five frames (20 s) before internalization ofthe pit In both cases histograms of CCP lifetimeswere well described by single exponential fitsindicating constant probabilities of internaliza-tion per unit time (fig S15A) The 1e lifetimes of564 plusmn 30 s for the Lifeact-associated CCPs and

672 plusmn 19 s for CCPs without Lifeact indicate thatactin when present indeed increases the inter-nalization probability Consistent with this themedianmaximum clathrin ring diameter for ringspersisting over at least five frames was slightlysmallerwith thanwithout associated Lifeact (160versus 168 nm) (fig S15B)Lifeact associatedwith CCPs usually approached

in a wave or filament (Fig 2E and fig S20A) Wealso observed rings of Lifeact (COS-7 cells) (Fig2G and fig S21 C and D) or F-tractin (U2OS cell)(fig S21 A and B) similar in size to the clathrinones and having lifetimes of several minutes (figS17B) However Lifeact rings were not as numer-ous as clathrin ones and were coincident withthem in only a few instances (fig S22) Althoughthey might be associated with other forms ofclathrin-independent endocytosis their role re-mains unclear

Live-cell nonlinear SIM via patternedactivation of photoswitchable fluorophores

To achieve even higher resolution than that ofhigh-NA TIRF-SIM we turned to nonlinear SIM(NL-SIM) The nonlinearity inherent in eitherthe patterned saturation of fluorescence excita-tion at high intensity (27 28) or the patterneddepletion of photoswitchable fluorophores (figS23) (29) introduces additional harmonicsH whichpermit resolution extension at wavelength l viaSIM to el=frac122NAethH thorn 1THORN with H ge 2 comparedwith H = 1 for the traditional linear form ofSIM andH = 0 for diffraction-limited widefield

aab3500-2 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 1 High-speed live-cell imaging at 37degC of associations between pro-teins at sub-100-nm resolution (A) Cytoskeletal proteins mApple-F-tractin(purple) and EGFP-myosin IIA (green) in a mouse embryonic fibroblast cell(Movie 1 and fig S9) (B) Magnified view of the boxed region in (A) show-ing bipolar myosin IIA filaments with clearly resolved opposed head groups(for example green arrowhead) (C) mApple-F-tractin (purple) and the fo-cal adhesion protein mEmerald-paxillin (green) in a U2OS cell (movie S2)

(D) Magnified view of the boxed region in (C) showing association of paxillinwith smaller actin fibers fanning out from the ends of larger stress fiber cables(E) Focal adhesion proteins mTagRFP-vinculin (purple) and mEmerald-paxillin (green) in a HFF-1 cell (Movie 2 and figs S10 and S13) (F) Magnifiedview of the boxed region in (E) showing a gradient of increased paxillinconcentration toward the cell periphery Scale bars 5 mm(A) (C) and (E) 1 mm(B) (D) and (F)

RESEARCH | RESEARCH ARTICLE

imaging with uniform illumination (fig S24 Aand B) Resolution of ~50 nm has been demon-strated with both approaches although not onliving cells saturated excitation was used to im-age densely labeled fluorescent beads at the like-ly phototoxic peak intensity of 8 MWcm2 (28)whereas saturated depletion (SD) imaged singlefixed cells at 945 sframe (29)mdashfar too slow tofollow most cellular processesSD provides the basis of resolution enhance-

ment in STED and RESOLFT as well as SD NL-SIM The degree of enhancement depends on thedegree of saturation (figs S23 and S25 and movieS6 part 3) defined in multiples of the saturationfactor (SF) for which 1e of the irradiated mol-

ecules remain in the original activated or excitedstate However high SFs are very photon ineffi-cient only a fraction of the photobleaching-dictated number of switching cycles for anygiven molecule then contributes useful signal(figs S25 to S27) Furthermore high SFs requirehigh intensities andor long exposures (fig S28)neither of which is compatible with fast non-invasive live-cell imagingWe addressed these issues by using patterned

activation (PA) followed with patterned excita-tion and readout of the green photoswitchableFP Skylan-NS (fig S29 andmovie S6 part 1) (30)rather than SD to generate H = 2 harmonicsyielding 62-nm resolution and subsecond acqui-

sition times in TIRF for live cells (Fig 3) Thisapproach termed PA NL-SIM allowed us toachieve large amplitudes in both the first andsecond harmonics of the emission pattern (figS29 G and H) leading to SR images of highsignal-to-noise ratio (SNR) even at low activa-tion and excitation saturation factors SFact andSFexc obtained with low intensities and shortexposures (table S1) Furthermore by keepingSFact low only a small fraction of the totalmolecular population needed to be activatedfor every raw image and with H = 2 only N =(2H + 1)2 = 25 such raw images needed to beacquired to reconstruct each SIM image frame(fig S30) Consequently we could acquire subs-tantially more frames at substantially higherSNR (fig S26) in far less time (table S1) bymeans of PANL-SIM (Movie 6) than SDNL-SIM(movie S7)PA NL-SIM of Skylan-NS-Lifeact (Fig 3 A and

BMovie 6 andmovies S8 to S10) in living COS-7cells revealed considerably more detail than didTIRF-SIM (Fig 3B) in dense peripheral actin arcsand star-like junctions of single actin filamentsWe were also able to resolve individual Lifeactrings once again including rings too small to seeclearly with high-NATIRF-SIM (fig S21 C andDandmovie S8) Furthermore we could follow thedynamics of the Lifeact-decorated actin cytoskel-eton for 30 image frames acquired in 12 s each(Movie 6) This is 1250times faster and used 20timeslower intensity (100 Wcm2) than was requiredfor an image of phalloidin-labeled actin at theventral surface of a fixed BSC-1 cell obtained bymeans of dual-objective localization microscopy(31) yet the level of detail seen by the two meth-ods was comparable (fig S31) even though our62-nm resolutionwas threefold coarser than thatreported in the localization imageWe also used PANL-SIM to image keratin (fig

S6 and movie S11) and caveolin (Fig 3 C to FMovie 7 figs S32 and S33 and movie S12) inliving COS-7 cells each with Skylan-NS at aresolution of 59 nm In the latter case this wassufficient to resolve numerous caveolae movingby less than their radii during the acquisitiontime as rings which is consistent with their in-vaginated appearance bymeans of EM (32) Suchrings were not observed at caveolae in a HeLacell imagedwith RESOLFT (fig S34) (33) despitea similar reported resolution Rings of Skylan-NS-caveolin were somewhat more abundant thanCCPs (figs S17 and S18) and althoughmost werebelow100nminsize their distributionwasbroader(Fig 3D) than the 60 to 80 nm range observedwith EM However some of the larger rings (Fig3E) may represent multiple caveolae clusteredaround surface-docked vesicles (34) Caveolaealso tended to loosely cluster in long narrow rib-bons although we saw tighter aggregations ofrings (Fig 3F) as well similar to those we saw inclathrin plaques (fig S19)Our time-lapse imaging showed that most

caveolae moved only a fraction of their size dur-ing the acquisition time although more met thiscondition when slowed (35) by operating at 23degC(Movie 7 and fig S32) than when imaged at 37degC

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-3

Fig 2 Dynamics of clathrin-mediated endocytosis and the cortical actin cytoskeleton (A) CCPsresolved as rings (fig S14 and movie S4) and color-coded according to their age since initial formation atone time point from amovie of CCPdynamics in a BSC-1 cell at 37degC stably expressing EGFP-clathrin lightchain a (Movie 3) (B) Histogram of maximum diameter of each CCP over its lifetime (C) Plot of CCPoverall lifetime versus CCP maximum diameter (D) Sequential production of multiple CCPs at a CCP-generating ldquohot spotrdquo identified with green arrowheads (Movie 4 and fig S16) (E) Formation growth anddissolution of a single CCP (right) and its relationship to cortical f-actin (left) in a COS-7 cell at 37degCtransfected with mEmerald-clathrin light chain b and mCherry-Lifeact Light blue arrowheads mark timepoints at which f-actin associates with the CCP (F) Individual CCPs and clathrin plaques (green) andcortical f-actin (red) at one time point during their evolution in a COS-7 cell (Movie 5 figs S17 and S18andmovie S5) (G) Formation of a nanoscale ring of f-actin (fig S17B) Scale bars 1 mm (A) and (F) and200 nm (D) (E) and (G)

RESEARCH | RESEARCH ARTICLE

(fig S33 and movie S12) The smaller laterallymobile fraction in each case appeared as distorteddiscontinuous rings or quasiperiodic patches(fig S35) These morphologies are indicative ofmotion-induced artifacts and underscore thedifficulty of live-cell SR imaging by any methodHigher resolution must be accompanied by pro-portionally faster acquisition times to followdynamic events of a given velocity yet higherresolution also requires a quadratically increasingnumber of raw measurements for each two-dimensional (2D) image frame Even the com-paratively brief 035 swe needed to acquireN= 25raw images for each PA NL-SIM image wasinsufficient to accurately depict caveolae mov-ing by much more than our 59-nm resolution inthis timeNevertheless by further increasing SFact we

were able to saturate the fraction of molecules inthe activated state near the maxima of the pat-terned activation light (movie S6 part 2) Saturated

PA NL-SIM generates an additional harmonic(H = 3) strong enough (fig S29) to further extendthe resolution to 45 nm (figs S36 and S37) andallowed us to identify even smaller Skylan-NS-caveolin rings unresolvable without the extraharmonic (Fig 3C) UsingN = 35 rather thanN =49 raw images per frame we balanced the re-sulting anisotropic resolution (fig S30) againstthe needs for rapid acquisition (049 sframe) andparsimonious use of the photon budget to im-age caveolin rings over 12 frames at 3 s intervals(movie S13)

Two-color live imaging via combinedTIRF-SIM and PA NL-SIM

By combining linear SIM and PA NL-SIM bothin TIRF we could study associations betweenfluorescent proteins one conventional and onephotoswtichable in two colors at higher resolu-tion than by means of linear TIRF-SIM aloneImages (Fig 4 A to C and figs S38 and S39) and

movies (Movie 8 and movie S14) of mCherry-Rab5amdasha regulator of the formation fusion andtransport of early endosomes (EEs) (36)mdashrevealedirregularly shaped dynamically remodelingpatchesof Rab5a (fig S39 A and B) consistent with thetubularvesicular architecture of EEs seen in EM(Fig 4B) (37) Numerous patches also featureddark spots (fig S39C) perhaps indicative of car-go or internal vacuoles depleted of Rab5a Mostpatchesmoved randomly between successive 20-stime intervals at velocities slow enough to avoidmotion artifacts during each 034-s acquisitionWe also observed a subpopulation of slowly grow-ing Skylan-NS-Lifeactndashassociated Rab5a patchesthat were constrained for minutes at a time (figS39D arrows) At the other extreme we occa-sionally observed streaks of Rab5amoving paral-lel to nearby actin filaments at velocities of 3 to5 mms (Fig 4C and fig S39E) These may repre-sent EEs actively transported alongmicrotubules(38) parallel to the filaments

aab3500-4 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 3 Live-cell non-linear structured illumi-nation microscopybased on patternedphotoactivation (A)Single time point from amovie of the evolution ofcortical f-actin in a COS-7cell at 23degC transfectedwith Skylan-NS-Lifeactseen at 62-nm resolution(Movie 6 fig S31 andmovie S8) (B) Magnifiedview from a different cellat 37degC comparingdiffraction-limited TIRFmicroscopy (top left)TIRF with deconvolution(top right) TIRF-SIM(bottom left) and non-linear TIRF-SIM withpatterned activation(PA NL-SIM bottomright) (movies S9 andS10) (C) Caveolae in aCOS-7 cell at 23degC trans-fected with Skylan-NS-caveolin comparing TIRFwith deconvolution (topleft 220-nm resolution)TIRF SIM (top right97-nm resolution) PANL-SIM (bottom left62-nm resolution) andsaturated PA NL-SIM(bottom right 45-nmresolution) (Insets) Asingle caveolae pit even-tually resolved as a ringby saturated PA NL-SIM(Movie 7 figs S34 toS37 and movie S13(D) Diversity of caveolae ring diameters as seen by means of PA NL-SIM (E) Larger rings that may represent surface-docked vesicles (F) Clusters of caveolaereminiscent of clathrin plaques (D) to (F) are from a different cell at 37degC (fig S33 and movie S12) Scale bars 3 mm (A) 1 mm (B) 200 nm (C) and 100 nm(D) (E) (F) and (C) inset

RESEARCH | RESEARCH ARTICLE

We also used PA NL-SIM and TIRF-SIM re-spectively to study the association of Skylan-NS-Lifeact with mCherry-a-actinin (Fig 4 D to Fand fig S40) Consistent with its role as an actin-bundling protein (39) in COS-7 cells we founda-actinin at the treadmilling edge of the lamelle-podium and at the basal surface in both filopodiaand the leading edges of growing membraneruffles (Fig 4F Movie 9 and movie S15) We alsoobserved concentrations of a-actinin along thesides (Fig 4E) and at the branching ends of stressfibers that likely attach to cell-substrate adhesions(40) Last a-actininwas present at dense junctionsof Lifeact-decorated filaments and Skylan-NS-Lifeact rings as described above were colocalizedin every instance with a mCherryndasha-actinin ringof similar size (fig S41) Septins another class ofactin-bundling proteins have been shown (41) toproduce f-actin rings in vitro (albeit of larger sizethan here) so perhaps a-actinin not only aids inbundling actin filaments in nanometric rings butalso contributes to their extreme curvature

3D live-cell imaging with combined PANL-SIM and lattice light sheet microscopy

Although the ~50- to 200-nm extent of the eva-nescent excitation field we used in the examplesabove eliminated out-of-focus background andconfined potentially phototoxic exposure to aminute fraction of the cellular volume it alsolimited our observations to this subvolume andseverely restricted the total photon budget avail-able for those targets unable to be replenishedfrom the cytosol during the imaging intervalTo extend our observations to the entire cell

we turned to live-cell 3D-SIM (14 15) Unfortu-nately traditional 3D-SIM with linear widefieldexcitation brings limitations of its own It is slow

(~20 s acquisition for whole adherent HeLa cells)limited to thin specimens (because of out-of-focusbackground) and requires high SNR for accurateimage reconstruction It is also potentially photo-toxic and bleaches specimens rapidly because ofcontinuous whole-cell illumination These prob-lems would all be greatly magnified in its directextension to PA NL-SIMThus to apply PA NL-SIM to living cells in

three dimensions (Fig 5) we used lattice lightsheet microscopy (42) In this technique an exci-tation objective (fig S42A) projects a thin sheetof light (fig S42A blue) through a specimen (figS42A orange) and the fluorescence generated inthe illuminated plane is collected by a detectionobjective and imaged onto a camera Repeatingthis process plane-by-plane through the specimenproduces a 3D image Restriction of the light tothe detection focal plane eliminates out-of-focusbackground increases the z axis resolution andgreatly reduces photobleaching and phototoxicityIn cross-section the light sheet has the 2D

periodic structure of an optical lattice (fig S42B)Sweeping the sheet back and forth along the xaxis produces time-averaged uniform illumina-tion offering high speed and diffraction-limitedxyz resolution of 230 by 230 by 370 nm as seenin a volume-rendered image of the actin cyto-skeleton (fig S43A) and its corresponding overalloptical transfer function (OTF) (fig S43D) Step-ping the sheet in x in five equal fractions of thelattice period and applying the algorithms of3D-SIM to the resulting five raw images perplane extends the xyz resolution to 150 by 230by 280 nm (fig S43 B and E) but at the cost of atleast 5times longer acquisition times (42)To further extend the 3D resolution via PA

NL-SIMwe first photoactivated targetmolecules

fused to Skylan-NS using a hexagonal lattice lightsheet of l = 405 nm wavelength having H = 2harmonics (fig S43E) We then imaged the fluo-rescence from the activated region exciting thefluorescencewith a lattice light sheet of l =488nmwavelength having the same hexagonal sym-metry and period (fig S42B bottom) as the ac-tivation lattice For activationwell below saturationthe product of the activation and excitation pat-terns creates a fluorescence emission patternwithin the specimen having H = 4 harmonics(fig S43F) Thus we stepped the sheet in x in2H + 1 = 9 equal fractions of the lattice periodwhile recording nine images Repeating this pro-cess for every plane within the specimen we thenreconstructed a 3D PA NL-SIM volume-renderedimage (fig S43C) with resolution extended to118 by 230 by 170 nmWe used this approach to image mitochondria

in COS-7 cells (Fig 5A) as well as the actin cyto-skeleton (fig S43 A to C and movie S16) andthe Golgi apparatus (Fig 5B) inU2OS cells all at23degC so as to simplify the overlap of the activa-tion and excitation patterns Time-lapse 3D im-ages (Fig 5A bottom) and movies (Movie 10) ofSkylan-NSndashtagged translocase of outer mitochon-drial membrane 20 (TOM20) revealed the mi-gration constriction before fission and fusionof individual mitochondria (43 44) each clearlyresolved as a hollow tubular structure The 3Dvolume rendering and the widths of mitochon-drial membranes in individual xy orthosliceswere both comparable with similar data from afixed cell imaged with 3D localization micros-copy (45) at a reported xyz resolution of ~20 by20 by 60 nm (fig S44)A volume-rendered movie (movie S17) of the

Golgi-resident enzyme Mannosidase II (MannII)

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-5

Fig 4 Combined TIRF-SIMand PA NL-SIM of protein-pair dynamics in livingcells (A) Skylan-NS-Lifeact(orange PA NL-SIM) andmCherry-Rab5a a marker ofearly endosomes (greenTIRF-SIM) in a COS-7 cellat 23degC (Movie 8 figs S38and S39 and movie S14)(B) Comparison of EMimages of early endosomes(37) with similarly shapedRab5a patches seen in (A)(C) Magnified view at threesuccessive time pointsshowing rapid transportof a Rab5a streak parallel tothe cytoskeleton (D) Skylan-NS-Lifeact (green PA NL-SIM) and mCherry-a-actinin(purpleTIRF-SIM) in a COS-7cell at 23degC (Movie 9 figsS40 and S41 and movieS15) (E) Magnified viewfrom (D) with Lifeact (top)a-actinin (middle) and overlay (bottom) showing paired association at focal adhesions and along the sides of large stress fibers (F) Evolution of amembrane ruffle showing a-actinin concentrated at the leading edge Scale bars 5 mm (A) (D) 200 nm (B) 1 mm (C) and (E) and 500 nm (F)

RESEARCH | RESEARCH ARTICLE

tagged with Skylan-NS in a U2OS cell as seenlooking into the cis-face from the nucleus showedMann II concentrated in a hollow sphere ofcisternae having a cis-facing void Time-lapse3D data (Fig 5B andmovie S18) color-coded forheight showed the docking of small vesicles (Fig5B white arrows) that may represent pre-Golgiintermediates (46) as well as the rapid export ofMann II in long tubular post-Golgi carriers (Fig5B red arrows) (47)The volumetric resolution of 3D lattice light

sheet PA NL-SIM at the 06-NA excitation and11-NA detection we used here is comparablewith the 105- by 105- by 369-nm xyz resolutionof widefield 3D-SIM at 12 NA However thelattice approach has twofold higher axial resolu-tion and fourfold better than traditional diffraction-limited microscopy It is therefore better suitedto problems in which its superior optical section-ing is essential such as in resolving heterogene-ities in nuclear architecture distinguishing eventsoccurring at the dorsal or ventral plasma mem-brane or as above tracking vesicles through thesecretory pathway Whole-cell acquisition times(705 and 327 s in Fig 5 A and B respectively)are slow compared with PA NL-SIM in TIRF butsimilar to widefield 3D-SIM However thanks tothe oblique imaging geometry (fig S42) restrictedxy fields of view can be imaged at proportion-ally faster speed through the entire thickness ofthe cell

Discussion

The above results provide but a brief glimpse ofthe biology that might be uncovered with thelive-cellndashcompatible SRmethods of high-NATIRF-SIM and PA NL-SIM We have measured andcorrelated the diameters and lifetimes of CCPsobserved at high resolution different forms ofCCP initiation and shown that CCP internaliza-

tion is aided by actin filaments in about half of allcases We have seen that caveolin localizes notonly to the 60- to 80-nm invaginated caveloaecommon in EM images but also to much largerring-like structures and have followed dynamicchanges in the shapes of early endosomes Lastwe have observed the nanoscale remodeling ofthe actin cytoskeleton in relation to clathrin andRab5a as well as cytoskeletal-related proteinssuch as myosin IIA a-actinin and paxillinHowever the above results also amply illus-

trate the trade-offs inherent in live SR imagingWith high-NA TIRF-SIM at 17 NA we could ac-quire up to 200 image frames in lt05 s each atintensities of 20 to 100 Wcm2 and a resolutionof 84 nm (for GFP) whereas extending the reso-lution to 62 nm with PA NL-SIM restricted us tono more than 40 frames and further extensionto 45 nm with saturated PA NL-SIM required490 Wcm2 and produced only 12 frames atuseful SNRIn short evenmodest gains in resolution come

at substantial cost in terms of the other metricsimportant for live-cell imaging These tradeoffsare not specific to SIM In fact our extensions ofSIM are far more compatible with live imagingthan any other form of SR fluorescence micros-copy of comparable resolution demonstrated todate In part this is because the OTF which de-fines the degree towhich different sample spatialfrequencies (representing differently sized struc-tures) are passed to the image is far stronger inthe 100-nm regime (fig S24B) for high-NA TIRF-SIM at 17 NA than other linear methods such asconfocal or image scanning microscopy (ISM)(48ndash50) and far stronger in the 50- to 100-nmregime (fig S24C) for PA NL-SIM than othernonlinear methods such as STED (8 9) point-scanning (PS) RESOLFT (10) or array-basedwide-field (WF) RESOLFT (11) As a result far fewer

photons need to be collected (fig S2) and far lesslight (fig S3) needs to be applied to the specimento see features in these regimes at acceptableSNR Localization microscopy is also photon in-efficient in that the density of localizedmoleculesis nearly always more limiting to the resolutionthan is the number of photons emittedper switch-ing cycle which dictates the localization preci-sion For example simulations (12) based on thetheoretical OTFs suggest that to resolve an 85-nmgrating PANL-SIM requires ~80times fewer photonsfrom the specimen per unit area than localiza-tionmicroscopy ~200times fewer thanWF-RESOLFTand ~15times fewer than PS-RESOLFT or STED eachat a depletion saturation factor of SFdepletion = 10(fig S2)Another reason for the greater compatibility of

high-NA TIRF-SIM and PA NL-SIM with livingcells is that they require much lower peak inten-sities of applied light High resolutionwith STEDor RESOLFT demands high factors of saturateddepletion (fig S25 A and C) that are wasteful ofthe photon budget (fig S25 B andD) and requireenormous intensities andor long exposures foractivation (fig S45) depletion (fig S28) and read-out of the final signal (fig S3) Localization mi-croscopy also requires high intensities to achievehigh photon emission and photoswitching ratesfrom single molecules For example extrapolat-ing from reported experimental values for live-cell imaging (table S1) the 08- to 35-Wcm2

activation intensity used over the 45- by 45-mmfield of view in Fig 3A in 12 s bymeans of PANL-SIM is 960000 times weaker than that whichwould be required to image the same area in thesame acquisition time by means of PS-RESOLFT(10) Similarly under the same parameters the100-Wcm2 read-out intensity used for PA NL-SIM shown in Fig 3A is 200 times weaker thanthat which would be required for localization

aab3500-6 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 5 Live-cell 3D PA NL-SIMvia lattice light sheet micros-copy (A) (Top) Membranemarker Skylan-NS-TOM20showing mitochondria in aCOS-7 cell at 23degC color-codedfor distance from the substrate(Bottom) Evolution of individualmitochondria showing fissionand fusion events the formerpreceded by mitochondrial con-striction (Movie 10 and fig S44)(B) Time-lapse distribution ofGolgi-resident enzyme Skylan-NS-Mann II in a U2OS cell at23degC showing centralizedcisternae surrounded byvesicles White arrowheads indi-cate a docking vesicle and redarrowheads highlight rapidexport of a long tubular vesicle(movies S17 and S18) Scalebars 5 mm (A) top 1 mm (A)bottom and 3 mm (B)

RESEARCH | RESEARCH ARTICLE

microscopy (6 7) and 640000 times less than PS-RESOLFT (10) Furthermore STED andRESOLFTrequire an additional depletion step not neededin PA NL-SIM which would further expose thesample to peak intensities of 807 MWcm2 forSTED (8) 17 MWcm2 for PS-RESOLFT (10) and3 kWcm2 forWF-RESOLFT (11) Even over smallimage fields nanoscopy with focused light suchas PS-RESOLFT and STED uses intensities 105-to 1010-fold larger than that of terrestrial solarflux and is thus ill-equipped to study live-celldynamics noninvasivelyOf course despite these gains no method of

live-cell fluorescencemicroscopy including high-NA TIRF-SIM and PA NL-SIM can claim to becompletely noninvasive owing to possible photo-induced physiological changes protein over-expression andor label-induced perturbationsFor example the gradual development of curvedfilopodia and membrane ruffles after the start ofimaging are shown in Movies 5 and 6 and movieS2 These may reflect a response to the illumina-tion although we have also commonly seen suchstructures under initial conditions when imagingwith diffraction-limited TIRF (fig S46) Anothercaveat is that all the cells except BSC-1 in thiswork were transiently transfected and henceexpression levels of the target proteins were un-controlled This could affect eithermorphologiessuch as the sizes of Rab5a-labeled endosomes(Fig 4 A to C and figs S38 and S39) (51) ordynamic phenotypes such as the growth rate ofmembrane ruffles inmCherryndasha-actininndashexpressingcells (Fig 4E Movie 9 and movie S15) Althoughendogenous expression levels can be achievedwith genome editing (52) even more light orlonger exposures would be needed for cases inwhich these levels are lower than those used hereThus the biological findings described in this workshould not be considered definitive More exten-sive measurements across multiple cell lines withcareful controls and targeted perturbation experi-ments will be needed to reach conclusive insightsThe lesson is that when addressing any biolog-

ical question by means of live-cell imaging it isprudent to startwith less invasive lower-resolutionmethods such aswidefield spinning disk confocalor lattice light sheetmicroscopy andmove progres-sively only as needed to more invasive higher-resolution methods such as 3D-SIM TIRF-SIMPANL-SIM and last localizationmicroscopy Seenfrom this perspective the two extended-resolutionmethods of high-NATIRF-SIMandPANL-SIMweintroduce here fill an important gap between the100-nm limit of traditional SIM and the macro-molecular level of localizationmicroscopy Togetherthey open the door to high-resolution minimallyinvasive studies of dynamic processes includingendocytosis exocytosis signal transduction proteindiffusion vesicle trafficking viral entry cytoskeletalremodeling interactions with the extracellularmatrix and the evolution of lipid rafts

Materials and methodsOptical path of the TIRF-SIM system

The schematic of TIRF-SIM system is presentedin fig S47A The beam from a laser combiner

equipped with 405 nm (250 mW RPMC OxxiusLBX-405-300-CIR-PP) 488 nm (500mW Coher-ent SAPPHIRE 488-500) and 560 nm (1W MPBCommunications 2RU-VFL-P-1000-560-B1R) lasersis passed through an acousto-optic tunable filter(AOTF AA Quanta Tech AOTFnC-400650-TN)The beam is then expanded to a 1e2 diameter of12 mm and sent to a phase-only modulator (13)consisting of a polarizing beam splitter a achro-matic half-wave plate (HWP Bolder Vision OptikBVO AHWP3) and a ferroelectric spatial lightmodulator (SLM ForthDimensionDisplays SXGA-3DM) Light diffracted by the grating patterndisplayed on SLM passes through a polarizationrotator (15) consisting of a liquid crystal cell (LCMeadowlark SWIFT) and an achromatic quarter-wave plate (QWP Bolder Vision Optik BVOAQWP3) which rotates the linear polarizationof the diffracted light so as to maintain thes-polarization necessary to maximize the patterncontrast for all pattern orientations A mask con-sisting of a hollow barrel with slots for differentpattern orientations (15) is driven by a galvano-metric scanner (Cambridge Technology 6230HB)to filter out all diffraction orders created by thebinary and pixelated nature of the SLM exceptfor the desired plusmn1 diffraction orders These arethen imaged at the back focal plane of the ob-jective (Olympus APON 100XHOTIRF 17 NA forhigh-NATIRF-SIMOlympusUAPON100XOTIRF149 NA for PA NL-SIM at 23degC or Zeiss Plan-Apochromat 100X Oil-HI 157 NA for high-NAPA-NL-SIM at 37degC) as two spots at oppositesides of the pupil After passage through the ob-jective the two beams intersect at the interfacebetween the coverslip and the sample at an angleexceeding the critical angle for total internal re-flection An evanescent standing wave penetrat-ing ~100 nm into the sample is thereby generatedconsisting of a sinusoidal pattern of excitationintensity that is a low-pass filtered image of theSLM pattern The period orientation and rela-tive phase of this excitation pattern can befinely tuned by altering the corresponding pat-tern displayed on SLM For each orientationand phase of the applied excitation pattern theresulting fluorescence is collected by the ob-jective focused by a tube lens at an interme-diate image plane separated from excitationlight by a dichroic mirror (Chroma ZT405488560tpc_225deg) placed between two relaylenses and reimaged onto a sCMOS camera(Hamamatsu Orca Flash 40 v2 sCMOS) wherethe structured fluorescence emission pattern isrecorded

Calibration of pattern overlap forPA NL-SIM

In order to maximize the amplitudes of the non-linear harmonics for PA NL-SIM to work efficient-ly the sinusoidal patterns of 405 nm activationlight and 488 nmexcitation and deactivation lightmust be aligned to precisely overlap one anotherAs noted above these patterns at the sampleplane are created by displaying correspondingbinary grating patterns on an SLM at a corre-sponding optically conjugate plane In this case

the period ps at the specimen is related to theperiod pSLM at the SLM by

ps =Ml middot pSLM eth1THORN

where M is the demagnification factor betweenthe two conjugate planes and is dictated to bethe focal lengths of the relay lenses between thetwo planes Unfortunately chromatic aberrationleads to slightly different focal lengths for evenachromatic relay lenses for different wavelengthsof light In particular in our system M405 andM488 vary by ~2 Considering that the sinusoi-dal interference pattern is composed of hundredsof periods across our 45- by 45-mm2 field-of-view(FOV) even this 2 difference results in sub-stantial drift in the relative phases of the 405-and 488-nmexcitationpatterns across the FOV (figS48 A to C) leading to spatially variable ampli-tudes for thenonlinearharmonics and correspond-ing spatially variable errors in the resultingSIM reconstructionsA straightforward way to compensate for

chromatic aberration and achieve identical peri-ods ps405 = ps488 at the sample (fig S47B) is tointroduce a period difference DpSLM between thetwo corresponding patterns at the SLM (figS47C) In fact in order to compensate completelyand achieve well-overlapped 405- and 488-nmexcitation patterns over the whole FOV we needto measure two parameters the initial perioddifference at the sampleDpi

s frac14 Dpis488 minus Dpis405

when pSLM is the same for bothwavelengths andthe phase differenceDfis frac14 Dfis488 minus Dfis405 whenps is the same Do to so we used a sample con-sisting of a dense but submonolayer spread ofgreen fluorescent beads excitable at both 405and 488 nm and proceeded as follows

Step 1

Keeping pSLM constant we acquired five imageseach of the sample under 405- and 488-nm sinus-oidal excitation with the phase shifted by pSLM5for each image at a given wavelength We then ap-plied the structured illumination (SI) reconstruc-tion algorithm (53) to each set of five images fromwhich pis405 and pi

s488 emerged as measuredoutputs For a given period pSLM488 used at theSLM for 488-nm excitation the correspondingperiod pSLM405 needed at the SLM for 405-nmexcitation to produce the same period ps at thesample for both wavelengths is then given by

pSLM405 frac14pis488pis405

pSLM488 eth2THORN

Step 2

After adjusting pSLM405 and pSLM488 to obtainthe same period ps at the sample for both wave-lengths a constant phase offset exists betweenthe two sinusoidal illumination patterns acrossthe FOV (fig S48 D and E) We measured thephase f for each wavelength by applying thesinusoidal illumination for that wavelength andthen recorded the position xn along the modu-lation direction and intensity In for each of Nbeads scattered across the FOV We then fit the

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-7

RESEARCH | RESEARCH ARTICLE

function I(x) = Imax[1 + sin(2pxps + f)]2 to thisdata to find f (fig S48F) A phase shift Df = f488 ndashf405 was then applied the SLM pattern for the405-nm illumination so as to bring it into phasewith the 488-nm illumination at the specimen(figs S48 G to I)

Step 3

Last we confirmed that both the period and phaseof the sinusoidal illumination patterns at the twowavelengths match across the entire FOV byremeasuring the periods ps488 ps405 and thephases f488 f405 as described above and con-firming that they are identical

SLM pattern generation

We generated the sinusoidal illumination pat-terns using a binary ferroelectric SLM (Forth Di-mension Displays SXGA-3DM) because it hasthe submillisecond switching times needed toacquire the nine (TIRF-SIM) 25 (PA NL-SIM) ormore (saturated PA NL-SIM) raw images of dif-ferent phase and orientation required to recon-struct a single SIM image in as fast as 100 to400msHowever care must be taken to account for thefinite pixel size of the SLM especially consideringthat subpixel adjustment accuracy is necessary toachieve precise pattern overlap at 405 and488nmas described in the previous section The SLMpattern-generation algorithms used in previouswork (13ndash15) do not provide such subpixel accu-racy Thus in this work we developed a newalgorithm that matches the two pattern periodsto 002 precision leading to a phase error nogreater than 18deg over the 45-mm FOVIn detail a set of radial vectors An define the

desired orientations of the grating pattern at theSLM The angular orientation of this radial setrelative to the x and y axes defined by pixel rowsand columns of the SLM is chosen so that eachvector is at least 4deg away from either axis This isessential to achieve subpixel precision in the ad-justment of the period For each orientation rep-resented by An we define a vector Bn that isorthogonal to An (fig S49) Likewise for everypixel of the SLM we define a pixel vector (suchas C1 or C2 in fig S49) from the point O at theintersection of An and Bn to the pixel We thencalculate F = [(C middotB)modp]p the fraction of theperiod p by which the pixel extends beyond anintegral number of periods on the SLM For apattern with a desired off fraction D per period(D = 05 in 2D SIM) the pixel is set to 0 if F lt Dand set to 1 otherwise Last to define the pixelpatterns required for the other N ndash 1 phases ofthe illumination for a given orientation the pointO is translated along Bn in steps of pN and thisprocess is repeatedwith the new vectorC for eachpixel Unlike the pixel assignment algorithm usedpreviously for SIM (15) this approach does notrely on unit-cell repetition and therefore doesnot succumb to error accumulation over theentire span of the SLM

Lattice light sheet PA NL-SIM system

To extend PA NL-SIM to three dimensions it isessential to minimize out-of-focus fluorescence

emission that can cause the shot noise in the DCharmonic to completely overwhelm the weaksignals in the nonlinear harmonics To accom-plish this we turned to the SIM mode of latticelight sheet microscopy (42) Just as in the case of2D-SIM and for the same reasons we chooseto introduce the nonlinear harmonics throughpatterned activation of Skylan-NS The excitationobjective (Special Optics 065 NA 374 mmWD)is placed perpendicular to the detection objective(Nikon CFI Apo LWD 25XW 11 NA 2 mmWD)to confine the illumination to the proximity ofthe latterrsquos focal plane (fig S42A) The latticepattern projected on the SLM (Forth DimensionDisplays SXGA-3DM) is imaged onto the focalplane of the excitation objective after the excita-tion is first spatially filtered by an annular mask(Photo-Science) and relayed by a pair of galva-nometers (Cambridge Technology 6215H) thatphase step the pattern in the x direction and scanthe light sheet in z Also as in 2D PA NL-SIM wematch the periods and phases of the 405- and488-nm lattices to exactly match by measuringtheir excitation profiles across the FOV using fluo-rescent beads (fig S42B) and adjusting accord-ingly The fluorescence emission is collected bythe detection objective and imaged by a tube lensonto a sCMOS camera (Hamamatsu Orca Flash40 v2) A 3D image is formed by repeating thisprocess as the sample is translated through thelight sheet with a piezoelectric stage (PhysikInstrumente P-6211CD) along an axis s in theplane of the cover slip and a 3D super-resolutionNL-SIM image is reconstructed as describedbelow

Data acquisitionHigh-NA TIRF SIM

All high-NA TIRF-SIM images were acquiredwith the Olympus 17-NA objective under thephysiological conditions of 37degC and 5 CO2 Ateach time point we acquired three raw images atsuccessive phase steps of 0 13 and 23 of theillumination period We then repeated this pro-cess with the standing wave excitation patternrotated plusmn120deg with respect to the first orienta-tion for a total of nine raw images The phasestepping and pattern rotation were accomplishedby rotating or translating the binary grating pat-terndisplayedon theSLMFormulticolor imagingwe acquired nine raw images at each excitationwavelength before moving to the next and thenrepeated this series at successive time points Wecould adjust the excitationNA for eachwavelengthby changing the period of the grating pattern at theSLM This allowed us to control penetration depthof the evanescent wave (fig S8) in order to ba-lance the number of excitable fluorescent mole-cules against the background fluorescence andpossible physiological effects of the excitation

PA NL-SIM and saturated PA NL-SIM

The high refractive index immersion oil requiredfor the Olympus 17-NA objective strongly ab-sorbs 405-nm light leading to a substantial reduc-tion in the modulation depth we could achieve inthe activation pattern at this wavelength Conse-

quently forNL-SIMwe first turned to theOlympus149-NA TIRF objective and imaged at room tem-perature (23degC) with L15 medium without phenolred having 10 fetal bovine serum (Life Technol-ogies) With this objective we were able to achievehigh modulation contrast while stably and pre-cisely overlapping the 405- and 488-nm standingwaves over the whole FOV An excitation NA of144 was used for both 488- and 560-nm light inthis case leading to 62-nm resolution for PANL-SIMwhen using green-emitting FPs Recently how-ever we found that the high refractive index im-mersion oil used for the Zeiss 157-NA objectivedid not absorb 405-nm light strongly and there-fore could be used to maintain precisely over-lapped 405- and 488-nm standing waves withhigh modulation contrast at 37degC and 5 CO2The excitation NA in this case was 152 for 488-nmlight leading to 59-nm resolution for PA NL-SIMwhen using green-emitting FPsThe exposure procedure for a single phase step

inNL-SIMconsists of (i) 405-nmpatterned illumi-nation for 1 ms to activate the fluorescent mol-ecules (ii) 488-nm patterned illumination for 5 to~30 ms to read-out the activated molecules and(iii) 488-nm uniform illumination for 2 to ~10 msto read-out the remaining activated molecules andreturn the sample back to the original unactivatedstate We collected the fluorescence from bothsteps (ii) and (iii) to reconstruct the SR imageDepending on the number of modulation har-monics H of non-negligible amplitude in theimage (H = 2 for PA-NL-SIM andH = 3 or possiblymore for saturated PA NL-SIM) we repeated thissequence for 2H + 1 raw images at each of 2H +1 angular orientations equally spaced around 360degfor a total of (2H + 1)2 raw images at each NL-SIMtime point An exceptionwas saturated PA-NL-SIMfor which to reduce the acquisition time weoften used only five orientations rather thansevenIn two-color imaging combining linear TIRF-SIM

and PA NL-SIM (Fig 4) at each time point weacquired the PANL-SIM image as discussed aboveHowever we acquired the TIRF-SIM image withfive instead of three orientations (15 raw images forthe TIRF-SIM channel at every time point) inorder to match the orientations of the five-slotgalvanometer-driven barrel mask used to pickout thedesireddiffractionorders for thePANL-SIMacquisition

3D PA NL-SIM with lattice lightsheet microscopy

Here we used a hexagonal lattice having aperiod large enough to contain two harmonicsfor each of the 405-nm activation and the 488-nm excitation (42)mdashone harmonic just belowthe Abbe limit of the 065-NA excitation objec-tive and the other at twice this period Theproduct of these patterns created a fluorescenceemission pattern containing H = 4 harmonics(fig S43F) However with a single excitation ob-jective we were limited to producing this pat-tern at only one orientation Therefore at eachplane of the 3D stack we acquired 2H + 1 = 9images resulting in improved resolution (Fig 5)

aab3500-8 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

RESEARCH | RESEARCH ARTICLE

in both the lateral and axial directions of thepattern

Reconstruction of SIM images

The raw image frames with patterned excitationwere processed and reconstructed into the super-resolved images by means of a previously de-scribed algorithm (53) In brief for each patternorientation with H modulation harmonics 2H +1 raw images are collected and Fourier transformedinto 2H + 1 information components These com-ponents are assembled by initially translating eachin Fourier space by a distance equal to the am-plitudeof the illuminationpatternvectornk0wherek0 is the spatial frequency of the illumination pat-tern and n = ndashH to H The pattern vector of eachinformation component is then fine-tuned byfinding the vector that maximizes the complexcross-correlation in the overlap region betweensuccessive components The modulation ampli-tude of the harmonic and its starting phase arefound through complex linear regression In linewith previous work (28) the modulation ampli-tudes for the highest harmonics are generally toolow for this empirical approach to work well sofor these the theoretical values of their complexamplitudes are used After fine-tuning the posi-tions and complex amplitudes of the informationcomponents in the overlap regions a generalizedWiener filter is applied to this expanded transferfunction to balance the amplitudes of the variousspatial frequencies against the underlying noiseNext an apodization function is applied to min-imize ringing artifacts when the result is Fourier-transformed back to real space However ratherthan the triangle apodization A(k) = 1 ndash kkmax

normally used (53) we applied a g apodizationA(k) = 1 ndash (kkmax)

g usually with g = 04 so thatthe higher spatial frequencies are not suppressedmore than necessary Furthermore we strictly fol-lowed the azimuthally dependent support kmax(q)of the expanded OTF (figs S7 and S30) to definethe endpoint of the apodization function This pro-vides additional suppression of ringing artifactsFor the time series data we independently imple-ment this reconstructionprocess for each timepoint

Cell culture transfection stainingand fixation

BSC-1 COS-7 U2OS andmouse embryonic fibro-blast (MEF) cells (American Type Culture Collec-tion) were grown to ~60 to 80 confluency inDulbeccorsquos modified eagle medium (DMEM) withhigh glucose and no phenol red supplementedwith 15 fetal bovine serum (Life Technologies)BSC-1 cells stably expressed EGFP-CLTA Othercells were transiently transfected with an AmaxaNucleofector 96-well shuttle system (Lonza) with1 mg DNA per 400000 cells with nucleofectionsolution and a program optimized for each cellline per the manufactures instructions Beforeimaging 25-mm or 5-mm coverslips were coatedwith 10 mgml fibronection (Millipore FC010) for24 hours before plating transfected cells Imagingwas performed in DMEM with HEPES if there isno CO2 control containing no phenol red at tem-peratures specifically stated in each case

In two-color imaging of CCPs and transferrinreceptors (TfRs) by means of high-NA TIRF-SIMMEF cells expressing clathrin light chain B fusedto the C terminal of mEmerald were incubatedwith DMEM medium containing 250 mgmLTfR bound to human transferrin conjugatedwith Alexa 568 (T23365 Life Technologies) for15 minFixed cells were treated for 15 min with fixa-

tion buffer containing 4 paraformaldehyde01 gluteraldehyde in PHEM buffer (25 mMHEPES 10mMEGTA 2mMMgCl2 and 120mMPIPES in pH 73)

Tracking analysis of CCPs

For each image frame we segmented the CCPsusing a watershed algorithm written in Matlab(MathWorks 2014a) and measured their cent-roids individually Subsequently the centroidpositionwas linked between time points using u-track 21 (54) This linking operation collectedsuccessive position information for each pit overthe entire endocytic process (Fig 2E) from ini-tiation to final internalization It was then straight-forward to determine the lifetime (Fig 2A) foreach endocytic eventIn order to precisely measure the pit diameter

(Fig 2 B and C) we first measured the systemmagnification to the camera by imaging a stan-dard fine counting grid (2280-32 Ted Pella) TheSIM image of each CCP was then deconvolvedwith the equivalent PSF of the SIM system tocompensate for the broadening due to the finiteresolution of the instrument Last we measuredthe diameter of each deconvolved pit using anintensity-weighted average radius relative to thecentroid of the pit In certain cases (Fig 2A andMovie 3) pits were color-coded at each timepoint based on the time since their initiation tothe current time pointOne challenge in this analysis was how to

identify isolated pits rather than aggregates andhow to be sure that these represented true pitsrather than noise or disorganized patches ofnonassembled clathrin To accomplish this weset some conditions during the analysis such asthat a pit must start as a spot and then evolveinto a ring at at least one time point When ana-lyzing the correlation between pit lifetime andmaximum diameter we added the further con-straint of including only those pits formed afterthe first frame in order to insure that we couldaccurately measure the entire lifetimeWhenmeasuring the associations of actinwith

clathrin we first implemented the tracking al-gorithm above to obtain time-lapse CCP imagesfor each endocytic eventWe then created amaskfor each CCP identified in each frame equal tothe CCP size plus an additional boundary of onepixelWe then applied thesemasks to each frameof Lifeact data and integrated the actin fluores-cence within each CCP-derivedmask If the actinsignal integrated over the area of a given maskincreased during the final five frames of the lifeof the associated CCP it was decided that actinwas recruited to the CCP during the final stage ofendocytosis

REFERENCES AND NOTES

1 L Schermelleh R Heintzmann H Leonhardt A guide to super-resolution fluorescence microscopy J Cell Biol 190 165ndash175(2010) doi 101083jcb201002018 pmid 20643879

2 U Schnell F Dijk K A Sjollema B N GiepmansImmunolabeling artifacts and the need for live-cell imagingNat Methods 9 152ndash158 (2012) doi 101038nmeth1855pmid 22290187

3 R P Nieuwenhuizen et al Measuring image resolution inoptical nanoscopy Nat Methods 10 557ndash562 (2013)doi 101038nmeth2448 pmid 23624665

4 X Shu et al A genetically encoded tag for correlated light andelectron microscopy of intact cells tissues and organismsPLOS Biol 9 e1001041 (2011) doi 101371journalpbio1001041 pmid 21483721

5 J D Martell et al Engineered ascorbate peroxidase as agenetically encoded reporter for electron microscopy NatBiotechnol 30 1143ndash1148 (2012) doi 101038nbt2375pmid 23086203

6 H Shroff C G Galbraith J A Galbraith E Betzig Live-cellphotoactivated localization microscopy of nanoscale adhesiondynamics Nat Methods 5 417ndash423 (2008) doi 101038nmeth1202 pmid 18408726

7 S H Shim et al Super-resolution fluorescence imaging oforganelles in live cells with photoswitchable membrane probesProc Natl Acad Sci USA 109 13978ndash13983 (2012)doi 101073pnas1201882109 pmid 22891300

8 B Hein K I Willig S W Hell Stimulated emission depletion(STED) nanoscopy of a fluorescent protein-labeled organelleinside a living cell Proc Natl Acad Sci USA 10514271ndash14276 (2008) doi 101073pnas0807705105pmid 18796604

9 V Westphal et al Video-rate far-field optical nanoscopydissects synaptic vesicle movement Science 320 246ndash249(2008) doi 101126science1154228 pmid 18292304

10 T Grotjohann et al rsEGFP2 enables fast RESOLFT nanoscopyof living cells eLife 1 e00248 (2012) doi 107554eLife00248 pmid 23330067

11 A Chmyrov et al Nanoscopy with more than 100000lsquodoughnutsrsquo Nat Methods 10 737ndash740 (2013) doi 101038nmeth2556 pmid 23832150

12 Materials and methods are available as supplementarymaterials on Science Online

13 P Kner B B Chhun E R Griffis L Winoto M G GustafssonSuper-resolution video microscopy of live cells by structuredillumination Nat Methods 6 339ndash342 (2009) doi 101038nmeth1324 pmid 19404253

14 L Shao P Kner E H Rego M G Gustafsson Super-resolution 3D microscopy of live whole cells using structuredillumination Nat Methods 8 1044ndash1046 (2011) doi 101038nmeth1734 pmid 22002026

15 R Fiolka L Shao E H Rego M W DavidsonM G Gustafsson Time-lapse two-color 3D imaging of live cellswith doubled resolution using structured illumination ProcNatl Acad Sci USA 109 5311ndash5315 (2012) doi 101073pnas1119262109 pmid 22431626

16 J Riedl et al Lifeact A versatile marker to visualize F-actinNat Methods 5 605ndash607 (2008) doi 101038nmeth1220pmid 18536722

17 H T McMahon E Boucrot Molecular mechanism andphysiological functions of clathrin-mediated endocytosis NatRev Mol Cell Biol 12 517ndash533 (2011) doi 101038nrm3151pmid 21779028

18 M Ehrlich et al Endocytosis by random initiation andstabilization of clathrin-coated pits Cell 118 591ndash605 (2004)doi 101016jcell200408017 pmid 15339664

19 I Gaidarov F Santini R A Warren J H Keen Spatial controlof coated-pit dynamics in living cells Nat Cell Biol 1 1ndash7(1999) pmid 10559856

20 S Saffarian E Cocucci T Kirchhausen Distinct dynamics ofendocytic clathrin-coated pits and coated plaques PLOS Biol7 e1000191 (2009) doi 101371journalpbio1000191pmid 19809571

21 J Grove et al Flat clathrin lattices Stable features of theplasma membrane Mol Biol Cell 25 3581ndash3594 (2014)doi 101091mbcE14-06-1154 pmid 25165141

22 J Heuser Effects of cytoplasmic acidification on clathrin latticemorphology J Cell Biol 108 401ndash411 (1989) doi 101083jcb1082401 pmid 2563729

23 M Kaksonen C P Toret D G Drubin Harnessing actindynamics for clathrin-mediated endocytosis Nat Rev Mol CellBiol 7 404ndash414 (2006) doi 101038nrm1940pmid 16723976

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-9

RESEARCH | RESEARCH ARTICLE

For clathrin (Fig 2) we saw that matureclathrin-coated pits (CCPs)were resolved as rings(fig S14 and movie S4) presumably because thedistal end of each resides outside the evanescentexcitation field When imaged at 37degC CCPs largeand stable enough to be resolved as a ring at one ormore time points grew to a ~152-nmmedian max-imum diameter (Fig 2B) (17) in BSC-1 cells ex-pressing EGFP-CLTA (Fig 2A and fig S14) withsimilar results inCOS-7 cells expressingmEmerald-CLTB (Fig 2F and fig S15B) and persisted foron the order of 1 min (Movie 3 and fig S15A)Linear regression revealed a positive correlationof 020 nms between maximum ring diameterand lifetime (Fig 2C) which is consistent withpreviously observed correlations between lifetimeand clathrin intensity or cargo size (18)In BSC-1 cells most CCPs were isolated at any

given time but the sites of their initiation did notfollow a Poisson distribution (fig S16) Insteadwhen the 1297 initiation events over the course ofMovie 3 were binned into 032- by 032-mm sitescorresponding to a mean density of 029 eventsper site 365 of all events occurred at sites ofone or more additional events compared with135 expected if they were Poisson distributedat this mean density Indeed at 36 of the sitesfive or more CCPs were generated sequentially(for example one marked by green arrows atdifferent time points in Fig 2D) which wouldotherwise be a very rare event (00053) assum-ing Poisson statistics Although such ldquohot spotsrdquohave been observed previously (19 20) in our case

we found that these consisted of single persistentsubdiffractive patches of clathrin fromwhichmul-tiple CCPs emerged (Movie 4) like bubblesIn COS-7 cells mEmerald-CLTB appeared as

both isolated rings and larger structures (Fig 2FMovie 5 figs S17 and S18 and movie S5) thelatter consisting of aggregates of rings (fig S19)They may be related to clathrin plaques made ofextended clathrin lattices of low curvature (20)also referred to elsewhere as flat clathrin lattices(21) Although they persisted far longer than iso-lated CCPs individual rings would occasionallydetach from these aggregates (fig S19) In no in-stance did we observe large homogenous patchesof clathrin as wemight expect for the flat clathrinlattices common in EM images (22)The role of actin in clathrin-mediated endocy-

tosis inmammalian cells remains an area of somedebate (17 23ndash26) Our two-color imaging ofCOS-7 cells by means of high-NA TIRF-SIM re-vealed that all aggregates of mEmerald-CLTBrings were associated with mCherry-Lifeact overat least part of their areas at all times (fig S19)In contrast approximately equal populations ofindividual CCPs completed endocytosis eitherwith (Fig 2E and fig S20 A and B) or without(fig S20 C and D) recruitment of Lifeact in thefinal five frames (20 s) before internalization ofthe pit In both cases histograms of CCP lifetimeswere well described by single exponential fitsindicating constant probabilities of internaliza-tion per unit time (fig S15A) The 1e lifetimes of564 plusmn 30 s for the Lifeact-associated CCPs and

672 plusmn 19 s for CCPs without Lifeact indicate thatactin when present indeed increases the inter-nalization probability Consistent with this themedianmaximum clathrin ring diameter for ringspersisting over at least five frames was slightlysmallerwith thanwithout associated Lifeact (160versus 168 nm) (fig S15B)Lifeact associatedwith CCPs usually approached

in a wave or filament (Fig 2E and fig S20A) Wealso observed rings of Lifeact (COS-7 cells) (Fig2G and fig S21 C and D) or F-tractin (U2OS cell)(fig S21 A and B) similar in size to the clathrinones and having lifetimes of several minutes (figS17B) However Lifeact rings were not as numer-ous as clathrin ones and were coincident withthem in only a few instances (fig S22) Althoughthey might be associated with other forms ofclathrin-independent endocytosis their role re-mains unclear

Live-cell nonlinear SIM via patternedactivation of photoswitchable fluorophores

To achieve even higher resolution than that ofhigh-NA TIRF-SIM we turned to nonlinear SIM(NL-SIM) The nonlinearity inherent in eitherthe patterned saturation of fluorescence excita-tion at high intensity (27 28) or the patterneddepletion of photoswitchable fluorophores (figS23) (29) introduces additional harmonicsH whichpermit resolution extension at wavelength l viaSIM to el=frac122NAethH thorn 1THORN with H ge 2 comparedwith H = 1 for the traditional linear form ofSIM andH = 0 for diffraction-limited widefield

aab3500-2 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 1 High-speed live-cell imaging at 37degC of associations between pro-teins at sub-100-nm resolution (A) Cytoskeletal proteins mApple-F-tractin(purple) and EGFP-myosin IIA (green) in a mouse embryonic fibroblast cell(Movie 1 and fig S9) (B) Magnified view of the boxed region in (A) show-ing bipolar myosin IIA filaments with clearly resolved opposed head groups(for example green arrowhead) (C) mApple-F-tractin (purple) and the fo-cal adhesion protein mEmerald-paxillin (green) in a U2OS cell (movie S2)

(D) Magnified view of the boxed region in (C) showing association of paxillinwith smaller actin fibers fanning out from the ends of larger stress fiber cables(E) Focal adhesion proteins mTagRFP-vinculin (purple) and mEmerald-paxillin (green) in a HFF-1 cell (Movie 2 and figs S10 and S13) (F) Magnifiedview of the boxed region in (E) showing a gradient of increased paxillinconcentration toward the cell periphery Scale bars 5 mm(A) (C) and (E) 1 mm(B) (D) and (F)

RESEARCH | RESEARCH ARTICLE

imaging with uniform illumination (fig S24 Aand B) Resolution of ~50 nm has been demon-strated with both approaches although not onliving cells saturated excitation was used to im-age densely labeled fluorescent beads at the like-ly phototoxic peak intensity of 8 MWcm2 (28)whereas saturated depletion (SD) imaged singlefixed cells at 945 sframe (29)mdashfar too slow tofollow most cellular processesSD provides the basis of resolution enhance-

ment in STED and RESOLFT as well as SD NL-SIM The degree of enhancement depends on thedegree of saturation (figs S23 and S25 and movieS6 part 3) defined in multiples of the saturationfactor (SF) for which 1e of the irradiated mol-

ecules remain in the original activated or excitedstate However high SFs are very photon ineffi-cient only a fraction of the photobleaching-dictated number of switching cycles for anygiven molecule then contributes useful signal(figs S25 to S27) Furthermore high SFs requirehigh intensities andor long exposures (fig S28)neither of which is compatible with fast non-invasive live-cell imagingWe addressed these issues by using patterned

activation (PA) followed with patterned excita-tion and readout of the green photoswitchableFP Skylan-NS (fig S29 andmovie S6 part 1) (30)rather than SD to generate H = 2 harmonicsyielding 62-nm resolution and subsecond acqui-

sition times in TIRF for live cells (Fig 3) Thisapproach termed PA NL-SIM allowed us toachieve large amplitudes in both the first andsecond harmonics of the emission pattern (figS29 G and H) leading to SR images of highsignal-to-noise ratio (SNR) even at low activa-tion and excitation saturation factors SFact andSFexc obtained with low intensities and shortexposures (table S1) Furthermore by keepingSFact low only a small fraction of the totalmolecular population needed to be activatedfor every raw image and with H = 2 only N =(2H + 1)2 = 25 such raw images needed to beacquired to reconstruct each SIM image frame(fig S30) Consequently we could acquire subs-tantially more frames at substantially higherSNR (fig S26) in far less time (table S1) bymeans of PANL-SIM (Movie 6) than SDNL-SIM(movie S7)PA NL-SIM of Skylan-NS-Lifeact (Fig 3 A and

BMovie 6 andmovies S8 to S10) in living COS-7cells revealed considerably more detail than didTIRF-SIM (Fig 3B) in dense peripheral actin arcsand star-like junctions of single actin filamentsWe were also able to resolve individual Lifeactrings once again including rings too small to seeclearly with high-NATIRF-SIM (fig S21 C andDandmovie S8) Furthermore we could follow thedynamics of the Lifeact-decorated actin cytoskel-eton for 30 image frames acquired in 12 s each(Movie 6) This is 1250times faster and used 20timeslower intensity (100 Wcm2) than was requiredfor an image of phalloidin-labeled actin at theventral surface of a fixed BSC-1 cell obtained bymeans of dual-objective localization microscopy(31) yet the level of detail seen by the two meth-ods was comparable (fig S31) even though our62-nm resolutionwas threefold coarser than thatreported in the localization imageWe also used PANL-SIM to image keratin (fig

S6 and movie S11) and caveolin (Fig 3 C to FMovie 7 figs S32 and S33 and movie S12) inliving COS-7 cells each with Skylan-NS at aresolution of 59 nm In the latter case this wassufficient to resolve numerous caveolae movingby less than their radii during the acquisitiontime as rings which is consistent with their in-vaginated appearance bymeans of EM (32) Suchrings were not observed at caveolae in a HeLacell imagedwith RESOLFT (fig S34) (33) despitea similar reported resolution Rings of Skylan-NS-caveolin were somewhat more abundant thanCCPs (figs S17 and S18) and althoughmost werebelow100nminsize their distributionwasbroader(Fig 3D) than the 60 to 80 nm range observedwith EM However some of the larger rings (Fig3E) may represent multiple caveolae clusteredaround surface-docked vesicles (34) Caveolaealso tended to loosely cluster in long narrow rib-bons although we saw tighter aggregations ofrings (Fig 3F) as well similar to those we saw inclathrin plaques (fig S19)Our time-lapse imaging showed that most

caveolae moved only a fraction of their size dur-ing the acquisition time although more met thiscondition when slowed (35) by operating at 23degC(Movie 7 and fig S32) than when imaged at 37degC

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-3

Fig 2 Dynamics of clathrin-mediated endocytosis and the cortical actin cytoskeleton (A) CCPsresolved as rings (fig S14 and movie S4) and color-coded according to their age since initial formation atone time point from amovie of CCPdynamics in a BSC-1 cell at 37degC stably expressing EGFP-clathrin lightchain a (Movie 3) (B) Histogram of maximum diameter of each CCP over its lifetime (C) Plot of CCPoverall lifetime versus CCP maximum diameter (D) Sequential production of multiple CCPs at a CCP-generating ldquohot spotrdquo identified with green arrowheads (Movie 4 and fig S16) (E) Formation growth anddissolution of a single CCP (right) and its relationship to cortical f-actin (left) in a COS-7 cell at 37degCtransfected with mEmerald-clathrin light chain b and mCherry-Lifeact Light blue arrowheads mark timepoints at which f-actin associates with the CCP (F) Individual CCPs and clathrin plaques (green) andcortical f-actin (red) at one time point during their evolution in a COS-7 cell (Movie 5 figs S17 and S18andmovie S5) (G) Formation of a nanoscale ring of f-actin (fig S17B) Scale bars 1 mm (A) and (F) and200 nm (D) (E) and (G)

RESEARCH | RESEARCH ARTICLE

(fig S33 and movie S12) The smaller laterallymobile fraction in each case appeared as distorteddiscontinuous rings or quasiperiodic patches(fig S35) These morphologies are indicative ofmotion-induced artifacts and underscore thedifficulty of live-cell SR imaging by any methodHigher resolution must be accompanied by pro-portionally faster acquisition times to followdynamic events of a given velocity yet higherresolution also requires a quadratically increasingnumber of raw measurements for each two-dimensional (2D) image frame Even the com-paratively brief 035 swe needed to acquireN= 25raw images for each PA NL-SIM image wasinsufficient to accurately depict caveolae mov-ing by much more than our 59-nm resolution inthis timeNevertheless by further increasing SFact we

were able to saturate the fraction of molecules inthe activated state near the maxima of the pat-terned activation light (movie S6 part 2) Saturated

PA NL-SIM generates an additional harmonic(H = 3) strong enough (fig S29) to further extendthe resolution to 45 nm (figs S36 and S37) andallowed us to identify even smaller Skylan-NS-caveolin rings unresolvable without the extraharmonic (Fig 3C) UsingN = 35 rather thanN =49 raw images per frame we balanced the re-sulting anisotropic resolution (fig S30) againstthe needs for rapid acquisition (049 sframe) andparsimonious use of the photon budget to im-age caveolin rings over 12 frames at 3 s intervals(movie S13)

Two-color live imaging via combinedTIRF-SIM and PA NL-SIM

By combining linear SIM and PA NL-SIM bothin TIRF we could study associations betweenfluorescent proteins one conventional and onephotoswtichable in two colors at higher resolu-tion than by means of linear TIRF-SIM aloneImages (Fig 4 A to C and figs S38 and S39) and

movies (Movie 8 and movie S14) of mCherry-Rab5amdasha regulator of the formation fusion andtransport of early endosomes (EEs) (36)mdashrevealedirregularly shaped dynamically remodelingpatchesof Rab5a (fig S39 A and B) consistent with thetubularvesicular architecture of EEs seen in EM(Fig 4B) (37) Numerous patches also featureddark spots (fig S39C) perhaps indicative of car-go or internal vacuoles depleted of Rab5a Mostpatchesmoved randomly between successive 20-stime intervals at velocities slow enough to avoidmotion artifacts during each 034-s acquisitionWe also observed a subpopulation of slowly grow-ing Skylan-NS-Lifeactndashassociated Rab5a patchesthat were constrained for minutes at a time (figS39D arrows) At the other extreme we occa-sionally observed streaks of Rab5amoving paral-lel to nearby actin filaments at velocities of 3 to5 mms (Fig 4C and fig S39E) These may repre-sent EEs actively transported alongmicrotubules(38) parallel to the filaments

aab3500-4 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 3 Live-cell non-linear structured illumi-nation microscopybased on patternedphotoactivation (A)Single time point from amovie of the evolution ofcortical f-actin in a COS-7cell at 23degC transfectedwith Skylan-NS-Lifeactseen at 62-nm resolution(Movie 6 fig S31 andmovie S8) (B) Magnifiedview from a different cellat 37degC comparingdiffraction-limited TIRFmicroscopy (top left)TIRF with deconvolution(top right) TIRF-SIM(bottom left) and non-linear TIRF-SIM withpatterned activation(PA NL-SIM bottomright) (movies S9 andS10) (C) Caveolae in aCOS-7 cell at 23degC trans-fected with Skylan-NS-caveolin comparing TIRFwith deconvolution (topleft 220-nm resolution)TIRF SIM (top right97-nm resolution) PANL-SIM (bottom left62-nm resolution) andsaturated PA NL-SIM(bottom right 45-nmresolution) (Insets) Asingle caveolae pit even-tually resolved as a ringby saturated PA NL-SIM(Movie 7 figs S34 toS37 and movie S13(D) Diversity of caveolae ring diameters as seen by means of PA NL-SIM (E) Larger rings that may represent surface-docked vesicles (F) Clusters of caveolaereminiscent of clathrin plaques (D) to (F) are from a different cell at 37degC (fig S33 and movie S12) Scale bars 3 mm (A) 1 mm (B) 200 nm (C) and 100 nm(D) (E) (F) and (C) inset

RESEARCH | RESEARCH ARTICLE

We also used PA NL-SIM and TIRF-SIM re-spectively to study the association of Skylan-NS-Lifeact with mCherry-a-actinin (Fig 4 D to Fand fig S40) Consistent with its role as an actin-bundling protein (39) in COS-7 cells we founda-actinin at the treadmilling edge of the lamelle-podium and at the basal surface in both filopodiaand the leading edges of growing membraneruffles (Fig 4F Movie 9 and movie S15) We alsoobserved concentrations of a-actinin along thesides (Fig 4E) and at the branching ends of stressfibers that likely attach to cell-substrate adhesions(40) Last a-actininwas present at dense junctionsof Lifeact-decorated filaments and Skylan-NS-Lifeact rings as described above were colocalizedin every instance with a mCherryndasha-actinin ringof similar size (fig S41) Septins another class ofactin-bundling proteins have been shown (41) toproduce f-actin rings in vitro (albeit of larger sizethan here) so perhaps a-actinin not only aids inbundling actin filaments in nanometric rings butalso contributes to their extreme curvature

3D live-cell imaging with combined PANL-SIM and lattice light sheet microscopy

Although the ~50- to 200-nm extent of the eva-nescent excitation field we used in the examplesabove eliminated out-of-focus background andconfined potentially phototoxic exposure to aminute fraction of the cellular volume it alsolimited our observations to this subvolume andseverely restricted the total photon budget avail-able for those targets unable to be replenishedfrom the cytosol during the imaging intervalTo extend our observations to the entire cell

we turned to live-cell 3D-SIM (14 15) Unfortu-nately traditional 3D-SIM with linear widefieldexcitation brings limitations of its own It is slow

(~20 s acquisition for whole adherent HeLa cells)limited to thin specimens (because of out-of-focusbackground) and requires high SNR for accurateimage reconstruction It is also potentially photo-toxic and bleaches specimens rapidly because ofcontinuous whole-cell illumination These prob-lems would all be greatly magnified in its directextension to PA NL-SIMThus to apply PA NL-SIM to living cells in

three dimensions (Fig 5) we used lattice lightsheet microscopy (42) In this technique an exci-tation objective (fig S42A) projects a thin sheetof light (fig S42A blue) through a specimen (figS42A orange) and the fluorescence generated inthe illuminated plane is collected by a detectionobjective and imaged onto a camera Repeatingthis process plane-by-plane through the specimenproduces a 3D image Restriction of the light tothe detection focal plane eliminates out-of-focusbackground increases the z axis resolution andgreatly reduces photobleaching and phototoxicityIn cross-section the light sheet has the 2D

periodic structure of an optical lattice (fig S42B)Sweeping the sheet back and forth along the xaxis produces time-averaged uniform illumina-tion offering high speed and diffraction-limitedxyz resolution of 230 by 230 by 370 nm as seenin a volume-rendered image of the actin cyto-skeleton (fig S43A) and its corresponding overalloptical transfer function (OTF) (fig S43D) Step-ping the sheet in x in five equal fractions of thelattice period and applying the algorithms of3D-SIM to the resulting five raw images perplane extends the xyz resolution to 150 by 230by 280 nm (fig S43 B and E) but at the cost of atleast 5times longer acquisition times (42)To further extend the 3D resolution via PA

NL-SIMwe first photoactivated targetmolecules

fused to Skylan-NS using a hexagonal lattice lightsheet of l = 405 nm wavelength having H = 2harmonics (fig S43E) We then imaged the fluo-rescence from the activated region exciting thefluorescencewith a lattice light sheet of l =488nmwavelength having the same hexagonal sym-metry and period (fig S42B bottom) as the ac-tivation lattice For activationwell below saturationthe product of the activation and excitation pat-terns creates a fluorescence emission patternwithin the specimen having H = 4 harmonics(fig S43F) Thus we stepped the sheet in x in2H + 1 = 9 equal fractions of the lattice periodwhile recording nine images Repeating this pro-cess for every plane within the specimen we thenreconstructed a 3D PA NL-SIM volume-renderedimage (fig S43C) with resolution extended to118 by 230 by 170 nmWe used this approach to image mitochondria

in COS-7 cells (Fig 5A) as well as the actin cyto-skeleton (fig S43 A to C and movie S16) andthe Golgi apparatus (Fig 5B) inU2OS cells all at23degC so as to simplify the overlap of the activa-tion and excitation patterns Time-lapse 3D im-ages (Fig 5A bottom) and movies (Movie 10) ofSkylan-NSndashtagged translocase of outer mitochon-drial membrane 20 (TOM20) revealed the mi-gration constriction before fission and fusionof individual mitochondria (43 44) each clearlyresolved as a hollow tubular structure The 3Dvolume rendering and the widths of mitochon-drial membranes in individual xy orthosliceswere both comparable with similar data from afixed cell imaged with 3D localization micros-copy (45) at a reported xyz resolution of ~20 by20 by 60 nm (fig S44)A volume-rendered movie (movie S17) of the

Golgi-resident enzyme Mannosidase II (MannII)

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-5

Fig 4 Combined TIRF-SIMand PA NL-SIM of protein-pair dynamics in livingcells (A) Skylan-NS-Lifeact(orange PA NL-SIM) andmCherry-Rab5a a marker ofearly endosomes (greenTIRF-SIM) in a COS-7 cellat 23degC (Movie 8 figs S38and S39 and movie S14)(B) Comparison of EMimages of early endosomes(37) with similarly shapedRab5a patches seen in (A)(C) Magnified view at threesuccessive time pointsshowing rapid transportof a Rab5a streak parallel tothe cytoskeleton (D) Skylan-NS-Lifeact (green PA NL-SIM) and mCherry-a-actinin(purpleTIRF-SIM) in a COS-7cell at 23degC (Movie 9 figsS40 and S41 and movieS15) (E) Magnified viewfrom (D) with Lifeact (top)a-actinin (middle) and overlay (bottom) showing paired association at focal adhesions and along the sides of large stress fibers (F) Evolution of amembrane ruffle showing a-actinin concentrated at the leading edge Scale bars 5 mm (A) (D) 200 nm (B) 1 mm (C) and (E) and 500 nm (F)

RESEARCH | RESEARCH ARTICLE

tagged with Skylan-NS in a U2OS cell as seenlooking into the cis-face from the nucleus showedMann II concentrated in a hollow sphere ofcisternae having a cis-facing void Time-lapse3D data (Fig 5B andmovie S18) color-coded forheight showed the docking of small vesicles (Fig5B white arrows) that may represent pre-Golgiintermediates (46) as well as the rapid export ofMann II in long tubular post-Golgi carriers (Fig5B red arrows) (47)The volumetric resolution of 3D lattice light

sheet PA NL-SIM at the 06-NA excitation and11-NA detection we used here is comparablewith the 105- by 105- by 369-nm xyz resolutionof widefield 3D-SIM at 12 NA However thelattice approach has twofold higher axial resolu-tion and fourfold better than traditional diffraction-limited microscopy It is therefore better suitedto problems in which its superior optical section-ing is essential such as in resolving heterogene-ities in nuclear architecture distinguishing eventsoccurring at the dorsal or ventral plasma mem-brane or as above tracking vesicles through thesecretory pathway Whole-cell acquisition times(705 and 327 s in Fig 5 A and B respectively)are slow compared with PA NL-SIM in TIRF butsimilar to widefield 3D-SIM However thanks tothe oblique imaging geometry (fig S42) restrictedxy fields of view can be imaged at proportion-ally faster speed through the entire thickness ofthe cell

Discussion

The above results provide but a brief glimpse ofthe biology that might be uncovered with thelive-cellndashcompatible SRmethods of high-NATIRF-SIM and PA NL-SIM We have measured andcorrelated the diameters and lifetimes of CCPsobserved at high resolution different forms ofCCP initiation and shown that CCP internaliza-

tion is aided by actin filaments in about half of allcases We have seen that caveolin localizes notonly to the 60- to 80-nm invaginated caveloaecommon in EM images but also to much largerring-like structures and have followed dynamicchanges in the shapes of early endosomes Lastwe have observed the nanoscale remodeling ofthe actin cytoskeleton in relation to clathrin andRab5a as well as cytoskeletal-related proteinssuch as myosin IIA a-actinin and paxillinHowever the above results also amply illus-

trate the trade-offs inherent in live SR imagingWith high-NA TIRF-SIM at 17 NA we could ac-quire up to 200 image frames in lt05 s each atintensities of 20 to 100 Wcm2 and a resolutionof 84 nm (for GFP) whereas extending the reso-lution to 62 nm with PA NL-SIM restricted us tono more than 40 frames and further extensionto 45 nm with saturated PA NL-SIM required490 Wcm2 and produced only 12 frames atuseful SNRIn short evenmodest gains in resolution come

at substantial cost in terms of the other metricsimportant for live-cell imaging These tradeoffsare not specific to SIM In fact our extensions ofSIM are far more compatible with live imagingthan any other form of SR fluorescence micros-copy of comparable resolution demonstrated todate In part this is because the OTF which de-fines the degree towhich different sample spatialfrequencies (representing differently sized struc-tures) are passed to the image is far stronger inthe 100-nm regime (fig S24B) for high-NA TIRF-SIM at 17 NA than other linear methods such asconfocal or image scanning microscopy (ISM)(48ndash50) and far stronger in the 50- to 100-nmregime (fig S24C) for PA NL-SIM than othernonlinear methods such as STED (8 9) point-scanning (PS) RESOLFT (10) or array-basedwide-field (WF) RESOLFT (11) As a result far fewer

photons need to be collected (fig S2) and far lesslight (fig S3) needs to be applied to the specimento see features in these regimes at acceptableSNR Localization microscopy is also photon in-efficient in that the density of localizedmoleculesis nearly always more limiting to the resolutionthan is the number of photons emittedper switch-ing cycle which dictates the localization preci-sion For example simulations (12) based on thetheoretical OTFs suggest that to resolve an 85-nmgrating PANL-SIM requires ~80times fewer photonsfrom the specimen per unit area than localiza-tionmicroscopy ~200times fewer thanWF-RESOLFTand ~15times fewer than PS-RESOLFT or STED eachat a depletion saturation factor of SFdepletion = 10(fig S2)Another reason for the greater compatibility of

high-NA TIRF-SIM and PA NL-SIM with livingcells is that they require much lower peak inten-sities of applied light High resolutionwith STEDor RESOLFT demands high factors of saturateddepletion (fig S25 A and C) that are wasteful ofthe photon budget (fig S25 B andD) and requireenormous intensities andor long exposures foractivation (fig S45) depletion (fig S28) and read-out of the final signal (fig S3) Localization mi-croscopy also requires high intensities to achievehigh photon emission and photoswitching ratesfrom single molecules For example extrapolat-ing from reported experimental values for live-cell imaging (table S1) the 08- to 35-Wcm2

activation intensity used over the 45- by 45-mmfield of view in Fig 3A in 12 s bymeans of PANL-SIM is 960000 times weaker than that whichwould be required to image the same area in thesame acquisition time by means of PS-RESOLFT(10) Similarly under the same parameters the100-Wcm2 read-out intensity used for PA NL-SIM shown in Fig 3A is 200 times weaker thanthat which would be required for localization

aab3500-6 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 5 Live-cell 3D PA NL-SIMvia lattice light sheet micros-copy (A) (Top) Membranemarker Skylan-NS-TOM20showing mitochondria in aCOS-7 cell at 23degC color-codedfor distance from the substrate(Bottom) Evolution of individualmitochondria showing fissionand fusion events the formerpreceded by mitochondrial con-striction (Movie 10 and fig S44)(B) Time-lapse distribution ofGolgi-resident enzyme Skylan-NS-Mann II in a U2OS cell at23degC showing centralizedcisternae surrounded byvesicles White arrowheads indi-cate a docking vesicle and redarrowheads highlight rapidexport of a long tubular vesicle(movies S17 and S18) Scalebars 5 mm (A) top 1 mm (A)bottom and 3 mm (B)

RESEARCH | RESEARCH ARTICLE

microscopy (6 7) and 640000 times less than PS-RESOLFT (10) Furthermore STED andRESOLFTrequire an additional depletion step not neededin PA NL-SIM which would further expose thesample to peak intensities of 807 MWcm2 forSTED (8) 17 MWcm2 for PS-RESOLFT (10) and3 kWcm2 forWF-RESOLFT (11) Even over smallimage fields nanoscopy with focused light suchas PS-RESOLFT and STED uses intensities 105-to 1010-fold larger than that of terrestrial solarflux and is thus ill-equipped to study live-celldynamics noninvasivelyOf course despite these gains no method of

live-cell fluorescencemicroscopy including high-NA TIRF-SIM and PA NL-SIM can claim to becompletely noninvasive owing to possible photo-induced physiological changes protein over-expression andor label-induced perturbationsFor example the gradual development of curvedfilopodia and membrane ruffles after the start ofimaging are shown in Movies 5 and 6 and movieS2 These may reflect a response to the illumina-tion although we have also commonly seen suchstructures under initial conditions when imagingwith diffraction-limited TIRF (fig S46) Anothercaveat is that all the cells except BSC-1 in thiswork were transiently transfected and henceexpression levels of the target proteins were un-controlled This could affect eithermorphologiessuch as the sizes of Rab5a-labeled endosomes(Fig 4 A to C and figs S38 and S39) (51) ordynamic phenotypes such as the growth rate ofmembrane ruffles inmCherryndasha-actininndashexpressingcells (Fig 4E Movie 9 and movie S15) Althoughendogenous expression levels can be achievedwith genome editing (52) even more light orlonger exposures would be needed for cases inwhich these levels are lower than those used hereThus the biological findings described in this workshould not be considered definitive More exten-sive measurements across multiple cell lines withcareful controls and targeted perturbation experi-ments will be needed to reach conclusive insightsThe lesson is that when addressing any biolog-

ical question by means of live-cell imaging it isprudent to startwith less invasive lower-resolutionmethods such aswidefield spinning disk confocalor lattice light sheetmicroscopy andmove progres-sively only as needed to more invasive higher-resolution methods such as 3D-SIM TIRF-SIMPANL-SIM and last localizationmicroscopy Seenfrom this perspective the two extended-resolutionmethods of high-NATIRF-SIMandPANL-SIMweintroduce here fill an important gap between the100-nm limit of traditional SIM and the macro-molecular level of localizationmicroscopy Togetherthey open the door to high-resolution minimallyinvasive studies of dynamic processes includingendocytosis exocytosis signal transduction proteindiffusion vesicle trafficking viral entry cytoskeletalremodeling interactions with the extracellularmatrix and the evolution of lipid rafts

Materials and methodsOptical path of the TIRF-SIM system

The schematic of TIRF-SIM system is presentedin fig S47A The beam from a laser combiner

equipped with 405 nm (250 mW RPMC OxxiusLBX-405-300-CIR-PP) 488 nm (500mW Coher-ent SAPPHIRE 488-500) and 560 nm (1W MPBCommunications 2RU-VFL-P-1000-560-B1R) lasersis passed through an acousto-optic tunable filter(AOTF AA Quanta Tech AOTFnC-400650-TN)The beam is then expanded to a 1e2 diameter of12 mm and sent to a phase-only modulator (13)consisting of a polarizing beam splitter a achro-matic half-wave plate (HWP Bolder Vision OptikBVO AHWP3) and a ferroelectric spatial lightmodulator (SLM ForthDimensionDisplays SXGA-3DM) Light diffracted by the grating patterndisplayed on SLM passes through a polarizationrotator (15) consisting of a liquid crystal cell (LCMeadowlark SWIFT) and an achromatic quarter-wave plate (QWP Bolder Vision Optik BVOAQWP3) which rotates the linear polarizationof the diffracted light so as to maintain thes-polarization necessary to maximize the patterncontrast for all pattern orientations A mask con-sisting of a hollow barrel with slots for differentpattern orientations (15) is driven by a galvano-metric scanner (Cambridge Technology 6230HB)to filter out all diffraction orders created by thebinary and pixelated nature of the SLM exceptfor the desired plusmn1 diffraction orders These arethen imaged at the back focal plane of the ob-jective (Olympus APON 100XHOTIRF 17 NA forhigh-NATIRF-SIMOlympusUAPON100XOTIRF149 NA for PA NL-SIM at 23degC or Zeiss Plan-Apochromat 100X Oil-HI 157 NA for high-NAPA-NL-SIM at 37degC) as two spots at oppositesides of the pupil After passage through the ob-jective the two beams intersect at the interfacebetween the coverslip and the sample at an angleexceeding the critical angle for total internal re-flection An evanescent standing wave penetrat-ing ~100 nm into the sample is thereby generatedconsisting of a sinusoidal pattern of excitationintensity that is a low-pass filtered image of theSLM pattern The period orientation and rela-tive phase of this excitation pattern can befinely tuned by altering the corresponding pat-tern displayed on SLM For each orientationand phase of the applied excitation pattern theresulting fluorescence is collected by the ob-jective focused by a tube lens at an interme-diate image plane separated from excitationlight by a dichroic mirror (Chroma ZT405488560tpc_225deg) placed between two relaylenses and reimaged onto a sCMOS camera(Hamamatsu Orca Flash 40 v2 sCMOS) wherethe structured fluorescence emission pattern isrecorded

Calibration of pattern overlap forPA NL-SIM

In order to maximize the amplitudes of the non-linear harmonics for PA NL-SIM to work efficient-ly the sinusoidal patterns of 405 nm activationlight and 488 nmexcitation and deactivation lightmust be aligned to precisely overlap one anotherAs noted above these patterns at the sampleplane are created by displaying correspondingbinary grating patterns on an SLM at a corre-sponding optically conjugate plane In this case

the period ps at the specimen is related to theperiod pSLM at the SLM by

ps =Ml middot pSLM eth1THORN

where M is the demagnification factor betweenthe two conjugate planes and is dictated to bethe focal lengths of the relay lenses between thetwo planes Unfortunately chromatic aberrationleads to slightly different focal lengths for evenachromatic relay lenses for different wavelengthsof light In particular in our system M405 andM488 vary by ~2 Considering that the sinusoi-dal interference pattern is composed of hundredsof periods across our 45- by 45-mm2 field-of-view(FOV) even this 2 difference results in sub-stantial drift in the relative phases of the 405-and 488-nmexcitationpatterns across the FOV (figS48 A to C) leading to spatially variable ampli-tudes for thenonlinearharmonics and correspond-ing spatially variable errors in the resultingSIM reconstructionsA straightforward way to compensate for

chromatic aberration and achieve identical peri-ods ps405 = ps488 at the sample (fig S47B) is tointroduce a period difference DpSLM between thetwo corresponding patterns at the SLM (figS47C) In fact in order to compensate completelyand achieve well-overlapped 405- and 488-nmexcitation patterns over the whole FOV we needto measure two parameters the initial perioddifference at the sampleDpi

s frac14 Dpis488 minus Dpis405

when pSLM is the same for bothwavelengths andthe phase differenceDfis frac14 Dfis488 minus Dfis405 whenps is the same Do to so we used a sample con-sisting of a dense but submonolayer spread ofgreen fluorescent beads excitable at both 405and 488 nm and proceeded as follows

Step 1

Keeping pSLM constant we acquired five imageseach of the sample under 405- and 488-nm sinus-oidal excitation with the phase shifted by pSLM5for each image at a given wavelength We then ap-plied the structured illumination (SI) reconstruc-tion algorithm (53) to each set of five images fromwhich pis405 and pi

s488 emerged as measuredoutputs For a given period pSLM488 used at theSLM for 488-nm excitation the correspondingperiod pSLM405 needed at the SLM for 405-nmexcitation to produce the same period ps at thesample for both wavelengths is then given by

pSLM405 frac14pis488pis405

pSLM488 eth2THORN

Step 2

After adjusting pSLM405 and pSLM488 to obtainthe same period ps at the sample for both wave-lengths a constant phase offset exists betweenthe two sinusoidal illumination patterns acrossthe FOV (fig S48 D and E) We measured thephase f for each wavelength by applying thesinusoidal illumination for that wavelength andthen recorded the position xn along the modu-lation direction and intensity In for each of Nbeads scattered across the FOV We then fit the

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-7

RESEARCH | RESEARCH ARTICLE

function I(x) = Imax[1 + sin(2pxps + f)]2 to thisdata to find f (fig S48F) A phase shift Df = f488 ndashf405 was then applied the SLM pattern for the405-nm illumination so as to bring it into phasewith the 488-nm illumination at the specimen(figs S48 G to I)

Step 3

Last we confirmed that both the period and phaseof the sinusoidal illumination patterns at the twowavelengths match across the entire FOV byremeasuring the periods ps488 ps405 and thephases f488 f405 as described above and con-firming that they are identical

SLM pattern generation

We generated the sinusoidal illumination pat-terns using a binary ferroelectric SLM (Forth Di-mension Displays SXGA-3DM) because it hasthe submillisecond switching times needed toacquire the nine (TIRF-SIM) 25 (PA NL-SIM) ormore (saturated PA NL-SIM) raw images of dif-ferent phase and orientation required to recon-struct a single SIM image in as fast as 100 to400msHowever care must be taken to account for thefinite pixel size of the SLM especially consideringthat subpixel adjustment accuracy is necessary toachieve precise pattern overlap at 405 and488nmas described in the previous section The SLMpattern-generation algorithms used in previouswork (13ndash15) do not provide such subpixel accu-racy Thus in this work we developed a newalgorithm that matches the two pattern periodsto 002 precision leading to a phase error nogreater than 18deg over the 45-mm FOVIn detail a set of radial vectors An define the

desired orientations of the grating pattern at theSLM The angular orientation of this radial setrelative to the x and y axes defined by pixel rowsand columns of the SLM is chosen so that eachvector is at least 4deg away from either axis This isessential to achieve subpixel precision in the ad-justment of the period For each orientation rep-resented by An we define a vector Bn that isorthogonal to An (fig S49) Likewise for everypixel of the SLM we define a pixel vector (suchas C1 or C2 in fig S49) from the point O at theintersection of An and Bn to the pixel We thencalculate F = [(C middotB)modp]p the fraction of theperiod p by which the pixel extends beyond anintegral number of periods on the SLM For apattern with a desired off fraction D per period(D = 05 in 2D SIM) the pixel is set to 0 if F lt Dand set to 1 otherwise Last to define the pixelpatterns required for the other N ndash 1 phases ofthe illumination for a given orientation the pointO is translated along Bn in steps of pN and thisprocess is repeatedwith the new vectorC for eachpixel Unlike the pixel assignment algorithm usedpreviously for SIM (15) this approach does notrely on unit-cell repetition and therefore doesnot succumb to error accumulation over theentire span of the SLM

Lattice light sheet PA NL-SIM system

To extend PA NL-SIM to three dimensions it isessential to minimize out-of-focus fluorescence

emission that can cause the shot noise in the DCharmonic to completely overwhelm the weaksignals in the nonlinear harmonics To accom-plish this we turned to the SIM mode of latticelight sheet microscopy (42) Just as in the case of2D-SIM and for the same reasons we chooseto introduce the nonlinear harmonics throughpatterned activation of Skylan-NS The excitationobjective (Special Optics 065 NA 374 mmWD)is placed perpendicular to the detection objective(Nikon CFI Apo LWD 25XW 11 NA 2 mmWD)to confine the illumination to the proximity ofthe latterrsquos focal plane (fig S42A) The latticepattern projected on the SLM (Forth DimensionDisplays SXGA-3DM) is imaged onto the focalplane of the excitation objective after the excita-tion is first spatially filtered by an annular mask(Photo-Science) and relayed by a pair of galva-nometers (Cambridge Technology 6215H) thatphase step the pattern in the x direction and scanthe light sheet in z Also as in 2D PA NL-SIM wematch the periods and phases of the 405- and488-nm lattices to exactly match by measuringtheir excitation profiles across the FOV using fluo-rescent beads (fig S42B) and adjusting accord-ingly The fluorescence emission is collected bythe detection objective and imaged by a tube lensonto a sCMOS camera (Hamamatsu Orca Flash40 v2) A 3D image is formed by repeating thisprocess as the sample is translated through thelight sheet with a piezoelectric stage (PhysikInstrumente P-6211CD) along an axis s in theplane of the cover slip and a 3D super-resolutionNL-SIM image is reconstructed as describedbelow

Data acquisitionHigh-NA TIRF SIM

All high-NA TIRF-SIM images were acquiredwith the Olympus 17-NA objective under thephysiological conditions of 37degC and 5 CO2 Ateach time point we acquired three raw images atsuccessive phase steps of 0 13 and 23 of theillumination period We then repeated this pro-cess with the standing wave excitation patternrotated plusmn120deg with respect to the first orienta-tion for a total of nine raw images The phasestepping and pattern rotation were accomplishedby rotating or translating the binary grating pat-terndisplayedon theSLMFormulticolor imagingwe acquired nine raw images at each excitationwavelength before moving to the next and thenrepeated this series at successive time points Wecould adjust the excitationNA for eachwavelengthby changing the period of the grating pattern at theSLM This allowed us to control penetration depthof the evanescent wave (fig S8) in order to ba-lance the number of excitable fluorescent mole-cules against the background fluorescence andpossible physiological effects of the excitation

PA NL-SIM and saturated PA NL-SIM

The high refractive index immersion oil requiredfor the Olympus 17-NA objective strongly ab-sorbs 405-nm light leading to a substantial reduc-tion in the modulation depth we could achieve inthe activation pattern at this wavelength Conse-

quently forNL-SIMwe first turned to theOlympus149-NA TIRF objective and imaged at room tem-perature (23degC) with L15 medium without phenolred having 10 fetal bovine serum (Life Technol-ogies) With this objective we were able to achievehigh modulation contrast while stably and pre-cisely overlapping the 405- and 488-nm standingwaves over the whole FOV An excitation NA of144 was used for both 488- and 560-nm light inthis case leading to 62-nm resolution for PANL-SIMwhen using green-emitting FPs Recently how-ever we found that the high refractive index im-mersion oil used for the Zeiss 157-NA objectivedid not absorb 405-nm light strongly and there-fore could be used to maintain precisely over-lapped 405- and 488-nm standing waves withhigh modulation contrast at 37degC and 5 CO2The excitation NA in this case was 152 for 488-nmlight leading to 59-nm resolution for PA NL-SIMwhen using green-emitting FPsThe exposure procedure for a single phase step

inNL-SIMconsists of (i) 405-nmpatterned illumi-nation for 1 ms to activate the fluorescent mol-ecules (ii) 488-nm patterned illumination for 5 to~30 ms to read-out the activated molecules and(iii) 488-nm uniform illumination for 2 to ~10 msto read-out the remaining activated molecules andreturn the sample back to the original unactivatedstate We collected the fluorescence from bothsteps (ii) and (iii) to reconstruct the SR imageDepending on the number of modulation har-monics H of non-negligible amplitude in theimage (H = 2 for PA-NL-SIM andH = 3 or possiblymore for saturated PA NL-SIM) we repeated thissequence for 2H + 1 raw images at each of 2H +1 angular orientations equally spaced around 360degfor a total of (2H + 1)2 raw images at each NL-SIMtime point An exceptionwas saturated PA-NL-SIMfor which to reduce the acquisition time weoften used only five orientations rather thansevenIn two-color imaging combining linear TIRF-SIM

and PA NL-SIM (Fig 4) at each time point weacquired the PANL-SIM image as discussed aboveHowever we acquired the TIRF-SIM image withfive instead of three orientations (15 raw images forthe TIRF-SIM channel at every time point) inorder to match the orientations of the five-slotgalvanometer-driven barrel mask used to pickout thedesireddiffractionorders for thePANL-SIMacquisition

3D PA NL-SIM with lattice lightsheet microscopy

Here we used a hexagonal lattice having aperiod large enough to contain two harmonicsfor each of the 405-nm activation and the 488-nm excitation (42)mdashone harmonic just belowthe Abbe limit of the 065-NA excitation objec-tive and the other at twice this period Theproduct of these patterns created a fluorescenceemission pattern containing H = 4 harmonics(fig S43F) However with a single excitation ob-jective we were limited to producing this pat-tern at only one orientation Therefore at eachplane of the 3D stack we acquired 2H + 1 = 9images resulting in improved resolution (Fig 5)

aab3500-8 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

RESEARCH | RESEARCH ARTICLE

in both the lateral and axial directions of thepattern

Reconstruction of SIM images

The raw image frames with patterned excitationwere processed and reconstructed into the super-resolved images by means of a previously de-scribed algorithm (53) In brief for each patternorientation with H modulation harmonics 2H +1 raw images are collected and Fourier transformedinto 2H + 1 information components These com-ponents are assembled by initially translating eachin Fourier space by a distance equal to the am-plitudeof the illuminationpatternvectornk0wherek0 is the spatial frequency of the illumination pat-tern and n = ndashH to H The pattern vector of eachinformation component is then fine-tuned byfinding the vector that maximizes the complexcross-correlation in the overlap region betweensuccessive components The modulation ampli-tude of the harmonic and its starting phase arefound through complex linear regression In linewith previous work (28) the modulation ampli-tudes for the highest harmonics are generally toolow for this empirical approach to work well sofor these the theoretical values of their complexamplitudes are used After fine-tuning the posi-tions and complex amplitudes of the informationcomponents in the overlap regions a generalizedWiener filter is applied to this expanded transferfunction to balance the amplitudes of the variousspatial frequencies against the underlying noiseNext an apodization function is applied to min-imize ringing artifacts when the result is Fourier-transformed back to real space However ratherthan the triangle apodization A(k) = 1 ndash kkmax

normally used (53) we applied a g apodizationA(k) = 1 ndash (kkmax)

g usually with g = 04 so thatthe higher spatial frequencies are not suppressedmore than necessary Furthermore we strictly fol-lowed the azimuthally dependent support kmax(q)of the expanded OTF (figs S7 and S30) to definethe endpoint of the apodization function This pro-vides additional suppression of ringing artifactsFor the time series data we independently imple-ment this reconstructionprocess for each timepoint

Cell culture transfection stainingand fixation

BSC-1 COS-7 U2OS andmouse embryonic fibro-blast (MEF) cells (American Type Culture Collec-tion) were grown to ~60 to 80 confluency inDulbeccorsquos modified eagle medium (DMEM) withhigh glucose and no phenol red supplementedwith 15 fetal bovine serum (Life Technologies)BSC-1 cells stably expressed EGFP-CLTA Othercells were transiently transfected with an AmaxaNucleofector 96-well shuttle system (Lonza) with1 mg DNA per 400000 cells with nucleofectionsolution and a program optimized for each cellline per the manufactures instructions Beforeimaging 25-mm or 5-mm coverslips were coatedwith 10 mgml fibronection (Millipore FC010) for24 hours before plating transfected cells Imagingwas performed in DMEM with HEPES if there isno CO2 control containing no phenol red at tem-peratures specifically stated in each case

In two-color imaging of CCPs and transferrinreceptors (TfRs) by means of high-NA TIRF-SIMMEF cells expressing clathrin light chain B fusedto the C terminal of mEmerald were incubatedwith DMEM medium containing 250 mgmLTfR bound to human transferrin conjugatedwith Alexa 568 (T23365 Life Technologies) for15 minFixed cells were treated for 15 min with fixa-

tion buffer containing 4 paraformaldehyde01 gluteraldehyde in PHEM buffer (25 mMHEPES 10mMEGTA 2mMMgCl2 and 120mMPIPES in pH 73)

Tracking analysis of CCPs

For each image frame we segmented the CCPsusing a watershed algorithm written in Matlab(MathWorks 2014a) and measured their cent-roids individually Subsequently the centroidpositionwas linked between time points using u-track 21 (54) This linking operation collectedsuccessive position information for each pit overthe entire endocytic process (Fig 2E) from ini-tiation to final internalization It was then straight-forward to determine the lifetime (Fig 2A) foreach endocytic eventIn order to precisely measure the pit diameter

(Fig 2 B and C) we first measured the systemmagnification to the camera by imaging a stan-dard fine counting grid (2280-32 Ted Pella) TheSIM image of each CCP was then deconvolvedwith the equivalent PSF of the SIM system tocompensate for the broadening due to the finiteresolution of the instrument Last we measuredthe diameter of each deconvolved pit using anintensity-weighted average radius relative to thecentroid of the pit In certain cases (Fig 2A andMovie 3) pits were color-coded at each timepoint based on the time since their initiation tothe current time pointOne challenge in this analysis was how to

identify isolated pits rather than aggregates andhow to be sure that these represented true pitsrather than noise or disorganized patches ofnonassembled clathrin To accomplish this weset some conditions during the analysis such asthat a pit must start as a spot and then evolveinto a ring at at least one time point When ana-lyzing the correlation between pit lifetime andmaximum diameter we added the further con-straint of including only those pits formed afterthe first frame in order to insure that we couldaccurately measure the entire lifetimeWhenmeasuring the associations of actinwith

clathrin we first implemented the tracking al-gorithm above to obtain time-lapse CCP imagesfor each endocytic eventWe then created amaskfor each CCP identified in each frame equal tothe CCP size plus an additional boundary of onepixelWe then applied thesemasks to each frameof Lifeact data and integrated the actin fluores-cence within each CCP-derivedmask If the actinsignal integrated over the area of a given maskincreased during the final five frames of the lifeof the associated CCP it was decided that actinwas recruited to the CCP during the final stage ofendocytosis

REFERENCES AND NOTES

1 L Schermelleh R Heintzmann H Leonhardt A guide to super-resolution fluorescence microscopy J Cell Biol 190 165ndash175(2010) doi 101083jcb201002018 pmid 20643879

2 U Schnell F Dijk K A Sjollema B N GiepmansImmunolabeling artifacts and the need for live-cell imagingNat Methods 9 152ndash158 (2012) doi 101038nmeth1855pmid 22290187

3 R P Nieuwenhuizen et al Measuring image resolution inoptical nanoscopy Nat Methods 10 557ndash562 (2013)doi 101038nmeth2448 pmid 23624665

4 X Shu et al A genetically encoded tag for correlated light andelectron microscopy of intact cells tissues and organismsPLOS Biol 9 e1001041 (2011) doi 101371journalpbio1001041 pmid 21483721

5 J D Martell et al Engineered ascorbate peroxidase as agenetically encoded reporter for electron microscopy NatBiotechnol 30 1143ndash1148 (2012) doi 101038nbt2375pmid 23086203

6 H Shroff C G Galbraith J A Galbraith E Betzig Live-cellphotoactivated localization microscopy of nanoscale adhesiondynamics Nat Methods 5 417ndash423 (2008) doi 101038nmeth1202 pmid 18408726

7 S H Shim et al Super-resolution fluorescence imaging oforganelles in live cells with photoswitchable membrane probesProc Natl Acad Sci USA 109 13978ndash13983 (2012)doi 101073pnas1201882109 pmid 22891300

8 B Hein K I Willig S W Hell Stimulated emission depletion(STED) nanoscopy of a fluorescent protein-labeled organelleinside a living cell Proc Natl Acad Sci USA 10514271ndash14276 (2008) doi 101073pnas0807705105pmid 18796604

9 V Westphal et al Video-rate far-field optical nanoscopydissects synaptic vesicle movement Science 320 246ndash249(2008) doi 101126science1154228 pmid 18292304

10 T Grotjohann et al rsEGFP2 enables fast RESOLFT nanoscopyof living cells eLife 1 e00248 (2012) doi 107554eLife00248 pmid 23330067

11 A Chmyrov et al Nanoscopy with more than 100000lsquodoughnutsrsquo Nat Methods 10 737ndash740 (2013) doi 101038nmeth2556 pmid 23832150

12 Materials and methods are available as supplementarymaterials on Science Online

13 P Kner B B Chhun E R Griffis L Winoto M G GustafssonSuper-resolution video microscopy of live cells by structuredillumination Nat Methods 6 339ndash342 (2009) doi 101038nmeth1324 pmid 19404253

14 L Shao P Kner E H Rego M G Gustafsson Super-resolution 3D microscopy of live whole cells using structuredillumination Nat Methods 8 1044ndash1046 (2011) doi 101038nmeth1734 pmid 22002026

15 R Fiolka L Shao E H Rego M W DavidsonM G Gustafsson Time-lapse two-color 3D imaging of live cellswith doubled resolution using structured illumination ProcNatl Acad Sci USA 109 5311ndash5315 (2012) doi 101073pnas1119262109 pmid 22431626

16 J Riedl et al Lifeact A versatile marker to visualize F-actinNat Methods 5 605ndash607 (2008) doi 101038nmeth1220pmid 18536722

17 H T McMahon E Boucrot Molecular mechanism andphysiological functions of clathrin-mediated endocytosis NatRev Mol Cell Biol 12 517ndash533 (2011) doi 101038nrm3151pmid 21779028

18 M Ehrlich et al Endocytosis by random initiation andstabilization of clathrin-coated pits Cell 118 591ndash605 (2004)doi 101016jcell200408017 pmid 15339664

19 I Gaidarov F Santini R A Warren J H Keen Spatial controlof coated-pit dynamics in living cells Nat Cell Biol 1 1ndash7(1999) pmid 10559856

20 S Saffarian E Cocucci T Kirchhausen Distinct dynamics ofendocytic clathrin-coated pits and coated plaques PLOS Biol7 e1000191 (2009) doi 101371journalpbio1000191pmid 19809571

21 J Grove et al Flat clathrin lattices Stable features of theplasma membrane Mol Biol Cell 25 3581ndash3594 (2014)doi 101091mbcE14-06-1154 pmid 25165141

22 J Heuser Effects of cytoplasmic acidification on clathrin latticemorphology J Cell Biol 108 401ndash411 (1989) doi 101083jcb1082401 pmid 2563729

23 M Kaksonen C P Toret D G Drubin Harnessing actindynamics for clathrin-mediated endocytosis Nat Rev Mol CellBiol 7 404ndash414 (2006) doi 101038nrm1940pmid 16723976

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-9

RESEARCH | RESEARCH ARTICLE

imaging with uniform illumination (fig S24 Aand B) Resolution of ~50 nm has been demon-strated with both approaches although not onliving cells saturated excitation was used to im-age densely labeled fluorescent beads at the like-ly phototoxic peak intensity of 8 MWcm2 (28)whereas saturated depletion (SD) imaged singlefixed cells at 945 sframe (29)mdashfar too slow tofollow most cellular processesSD provides the basis of resolution enhance-

ment in STED and RESOLFT as well as SD NL-SIM The degree of enhancement depends on thedegree of saturation (figs S23 and S25 and movieS6 part 3) defined in multiples of the saturationfactor (SF) for which 1e of the irradiated mol-

ecules remain in the original activated or excitedstate However high SFs are very photon ineffi-cient only a fraction of the photobleaching-dictated number of switching cycles for anygiven molecule then contributes useful signal(figs S25 to S27) Furthermore high SFs requirehigh intensities andor long exposures (fig S28)neither of which is compatible with fast non-invasive live-cell imagingWe addressed these issues by using patterned

activation (PA) followed with patterned excita-tion and readout of the green photoswitchableFP Skylan-NS (fig S29 andmovie S6 part 1) (30)rather than SD to generate H = 2 harmonicsyielding 62-nm resolution and subsecond acqui-

sition times in TIRF for live cells (Fig 3) Thisapproach termed PA NL-SIM allowed us toachieve large amplitudes in both the first andsecond harmonics of the emission pattern (figS29 G and H) leading to SR images of highsignal-to-noise ratio (SNR) even at low activa-tion and excitation saturation factors SFact andSFexc obtained with low intensities and shortexposures (table S1) Furthermore by keepingSFact low only a small fraction of the totalmolecular population needed to be activatedfor every raw image and with H = 2 only N =(2H + 1)2 = 25 such raw images needed to beacquired to reconstruct each SIM image frame(fig S30) Consequently we could acquire subs-tantially more frames at substantially higherSNR (fig S26) in far less time (table S1) bymeans of PANL-SIM (Movie 6) than SDNL-SIM(movie S7)PA NL-SIM of Skylan-NS-Lifeact (Fig 3 A and

BMovie 6 andmovies S8 to S10) in living COS-7cells revealed considerably more detail than didTIRF-SIM (Fig 3B) in dense peripheral actin arcsand star-like junctions of single actin filamentsWe were also able to resolve individual Lifeactrings once again including rings too small to seeclearly with high-NATIRF-SIM (fig S21 C andDandmovie S8) Furthermore we could follow thedynamics of the Lifeact-decorated actin cytoskel-eton for 30 image frames acquired in 12 s each(Movie 6) This is 1250times faster and used 20timeslower intensity (100 Wcm2) than was requiredfor an image of phalloidin-labeled actin at theventral surface of a fixed BSC-1 cell obtained bymeans of dual-objective localization microscopy(31) yet the level of detail seen by the two meth-ods was comparable (fig S31) even though our62-nm resolutionwas threefold coarser than thatreported in the localization imageWe also used PANL-SIM to image keratin (fig

S6 and movie S11) and caveolin (Fig 3 C to FMovie 7 figs S32 and S33 and movie S12) inliving COS-7 cells each with Skylan-NS at aresolution of 59 nm In the latter case this wassufficient to resolve numerous caveolae movingby less than their radii during the acquisitiontime as rings which is consistent with their in-vaginated appearance bymeans of EM (32) Suchrings were not observed at caveolae in a HeLacell imagedwith RESOLFT (fig S34) (33) despitea similar reported resolution Rings of Skylan-NS-caveolin were somewhat more abundant thanCCPs (figs S17 and S18) and althoughmost werebelow100nminsize their distributionwasbroader(Fig 3D) than the 60 to 80 nm range observedwith EM However some of the larger rings (Fig3E) may represent multiple caveolae clusteredaround surface-docked vesicles (34) Caveolaealso tended to loosely cluster in long narrow rib-bons although we saw tighter aggregations ofrings (Fig 3F) as well similar to those we saw inclathrin plaques (fig S19)Our time-lapse imaging showed that most

caveolae moved only a fraction of their size dur-ing the acquisition time although more met thiscondition when slowed (35) by operating at 23degC(Movie 7 and fig S32) than when imaged at 37degC

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-3

Fig 2 Dynamics of clathrin-mediated endocytosis and the cortical actin cytoskeleton (A) CCPsresolved as rings (fig S14 and movie S4) and color-coded according to their age since initial formation atone time point from amovie of CCPdynamics in a BSC-1 cell at 37degC stably expressing EGFP-clathrin lightchain a (Movie 3) (B) Histogram of maximum diameter of each CCP over its lifetime (C) Plot of CCPoverall lifetime versus CCP maximum diameter (D) Sequential production of multiple CCPs at a CCP-generating ldquohot spotrdquo identified with green arrowheads (Movie 4 and fig S16) (E) Formation growth anddissolution of a single CCP (right) and its relationship to cortical f-actin (left) in a COS-7 cell at 37degCtransfected with mEmerald-clathrin light chain b and mCherry-Lifeact Light blue arrowheads mark timepoints at which f-actin associates with the CCP (F) Individual CCPs and clathrin plaques (green) andcortical f-actin (red) at one time point during their evolution in a COS-7 cell (Movie 5 figs S17 and S18andmovie S5) (G) Formation of a nanoscale ring of f-actin (fig S17B) Scale bars 1 mm (A) and (F) and200 nm (D) (E) and (G)

RESEARCH | RESEARCH ARTICLE

(fig S33 and movie S12) The smaller laterallymobile fraction in each case appeared as distorteddiscontinuous rings or quasiperiodic patches(fig S35) These morphologies are indicative ofmotion-induced artifacts and underscore thedifficulty of live-cell SR imaging by any methodHigher resolution must be accompanied by pro-portionally faster acquisition times to followdynamic events of a given velocity yet higherresolution also requires a quadratically increasingnumber of raw measurements for each two-dimensional (2D) image frame Even the com-paratively brief 035 swe needed to acquireN= 25raw images for each PA NL-SIM image wasinsufficient to accurately depict caveolae mov-ing by much more than our 59-nm resolution inthis timeNevertheless by further increasing SFact we

were able to saturate the fraction of molecules inthe activated state near the maxima of the pat-terned activation light (movie S6 part 2) Saturated

PA NL-SIM generates an additional harmonic(H = 3) strong enough (fig S29) to further extendthe resolution to 45 nm (figs S36 and S37) andallowed us to identify even smaller Skylan-NS-caveolin rings unresolvable without the extraharmonic (Fig 3C) UsingN = 35 rather thanN =49 raw images per frame we balanced the re-sulting anisotropic resolution (fig S30) againstthe needs for rapid acquisition (049 sframe) andparsimonious use of the photon budget to im-age caveolin rings over 12 frames at 3 s intervals(movie S13)

Two-color live imaging via combinedTIRF-SIM and PA NL-SIM

By combining linear SIM and PA NL-SIM bothin TIRF we could study associations betweenfluorescent proteins one conventional and onephotoswtichable in two colors at higher resolu-tion than by means of linear TIRF-SIM aloneImages (Fig 4 A to C and figs S38 and S39) and

movies (Movie 8 and movie S14) of mCherry-Rab5amdasha regulator of the formation fusion andtransport of early endosomes (EEs) (36)mdashrevealedirregularly shaped dynamically remodelingpatchesof Rab5a (fig S39 A and B) consistent with thetubularvesicular architecture of EEs seen in EM(Fig 4B) (37) Numerous patches also featureddark spots (fig S39C) perhaps indicative of car-go or internal vacuoles depleted of Rab5a Mostpatchesmoved randomly between successive 20-stime intervals at velocities slow enough to avoidmotion artifacts during each 034-s acquisitionWe also observed a subpopulation of slowly grow-ing Skylan-NS-Lifeactndashassociated Rab5a patchesthat were constrained for minutes at a time (figS39D arrows) At the other extreme we occa-sionally observed streaks of Rab5amoving paral-lel to nearby actin filaments at velocities of 3 to5 mms (Fig 4C and fig S39E) These may repre-sent EEs actively transported alongmicrotubules(38) parallel to the filaments

aab3500-4 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 3 Live-cell non-linear structured illumi-nation microscopybased on patternedphotoactivation (A)Single time point from amovie of the evolution ofcortical f-actin in a COS-7cell at 23degC transfectedwith Skylan-NS-Lifeactseen at 62-nm resolution(Movie 6 fig S31 andmovie S8) (B) Magnifiedview from a different cellat 37degC comparingdiffraction-limited TIRFmicroscopy (top left)TIRF with deconvolution(top right) TIRF-SIM(bottom left) and non-linear TIRF-SIM withpatterned activation(PA NL-SIM bottomright) (movies S9 andS10) (C) Caveolae in aCOS-7 cell at 23degC trans-fected with Skylan-NS-caveolin comparing TIRFwith deconvolution (topleft 220-nm resolution)TIRF SIM (top right97-nm resolution) PANL-SIM (bottom left62-nm resolution) andsaturated PA NL-SIM(bottom right 45-nmresolution) (Insets) Asingle caveolae pit even-tually resolved as a ringby saturated PA NL-SIM(Movie 7 figs S34 toS37 and movie S13(D) Diversity of caveolae ring diameters as seen by means of PA NL-SIM (E) Larger rings that may represent surface-docked vesicles (F) Clusters of caveolaereminiscent of clathrin plaques (D) to (F) are from a different cell at 37degC (fig S33 and movie S12) Scale bars 3 mm (A) 1 mm (B) 200 nm (C) and 100 nm(D) (E) (F) and (C) inset

RESEARCH | RESEARCH ARTICLE

We also used PA NL-SIM and TIRF-SIM re-spectively to study the association of Skylan-NS-Lifeact with mCherry-a-actinin (Fig 4 D to Fand fig S40) Consistent with its role as an actin-bundling protein (39) in COS-7 cells we founda-actinin at the treadmilling edge of the lamelle-podium and at the basal surface in both filopodiaand the leading edges of growing membraneruffles (Fig 4F Movie 9 and movie S15) We alsoobserved concentrations of a-actinin along thesides (Fig 4E) and at the branching ends of stressfibers that likely attach to cell-substrate adhesions(40) Last a-actininwas present at dense junctionsof Lifeact-decorated filaments and Skylan-NS-Lifeact rings as described above were colocalizedin every instance with a mCherryndasha-actinin ringof similar size (fig S41) Septins another class ofactin-bundling proteins have been shown (41) toproduce f-actin rings in vitro (albeit of larger sizethan here) so perhaps a-actinin not only aids inbundling actin filaments in nanometric rings butalso contributes to their extreme curvature

3D live-cell imaging with combined PANL-SIM and lattice light sheet microscopy

Although the ~50- to 200-nm extent of the eva-nescent excitation field we used in the examplesabove eliminated out-of-focus background andconfined potentially phototoxic exposure to aminute fraction of the cellular volume it alsolimited our observations to this subvolume andseverely restricted the total photon budget avail-able for those targets unable to be replenishedfrom the cytosol during the imaging intervalTo extend our observations to the entire cell

we turned to live-cell 3D-SIM (14 15) Unfortu-nately traditional 3D-SIM with linear widefieldexcitation brings limitations of its own It is slow

(~20 s acquisition for whole adherent HeLa cells)limited to thin specimens (because of out-of-focusbackground) and requires high SNR for accurateimage reconstruction It is also potentially photo-toxic and bleaches specimens rapidly because ofcontinuous whole-cell illumination These prob-lems would all be greatly magnified in its directextension to PA NL-SIMThus to apply PA NL-SIM to living cells in

three dimensions (Fig 5) we used lattice lightsheet microscopy (42) In this technique an exci-tation objective (fig S42A) projects a thin sheetof light (fig S42A blue) through a specimen (figS42A orange) and the fluorescence generated inthe illuminated plane is collected by a detectionobjective and imaged onto a camera Repeatingthis process plane-by-plane through the specimenproduces a 3D image Restriction of the light tothe detection focal plane eliminates out-of-focusbackground increases the z axis resolution andgreatly reduces photobleaching and phototoxicityIn cross-section the light sheet has the 2D

periodic structure of an optical lattice (fig S42B)Sweeping the sheet back and forth along the xaxis produces time-averaged uniform illumina-tion offering high speed and diffraction-limitedxyz resolution of 230 by 230 by 370 nm as seenin a volume-rendered image of the actin cyto-skeleton (fig S43A) and its corresponding overalloptical transfer function (OTF) (fig S43D) Step-ping the sheet in x in five equal fractions of thelattice period and applying the algorithms of3D-SIM to the resulting five raw images perplane extends the xyz resolution to 150 by 230by 280 nm (fig S43 B and E) but at the cost of atleast 5times longer acquisition times (42)To further extend the 3D resolution via PA

NL-SIMwe first photoactivated targetmolecules

fused to Skylan-NS using a hexagonal lattice lightsheet of l = 405 nm wavelength having H = 2harmonics (fig S43E) We then imaged the fluo-rescence from the activated region exciting thefluorescencewith a lattice light sheet of l =488nmwavelength having the same hexagonal sym-metry and period (fig S42B bottom) as the ac-tivation lattice For activationwell below saturationthe product of the activation and excitation pat-terns creates a fluorescence emission patternwithin the specimen having H = 4 harmonics(fig S43F) Thus we stepped the sheet in x in2H + 1 = 9 equal fractions of the lattice periodwhile recording nine images Repeating this pro-cess for every plane within the specimen we thenreconstructed a 3D PA NL-SIM volume-renderedimage (fig S43C) with resolution extended to118 by 230 by 170 nmWe used this approach to image mitochondria

in COS-7 cells (Fig 5A) as well as the actin cyto-skeleton (fig S43 A to C and movie S16) andthe Golgi apparatus (Fig 5B) inU2OS cells all at23degC so as to simplify the overlap of the activa-tion and excitation patterns Time-lapse 3D im-ages (Fig 5A bottom) and movies (Movie 10) ofSkylan-NSndashtagged translocase of outer mitochon-drial membrane 20 (TOM20) revealed the mi-gration constriction before fission and fusionof individual mitochondria (43 44) each clearlyresolved as a hollow tubular structure The 3Dvolume rendering and the widths of mitochon-drial membranes in individual xy orthosliceswere both comparable with similar data from afixed cell imaged with 3D localization micros-copy (45) at a reported xyz resolution of ~20 by20 by 60 nm (fig S44)A volume-rendered movie (movie S17) of the

Golgi-resident enzyme Mannosidase II (MannII)

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-5

Fig 4 Combined TIRF-SIMand PA NL-SIM of protein-pair dynamics in livingcells (A) Skylan-NS-Lifeact(orange PA NL-SIM) andmCherry-Rab5a a marker ofearly endosomes (greenTIRF-SIM) in a COS-7 cellat 23degC (Movie 8 figs S38and S39 and movie S14)(B) Comparison of EMimages of early endosomes(37) with similarly shapedRab5a patches seen in (A)(C) Magnified view at threesuccessive time pointsshowing rapid transportof a Rab5a streak parallel tothe cytoskeleton (D) Skylan-NS-Lifeact (green PA NL-SIM) and mCherry-a-actinin(purpleTIRF-SIM) in a COS-7cell at 23degC (Movie 9 figsS40 and S41 and movieS15) (E) Magnified viewfrom (D) with Lifeact (top)a-actinin (middle) and overlay (bottom) showing paired association at focal adhesions and along the sides of large stress fibers (F) Evolution of amembrane ruffle showing a-actinin concentrated at the leading edge Scale bars 5 mm (A) (D) 200 nm (B) 1 mm (C) and (E) and 500 nm (F)

RESEARCH | RESEARCH ARTICLE

tagged with Skylan-NS in a U2OS cell as seenlooking into the cis-face from the nucleus showedMann II concentrated in a hollow sphere ofcisternae having a cis-facing void Time-lapse3D data (Fig 5B andmovie S18) color-coded forheight showed the docking of small vesicles (Fig5B white arrows) that may represent pre-Golgiintermediates (46) as well as the rapid export ofMann II in long tubular post-Golgi carriers (Fig5B red arrows) (47)The volumetric resolution of 3D lattice light

sheet PA NL-SIM at the 06-NA excitation and11-NA detection we used here is comparablewith the 105- by 105- by 369-nm xyz resolutionof widefield 3D-SIM at 12 NA However thelattice approach has twofold higher axial resolu-tion and fourfold better than traditional diffraction-limited microscopy It is therefore better suitedto problems in which its superior optical section-ing is essential such as in resolving heterogene-ities in nuclear architecture distinguishing eventsoccurring at the dorsal or ventral plasma mem-brane or as above tracking vesicles through thesecretory pathway Whole-cell acquisition times(705 and 327 s in Fig 5 A and B respectively)are slow compared with PA NL-SIM in TIRF butsimilar to widefield 3D-SIM However thanks tothe oblique imaging geometry (fig S42) restrictedxy fields of view can be imaged at proportion-ally faster speed through the entire thickness ofthe cell

Discussion

The above results provide but a brief glimpse ofthe biology that might be uncovered with thelive-cellndashcompatible SRmethods of high-NATIRF-SIM and PA NL-SIM We have measured andcorrelated the diameters and lifetimes of CCPsobserved at high resolution different forms ofCCP initiation and shown that CCP internaliza-

tion is aided by actin filaments in about half of allcases We have seen that caveolin localizes notonly to the 60- to 80-nm invaginated caveloaecommon in EM images but also to much largerring-like structures and have followed dynamicchanges in the shapes of early endosomes Lastwe have observed the nanoscale remodeling ofthe actin cytoskeleton in relation to clathrin andRab5a as well as cytoskeletal-related proteinssuch as myosin IIA a-actinin and paxillinHowever the above results also amply illus-

trate the trade-offs inherent in live SR imagingWith high-NA TIRF-SIM at 17 NA we could ac-quire up to 200 image frames in lt05 s each atintensities of 20 to 100 Wcm2 and a resolutionof 84 nm (for GFP) whereas extending the reso-lution to 62 nm with PA NL-SIM restricted us tono more than 40 frames and further extensionto 45 nm with saturated PA NL-SIM required490 Wcm2 and produced only 12 frames atuseful SNRIn short evenmodest gains in resolution come

at substantial cost in terms of the other metricsimportant for live-cell imaging These tradeoffsare not specific to SIM In fact our extensions ofSIM are far more compatible with live imagingthan any other form of SR fluorescence micros-copy of comparable resolution demonstrated todate In part this is because the OTF which de-fines the degree towhich different sample spatialfrequencies (representing differently sized struc-tures) are passed to the image is far stronger inthe 100-nm regime (fig S24B) for high-NA TIRF-SIM at 17 NA than other linear methods such asconfocal or image scanning microscopy (ISM)(48ndash50) and far stronger in the 50- to 100-nmregime (fig S24C) for PA NL-SIM than othernonlinear methods such as STED (8 9) point-scanning (PS) RESOLFT (10) or array-basedwide-field (WF) RESOLFT (11) As a result far fewer

photons need to be collected (fig S2) and far lesslight (fig S3) needs to be applied to the specimento see features in these regimes at acceptableSNR Localization microscopy is also photon in-efficient in that the density of localizedmoleculesis nearly always more limiting to the resolutionthan is the number of photons emittedper switch-ing cycle which dictates the localization preci-sion For example simulations (12) based on thetheoretical OTFs suggest that to resolve an 85-nmgrating PANL-SIM requires ~80times fewer photonsfrom the specimen per unit area than localiza-tionmicroscopy ~200times fewer thanWF-RESOLFTand ~15times fewer than PS-RESOLFT or STED eachat a depletion saturation factor of SFdepletion = 10(fig S2)Another reason for the greater compatibility of

high-NA TIRF-SIM and PA NL-SIM with livingcells is that they require much lower peak inten-sities of applied light High resolutionwith STEDor RESOLFT demands high factors of saturateddepletion (fig S25 A and C) that are wasteful ofthe photon budget (fig S25 B andD) and requireenormous intensities andor long exposures foractivation (fig S45) depletion (fig S28) and read-out of the final signal (fig S3) Localization mi-croscopy also requires high intensities to achievehigh photon emission and photoswitching ratesfrom single molecules For example extrapolat-ing from reported experimental values for live-cell imaging (table S1) the 08- to 35-Wcm2

activation intensity used over the 45- by 45-mmfield of view in Fig 3A in 12 s bymeans of PANL-SIM is 960000 times weaker than that whichwould be required to image the same area in thesame acquisition time by means of PS-RESOLFT(10) Similarly under the same parameters the100-Wcm2 read-out intensity used for PA NL-SIM shown in Fig 3A is 200 times weaker thanthat which would be required for localization

aab3500-6 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 5 Live-cell 3D PA NL-SIMvia lattice light sheet micros-copy (A) (Top) Membranemarker Skylan-NS-TOM20showing mitochondria in aCOS-7 cell at 23degC color-codedfor distance from the substrate(Bottom) Evolution of individualmitochondria showing fissionand fusion events the formerpreceded by mitochondrial con-striction (Movie 10 and fig S44)(B) Time-lapse distribution ofGolgi-resident enzyme Skylan-NS-Mann II in a U2OS cell at23degC showing centralizedcisternae surrounded byvesicles White arrowheads indi-cate a docking vesicle and redarrowheads highlight rapidexport of a long tubular vesicle(movies S17 and S18) Scalebars 5 mm (A) top 1 mm (A)bottom and 3 mm (B)

RESEARCH | RESEARCH ARTICLE

microscopy (6 7) and 640000 times less than PS-RESOLFT (10) Furthermore STED andRESOLFTrequire an additional depletion step not neededin PA NL-SIM which would further expose thesample to peak intensities of 807 MWcm2 forSTED (8) 17 MWcm2 for PS-RESOLFT (10) and3 kWcm2 forWF-RESOLFT (11) Even over smallimage fields nanoscopy with focused light suchas PS-RESOLFT and STED uses intensities 105-to 1010-fold larger than that of terrestrial solarflux and is thus ill-equipped to study live-celldynamics noninvasivelyOf course despite these gains no method of

live-cell fluorescencemicroscopy including high-NA TIRF-SIM and PA NL-SIM can claim to becompletely noninvasive owing to possible photo-induced physiological changes protein over-expression andor label-induced perturbationsFor example the gradual development of curvedfilopodia and membrane ruffles after the start ofimaging are shown in Movies 5 and 6 and movieS2 These may reflect a response to the illumina-tion although we have also commonly seen suchstructures under initial conditions when imagingwith diffraction-limited TIRF (fig S46) Anothercaveat is that all the cells except BSC-1 in thiswork were transiently transfected and henceexpression levels of the target proteins were un-controlled This could affect eithermorphologiessuch as the sizes of Rab5a-labeled endosomes(Fig 4 A to C and figs S38 and S39) (51) ordynamic phenotypes such as the growth rate ofmembrane ruffles inmCherryndasha-actininndashexpressingcells (Fig 4E Movie 9 and movie S15) Althoughendogenous expression levels can be achievedwith genome editing (52) even more light orlonger exposures would be needed for cases inwhich these levels are lower than those used hereThus the biological findings described in this workshould not be considered definitive More exten-sive measurements across multiple cell lines withcareful controls and targeted perturbation experi-ments will be needed to reach conclusive insightsThe lesson is that when addressing any biolog-

ical question by means of live-cell imaging it isprudent to startwith less invasive lower-resolutionmethods such aswidefield spinning disk confocalor lattice light sheetmicroscopy andmove progres-sively only as needed to more invasive higher-resolution methods such as 3D-SIM TIRF-SIMPANL-SIM and last localizationmicroscopy Seenfrom this perspective the two extended-resolutionmethods of high-NATIRF-SIMandPANL-SIMweintroduce here fill an important gap between the100-nm limit of traditional SIM and the macro-molecular level of localizationmicroscopy Togetherthey open the door to high-resolution minimallyinvasive studies of dynamic processes includingendocytosis exocytosis signal transduction proteindiffusion vesicle trafficking viral entry cytoskeletalremodeling interactions with the extracellularmatrix and the evolution of lipid rafts

Materials and methodsOptical path of the TIRF-SIM system

The schematic of TIRF-SIM system is presentedin fig S47A The beam from a laser combiner

equipped with 405 nm (250 mW RPMC OxxiusLBX-405-300-CIR-PP) 488 nm (500mW Coher-ent SAPPHIRE 488-500) and 560 nm (1W MPBCommunications 2RU-VFL-P-1000-560-B1R) lasersis passed through an acousto-optic tunable filter(AOTF AA Quanta Tech AOTFnC-400650-TN)The beam is then expanded to a 1e2 diameter of12 mm and sent to a phase-only modulator (13)consisting of a polarizing beam splitter a achro-matic half-wave plate (HWP Bolder Vision OptikBVO AHWP3) and a ferroelectric spatial lightmodulator (SLM ForthDimensionDisplays SXGA-3DM) Light diffracted by the grating patterndisplayed on SLM passes through a polarizationrotator (15) consisting of a liquid crystal cell (LCMeadowlark SWIFT) and an achromatic quarter-wave plate (QWP Bolder Vision Optik BVOAQWP3) which rotates the linear polarizationof the diffracted light so as to maintain thes-polarization necessary to maximize the patterncontrast for all pattern orientations A mask con-sisting of a hollow barrel with slots for differentpattern orientations (15) is driven by a galvano-metric scanner (Cambridge Technology 6230HB)to filter out all diffraction orders created by thebinary and pixelated nature of the SLM exceptfor the desired plusmn1 diffraction orders These arethen imaged at the back focal plane of the ob-jective (Olympus APON 100XHOTIRF 17 NA forhigh-NATIRF-SIMOlympusUAPON100XOTIRF149 NA for PA NL-SIM at 23degC or Zeiss Plan-Apochromat 100X Oil-HI 157 NA for high-NAPA-NL-SIM at 37degC) as two spots at oppositesides of the pupil After passage through the ob-jective the two beams intersect at the interfacebetween the coverslip and the sample at an angleexceeding the critical angle for total internal re-flection An evanescent standing wave penetrat-ing ~100 nm into the sample is thereby generatedconsisting of a sinusoidal pattern of excitationintensity that is a low-pass filtered image of theSLM pattern The period orientation and rela-tive phase of this excitation pattern can befinely tuned by altering the corresponding pat-tern displayed on SLM For each orientationand phase of the applied excitation pattern theresulting fluorescence is collected by the ob-jective focused by a tube lens at an interme-diate image plane separated from excitationlight by a dichroic mirror (Chroma ZT405488560tpc_225deg) placed between two relaylenses and reimaged onto a sCMOS camera(Hamamatsu Orca Flash 40 v2 sCMOS) wherethe structured fluorescence emission pattern isrecorded

Calibration of pattern overlap forPA NL-SIM

In order to maximize the amplitudes of the non-linear harmonics for PA NL-SIM to work efficient-ly the sinusoidal patterns of 405 nm activationlight and 488 nmexcitation and deactivation lightmust be aligned to precisely overlap one anotherAs noted above these patterns at the sampleplane are created by displaying correspondingbinary grating patterns on an SLM at a corre-sponding optically conjugate plane In this case

the period ps at the specimen is related to theperiod pSLM at the SLM by

ps =Ml middot pSLM eth1THORN

where M is the demagnification factor betweenthe two conjugate planes and is dictated to bethe focal lengths of the relay lenses between thetwo planes Unfortunately chromatic aberrationleads to slightly different focal lengths for evenachromatic relay lenses for different wavelengthsof light In particular in our system M405 andM488 vary by ~2 Considering that the sinusoi-dal interference pattern is composed of hundredsof periods across our 45- by 45-mm2 field-of-view(FOV) even this 2 difference results in sub-stantial drift in the relative phases of the 405-and 488-nmexcitationpatterns across the FOV (figS48 A to C) leading to spatially variable ampli-tudes for thenonlinearharmonics and correspond-ing spatially variable errors in the resultingSIM reconstructionsA straightforward way to compensate for

chromatic aberration and achieve identical peri-ods ps405 = ps488 at the sample (fig S47B) is tointroduce a period difference DpSLM between thetwo corresponding patterns at the SLM (figS47C) In fact in order to compensate completelyand achieve well-overlapped 405- and 488-nmexcitation patterns over the whole FOV we needto measure two parameters the initial perioddifference at the sampleDpi

s frac14 Dpis488 minus Dpis405

when pSLM is the same for bothwavelengths andthe phase differenceDfis frac14 Dfis488 minus Dfis405 whenps is the same Do to so we used a sample con-sisting of a dense but submonolayer spread ofgreen fluorescent beads excitable at both 405and 488 nm and proceeded as follows

Step 1

Keeping pSLM constant we acquired five imageseach of the sample under 405- and 488-nm sinus-oidal excitation with the phase shifted by pSLM5for each image at a given wavelength We then ap-plied the structured illumination (SI) reconstruc-tion algorithm (53) to each set of five images fromwhich pis405 and pi

s488 emerged as measuredoutputs For a given period pSLM488 used at theSLM for 488-nm excitation the correspondingperiod pSLM405 needed at the SLM for 405-nmexcitation to produce the same period ps at thesample for both wavelengths is then given by

pSLM405 frac14pis488pis405

pSLM488 eth2THORN

Step 2

After adjusting pSLM405 and pSLM488 to obtainthe same period ps at the sample for both wave-lengths a constant phase offset exists betweenthe two sinusoidal illumination patterns acrossthe FOV (fig S48 D and E) We measured thephase f for each wavelength by applying thesinusoidal illumination for that wavelength andthen recorded the position xn along the modu-lation direction and intensity In for each of Nbeads scattered across the FOV We then fit the

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-7

RESEARCH | RESEARCH ARTICLE

function I(x) = Imax[1 + sin(2pxps + f)]2 to thisdata to find f (fig S48F) A phase shift Df = f488 ndashf405 was then applied the SLM pattern for the405-nm illumination so as to bring it into phasewith the 488-nm illumination at the specimen(figs S48 G to I)

Step 3

Last we confirmed that both the period and phaseof the sinusoidal illumination patterns at the twowavelengths match across the entire FOV byremeasuring the periods ps488 ps405 and thephases f488 f405 as described above and con-firming that they are identical

SLM pattern generation

We generated the sinusoidal illumination pat-terns using a binary ferroelectric SLM (Forth Di-mension Displays SXGA-3DM) because it hasthe submillisecond switching times needed toacquire the nine (TIRF-SIM) 25 (PA NL-SIM) ormore (saturated PA NL-SIM) raw images of dif-ferent phase and orientation required to recon-struct a single SIM image in as fast as 100 to400msHowever care must be taken to account for thefinite pixel size of the SLM especially consideringthat subpixel adjustment accuracy is necessary toachieve precise pattern overlap at 405 and488nmas described in the previous section The SLMpattern-generation algorithms used in previouswork (13ndash15) do not provide such subpixel accu-racy Thus in this work we developed a newalgorithm that matches the two pattern periodsto 002 precision leading to a phase error nogreater than 18deg over the 45-mm FOVIn detail a set of radial vectors An define the

desired orientations of the grating pattern at theSLM The angular orientation of this radial setrelative to the x and y axes defined by pixel rowsand columns of the SLM is chosen so that eachvector is at least 4deg away from either axis This isessential to achieve subpixel precision in the ad-justment of the period For each orientation rep-resented by An we define a vector Bn that isorthogonal to An (fig S49) Likewise for everypixel of the SLM we define a pixel vector (suchas C1 or C2 in fig S49) from the point O at theintersection of An and Bn to the pixel We thencalculate F = [(C middotB)modp]p the fraction of theperiod p by which the pixel extends beyond anintegral number of periods on the SLM For apattern with a desired off fraction D per period(D = 05 in 2D SIM) the pixel is set to 0 if F lt Dand set to 1 otherwise Last to define the pixelpatterns required for the other N ndash 1 phases ofthe illumination for a given orientation the pointO is translated along Bn in steps of pN and thisprocess is repeatedwith the new vectorC for eachpixel Unlike the pixel assignment algorithm usedpreviously for SIM (15) this approach does notrely on unit-cell repetition and therefore doesnot succumb to error accumulation over theentire span of the SLM

Lattice light sheet PA NL-SIM system

To extend PA NL-SIM to three dimensions it isessential to minimize out-of-focus fluorescence

emission that can cause the shot noise in the DCharmonic to completely overwhelm the weaksignals in the nonlinear harmonics To accom-plish this we turned to the SIM mode of latticelight sheet microscopy (42) Just as in the case of2D-SIM and for the same reasons we chooseto introduce the nonlinear harmonics throughpatterned activation of Skylan-NS The excitationobjective (Special Optics 065 NA 374 mmWD)is placed perpendicular to the detection objective(Nikon CFI Apo LWD 25XW 11 NA 2 mmWD)to confine the illumination to the proximity ofthe latterrsquos focal plane (fig S42A) The latticepattern projected on the SLM (Forth DimensionDisplays SXGA-3DM) is imaged onto the focalplane of the excitation objective after the excita-tion is first spatially filtered by an annular mask(Photo-Science) and relayed by a pair of galva-nometers (Cambridge Technology 6215H) thatphase step the pattern in the x direction and scanthe light sheet in z Also as in 2D PA NL-SIM wematch the periods and phases of the 405- and488-nm lattices to exactly match by measuringtheir excitation profiles across the FOV using fluo-rescent beads (fig S42B) and adjusting accord-ingly The fluorescence emission is collected bythe detection objective and imaged by a tube lensonto a sCMOS camera (Hamamatsu Orca Flash40 v2) A 3D image is formed by repeating thisprocess as the sample is translated through thelight sheet with a piezoelectric stage (PhysikInstrumente P-6211CD) along an axis s in theplane of the cover slip and a 3D super-resolutionNL-SIM image is reconstructed as describedbelow

Data acquisitionHigh-NA TIRF SIM

All high-NA TIRF-SIM images were acquiredwith the Olympus 17-NA objective under thephysiological conditions of 37degC and 5 CO2 Ateach time point we acquired three raw images atsuccessive phase steps of 0 13 and 23 of theillumination period We then repeated this pro-cess with the standing wave excitation patternrotated plusmn120deg with respect to the first orienta-tion for a total of nine raw images The phasestepping and pattern rotation were accomplishedby rotating or translating the binary grating pat-terndisplayedon theSLMFormulticolor imagingwe acquired nine raw images at each excitationwavelength before moving to the next and thenrepeated this series at successive time points Wecould adjust the excitationNA for eachwavelengthby changing the period of the grating pattern at theSLM This allowed us to control penetration depthof the evanescent wave (fig S8) in order to ba-lance the number of excitable fluorescent mole-cules against the background fluorescence andpossible physiological effects of the excitation

PA NL-SIM and saturated PA NL-SIM

The high refractive index immersion oil requiredfor the Olympus 17-NA objective strongly ab-sorbs 405-nm light leading to a substantial reduc-tion in the modulation depth we could achieve inthe activation pattern at this wavelength Conse-

quently forNL-SIMwe first turned to theOlympus149-NA TIRF objective and imaged at room tem-perature (23degC) with L15 medium without phenolred having 10 fetal bovine serum (Life Technol-ogies) With this objective we were able to achievehigh modulation contrast while stably and pre-cisely overlapping the 405- and 488-nm standingwaves over the whole FOV An excitation NA of144 was used for both 488- and 560-nm light inthis case leading to 62-nm resolution for PANL-SIMwhen using green-emitting FPs Recently how-ever we found that the high refractive index im-mersion oil used for the Zeiss 157-NA objectivedid not absorb 405-nm light strongly and there-fore could be used to maintain precisely over-lapped 405- and 488-nm standing waves withhigh modulation contrast at 37degC and 5 CO2The excitation NA in this case was 152 for 488-nmlight leading to 59-nm resolution for PA NL-SIMwhen using green-emitting FPsThe exposure procedure for a single phase step

inNL-SIMconsists of (i) 405-nmpatterned illumi-nation for 1 ms to activate the fluorescent mol-ecules (ii) 488-nm patterned illumination for 5 to~30 ms to read-out the activated molecules and(iii) 488-nm uniform illumination for 2 to ~10 msto read-out the remaining activated molecules andreturn the sample back to the original unactivatedstate We collected the fluorescence from bothsteps (ii) and (iii) to reconstruct the SR imageDepending on the number of modulation har-monics H of non-negligible amplitude in theimage (H = 2 for PA-NL-SIM andH = 3 or possiblymore for saturated PA NL-SIM) we repeated thissequence for 2H + 1 raw images at each of 2H +1 angular orientations equally spaced around 360degfor a total of (2H + 1)2 raw images at each NL-SIMtime point An exceptionwas saturated PA-NL-SIMfor which to reduce the acquisition time weoften used only five orientations rather thansevenIn two-color imaging combining linear TIRF-SIM

and PA NL-SIM (Fig 4) at each time point weacquired the PANL-SIM image as discussed aboveHowever we acquired the TIRF-SIM image withfive instead of three orientations (15 raw images forthe TIRF-SIM channel at every time point) inorder to match the orientations of the five-slotgalvanometer-driven barrel mask used to pickout thedesireddiffractionorders for thePANL-SIMacquisition

3D PA NL-SIM with lattice lightsheet microscopy

Here we used a hexagonal lattice having aperiod large enough to contain two harmonicsfor each of the 405-nm activation and the 488-nm excitation (42)mdashone harmonic just belowthe Abbe limit of the 065-NA excitation objec-tive and the other at twice this period Theproduct of these patterns created a fluorescenceemission pattern containing H = 4 harmonics(fig S43F) However with a single excitation ob-jective we were limited to producing this pat-tern at only one orientation Therefore at eachplane of the 3D stack we acquired 2H + 1 = 9images resulting in improved resolution (Fig 5)

aab3500-8 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

RESEARCH | RESEARCH ARTICLE

in both the lateral and axial directions of thepattern

Reconstruction of SIM images

The raw image frames with patterned excitationwere processed and reconstructed into the super-resolved images by means of a previously de-scribed algorithm (53) In brief for each patternorientation with H modulation harmonics 2H +1 raw images are collected and Fourier transformedinto 2H + 1 information components These com-ponents are assembled by initially translating eachin Fourier space by a distance equal to the am-plitudeof the illuminationpatternvectornk0wherek0 is the spatial frequency of the illumination pat-tern and n = ndashH to H The pattern vector of eachinformation component is then fine-tuned byfinding the vector that maximizes the complexcross-correlation in the overlap region betweensuccessive components The modulation ampli-tude of the harmonic and its starting phase arefound through complex linear regression In linewith previous work (28) the modulation ampli-tudes for the highest harmonics are generally toolow for this empirical approach to work well sofor these the theoretical values of their complexamplitudes are used After fine-tuning the posi-tions and complex amplitudes of the informationcomponents in the overlap regions a generalizedWiener filter is applied to this expanded transferfunction to balance the amplitudes of the variousspatial frequencies against the underlying noiseNext an apodization function is applied to min-imize ringing artifacts when the result is Fourier-transformed back to real space However ratherthan the triangle apodization A(k) = 1 ndash kkmax

normally used (53) we applied a g apodizationA(k) = 1 ndash (kkmax)

g usually with g = 04 so thatthe higher spatial frequencies are not suppressedmore than necessary Furthermore we strictly fol-lowed the azimuthally dependent support kmax(q)of the expanded OTF (figs S7 and S30) to definethe endpoint of the apodization function This pro-vides additional suppression of ringing artifactsFor the time series data we independently imple-ment this reconstructionprocess for each timepoint

Cell culture transfection stainingand fixation

BSC-1 COS-7 U2OS andmouse embryonic fibro-blast (MEF) cells (American Type Culture Collec-tion) were grown to ~60 to 80 confluency inDulbeccorsquos modified eagle medium (DMEM) withhigh glucose and no phenol red supplementedwith 15 fetal bovine serum (Life Technologies)BSC-1 cells stably expressed EGFP-CLTA Othercells were transiently transfected with an AmaxaNucleofector 96-well shuttle system (Lonza) with1 mg DNA per 400000 cells with nucleofectionsolution and a program optimized for each cellline per the manufactures instructions Beforeimaging 25-mm or 5-mm coverslips were coatedwith 10 mgml fibronection (Millipore FC010) for24 hours before plating transfected cells Imagingwas performed in DMEM with HEPES if there isno CO2 control containing no phenol red at tem-peratures specifically stated in each case

In two-color imaging of CCPs and transferrinreceptors (TfRs) by means of high-NA TIRF-SIMMEF cells expressing clathrin light chain B fusedto the C terminal of mEmerald were incubatedwith DMEM medium containing 250 mgmLTfR bound to human transferrin conjugatedwith Alexa 568 (T23365 Life Technologies) for15 minFixed cells were treated for 15 min with fixa-

tion buffer containing 4 paraformaldehyde01 gluteraldehyde in PHEM buffer (25 mMHEPES 10mMEGTA 2mMMgCl2 and 120mMPIPES in pH 73)

Tracking analysis of CCPs

For each image frame we segmented the CCPsusing a watershed algorithm written in Matlab(MathWorks 2014a) and measured their cent-roids individually Subsequently the centroidpositionwas linked between time points using u-track 21 (54) This linking operation collectedsuccessive position information for each pit overthe entire endocytic process (Fig 2E) from ini-tiation to final internalization It was then straight-forward to determine the lifetime (Fig 2A) foreach endocytic eventIn order to precisely measure the pit diameter

(Fig 2 B and C) we first measured the systemmagnification to the camera by imaging a stan-dard fine counting grid (2280-32 Ted Pella) TheSIM image of each CCP was then deconvolvedwith the equivalent PSF of the SIM system tocompensate for the broadening due to the finiteresolution of the instrument Last we measuredthe diameter of each deconvolved pit using anintensity-weighted average radius relative to thecentroid of the pit In certain cases (Fig 2A andMovie 3) pits were color-coded at each timepoint based on the time since their initiation tothe current time pointOne challenge in this analysis was how to

identify isolated pits rather than aggregates andhow to be sure that these represented true pitsrather than noise or disorganized patches ofnonassembled clathrin To accomplish this weset some conditions during the analysis such asthat a pit must start as a spot and then evolveinto a ring at at least one time point When ana-lyzing the correlation between pit lifetime andmaximum diameter we added the further con-straint of including only those pits formed afterthe first frame in order to insure that we couldaccurately measure the entire lifetimeWhenmeasuring the associations of actinwith

clathrin we first implemented the tracking al-gorithm above to obtain time-lapse CCP imagesfor each endocytic eventWe then created amaskfor each CCP identified in each frame equal tothe CCP size plus an additional boundary of onepixelWe then applied thesemasks to each frameof Lifeact data and integrated the actin fluores-cence within each CCP-derivedmask If the actinsignal integrated over the area of a given maskincreased during the final five frames of the lifeof the associated CCP it was decided that actinwas recruited to the CCP during the final stage ofendocytosis

REFERENCES AND NOTES

1 L Schermelleh R Heintzmann H Leonhardt A guide to super-resolution fluorescence microscopy J Cell Biol 190 165ndash175(2010) doi 101083jcb201002018 pmid 20643879

2 U Schnell F Dijk K A Sjollema B N GiepmansImmunolabeling artifacts and the need for live-cell imagingNat Methods 9 152ndash158 (2012) doi 101038nmeth1855pmid 22290187

3 R P Nieuwenhuizen et al Measuring image resolution inoptical nanoscopy Nat Methods 10 557ndash562 (2013)doi 101038nmeth2448 pmid 23624665

4 X Shu et al A genetically encoded tag for correlated light andelectron microscopy of intact cells tissues and organismsPLOS Biol 9 e1001041 (2011) doi 101371journalpbio1001041 pmid 21483721

5 J D Martell et al Engineered ascorbate peroxidase as agenetically encoded reporter for electron microscopy NatBiotechnol 30 1143ndash1148 (2012) doi 101038nbt2375pmid 23086203

6 H Shroff C G Galbraith J A Galbraith E Betzig Live-cellphotoactivated localization microscopy of nanoscale adhesiondynamics Nat Methods 5 417ndash423 (2008) doi 101038nmeth1202 pmid 18408726

7 S H Shim et al Super-resolution fluorescence imaging oforganelles in live cells with photoswitchable membrane probesProc Natl Acad Sci USA 109 13978ndash13983 (2012)doi 101073pnas1201882109 pmid 22891300

8 B Hein K I Willig S W Hell Stimulated emission depletion(STED) nanoscopy of a fluorescent protein-labeled organelleinside a living cell Proc Natl Acad Sci USA 10514271ndash14276 (2008) doi 101073pnas0807705105pmid 18796604

9 V Westphal et al Video-rate far-field optical nanoscopydissects synaptic vesicle movement Science 320 246ndash249(2008) doi 101126science1154228 pmid 18292304

10 T Grotjohann et al rsEGFP2 enables fast RESOLFT nanoscopyof living cells eLife 1 e00248 (2012) doi 107554eLife00248 pmid 23330067

11 A Chmyrov et al Nanoscopy with more than 100000lsquodoughnutsrsquo Nat Methods 10 737ndash740 (2013) doi 101038nmeth2556 pmid 23832150

12 Materials and methods are available as supplementarymaterials on Science Online

13 P Kner B B Chhun E R Griffis L Winoto M G GustafssonSuper-resolution video microscopy of live cells by structuredillumination Nat Methods 6 339ndash342 (2009) doi 101038nmeth1324 pmid 19404253

14 L Shao P Kner E H Rego M G Gustafsson Super-resolution 3D microscopy of live whole cells using structuredillumination Nat Methods 8 1044ndash1046 (2011) doi 101038nmeth1734 pmid 22002026

15 R Fiolka L Shao E H Rego M W DavidsonM G Gustafsson Time-lapse two-color 3D imaging of live cellswith doubled resolution using structured illumination ProcNatl Acad Sci USA 109 5311ndash5315 (2012) doi 101073pnas1119262109 pmid 22431626

16 J Riedl et al Lifeact A versatile marker to visualize F-actinNat Methods 5 605ndash607 (2008) doi 101038nmeth1220pmid 18536722

17 H T McMahon E Boucrot Molecular mechanism andphysiological functions of clathrin-mediated endocytosis NatRev Mol Cell Biol 12 517ndash533 (2011) doi 101038nrm3151pmid 21779028

18 M Ehrlich et al Endocytosis by random initiation andstabilization of clathrin-coated pits Cell 118 591ndash605 (2004)doi 101016jcell200408017 pmid 15339664

19 I Gaidarov F Santini R A Warren J H Keen Spatial controlof coated-pit dynamics in living cells Nat Cell Biol 1 1ndash7(1999) pmid 10559856

20 S Saffarian E Cocucci T Kirchhausen Distinct dynamics ofendocytic clathrin-coated pits and coated plaques PLOS Biol7 e1000191 (2009) doi 101371journalpbio1000191pmid 19809571

21 J Grove et al Flat clathrin lattices Stable features of theplasma membrane Mol Biol Cell 25 3581ndash3594 (2014)doi 101091mbcE14-06-1154 pmid 25165141

22 J Heuser Effects of cytoplasmic acidification on clathrin latticemorphology J Cell Biol 108 401ndash411 (1989) doi 101083jcb1082401 pmid 2563729

23 M Kaksonen C P Toret D G Drubin Harnessing actindynamics for clathrin-mediated endocytosis Nat Rev Mol CellBiol 7 404ndash414 (2006) doi 101038nrm1940pmid 16723976

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-9

RESEARCH | RESEARCH ARTICLE

(fig S33 and movie S12) The smaller laterallymobile fraction in each case appeared as distorteddiscontinuous rings or quasiperiodic patches(fig S35) These morphologies are indicative ofmotion-induced artifacts and underscore thedifficulty of live-cell SR imaging by any methodHigher resolution must be accompanied by pro-portionally faster acquisition times to followdynamic events of a given velocity yet higherresolution also requires a quadratically increasingnumber of raw measurements for each two-dimensional (2D) image frame Even the com-paratively brief 035 swe needed to acquireN= 25raw images for each PA NL-SIM image wasinsufficient to accurately depict caveolae mov-ing by much more than our 59-nm resolution inthis timeNevertheless by further increasing SFact we

were able to saturate the fraction of molecules inthe activated state near the maxima of the pat-terned activation light (movie S6 part 2) Saturated

PA NL-SIM generates an additional harmonic(H = 3) strong enough (fig S29) to further extendthe resolution to 45 nm (figs S36 and S37) andallowed us to identify even smaller Skylan-NS-caveolin rings unresolvable without the extraharmonic (Fig 3C) UsingN = 35 rather thanN =49 raw images per frame we balanced the re-sulting anisotropic resolution (fig S30) againstthe needs for rapid acquisition (049 sframe) andparsimonious use of the photon budget to im-age caveolin rings over 12 frames at 3 s intervals(movie S13)

Two-color live imaging via combinedTIRF-SIM and PA NL-SIM

By combining linear SIM and PA NL-SIM bothin TIRF we could study associations betweenfluorescent proteins one conventional and onephotoswtichable in two colors at higher resolu-tion than by means of linear TIRF-SIM aloneImages (Fig 4 A to C and figs S38 and S39) and

movies (Movie 8 and movie S14) of mCherry-Rab5amdasha regulator of the formation fusion andtransport of early endosomes (EEs) (36)mdashrevealedirregularly shaped dynamically remodelingpatchesof Rab5a (fig S39 A and B) consistent with thetubularvesicular architecture of EEs seen in EM(Fig 4B) (37) Numerous patches also featureddark spots (fig S39C) perhaps indicative of car-go or internal vacuoles depleted of Rab5a Mostpatchesmoved randomly between successive 20-stime intervals at velocities slow enough to avoidmotion artifacts during each 034-s acquisitionWe also observed a subpopulation of slowly grow-ing Skylan-NS-Lifeactndashassociated Rab5a patchesthat were constrained for minutes at a time (figS39D arrows) At the other extreme we occa-sionally observed streaks of Rab5amoving paral-lel to nearby actin filaments at velocities of 3 to5 mms (Fig 4C and fig S39E) These may repre-sent EEs actively transported alongmicrotubules(38) parallel to the filaments

aab3500-4 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 3 Live-cell non-linear structured illumi-nation microscopybased on patternedphotoactivation (A)Single time point from amovie of the evolution ofcortical f-actin in a COS-7cell at 23degC transfectedwith Skylan-NS-Lifeactseen at 62-nm resolution(Movie 6 fig S31 andmovie S8) (B) Magnifiedview from a different cellat 37degC comparingdiffraction-limited TIRFmicroscopy (top left)TIRF with deconvolution(top right) TIRF-SIM(bottom left) and non-linear TIRF-SIM withpatterned activation(PA NL-SIM bottomright) (movies S9 andS10) (C) Caveolae in aCOS-7 cell at 23degC trans-fected with Skylan-NS-caveolin comparing TIRFwith deconvolution (topleft 220-nm resolution)TIRF SIM (top right97-nm resolution) PANL-SIM (bottom left62-nm resolution) andsaturated PA NL-SIM(bottom right 45-nmresolution) (Insets) Asingle caveolae pit even-tually resolved as a ringby saturated PA NL-SIM(Movie 7 figs S34 toS37 and movie S13(D) Diversity of caveolae ring diameters as seen by means of PA NL-SIM (E) Larger rings that may represent surface-docked vesicles (F) Clusters of caveolaereminiscent of clathrin plaques (D) to (F) are from a different cell at 37degC (fig S33 and movie S12) Scale bars 3 mm (A) 1 mm (B) 200 nm (C) and 100 nm(D) (E) (F) and (C) inset

RESEARCH | RESEARCH ARTICLE

We also used PA NL-SIM and TIRF-SIM re-spectively to study the association of Skylan-NS-Lifeact with mCherry-a-actinin (Fig 4 D to Fand fig S40) Consistent with its role as an actin-bundling protein (39) in COS-7 cells we founda-actinin at the treadmilling edge of the lamelle-podium and at the basal surface in both filopodiaand the leading edges of growing membraneruffles (Fig 4F Movie 9 and movie S15) We alsoobserved concentrations of a-actinin along thesides (Fig 4E) and at the branching ends of stressfibers that likely attach to cell-substrate adhesions(40) Last a-actininwas present at dense junctionsof Lifeact-decorated filaments and Skylan-NS-Lifeact rings as described above were colocalizedin every instance with a mCherryndasha-actinin ringof similar size (fig S41) Septins another class ofactin-bundling proteins have been shown (41) toproduce f-actin rings in vitro (albeit of larger sizethan here) so perhaps a-actinin not only aids inbundling actin filaments in nanometric rings butalso contributes to their extreme curvature

3D live-cell imaging with combined PANL-SIM and lattice light sheet microscopy

Although the ~50- to 200-nm extent of the eva-nescent excitation field we used in the examplesabove eliminated out-of-focus background andconfined potentially phototoxic exposure to aminute fraction of the cellular volume it alsolimited our observations to this subvolume andseverely restricted the total photon budget avail-able for those targets unable to be replenishedfrom the cytosol during the imaging intervalTo extend our observations to the entire cell

we turned to live-cell 3D-SIM (14 15) Unfortu-nately traditional 3D-SIM with linear widefieldexcitation brings limitations of its own It is slow

(~20 s acquisition for whole adherent HeLa cells)limited to thin specimens (because of out-of-focusbackground) and requires high SNR for accurateimage reconstruction It is also potentially photo-toxic and bleaches specimens rapidly because ofcontinuous whole-cell illumination These prob-lems would all be greatly magnified in its directextension to PA NL-SIMThus to apply PA NL-SIM to living cells in

three dimensions (Fig 5) we used lattice lightsheet microscopy (42) In this technique an exci-tation objective (fig S42A) projects a thin sheetof light (fig S42A blue) through a specimen (figS42A orange) and the fluorescence generated inthe illuminated plane is collected by a detectionobjective and imaged onto a camera Repeatingthis process plane-by-plane through the specimenproduces a 3D image Restriction of the light tothe detection focal plane eliminates out-of-focusbackground increases the z axis resolution andgreatly reduces photobleaching and phototoxicityIn cross-section the light sheet has the 2D

periodic structure of an optical lattice (fig S42B)Sweeping the sheet back and forth along the xaxis produces time-averaged uniform illumina-tion offering high speed and diffraction-limitedxyz resolution of 230 by 230 by 370 nm as seenin a volume-rendered image of the actin cyto-skeleton (fig S43A) and its corresponding overalloptical transfer function (OTF) (fig S43D) Step-ping the sheet in x in five equal fractions of thelattice period and applying the algorithms of3D-SIM to the resulting five raw images perplane extends the xyz resolution to 150 by 230by 280 nm (fig S43 B and E) but at the cost of atleast 5times longer acquisition times (42)To further extend the 3D resolution via PA

NL-SIMwe first photoactivated targetmolecules

fused to Skylan-NS using a hexagonal lattice lightsheet of l = 405 nm wavelength having H = 2harmonics (fig S43E) We then imaged the fluo-rescence from the activated region exciting thefluorescencewith a lattice light sheet of l =488nmwavelength having the same hexagonal sym-metry and period (fig S42B bottom) as the ac-tivation lattice For activationwell below saturationthe product of the activation and excitation pat-terns creates a fluorescence emission patternwithin the specimen having H = 4 harmonics(fig S43F) Thus we stepped the sheet in x in2H + 1 = 9 equal fractions of the lattice periodwhile recording nine images Repeating this pro-cess for every plane within the specimen we thenreconstructed a 3D PA NL-SIM volume-renderedimage (fig S43C) with resolution extended to118 by 230 by 170 nmWe used this approach to image mitochondria

in COS-7 cells (Fig 5A) as well as the actin cyto-skeleton (fig S43 A to C and movie S16) andthe Golgi apparatus (Fig 5B) inU2OS cells all at23degC so as to simplify the overlap of the activa-tion and excitation patterns Time-lapse 3D im-ages (Fig 5A bottom) and movies (Movie 10) ofSkylan-NSndashtagged translocase of outer mitochon-drial membrane 20 (TOM20) revealed the mi-gration constriction before fission and fusionof individual mitochondria (43 44) each clearlyresolved as a hollow tubular structure The 3Dvolume rendering and the widths of mitochon-drial membranes in individual xy orthosliceswere both comparable with similar data from afixed cell imaged with 3D localization micros-copy (45) at a reported xyz resolution of ~20 by20 by 60 nm (fig S44)A volume-rendered movie (movie S17) of the

Golgi-resident enzyme Mannosidase II (MannII)

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-5

Fig 4 Combined TIRF-SIMand PA NL-SIM of protein-pair dynamics in livingcells (A) Skylan-NS-Lifeact(orange PA NL-SIM) andmCherry-Rab5a a marker ofearly endosomes (greenTIRF-SIM) in a COS-7 cellat 23degC (Movie 8 figs S38and S39 and movie S14)(B) Comparison of EMimages of early endosomes(37) with similarly shapedRab5a patches seen in (A)(C) Magnified view at threesuccessive time pointsshowing rapid transportof a Rab5a streak parallel tothe cytoskeleton (D) Skylan-NS-Lifeact (green PA NL-SIM) and mCherry-a-actinin(purpleTIRF-SIM) in a COS-7cell at 23degC (Movie 9 figsS40 and S41 and movieS15) (E) Magnified viewfrom (D) with Lifeact (top)a-actinin (middle) and overlay (bottom) showing paired association at focal adhesions and along the sides of large stress fibers (F) Evolution of amembrane ruffle showing a-actinin concentrated at the leading edge Scale bars 5 mm (A) (D) 200 nm (B) 1 mm (C) and (E) and 500 nm (F)

RESEARCH | RESEARCH ARTICLE

tagged with Skylan-NS in a U2OS cell as seenlooking into the cis-face from the nucleus showedMann II concentrated in a hollow sphere ofcisternae having a cis-facing void Time-lapse3D data (Fig 5B andmovie S18) color-coded forheight showed the docking of small vesicles (Fig5B white arrows) that may represent pre-Golgiintermediates (46) as well as the rapid export ofMann II in long tubular post-Golgi carriers (Fig5B red arrows) (47)The volumetric resolution of 3D lattice light

sheet PA NL-SIM at the 06-NA excitation and11-NA detection we used here is comparablewith the 105- by 105- by 369-nm xyz resolutionof widefield 3D-SIM at 12 NA However thelattice approach has twofold higher axial resolu-tion and fourfold better than traditional diffraction-limited microscopy It is therefore better suitedto problems in which its superior optical section-ing is essential such as in resolving heterogene-ities in nuclear architecture distinguishing eventsoccurring at the dorsal or ventral plasma mem-brane or as above tracking vesicles through thesecretory pathway Whole-cell acquisition times(705 and 327 s in Fig 5 A and B respectively)are slow compared with PA NL-SIM in TIRF butsimilar to widefield 3D-SIM However thanks tothe oblique imaging geometry (fig S42) restrictedxy fields of view can be imaged at proportion-ally faster speed through the entire thickness ofthe cell

Discussion

The above results provide but a brief glimpse ofthe biology that might be uncovered with thelive-cellndashcompatible SRmethods of high-NATIRF-SIM and PA NL-SIM We have measured andcorrelated the diameters and lifetimes of CCPsobserved at high resolution different forms ofCCP initiation and shown that CCP internaliza-

tion is aided by actin filaments in about half of allcases We have seen that caveolin localizes notonly to the 60- to 80-nm invaginated caveloaecommon in EM images but also to much largerring-like structures and have followed dynamicchanges in the shapes of early endosomes Lastwe have observed the nanoscale remodeling ofthe actin cytoskeleton in relation to clathrin andRab5a as well as cytoskeletal-related proteinssuch as myosin IIA a-actinin and paxillinHowever the above results also amply illus-

trate the trade-offs inherent in live SR imagingWith high-NA TIRF-SIM at 17 NA we could ac-quire up to 200 image frames in lt05 s each atintensities of 20 to 100 Wcm2 and a resolutionof 84 nm (for GFP) whereas extending the reso-lution to 62 nm with PA NL-SIM restricted us tono more than 40 frames and further extensionto 45 nm with saturated PA NL-SIM required490 Wcm2 and produced only 12 frames atuseful SNRIn short evenmodest gains in resolution come

at substantial cost in terms of the other metricsimportant for live-cell imaging These tradeoffsare not specific to SIM In fact our extensions ofSIM are far more compatible with live imagingthan any other form of SR fluorescence micros-copy of comparable resolution demonstrated todate In part this is because the OTF which de-fines the degree towhich different sample spatialfrequencies (representing differently sized struc-tures) are passed to the image is far stronger inthe 100-nm regime (fig S24B) for high-NA TIRF-SIM at 17 NA than other linear methods such asconfocal or image scanning microscopy (ISM)(48ndash50) and far stronger in the 50- to 100-nmregime (fig S24C) for PA NL-SIM than othernonlinear methods such as STED (8 9) point-scanning (PS) RESOLFT (10) or array-basedwide-field (WF) RESOLFT (11) As a result far fewer

photons need to be collected (fig S2) and far lesslight (fig S3) needs to be applied to the specimento see features in these regimes at acceptableSNR Localization microscopy is also photon in-efficient in that the density of localizedmoleculesis nearly always more limiting to the resolutionthan is the number of photons emittedper switch-ing cycle which dictates the localization preci-sion For example simulations (12) based on thetheoretical OTFs suggest that to resolve an 85-nmgrating PANL-SIM requires ~80times fewer photonsfrom the specimen per unit area than localiza-tionmicroscopy ~200times fewer thanWF-RESOLFTand ~15times fewer than PS-RESOLFT or STED eachat a depletion saturation factor of SFdepletion = 10(fig S2)Another reason for the greater compatibility of

high-NA TIRF-SIM and PA NL-SIM with livingcells is that they require much lower peak inten-sities of applied light High resolutionwith STEDor RESOLFT demands high factors of saturateddepletion (fig S25 A and C) that are wasteful ofthe photon budget (fig S25 B andD) and requireenormous intensities andor long exposures foractivation (fig S45) depletion (fig S28) and read-out of the final signal (fig S3) Localization mi-croscopy also requires high intensities to achievehigh photon emission and photoswitching ratesfrom single molecules For example extrapolat-ing from reported experimental values for live-cell imaging (table S1) the 08- to 35-Wcm2

activation intensity used over the 45- by 45-mmfield of view in Fig 3A in 12 s bymeans of PANL-SIM is 960000 times weaker than that whichwould be required to image the same area in thesame acquisition time by means of PS-RESOLFT(10) Similarly under the same parameters the100-Wcm2 read-out intensity used for PA NL-SIM shown in Fig 3A is 200 times weaker thanthat which would be required for localization

aab3500-6 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 5 Live-cell 3D PA NL-SIMvia lattice light sheet micros-copy (A) (Top) Membranemarker Skylan-NS-TOM20showing mitochondria in aCOS-7 cell at 23degC color-codedfor distance from the substrate(Bottom) Evolution of individualmitochondria showing fissionand fusion events the formerpreceded by mitochondrial con-striction (Movie 10 and fig S44)(B) Time-lapse distribution ofGolgi-resident enzyme Skylan-NS-Mann II in a U2OS cell at23degC showing centralizedcisternae surrounded byvesicles White arrowheads indi-cate a docking vesicle and redarrowheads highlight rapidexport of a long tubular vesicle(movies S17 and S18) Scalebars 5 mm (A) top 1 mm (A)bottom and 3 mm (B)

RESEARCH | RESEARCH ARTICLE

microscopy (6 7) and 640000 times less than PS-RESOLFT (10) Furthermore STED andRESOLFTrequire an additional depletion step not neededin PA NL-SIM which would further expose thesample to peak intensities of 807 MWcm2 forSTED (8) 17 MWcm2 for PS-RESOLFT (10) and3 kWcm2 forWF-RESOLFT (11) Even over smallimage fields nanoscopy with focused light suchas PS-RESOLFT and STED uses intensities 105-to 1010-fold larger than that of terrestrial solarflux and is thus ill-equipped to study live-celldynamics noninvasivelyOf course despite these gains no method of

live-cell fluorescencemicroscopy including high-NA TIRF-SIM and PA NL-SIM can claim to becompletely noninvasive owing to possible photo-induced physiological changes protein over-expression andor label-induced perturbationsFor example the gradual development of curvedfilopodia and membrane ruffles after the start ofimaging are shown in Movies 5 and 6 and movieS2 These may reflect a response to the illumina-tion although we have also commonly seen suchstructures under initial conditions when imagingwith diffraction-limited TIRF (fig S46) Anothercaveat is that all the cells except BSC-1 in thiswork were transiently transfected and henceexpression levels of the target proteins were un-controlled This could affect eithermorphologiessuch as the sizes of Rab5a-labeled endosomes(Fig 4 A to C and figs S38 and S39) (51) ordynamic phenotypes such as the growth rate ofmembrane ruffles inmCherryndasha-actininndashexpressingcells (Fig 4E Movie 9 and movie S15) Althoughendogenous expression levels can be achievedwith genome editing (52) even more light orlonger exposures would be needed for cases inwhich these levels are lower than those used hereThus the biological findings described in this workshould not be considered definitive More exten-sive measurements across multiple cell lines withcareful controls and targeted perturbation experi-ments will be needed to reach conclusive insightsThe lesson is that when addressing any biolog-

ical question by means of live-cell imaging it isprudent to startwith less invasive lower-resolutionmethods such aswidefield spinning disk confocalor lattice light sheetmicroscopy andmove progres-sively only as needed to more invasive higher-resolution methods such as 3D-SIM TIRF-SIMPANL-SIM and last localizationmicroscopy Seenfrom this perspective the two extended-resolutionmethods of high-NATIRF-SIMandPANL-SIMweintroduce here fill an important gap between the100-nm limit of traditional SIM and the macro-molecular level of localizationmicroscopy Togetherthey open the door to high-resolution minimallyinvasive studies of dynamic processes includingendocytosis exocytosis signal transduction proteindiffusion vesicle trafficking viral entry cytoskeletalremodeling interactions with the extracellularmatrix and the evolution of lipid rafts

Materials and methodsOptical path of the TIRF-SIM system

The schematic of TIRF-SIM system is presentedin fig S47A The beam from a laser combiner

equipped with 405 nm (250 mW RPMC OxxiusLBX-405-300-CIR-PP) 488 nm (500mW Coher-ent SAPPHIRE 488-500) and 560 nm (1W MPBCommunications 2RU-VFL-P-1000-560-B1R) lasersis passed through an acousto-optic tunable filter(AOTF AA Quanta Tech AOTFnC-400650-TN)The beam is then expanded to a 1e2 diameter of12 mm and sent to a phase-only modulator (13)consisting of a polarizing beam splitter a achro-matic half-wave plate (HWP Bolder Vision OptikBVO AHWP3) and a ferroelectric spatial lightmodulator (SLM ForthDimensionDisplays SXGA-3DM) Light diffracted by the grating patterndisplayed on SLM passes through a polarizationrotator (15) consisting of a liquid crystal cell (LCMeadowlark SWIFT) and an achromatic quarter-wave plate (QWP Bolder Vision Optik BVOAQWP3) which rotates the linear polarizationof the diffracted light so as to maintain thes-polarization necessary to maximize the patterncontrast for all pattern orientations A mask con-sisting of a hollow barrel with slots for differentpattern orientations (15) is driven by a galvano-metric scanner (Cambridge Technology 6230HB)to filter out all diffraction orders created by thebinary and pixelated nature of the SLM exceptfor the desired plusmn1 diffraction orders These arethen imaged at the back focal plane of the ob-jective (Olympus APON 100XHOTIRF 17 NA forhigh-NATIRF-SIMOlympusUAPON100XOTIRF149 NA for PA NL-SIM at 23degC or Zeiss Plan-Apochromat 100X Oil-HI 157 NA for high-NAPA-NL-SIM at 37degC) as two spots at oppositesides of the pupil After passage through the ob-jective the two beams intersect at the interfacebetween the coverslip and the sample at an angleexceeding the critical angle for total internal re-flection An evanescent standing wave penetrat-ing ~100 nm into the sample is thereby generatedconsisting of a sinusoidal pattern of excitationintensity that is a low-pass filtered image of theSLM pattern The period orientation and rela-tive phase of this excitation pattern can befinely tuned by altering the corresponding pat-tern displayed on SLM For each orientationand phase of the applied excitation pattern theresulting fluorescence is collected by the ob-jective focused by a tube lens at an interme-diate image plane separated from excitationlight by a dichroic mirror (Chroma ZT405488560tpc_225deg) placed between two relaylenses and reimaged onto a sCMOS camera(Hamamatsu Orca Flash 40 v2 sCMOS) wherethe structured fluorescence emission pattern isrecorded

Calibration of pattern overlap forPA NL-SIM

In order to maximize the amplitudes of the non-linear harmonics for PA NL-SIM to work efficient-ly the sinusoidal patterns of 405 nm activationlight and 488 nmexcitation and deactivation lightmust be aligned to precisely overlap one anotherAs noted above these patterns at the sampleplane are created by displaying correspondingbinary grating patterns on an SLM at a corre-sponding optically conjugate plane In this case

the period ps at the specimen is related to theperiod pSLM at the SLM by

ps =Ml middot pSLM eth1THORN

where M is the demagnification factor betweenthe two conjugate planes and is dictated to bethe focal lengths of the relay lenses between thetwo planes Unfortunately chromatic aberrationleads to slightly different focal lengths for evenachromatic relay lenses for different wavelengthsof light In particular in our system M405 andM488 vary by ~2 Considering that the sinusoi-dal interference pattern is composed of hundredsof periods across our 45- by 45-mm2 field-of-view(FOV) even this 2 difference results in sub-stantial drift in the relative phases of the 405-and 488-nmexcitationpatterns across the FOV (figS48 A to C) leading to spatially variable ampli-tudes for thenonlinearharmonics and correspond-ing spatially variable errors in the resultingSIM reconstructionsA straightforward way to compensate for

chromatic aberration and achieve identical peri-ods ps405 = ps488 at the sample (fig S47B) is tointroduce a period difference DpSLM between thetwo corresponding patterns at the SLM (figS47C) In fact in order to compensate completelyand achieve well-overlapped 405- and 488-nmexcitation patterns over the whole FOV we needto measure two parameters the initial perioddifference at the sampleDpi

s frac14 Dpis488 minus Dpis405

when pSLM is the same for bothwavelengths andthe phase differenceDfis frac14 Dfis488 minus Dfis405 whenps is the same Do to so we used a sample con-sisting of a dense but submonolayer spread ofgreen fluorescent beads excitable at both 405and 488 nm and proceeded as follows

Step 1

Keeping pSLM constant we acquired five imageseach of the sample under 405- and 488-nm sinus-oidal excitation with the phase shifted by pSLM5for each image at a given wavelength We then ap-plied the structured illumination (SI) reconstruc-tion algorithm (53) to each set of five images fromwhich pis405 and pi

s488 emerged as measuredoutputs For a given period pSLM488 used at theSLM for 488-nm excitation the correspondingperiod pSLM405 needed at the SLM for 405-nmexcitation to produce the same period ps at thesample for both wavelengths is then given by

pSLM405 frac14pis488pis405

pSLM488 eth2THORN

Step 2

After adjusting pSLM405 and pSLM488 to obtainthe same period ps at the sample for both wave-lengths a constant phase offset exists betweenthe two sinusoidal illumination patterns acrossthe FOV (fig S48 D and E) We measured thephase f for each wavelength by applying thesinusoidal illumination for that wavelength andthen recorded the position xn along the modu-lation direction and intensity In for each of Nbeads scattered across the FOV We then fit the

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-7

RESEARCH | RESEARCH ARTICLE

function I(x) = Imax[1 + sin(2pxps + f)]2 to thisdata to find f (fig S48F) A phase shift Df = f488 ndashf405 was then applied the SLM pattern for the405-nm illumination so as to bring it into phasewith the 488-nm illumination at the specimen(figs S48 G to I)

Step 3

Last we confirmed that both the period and phaseof the sinusoidal illumination patterns at the twowavelengths match across the entire FOV byremeasuring the periods ps488 ps405 and thephases f488 f405 as described above and con-firming that they are identical

SLM pattern generation

We generated the sinusoidal illumination pat-terns using a binary ferroelectric SLM (Forth Di-mension Displays SXGA-3DM) because it hasthe submillisecond switching times needed toacquire the nine (TIRF-SIM) 25 (PA NL-SIM) ormore (saturated PA NL-SIM) raw images of dif-ferent phase and orientation required to recon-struct a single SIM image in as fast as 100 to400msHowever care must be taken to account for thefinite pixel size of the SLM especially consideringthat subpixel adjustment accuracy is necessary toachieve precise pattern overlap at 405 and488nmas described in the previous section The SLMpattern-generation algorithms used in previouswork (13ndash15) do not provide such subpixel accu-racy Thus in this work we developed a newalgorithm that matches the two pattern periodsto 002 precision leading to a phase error nogreater than 18deg over the 45-mm FOVIn detail a set of radial vectors An define the

desired orientations of the grating pattern at theSLM The angular orientation of this radial setrelative to the x and y axes defined by pixel rowsand columns of the SLM is chosen so that eachvector is at least 4deg away from either axis This isessential to achieve subpixel precision in the ad-justment of the period For each orientation rep-resented by An we define a vector Bn that isorthogonal to An (fig S49) Likewise for everypixel of the SLM we define a pixel vector (suchas C1 or C2 in fig S49) from the point O at theintersection of An and Bn to the pixel We thencalculate F = [(C middotB)modp]p the fraction of theperiod p by which the pixel extends beyond anintegral number of periods on the SLM For apattern with a desired off fraction D per period(D = 05 in 2D SIM) the pixel is set to 0 if F lt Dand set to 1 otherwise Last to define the pixelpatterns required for the other N ndash 1 phases ofthe illumination for a given orientation the pointO is translated along Bn in steps of pN and thisprocess is repeatedwith the new vectorC for eachpixel Unlike the pixel assignment algorithm usedpreviously for SIM (15) this approach does notrely on unit-cell repetition and therefore doesnot succumb to error accumulation over theentire span of the SLM

Lattice light sheet PA NL-SIM system

To extend PA NL-SIM to three dimensions it isessential to minimize out-of-focus fluorescence

emission that can cause the shot noise in the DCharmonic to completely overwhelm the weaksignals in the nonlinear harmonics To accom-plish this we turned to the SIM mode of latticelight sheet microscopy (42) Just as in the case of2D-SIM and for the same reasons we chooseto introduce the nonlinear harmonics throughpatterned activation of Skylan-NS The excitationobjective (Special Optics 065 NA 374 mmWD)is placed perpendicular to the detection objective(Nikon CFI Apo LWD 25XW 11 NA 2 mmWD)to confine the illumination to the proximity ofthe latterrsquos focal plane (fig S42A) The latticepattern projected on the SLM (Forth DimensionDisplays SXGA-3DM) is imaged onto the focalplane of the excitation objective after the excita-tion is first spatially filtered by an annular mask(Photo-Science) and relayed by a pair of galva-nometers (Cambridge Technology 6215H) thatphase step the pattern in the x direction and scanthe light sheet in z Also as in 2D PA NL-SIM wematch the periods and phases of the 405- and488-nm lattices to exactly match by measuringtheir excitation profiles across the FOV using fluo-rescent beads (fig S42B) and adjusting accord-ingly The fluorescence emission is collected bythe detection objective and imaged by a tube lensonto a sCMOS camera (Hamamatsu Orca Flash40 v2) A 3D image is formed by repeating thisprocess as the sample is translated through thelight sheet with a piezoelectric stage (PhysikInstrumente P-6211CD) along an axis s in theplane of the cover slip and a 3D super-resolutionNL-SIM image is reconstructed as describedbelow

Data acquisitionHigh-NA TIRF SIM

All high-NA TIRF-SIM images were acquiredwith the Olympus 17-NA objective under thephysiological conditions of 37degC and 5 CO2 Ateach time point we acquired three raw images atsuccessive phase steps of 0 13 and 23 of theillumination period We then repeated this pro-cess with the standing wave excitation patternrotated plusmn120deg with respect to the first orienta-tion for a total of nine raw images The phasestepping and pattern rotation were accomplishedby rotating or translating the binary grating pat-terndisplayedon theSLMFormulticolor imagingwe acquired nine raw images at each excitationwavelength before moving to the next and thenrepeated this series at successive time points Wecould adjust the excitationNA for eachwavelengthby changing the period of the grating pattern at theSLM This allowed us to control penetration depthof the evanescent wave (fig S8) in order to ba-lance the number of excitable fluorescent mole-cules against the background fluorescence andpossible physiological effects of the excitation

PA NL-SIM and saturated PA NL-SIM

The high refractive index immersion oil requiredfor the Olympus 17-NA objective strongly ab-sorbs 405-nm light leading to a substantial reduc-tion in the modulation depth we could achieve inthe activation pattern at this wavelength Conse-

quently forNL-SIMwe first turned to theOlympus149-NA TIRF objective and imaged at room tem-perature (23degC) with L15 medium without phenolred having 10 fetal bovine serum (Life Technol-ogies) With this objective we were able to achievehigh modulation contrast while stably and pre-cisely overlapping the 405- and 488-nm standingwaves over the whole FOV An excitation NA of144 was used for both 488- and 560-nm light inthis case leading to 62-nm resolution for PANL-SIMwhen using green-emitting FPs Recently how-ever we found that the high refractive index im-mersion oil used for the Zeiss 157-NA objectivedid not absorb 405-nm light strongly and there-fore could be used to maintain precisely over-lapped 405- and 488-nm standing waves withhigh modulation contrast at 37degC and 5 CO2The excitation NA in this case was 152 for 488-nmlight leading to 59-nm resolution for PA NL-SIMwhen using green-emitting FPsThe exposure procedure for a single phase step

inNL-SIMconsists of (i) 405-nmpatterned illumi-nation for 1 ms to activate the fluorescent mol-ecules (ii) 488-nm patterned illumination for 5 to~30 ms to read-out the activated molecules and(iii) 488-nm uniform illumination for 2 to ~10 msto read-out the remaining activated molecules andreturn the sample back to the original unactivatedstate We collected the fluorescence from bothsteps (ii) and (iii) to reconstruct the SR imageDepending on the number of modulation har-monics H of non-negligible amplitude in theimage (H = 2 for PA-NL-SIM andH = 3 or possiblymore for saturated PA NL-SIM) we repeated thissequence for 2H + 1 raw images at each of 2H +1 angular orientations equally spaced around 360degfor a total of (2H + 1)2 raw images at each NL-SIMtime point An exceptionwas saturated PA-NL-SIMfor which to reduce the acquisition time weoften used only five orientations rather thansevenIn two-color imaging combining linear TIRF-SIM

and PA NL-SIM (Fig 4) at each time point weacquired the PANL-SIM image as discussed aboveHowever we acquired the TIRF-SIM image withfive instead of three orientations (15 raw images forthe TIRF-SIM channel at every time point) inorder to match the orientations of the five-slotgalvanometer-driven barrel mask used to pickout thedesireddiffractionorders for thePANL-SIMacquisition

3D PA NL-SIM with lattice lightsheet microscopy

Here we used a hexagonal lattice having aperiod large enough to contain two harmonicsfor each of the 405-nm activation and the 488-nm excitation (42)mdashone harmonic just belowthe Abbe limit of the 065-NA excitation objec-tive and the other at twice this period Theproduct of these patterns created a fluorescenceemission pattern containing H = 4 harmonics(fig S43F) However with a single excitation ob-jective we were limited to producing this pat-tern at only one orientation Therefore at eachplane of the 3D stack we acquired 2H + 1 = 9images resulting in improved resolution (Fig 5)

aab3500-8 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

RESEARCH | RESEARCH ARTICLE

in both the lateral and axial directions of thepattern

Reconstruction of SIM images

The raw image frames with patterned excitationwere processed and reconstructed into the super-resolved images by means of a previously de-scribed algorithm (53) In brief for each patternorientation with H modulation harmonics 2H +1 raw images are collected and Fourier transformedinto 2H + 1 information components These com-ponents are assembled by initially translating eachin Fourier space by a distance equal to the am-plitudeof the illuminationpatternvectornk0wherek0 is the spatial frequency of the illumination pat-tern and n = ndashH to H The pattern vector of eachinformation component is then fine-tuned byfinding the vector that maximizes the complexcross-correlation in the overlap region betweensuccessive components The modulation ampli-tude of the harmonic and its starting phase arefound through complex linear regression In linewith previous work (28) the modulation ampli-tudes for the highest harmonics are generally toolow for this empirical approach to work well sofor these the theoretical values of their complexamplitudes are used After fine-tuning the posi-tions and complex amplitudes of the informationcomponents in the overlap regions a generalizedWiener filter is applied to this expanded transferfunction to balance the amplitudes of the variousspatial frequencies against the underlying noiseNext an apodization function is applied to min-imize ringing artifacts when the result is Fourier-transformed back to real space However ratherthan the triangle apodization A(k) = 1 ndash kkmax

normally used (53) we applied a g apodizationA(k) = 1 ndash (kkmax)

g usually with g = 04 so thatthe higher spatial frequencies are not suppressedmore than necessary Furthermore we strictly fol-lowed the azimuthally dependent support kmax(q)of the expanded OTF (figs S7 and S30) to definethe endpoint of the apodization function This pro-vides additional suppression of ringing artifactsFor the time series data we independently imple-ment this reconstructionprocess for each timepoint

Cell culture transfection stainingand fixation

BSC-1 COS-7 U2OS andmouse embryonic fibro-blast (MEF) cells (American Type Culture Collec-tion) were grown to ~60 to 80 confluency inDulbeccorsquos modified eagle medium (DMEM) withhigh glucose and no phenol red supplementedwith 15 fetal bovine serum (Life Technologies)BSC-1 cells stably expressed EGFP-CLTA Othercells were transiently transfected with an AmaxaNucleofector 96-well shuttle system (Lonza) with1 mg DNA per 400000 cells with nucleofectionsolution and a program optimized for each cellline per the manufactures instructions Beforeimaging 25-mm or 5-mm coverslips were coatedwith 10 mgml fibronection (Millipore FC010) for24 hours before plating transfected cells Imagingwas performed in DMEM with HEPES if there isno CO2 control containing no phenol red at tem-peratures specifically stated in each case

In two-color imaging of CCPs and transferrinreceptors (TfRs) by means of high-NA TIRF-SIMMEF cells expressing clathrin light chain B fusedto the C terminal of mEmerald were incubatedwith DMEM medium containing 250 mgmLTfR bound to human transferrin conjugatedwith Alexa 568 (T23365 Life Technologies) for15 minFixed cells were treated for 15 min with fixa-

tion buffer containing 4 paraformaldehyde01 gluteraldehyde in PHEM buffer (25 mMHEPES 10mMEGTA 2mMMgCl2 and 120mMPIPES in pH 73)

Tracking analysis of CCPs

For each image frame we segmented the CCPsusing a watershed algorithm written in Matlab(MathWorks 2014a) and measured their cent-roids individually Subsequently the centroidpositionwas linked between time points using u-track 21 (54) This linking operation collectedsuccessive position information for each pit overthe entire endocytic process (Fig 2E) from ini-tiation to final internalization It was then straight-forward to determine the lifetime (Fig 2A) foreach endocytic eventIn order to precisely measure the pit diameter

(Fig 2 B and C) we first measured the systemmagnification to the camera by imaging a stan-dard fine counting grid (2280-32 Ted Pella) TheSIM image of each CCP was then deconvolvedwith the equivalent PSF of the SIM system tocompensate for the broadening due to the finiteresolution of the instrument Last we measuredthe diameter of each deconvolved pit using anintensity-weighted average radius relative to thecentroid of the pit In certain cases (Fig 2A andMovie 3) pits were color-coded at each timepoint based on the time since their initiation tothe current time pointOne challenge in this analysis was how to

identify isolated pits rather than aggregates andhow to be sure that these represented true pitsrather than noise or disorganized patches ofnonassembled clathrin To accomplish this weset some conditions during the analysis such asthat a pit must start as a spot and then evolveinto a ring at at least one time point When ana-lyzing the correlation between pit lifetime andmaximum diameter we added the further con-straint of including only those pits formed afterthe first frame in order to insure that we couldaccurately measure the entire lifetimeWhenmeasuring the associations of actinwith

clathrin we first implemented the tracking al-gorithm above to obtain time-lapse CCP imagesfor each endocytic eventWe then created amaskfor each CCP identified in each frame equal tothe CCP size plus an additional boundary of onepixelWe then applied thesemasks to each frameof Lifeact data and integrated the actin fluores-cence within each CCP-derivedmask If the actinsignal integrated over the area of a given maskincreased during the final five frames of the lifeof the associated CCP it was decided that actinwas recruited to the CCP during the final stage ofendocytosis

REFERENCES AND NOTES

1 L Schermelleh R Heintzmann H Leonhardt A guide to super-resolution fluorescence microscopy J Cell Biol 190 165ndash175(2010) doi 101083jcb201002018 pmid 20643879

2 U Schnell F Dijk K A Sjollema B N GiepmansImmunolabeling artifacts and the need for live-cell imagingNat Methods 9 152ndash158 (2012) doi 101038nmeth1855pmid 22290187

3 R P Nieuwenhuizen et al Measuring image resolution inoptical nanoscopy Nat Methods 10 557ndash562 (2013)doi 101038nmeth2448 pmid 23624665

4 X Shu et al A genetically encoded tag for correlated light andelectron microscopy of intact cells tissues and organismsPLOS Biol 9 e1001041 (2011) doi 101371journalpbio1001041 pmid 21483721

5 J D Martell et al Engineered ascorbate peroxidase as agenetically encoded reporter for electron microscopy NatBiotechnol 30 1143ndash1148 (2012) doi 101038nbt2375pmid 23086203

6 H Shroff C G Galbraith J A Galbraith E Betzig Live-cellphotoactivated localization microscopy of nanoscale adhesiondynamics Nat Methods 5 417ndash423 (2008) doi 101038nmeth1202 pmid 18408726

7 S H Shim et al Super-resolution fluorescence imaging oforganelles in live cells with photoswitchable membrane probesProc Natl Acad Sci USA 109 13978ndash13983 (2012)doi 101073pnas1201882109 pmid 22891300

8 B Hein K I Willig S W Hell Stimulated emission depletion(STED) nanoscopy of a fluorescent protein-labeled organelleinside a living cell Proc Natl Acad Sci USA 10514271ndash14276 (2008) doi 101073pnas0807705105pmid 18796604

9 V Westphal et al Video-rate far-field optical nanoscopydissects synaptic vesicle movement Science 320 246ndash249(2008) doi 101126science1154228 pmid 18292304

10 T Grotjohann et al rsEGFP2 enables fast RESOLFT nanoscopyof living cells eLife 1 e00248 (2012) doi 107554eLife00248 pmid 23330067

11 A Chmyrov et al Nanoscopy with more than 100000lsquodoughnutsrsquo Nat Methods 10 737ndash740 (2013) doi 101038nmeth2556 pmid 23832150

12 Materials and methods are available as supplementarymaterials on Science Online

13 P Kner B B Chhun E R Griffis L Winoto M G GustafssonSuper-resolution video microscopy of live cells by structuredillumination Nat Methods 6 339ndash342 (2009) doi 101038nmeth1324 pmid 19404253

14 L Shao P Kner E H Rego M G Gustafsson Super-resolution 3D microscopy of live whole cells using structuredillumination Nat Methods 8 1044ndash1046 (2011) doi 101038nmeth1734 pmid 22002026

15 R Fiolka L Shao E H Rego M W DavidsonM G Gustafsson Time-lapse two-color 3D imaging of live cellswith doubled resolution using structured illumination ProcNatl Acad Sci USA 109 5311ndash5315 (2012) doi 101073pnas1119262109 pmid 22431626

16 J Riedl et al Lifeact A versatile marker to visualize F-actinNat Methods 5 605ndash607 (2008) doi 101038nmeth1220pmid 18536722

17 H T McMahon E Boucrot Molecular mechanism andphysiological functions of clathrin-mediated endocytosis NatRev Mol Cell Biol 12 517ndash533 (2011) doi 101038nrm3151pmid 21779028

18 M Ehrlich et al Endocytosis by random initiation andstabilization of clathrin-coated pits Cell 118 591ndash605 (2004)doi 101016jcell200408017 pmid 15339664

19 I Gaidarov F Santini R A Warren J H Keen Spatial controlof coated-pit dynamics in living cells Nat Cell Biol 1 1ndash7(1999) pmid 10559856

20 S Saffarian E Cocucci T Kirchhausen Distinct dynamics ofendocytic clathrin-coated pits and coated plaques PLOS Biol7 e1000191 (2009) doi 101371journalpbio1000191pmid 19809571

21 J Grove et al Flat clathrin lattices Stable features of theplasma membrane Mol Biol Cell 25 3581ndash3594 (2014)doi 101091mbcE14-06-1154 pmid 25165141

22 J Heuser Effects of cytoplasmic acidification on clathrin latticemorphology J Cell Biol 108 401ndash411 (1989) doi 101083jcb1082401 pmid 2563729

23 M Kaksonen C P Toret D G Drubin Harnessing actindynamics for clathrin-mediated endocytosis Nat Rev Mol CellBiol 7 404ndash414 (2006) doi 101038nrm1940pmid 16723976

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-9

RESEARCH | RESEARCH ARTICLE

We also used PA NL-SIM and TIRF-SIM re-spectively to study the association of Skylan-NS-Lifeact with mCherry-a-actinin (Fig 4 D to Fand fig S40) Consistent with its role as an actin-bundling protein (39) in COS-7 cells we founda-actinin at the treadmilling edge of the lamelle-podium and at the basal surface in both filopodiaand the leading edges of growing membraneruffles (Fig 4F Movie 9 and movie S15) We alsoobserved concentrations of a-actinin along thesides (Fig 4E) and at the branching ends of stressfibers that likely attach to cell-substrate adhesions(40) Last a-actininwas present at dense junctionsof Lifeact-decorated filaments and Skylan-NS-Lifeact rings as described above were colocalizedin every instance with a mCherryndasha-actinin ringof similar size (fig S41) Septins another class ofactin-bundling proteins have been shown (41) toproduce f-actin rings in vitro (albeit of larger sizethan here) so perhaps a-actinin not only aids inbundling actin filaments in nanometric rings butalso contributes to their extreme curvature

3D live-cell imaging with combined PANL-SIM and lattice light sheet microscopy

Although the ~50- to 200-nm extent of the eva-nescent excitation field we used in the examplesabove eliminated out-of-focus background andconfined potentially phototoxic exposure to aminute fraction of the cellular volume it alsolimited our observations to this subvolume andseverely restricted the total photon budget avail-able for those targets unable to be replenishedfrom the cytosol during the imaging intervalTo extend our observations to the entire cell

we turned to live-cell 3D-SIM (14 15) Unfortu-nately traditional 3D-SIM with linear widefieldexcitation brings limitations of its own It is slow

(~20 s acquisition for whole adherent HeLa cells)limited to thin specimens (because of out-of-focusbackground) and requires high SNR for accurateimage reconstruction It is also potentially photo-toxic and bleaches specimens rapidly because ofcontinuous whole-cell illumination These prob-lems would all be greatly magnified in its directextension to PA NL-SIMThus to apply PA NL-SIM to living cells in

three dimensions (Fig 5) we used lattice lightsheet microscopy (42) In this technique an exci-tation objective (fig S42A) projects a thin sheetof light (fig S42A blue) through a specimen (figS42A orange) and the fluorescence generated inthe illuminated plane is collected by a detectionobjective and imaged onto a camera Repeatingthis process plane-by-plane through the specimenproduces a 3D image Restriction of the light tothe detection focal plane eliminates out-of-focusbackground increases the z axis resolution andgreatly reduces photobleaching and phototoxicityIn cross-section the light sheet has the 2D

periodic structure of an optical lattice (fig S42B)Sweeping the sheet back and forth along the xaxis produces time-averaged uniform illumina-tion offering high speed and diffraction-limitedxyz resolution of 230 by 230 by 370 nm as seenin a volume-rendered image of the actin cyto-skeleton (fig S43A) and its corresponding overalloptical transfer function (OTF) (fig S43D) Step-ping the sheet in x in five equal fractions of thelattice period and applying the algorithms of3D-SIM to the resulting five raw images perplane extends the xyz resolution to 150 by 230by 280 nm (fig S43 B and E) but at the cost of atleast 5times longer acquisition times (42)To further extend the 3D resolution via PA

NL-SIMwe first photoactivated targetmolecules

fused to Skylan-NS using a hexagonal lattice lightsheet of l = 405 nm wavelength having H = 2harmonics (fig S43E) We then imaged the fluo-rescence from the activated region exciting thefluorescencewith a lattice light sheet of l =488nmwavelength having the same hexagonal sym-metry and period (fig S42B bottom) as the ac-tivation lattice For activationwell below saturationthe product of the activation and excitation pat-terns creates a fluorescence emission patternwithin the specimen having H = 4 harmonics(fig S43F) Thus we stepped the sheet in x in2H + 1 = 9 equal fractions of the lattice periodwhile recording nine images Repeating this pro-cess for every plane within the specimen we thenreconstructed a 3D PA NL-SIM volume-renderedimage (fig S43C) with resolution extended to118 by 230 by 170 nmWe used this approach to image mitochondria

in COS-7 cells (Fig 5A) as well as the actin cyto-skeleton (fig S43 A to C and movie S16) andthe Golgi apparatus (Fig 5B) inU2OS cells all at23degC so as to simplify the overlap of the activa-tion and excitation patterns Time-lapse 3D im-ages (Fig 5A bottom) and movies (Movie 10) ofSkylan-NSndashtagged translocase of outer mitochon-drial membrane 20 (TOM20) revealed the mi-gration constriction before fission and fusionof individual mitochondria (43 44) each clearlyresolved as a hollow tubular structure The 3Dvolume rendering and the widths of mitochon-drial membranes in individual xy orthosliceswere both comparable with similar data from afixed cell imaged with 3D localization micros-copy (45) at a reported xyz resolution of ~20 by20 by 60 nm (fig S44)A volume-rendered movie (movie S17) of the

Golgi-resident enzyme Mannosidase II (MannII)

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-5

Fig 4 Combined TIRF-SIMand PA NL-SIM of protein-pair dynamics in livingcells (A) Skylan-NS-Lifeact(orange PA NL-SIM) andmCherry-Rab5a a marker ofearly endosomes (greenTIRF-SIM) in a COS-7 cellat 23degC (Movie 8 figs S38and S39 and movie S14)(B) Comparison of EMimages of early endosomes(37) with similarly shapedRab5a patches seen in (A)(C) Magnified view at threesuccessive time pointsshowing rapid transportof a Rab5a streak parallel tothe cytoskeleton (D) Skylan-NS-Lifeact (green PA NL-SIM) and mCherry-a-actinin(purpleTIRF-SIM) in a COS-7cell at 23degC (Movie 9 figsS40 and S41 and movieS15) (E) Magnified viewfrom (D) with Lifeact (top)a-actinin (middle) and overlay (bottom) showing paired association at focal adhesions and along the sides of large stress fibers (F) Evolution of amembrane ruffle showing a-actinin concentrated at the leading edge Scale bars 5 mm (A) (D) 200 nm (B) 1 mm (C) and (E) and 500 nm (F)

RESEARCH | RESEARCH ARTICLE

tagged with Skylan-NS in a U2OS cell as seenlooking into the cis-face from the nucleus showedMann II concentrated in a hollow sphere ofcisternae having a cis-facing void Time-lapse3D data (Fig 5B andmovie S18) color-coded forheight showed the docking of small vesicles (Fig5B white arrows) that may represent pre-Golgiintermediates (46) as well as the rapid export ofMann II in long tubular post-Golgi carriers (Fig5B red arrows) (47)The volumetric resolution of 3D lattice light

sheet PA NL-SIM at the 06-NA excitation and11-NA detection we used here is comparablewith the 105- by 105- by 369-nm xyz resolutionof widefield 3D-SIM at 12 NA However thelattice approach has twofold higher axial resolu-tion and fourfold better than traditional diffraction-limited microscopy It is therefore better suitedto problems in which its superior optical section-ing is essential such as in resolving heterogene-ities in nuclear architecture distinguishing eventsoccurring at the dorsal or ventral plasma mem-brane or as above tracking vesicles through thesecretory pathway Whole-cell acquisition times(705 and 327 s in Fig 5 A and B respectively)are slow compared with PA NL-SIM in TIRF butsimilar to widefield 3D-SIM However thanks tothe oblique imaging geometry (fig S42) restrictedxy fields of view can be imaged at proportion-ally faster speed through the entire thickness ofthe cell

Discussion

The above results provide but a brief glimpse ofthe biology that might be uncovered with thelive-cellndashcompatible SRmethods of high-NATIRF-SIM and PA NL-SIM We have measured andcorrelated the diameters and lifetimes of CCPsobserved at high resolution different forms ofCCP initiation and shown that CCP internaliza-

tion is aided by actin filaments in about half of allcases We have seen that caveolin localizes notonly to the 60- to 80-nm invaginated caveloaecommon in EM images but also to much largerring-like structures and have followed dynamicchanges in the shapes of early endosomes Lastwe have observed the nanoscale remodeling ofthe actin cytoskeleton in relation to clathrin andRab5a as well as cytoskeletal-related proteinssuch as myosin IIA a-actinin and paxillinHowever the above results also amply illus-

trate the trade-offs inherent in live SR imagingWith high-NA TIRF-SIM at 17 NA we could ac-quire up to 200 image frames in lt05 s each atintensities of 20 to 100 Wcm2 and a resolutionof 84 nm (for GFP) whereas extending the reso-lution to 62 nm with PA NL-SIM restricted us tono more than 40 frames and further extensionto 45 nm with saturated PA NL-SIM required490 Wcm2 and produced only 12 frames atuseful SNRIn short evenmodest gains in resolution come

at substantial cost in terms of the other metricsimportant for live-cell imaging These tradeoffsare not specific to SIM In fact our extensions ofSIM are far more compatible with live imagingthan any other form of SR fluorescence micros-copy of comparable resolution demonstrated todate In part this is because the OTF which de-fines the degree towhich different sample spatialfrequencies (representing differently sized struc-tures) are passed to the image is far stronger inthe 100-nm regime (fig S24B) for high-NA TIRF-SIM at 17 NA than other linear methods such asconfocal or image scanning microscopy (ISM)(48ndash50) and far stronger in the 50- to 100-nmregime (fig S24C) for PA NL-SIM than othernonlinear methods such as STED (8 9) point-scanning (PS) RESOLFT (10) or array-basedwide-field (WF) RESOLFT (11) As a result far fewer

photons need to be collected (fig S2) and far lesslight (fig S3) needs to be applied to the specimento see features in these regimes at acceptableSNR Localization microscopy is also photon in-efficient in that the density of localizedmoleculesis nearly always more limiting to the resolutionthan is the number of photons emittedper switch-ing cycle which dictates the localization preci-sion For example simulations (12) based on thetheoretical OTFs suggest that to resolve an 85-nmgrating PANL-SIM requires ~80times fewer photonsfrom the specimen per unit area than localiza-tionmicroscopy ~200times fewer thanWF-RESOLFTand ~15times fewer than PS-RESOLFT or STED eachat a depletion saturation factor of SFdepletion = 10(fig S2)Another reason for the greater compatibility of

high-NA TIRF-SIM and PA NL-SIM with livingcells is that they require much lower peak inten-sities of applied light High resolutionwith STEDor RESOLFT demands high factors of saturateddepletion (fig S25 A and C) that are wasteful ofthe photon budget (fig S25 B andD) and requireenormous intensities andor long exposures foractivation (fig S45) depletion (fig S28) and read-out of the final signal (fig S3) Localization mi-croscopy also requires high intensities to achievehigh photon emission and photoswitching ratesfrom single molecules For example extrapolat-ing from reported experimental values for live-cell imaging (table S1) the 08- to 35-Wcm2

activation intensity used over the 45- by 45-mmfield of view in Fig 3A in 12 s bymeans of PANL-SIM is 960000 times weaker than that whichwould be required to image the same area in thesame acquisition time by means of PS-RESOLFT(10) Similarly under the same parameters the100-Wcm2 read-out intensity used for PA NL-SIM shown in Fig 3A is 200 times weaker thanthat which would be required for localization

aab3500-6 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 5 Live-cell 3D PA NL-SIMvia lattice light sheet micros-copy (A) (Top) Membranemarker Skylan-NS-TOM20showing mitochondria in aCOS-7 cell at 23degC color-codedfor distance from the substrate(Bottom) Evolution of individualmitochondria showing fissionand fusion events the formerpreceded by mitochondrial con-striction (Movie 10 and fig S44)(B) Time-lapse distribution ofGolgi-resident enzyme Skylan-NS-Mann II in a U2OS cell at23degC showing centralizedcisternae surrounded byvesicles White arrowheads indi-cate a docking vesicle and redarrowheads highlight rapidexport of a long tubular vesicle(movies S17 and S18) Scalebars 5 mm (A) top 1 mm (A)bottom and 3 mm (B)

RESEARCH | RESEARCH ARTICLE

microscopy (6 7) and 640000 times less than PS-RESOLFT (10) Furthermore STED andRESOLFTrequire an additional depletion step not neededin PA NL-SIM which would further expose thesample to peak intensities of 807 MWcm2 forSTED (8) 17 MWcm2 for PS-RESOLFT (10) and3 kWcm2 forWF-RESOLFT (11) Even over smallimage fields nanoscopy with focused light suchas PS-RESOLFT and STED uses intensities 105-to 1010-fold larger than that of terrestrial solarflux and is thus ill-equipped to study live-celldynamics noninvasivelyOf course despite these gains no method of

live-cell fluorescencemicroscopy including high-NA TIRF-SIM and PA NL-SIM can claim to becompletely noninvasive owing to possible photo-induced physiological changes protein over-expression andor label-induced perturbationsFor example the gradual development of curvedfilopodia and membrane ruffles after the start ofimaging are shown in Movies 5 and 6 and movieS2 These may reflect a response to the illumina-tion although we have also commonly seen suchstructures under initial conditions when imagingwith diffraction-limited TIRF (fig S46) Anothercaveat is that all the cells except BSC-1 in thiswork were transiently transfected and henceexpression levels of the target proteins were un-controlled This could affect eithermorphologiessuch as the sizes of Rab5a-labeled endosomes(Fig 4 A to C and figs S38 and S39) (51) ordynamic phenotypes such as the growth rate ofmembrane ruffles inmCherryndasha-actininndashexpressingcells (Fig 4E Movie 9 and movie S15) Althoughendogenous expression levels can be achievedwith genome editing (52) even more light orlonger exposures would be needed for cases inwhich these levels are lower than those used hereThus the biological findings described in this workshould not be considered definitive More exten-sive measurements across multiple cell lines withcareful controls and targeted perturbation experi-ments will be needed to reach conclusive insightsThe lesson is that when addressing any biolog-

ical question by means of live-cell imaging it isprudent to startwith less invasive lower-resolutionmethods such aswidefield spinning disk confocalor lattice light sheetmicroscopy andmove progres-sively only as needed to more invasive higher-resolution methods such as 3D-SIM TIRF-SIMPANL-SIM and last localizationmicroscopy Seenfrom this perspective the two extended-resolutionmethods of high-NATIRF-SIMandPANL-SIMweintroduce here fill an important gap between the100-nm limit of traditional SIM and the macro-molecular level of localizationmicroscopy Togetherthey open the door to high-resolution minimallyinvasive studies of dynamic processes includingendocytosis exocytosis signal transduction proteindiffusion vesicle trafficking viral entry cytoskeletalremodeling interactions with the extracellularmatrix and the evolution of lipid rafts

Materials and methodsOptical path of the TIRF-SIM system

The schematic of TIRF-SIM system is presentedin fig S47A The beam from a laser combiner

equipped with 405 nm (250 mW RPMC OxxiusLBX-405-300-CIR-PP) 488 nm (500mW Coher-ent SAPPHIRE 488-500) and 560 nm (1W MPBCommunications 2RU-VFL-P-1000-560-B1R) lasersis passed through an acousto-optic tunable filter(AOTF AA Quanta Tech AOTFnC-400650-TN)The beam is then expanded to a 1e2 diameter of12 mm and sent to a phase-only modulator (13)consisting of a polarizing beam splitter a achro-matic half-wave plate (HWP Bolder Vision OptikBVO AHWP3) and a ferroelectric spatial lightmodulator (SLM ForthDimensionDisplays SXGA-3DM) Light diffracted by the grating patterndisplayed on SLM passes through a polarizationrotator (15) consisting of a liquid crystal cell (LCMeadowlark SWIFT) and an achromatic quarter-wave plate (QWP Bolder Vision Optik BVOAQWP3) which rotates the linear polarizationof the diffracted light so as to maintain thes-polarization necessary to maximize the patterncontrast for all pattern orientations A mask con-sisting of a hollow barrel with slots for differentpattern orientations (15) is driven by a galvano-metric scanner (Cambridge Technology 6230HB)to filter out all diffraction orders created by thebinary and pixelated nature of the SLM exceptfor the desired plusmn1 diffraction orders These arethen imaged at the back focal plane of the ob-jective (Olympus APON 100XHOTIRF 17 NA forhigh-NATIRF-SIMOlympusUAPON100XOTIRF149 NA for PA NL-SIM at 23degC or Zeiss Plan-Apochromat 100X Oil-HI 157 NA for high-NAPA-NL-SIM at 37degC) as two spots at oppositesides of the pupil After passage through the ob-jective the two beams intersect at the interfacebetween the coverslip and the sample at an angleexceeding the critical angle for total internal re-flection An evanescent standing wave penetrat-ing ~100 nm into the sample is thereby generatedconsisting of a sinusoidal pattern of excitationintensity that is a low-pass filtered image of theSLM pattern The period orientation and rela-tive phase of this excitation pattern can befinely tuned by altering the corresponding pat-tern displayed on SLM For each orientationand phase of the applied excitation pattern theresulting fluorescence is collected by the ob-jective focused by a tube lens at an interme-diate image plane separated from excitationlight by a dichroic mirror (Chroma ZT405488560tpc_225deg) placed between two relaylenses and reimaged onto a sCMOS camera(Hamamatsu Orca Flash 40 v2 sCMOS) wherethe structured fluorescence emission pattern isrecorded

Calibration of pattern overlap forPA NL-SIM

In order to maximize the amplitudes of the non-linear harmonics for PA NL-SIM to work efficient-ly the sinusoidal patterns of 405 nm activationlight and 488 nmexcitation and deactivation lightmust be aligned to precisely overlap one anotherAs noted above these patterns at the sampleplane are created by displaying correspondingbinary grating patterns on an SLM at a corre-sponding optically conjugate plane In this case

the period ps at the specimen is related to theperiod pSLM at the SLM by

ps =Ml middot pSLM eth1THORN

where M is the demagnification factor betweenthe two conjugate planes and is dictated to bethe focal lengths of the relay lenses between thetwo planes Unfortunately chromatic aberrationleads to slightly different focal lengths for evenachromatic relay lenses for different wavelengthsof light In particular in our system M405 andM488 vary by ~2 Considering that the sinusoi-dal interference pattern is composed of hundredsof periods across our 45- by 45-mm2 field-of-view(FOV) even this 2 difference results in sub-stantial drift in the relative phases of the 405-and 488-nmexcitationpatterns across the FOV (figS48 A to C) leading to spatially variable ampli-tudes for thenonlinearharmonics and correspond-ing spatially variable errors in the resultingSIM reconstructionsA straightforward way to compensate for

chromatic aberration and achieve identical peri-ods ps405 = ps488 at the sample (fig S47B) is tointroduce a period difference DpSLM between thetwo corresponding patterns at the SLM (figS47C) In fact in order to compensate completelyand achieve well-overlapped 405- and 488-nmexcitation patterns over the whole FOV we needto measure two parameters the initial perioddifference at the sampleDpi

s frac14 Dpis488 minus Dpis405

when pSLM is the same for bothwavelengths andthe phase differenceDfis frac14 Dfis488 minus Dfis405 whenps is the same Do to so we used a sample con-sisting of a dense but submonolayer spread ofgreen fluorescent beads excitable at both 405and 488 nm and proceeded as follows

Step 1

Keeping pSLM constant we acquired five imageseach of the sample under 405- and 488-nm sinus-oidal excitation with the phase shifted by pSLM5for each image at a given wavelength We then ap-plied the structured illumination (SI) reconstruc-tion algorithm (53) to each set of five images fromwhich pis405 and pi

s488 emerged as measuredoutputs For a given period pSLM488 used at theSLM for 488-nm excitation the correspondingperiod pSLM405 needed at the SLM for 405-nmexcitation to produce the same period ps at thesample for both wavelengths is then given by

pSLM405 frac14pis488pis405

pSLM488 eth2THORN

Step 2

After adjusting pSLM405 and pSLM488 to obtainthe same period ps at the sample for both wave-lengths a constant phase offset exists betweenthe two sinusoidal illumination patterns acrossthe FOV (fig S48 D and E) We measured thephase f for each wavelength by applying thesinusoidal illumination for that wavelength andthen recorded the position xn along the modu-lation direction and intensity In for each of Nbeads scattered across the FOV We then fit the

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-7

RESEARCH | RESEARCH ARTICLE

function I(x) = Imax[1 + sin(2pxps + f)]2 to thisdata to find f (fig S48F) A phase shift Df = f488 ndashf405 was then applied the SLM pattern for the405-nm illumination so as to bring it into phasewith the 488-nm illumination at the specimen(figs S48 G to I)

Step 3

Last we confirmed that both the period and phaseof the sinusoidal illumination patterns at the twowavelengths match across the entire FOV byremeasuring the periods ps488 ps405 and thephases f488 f405 as described above and con-firming that they are identical

SLM pattern generation

We generated the sinusoidal illumination pat-terns using a binary ferroelectric SLM (Forth Di-mension Displays SXGA-3DM) because it hasthe submillisecond switching times needed toacquire the nine (TIRF-SIM) 25 (PA NL-SIM) ormore (saturated PA NL-SIM) raw images of dif-ferent phase and orientation required to recon-struct a single SIM image in as fast as 100 to400msHowever care must be taken to account for thefinite pixel size of the SLM especially consideringthat subpixel adjustment accuracy is necessary toachieve precise pattern overlap at 405 and488nmas described in the previous section The SLMpattern-generation algorithms used in previouswork (13ndash15) do not provide such subpixel accu-racy Thus in this work we developed a newalgorithm that matches the two pattern periodsto 002 precision leading to a phase error nogreater than 18deg over the 45-mm FOVIn detail a set of radial vectors An define the

desired orientations of the grating pattern at theSLM The angular orientation of this radial setrelative to the x and y axes defined by pixel rowsand columns of the SLM is chosen so that eachvector is at least 4deg away from either axis This isessential to achieve subpixel precision in the ad-justment of the period For each orientation rep-resented by An we define a vector Bn that isorthogonal to An (fig S49) Likewise for everypixel of the SLM we define a pixel vector (suchas C1 or C2 in fig S49) from the point O at theintersection of An and Bn to the pixel We thencalculate F = [(C middotB)modp]p the fraction of theperiod p by which the pixel extends beyond anintegral number of periods on the SLM For apattern with a desired off fraction D per period(D = 05 in 2D SIM) the pixel is set to 0 if F lt Dand set to 1 otherwise Last to define the pixelpatterns required for the other N ndash 1 phases ofthe illumination for a given orientation the pointO is translated along Bn in steps of pN and thisprocess is repeatedwith the new vectorC for eachpixel Unlike the pixel assignment algorithm usedpreviously for SIM (15) this approach does notrely on unit-cell repetition and therefore doesnot succumb to error accumulation over theentire span of the SLM

Lattice light sheet PA NL-SIM system

To extend PA NL-SIM to three dimensions it isessential to minimize out-of-focus fluorescence

emission that can cause the shot noise in the DCharmonic to completely overwhelm the weaksignals in the nonlinear harmonics To accom-plish this we turned to the SIM mode of latticelight sheet microscopy (42) Just as in the case of2D-SIM and for the same reasons we chooseto introduce the nonlinear harmonics throughpatterned activation of Skylan-NS The excitationobjective (Special Optics 065 NA 374 mmWD)is placed perpendicular to the detection objective(Nikon CFI Apo LWD 25XW 11 NA 2 mmWD)to confine the illumination to the proximity ofthe latterrsquos focal plane (fig S42A) The latticepattern projected on the SLM (Forth DimensionDisplays SXGA-3DM) is imaged onto the focalplane of the excitation objective after the excita-tion is first spatially filtered by an annular mask(Photo-Science) and relayed by a pair of galva-nometers (Cambridge Technology 6215H) thatphase step the pattern in the x direction and scanthe light sheet in z Also as in 2D PA NL-SIM wematch the periods and phases of the 405- and488-nm lattices to exactly match by measuringtheir excitation profiles across the FOV using fluo-rescent beads (fig S42B) and adjusting accord-ingly The fluorescence emission is collected bythe detection objective and imaged by a tube lensonto a sCMOS camera (Hamamatsu Orca Flash40 v2) A 3D image is formed by repeating thisprocess as the sample is translated through thelight sheet with a piezoelectric stage (PhysikInstrumente P-6211CD) along an axis s in theplane of the cover slip and a 3D super-resolutionNL-SIM image is reconstructed as describedbelow

Data acquisitionHigh-NA TIRF SIM

All high-NA TIRF-SIM images were acquiredwith the Olympus 17-NA objective under thephysiological conditions of 37degC and 5 CO2 Ateach time point we acquired three raw images atsuccessive phase steps of 0 13 and 23 of theillumination period We then repeated this pro-cess with the standing wave excitation patternrotated plusmn120deg with respect to the first orienta-tion for a total of nine raw images The phasestepping and pattern rotation were accomplishedby rotating or translating the binary grating pat-terndisplayedon theSLMFormulticolor imagingwe acquired nine raw images at each excitationwavelength before moving to the next and thenrepeated this series at successive time points Wecould adjust the excitationNA for eachwavelengthby changing the period of the grating pattern at theSLM This allowed us to control penetration depthof the evanescent wave (fig S8) in order to ba-lance the number of excitable fluorescent mole-cules against the background fluorescence andpossible physiological effects of the excitation

PA NL-SIM and saturated PA NL-SIM

The high refractive index immersion oil requiredfor the Olympus 17-NA objective strongly ab-sorbs 405-nm light leading to a substantial reduc-tion in the modulation depth we could achieve inthe activation pattern at this wavelength Conse-

quently forNL-SIMwe first turned to theOlympus149-NA TIRF objective and imaged at room tem-perature (23degC) with L15 medium without phenolred having 10 fetal bovine serum (Life Technol-ogies) With this objective we were able to achievehigh modulation contrast while stably and pre-cisely overlapping the 405- and 488-nm standingwaves over the whole FOV An excitation NA of144 was used for both 488- and 560-nm light inthis case leading to 62-nm resolution for PANL-SIMwhen using green-emitting FPs Recently how-ever we found that the high refractive index im-mersion oil used for the Zeiss 157-NA objectivedid not absorb 405-nm light strongly and there-fore could be used to maintain precisely over-lapped 405- and 488-nm standing waves withhigh modulation contrast at 37degC and 5 CO2The excitation NA in this case was 152 for 488-nmlight leading to 59-nm resolution for PA NL-SIMwhen using green-emitting FPsThe exposure procedure for a single phase step

inNL-SIMconsists of (i) 405-nmpatterned illumi-nation for 1 ms to activate the fluorescent mol-ecules (ii) 488-nm patterned illumination for 5 to~30 ms to read-out the activated molecules and(iii) 488-nm uniform illumination for 2 to ~10 msto read-out the remaining activated molecules andreturn the sample back to the original unactivatedstate We collected the fluorescence from bothsteps (ii) and (iii) to reconstruct the SR imageDepending on the number of modulation har-monics H of non-negligible amplitude in theimage (H = 2 for PA-NL-SIM andH = 3 or possiblymore for saturated PA NL-SIM) we repeated thissequence for 2H + 1 raw images at each of 2H +1 angular orientations equally spaced around 360degfor a total of (2H + 1)2 raw images at each NL-SIMtime point An exceptionwas saturated PA-NL-SIMfor which to reduce the acquisition time weoften used only five orientations rather thansevenIn two-color imaging combining linear TIRF-SIM

and PA NL-SIM (Fig 4) at each time point weacquired the PANL-SIM image as discussed aboveHowever we acquired the TIRF-SIM image withfive instead of three orientations (15 raw images forthe TIRF-SIM channel at every time point) inorder to match the orientations of the five-slotgalvanometer-driven barrel mask used to pickout thedesireddiffractionorders for thePANL-SIMacquisition

3D PA NL-SIM with lattice lightsheet microscopy

Here we used a hexagonal lattice having aperiod large enough to contain two harmonicsfor each of the 405-nm activation and the 488-nm excitation (42)mdashone harmonic just belowthe Abbe limit of the 065-NA excitation objec-tive and the other at twice this period Theproduct of these patterns created a fluorescenceemission pattern containing H = 4 harmonics(fig S43F) However with a single excitation ob-jective we were limited to producing this pat-tern at only one orientation Therefore at eachplane of the 3D stack we acquired 2H + 1 = 9images resulting in improved resolution (Fig 5)

aab3500-8 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

RESEARCH | RESEARCH ARTICLE

in both the lateral and axial directions of thepattern

Reconstruction of SIM images

The raw image frames with patterned excitationwere processed and reconstructed into the super-resolved images by means of a previously de-scribed algorithm (53) In brief for each patternorientation with H modulation harmonics 2H +1 raw images are collected and Fourier transformedinto 2H + 1 information components These com-ponents are assembled by initially translating eachin Fourier space by a distance equal to the am-plitudeof the illuminationpatternvectornk0wherek0 is the spatial frequency of the illumination pat-tern and n = ndashH to H The pattern vector of eachinformation component is then fine-tuned byfinding the vector that maximizes the complexcross-correlation in the overlap region betweensuccessive components The modulation ampli-tude of the harmonic and its starting phase arefound through complex linear regression In linewith previous work (28) the modulation ampli-tudes for the highest harmonics are generally toolow for this empirical approach to work well sofor these the theoretical values of their complexamplitudes are used After fine-tuning the posi-tions and complex amplitudes of the informationcomponents in the overlap regions a generalizedWiener filter is applied to this expanded transferfunction to balance the amplitudes of the variousspatial frequencies against the underlying noiseNext an apodization function is applied to min-imize ringing artifacts when the result is Fourier-transformed back to real space However ratherthan the triangle apodization A(k) = 1 ndash kkmax

normally used (53) we applied a g apodizationA(k) = 1 ndash (kkmax)

g usually with g = 04 so thatthe higher spatial frequencies are not suppressedmore than necessary Furthermore we strictly fol-lowed the azimuthally dependent support kmax(q)of the expanded OTF (figs S7 and S30) to definethe endpoint of the apodization function This pro-vides additional suppression of ringing artifactsFor the time series data we independently imple-ment this reconstructionprocess for each timepoint

Cell culture transfection stainingand fixation

BSC-1 COS-7 U2OS andmouse embryonic fibro-blast (MEF) cells (American Type Culture Collec-tion) were grown to ~60 to 80 confluency inDulbeccorsquos modified eagle medium (DMEM) withhigh glucose and no phenol red supplementedwith 15 fetal bovine serum (Life Technologies)BSC-1 cells stably expressed EGFP-CLTA Othercells were transiently transfected with an AmaxaNucleofector 96-well shuttle system (Lonza) with1 mg DNA per 400000 cells with nucleofectionsolution and a program optimized for each cellline per the manufactures instructions Beforeimaging 25-mm or 5-mm coverslips were coatedwith 10 mgml fibronection (Millipore FC010) for24 hours before plating transfected cells Imagingwas performed in DMEM with HEPES if there isno CO2 control containing no phenol red at tem-peratures specifically stated in each case

In two-color imaging of CCPs and transferrinreceptors (TfRs) by means of high-NA TIRF-SIMMEF cells expressing clathrin light chain B fusedto the C terminal of mEmerald were incubatedwith DMEM medium containing 250 mgmLTfR bound to human transferrin conjugatedwith Alexa 568 (T23365 Life Technologies) for15 minFixed cells were treated for 15 min with fixa-

tion buffer containing 4 paraformaldehyde01 gluteraldehyde in PHEM buffer (25 mMHEPES 10mMEGTA 2mMMgCl2 and 120mMPIPES in pH 73)

Tracking analysis of CCPs

For each image frame we segmented the CCPsusing a watershed algorithm written in Matlab(MathWorks 2014a) and measured their cent-roids individually Subsequently the centroidpositionwas linked between time points using u-track 21 (54) This linking operation collectedsuccessive position information for each pit overthe entire endocytic process (Fig 2E) from ini-tiation to final internalization It was then straight-forward to determine the lifetime (Fig 2A) foreach endocytic eventIn order to precisely measure the pit diameter

(Fig 2 B and C) we first measured the systemmagnification to the camera by imaging a stan-dard fine counting grid (2280-32 Ted Pella) TheSIM image of each CCP was then deconvolvedwith the equivalent PSF of the SIM system tocompensate for the broadening due to the finiteresolution of the instrument Last we measuredthe diameter of each deconvolved pit using anintensity-weighted average radius relative to thecentroid of the pit In certain cases (Fig 2A andMovie 3) pits were color-coded at each timepoint based on the time since their initiation tothe current time pointOne challenge in this analysis was how to

identify isolated pits rather than aggregates andhow to be sure that these represented true pitsrather than noise or disorganized patches ofnonassembled clathrin To accomplish this weset some conditions during the analysis such asthat a pit must start as a spot and then evolveinto a ring at at least one time point When ana-lyzing the correlation between pit lifetime andmaximum diameter we added the further con-straint of including only those pits formed afterthe first frame in order to insure that we couldaccurately measure the entire lifetimeWhenmeasuring the associations of actinwith

clathrin we first implemented the tracking al-gorithm above to obtain time-lapse CCP imagesfor each endocytic eventWe then created amaskfor each CCP identified in each frame equal tothe CCP size plus an additional boundary of onepixelWe then applied thesemasks to each frameof Lifeact data and integrated the actin fluores-cence within each CCP-derivedmask If the actinsignal integrated over the area of a given maskincreased during the final five frames of the lifeof the associated CCP it was decided that actinwas recruited to the CCP during the final stage ofendocytosis

REFERENCES AND NOTES

1 L Schermelleh R Heintzmann H Leonhardt A guide to super-resolution fluorescence microscopy J Cell Biol 190 165ndash175(2010) doi 101083jcb201002018 pmid 20643879

2 U Schnell F Dijk K A Sjollema B N GiepmansImmunolabeling artifacts and the need for live-cell imagingNat Methods 9 152ndash158 (2012) doi 101038nmeth1855pmid 22290187

3 R P Nieuwenhuizen et al Measuring image resolution inoptical nanoscopy Nat Methods 10 557ndash562 (2013)doi 101038nmeth2448 pmid 23624665

4 X Shu et al A genetically encoded tag for correlated light andelectron microscopy of intact cells tissues and organismsPLOS Biol 9 e1001041 (2011) doi 101371journalpbio1001041 pmid 21483721

5 J D Martell et al Engineered ascorbate peroxidase as agenetically encoded reporter for electron microscopy NatBiotechnol 30 1143ndash1148 (2012) doi 101038nbt2375pmid 23086203

6 H Shroff C G Galbraith J A Galbraith E Betzig Live-cellphotoactivated localization microscopy of nanoscale adhesiondynamics Nat Methods 5 417ndash423 (2008) doi 101038nmeth1202 pmid 18408726

7 S H Shim et al Super-resolution fluorescence imaging oforganelles in live cells with photoswitchable membrane probesProc Natl Acad Sci USA 109 13978ndash13983 (2012)doi 101073pnas1201882109 pmid 22891300

8 B Hein K I Willig S W Hell Stimulated emission depletion(STED) nanoscopy of a fluorescent protein-labeled organelleinside a living cell Proc Natl Acad Sci USA 10514271ndash14276 (2008) doi 101073pnas0807705105pmid 18796604

9 V Westphal et al Video-rate far-field optical nanoscopydissects synaptic vesicle movement Science 320 246ndash249(2008) doi 101126science1154228 pmid 18292304

10 T Grotjohann et al rsEGFP2 enables fast RESOLFT nanoscopyof living cells eLife 1 e00248 (2012) doi 107554eLife00248 pmid 23330067

11 A Chmyrov et al Nanoscopy with more than 100000lsquodoughnutsrsquo Nat Methods 10 737ndash740 (2013) doi 101038nmeth2556 pmid 23832150

12 Materials and methods are available as supplementarymaterials on Science Online

13 P Kner B B Chhun E R Griffis L Winoto M G GustafssonSuper-resolution video microscopy of live cells by structuredillumination Nat Methods 6 339ndash342 (2009) doi 101038nmeth1324 pmid 19404253

14 L Shao P Kner E H Rego M G Gustafsson Super-resolution 3D microscopy of live whole cells using structuredillumination Nat Methods 8 1044ndash1046 (2011) doi 101038nmeth1734 pmid 22002026

15 R Fiolka L Shao E H Rego M W DavidsonM G Gustafsson Time-lapse two-color 3D imaging of live cellswith doubled resolution using structured illumination ProcNatl Acad Sci USA 109 5311ndash5315 (2012) doi 101073pnas1119262109 pmid 22431626

16 J Riedl et al Lifeact A versatile marker to visualize F-actinNat Methods 5 605ndash607 (2008) doi 101038nmeth1220pmid 18536722

17 H T McMahon E Boucrot Molecular mechanism andphysiological functions of clathrin-mediated endocytosis NatRev Mol Cell Biol 12 517ndash533 (2011) doi 101038nrm3151pmid 21779028

18 M Ehrlich et al Endocytosis by random initiation andstabilization of clathrin-coated pits Cell 118 591ndash605 (2004)doi 101016jcell200408017 pmid 15339664

19 I Gaidarov F Santini R A Warren J H Keen Spatial controlof coated-pit dynamics in living cells Nat Cell Biol 1 1ndash7(1999) pmid 10559856

20 S Saffarian E Cocucci T Kirchhausen Distinct dynamics ofendocytic clathrin-coated pits and coated plaques PLOS Biol7 e1000191 (2009) doi 101371journalpbio1000191pmid 19809571

21 J Grove et al Flat clathrin lattices Stable features of theplasma membrane Mol Biol Cell 25 3581ndash3594 (2014)doi 101091mbcE14-06-1154 pmid 25165141

22 J Heuser Effects of cytoplasmic acidification on clathrin latticemorphology J Cell Biol 108 401ndash411 (1989) doi 101083jcb1082401 pmid 2563729

23 M Kaksonen C P Toret D G Drubin Harnessing actindynamics for clathrin-mediated endocytosis Nat Rev Mol CellBiol 7 404ndash414 (2006) doi 101038nrm1940pmid 16723976

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-9

RESEARCH | RESEARCH ARTICLE

tagged with Skylan-NS in a U2OS cell as seenlooking into the cis-face from the nucleus showedMann II concentrated in a hollow sphere ofcisternae having a cis-facing void Time-lapse3D data (Fig 5B andmovie S18) color-coded forheight showed the docking of small vesicles (Fig5B white arrows) that may represent pre-Golgiintermediates (46) as well as the rapid export ofMann II in long tubular post-Golgi carriers (Fig5B red arrows) (47)The volumetric resolution of 3D lattice light

sheet PA NL-SIM at the 06-NA excitation and11-NA detection we used here is comparablewith the 105- by 105- by 369-nm xyz resolutionof widefield 3D-SIM at 12 NA However thelattice approach has twofold higher axial resolu-tion and fourfold better than traditional diffraction-limited microscopy It is therefore better suitedto problems in which its superior optical section-ing is essential such as in resolving heterogene-ities in nuclear architecture distinguishing eventsoccurring at the dorsal or ventral plasma mem-brane or as above tracking vesicles through thesecretory pathway Whole-cell acquisition times(705 and 327 s in Fig 5 A and B respectively)are slow compared with PA NL-SIM in TIRF butsimilar to widefield 3D-SIM However thanks tothe oblique imaging geometry (fig S42) restrictedxy fields of view can be imaged at proportion-ally faster speed through the entire thickness ofthe cell

Discussion

The above results provide but a brief glimpse ofthe biology that might be uncovered with thelive-cellndashcompatible SRmethods of high-NATIRF-SIM and PA NL-SIM We have measured andcorrelated the diameters and lifetimes of CCPsobserved at high resolution different forms ofCCP initiation and shown that CCP internaliza-

tion is aided by actin filaments in about half of allcases We have seen that caveolin localizes notonly to the 60- to 80-nm invaginated caveloaecommon in EM images but also to much largerring-like structures and have followed dynamicchanges in the shapes of early endosomes Lastwe have observed the nanoscale remodeling ofthe actin cytoskeleton in relation to clathrin andRab5a as well as cytoskeletal-related proteinssuch as myosin IIA a-actinin and paxillinHowever the above results also amply illus-

trate the trade-offs inherent in live SR imagingWith high-NA TIRF-SIM at 17 NA we could ac-quire up to 200 image frames in lt05 s each atintensities of 20 to 100 Wcm2 and a resolutionof 84 nm (for GFP) whereas extending the reso-lution to 62 nm with PA NL-SIM restricted us tono more than 40 frames and further extensionto 45 nm with saturated PA NL-SIM required490 Wcm2 and produced only 12 frames atuseful SNRIn short evenmodest gains in resolution come

at substantial cost in terms of the other metricsimportant for live-cell imaging These tradeoffsare not specific to SIM In fact our extensions ofSIM are far more compatible with live imagingthan any other form of SR fluorescence micros-copy of comparable resolution demonstrated todate In part this is because the OTF which de-fines the degree towhich different sample spatialfrequencies (representing differently sized struc-tures) are passed to the image is far stronger inthe 100-nm regime (fig S24B) for high-NA TIRF-SIM at 17 NA than other linear methods such asconfocal or image scanning microscopy (ISM)(48ndash50) and far stronger in the 50- to 100-nmregime (fig S24C) for PA NL-SIM than othernonlinear methods such as STED (8 9) point-scanning (PS) RESOLFT (10) or array-basedwide-field (WF) RESOLFT (11) As a result far fewer

photons need to be collected (fig S2) and far lesslight (fig S3) needs to be applied to the specimento see features in these regimes at acceptableSNR Localization microscopy is also photon in-efficient in that the density of localizedmoleculesis nearly always more limiting to the resolutionthan is the number of photons emittedper switch-ing cycle which dictates the localization preci-sion For example simulations (12) based on thetheoretical OTFs suggest that to resolve an 85-nmgrating PANL-SIM requires ~80times fewer photonsfrom the specimen per unit area than localiza-tionmicroscopy ~200times fewer thanWF-RESOLFTand ~15times fewer than PS-RESOLFT or STED eachat a depletion saturation factor of SFdepletion = 10(fig S2)Another reason for the greater compatibility of

high-NA TIRF-SIM and PA NL-SIM with livingcells is that they require much lower peak inten-sities of applied light High resolutionwith STEDor RESOLFT demands high factors of saturateddepletion (fig S25 A and C) that are wasteful ofthe photon budget (fig S25 B andD) and requireenormous intensities andor long exposures foractivation (fig S45) depletion (fig S28) and read-out of the final signal (fig S3) Localization mi-croscopy also requires high intensities to achievehigh photon emission and photoswitching ratesfrom single molecules For example extrapolat-ing from reported experimental values for live-cell imaging (table S1) the 08- to 35-Wcm2

activation intensity used over the 45- by 45-mmfield of view in Fig 3A in 12 s bymeans of PANL-SIM is 960000 times weaker than that whichwould be required to image the same area in thesame acquisition time by means of PS-RESOLFT(10) Similarly under the same parameters the100-Wcm2 read-out intensity used for PA NL-SIM shown in Fig 3A is 200 times weaker thanthat which would be required for localization

aab3500-6 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

Fig 5 Live-cell 3D PA NL-SIMvia lattice light sheet micros-copy (A) (Top) Membranemarker Skylan-NS-TOM20showing mitochondria in aCOS-7 cell at 23degC color-codedfor distance from the substrate(Bottom) Evolution of individualmitochondria showing fissionand fusion events the formerpreceded by mitochondrial con-striction (Movie 10 and fig S44)(B) Time-lapse distribution ofGolgi-resident enzyme Skylan-NS-Mann II in a U2OS cell at23degC showing centralizedcisternae surrounded byvesicles White arrowheads indi-cate a docking vesicle and redarrowheads highlight rapidexport of a long tubular vesicle(movies S17 and S18) Scalebars 5 mm (A) top 1 mm (A)bottom and 3 mm (B)

RESEARCH | RESEARCH ARTICLE

microscopy (6 7) and 640000 times less than PS-RESOLFT (10) Furthermore STED andRESOLFTrequire an additional depletion step not neededin PA NL-SIM which would further expose thesample to peak intensities of 807 MWcm2 forSTED (8) 17 MWcm2 for PS-RESOLFT (10) and3 kWcm2 forWF-RESOLFT (11) Even over smallimage fields nanoscopy with focused light suchas PS-RESOLFT and STED uses intensities 105-to 1010-fold larger than that of terrestrial solarflux and is thus ill-equipped to study live-celldynamics noninvasivelyOf course despite these gains no method of

live-cell fluorescencemicroscopy including high-NA TIRF-SIM and PA NL-SIM can claim to becompletely noninvasive owing to possible photo-induced physiological changes protein over-expression andor label-induced perturbationsFor example the gradual development of curvedfilopodia and membrane ruffles after the start ofimaging are shown in Movies 5 and 6 and movieS2 These may reflect a response to the illumina-tion although we have also commonly seen suchstructures under initial conditions when imagingwith diffraction-limited TIRF (fig S46) Anothercaveat is that all the cells except BSC-1 in thiswork were transiently transfected and henceexpression levels of the target proteins were un-controlled This could affect eithermorphologiessuch as the sizes of Rab5a-labeled endosomes(Fig 4 A to C and figs S38 and S39) (51) ordynamic phenotypes such as the growth rate ofmembrane ruffles inmCherryndasha-actininndashexpressingcells (Fig 4E Movie 9 and movie S15) Althoughendogenous expression levels can be achievedwith genome editing (52) even more light orlonger exposures would be needed for cases inwhich these levels are lower than those used hereThus the biological findings described in this workshould not be considered definitive More exten-sive measurements across multiple cell lines withcareful controls and targeted perturbation experi-ments will be needed to reach conclusive insightsThe lesson is that when addressing any biolog-

ical question by means of live-cell imaging it isprudent to startwith less invasive lower-resolutionmethods such aswidefield spinning disk confocalor lattice light sheetmicroscopy andmove progres-sively only as needed to more invasive higher-resolution methods such as 3D-SIM TIRF-SIMPANL-SIM and last localizationmicroscopy Seenfrom this perspective the two extended-resolutionmethods of high-NATIRF-SIMandPANL-SIMweintroduce here fill an important gap between the100-nm limit of traditional SIM and the macro-molecular level of localizationmicroscopy Togetherthey open the door to high-resolution minimallyinvasive studies of dynamic processes includingendocytosis exocytosis signal transduction proteindiffusion vesicle trafficking viral entry cytoskeletalremodeling interactions with the extracellularmatrix and the evolution of lipid rafts

Materials and methodsOptical path of the TIRF-SIM system

The schematic of TIRF-SIM system is presentedin fig S47A The beam from a laser combiner

equipped with 405 nm (250 mW RPMC OxxiusLBX-405-300-CIR-PP) 488 nm (500mW Coher-ent SAPPHIRE 488-500) and 560 nm (1W MPBCommunications 2RU-VFL-P-1000-560-B1R) lasersis passed through an acousto-optic tunable filter(AOTF AA Quanta Tech AOTFnC-400650-TN)The beam is then expanded to a 1e2 diameter of12 mm and sent to a phase-only modulator (13)consisting of a polarizing beam splitter a achro-matic half-wave plate (HWP Bolder Vision OptikBVO AHWP3) and a ferroelectric spatial lightmodulator (SLM ForthDimensionDisplays SXGA-3DM) Light diffracted by the grating patterndisplayed on SLM passes through a polarizationrotator (15) consisting of a liquid crystal cell (LCMeadowlark SWIFT) and an achromatic quarter-wave plate (QWP Bolder Vision Optik BVOAQWP3) which rotates the linear polarizationof the diffracted light so as to maintain thes-polarization necessary to maximize the patterncontrast for all pattern orientations A mask con-sisting of a hollow barrel with slots for differentpattern orientations (15) is driven by a galvano-metric scanner (Cambridge Technology 6230HB)to filter out all diffraction orders created by thebinary and pixelated nature of the SLM exceptfor the desired plusmn1 diffraction orders These arethen imaged at the back focal plane of the ob-jective (Olympus APON 100XHOTIRF 17 NA forhigh-NATIRF-SIMOlympusUAPON100XOTIRF149 NA for PA NL-SIM at 23degC or Zeiss Plan-Apochromat 100X Oil-HI 157 NA for high-NAPA-NL-SIM at 37degC) as two spots at oppositesides of the pupil After passage through the ob-jective the two beams intersect at the interfacebetween the coverslip and the sample at an angleexceeding the critical angle for total internal re-flection An evanescent standing wave penetrat-ing ~100 nm into the sample is thereby generatedconsisting of a sinusoidal pattern of excitationintensity that is a low-pass filtered image of theSLM pattern The period orientation and rela-tive phase of this excitation pattern can befinely tuned by altering the corresponding pat-tern displayed on SLM For each orientationand phase of the applied excitation pattern theresulting fluorescence is collected by the ob-jective focused by a tube lens at an interme-diate image plane separated from excitationlight by a dichroic mirror (Chroma ZT405488560tpc_225deg) placed between two relaylenses and reimaged onto a sCMOS camera(Hamamatsu Orca Flash 40 v2 sCMOS) wherethe structured fluorescence emission pattern isrecorded

Calibration of pattern overlap forPA NL-SIM

In order to maximize the amplitudes of the non-linear harmonics for PA NL-SIM to work efficient-ly the sinusoidal patterns of 405 nm activationlight and 488 nmexcitation and deactivation lightmust be aligned to precisely overlap one anotherAs noted above these patterns at the sampleplane are created by displaying correspondingbinary grating patterns on an SLM at a corre-sponding optically conjugate plane In this case

the period ps at the specimen is related to theperiod pSLM at the SLM by

ps =Ml middot pSLM eth1THORN

where M is the demagnification factor betweenthe two conjugate planes and is dictated to bethe focal lengths of the relay lenses between thetwo planes Unfortunately chromatic aberrationleads to slightly different focal lengths for evenachromatic relay lenses for different wavelengthsof light In particular in our system M405 andM488 vary by ~2 Considering that the sinusoi-dal interference pattern is composed of hundredsof periods across our 45- by 45-mm2 field-of-view(FOV) even this 2 difference results in sub-stantial drift in the relative phases of the 405-and 488-nmexcitationpatterns across the FOV (figS48 A to C) leading to spatially variable ampli-tudes for thenonlinearharmonics and correspond-ing spatially variable errors in the resultingSIM reconstructionsA straightforward way to compensate for

chromatic aberration and achieve identical peri-ods ps405 = ps488 at the sample (fig S47B) is tointroduce a period difference DpSLM between thetwo corresponding patterns at the SLM (figS47C) In fact in order to compensate completelyand achieve well-overlapped 405- and 488-nmexcitation patterns over the whole FOV we needto measure two parameters the initial perioddifference at the sampleDpi

s frac14 Dpis488 minus Dpis405

when pSLM is the same for bothwavelengths andthe phase differenceDfis frac14 Dfis488 minus Dfis405 whenps is the same Do to so we used a sample con-sisting of a dense but submonolayer spread ofgreen fluorescent beads excitable at both 405and 488 nm and proceeded as follows

Step 1

Keeping pSLM constant we acquired five imageseach of the sample under 405- and 488-nm sinus-oidal excitation with the phase shifted by pSLM5for each image at a given wavelength We then ap-plied the structured illumination (SI) reconstruc-tion algorithm (53) to each set of five images fromwhich pis405 and pi

s488 emerged as measuredoutputs For a given period pSLM488 used at theSLM for 488-nm excitation the correspondingperiod pSLM405 needed at the SLM for 405-nmexcitation to produce the same period ps at thesample for both wavelengths is then given by

pSLM405 frac14pis488pis405

pSLM488 eth2THORN

Step 2

After adjusting pSLM405 and pSLM488 to obtainthe same period ps at the sample for both wave-lengths a constant phase offset exists betweenthe two sinusoidal illumination patterns acrossthe FOV (fig S48 D and E) We measured thephase f for each wavelength by applying thesinusoidal illumination for that wavelength andthen recorded the position xn along the modu-lation direction and intensity In for each of Nbeads scattered across the FOV We then fit the

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-7

RESEARCH | RESEARCH ARTICLE

function I(x) = Imax[1 + sin(2pxps + f)]2 to thisdata to find f (fig S48F) A phase shift Df = f488 ndashf405 was then applied the SLM pattern for the405-nm illumination so as to bring it into phasewith the 488-nm illumination at the specimen(figs S48 G to I)

Step 3

Last we confirmed that both the period and phaseof the sinusoidal illumination patterns at the twowavelengths match across the entire FOV byremeasuring the periods ps488 ps405 and thephases f488 f405 as described above and con-firming that they are identical

SLM pattern generation

We generated the sinusoidal illumination pat-terns using a binary ferroelectric SLM (Forth Di-mension Displays SXGA-3DM) because it hasthe submillisecond switching times needed toacquire the nine (TIRF-SIM) 25 (PA NL-SIM) ormore (saturated PA NL-SIM) raw images of dif-ferent phase and orientation required to recon-struct a single SIM image in as fast as 100 to400msHowever care must be taken to account for thefinite pixel size of the SLM especially consideringthat subpixel adjustment accuracy is necessary toachieve precise pattern overlap at 405 and488nmas described in the previous section The SLMpattern-generation algorithms used in previouswork (13ndash15) do not provide such subpixel accu-racy Thus in this work we developed a newalgorithm that matches the two pattern periodsto 002 precision leading to a phase error nogreater than 18deg over the 45-mm FOVIn detail a set of radial vectors An define the

desired orientations of the grating pattern at theSLM The angular orientation of this radial setrelative to the x and y axes defined by pixel rowsand columns of the SLM is chosen so that eachvector is at least 4deg away from either axis This isessential to achieve subpixel precision in the ad-justment of the period For each orientation rep-resented by An we define a vector Bn that isorthogonal to An (fig S49) Likewise for everypixel of the SLM we define a pixel vector (suchas C1 or C2 in fig S49) from the point O at theintersection of An and Bn to the pixel We thencalculate F = [(C middotB)modp]p the fraction of theperiod p by which the pixel extends beyond anintegral number of periods on the SLM For apattern with a desired off fraction D per period(D = 05 in 2D SIM) the pixel is set to 0 if F lt Dand set to 1 otherwise Last to define the pixelpatterns required for the other N ndash 1 phases ofthe illumination for a given orientation the pointO is translated along Bn in steps of pN and thisprocess is repeatedwith the new vectorC for eachpixel Unlike the pixel assignment algorithm usedpreviously for SIM (15) this approach does notrely on unit-cell repetition and therefore doesnot succumb to error accumulation over theentire span of the SLM

Lattice light sheet PA NL-SIM system

To extend PA NL-SIM to three dimensions it isessential to minimize out-of-focus fluorescence

emission that can cause the shot noise in the DCharmonic to completely overwhelm the weaksignals in the nonlinear harmonics To accom-plish this we turned to the SIM mode of latticelight sheet microscopy (42) Just as in the case of2D-SIM and for the same reasons we chooseto introduce the nonlinear harmonics throughpatterned activation of Skylan-NS The excitationobjective (Special Optics 065 NA 374 mmWD)is placed perpendicular to the detection objective(Nikon CFI Apo LWD 25XW 11 NA 2 mmWD)to confine the illumination to the proximity ofthe latterrsquos focal plane (fig S42A) The latticepattern projected on the SLM (Forth DimensionDisplays SXGA-3DM) is imaged onto the focalplane of the excitation objective after the excita-tion is first spatially filtered by an annular mask(Photo-Science) and relayed by a pair of galva-nometers (Cambridge Technology 6215H) thatphase step the pattern in the x direction and scanthe light sheet in z Also as in 2D PA NL-SIM wematch the periods and phases of the 405- and488-nm lattices to exactly match by measuringtheir excitation profiles across the FOV using fluo-rescent beads (fig S42B) and adjusting accord-ingly The fluorescence emission is collected bythe detection objective and imaged by a tube lensonto a sCMOS camera (Hamamatsu Orca Flash40 v2) A 3D image is formed by repeating thisprocess as the sample is translated through thelight sheet with a piezoelectric stage (PhysikInstrumente P-6211CD) along an axis s in theplane of the cover slip and a 3D super-resolutionNL-SIM image is reconstructed as describedbelow

Data acquisitionHigh-NA TIRF SIM

All high-NA TIRF-SIM images were acquiredwith the Olympus 17-NA objective under thephysiological conditions of 37degC and 5 CO2 Ateach time point we acquired three raw images atsuccessive phase steps of 0 13 and 23 of theillumination period We then repeated this pro-cess with the standing wave excitation patternrotated plusmn120deg with respect to the first orienta-tion for a total of nine raw images The phasestepping and pattern rotation were accomplishedby rotating or translating the binary grating pat-terndisplayedon theSLMFormulticolor imagingwe acquired nine raw images at each excitationwavelength before moving to the next and thenrepeated this series at successive time points Wecould adjust the excitationNA for eachwavelengthby changing the period of the grating pattern at theSLM This allowed us to control penetration depthof the evanescent wave (fig S8) in order to ba-lance the number of excitable fluorescent mole-cules against the background fluorescence andpossible physiological effects of the excitation

PA NL-SIM and saturated PA NL-SIM

The high refractive index immersion oil requiredfor the Olympus 17-NA objective strongly ab-sorbs 405-nm light leading to a substantial reduc-tion in the modulation depth we could achieve inthe activation pattern at this wavelength Conse-

quently forNL-SIMwe first turned to theOlympus149-NA TIRF objective and imaged at room tem-perature (23degC) with L15 medium without phenolred having 10 fetal bovine serum (Life Technol-ogies) With this objective we were able to achievehigh modulation contrast while stably and pre-cisely overlapping the 405- and 488-nm standingwaves over the whole FOV An excitation NA of144 was used for both 488- and 560-nm light inthis case leading to 62-nm resolution for PANL-SIMwhen using green-emitting FPs Recently how-ever we found that the high refractive index im-mersion oil used for the Zeiss 157-NA objectivedid not absorb 405-nm light strongly and there-fore could be used to maintain precisely over-lapped 405- and 488-nm standing waves withhigh modulation contrast at 37degC and 5 CO2The excitation NA in this case was 152 for 488-nmlight leading to 59-nm resolution for PA NL-SIMwhen using green-emitting FPsThe exposure procedure for a single phase step

inNL-SIMconsists of (i) 405-nmpatterned illumi-nation for 1 ms to activate the fluorescent mol-ecules (ii) 488-nm patterned illumination for 5 to~30 ms to read-out the activated molecules and(iii) 488-nm uniform illumination for 2 to ~10 msto read-out the remaining activated molecules andreturn the sample back to the original unactivatedstate We collected the fluorescence from bothsteps (ii) and (iii) to reconstruct the SR imageDepending on the number of modulation har-monics H of non-negligible amplitude in theimage (H = 2 for PA-NL-SIM andH = 3 or possiblymore for saturated PA NL-SIM) we repeated thissequence for 2H + 1 raw images at each of 2H +1 angular orientations equally spaced around 360degfor a total of (2H + 1)2 raw images at each NL-SIMtime point An exceptionwas saturated PA-NL-SIMfor which to reduce the acquisition time weoften used only five orientations rather thansevenIn two-color imaging combining linear TIRF-SIM

and PA NL-SIM (Fig 4) at each time point weacquired the PANL-SIM image as discussed aboveHowever we acquired the TIRF-SIM image withfive instead of three orientations (15 raw images forthe TIRF-SIM channel at every time point) inorder to match the orientations of the five-slotgalvanometer-driven barrel mask used to pickout thedesireddiffractionorders for thePANL-SIMacquisition

3D PA NL-SIM with lattice lightsheet microscopy

Here we used a hexagonal lattice having aperiod large enough to contain two harmonicsfor each of the 405-nm activation and the 488-nm excitation (42)mdashone harmonic just belowthe Abbe limit of the 065-NA excitation objec-tive and the other at twice this period Theproduct of these patterns created a fluorescenceemission pattern containing H = 4 harmonics(fig S43F) However with a single excitation ob-jective we were limited to producing this pat-tern at only one orientation Therefore at eachplane of the 3D stack we acquired 2H + 1 = 9images resulting in improved resolution (Fig 5)

aab3500-8 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

RESEARCH | RESEARCH ARTICLE

in both the lateral and axial directions of thepattern

Reconstruction of SIM images

The raw image frames with patterned excitationwere processed and reconstructed into the super-resolved images by means of a previously de-scribed algorithm (53) In brief for each patternorientation with H modulation harmonics 2H +1 raw images are collected and Fourier transformedinto 2H + 1 information components These com-ponents are assembled by initially translating eachin Fourier space by a distance equal to the am-plitudeof the illuminationpatternvectornk0wherek0 is the spatial frequency of the illumination pat-tern and n = ndashH to H The pattern vector of eachinformation component is then fine-tuned byfinding the vector that maximizes the complexcross-correlation in the overlap region betweensuccessive components The modulation ampli-tude of the harmonic and its starting phase arefound through complex linear regression In linewith previous work (28) the modulation ampli-tudes for the highest harmonics are generally toolow for this empirical approach to work well sofor these the theoretical values of their complexamplitudes are used After fine-tuning the posi-tions and complex amplitudes of the informationcomponents in the overlap regions a generalizedWiener filter is applied to this expanded transferfunction to balance the amplitudes of the variousspatial frequencies against the underlying noiseNext an apodization function is applied to min-imize ringing artifacts when the result is Fourier-transformed back to real space However ratherthan the triangle apodization A(k) = 1 ndash kkmax

normally used (53) we applied a g apodizationA(k) = 1 ndash (kkmax)

g usually with g = 04 so thatthe higher spatial frequencies are not suppressedmore than necessary Furthermore we strictly fol-lowed the azimuthally dependent support kmax(q)of the expanded OTF (figs S7 and S30) to definethe endpoint of the apodization function This pro-vides additional suppression of ringing artifactsFor the time series data we independently imple-ment this reconstructionprocess for each timepoint

Cell culture transfection stainingand fixation

BSC-1 COS-7 U2OS andmouse embryonic fibro-blast (MEF) cells (American Type Culture Collec-tion) were grown to ~60 to 80 confluency inDulbeccorsquos modified eagle medium (DMEM) withhigh glucose and no phenol red supplementedwith 15 fetal bovine serum (Life Technologies)BSC-1 cells stably expressed EGFP-CLTA Othercells were transiently transfected with an AmaxaNucleofector 96-well shuttle system (Lonza) with1 mg DNA per 400000 cells with nucleofectionsolution and a program optimized for each cellline per the manufactures instructions Beforeimaging 25-mm or 5-mm coverslips were coatedwith 10 mgml fibronection (Millipore FC010) for24 hours before plating transfected cells Imagingwas performed in DMEM with HEPES if there isno CO2 control containing no phenol red at tem-peratures specifically stated in each case

In two-color imaging of CCPs and transferrinreceptors (TfRs) by means of high-NA TIRF-SIMMEF cells expressing clathrin light chain B fusedto the C terminal of mEmerald were incubatedwith DMEM medium containing 250 mgmLTfR bound to human transferrin conjugatedwith Alexa 568 (T23365 Life Technologies) for15 minFixed cells were treated for 15 min with fixa-

tion buffer containing 4 paraformaldehyde01 gluteraldehyde in PHEM buffer (25 mMHEPES 10mMEGTA 2mMMgCl2 and 120mMPIPES in pH 73)

Tracking analysis of CCPs

For each image frame we segmented the CCPsusing a watershed algorithm written in Matlab(MathWorks 2014a) and measured their cent-roids individually Subsequently the centroidpositionwas linked between time points using u-track 21 (54) This linking operation collectedsuccessive position information for each pit overthe entire endocytic process (Fig 2E) from ini-tiation to final internalization It was then straight-forward to determine the lifetime (Fig 2A) foreach endocytic eventIn order to precisely measure the pit diameter

(Fig 2 B and C) we first measured the systemmagnification to the camera by imaging a stan-dard fine counting grid (2280-32 Ted Pella) TheSIM image of each CCP was then deconvolvedwith the equivalent PSF of the SIM system tocompensate for the broadening due to the finiteresolution of the instrument Last we measuredthe diameter of each deconvolved pit using anintensity-weighted average radius relative to thecentroid of the pit In certain cases (Fig 2A andMovie 3) pits were color-coded at each timepoint based on the time since their initiation tothe current time pointOne challenge in this analysis was how to

identify isolated pits rather than aggregates andhow to be sure that these represented true pitsrather than noise or disorganized patches ofnonassembled clathrin To accomplish this weset some conditions during the analysis such asthat a pit must start as a spot and then evolveinto a ring at at least one time point When ana-lyzing the correlation between pit lifetime andmaximum diameter we added the further con-straint of including only those pits formed afterthe first frame in order to insure that we couldaccurately measure the entire lifetimeWhenmeasuring the associations of actinwith

clathrin we first implemented the tracking al-gorithm above to obtain time-lapse CCP imagesfor each endocytic eventWe then created amaskfor each CCP identified in each frame equal tothe CCP size plus an additional boundary of onepixelWe then applied thesemasks to each frameof Lifeact data and integrated the actin fluores-cence within each CCP-derivedmask If the actinsignal integrated over the area of a given maskincreased during the final five frames of the lifeof the associated CCP it was decided that actinwas recruited to the CCP during the final stage ofendocytosis

REFERENCES AND NOTES

1 L Schermelleh R Heintzmann H Leonhardt A guide to super-resolution fluorescence microscopy J Cell Biol 190 165ndash175(2010) doi 101083jcb201002018 pmid 20643879

2 U Schnell F Dijk K A Sjollema B N GiepmansImmunolabeling artifacts and the need for live-cell imagingNat Methods 9 152ndash158 (2012) doi 101038nmeth1855pmid 22290187

3 R P Nieuwenhuizen et al Measuring image resolution inoptical nanoscopy Nat Methods 10 557ndash562 (2013)doi 101038nmeth2448 pmid 23624665

4 X Shu et al A genetically encoded tag for correlated light andelectron microscopy of intact cells tissues and organismsPLOS Biol 9 e1001041 (2011) doi 101371journalpbio1001041 pmid 21483721

5 J D Martell et al Engineered ascorbate peroxidase as agenetically encoded reporter for electron microscopy NatBiotechnol 30 1143ndash1148 (2012) doi 101038nbt2375pmid 23086203

6 H Shroff C G Galbraith J A Galbraith E Betzig Live-cellphotoactivated localization microscopy of nanoscale adhesiondynamics Nat Methods 5 417ndash423 (2008) doi 101038nmeth1202 pmid 18408726

7 S H Shim et al Super-resolution fluorescence imaging oforganelles in live cells with photoswitchable membrane probesProc Natl Acad Sci USA 109 13978ndash13983 (2012)doi 101073pnas1201882109 pmid 22891300

8 B Hein K I Willig S W Hell Stimulated emission depletion(STED) nanoscopy of a fluorescent protein-labeled organelleinside a living cell Proc Natl Acad Sci USA 10514271ndash14276 (2008) doi 101073pnas0807705105pmid 18796604

9 V Westphal et al Video-rate far-field optical nanoscopydissects synaptic vesicle movement Science 320 246ndash249(2008) doi 101126science1154228 pmid 18292304

10 T Grotjohann et al rsEGFP2 enables fast RESOLFT nanoscopyof living cells eLife 1 e00248 (2012) doi 107554eLife00248 pmid 23330067

11 A Chmyrov et al Nanoscopy with more than 100000lsquodoughnutsrsquo Nat Methods 10 737ndash740 (2013) doi 101038nmeth2556 pmid 23832150

12 Materials and methods are available as supplementarymaterials on Science Online

13 P Kner B B Chhun E R Griffis L Winoto M G GustafssonSuper-resolution video microscopy of live cells by structuredillumination Nat Methods 6 339ndash342 (2009) doi 101038nmeth1324 pmid 19404253

14 L Shao P Kner E H Rego M G Gustafsson Super-resolution 3D microscopy of live whole cells using structuredillumination Nat Methods 8 1044ndash1046 (2011) doi 101038nmeth1734 pmid 22002026

15 R Fiolka L Shao E H Rego M W DavidsonM G Gustafsson Time-lapse two-color 3D imaging of live cellswith doubled resolution using structured illumination ProcNatl Acad Sci USA 109 5311ndash5315 (2012) doi 101073pnas1119262109 pmid 22431626

16 J Riedl et al Lifeact A versatile marker to visualize F-actinNat Methods 5 605ndash607 (2008) doi 101038nmeth1220pmid 18536722

17 H T McMahon E Boucrot Molecular mechanism andphysiological functions of clathrin-mediated endocytosis NatRev Mol Cell Biol 12 517ndash533 (2011) doi 101038nrm3151pmid 21779028

18 M Ehrlich et al Endocytosis by random initiation andstabilization of clathrin-coated pits Cell 118 591ndash605 (2004)doi 101016jcell200408017 pmid 15339664

19 I Gaidarov F Santini R A Warren J H Keen Spatial controlof coated-pit dynamics in living cells Nat Cell Biol 1 1ndash7(1999) pmid 10559856

20 S Saffarian E Cocucci T Kirchhausen Distinct dynamics ofendocytic clathrin-coated pits and coated plaques PLOS Biol7 e1000191 (2009) doi 101371journalpbio1000191pmid 19809571

21 J Grove et al Flat clathrin lattices Stable features of theplasma membrane Mol Biol Cell 25 3581ndash3594 (2014)doi 101091mbcE14-06-1154 pmid 25165141

22 J Heuser Effects of cytoplasmic acidification on clathrin latticemorphology J Cell Biol 108 401ndash411 (1989) doi 101083jcb1082401 pmid 2563729

23 M Kaksonen C P Toret D G Drubin Harnessing actindynamics for clathrin-mediated endocytosis Nat Rev Mol CellBiol 7 404ndash414 (2006) doi 101038nrm1940pmid 16723976

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-9

RESEARCH | RESEARCH ARTICLE

microscopy (6 7) and 640000 times less than PS-RESOLFT (10) Furthermore STED andRESOLFTrequire an additional depletion step not neededin PA NL-SIM which would further expose thesample to peak intensities of 807 MWcm2 forSTED (8) 17 MWcm2 for PS-RESOLFT (10) and3 kWcm2 forWF-RESOLFT (11) Even over smallimage fields nanoscopy with focused light suchas PS-RESOLFT and STED uses intensities 105-to 1010-fold larger than that of terrestrial solarflux and is thus ill-equipped to study live-celldynamics noninvasivelyOf course despite these gains no method of

live-cell fluorescencemicroscopy including high-NA TIRF-SIM and PA NL-SIM can claim to becompletely noninvasive owing to possible photo-induced physiological changes protein over-expression andor label-induced perturbationsFor example the gradual development of curvedfilopodia and membrane ruffles after the start ofimaging are shown in Movies 5 and 6 and movieS2 These may reflect a response to the illumina-tion although we have also commonly seen suchstructures under initial conditions when imagingwith diffraction-limited TIRF (fig S46) Anothercaveat is that all the cells except BSC-1 in thiswork were transiently transfected and henceexpression levels of the target proteins were un-controlled This could affect eithermorphologiessuch as the sizes of Rab5a-labeled endosomes(Fig 4 A to C and figs S38 and S39) (51) ordynamic phenotypes such as the growth rate ofmembrane ruffles inmCherryndasha-actininndashexpressingcells (Fig 4E Movie 9 and movie S15) Althoughendogenous expression levels can be achievedwith genome editing (52) even more light orlonger exposures would be needed for cases inwhich these levels are lower than those used hereThus the biological findings described in this workshould not be considered definitive More exten-sive measurements across multiple cell lines withcareful controls and targeted perturbation experi-ments will be needed to reach conclusive insightsThe lesson is that when addressing any biolog-

ical question by means of live-cell imaging it isprudent to startwith less invasive lower-resolutionmethods such aswidefield spinning disk confocalor lattice light sheetmicroscopy andmove progres-sively only as needed to more invasive higher-resolution methods such as 3D-SIM TIRF-SIMPANL-SIM and last localizationmicroscopy Seenfrom this perspective the two extended-resolutionmethods of high-NATIRF-SIMandPANL-SIMweintroduce here fill an important gap between the100-nm limit of traditional SIM and the macro-molecular level of localizationmicroscopy Togetherthey open the door to high-resolution minimallyinvasive studies of dynamic processes includingendocytosis exocytosis signal transduction proteindiffusion vesicle trafficking viral entry cytoskeletalremodeling interactions with the extracellularmatrix and the evolution of lipid rafts

Materials and methodsOptical path of the TIRF-SIM system

The schematic of TIRF-SIM system is presentedin fig S47A The beam from a laser combiner

equipped with 405 nm (250 mW RPMC OxxiusLBX-405-300-CIR-PP) 488 nm (500mW Coher-ent SAPPHIRE 488-500) and 560 nm (1W MPBCommunications 2RU-VFL-P-1000-560-B1R) lasersis passed through an acousto-optic tunable filter(AOTF AA Quanta Tech AOTFnC-400650-TN)The beam is then expanded to a 1e2 diameter of12 mm and sent to a phase-only modulator (13)consisting of a polarizing beam splitter a achro-matic half-wave plate (HWP Bolder Vision OptikBVO AHWP3) and a ferroelectric spatial lightmodulator (SLM ForthDimensionDisplays SXGA-3DM) Light diffracted by the grating patterndisplayed on SLM passes through a polarizationrotator (15) consisting of a liquid crystal cell (LCMeadowlark SWIFT) and an achromatic quarter-wave plate (QWP Bolder Vision Optik BVOAQWP3) which rotates the linear polarizationof the diffracted light so as to maintain thes-polarization necessary to maximize the patterncontrast for all pattern orientations A mask con-sisting of a hollow barrel with slots for differentpattern orientations (15) is driven by a galvano-metric scanner (Cambridge Technology 6230HB)to filter out all diffraction orders created by thebinary and pixelated nature of the SLM exceptfor the desired plusmn1 diffraction orders These arethen imaged at the back focal plane of the ob-jective (Olympus APON 100XHOTIRF 17 NA forhigh-NATIRF-SIMOlympusUAPON100XOTIRF149 NA for PA NL-SIM at 23degC or Zeiss Plan-Apochromat 100X Oil-HI 157 NA for high-NAPA-NL-SIM at 37degC) as two spots at oppositesides of the pupil After passage through the ob-jective the two beams intersect at the interfacebetween the coverslip and the sample at an angleexceeding the critical angle for total internal re-flection An evanescent standing wave penetrat-ing ~100 nm into the sample is thereby generatedconsisting of a sinusoidal pattern of excitationintensity that is a low-pass filtered image of theSLM pattern The period orientation and rela-tive phase of this excitation pattern can befinely tuned by altering the corresponding pat-tern displayed on SLM For each orientationand phase of the applied excitation pattern theresulting fluorescence is collected by the ob-jective focused by a tube lens at an interme-diate image plane separated from excitationlight by a dichroic mirror (Chroma ZT405488560tpc_225deg) placed between two relaylenses and reimaged onto a sCMOS camera(Hamamatsu Orca Flash 40 v2 sCMOS) wherethe structured fluorescence emission pattern isrecorded

Calibration of pattern overlap forPA NL-SIM

In order to maximize the amplitudes of the non-linear harmonics for PA NL-SIM to work efficient-ly the sinusoidal patterns of 405 nm activationlight and 488 nmexcitation and deactivation lightmust be aligned to precisely overlap one anotherAs noted above these patterns at the sampleplane are created by displaying correspondingbinary grating patterns on an SLM at a corre-sponding optically conjugate plane In this case

the period ps at the specimen is related to theperiod pSLM at the SLM by

ps =Ml middot pSLM eth1THORN

where M is the demagnification factor betweenthe two conjugate planes and is dictated to bethe focal lengths of the relay lenses between thetwo planes Unfortunately chromatic aberrationleads to slightly different focal lengths for evenachromatic relay lenses for different wavelengthsof light In particular in our system M405 andM488 vary by ~2 Considering that the sinusoi-dal interference pattern is composed of hundredsof periods across our 45- by 45-mm2 field-of-view(FOV) even this 2 difference results in sub-stantial drift in the relative phases of the 405-and 488-nmexcitationpatterns across the FOV (figS48 A to C) leading to spatially variable ampli-tudes for thenonlinearharmonics and correspond-ing spatially variable errors in the resultingSIM reconstructionsA straightforward way to compensate for

chromatic aberration and achieve identical peri-ods ps405 = ps488 at the sample (fig S47B) is tointroduce a period difference DpSLM between thetwo corresponding patterns at the SLM (figS47C) In fact in order to compensate completelyand achieve well-overlapped 405- and 488-nmexcitation patterns over the whole FOV we needto measure two parameters the initial perioddifference at the sampleDpi

s frac14 Dpis488 minus Dpis405

when pSLM is the same for bothwavelengths andthe phase differenceDfis frac14 Dfis488 minus Dfis405 whenps is the same Do to so we used a sample con-sisting of a dense but submonolayer spread ofgreen fluorescent beads excitable at both 405and 488 nm and proceeded as follows

Step 1

Keeping pSLM constant we acquired five imageseach of the sample under 405- and 488-nm sinus-oidal excitation with the phase shifted by pSLM5for each image at a given wavelength We then ap-plied the structured illumination (SI) reconstruc-tion algorithm (53) to each set of five images fromwhich pis405 and pi

s488 emerged as measuredoutputs For a given period pSLM488 used at theSLM for 488-nm excitation the correspondingperiod pSLM405 needed at the SLM for 405-nmexcitation to produce the same period ps at thesample for both wavelengths is then given by

pSLM405 frac14pis488pis405

pSLM488 eth2THORN

Step 2

After adjusting pSLM405 and pSLM488 to obtainthe same period ps at the sample for both wave-lengths a constant phase offset exists betweenthe two sinusoidal illumination patterns acrossthe FOV (fig S48 D and E) We measured thephase f for each wavelength by applying thesinusoidal illumination for that wavelength andthen recorded the position xn along the modu-lation direction and intensity In for each of Nbeads scattered across the FOV We then fit the

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-7

RESEARCH | RESEARCH ARTICLE

function I(x) = Imax[1 + sin(2pxps + f)]2 to thisdata to find f (fig S48F) A phase shift Df = f488 ndashf405 was then applied the SLM pattern for the405-nm illumination so as to bring it into phasewith the 488-nm illumination at the specimen(figs S48 G to I)

Step 3

Last we confirmed that both the period and phaseof the sinusoidal illumination patterns at the twowavelengths match across the entire FOV byremeasuring the periods ps488 ps405 and thephases f488 f405 as described above and con-firming that they are identical

SLM pattern generation

We generated the sinusoidal illumination pat-terns using a binary ferroelectric SLM (Forth Di-mension Displays SXGA-3DM) because it hasthe submillisecond switching times needed toacquire the nine (TIRF-SIM) 25 (PA NL-SIM) ormore (saturated PA NL-SIM) raw images of dif-ferent phase and orientation required to recon-struct a single SIM image in as fast as 100 to400msHowever care must be taken to account for thefinite pixel size of the SLM especially consideringthat subpixel adjustment accuracy is necessary toachieve precise pattern overlap at 405 and488nmas described in the previous section The SLMpattern-generation algorithms used in previouswork (13ndash15) do not provide such subpixel accu-racy Thus in this work we developed a newalgorithm that matches the two pattern periodsto 002 precision leading to a phase error nogreater than 18deg over the 45-mm FOVIn detail a set of radial vectors An define the

desired orientations of the grating pattern at theSLM The angular orientation of this radial setrelative to the x and y axes defined by pixel rowsand columns of the SLM is chosen so that eachvector is at least 4deg away from either axis This isessential to achieve subpixel precision in the ad-justment of the period For each orientation rep-resented by An we define a vector Bn that isorthogonal to An (fig S49) Likewise for everypixel of the SLM we define a pixel vector (suchas C1 or C2 in fig S49) from the point O at theintersection of An and Bn to the pixel We thencalculate F = [(C middotB)modp]p the fraction of theperiod p by which the pixel extends beyond anintegral number of periods on the SLM For apattern with a desired off fraction D per period(D = 05 in 2D SIM) the pixel is set to 0 if F lt Dand set to 1 otherwise Last to define the pixelpatterns required for the other N ndash 1 phases ofthe illumination for a given orientation the pointO is translated along Bn in steps of pN and thisprocess is repeatedwith the new vectorC for eachpixel Unlike the pixel assignment algorithm usedpreviously for SIM (15) this approach does notrely on unit-cell repetition and therefore doesnot succumb to error accumulation over theentire span of the SLM

Lattice light sheet PA NL-SIM system

To extend PA NL-SIM to three dimensions it isessential to minimize out-of-focus fluorescence

emission that can cause the shot noise in the DCharmonic to completely overwhelm the weaksignals in the nonlinear harmonics To accom-plish this we turned to the SIM mode of latticelight sheet microscopy (42) Just as in the case of2D-SIM and for the same reasons we chooseto introduce the nonlinear harmonics throughpatterned activation of Skylan-NS The excitationobjective (Special Optics 065 NA 374 mmWD)is placed perpendicular to the detection objective(Nikon CFI Apo LWD 25XW 11 NA 2 mmWD)to confine the illumination to the proximity ofthe latterrsquos focal plane (fig S42A) The latticepattern projected on the SLM (Forth DimensionDisplays SXGA-3DM) is imaged onto the focalplane of the excitation objective after the excita-tion is first spatially filtered by an annular mask(Photo-Science) and relayed by a pair of galva-nometers (Cambridge Technology 6215H) thatphase step the pattern in the x direction and scanthe light sheet in z Also as in 2D PA NL-SIM wematch the periods and phases of the 405- and488-nm lattices to exactly match by measuringtheir excitation profiles across the FOV using fluo-rescent beads (fig S42B) and adjusting accord-ingly The fluorescence emission is collected bythe detection objective and imaged by a tube lensonto a sCMOS camera (Hamamatsu Orca Flash40 v2) A 3D image is formed by repeating thisprocess as the sample is translated through thelight sheet with a piezoelectric stage (PhysikInstrumente P-6211CD) along an axis s in theplane of the cover slip and a 3D super-resolutionNL-SIM image is reconstructed as describedbelow

Data acquisitionHigh-NA TIRF SIM

All high-NA TIRF-SIM images were acquiredwith the Olympus 17-NA objective under thephysiological conditions of 37degC and 5 CO2 Ateach time point we acquired three raw images atsuccessive phase steps of 0 13 and 23 of theillumination period We then repeated this pro-cess with the standing wave excitation patternrotated plusmn120deg with respect to the first orienta-tion for a total of nine raw images The phasestepping and pattern rotation were accomplishedby rotating or translating the binary grating pat-terndisplayedon theSLMFormulticolor imagingwe acquired nine raw images at each excitationwavelength before moving to the next and thenrepeated this series at successive time points Wecould adjust the excitationNA for eachwavelengthby changing the period of the grating pattern at theSLM This allowed us to control penetration depthof the evanescent wave (fig S8) in order to ba-lance the number of excitable fluorescent mole-cules against the background fluorescence andpossible physiological effects of the excitation

PA NL-SIM and saturated PA NL-SIM

The high refractive index immersion oil requiredfor the Olympus 17-NA objective strongly ab-sorbs 405-nm light leading to a substantial reduc-tion in the modulation depth we could achieve inthe activation pattern at this wavelength Conse-

quently forNL-SIMwe first turned to theOlympus149-NA TIRF objective and imaged at room tem-perature (23degC) with L15 medium without phenolred having 10 fetal bovine serum (Life Technol-ogies) With this objective we were able to achievehigh modulation contrast while stably and pre-cisely overlapping the 405- and 488-nm standingwaves over the whole FOV An excitation NA of144 was used for both 488- and 560-nm light inthis case leading to 62-nm resolution for PANL-SIMwhen using green-emitting FPs Recently how-ever we found that the high refractive index im-mersion oil used for the Zeiss 157-NA objectivedid not absorb 405-nm light strongly and there-fore could be used to maintain precisely over-lapped 405- and 488-nm standing waves withhigh modulation contrast at 37degC and 5 CO2The excitation NA in this case was 152 for 488-nmlight leading to 59-nm resolution for PA NL-SIMwhen using green-emitting FPsThe exposure procedure for a single phase step

inNL-SIMconsists of (i) 405-nmpatterned illumi-nation for 1 ms to activate the fluorescent mol-ecules (ii) 488-nm patterned illumination for 5 to~30 ms to read-out the activated molecules and(iii) 488-nm uniform illumination for 2 to ~10 msto read-out the remaining activated molecules andreturn the sample back to the original unactivatedstate We collected the fluorescence from bothsteps (ii) and (iii) to reconstruct the SR imageDepending on the number of modulation har-monics H of non-negligible amplitude in theimage (H = 2 for PA-NL-SIM andH = 3 or possiblymore for saturated PA NL-SIM) we repeated thissequence for 2H + 1 raw images at each of 2H +1 angular orientations equally spaced around 360degfor a total of (2H + 1)2 raw images at each NL-SIMtime point An exceptionwas saturated PA-NL-SIMfor which to reduce the acquisition time weoften used only five orientations rather thansevenIn two-color imaging combining linear TIRF-SIM

and PA NL-SIM (Fig 4) at each time point weacquired the PANL-SIM image as discussed aboveHowever we acquired the TIRF-SIM image withfive instead of three orientations (15 raw images forthe TIRF-SIM channel at every time point) inorder to match the orientations of the five-slotgalvanometer-driven barrel mask used to pickout thedesireddiffractionorders for thePANL-SIMacquisition

3D PA NL-SIM with lattice lightsheet microscopy

Here we used a hexagonal lattice having aperiod large enough to contain two harmonicsfor each of the 405-nm activation and the 488-nm excitation (42)mdashone harmonic just belowthe Abbe limit of the 065-NA excitation objec-tive and the other at twice this period Theproduct of these patterns created a fluorescenceemission pattern containing H = 4 harmonics(fig S43F) However with a single excitation ob-jective we were limited to producing this pat-tern at only one orientation Therefore at eachplane of the 3D stack we acquired 2H + 1 = 9images resulting in improved resolution (Fig 5)

aab3500-8 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

RESEARCH | RESEARCH ARTICLE

in both the lateral and axial directions of thepattern

Reconstruction of SIM images

The raw image frames with patterned excitationwere processed and reconstructed into the super-resolved images by means of a previously de-scribed algorithm (53) In brief for each patternorientation with H modulation harmonics 2H +1 raw images are collected and Fourier transformedinto 2H + 1 information components These com-ponents are assembled by initially translating eachin Fourier space by a distance equal to the am-plitudeof the illuminationpatternvectornk0wherek0 is the spatial frequency of the illumination pat-tern and n = ndashH to H The pattern vector of eachinformation component is then fine-tuned byfinding the vector that maximizes the complexcross-correlation in the overlap region betweensuccessive components The modulation ampli-tude of the harmonic and its starting phase arefound through complex linear regression In linewith previous work (28) the modulation ampli-tudes for the highest harmonics are generally toolow for this empirical approach to work well sofor these the theoretical values of their complexamplitudes are used After fine-tuning the posi-tions and complex amplitudes of the informationcomponents in the overlap regions a generalizedWiener filter is applied to this expanded transferfunction to balance the amplitudes of the variousspatial frequencies against the underlying noiseNext an apodization function is applied to min-imize ringing artifacts when the result is Fourier-transformed back to real space However ratherthan the triangle apodization A(k) = 1 ndash kkmax

normally used (53) we applied a g apodizationA(k) = 1 ndash (kkmax)

g usually with g = 04 so thatthe higher spatial frequencies are not suppressedmore than necessary Furthermore we strictly fol-lowed the azimuthally dependent support kmax(q)of the expanded OTF (figs S7 and S30) to definethe endpoint of the apodization function This pro-vides additional suppression of ringing artifactsFor the time series data we independently imple-ment this reconstructionprocess for each timepoint

Cell culture transfection stainingand fixation

BSC-1 COS-7 U2OS andmouse embryonic fibro-blast (MEF) cells (American Type Culture Collec-tion) were grown to ~60 to 80 confluency inDulbeccorsquos modified eagle medium (DMEM) withhigh glucose and no phenol red supplementedwith 15 fetal bovine serum (Life Technologies)BSC-1 cells stably expressed EGFP-CLTA Othercells were transiently transfected with an AmaxaNucleofector 96-well shuttle system (Lonza) with1 mg DNA per 400000 cells with nucleofectionsolution and a program optimized for each cellline per the manufactures instructions Beforeimaging 25-mm or 5-mm coverslips were coatedwith 10 mgml fibronection (Millipore FC010) for24 hours before plating transfected cells Imagingwas performed in DMEM with HEPES if there isno CO2 control containing no phenol red at tem-peratures specifically stated in each case

In two-color imaging of CCPs and transferrinreceptors (TfRs) by means of high-NA TIRF-SIMMEF cells expressing clathrin light chain B fusedto the C terminal of mEmerald were incubatedwith DMEM medium containing 250 mgmLTfR bound to human transferrin conjugatedwith Alexa 568 (T23365 Life Technologies) for15 minFixed cells were treated for 15 min with fixa-

tion buffer containing 4 paraformaldehyde01 gluteraldehyde in PHEM buffer (25 mMHEPES 10mMEGTA 2mMMgCl2 and 120mMPIPES in pH 73)

Tracking analysis of CCPs

For each image frame we segmented the CCPsusing a watershed algorithm written in Matlab(MathWorks 2014a) and measured their cent-roids individually Subsequently the centroidpositionwas linked between time points using u-track 21 (54) This linking operation collectedsuccessive position information for each pit overthe entire endocytic process (Fig 2E) from ini-tiation to final internalization It was then straight-forward to determine the lifetime (Fig 2A) foreach endocytic eventIn order to precisely measure the pit diameter

(Fig 2 B and C) we first measured the systemmagnification to the camera by imaging a stan-dard fine counting grid (2280-32 Ted Pella) TheSIM image of each CCP was then deconvolvedwith the equivalent PSF of the SIM system tocompensate for the broadening due to the finiteresolution of the instrument Last we measuredthe diameter of each deconvolved pit using anintensity-weighted average radius relative to thecentroid of the pit In certain cases (Fig 2A andMovie 3) pits were color-coded at each timepoint based on the time since their initiation tothe current time pointOne challenge in this analysis was how to

identify isolated pits rather than aggregates andhow to be sure that these represented true pitsrather than noise or disorganized patches ofnonassembled clathrin To accomplish this weset some conditions during the analysis such asthat a pit must start as a spot and then evolveinto a ring at at least one time point When ana-lyzing the correlation between pit lifetime andmaximum diameter we added the further con-straint of including only those pits formed afterthe first frame in order to insure that we couldaccurately measure the entire lifetimeWhenmeasuring the associations of actinwith

clathrin we first implemented the tracking al-gorithm above to obtain time-lapse CCP imagesfor each endocytic eventWe then created amaskfor each CCP identified in each frame equal tothe CCP size plus an additional boundary of onepixelWe then applied thesemasks to each frameof Lifeact data and integrated the actin fluores-cence within each CCP-derivedmask If the actinsignal integrated over the area of a given maskincreased during the final five frames of the lifeof the associated CCP it was decided that actinwas recruited to the CCP during the final stage ofendocytosis

REFERENCES AND NOTES

1 L Schermelleh R Heintzmann H Leonhardt A guide to super-resolution fluorescence microscopy J Cell Biol 190 165ndash175(2010) doi 101083jcb201002018 pmid 20643879

2 U Schnell F Dijk K A Sjollema B N GiepmansImmunolabeling artifacts and the need for live-cell imagingNat Methods 9 152ndash158 (2012) doi 101038nmeth1855pmid 22290187

3 R P Nieuwenhuizen et al Measuring image resolution inoptical nanoscopy Nat Methods 10 557ndash562 (2013)doi 101038nmeth2448 pmid 23624665

4 X Shu et al A genetically encoded tag for correlated light andelectron microscopy of intact cells tissues and organismsPLOS Biol 9 e1001041 (2011) doi 101371journalpbio1001041 pmid 21483721

5 J D Martell et al Engineered ascorbate peroxidase as agenetically encoded reporter for electron microscopy NatBiotechnol 30 1143ndash1148 (2012) doi 101038nbt2375pmid 23086203

6 H Shroff C G Galbraith J A Galbraith E Betzig Live-cellphotoactivated localization microscopy of nanoscale adhesiondynamics Nat Methods 5 417ndash423 (2008) doi 101038nmeth1202 pmid 18408726

7 S H Shim et al Super-resolution fluorescence imaging oforganelles in live cells with photoswitchable membrane probesProc Natl Acad Sci USA 109 13978ndash13983 (2012)doi 101073pnas1201882109 pmid 22891300

8 B Hein K I Willig S W Hell Stimulated emission depletion(STED) nanoscopy of a fluorescent protein-labeled organelleinside a living cell Proc Natl Acad Sci USA 10514271ndash14276 (2008) doi 101073pnas0807705105pmid 18796604

9 V Westphal et al Video-rate far-field optical nanoscopydissects synaptic vesicle movement Science 320 246ndash249(2008) doi 101126science1154228 pmid 18292304

10 T Grotjohann et al rsEGFP2 enables fast RESOLFT nanoscopyof living cells eLife 1 e00248 (2012) doi 107554eLife00248 pmid 23330067

11 A Chmyrov et al Nanoscopy with more than 100000lsquodoughnutsrsquo Nat Methods 10 737ndash740 (2013) doi 101038nmeth2556 pmid 23832150

12 Materials and methods are available as supplementarymaterials on Science Online

13 P Kner B B Chhun E R Griffis L Winoto M G GustafssonSuper-resolution video microscopy of live cells by structuredillumination Nat Methods 6 339ndash342 (2009) doi 101038nmeth1324 pmid 19404253

14 L Shao P Kner E H Rego M G Gustafsson Super-resolution 3D microscopy of live whole cells using structuredillumination Nat Methods 8 1044ndash1046 (2011) doi 101038nmeth1734 pmid 22002026

15 R Fiolka L Shao E H Rego M W DavidsonM G Gustafsson Time-lapse two-color 3D imaging of live cellswith doubled resolution using structured illumination ProcNatl Acad Sci USA 109 5311ndash5315 (2012) doi 101073pnas1119262109 pmid 22431626

16 J Riedl et al Lifeact A versatile marker to visualize F-actinNat Methods 5 605ndash607 (2008) doi 101038nmeth1220pmid 18536722

17 H T McMahon E Boucrot Molecular mechanism andphysiological functions of clathrin-mediated endocytosis NatRev Mol Cell Biol 12 517ndash533 (2011) doi 101038nrm3151pmid 21779028

18 M Ehrlich et al Endocytosis by random initiation andstabilization of clathrin-coated pits Cell 118 591ndash605 (2004)doi 101016jcell200408017 pmid 15339664

19 I Gaidarov F Santini R A Warren J H Keen Spatial controlof coated-pit dynamics in living cells Nat Cell Biol 1 1ndash7(1999) pmid 10559856

20 S Saffarian E Cocucci T Kirchhausen Distinct dynamics ofendocytic clathrin-coated pits and coated plaques PLOS Biol7 e1000191 (2009) doi 101371journalpbio1000191pmid 19809571

21 J Grove et al Flat clathrin lattices Stable features of theplasma membrane Mol Biol Cell 25 3581ndash3594 (2014)doi 101091mbcE14-06-1154 pmid 25165141

22 J Heuser Effects of cytoplasmic acidification on clathrin latticemorphology J Cell Biol 108 401ndash411 (1989) doi 101083jcb1082401 pmid 2563729

23 M Kaksonen C P Toret D G Drubin Harnessing actindynamics for clathrin-mediated endocytosis Nat Rev Mol CellBiol 7 404ndash414 (2006) doi 101038nrm1940pmid 16723976

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-9

RESEARCH | RESEARCH ARTICLE

function I(x) = Imax[1 + sin(2pxps + f)]2 to thisdata to find f (fig S48F) A phase shift Df = f488 ndashf405 was then applied the SLM pattern for the405-nm illumination so as to bring it into phasewith the 488-nm illumination at the specimen(figs S48 G to I)

Step 3

Last we confirmed that both the period and phaseof the sinusoidal illumination patterns at the twowavelengths match across the entire FOV byremeasuring the periods ps488 ps405 and thephases f488 f405 as described above and con-firming that they are identical

SLM pattern generation

We generated the sinusoidal illumination pat-terns using a binary ferroelectric SLM (Forth Di-mension Displays SXGA-3DM) because it hasthe submillisecond switching times needed toacquire the nine (TIRF-SIM) 25 (PA NL-SIM) ormore (saturated PA NL-SIM) raw images of dif-ferent phase and orientation required to recon-struct a single SIM image in as fast as 100 to400msHowever care must be taken to account for thefinite pixel size of the SLM especially consideringthat subpixel adjustment accuracy is necessary toachieve precise pattern overlap at 405 and488nmas described in the previous section The SLMpattern-generation algorithms used in previouswork (13ndash15) do not provide such subpixel accu-racy Thus in this work we developed a newalgorithm that matches the two pattern periodsto 002 precision leading to a phase error nogreater than 18deg over the 45-mm FOVIn detail a set of radial vectors An define the

desired orientations of the grating pattern at theSLM The angular orientation of this radial setrelative to the x and y axes defined by pixel rowsand columns of the SLM is chosen so that eachvector is at least 4deg away from either axis This isessential to achieve subpixel precision in the ad-justment of the period For each orientation rep-resented by An we define a vector Bn that isorthogonal to An (fig S49) Likewise for everypixel of the SLM we define a pixel vector (suchas C1 or C2 in fig S49) from the point O at theintersection of An and Bn to the pixel We thencalculate F = [(C middotB)modp]p the fraction of theperiod p by which the pixel extends beyond anintegral number of periods on the SLM For apattern with a desired off fraction D per period(D = 05 in 2D SIM) the pixel is set to 0 if F lt Dand set to 1 otherwise Last to define the pixelpatterns required for the other N ndash 1 phases ofthe illumination for a given orientation the pointO is translated along Bn in steps of pN and thisprocess is repeatedwith the new vectorC for eachpixel Unlike the pixel assignment algorithm usedpreviously for SIM (15) this approach does notrely on unit-cell repetition and therefore doesnot succumb to error accumulation over theentire span of the SLM

Lattice light sheet PA NL-SIM system

To extend PA NL-SIM to three dimensions it isessential to minimize out-of-focus fluorescence

emission that can cause the shot noise in the DCharmonic to completely overwhelm the weaksignals in the nonlinear harmonics To accom-plish this we turned to the SIM mode of latticelight sheet microscopy (42) Just as in the case of2D-SIM and for the same reasons we chooseto introduce the nonlinear harmonics throughpatterned activation of Skylan-NS The excitationobjective (Special Optics 065 NA 374 mmWD)is placed perpendicular to the detection objective(Nikon CFI Apo LWD 25XW 11 NA 2 mmWD)to confine the illumination to the proximity ofthe latterrsquos focal plane (fig S42A) The latticepattern projected on the SLM (Forth DimensionDisplays SXGA-3DM) is imaged onto the focalplane of the excitation objective after the excita-tion is first spatially filtered by an annular mask(Photo-Science) and relayed by a pair of galva-nometers (Cambridge Technology 6215H) thatphase step the pattern in the x direction and scanthe light sheet in z Also as in 2D PA NL-SIM wematch the periods and phases of the 405- and488-nm lattices to exactly match by measuringtheir excitation profiles across the FOV using fluo-rescent beads (fig S42B) and adjusting accord-ingly The fluorescence emission is collected bythe detection objective and imaged by a tube lensonto a sCMOS camera (Hamamatsu Orca Flash40 v2) A 3D image is formed by repeating thisprocess as the sample is translated through thelight sheet with a piezoelectric stage (PhysikInstrumente P-6211CD) along an axis s in theplane of the cover slip and a 3D super-resolutionNL-SIM image is reconstructed as describedbelow

Data acquisitionHigh-NA TIRF SIM

All high-NA TIRF-SIM images were acquiredwith the Olympus 17-NA objective under thephysiological conditions of 37degC and 5 CO2 Ateach time point we acquired three raw images atsuccessive phase steps of 0 13 and 23 of theillumination period We then repeated this pro-cess with the standing wave excitation patternrotated plusmn120deg with respect to the first orienta-tion for a total of nine raw images The phasestepping and pattern rotation were accomplishedby rotating or translating the binary grating pat-terndisplayedon theSLMFormulticolor imagingwe acquired nine raw images at each excitationwavelength before moving to the next and thenrepeated this series at successive time points Wecould adjust the excitationNA for eachwavelengthby changing the period of the grating pattern at theSLM This allowed us to control penetration depthof the evanescent wave (fig S8) in order to ba-lance the number of excitable fluorescent mole-cules against the background fluorescence andpossible physiological effects of the excitation

PA NL-SIM and saturated PA NL-SIM

The high refractive index immersion oil requiredfor the Olympus 17-NA objective strongly ab-sorbs 405-nm light leading to a substantial reduc-tion in the modulation depth we could achieve inthe activation pattern at this wavelength Conse-

quently forNL-SIMwe first turned to theOlympus149-NA TIRF objective and imaged at room tem-perature (23degC) with L15 medium without phenolred having 10 fetal bovine serum (Life Technol-ogies) With this objective we were able to achievehigh modulation contrast while stably and pre-cisely overlapping the 405- and 488-nm standingwaves over the whole FOV An excitation NA of144 was used for both 488- and 560-nm light inthis case leading to 62-nm resolution for PANL-SIMwhen using green-emitting FPs Recently how-ever we found that the high refractive index im-mersion oil used for the Zeiss 157-NA objectivedid not absorb 405-nm light strongly and there-fore could be used to maintain precisely over-lapped 405- and 488-nm standing waves withhigh modulation contrast at 37degC and 5 CO2The excitation NA in this case was 152 for 488-nmlight leading to 59-nm resolution for PA NL-SIMwhen using green-emitting FPsThe exposure procedure for a single phase step

inNL-SIMconsists of (i) 405-nmpatterned illumi-nation for 1 ms to activate the fluorescent mol-ecules (ii) 488-nm patterned illumination for 5 to~30 ms to read-out the activated molecules and(iii) 488-nm uniform illumination for 2 to ~10 msto read-out the remaining activated molecules andreturn the sample back to the original unactivatedstate We collected the fluorescence from bothsteps (ii) and (iii) to reconstruct the SR imageDepending on the number of modulation har-monics H of non-negligible amplitude in theimage (H = 2 for PA-NL-SIM andH = 3 or possiblymore for saturated PA NL-SIM) we repeated thissequence for 2H + 1 raw images at each of 2H +1 angular orientations equally spaced around 360degfor a total of (2H + 1)2 raw images at each NL-SIMtime point An exceptionwas saturated PA-NL-SIMfor which to reduce the acquisition time weoften used only five orientations rather thansevenIn two-color imaging combining linear TIRF-SIM

and PA NL-SIM (Fig 4) at each time point weacquired the PANL-SIM image as discussed aboveHowever we acquired the TIRF-SIM image withfive instead of three orientations (15 raw images forthe TIRF-SIM channel at every time point) inorder to match the orientations of the five-slotgalvanometer-driven barrel mask used to pickout thedesireddiffractionorders for thePANL-SIMacquisition

3D PA NL-SIM with lattice lightsheet microscopy

Here we used a hexagonal lattice having aperiod large enough to contain two harmonicsfor each of the 405-nm activation and the 488-nm excitation (42)mdashone harmonic just belowthe Abbe limit of the 065-NA excitation objec-tive and the other at twice this period Theproduct of these patterns created a fluorescenceemission pattern containing H = 4 harmonics(fig S43F) However with a single excitation ob-jective we were limited to producing this pat-tern at only one orientation Therefore at eachplane of the 3D stack we acquired 2H + 1 = 9images resulting in improved resolution (Fig 5)

aab3500-8 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

RESEARCH | RESEARCH ARTICLE

in both the lateral and axial directions of thepattern

Reconstruction of SIM images

The raw image frames with patterned excitationwere processed and reconstructed into the super-resolved images by means of a previously de-scribed algorithm (53) In brief for each patternorientation with H modulation harmonics 2H +1 raw images are collected and Fourier transformedinto 2H + 1 information components These com-ponents are assembled by initially translating eachin Fourier space by a distance equal to the am-plitudeof the illuminationpatternvectornk0wherek0 is the spatial frequency of the illumination pat-tern and n = ndashH to H The pattern vector of eachinformation component is then fine-tuned byfinding the vector that maximizes the complexcross-correlation in the overlap region betweensuccessive components The modulation ampli-tude of the harmonic and its starting phase arefound through complex linear regression In linewith previous work (28) the modulation ampli-tudes for the highest harmonics are generally toolow for this empirical approach to work well sofor these the theoretical values of their complexamplitudes are used After fine-tuning the posi-tions and complex amplitudes of the informationcomponents in the overlap regions a generalizedWiener filter is applied to this expanded transferfunction to balance the amplitudes of the variousspatial frequencies against the underlying noiseNext an apodization function is applied to min-imize ringing artifacts when the result is Fourier-transformed back to real space However ratherthan the triangle apodization A(k) = 1 ndash kkmax

normally used (53) we applied a g apodizationA(k) = 1 ndash (kkmax)

g usually with g = 04 so thatthe higher spatial frequencies are not suppressedmore than necessary Furthermore we strictly fol-lowed the azimuthally dependent support kmax(q)of the expanded OTF (figs S7 and S30) to definethe endpoint of the apodization function This pro-vides additional suppression of ringing artifactsFor the time series data we independently imple-ment this reconstructionprocess for each timepoint

Cell culture transfection stainingand fixation

BSC-1 COS-7 U2OS andmouse embryonic fibro-blast (MEF) cells (American Type Culture Collec-tion) were grown to ~60 to 80 confluency inDulbeccorsquos modified eagle medium (DMEM) withhigh glucose and no phenol red supplementedwith 15 fetal bovine serum (Life Technologies)BSC-1 cells stably expressed EGFP-CLTA Othercells were transiently transfected with an AmaxaNucleofector 96-well shuttle system (Lonza) with1 mg DNA per 400000 cells with nucleofectionsolution and a program optimized for each cellline per the manufactures instructions Beforeimaging 25-mm or 5-mm coverslips were coatedwith 10 mgml fibronection (Millipore FC010) for24 hours before plating transfected cells Imagingwas performed in DMEM with HEPES if there isno CO2 control containing no phenol red at tem-peratures specifically stated in each case

In two-color imaging of CCPs and transferrinreceptors (TfRs) by means of high-NA TIRF-SIMMEF cells expressing clathrin light chain B fusedto the C terminal of mEmerald were incubatedwith DMEM medium containing 250 mgmLTfR bound to human transferrin conjugatedwith Alexa 568 (T23365 Life Technologies) for15 minFixed cells were treated for 15 min with fixa-

tion buffer containing 4 paraformaldehyde01 gluteraldehyde in PHEM buffer (25 mMHEPES 10mMEGTA 2mMMgCl2 and 120mMPIPES in pH 73)

Tracking analysis of CCPs

For each image frame we segmented the CCPsusing a watershed algorithm written in Matlab(MathWorks 2014a) and measured their cent-roids individually Subsequently the centroidpositionwas linked between time points using u-track 21 (54) This linking operation collectedsuccessive position information for each pit overthe entire endocytic process (Fig 2E) from ini-tiation to final internalization It was then straight-forward to determine the lifetime (Fig 2A) foreach endocytic eventIn order to precisely measure the pit diameter

(Fig 2 B and C) we first measured the systemmagnification to the camera by imaging a stan-dard fine counting grid (2280-32 Ted Pella) TheSIM image of each CCP was then deconvolvedwith the equivalent PSF of the SIM system tocompensate for the broadening due to the finiteresolution of the instrument Last we measuredthe diameter of each deconvolved pit using anintensity-weighted average radius relative to thecentroid of the pit In certain cases (Fig 2A andMovie 3) pits were color-coded at each timepoint based on the time since their initiation tothe current time pointOne challenge in this analysis was how to

identify isolated pits rather than aggregates andhow to be sure that these represented true pitsrather than noise or disorganized patches ofnonassembled clathrin To accomplish this weset some conditions during the analysis such asthat a pit must start as a spot and then evolveinto a ring at at least one time point When ana-lyzing the correlation between pit lifetime andmaximum diameter we added the further con-straint of including only those pits formed afterthe first frame in order to insure that we couldaccurately measure the entire lifetimeWhenmeasuring the associations of actinwith

clathrin we first implemented the tracking al-gorithm above to obtain time-lapse CCP imagesfor each endocytic eventWe then created amaskfor each CCP identified in each frame equal tothe CCP size plus an additional boundary of onepixelWe then applied thesemasks to each frameof Lifeact data and integrated the actin fluores-cence within each CCP-derivedmask If the actinsignal integrated over the area of a given maskincreased during the final five frames of the lifeof the associated CCP it was decided that actinwas recruited to the CCP during the final stage ofendocytosis

REFERENCES AND NOTES

1 L Schermelleh R Heintzmann H Leonhardt A guide to super-resolution fluorescence microscopy J Cell Biol 190 165ndash175(2010) doi 101083jcb201002018 pmid 20643879

2 U Schnell F Dijk K A Sjollema B N GiepmansImmunolabeling artifacts and the need for live-cell imagingNat Methods 9 152ndash158 (2012) doi 101038nmeth1855pmid 22290187

3 R P Nieuwenhuizen et al Measuring image resolution inoptical nanoscopy Nat Methods 10 557ndash562 (2013)doi 101038nmeth2448 pmid 23624665

4 X Shu et al A genetically encoded tag for correlated light andelectron microscopy of intact cells tissues and organismsPLOS Biol 9 e1001041 (2011) doi 101371journalpbio1001041 pmid 21483721

5 J D Martell et al Engineered ascorbate peroxidase as agenetically encoded reporter for electron microscopy NatBiotechnol 30 1143ndash1148 (2012) doi 101038nbt2375pmid 23086203

6 H Shroff C G Galbraith J A Galbraith E Betzig Live-cellphotoactivated localization microscopy of nanoscale adhesiondynamics Nat Methods 5 417ndash423 (2008) doi 101038nmeth1202 pmid 18408726

7 S H Shim et al Super-resolution fluorescence imaging oforganelles in live cells with photoswitchable membrane probesProc Natl Acad Sci USA 109 13978ndash13983 (2012)doi 101073pnas1201882109 pmid 22891300

8 B Hein K I Willig S W Hell Stimulated emission depletion(STED) nanoscopy of a fluorescent protein-labeled organelleinside a living cell Proc Natl Acad Sci USA 10514271ndash14276 (2008) doi 101073pnas0807705105pmid 18796604

9 V Westphal et al Video-rate far-field optical nanoscopydissects synaptic vesicle movement Science 320 246ndash249(2008) doi 101126science1154228 pmid 18292304

10 T Grotjohann et al rsEGFP2 enables fast RESOLFT nanoscopyof living cells eLife 1 e00248 (2012) doi 107554eLife00248 pmid 23330067

11 A Chmyrov et al Nanoscopy with more than 100000lsquodoughnutsrsquo Nat Methods 10 737ndash740 (2013) doi 101038nmeth2556 pmid 23832150

12 Materials and methods are available as supplementarymaterials on Science Online

13 P Kner B B Chhun E R Griffis L Winoto M G GustafssonSuper-resolution video microscopy of live cells by structuredillumination Nat Methods 6 339ndash342 (2009) doi 101038nmeth1324 pmid 19404253

14 L Shao P Kner E H Rego M G Gustafsson Super-resolution 3D microscopy of live whole cells using structuredillumination Nat Methods 8 1044ndash1046 (2011) doi 101038nmeth1734 pmid 22002026

15 R Fiolka L Shao E H Rego M W DavidsonM G Gustafsson Time-lapse two-color 3D imaging of live cellswith doubled resolution using structured illumination ProcNatl Acad Sci USA 109 5311ndash5315 (2012) doi 101073pnas1119262109 pmid 22431626

16 J Riedl et al Lifeact A versatile marker to visualize F-actinNat Methods 5 605ndash607 (2008) doi 101038nmeth1220pmid 18536722

17 H T McMahon E Boucrot Molecular mechanism andphysiological functions of clathrin-mediated endocytosis NatRev Mol Cell Biol 12 517ndash533 (2011) doi 101038nrm3151pmid 21779028

18 M Ehrlich et al Endocytosis by random initiation andstabilization of clathrin-coated pits Cell 118 591ndash605 (2004)doi 101016jcell200408017 pmid 15339664

19 I Gaidarov F Santini R A Warren J H Keen Spatial controlof coated-pit dynamics in living cells Nat Cell Biol 1 1ndash7(1999) pmid 10559856

20 S Saffarian E Cocucci T Kirchhausen Distinct dynamics ofendocytic clathrin-coated pits and coated plaques PLOS Biol7 e1000191 (2009) doi 101371journalpbio1000191pmid 19809571

21 J Grove et al Flat clathrin lattices Stable features of theplasma membrane Mol Biol Cell 25 3581ndash3594 (2014)doi 101091mbcE14-06-1154 pmid 25165141

22 J Heuser Effects of cytoplasmic acidification on clathrin latticemorphology J Cell Biol 108 401ndash411 (1989) doi 101083jcb1082401 pmid 2563729

23 M Kaksonen C P Toret D G Drubin Harnessing actindynamics for clathrin-mediated endocytosis Nat Rev Mol CellBiol 7 404ndash414 (2006) doi 101038nrm1940pmid 16723976

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-9

RESEARCH | RESEARCH ARTICLE

in both the lateral and axial directions of thepattern

Reconstruction of SIM images

The raw image frames with patterned excitationwere processed and reconstructed into the super-resolved images by means of a previously de-scribed algorithm (53) In brief for each patternorientation with H modulation harmonics 2H +1 raw images are collected and Fourier transformedinto 2H + 1 information components These com-ponents are assembled by initially translating eachin Fourier space by a distance equal to the am-plitudeof the illuminationpatternvectornk0wherek0 is the spatial frequency of the illumination pat-tern and n = ndashH to H The pattern vector of eachinformation component is then fine-tuned byfinding the vector that maximizes the complexcross-correlation in the overlap region betweensuccessive components The modulation ampli-tude of the harmonic and its starting phase arefound through complex linear regression In linewith previous work (28) the modulation ampli-tudes for the highest harmonics are generally toolow for this empirical approach to work well sofor these the theoretical values of their complexamplitudes are used After fine-tuning the posi-tions and complex amplitudes of the informationcomponents in the overlap regions a generalizedWiener filter is applied to this expanded transferfunction to balance the amplitudes of the variousspatial frequencies against the underlying noiseNext an apodization function is applied to min-imize ringing artifacts when the result is Fourier-transformed back to real space However ratherthan the triangle apodization A(k) = 1 ndash kkmax

normally used (53) we applied a g apodizationA(k) = 1 ndash (kkmax)

g usually with g = 04 so thatthe higher spatial frequencies are not suppressedmore than necessary Furthermore we strictly fol-lowed the azimuthally dependent support kmax(q)of the expanded OTF (figs S7 and S30) to definethe endpoint of the apodization function This pro-vides additional suppression of ringing artifactsFor the time series data we independently imple-ment this reconstructionprocess for each timepoint

Cell culture transfection stainingand fixation

BSC-1 COS-7 U2OS andmouse embryonic fibro-blast (MEF) cells (American Type Culture Collec-tion) were grown to ~60 to 80 confluency inDulbeccorsquos modified eagle medium (DMEM) withhigh glucose and no phenol red supplementedwith 15 fetal bovine serum (Life Technologies)BSC-1 cells stably expressed EGFP-CLTA Othercells were transiently transfected with an AmaxaNucleofector 96-well shuttle system (Lonza) with1 mg DNA per 400000 cells with nucleofectionsolution and a program optimized for each cellline per the manufactures instructions Beforeimaging 25-mm or 5-mm coverslips were coatedwith 10 mgml fibronection (Millipore FC010) for24 hours before plating transfected cells Imagingwas performed in DMEM with HEPES if there isno CO2 control containing no phenol red at tem-peratures specifically stated in each case

In two-color imaging of CCPs and transferrinreceptors (TfRs) by means of high-NA TIRF-SIMMEF cells expressing clathrin light chain B fusedto the C terminal of mEmerald were incubatedwith DMEM medium containing 250 mgmLTfR bound to human transferrin conjugatedwith Alexa 568 (T23365 Life Technologies) for15 minFixed cells were treated for 15 min with fixa-

tion buffer containing 4 paraformaldehyde01 gluteraldehyde in PHEM buffer (25 mMHEPES 10mMEGTA 2mMMgCl2 and 120mMPIPES in pH 73)

Tracking analysis of CCPs

For each image frame we segmented the CCPsusing a watershed algorithm written in Matlab(MathWorks 2014a) and measured their cent-roids individually Subsequently the centroidpositionwas linked between time points using u-track 21 (54) This linking operation collectedsuccessive position information for each pit overthe entire endocytic process (Fig 2E) from ini-tiation to final internalization It was then straight-forward to determine the lifetime (Fig 2A) foreach endocytic eventIn order to precisely measure the pit diameter

(Fig 2 B and C) we first measured the systemmagnification to the camera by imaging a stan-dard fine counting grid (2280-32 Ted Pella) TheSIM image of each CCP was then deconvolvedwith the equivalent PSF of the SIM system tocompensate for the broadening due to the finiteresolution of the instrument Last we measuredthe diameter of each deconvolved pit using anintensity-weighted average radius relative to thecentroid of the pit In certain cases (Fig 2A andMovie 3) pits were color-coded at each timepoint based on the time since their initiation tothe current time pointOne challenge in this analysis was how to

identify isolated pits rather than aggregates andhow to be sure that these represented true pitsrather than noise or disorganized patches ofnonassembled clathrin To accomplish this weset some conditions during the analysis such asthat a pit must start as a spot and then evolveinto a ring at at least one time point When ana-lyzing the correlation between pit lifetime andmaximum diameter we added the further con-straint of including only those pits formed afterthe first frame in order to insure that we couldaccurately measure the entire lifetimeWhenmeasuring the associations of actinwith

clathrin we first implemented the tracking al-gorithm above to obtain time-lapse CCP imagesfor each endocytic eventWe then created amaskfor each CCP identified in each frame equal tothe CCP size plus an additional boundary of onepixelWe then applied thesemasks to each frameof Lifeact data and integrated the actin fluores-cence within each CCP-derivedmask If the actinsignal integrated over the area of a given maskincreased during the final five frames of the lifeof the associated CCP it was decided that actinwas recruited to the CCP during the final stage ofendocytosis

REFERENCES AND NOTES

1 L Schermelleh R Heintzmann H Leonhardt A guide to super-resolution fluorescence microscopy J Cell Biol 190 165ndash175(2010) doi 101083jcb201002018 pmid 20643879

2 U Schnell F Dijk K A Sjollema B N GiepmansImmunolabeling artifacts and the need for live-cell imagingNat Methods 9 152ndash158 (2012) doi 101038nmeth1855pmid 22290187

3 R P Nieuwenhuizen et al Measuring image resolution inoptical nanoscopy Nat Methods 10 557ndash562 (2013)doi 101038nmeth2448 pmid 23624665

4 X Shu et al A genetically encoded tag for correlated light andelectron microscopy of intact cells tissues and organismsPLOS Biol 9 e1001041 (2011) doi 101371journalpbio1001041 pmid 21483721

5 J D Martell et al Engineered ascorbate peroxidase as agenetically encoded reporter for electron microscopy NatBiotechnol 30 1143ndash1148 (2012) doi 101038nbt2375pmid 23086203

6 H Shroff C G Galbraith J A Galbraith E Betzig Live-cellphotoactivated localization microscopy of nanoscale adhesiondynamics Nat Methods 5 417ndash423 (2008) doi 101038nmeth1202 pmid 18408726

7 S H Shim et al Super-resolution fluorescence imaging oforganelles in live cells with photoswitchable membrane probesProc Natl Acad Sci USA 109 13978ndash13983 (2012)doi 101073pnas1201882109 pmid 22891300

8 B Hein K I Willig S W Hell Stimulated emission depletion(STED) nanoscopy of a fluorescent protein-labeled organelleinside a living cell Proc Natl Acad Sci USA 10514271ndash14276 (2008) doi 101073pnas0807705105pmid 18796604

9 V Westphal et al Video-rate far-field optical nanoscopydissects synaptic vesicle movement Science 320 246ndash249(2008) doi 101126science1154228 pmid 18292304

10 T Grotjohann et al rsEGFP2 enables fast RESOLFT nanoscopyof living cells eLife 1 e00248 (2012) doi 107554eLife00248 pmid 23330067

11 A Chmyrov et al Nanoscopy with more than 100000lsquodoughnutsrsquo Nat Methods 10 737ndash740 (2013) doi 101038nmeth2556 pmid 23832150

12 Materials and methods are available as supplementarymaterials on Science Online

13 P Kner B B Chhun E R Griffis L Winoto M G GustafssonSuper-resolution video microscopy of live cells by structuredillumination Nat Methods 6 339ndash342 (2009) doi 101038nmeth1324 pmid 19404253

14 L Shao P Kner E H Rego M G Gustafsson Super-resolution 3D microscopy of live whole cells using structuredillumination Nat Methods 8 1044ndash1046 (2011) doi 101038nmeth1734 pmid 22002026

15 R Fiolka L Shao E H Rego M W DavidsonM G Gustafsson Time-lapse two-color 3D imaging of live cellswith doubled resolution using structured illumination ProcNatl Acad Sci USA 109 5311ndash5315 (2012) doi 101073pnas1119262109 pmid 22431626

16 J Riedl et al Lifeact A versatile marker to visualize F-actinNat Methods 5 605ndash607 (2008) doi 101038nmeth1220pmid 18536722

17 H T McMahon E Boucrot Molecular mechanism andphysiological functions of clathrin-mediated endocytosis NatRev Mol Cell Biol 12 517ndash533 (2011) doi 101038nrm3151pmid 21779028

18 M Ehrlich et al Endocytosis by random initiation andstabilization of clathrin-coated pits Cell 118 591ndash605 (2004)doi 101016jcell200408017 pmid 15339664

19 I Gaidarov F Santini R A Warren J H Keen Spatial controlof coated-pit dynamics in living cells Nat Cell Biol 1 1ndash7(1999) pmid 10559856

20 S Saffarian E Cocucci T Kirchhausen Distinct dynamics ofendocytic clathrin-coated pits and coated plaques PLOS Biol7 e1000191 (2009) doi 101371journalpbio1000191pmid 19809571

21 J Grove et al Flat clathrin lattices Stable features of theplasma membrane Mol Biol Cell 25 3581ndash3594 (2014)doi 101091mbcE14-06-1154 pmid 25165141

22 J Heuser Effects of cytoplasmic acidification on clathrin latticemorphology J Cell Biol 108 401ndash411 (1989) doi 101083jcb1082401 pmid 2563729

23 M Kaksonen C P Toret D G Drubin Harnessing actindynamics for clathrin-mediated endocytosis Nat Rev Mol CellBiol 7 404ndash414 (2006) doi 101038nrm1940pmid 16723976

SCIENCE sciencemagorg 28 AUGUST 2015 bull VOL 349 ISSUE 6251 aab3500-9

RESEARCH | RESEARCH ARTICLE

24 D K Cureton R H Massol S Saffarian T L KirchhausenS P Whelan Vesicular stomatitis virus enters cells throughvesicles incompletely coated with clathrin that depend uponactin for internalization PLOS Pathog 5 e1000394 (2009)doi 101371journalppat1000394 pmid 19390604

25 S Boulant C Kural J C Zeeh F Ubelmann T KirchhausenActin dynamics counteract membrane tension during clathrin-mediated endocytosis Nat Cell Biol 13 1124ndash1131 (2011)doi 101038ncb2307 pmid 21841790

26 A I Shevchuk et al An alternative mechanism of clathrin-coated pitclosure revealed by ion conductance microscopy J Cell Biol 197499ndash508 (2012) doi 101083jcb201109130 pmid 22564416

27 R Heintzmann T M Jovin C Cremer Saturated patternedexcitation microscopymdasha concept for optical resolutionimprovement J Opt Soc Am A Opt Image Sci Vis 191599ndash1609 (2002) doi 101364JOSAA19001599pmid 12152701

28 M G Gustafsson Nonlinear structured-illuminationmicroscopy Wide-field fluorescence imaging with theoreticallyunlimited resolution Proc Natl Acad Sci USA 10213081ndash13086 (2005) doi 101073pnas0406877102pmid 16141335

29 E H Rego et al Nonlinear structured-illumination microscopywith a photoswitchable protein reveals cellular structures at50-nm resolution Proc Natl Acad Sci USA 109 E135ndashE143(2012) doi 101073pnas1107547108 pmid 22160683

30 X Zhang et al Development of a reversibly switchablefluorescent protein for super-resolution optical fluctuationimaging (SOFI) ACS Nano 9 2659ndash2667 (2015) doi 101021nn5064387 pmid 25695314

31 K Xu H P Babcock X Zhuang Dual-objective STORM revealsthree-dimensional filament organization in the actincytoskeleton Nat Methods 9 185ndash188 (2012) doi 101038nmeth1841 pmid 22231642

32 R G Parton K Simons The multiple faces of caveolae NatRev Mol Cell Biol 8 185ndash194 (2007) doi 101038nrm2122pmid 17318224

33 F Lavoie-Cardinal et al Two-color RESOLFT nanoscopy withgreen and red fluorescent photochromic proteinsChemPhysChem 15 655ndash663 (2014) doi 101002cphc201301016 pmid 24449030

34 R G Parton M Hanzal-Bayer J F Hancock Biogenesis ofcaveolae A structural model for caveolin-induced domainformation J Cell Sci 119 787ndash796 (2006) doi 101242jcs02853 pmid 16495479

35 E Boucrot M T Howes T Kirchhausen R G PartonRedistribution of caveolae during mitosis J Cell Sci 1241965ndash1972 (2011) doi 101242jcs076570 pmid 21625007

36 M Jovic M Sharma J Rahajeng S Caplan The early endosomeA busy sorting station for proteins at the crossroads HistolHistopathol 25 99ndash112 (2010) pmid 19924646

37 J Tooze M Hollinshead In AtT20 and HeLa cells brefeldin Ainduces the fusion of tubular endosomes and changes theirdistribution and some of their endocytic properties J Cell Biol118 813ndash830 (1992) doi 101083jcb1184813pmid 1500425

38 E Nielsen F Severin J M Backer A A Hyman M ZerialRab5 regulates motility of early endosomes on microtubulesNat Cell Biol 1 376ndash382 (1999) doi 10103814075pmid 10559966

39 C A Otey O Carpen Alpha-actinin revisited A fresh look at anold player Cell Motil Cytoskeleton 58 104ndash111 (2004)doi 101002cm20007 pmid 15083532

40 C K Choi et al Actin and alpha-actinin orchestratethe assembly and maturation of nascent adhesions ina myosin II motor-independent manner Nat Cell Biol10 1039ndash1050 (2008) doi 101038ncb1763pmid 19160484

41 M Mavrakis et al Septins promote F-actin ring formation bycrosslinking actin filaments into curved bundles Nat Cell Biol16 322ndash334 (2014) doi 101038ncb2921 pmid 24633326

42 B C Chen et al Lattice light-sheet microscopy Imagingmolecules to embryos at high spatiotemporal resolutionScience 346 1257998 (2014) doi 101126science1257998pmid 25342811

43 A Legesse-Miller R H Massol T Kirchhausen Constrictionand Dnm1p recruitment are distinct processes in mitochondrialfission Mol Biol Cell 14 1953ndash1963 (2003) doi 101091mbcE02-10-0657 pmid 12802067

44 J R Friedman et al ER tubules mark sites of mitochondrialdivision Science 334 358ndash362 (2011) doi 101126science1207385 pmid 21885730

45 B Huang S A Jones B Brandenburg X Zhuang Whole-cell3D STORM reveals interactions between cellular structureswith nanometer-scale resolution Nat Methods 5 1047ndash1052(2008) doi 101038nmeth1274 pmid 19029906

46 J F Presley et al ER-to-Golgi transport visualized in livingcells Nature 389 81ndash85 (1997) doi 10103838891pmid 9288971

47 K Hirschberg et al Kinetic analysis of secretory protein trafficand characterization of golgi to plasma membrane transportintermediates in living cells J Cell Biol 143 1485ndash1503(1998) doi 101083jcb14361485 pmid 9852146

48 C J R Sheppard Super-resolution in confocal imaging Optik(Stuttg) 80 53 (1988)

49 C B Muumlller J Enderlein Image scanning microscopy PhysRev Lett 104 198101 (2010) doi 101103PhysRevLett104198101 pmid 20867000

50 A G York et al Resolution doubling in live multicellularorganisms via multifocal structured illumination microscopyNat Methods 9 749ndash754 (2012) doi 101038nmeth2025pmid 22581372

51 R L Roberts et al Endosome fusion in living cellsoverexpressing GFP-rab5 J Cell Sci 112 3667ndash3675 (1999)pmid 10523503

52 J D Sander J K Joung CRISPR-Cas systems for editingregulating and targeting genomes Nat Biotechnol 32347ndash355(2014) doi 101038nbt2842 pmid 24584096

53 M G L Gustafsson et al Three-dimensional resolutiondoubling in wide-field fluorescence microscopy by structuredillumination Biophys J 94 4957ndash4970(2008) doi 101529biophysj107120345 pmid 18326650

54 K Jaqaman et al Robust single-particle tracking in live-celltime-lapse sequences Nat Methods 5 695ndash702 (2008)doi 101038nmeth1237 pmid 18641657

ACKNOWLEDGMENTS

We thank the Shared Resource teams at Janelia for their skill anddedication in specimen handling and preparation and theInstrument Design and Fabrication team for their manufacturingexpertise DL LS B-CC and EB are funded by the HowardHughes Medical Institute (HHMI) XZ MZ and PX are funded bythe National Basic Research Program (973 Program) of China(2013CB910103) the National Natural Science Foundation of China(31370851) and the Beijing Natural Science FoundationChina (7131011) MP and TK were funded in part by NIH grantGM-075252 Skylan-NS is available from PX upon executionof a materials transfer agreement with the Institute of BiophysicsOther fluorescent protein constructs used in this work are from theMichael Davidson Collection and are available along with sequenceinformation from Addgene (wwwaddgeneorgfluorescent-proteinsdavidson) Researchers can apply to access themicroscope as visitors through the Advanced Imaging Center atJanelia (wwwjaneliaorgopen-scienceadvanced-imaging-center)Technical information for the construction of a copy of themicroscope is available to nonprofit entities upon execution of ano-cost Research License with HHMI Nonlinear SIM with patternedactivation in two or three dimensions as described here is coveredwithin US provisional patent application 62057220 filed by EBand DL and assigned to HHMI

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3496251aab3500supplDC1Supplementary TextFigs S1 to S55Table S1 and S2References (55ndash68)Movies S1 to S18

15 April 2015 accepted 20 July 2015101126scienceaab3500

aab3500-10 28 AUGUST 2015 bull VOL 349 ISSUE 6251 sciencemagorg SCIENCE

RESEARCH | RESEARCH ARTICLE

DOI 101126scienceaab3500 (2015)349 Science

et alDong Licytoskeletal dynamicsExtended-resolution structured illumination imaging of endocytic and

This copy is for your personal non-commercial use only

clicking herecolleagues clients or customers by you can order high-quality copies for yourIf you wish to distribute this article to others

herefollowing the guidelines

can be obtained byPermission to republish or repurpose articles or portions of articles

) September 7 2015 wwwsciencemagorg (this information is current as of

The following resources related to this article are available online at

httpwwwsciencemagorgcontent3496251aab3500fullhtmlversion of this article at

including high-resolution figures can be found in the onlineUpdated information and services

httpwwwsciencemagorgcontentsuppl201508263496251aab3500DC1html can be found at Supporting Online Material

httpwwwsciencemagorgcontent3496251aab3500fullhtmlrelatedfound at

can berelated to this article A list of selected additional articles on the Science Web sites

httpwwwsciencemagorgcontent3496251aab3500fullhtmlref-list-1 20 of which can be accessed freecites 62 articlesThis article

httpwwwsciencemagorgcgicollectioncell_biolCell Biology

subject collectionsThis article appears in the following

registered trademark of AAAS is aScience2015 by the American Association for the Advancement of Science all rights reserved The title

CopyrightAmerican Association for the Advancement of Science 1200 New York Avenue NW Washington DC 20005 (print ISSN 0036-8075 online ISSN 1095-9203) is published weekly except the last week in December by theScience

on

Sep

tem

ber

8 2

015

ww

ws

cien

cem

ago

rgD

ownl

oade

d fr

om