RESEARCH ARTICLE SUMMARY◥

INNATE IMMUNITY

Phosphorylation of innate immuneadaptor proteins MAVS, STING, andTRIF induces IRF3 activationSiqi Liu, Xin Cai, Jiaxi Wu, Qian Cong, Xiang Chen, Tuo Li, Fenghe Du,Junyao Ren, You-Tong Wu, Nick V. Grishin, Zhijian J. Chen*

INTRODUCTION: Sensing of pathogenicmi-crobes and tissue damage by the innate im-mune system triggers immune cells to secretecytokines that promote host defense. ViralRNA, cytosolic DNA, and the bacterial cell wallcomponent lipopolysaccharide activate sig-naling cascades through a number of patternrecognition receptor (PRR)–adaptor proteinpairs, includingRIG-I–MAVS, cGAS-STING, andTLR3/4-TRIF (TLR3/4, Toll-like receptors 3 and4). Activation of these signalingmodules resultsin the production of type I interferons (IFNs),a family of cytokines that are essential forhost protection. The adaptor proteins MAVS,STING, and TRIF each activate the downstreamprotein kinase TBK1, which then phospho-rylates the transcription factor interferon regu-latory factor 3 (IRF3), which drives type I IFNproduction. Although much progress has been

made in our understanding of PRR and adap-tor protein activation, themechanism by whichthe adaptor proteins activate TBK1 and IRF3remains unclear.

RATIONALE: Other signaling pathways be-sides the RIG-I–MAVS, cGAS-STING, andTLR3/4-TRIF pathways activate TBK1. How-ever, IRF3 phosphorylation by TBK1 is ob-served only in the IFN-producing pathwaysthat use MAVS, STING, or TRIF as the adap-tor protein. The discrepant activation ofTBK1 and IRF3 implies the existence of akinase-substrate specification mechanismexclusive to the IFN-producing pathways. Spec-ification of TBK1-mediated IRF3 activationis essential for the tight regulation of IFNproduction, which would otherwise lead toautoimmune diseases.

RESULTS: Using biochemical andmouse cell–and human cell–based assays, we found thatboth MAVS and STING interacted with IRF3in a phosphorylation-dependent manner. Weshow that both MAVS and STING are phos-phorylated in response to stimulation at theirrespective C-terminal consensus motif, pLxIS(p, hydrophilic residue; x, any residue; S, phos-phorylation site). This phosphorylation eventthen recruits IRF3 to the active adaptor proteinand is essential for IRF3 activation. Pointmuta-tions that impair the phosphorylation ofMAVSor STING at their consensus motif abrogatedIRF3 binding and subsequent IFN induction.We found that MAVS is phosphorylated by

the kinases TBK1 and IKK, whereas STING isphosphorylated by TBK1. PhosphorylatedMAVS and STING subsequently bind to con-served, positively charged surfaces of IRF3,

thereby recruiting IRF3 forits phosphorylation andactivation by TBK1. Pointmutations at IRF3’s posi-tively charged surfaces ab-rogated IRF3 binding toMAVS and STING and

subsequent IRF3 phosphorylation and ac-tivation. We further show that TRIF-mediatedactivation of IRF3 depends on TRIF phospho-rylation at the pLxIS motif commonly found inMAVS, STING, and IRF3. These results revealthat phosphorylation of innate immune adap-tor proteins is an essential and conservedmech-anism that selectively recruits IRF3 to activatetype I IFN production.

CONCLUSION:Weuncovered a general mech-anism of IRF3 activation by the innate im-mune adaptor proteins MAVS, STING, andTRIF, which functions in three distinct pat-tern recognition pathways. Following its ac-tivation, each adaptor protein recruits andactivates downstreamkinaseTBK1,which phos-phorylates the cognate upstream adaptor pro-tein at a consensusmotif. PhosphorylatedMAVS,STING, or TRIF in turn recruits IRF3 throughits conserved, positively chargedphospho-bindingdomain, allowing IRF3 phosphorylation byTBK1. Phosphorylated IRF3 subsequently dis-sociates from the adaptor protein and dimer-izes though the same phospho-binding domainbefore translocating into the nucleus to induceIFN. These results elucidate how IRF3 activa-tion and IFN production are tightly controlledand explain why TBK1 is necessary but not suf-ficient to phosphorylate IRF3: Phosphorylationof IRF3byTBK1 occurs onlywith the assistanceof an adaptor protein such asMAVS, STING, orTRIF, which also must be phosphorylated. ▪

RESEARCH

SCIENCE sciencemag.org 13 MARCH 2015 • VOL 347 ISSUE 6227 1217

The list of author affiliations is available in the full article online.*Corresponding author. E-mail: zhijian.chen@utsouthwestern.eduCite this article as S. Liu et al., Science 347, aaa2630 (2015).DOI: 10.1126/science.aaa2630

Phosphorylation of innate immune adaptor proteins licenses IRF3 activation. MAVS, STING,and TRIF—which are activated by viral RNA, cytosolic DNA, and bacterial lipopolysaccharide (LPS),respectively—activate the kinases IKK and TBK1. These kinases then phosphorylate the adaptorproteins, which in turn recruit IRF3, thereby licensing IRF3 for phosphorylation (P) by TBK1.Phosphorylated IRF3 dissociates from the adaptor proteins, dimerizes, and then enters the nucleusto induce IFNs.

ON OUR WEB SITE◥

Read the full articleat http://dx.doi.org/10.1126/science.aaa2630..................................................

on March 15, 2020

http://science.sciencem

ag.org/D

ownloaded from

RESEARCH ARTICLE◥

INNATE IMMUNITY

Phosphorylation of innate immuneadaptor proteins MAVS, STING,and TRIF induces IRF3 activationSiqi Liu,1 Xin Cai,1 Jiaxi Wu,1 Qian Cong,2 Xiang Chen,1,3 Tuo Li,1 Fenghe Du,1,3

Junyao Ren,1 You-Tong Wu,1 Nick V. Grishin,2,3 Zhijian J. Chen1,3*

During virus infection, the adaptor proteins MAVS and STING transduce signals from thecytosolic nucleic acid sensors RIG-I and cGAS, respectively, to induce type I interferons(IFNs) and other antiviral molecules. Here we show that MAVS and STING harbor twoconserved serine and threonine clusters that are phosphorylated by the kinases IKK and/orTBK1 in response to stimulation. Phosphorylated MAVS and STING then bind to a positivelycharged surface of interferon regulatory factor 3 (IRF3) and thereby recruit IRF3 for itsphosphorylation and activation by TBK1. We further show that TRIF, an adaptor protein inToll-like receptor signaling, activates IRF3 through a similar phosphorylation-dependentmechanism. These results reveal that phosphorylation of innate adaptor proteins is anessential and conserved mechanism that selectively recruits IRF3 to activate the type IIFN pathway.

The innate immune system employs germline-encoded pattern recognition receptors todetect common pathogenic molecular fea-tures (1). Cytosolic DNA and viral RNA aredetected by cGAS and RIG-I–like recep-

tors, respectively, to activate several convergentsignaling pathways to produce type I interferons(IFNs) (2–6). After ligand binding, cGAS andRIG-I signal through respective adaptor proteinsSTING and MAVS to recruit the kinases IKKand TBK1, which then activate the transcriptionfactors NF-kB and interferon regulatory factor 3(IRF3), respectively. Recent studies on the RIG-Ipathway have providedmechanistic insights intoinnate immune signaling. Specifically, activatedRIG-I forms oligomers to convert MAVS intoprion-like polymers, which then recruit ubiq-uitin E3 ligases TRAF2, TRAF5, and TRAF6 tosynthesize polyubiquitin chains (7–10); in turn,these ubiquitin chains recruit and activate IKKand TBK1 to trigger IFN production (9, 11).The critical role of TRAF2, TRAF5, and TRAF6

in MAVS downstream signaling closely resem-bles that of TRAF recruitment in other NF-kB ac-tivating pathways such as those emanating frominterleukin-1b (IL-1b), tumor necrosis factor–a(TNF-a), T cell receptor, and CD40 (12). How-ever, stimulation of cells with IL-1b or TNF-aactivates only IKK and TBK1, but not IRF3 (fig.

S1A, lanes 1 to 5) (13). In contrast, activation ofthe RIG-I–MAVS and cGAS-STING pathwaysthrough vesicular stomatitis virus (VSV) infec-tion and herring testis DNA (HT-DNA) trans-fection, respectively, leads to activation of IKKand TBK1, as well as IRF3 (fig. S1A, lanes 6 to8). Similarly, Toll-like receptors 3 and 4 (TLR3and TLR4) signal through the adaptor proteinTRIF to active TBK1 and IRF3 (1). However,other TLRs that do not signal through TRIFcan activate TBK1 but are unable to activateIRF3. How MAVS, STING, and TRIF possessthe ability to activate both TBK1 and IRF3 isunknown.

MAVS and IRF3 form a ubiquitination- andphosphorylation-dependent complex

To understand how MAVS activates IRF3, weused cell-free assays that recapitulate IRF3 andIkBa activation by different upstream activators(8, 11, 14). We have previously shown that a re-combinant MAVS protein, MAVSDTM (MAVSlacking the C-terminal transmembrane domain)(8), spontaneously forms prion-like fibers to ac-tivate IKK, TBK1, and IRF3 in HeLa cytosolicextracts (S100), which could be detected by im-munoblotting with phospho-specific antibodiesor by native gel electrophoresis that reveals IRF3dimerization (Fig. 1A, top). However, recombi-nant TRAF6 protein led to the activation of IKKand TBK1, but not IRF3 (Fig. 1A, bottom), sug-gesting that TBK1 activation alone is insufficientto activate IRF3.Next, to examine potential interactions be-

tween downstream effectors (IRF3 or IkBa) andupstream activators (MAVSDTM or TRAF6), HeLaS100 was incubated with MAVSDTM or TRAF6

at 0°C (control) or 30°C. Flag-tagged IRF3 (Flag-IRF3) or IkBa (Flag-IkBa) was then added to thereaction mixtures for immunoprecipitation (IP)(Fig. 1B). After an in vitro assay with MAVSDTM,a smear of protein bands in complex with Flag-IRF3 was detected with a MAVS antibody. TBK1was also present in the MAVS-IRF3 complex (Fig.1B, lane 3). In contrast, in cell extracts incubatedwith TRAF6, Flag-IRF3 was unable to pull downTBK1 (Fig. 1B, lane 9), suggesting that MAVS,but not TRAF6, induced interaction betweenIRF3 and TBK1.To determine if IRF3 and MAVS form a com-

plex in virus-infected cells, we expressed IRF3Ser385→Ala385 (S385A)/S386A (15) (Flag-IRF3 2A),which is unable to form a homodimer and maytherefore associatewithTBK1 orMAVSmore tight-ly, in human embryonic kidney 293T (HEK293T)cells. Infection of these cells with Sendai virusled to the association of Flag-IRF3 2A with en-dogenousMAVS,TRAF2,TRAF6, andTBK1 (Fig. 1C,lane 12). These results suggest that MAVS mayserve as a scaffold to bring IRF3 and TBK1 intoproximity, thereby facilitating IRF3 phosphoryl-ation by TBK1.Because multiple E3 ubiquitin ligases are

involved in MAVS downstream signaling (9),we next examined the role of ubiquitination inMAVS-IRF3 interaction. A deubiquitination en-zyme containing the ovarian tumor type domainof the Crimean Congo hemorrhagic fever virus(vOTU) completely blocked MAVS-IRF3 interac-tion when it was added to the cell-free reaction(Fig. 1D, lane 3). However, vOTU no longer af-fected MAVS-IRF3 complex formation whenadded after the reaction, suggesting that ubiq-uitination is required only for initiating but notmaintaining MAVS-IRF3 interaction (Fig. 1D,lane 4). In contrast, treatment with lambda pro-tein phosphatase after the reaction abolishedMAVS-IRF3 interaction, suggesting that MAVS-IRF3 interaction is dependent on phosphoryla-tion (Fig. 1D, lane 5). Consistently, MAVS boundto IRF3 appeared on SDS–polyacrylamide gelelectrophoresis (PAGE) as a slower-migratingsmear, which was sensitive to phosphatase butnot vOTU treatment (fig. S1B). These results sug-gest that the initial ubiquitination and subse-quent phosphorylation onMAVS are necessaryfor MAVS-IRF3 interaction.

MAVS serine-rich clusterscontaining Ser442 are essentialfor IRF3 binding and activation

To map the MAVS phosphorylation site(s) es-sential for downstream signaling, we tested apanel of MAVS truncation mutants for their abil-ity to bind and activate IRF3 (fig. S1C). We havepreviously shown that whereas the MAVS polym-erization domain CARD is essential, the middleproline-rich (residues 94 to 153) and C-terminaltransmembrane (510 to 540) regions of MAVSare dispensable for recombinant MAVS to ac-tivate IRF3 in the cell-free assay (8). Through aseries of deletion analyses, we found that a trun-cated MAVS harboring the first 130 amino acids(containing CARD) and a C-terminal 61–amino

RESEARCH

SCIENCE sciencemag.org 13 MARCH 2015 • VOL 347 ISSUE 6227 aaa2630-1

1Department of Molecular Biology, University of TexasSouthwestern Medical Center, Dallas, TX 75390-9148,USA. 2Departments of Biophysics and Biochemistry,University of Texas Southwestern Medical Center, Dallas,TX 75390-9148, USA. 3Howard Hughes Medical Institute(HHMI), University of Texas Southwestern Medical Center,Dallas, TX 75390-9148, USA.*Corresponding author. E-mail: zhijian.chen@utsouthwestern.edu

on March 15, 2020

http://science.sciencem

ag.org/D

ownloaded from

acid fragment (MAVS-N460 D131-398) was suf-ficient for both IRF3 binding and activation(fig. S1, C to E). Further mutagenesis revealedtwo MAVS C-terminal serine/threonine clus-ters, S426/S430/S433 (3S) and S442/S444/T445/S446 (4T/S), to be essential for IRF3 activation(fig. S1, F to H, and fig. S2, A and B). An un-biased structure-guided alignment of full-lengthIRF3, MAVS, and STING across species revealedsequence similarity among the MAVS 4T/S clus-

ter, STING C terminus and the IRF3 5T/S clus-ter (396 to 405), whose phosphorylation is knownto be essential for IRF3 dimerization (Fig. 2, Aand B). Specifically, not only are MAVS S442/S444aligned with IRF3 S396/S398, the essential IRF3phosphorylation sites, but the charged and hydro-phobic residues surrounding MAVS 4T/S (suchas D438, L439, and I441) also aligned well withthose around IRF3 S396/S398, suggesting thatthese two regions take on a similar structural

fold by sharing a DLxIS (where x is any aminoacid) consensus motif. Mutagenesis of the hy-drophobic resides L439 and I441 to Asp or Alaabolished the ability of MAVS to activate IRF3(fig. S2C). Furthermore, the location of MAVSS426/S430/S433 (3S) relative to MAVS S442 re-sembles the position of IRF3 S385/S386 withrespect to IRF3 S396 (fig. S2A); phosphorylationof both IRF3 serine clusters is essential for itsdimerization and activation (16, 17).

aaa2630-2 13 MARCH 2015 • VOL 347 ISSUE 6227 sciencemag.org SCIENCE

Fig. 1. IRF3 and MAVS form a ubiquitination- andphosphorylation-dependent complex. (A) HeLa cellextracts were incubated with MAVSDTM or TRAF6 inthe presence of ATP at 30°C for the indicated time.IRF3 dimerization (which indicates activation) in thisand other panels was analyzed by native gel electro-phoresis followed by immunoblotting (IB). Protein phos-phorylation was detected by immunoblotting with theindicated antibodies. (B) After incubation of MAVSDTMor TRAF6 in MEF extracts, Flag-IRF3 or Flag-IkBa wasadded before immunoprecipitation (Flag-IP). Coprecip-itated proteins were detected by immunoblotting.

(C) Twelve hours after Flag-IRF3-S385A/S386A (2A) transfection, HEK293Tcells were infected with Sendai virus as indicated. Flag-IP was carried out in whole-cell lysates, and coprecipitated proteins were detected by immunoblotting. (D) Indicated enzymes (vOTUand phosphatase) were added during or after (indicatedby asterisks) incubation of recombinant MAVSDTM in HEK293T extracts. GST-IRF3 2A was then added as indicated, followed by GST-pull down. MAVS-IRF3interaction was examined by immunoblotting.The data presented in this and all subsequent figures were reproduced in at least two independent experiments.

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

To examine whether phosphorylation of theMAVS 4T/S cluster is important for IRF3 activa-tion, wemutated the residues to alanine (4T/S→4A)or to aspartic acid (4T/S→4D). Recombinant MAVS4T/S→4A with the transmembrane domain de-letion (MAVS-N460 4T/S→4A) failed to bind oractivate IRF3 (fig. S2, B and D). MAVS-N4604T/S→4D, however, resulted in weak IRF3 inter-action and activation, probably due to incompletephosphate mimic by aspartic acid, analogous tothe loss of function of IRF3 S385D/S386D (18, 19).

Similar to the MAVS wild type (WT), the MAVS4A mutant still triggered phosphorylation ofTBK1 and IkBa, and it also directly interactedwith TRAF2 and TRAF6 (fig. S2, B and E), indi-cating that the observed defect in IRF3 bindingand activation is specific and not due to mis-folding of MAVS.Subsequent mutation of each T/S residue re-

vealed that MAVS S422A alone abolished IRF3dimerization (Fig. 2C). To examine the possibilitythat S442 may be important structurally (e.g.,

for hydrogen bonding or polarity), we mutatedS442 into other amino acids. Mutating S442 intocysteine (S442C) or asparagine (S442N), whichretains serine’s hydrogen bonding ability orpolarity but can no longer be phosphorylated,abolished MAVS’ ability to activate IRF3 (fig.S2F). In contrast, serine-to-threonine (S442T) orserine–to–glutamic acid (S442E) mutants large-ly retained the activity, suggesting that MAVSS442 phosphorylation is essential for IRF3 ac-tivation in vitro.

SCIENCE sciencemag.org 13 MARCH 2015 • VOL 347 ISSUE 6227 aaa2630-3

Fig. 2. MAVS C terminus harbors a conserved serine-rich region con-taining Ser442 essential for IRF3 binding and activation. (A) A structure-guided cross-species sequence alignment of full-length IRF3, STING, andMAVS using PROMALS3D (32) revealed a cLxIS (c, charged residues; x, anyamino acid) consensus motif in the C-terminal regions. Conservation indexscore: 9 is the highest, ≥5 is significant (46). IRF3 structures were used toassist alignment by the software. (B) Diagrams of MAVS domains and its4T/S site compared to IRF3 5T/S site. TM, transmembrane domain. (C)Recombinant MAVS WTand point mutants at the 4T/S site were tested for

their ability to activate IRF3 in the cell-free assay. MAVS protein level wasanalyzed by immunoblotting. (D and E) Mavs−/− MEFs reconstituted withMAVSWTor 4T/S mutants were infected with VSV for the indicated time. IRF3dimerization, IkBa phosphorylation, and protein expression were analyzed byimmunoblotting. IFN-b mRNA levels were measured by q-RT-PCR. Error bars inthis and other figures represent SDs of triplicates. (F) HA-tagged IRF3 2A wasfurther stably expressed in the reconstituted cells described in (D) and (E).After VSV infection, HA-IP was carried out using a HA antibody to examine theinteraction between IRF3 2A and the MAVS proteins.

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

To investigate the role of MAVS S442 duringvirus infection, we reconstituted MAVS WT ormutants into Mavs−/− mouse embryonic fibro-blasts (MEFs). Mavs−/− MEFs expressing MAVSS442A failed to activate IRF3 or produce IFN-aor -b in response to VSV, whereas IkBa phos-phorylation was unaffected (Fig. 2, D and E, andfig. S3A). Virus infection, however, induced com-parable TBK1 phosphorylation at serine 172 inboth MAVS WT and S442A-expressing cells, in-dicative of normal TBK1 activation (fig. S3B).Point mutation of S442 to Cys or Asn, but notThr or Asp, abolished MAVS’ ability to induceIRF3 dimerization in cells upon virus infection,suggesting that MAVS S442 is essential due toits phosphorylation, but not the serine’s otherstructural roles (fig. S3, D to F). Both MAVS 4Aand S442A mutations also diminished the abil-ity of MAVS to interact with IRF3 2A in responseto VSV infection (Fig. 2F and fig S3C). Altogether,these results indicate that MAVS phosphoryl-ation at S442 is critical for IRF3 activation butdispensable for TBK1 and IKK activation.In contrast to the S442 point mutants, indi-

vidual mutation of the other three Ser/Thr atthe 4T/S site (S444A, T445A, or S446A) did notimpair IRF3 dimerization or IFN-a or -b induc-tion (Fig. 2, D and E, and fig. S3A). However,simultaneous mutations of these residues (S444A/T445A/S446A) strongly diminished IRF3 dimer-ization and IFN-a or -b induction, indicating thatresidues in the 4T/S site function cooperativelyin IRF3 activation.To examine the role of the first serine cluster

(3S) of MAVS in IRF3 activation, we reconstitutedMavs−/− MEFs with MAVS harboring simulta-neous (S426A/S430A/S433A, or 3S→3A) or singlemutations. MAVS 3S→3A, but not individualmutants, abolished virus induction of IFN-b (fig.S3G). Additionally, similar to MAVS 4T/S→4A,the MAVS 3S→3A mutation also completely abol-ished MAVS’ ability to interact with IRF3 uponVSV infection (fig. S3C). Hence, the MAVS 3S site,in addition to the 4T/S site, is also importantfor virus-induced IRF3 activation. This patternof two key serine/threonine clusters on MAVS isreminiscent of those on IRF3, where phosphoryl-ation at S385 and S386, in addition to S396/S398-containing 5T/S phosphorylation, is required forIRF3 dimerization.

MAVS 4T/S site, including Ser442,is phosphorylated redundantlyby TBK1 and IKK

To monitor Ser442 phosphorylation, we gener-ated a polyclonal antibody specific for humanMAVS phospho-serine 442 (p-S442). After IRF3activation in the cell-free assay, the MAVS p-S442antibody recognized only MAVS-N460 WT, butnot S442A (fig. S4A). MAVS S442 phosphorylationwas not observed in assays with MAVS E26A, apolymerization-defective mutant unable to acti-vate downstream kinases (fig. S4B, lane 3) (9).MAVS was also not phosphorylated in cell ex-tracts treated with vOTU or deficient in TRAF2/5or NEMO, which were unable to support IRF3activation (fig. S4B, lanes 2, 6, and 7). However,

S442-phosphorylated MAVS was strongly en-riched in the IRF3 immunoprecipitates afterthe cell-free assay (fig. S4A, lane 3, top). Hence,MAVS S442 phosphorylation depends on MAVSpolymerization and TRAF-mediated polyubiquitinsynthesis and correlates with MAVS-IRF3 com-plex formation.We then examined whether MAVS S442 phos-

phorylation could be detected in virus-infectedcells. The p-S442 antibody was unable to detect aclear MAVS signal in whole-cell lysates beforeor after Sendai virus infection. However, IP withthe p-S442 antibody revealed a MAVS smear thatappeared only after virus infection, indicatingthat MAVS S442 phosphorylation was inducedby the virus (Fig. 3A). Using targeted quanti-fication by mass spectrometry, we observed arobust induction of phosphorylated peptidescontaining the MAVS 4T/S site, including S442after virus infection (Fig. 3B and fig. S4, D to F;see also Materials and methods section on Massspectrometry). Our mass spectrometry resultsalso revealed that MAVS became phosphorylatedat many other Ser/Thr residues after virus infec-tion (table S2), which may explain its formationof a smear on SDS-PAGE (Fig. 3A, lane 4).Virus infection recruits TBK1 to MAVS. How-

ever, in Tbk1−/− extracts, MAVS S442 phosphoryl-ation and interaction with IRF3 were unaffected,whereas Ikka/Ikkb-deficient extracts showedreduced IRF3 activation and binding to MAVS(fig. S4B and fig. S5A). MAVS S442 phosphoryl-ation in Tbk1−/− extracts was completely blockedby an IKK inhibitor, TPCA-1, but not by a TBK1inhibitor BX795 (fig. S5B). Subsequent analysisindicated that the combination of IKK and TBK1inhibitors, but not each alone, blocked MAVS-IRF3 interaction and MAVS S442 phosphoryl-ation in WT cell extracts (fig. S5, C and D).In U2OS cells infected with Sendai virus, the

combination of IKK and TBK1 inhibitors, but noteach alone, abolished MAVS phosphorylation,as revealed by IP with the p-S442 antibody (Fig.3C). Mass spectrometry analysis indicated thatthe abundance of singly phosphorylated MAVS4T/S peptide (including p-S442) was severelydiminished only when both inhibitors wereadded to virus-infected HEK293T cells (Fig. 3D,and fig. S5E).In vitro kinase assays revealed that recombi-

nant TBK1 and IKKb directly phosphorylatedMAVS S442 (Fig. 3E). TBK1 also induced MAVS-IRF3 interaction, which was abolished by phos-phatase treatments (fig. S5F). Altogether, theseresults indicate that TBK1 and IKK are capableof directly phosphorylating MAVS C terminusupon virus infection. Phosphorylated MAVS maythen recruit IRF3 for its phosphorylation byTBK1 (see below).

IRF3 C terminus harbors positively chargedsurfaces important for MAVS-IRF3interaction and IRF3 dimerization

To map the IRF3 region responsible for bindingto phosphorylated MAVS, we made a series ofrecombinant IRF3 truncation mutants and testedtheir MAVS binding ability (fig. S6, A and B).

In vitro IRF3 IP revealed that the entire IRF3C terminus (190 to 427, IRF3-C), containing theIRF3 association domain and the serine-rich re-gion, is necessary and sufficient for MAVS bind-ing (fig. S6, A and B). IRF3-C crystal structuresrevealed a similar fold to the Mad homology 2(MH2) domain of the SMAD family of proteins(20, 21). As a well-known phospho-binding do-main, the MH2 domain contains positively chargedsurfaces composed of conserved basic residuesthat are important for both phosphorylated trans-forming growth factor–b (TGF-b) receptor bindingand subsequent dimerization of the phosphoryl-ated SMADs (22). Similarly, IRF3-C contains pos-itively charged patches composed of conservedbasic resides (Fig. 4A, and fig. S7), mutations ofwhich were shown to abolish phosphorylation anddimerization of IRF3 in virus-infected cells (21).To determine whether the observed IRF3 phos-

phorylation defect was due to impaired MAVSbinding, we tested IRF3 2A proteins containingpositively charged surface mutations for theirability to bind MAVS. Compared to IRF3 2A, allof the positively charged surface mutants hadreduced interaction with MAVS and TBK1 inour cell-free assay (Fig. 4B). Additionally, whenreconstituted into Irf3−/−Irf7−/− MEF cells, onlyIRF3 2A but not 2A proteins containing posi-tively charged surface mutations interacted withendogenous MAVS in response to VSV infection(Fig. 4C and fig. S6C). These results indicatethat the positively charged surfaces on IRF3-Care important for two interactions: the bindingto phosphorylated MAVS and to a second phos-phorylated IRF3 molecule (i.e., dimerization ofphosphorylated IRF3).To uncouple the steps of MAVS-IRF3 bind-

ing and IRF3 dimerization, Flag-IRF3 WT or 2A(S385A/S386A, which cannot dimerize) was addedto cell extracts with active MAVS either beforethe 30°C incubation, so that the reaction couldgo to completion, or afterward at 4°C where MAVShad been phosphorylated but IRF3 remainedunphosphorylated (Fig. 4D). When added to thereaction at 30°C, IRF3 WT dimerized exclusively,whereas IRF3 2A stably interacted with MAVS(Fig. 4D, lane 5 and 6). However, when added at4°C after the 30°C assay, both IRF3 WT and 2Ainteracted with MAVS comparably (lanes 7 and8), suggesting that IRF3 first binds to phospho-rylated MAVS (lane 7) before dissociating to forma dimer (lane 5). To further validate this mod-el, Flag-IRF3 WT or 2A was added to Tbk1−/− cellextracts before the 30°C incubation, in whichMAVS was still phosphorylated at S442 by IKKbut IRF3 could not be phosphorylated (fig. S4Band fig. S6D, bottom). Here, unphosphorylatedIRF3 exclusively interacted with MAVS withoutforming a dimer (fig. S6D, lane 2). This is in con-trast to IRF3 WT in WT cell extracts, which wasphosphorylated and formed a dimer (fig. S6D,lane 1). The phosphorylation-defective IRF3 2Amutant, however, exclusively interacted withMAVS (fig. S6D, lane 3). Hence, IRF3 phospho-rylation causes its dissociation from MAVS. Con-sistently, compared with IRF3 2A, IRF3 WT pulleddown less MAVS in response to VSV infection

aaa2630-4 13 MARCH 2015 • VOL 347 ISSUE 6227 sciencemag.org SCIENCE

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

from cells, suggesting that the majority of IRF3WT proteins had dissociated from MAVS andundergone phosphorylation-induced dimeriza-tion (Fig. 4C, lanes 2 and 7). Taken together,these results indicate that IRF3 directly binds tophosphorylated MAVS through conserved, pos-itively charged surfaces within the IRF3-C domain,which also mediate interaction with another phos-phorylated IRF3 monomer after its dissociationfrom MAVS.

Both MAVS polymerizationand phosphorylation are requiredfor MAVS-IRF3 interaction

In vitro kinase assay revealed that recombinantMAVS 361-460 could be directly phosphorylated

by recombinant TBK1 and IKK, but the phos-phorylated fragment failed to interact with IRF3,suggesting that phosphorylation alone is insuffi-cient for IRF3 recruitment. Given that recombi-nant MAVS containing CARD and a C-terminalfragment (399 to 460) interacted with IRF3 in thecell-free assay (fig. S1E), we tested whether MAVS-CARD–mediated polymerization is important forIRF3 binding. Active recombinant TBK1 inducedS442 phosphorylation in both MAVS-N460 WTand polymerization-defective mutants (Fig. 4E).However, only MAVS WT, but not the CARDpolymerization mutants, interacted with IRF3in a MAVS phosphorylation–dependent manner(Fig. 4E). These results indicate that MAVS-CARD polymerization is not only important for

TRAF and kinase recruitment (9) but also requiredfor subsequent IRF3 binding and activation.

STING phosphorylation at Ser366 isrequired for IRF3 binding and activationin the DNA-sensing pathway

STING is an essential adaptor protein in thecytosolic DNA sensing pathway that also ac-tivates IRF3 (23). Recently, STING Ser366 wasshown to be important in IRF3 binding and ac-tivation because a mutation to alanine (S366A)abolished DNA induced IRF3 activation (24). Incontrast, another study suggested that Ser366

phosphorylation by ULK1 negatively regulatesSTING because an aspartic acid (S366D) muta-tion renders the protein inactive (25). Thus, the

SCIENCE sciencemag.org 13 MARCH 2015 • VOL 347 ISSUE 6227 aaa2630-5

Fig. 3. The MAVS 4T/S cluster including Ser442 is redundantly phosphoryl-ated by TBK1 and IKK. (A) U2OS cells were infected with Sendai virus (SeV)for 12 hours, then whole-cell extracts were immunoprecipitated with a MAVSp-S442 specific antibody under denaturing conditions. The whole-cell extractsand IP products were then immunoblotted with a MAVS antibody. IRF3 dimeriza-tion was monitored by native gel electrophoresis and immunoblotting. (B) Tar-geted MS2 quantification of the IS442ASTSLGR peptide containing MAVS 4T/Ssite before or after SeV infection. The relative abundance of nonphosphorylated4T/S (4T/S), singly phosphorylated 4T/S (p-4T/S), or phosphorylated S442

(p-S442) peptides were selectively monitored using ions specific for each spe-cies (also see table S1 and Materials and methods).The intensity ratio (in red) =MA+SeV/MA-SeV (MA, mass area). (C and D) Similar to (A) and (B), except thatthe cells were treated with a TBK1 inhibitor (BX795), an IKK inhibitor (TPCA-1),or both 1 hour before virus infection. In (D), the relative intensity of eachpeak is shown in red (normalized with peptide MA of DMSO-treated sam-ples). (E) Recombinant TBK1 or IKKbwas incubated with purified MAVS N460in the presence of ATP at 30°C. The reaction products were analyzed by im-munoblotting with an antibody specific for p-S442 MAVS or total MAVS.

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

aaa2630-6 13 MARCH 2015 • VOL 347 ISSUE 6227 sciencemag.org SCIENCE

Fig. 4. IRF3 positively charged surfaces and MAVS aggregation are im-portant for the interaction between IRF3 and phosphorylated MAVS.(A) Diagrams of IRF3 domains and conserved, positively charged residues.DBD, DNA binding domain; IAD, IRF3 association domain; SR, serine-richregion. (B) Purified Flag-IRF3 2A or 2A containing positively charged sur-

face mutations was examined for the ability to form a complex with MAVS and TBK1 in cell extracts. (C) Irf3−/−Irf7−/− (IRF DKO) or Mavs−/− MEF cellsstably expressing HA-tagged IRF3 WT, 2A, or 2A containing mutations at the positively charged surfaces were infected with VSV, then HA-IP was carriedout to examine the interaction between MAVS and IRF3 WTor mutants. (D) Flag-IRF3 WTand 2A proteins were either incubated with active recombinantMAVS in HEK293Textracts at 30°C (lanes 5 and 6) or added after the reaction and then incubated at 4°C (lanes 7 and 8; indicated by asterisks). Flag-IPwas then carried out to examine IRF3-MAVS-TBK1 complex formation. IRF3 dimerization after the reactions was monitored by immunoblotting. (E) MAVSWT and CARD mutants were incubated with recombinant TBK1 in the presence of ATP followed by the addition of phosphatase (CIP) when indicated.Flag-IRF3 2A was then added followed by Flag-IP to examine the interaction between MAVS and IRF3. MAVS proteins and their phosphorylation at S442after the reaction were analyzed by immunoblotting.

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

SCIENCE sciencemag.org 13 MARCH 2015 • VOL 347 ISSUE 6227 aaa2630-7

Fig. 5. STING Ser366 phos-phorylation by TBK1 re-cruits IRF3. (A) Diagramsof the STING domains andits Ser366-containing motifcompared with the IRF3C-terminal 5T/S region.(B) Sting−/− Raw264.7macrophages were recon-stituted with C-terminalFlag-tagged human STINGWT and S366 mutants.Three hours after HT-DNAtransfection, IRF3 dimer-ization and STING expres-sion were analyzed byimmunoblotting. (C) TargetedMS2 quantification of theLLIS365GMDQPLPLR pep-tide containing mouse STINGSer365 from DNA-stimulated

(+HT-DNA), unstimulated (-HT-DNA), or DNA-stimulated cells treated with the TBK1 inhibitor (+HT-DNA+BX795). Phosphorylated (p-S365) or non-phosphorylated peptides (S365) were calculated by selectively using ions specific for each species (also see table S1 andMaterials andmethods). (D) L929cells depleted of endogenous STING were reconstituted with hSTING WTor S366A. STING S366 phosphorylation after HT-DNA transfection was detectedwith the p-S366 antibody (fig. S8E). (E) S366 phosphorylation of endogenous STING in THP-1 cells after HT-DNA transfection was examined byimmunoblotting with the p-S366 antibody. (F) WTor Tbk1−/− HEK293Tcells stably expressing hSTING were stimulated by cytosolic delivery of cGAMP, thenIRF3 dimerization and STING phosphorylation at Ser366 were analyzed by immunoblotting. (G) HA-tagged IRF3WT, 2A, or 2A proteins containing mutationsat the positively charged surfaces were stably expressed in the L929 STING-reconstituted cells, as described in (D). After HT-DNA transfection, HA-IP wascarried out to examine the interaction between IRF3 and phosphorylated STING. (H) L929 cells stably expressing hSTING-Flag andHA-IRF3 2A as describedin (G) were transfected with HT-DNA for 2 hours or mock transfected. Interaction between IRF3 2A and p-S366-STING was visualized under confocalmicroscopy, using PLA with an anti-HA monoclonal antibody for IRF3 and an anti–p-S366 rabbit polyclonal antibody (top). The cell in the top right wasfurther subjected to a z-stack collection, which obtains and combines multiple images at different focal distances along the z axis (bottom).The image wasshown as three-dimensional, maximum projection mode in the Zen software. No PLA-positive cells were observed without DNA transfection, and ~50% ofthe HT-DNA–stimulated cells are PLA-positive (red). Blue, DAPI; green, phalloidin labeled actin filaments. The images are representative of >100 cellsexamined.

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

mechanism of STING-mediated IRF3 activationand the role of Ser366 need to be clarified.Structure-guided sequence alignment of IRF3,

MAVS, and STING revealed that STING S366 ispositioned within the cLxIS (c, charged residue;x, any residue) consensus motif that is also foundaround MAVS S442 and IRF3 S396 (Figs. 2A and5A). Moreover, STING also harbors another se-rine cluster (S353/S358 in human STING; S354/S357 in mouse STING) upstream of S366, posi-tioned similarly to IRF3 S385/S386 and MAVS3S with respect to IRF3 S396 and MAVS S442(fig. S8A). These observations suggest that STINGand MAVS may activate IRF3 through similarmechanisms. Consistently, L929 cells in whichendogenous STING was depleted by short hairpinRNA (shRNA) and replaced with mouse STING(mSTING) S365A (corresponding to hSTINGS366A) failed to activate IRF3 after HT-DNAtransfection (fig. S8, B and C). mSTING S357A(corresponding to hSTING S358A) also had dimin-ished ability to activate IRF3, whereas muta-tions of other Ser or Thr residues had little effect.mSTING S365A and S357A mutations did notaffect DNA-induced activation of TBK1 and IKK(fig. S8C). Contrary to a previous report (25), theS365A mutation of mSTING also did not inhibitDNA-induced degradation of STING (fig. S8D).When reconstituted in Sting−/− Raw264.7 mac-rophages or L929 cells depleted of endogenousSTING, hSTING S366A, S366C, or S366N failedto activate IRF3 in response to DNA, whereasS366D retained weak activity (Fig. 5B and fig.S8E). These results suggest that, like MAVS S442,STING S366 is important because of its phospho-rylation rather than other structural roles. Similarto the MAVS 4T/S→4D and IRF3 S385D/S386Dmutants, the weak activity of S366D may be dueto incomplete phosphate mimic by aspartic acid.Quantitative mass spectrometry analysis con-

firmed that phosphorylation of mSTING S365 wasinduced by more than 200-fold after HT-DNAtransfection and that this phosphorylation wasabolished by the TBK1 inhibitor BX795 (Fig. 5Cand fig. S8F; see also Materials and methods sec-tion on Mass spectrometry). We also generateda rabbit polyclonal antibody specific for S366-phosphorylated hSTING (p-S366); this antibodydetected phosphorylated hSTING in HT-DNA–transfected cells but not in unstimulated cells(Fig. 5, D and E). The S366A hSTING mutant wasnot detected by the antibody. Additionally, thehSTING p-S366 antibody detected a perinuclearpunctate structure only in DNA-stimulated cells(fig. S8G, top), similar to the previously observedSTING foci (fig. S8G, bottom) (26). To rule out anynonspecific staining by the phospho-specific anti-body, we used proximity ligation assay (PLA) inL929-hSTING-Flag cells (Fig. S8H), where positivesignals (red immunofluorescence) occur only whenthe hSTING p-S366 antibody is in close proximityto the Flag-antibody. We observed peri-nuclearsignals only in cells after DNA transfection butnot after poly(I:C) or mock transfection, indicatingthat our antibody specifically recognizes hSTINGp-S366. After reconstitution of hSTING into WTand Tbk1−/− HEK293T cells (which lack endog-

enous STING), cGAMP stimulation induced STINGp-S366 and IRF3 dimerization in WT but notTbk1−/− cells, suggesting that TBK1 is essentialfor STING S366 phosphorylation in cells (Fig.5F). These results indicate that STING under-goes TBK1-dependent phosphorylation at Ser366

only in DNA-stimulated cells.We have previously shown that a STING

C-terminal fragment (residues 281 to 379) couldactivate IRF3 in cytosolic cell extracts (24). ThisSTING fragment was phosphorylated at S366 inthe cell extracts, as revealed by immunoblottingwith thep-S366antibody (fig. S9A). Phosphorylationof STING depended on TBK1 but not on NEMO(fig. S9A). To test whether STING phosphoryl-ation by TBK1 is important for IRF3 associationwith TBK1, we incubated recombinant TBK1 withWT or S366Amutant STING (281 to 379) and thenadded Flag-IRF3 for IP (fig. S9B). WT but notS366A STING formed a complex with IRF3 andTBK1; the formation of this complexwas abolishedby phosphatase treatment (fig. S9B, lanes 5, 6, and11). Similar results were obtained after incubatingthe recombinant STING (281 to 379) with crudecell extracts (fig. S9C). These results demonstratethat TBK1 directly phosphorylates STING at Ser366

and phosphorylated STING recruits IRF3, therebyfacilitating IRF3 phosphorylation by TBK1.We next examined whether full-length STING

interacts with IRF3 in response to DNA stimulationin cells. Upon DNA transfection, IRF3 2A inter-acted with hSTINGWT but not S366A, suggestingthat hSTING S366 is essential for DNA-inducedinteraction between STING and IRF3 (Fig. 5G,lane 2 and 3). Additionally, IRF3WT and IRF3 2A,but not IRF3 2A bearing positively charged surfacemutations, interacted with S366-phosphorylatedSTING, suggesting that the phospho-binding do-main of IRF3 is crucial for binding to phosphoryl-ated STING. Notably, compared with IRF3 2A,IRF3 WT pulled down substantially less STING.This suggests that analogous to MAVS-IRF3interaction, phosphorylated IRF3 dissociates fromSTING and forms a dimer (Fig. 5G, lane 8).Using the PLA assay, we examined the inter-

action between phosphorylated STING and IRF3in L929 cells in which endogenous STING wasdepleted by shRNA and replaced by hSTING.These cells also stably expressed HA-IRF3 2A.The PLA assay using the HA antibody and thep-S366 STING antibody revealed positive sig-nals (red immunofluorescent dots) in the peri-nuclear regions only after DNA stimulation(Fig. 5H; also see Materials and methods). Incontrast, no PLA-positive signals were observedin DNA-stimulated cells expressing STING S366Aor IRF3 2A R211/R213A (a positively charged sur-face mutant) (fig. S9D). Altogether, these resultsindicate that analogous to the role of MAVS S442phosphorylation, STING phosphorylation at S366is critical for direct IRF3 recruitment and activa-tion through IRF3’s phospho-binding domain.

Phosphorylation of TRIF recruitsand activates IRF3

Besides the cytosolic nucleic acid–sensing path-ways, stimulation of certain TLRs (namely TLR3

and TLR4) also activates IRF3 and induces type IIFNs.TLR3andTLR4caneachactivate the adaptorprotein TRIF, which in turn activates IRF3 (27).TRIF contains anN-terminal domain that includesthe TIR domain important for interaction withTLRs and a C-terminal RIP homotypic interac-tionmotif, which can activate cell death pathways(fig. S10A). The N-terminal fragment containing~540 amino acids of TRIF (TRIF-N540) has beenshown to bind IRF3 and activate the interferonpromoter when transiently expressed in HEK293Tcells (28, 29). To avoid triggering cell death, wechose TRIF-N540 to investigate the mechanismbywhich IRF3 is activated in theTLR3/4 pathways.To examine whether TRIF interacts with IRF3

in a phosphorylation-dependentmanner, we tran-siently expressed Flag-tagged IRF3 2A and TRIF-N540 in HEK293T cells. IP of IRF3 revealed aninteraction between IRF3 2A and TRIF-N540 thatwas greatly diminished after pretreatment of cellswith the TBK1 inhibitor BX795, suggesting thatTBK1 is important for inducing TRIF-IRF3 interac-tion (Fig. 6A, lane 2 and 3).Moreover, phosphatasetreatment of cell lysates using calf intestinal phos-phatase (CIP) or Lambda phosphatase (LambdaPP) before Flag-IRF3 IP completely abolished TRIFbinding to IRF3-2A, suggesting that a phosphoryl-ation event induced the TRIF-IRF3 interaction(Fig. 6A, lanes 4 and 5). IRF3 WT and IRF3 2Acontaining mutations in the positively chargedsurface (described above) all failed to pull downTRIF-N540 (Fig. 6B). Collectively, these data in-dicate that, similar to MAVS and STING, TRIFalso recruits IRF3 through a mechanism that de-pends on the kinase TBK1, its phosphorylation,and the phospho-binding domain of IRF3.Through examination of the TRIF-N540 se-

quence, we found a conserved Ser/Thr cluster(S210/S212/T214) whose surrounding sequencepLEIS (p, hydrophilic amino acid) sharesmarkedsimilarity to the consensus cLxISmotif (c, charged;x, any) found in IRF3,MAVS, and STING (Fig. 6C).TRIF S210 is positioned similarly to IRF3 S396,MAVS S442, and STING S366 (Fig. 6C). On thebasis of this sequence analysis, we expressed apanel of TRIF mutants, including S210A/S212A/T214A (TRIF 3A), in HEK293T cells and testedtheir ability to interact with IRF3 2A. Only TRIF3A, but not other TRIFmutants, failed to interactwith IRF3 2A (Fig. 6D). These results suggest thattheTRIFS210/S212/T214clusterprobably functionssimilarly toMAVS 4T/S and STING S366 in IRF3binding and activation.Quantitative mass spectrometry analysis iden-

tified phosphorylated peptides containing S210in TRIF protein that was transiently expressed inHEK293T cells (fig. S10B). The intensity of phos-phorylated peptides containing S210 increasedby >11-fold after TBK1 coexpression (fig. S10B),suggesting that TBK1 mediates TRIF phospho-rylation at the S210/S212/T214 cluster. Overexpres-sion of WT, but not kinase-dead mutant of TBK1(TBK1 Mut), induced a robust and phosphatase-sensitive interaction between TRIF-N540 andHA-IRF3 2A (Fig. 6E, lanes 3 to 6). In contrast, TBK1failed to induce interaction between TRIF 3A andHA-IRF3 2A, suggesting that TRIF phosphorylation

aaa2630-8 13 MARCH 2015 • VOL 347 ISSUE 6227 sciencemag.org SCIENCE

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

at S210/S212/T214 by TBK1 probably induced theinteraction between TRIF and IRF3 (Fig. 6E,lane 7). Taken together, these data indicate that

TRIF is phosphorylated at its consensus motifS210/S212/T214 by the kinase TBK1, which leadsto IRF3 recruitment.

When transiently expressed in HEK293T cells,only TRIF 3A, but not other TRIFmutants, failedto induce IRF3 dimerization (Fig. 6F). TRIF 3A

SCIENCE sciencemag.org 13 MARCH 2015 • VOL 347 ISSUE 6227 aaa2630-9

Fig. 6. TRIF recruits and activates IRF3 through phosphorylation at thepLxIS motif. (A) TRIF-N540 and Flag-tagged IRF3 2A were transientlyexpressed in HEK293T cells or cells that were treated with or without theTBK1 inhibitor BX795. Cell extracts were treated with different phosphatases,as indicated, before they were subjected to Flag-IP to examine the interactionbetween TRIF and IRF3 2A. (B) Similar to (A), except that IRF3 WTor IRF3 2Aproteins containing mutations at the positively charged surfaces were exam-ined for their ability to bind TRIF. (C) A cross-species sequence alignment

between TRIF, IRF3, MAVS, and STING showing the consensus phosphoryl-ation motif in TRIF. (D) Similar to A, except that different TRIF mutants wereexamined for their ability to bind IRF3 2A. (E) Similar to (A), except that Flag-tagged TRIF WT or mutant, HA-IRF3 2A, and Flag-TBK1 WT or kinase-deadmutant (Mut) were transiently expressed in HEK293Tcells. HA-IP was carriedout to examine the TRIF-IRF3 interaction. (F and G) After transient expressionof TRIF mutants in HEK293T cells, IRF3 dimerization, TBK1 phosphorylation,and TRIF expression were analyzed by immunoblotting.

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

led to normal TBK1 phosphorylation at S172,indicating that the observed IRF3 dimerizationdefect is specific and not due to misfolding ofTRIF (Fig. 6F). Moreover, mutagenesis of S210 toalanine alone in TRIF-N540 or full-length TRIFabolished its ability to induce IRF3 dimerization,suggesting that, like MAVS S442 and STINGS366, TRIF S210 is the critical phosphorylationsite that mediates IRF3 activation (Fig. 6G).Altogether, our data indicate that TRIF bindsand activates IRF3 through a phosphorylation-dependent mechanism that is similar to that ofMAVS and STING.

Discussion

Through in vitro reconstitution and cell-basedassays, we reveal that phosphorylation of a con-sensus motif in MAVS, STING, and TRIF is es-sential for IRF3 recruitment and subsequentphosphorylation by TBK1. Based on previouslypublished results and those presented here, wepropose the following model for MAVS/STING/TRIF-mediated IRF3 activation (Fig. 7): (i) Adap-tor protein activation: after ligand binding, RIG-I,cGAS, and TLR3/4 activate downstream adaptorproteins MAVS, STING, and TRIF, respectively.(ii) Kinase activation: active MAVS polymers orTRIF proteins recruit TRAF family ubiquitin E3ligases (i.e., TRAF2, 5 and 6) to synthesize poly-ubiquitin chains to activate IKKandTBK1,whereasactive STINGdirectly recruits and activates TBK1.(iii) Adaptor protein phosphorylation: the re-cruited kinases thenphosphorylateMAVS, STING,or TRIF at their conserved pLxIS motif. (iv) IRF3recruitment: IRF3binds to phosphorylatedMAVS,STING, or TRIF through IRF3’s conserved, posi-tively charged surface. (v) IRF3 phosphorylation:IRF3 is efficiently phosphorylated by TBK1 oncethey are in closeproximity. (vi) IRF3 self-association:phosphorylated IRF3 dissociates from the adap-tors and dimerizes though the same positivelycharged surfaces. IRF3 dimer then enters the nu-cleus, where it functions together with NF-kB toturn on type I interferons and other cytokines.The recruitment of IRF3 to phosphorylated

MAVS, STING, and TRIF for its subsequent ac-tivation by TBK1 ismarkedly similar to themecha-nism of SMAD activation by the TGF-b receptor(22). Analogous to TGF-b receptor I (TbR-I) phos-phorylation by recruited TbR-II, MAVS, STING,and TRIF are phosphorylated at their conservedconsensus serine-rich clusters by recruited TBK1and/or IKK. Phosphorylated TbR-I and MAVS/STING/TRIF then recruit SMAD and IRF3, re-spectively, through structurally similar phospho-binding domains on SMAD and IRF3 that alsomediate their respective homodimerization. Thisallows phosphorylated IRF3 or SMAD to dimer-ize and then enter the nucleus (30). In contrast,IL-1b, TNF-a, and other TLRs (e.g, TLR2) onlyactivate IKK and TBK1, but not IRF3, due to thelack of a phosphorylated adaptor protein that canrecruit IRF3.Sequence profiles of IRF3, MAVS, STING, and

TRIF across mammalian species revealed a con-sensus IRF3 binding motif: pLxIS (p, hydrophi-lic; x, nonaromatic) (fig. S11), which is always

located in a disordered region of the respectiveprotein. In addition, the presence of another con-served serine(s) can be found within 15 aminoacids upstreamof the pLxISmotif and serves as asecond phosphorylation site (i.e., S385/S386 inIRF3 and S426/S430/S433 in MAVS) for IKK-related kinases. By taking these features intoaccount, we performed a mammalian proteome-wide bioinformatic search to identify other pro-

teins that harbor a similarmotif (31, 32). Twenty-one mammalian gene candidates were found tocontain such a motif, and they were groupedbased on their functional similarity (fig. S12, andtable S3, B andC; also seeMaterials andmethodssection on Computational biology). Among these21 candidates, proteins related to innate immu-nity form the largest cluster, suggesting that theconsensus motif found in MAVS, STING, TRIF,

aaa2630-10 13 MARCH 2015 • VOL 347 ISSUE 6227 sciencemag.org SCIENCE

Fig. 7. A model of kinase-substrate specification by phosphorylated adaptors. (A) After its bindingto viral RNA, RIG-I induces the polymerization of MAVS on the mitochondrial outer membrane. MAVSthen recruits TRAF proteins to activate IKK and TBK1, which in turn phosphorylate MAVS at theconsensus pLxIS motif. Phosphorylated MAVS binds to the C-terminal positively charged region of IRF3,thereby recruiting IRF3 for phosphorylation (P) by TBK1 through induced proximity. Phosphorylated IRF3then forms a homodimer that enters the nucleus to turn on transcription. See the Discussion section fora more detailed description of the steps in the pathway. (B) Similar to (A), except that STING isphosphorylated by TBK1 in response to stimulation by DNA. DNA in the cytosol activates cGAS toproduce the second messenger cGAMP, which then binds and activates STING. (C) lipopolysaccharide(LPS) stimulation activates TLR4, which in turn activates the adaptor proteins MyD88 and TRIF. TRIFactivates TBK1, which in turn phosphorylates TRIF at the consensus motif. Phosphorylated TRIF thenrecruits IRF3 to facilitate IRF3 phosphorylation by TBK1.

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

and IRF3 is mostly related to innate immuneresponses, possibly as a conservedmechanism toregulate the IRF family of transcription factors.In this regard, it is interesting to note that IRF5,a key transcription factor that regulates inflam-mation, also contains the pLxISmotif. The serineresidue within this motif (S445 in mouse IRF5and S446 in human IRF5) has recently beenshown to be phosphorylated by IKKb for IRF5activation (33, 34). In summary, phosphorylationof innate immune adaptor proteins (e.g, MAVS,STING, and TRIF) and their recruitment of ki-nase substrates (e.g, IRF3) may be a general andconserved mechanism that provides signalingspecificity in innate immunity.

Materials and methods

Antibodies

Rabbit antibodies against human IRF3, TRAF2,TRAF6, and IKKa/b andmouse antibody againsthuman MAVS (residues 1 to 130) were obtainedfrom Santa Cruz Biotechnology; Flag antibody(M2), M2-conjugated agarose, and tubulin anti-body were purchased from Sigma; HA antibodyand anti-HA–conjugated agarose were from Co-vance; antibodies against p-IRF3 Ser396, p-TBK1Ser172, p-IkBa Ser32/36, and p-IKKa Ser176/p-IKKbSer177 were from Cell Signaling; and mouse IRF3antibody was from Invitrogen. Anti-TBK1 mono-clonal antibody was from IMGENEX. Polyclonalantibody against human TRIF was from Cell Sig-naling.Mouse immunoglobulinGTrueBlotULTRAwas fromRockland. The rabbit antibodies againsthuman MAVS and STING were generated asdescribed before (24, 35). Rabbit antibodiesagainst MAVS p-Ser442 and STING p-Ser366 weregenerated by immunizing rabbits with KLH-conjugated, chemically synthesized peptides “acetyl-EDLAIS(phospho)ASTSC” and “acetyl-PELLIS(phospho)GMEKC,” respectively. The antibodieswere affinity purifiedwith corresponding phospho-peptide columns.

Expression constructs, recombinantproteins, and RNAi

For expression inmammalian cells, humancDNAsencoding N-terminal Flag-tagged IRF3 and IkBawere cloned into pcDNA3. After overexpres-sion of these constructs in HEK293T cells, theencoded Flag-tagged proteins were further puri-fied with M2 agarose, followed by Flag peptideelution. Human cDNA encoding MAVS WT andmutants were cloned into pTY-EF1A-puroR-2alentiviral vector. Flag- or HA-tagged human IRF3-S385AS386A (2A) and other mutants were clonedinto pTY-EF1A-HygromycinR-2a lentiviral vector.A human MAVS shRNA sequence (5′-GGAGA-GAATTCAGAGCAAG-3′) containingU6 promoter,alongwithFlag-taggedhumanMAVSA440RM449R(2R), was cloned into pTY-U6-shRNA-EF1A-puroR-2a. STING lentiviral vectors pTY-U6-sh-mSTING-EF1A-puroR-2a-STING were cloned as describedpreviously (24). These lentiviruses were trans-duced intoMavs–/–MEF cells, HEK293T cells, orL929 cells as described previously (24). HA- orFlag-tagged TRIF-N540 andmutants were clonedinto pcDNA3. Flag-tagged TRIF-FLwas in pEF-BOS

vector. Mutants were constructed with the Quik-Change Site-DirectedMutagenesisKit (Stratagene;also see table S4 for the primer information). Forexpression inEscherichia coli, pET23a-His6-MAVSDTM [amino acids (aa) 1 to 510] and pET28a-His6-SUMO-MAVS-N460 (aa 1 to 460; WT andmutants) were transformed and expressed inE. coli BL21(DE3)-pLysS strain. These His-taggedproteins were purified as described previously(8). Sumo protease was subsequently used to cutoff His6-SUMO tag, yielding nontagged pro-teins. In addition, His8-IRF3 (from E. coli), His6-TRAF6 (from insect Sf9 cells), GST-TBK1 (Sf9cells; GST, glutathione S-transferase), and GST-IKKb (Sf9 cells) were purified as described pre-viously (8, 11, 14).

Viruses, cell culture, and transfection

Sendai virus (Cantell strain, Charles River Labo-ratories) was used at a final concentration of 100hemagglutinating units/ml. VSV (DM51)-GFP vi-rus was from J. Bell (Univ. of Ottawa) and waspropagated in Vero cells. Plasmids and HT-DNAwere transfected into cells using lipofectamine2000 (Life Technologies). Digitonin permeabili-zation was used to deliver cGAMP into culturedcells as previously described (36). A lentiviral sys-tem for stable gene expression and shRNAknock-downwere used as described before (24). All cellswere cultured at 37°C in an atmosphere of 5%(v/v) CO2. HEK293T and Raw264.7 macrophageswere cultured in Dulbecco’s modified Eagle’s me-dium (DMEM) supplemented with 10% (v/v) cos-mic calf serum (Hyclone), penicillin (100 U/ml),and streptomycin (100 mg/ml). HeLa, MEF, U2OS,L929, and BJ cells were cultured in DMEM sup-plemented with 10% (v/v) fetal bovine serum(Atlanta) and antibiotics. THP1 cells were culturedin RPMI 1640 supplemented with 10% fetal bo-vine serum, 2 mM b-mercaptoethanol, and anti-biotics. Sting–/– Raw264.7 macrophages werepurchased from InvivoGen.

Generation of TBK1 knockout HEK293Tcells by CRISPR/Cas9

Single-guide RNA (sgRNA) with the sequence5′-CATAAGCTTCCTTCGTCCAG-3′ was designedfor targeting exon 2 of the human TBK1 genomiclocus. The sgRNA sequence driven by a U6 pro-moter was cloned into a lentiCRISPR vectorthat also expresses Cas9 as previously described(37). The lentiviral plasmid DNA was then pack-aged into a lentivirus for infection in HEK293Tcells. Infected cells were selected in puromycin(2 mg/ml) for 2 weeks before single colonieswere selected and tested for TBK1 expressionby immunoblotting.

Biochemical assays for IRF3 activationand MAVS-IRF3 complex formation

Cell-free assays for IRF3 activation and phospho-rylation of IkBa, IKKa/b, and TBK1 were pre-formed as described previously (8, 9, 11). Similarly,MAVS S442 and STING S366 phosphorylationwas detected by immunoblotting with the p-S442and p-S366 antibodies, respectively, after the cell-free assay. For a better MAVS p-S442 detection,

His6-tagged MAVS-N460 in 100-ml reaction mix-tures was pulled down with the Ni-NTA agarosein 8 M urea. After washing the agarose with buf-fer A [20 mM Tris-HCl (pH 7.0), 1 M NaCl, and0.5%NP-40],MAVSphosphorylationwasdetectedby immunoblotting with the p-S442 antibody.To determineMAVS-IRF3 or STING-IRF3 com-

plex formation in the cell-free assay, a reactionmixture (100 ml) containing buffer B [20 mMHEPES-KOH (pH 7.0), 2 mM adenosine triphos-phate (ATP), 5 mM MgCl2, and 0.5 mM dithio-threitol (DTT)], 200 ng recombinant MAVSDTMorMAVS-N460 orHis6-STING (281 to 379), 200 mgcytosolic extracts (S5 or S100), and 200 ng Flag-IRF3 2A or GST-Flag-IRF3 2A was incubated at30°C for 1 hour. In some experiments, Flag-IRF3WT or 2A was added into the reactionmixture at4°C right before IP as indicated. Flag-IP was thencarried out using Flag antibody (M2) agarose at4°C for 2 hours in the presence of buffer C [20mMTris-HCl (PH 7.5), 150mMNaCl, 0.5%NP-40] andthe protease inhibitor cocktail (Roche). The agar-ose beads were washed three times with bufferC, and coprecipitated proteins were detected byimmunoblotting.For MAVS or STING phosphorylation by puri-

fied kinases TBK1 or IKKb, a reaction mixture(10 ml) containing buffer A, 25 ng recombinantMAVS protein, and 40 ng GST-TBK1 or GST-IKKb was incubated at 30°C for 40 min. Phos-phorylation of MAVS and STING were detectedby immunoblotting with the p-S442 and p-S366antibodies, respectively.To determineMAVS-IRF3 or STING-IRF3 com-

plex formation in the purified kinase assay, areaction mixture (100 ml) containing buffer A,250 ng recombinant MAVS or STING protein,and 400 ng GST-TBK1 or GST-IKKb was incu-bated at 30°C for 40 min. After stopping the re-actionwith 5mMEDTA, Flag-IRF3 2Awas addedinto the reaction mix at 4°C before IP. Flag-IPwas then carried out usingM2-agarose at 4°C for1 hour in the presence of buffer C and 0.2 mg/mlbovine serum albumin. The agarose beads werewashed three times with buffer C, and coprecipi-tated proteins were detected by immunoblotting.

IP assay for adaptor-IRF3 interactionsin cells

To determine MAVS-IRF3 complex formation incells, tagged IRF3WT or 2Awas stably expressedin the indicated cells. After virus infection, whole-cell lysates were prepared in the presence of buf-fer D [20 mM Tris-HCl (PH 7.5), 100 mM NaCl,10% glycerol, 0.5% NP-40, and 0.5% DDM] andthe protease inhibitor cocktail (Roche). Anti-Flag(M2) agarose or anti-HA agarose was added intothe whole-cell lysate and incubated at 4°C for2 hours. The agarose beads were washed threetimes with buffer D, and coprecipitated proteinswere detected by immunoblotting.To determine STING-IRF3 complex formation

in cells, HA-tagged IRF3 WT or 2A was stablyexpressed in the indicated cells. After HT-DNAtransfection, cells were homogenized in hy-potonic buffer (10 mM Tris-Cl [pH 7.5], 10 mMKCl, 0.5 mM EGTA, 1.5 mM MgCl2, and Roche

SCIENCE sciencemag.org 13 MARCH 2015 • VOL 347 ISSUE 6227 aaa2630-11

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

EDTA-free protease inhibitor cocktail). The ho-mogenates were centrifuged at 1000 × g for 5 minto pellet nuclei and unbroken cells (P1). The super-natant (S1) was subjected to centrifugation at5000 × g for 10 min to separate crude mitochon-drial pellet from cytosolic supernatant (S5). Anti-HA agarose beads were added into the S5 andincubated at 4°C for 2 hours. The agarose beadswere washed three times with buffer C, and copre-cipitatedproteinswere detectedby immunoblotting.To determine TRIF-IRF3 complex formation

in cells, tagged IRF3 and TRIF-N540 were tran-siently expressed in HEK293T cells. 24 hours af-ter transfection, whole-cell lysates were preparedin the presence of buffer E [20 mM Tris-HCl (PH7.5), 150 mM NaCl, 10% glycerol, 0.5% CHAPS]and the protease inhibitor cocktail (Roche). Anti-Flag (M2) agarose or anti-HA agarose beadswereadded into the whole-cell lysates and incubatedat 4°C for 2 hours. The agarose beads werewashed three times with buffer E, and coprecipi-tated proteins were detected by immunoblotting.

Quantitative reverse transcriptionpolymerase chain reaction (q-RT-PCR)

Total RNAwas isolated using TRIzol (Invitrogen).0.1 mg total RNA was reverse-transcribed intocDNA using iScript Kit (Bio-Rad). The resultingcDNA served as the template for quantitativePCR analysis using iTaq Universal SYBR GreenSupermix (Bio-Rad) and Real-Time PCR System(ABI). Primers for specific genes are listed as fol-lows:mouse b-actin, 5′-TGACGTTGACATCCGTAA-AGACC-3′ and 5′-AAGGGTGTAAAACGCAGCTCA-3′;mouse IFN-b, 5′CCCTATGGAGATGACGGAGA-3′and 5′-CTGTCTGCTGGTGGAGTTCA-3′; mouseIFN-a, 5′-ATTTTGGATTCCCCTTGGAG-3′ and5′-TATGTCCTCACAGCCAGCAG-3′.

Purification of MAVS, STING, and TRIFfor phosphorylation site identification bymass spectrometry

For human MAVS, HEK293T cells were stablyinfectedwith a lentiviral vector, pTY-U6-shMAVS-EF1A-Flag-MAVS-2R, which depleted endogenousMAVS by shRNA and replaced it with a Flag-taggedMAVS containing two amino acids substi-tution (A440R/M449RorMAVS-2R),whichpermitstrypsin digestion and therefore facilitates phos-phorylation site mapping by mass spectrometry(see below). The cells were infected with Sendaivirus to activate the RIG-I–MAVS pathway. ForSTING, HT-DNA was transfected into L929 cellswith endogenous STING depleted by shRNA andreplacedwithmouseSTING-Flag (pTY-U6-shSTING-EF1A-mSTING-flag). For TRIF, HA-TRIF-N540was transfected alone or cotransfectedwith Flag-TBK1 in HEK293T cells. Cells were then har-vested in PBS containing 3 mM EDTA, followedby addition of 1% SDS to denature proteins. Thedenatured cell lysates were sonicated to shearDNA and boiled at 95°C for 10 minutes. ExcessSDS was removed with SDS-OUT precipitationkit fromPierce and Flag-IP forMAVS and STINGor HA-IP for TRIF was carried out in the presenceof buffer A [20 mM Tris-HCl (pH 7.0), 1M NaCl,0.5% NP-40] at 4°C overnight. The anti-Flag M2

agarose beads were thenwashed three times withbuffer A and buffer C. Bound STING or MAVSproteins were eluted with the Flag peptide. Elutedproteins were buffer-exchanged into 50 mMNH4HCO3, reduced in 5 mM DTT, and alkylatedin 2.7 mM iodoacetamide. After trypsin diges-tion (1:20) at 30°C overnight, the peptidemixturewas acidified with 1% formic acid, purified withC18 Zip-tip (Millipore), and then analyzed by nano–liquid chromatography–mass spectrometry (nano-LC-MS). ForHA-TRIF, anti-HA agarose beadswithbound proteins were directly boiled in SDS sam-ple buffer containing 2% SDS and subjected toSDS-PAGE and Coomassie-blue staining. Gel slicesnear 75 kDwere subsequently subjected to trypsindigestion and C18 Zip-tip purification, followed bynano-LC-MS analysis.

Mass spectrometry

Mass spectrometry analyses and targeted quan-tification of tryptic MAVS, STING, or TRIF pep-tideswere conducted on a Dionex Ultimate 3000nanoLC system coupled to a Q-Exactive massspectrometer (Thermo Scientific). The LC con-ditions and ion source parameters have beendescribed before (4, 9). For tandemMS/MS analy-ses, full-scan mass spectra were acquired in therange of mass/charge ratio (m/z) = 300 to 1500,with a resolution of 70,000 at m/z = 200 in theOrbitrap. MS/MS spectra (resolution: 17,500 atm/z = 200) were acquired in a data-dependentmode whereby the top 15 most abundant parentions were subjected to further fragmentation byhigher-energy collision dissociation (HCD). Phos-phorylated and nonphosphorylated TRIF S210–containing peptides were directly analyzed andquantified with Xcalibar 2.2 (Thermo Scientific),according to the specific ions indicated in table S1.To quantify phosphorylated and nonphospho-

rylatedMAVS and STING peptides, targeted SIMand targeted MS2 assays were developed on theQ-Exactivemass spectrometer. The precursor ionsin the inclusion list for both of the assays areshown in table S1. Settings for targeted SIMwere:resolution, 70,000; AGC target, 5E4; maximuminjection time, 250ms; isolation window, 0.5m/z.The targeted MS2 settings were as follows: res-olution, 35,000; AGC target, 2E4; maximum in-jection time, 120 ms; isolation window, 0.5 m/z;and normalized collision energy, 30. Data ac-quisition and analyses were performed withXcalibar 2.2. The relative abundance of eachpeptide or site-specific phosphorylation on thesame peptide was represented by the intensityof product ions that are specific to each phos-phorylation site (see table S1).

Targeted mass spectrometryidentifies MAVS phosphorylationsites in virus-infected cells

The MAVS C-terminal region containing the 3Sand 4T/S sites (aa 420 to 460) lacks lysine andarginine (fig. S2A). To facilitate trypsin digestionand mass spectrometry analysis, a Flag-taggedMAVS mutant (A440R/M449R or MAVS-2R) wasintroduced into a HEK293T cell line that wasdepleted of endogenous MAVS by shRNA. The

MAVS-2R mutant robustly rescued IRF3 dimer-ization after viral infection (fig. S4C), indicatingthat the substitutions do not alter the function ofMAVS. Trypsin digestion of purified Flag-MAVS-2Rfrom cells created a short peptide (“ISASTSLGR”)including the 4T/S site (4T/S peptide), whichwasidentified by nano-LC-MS. Following viral infec-tion, we observed substantially increased abun-dance of an ion with m/z of 486.23 (z = 2+), thefragmentation pattern of which matched to sin-gly phosphorylated 4T/S peptides (fig. S4D). Fur-thermore, by using targeted quantification, weobserved robust induction of singly phosphoryl-ated 4T/S peptides from the virus-infected sample(Fig. 3B and fig. S4D). The singly phosphorylatedpeptides eluted as two peaks (peak a and b). Byexamining site-localizing fragment ions specificto eachphospho-serine in theMS2 spectra,we foundthat peak a contains mainly p-S446 peptides,whereas peak b is amixture of p-S442 and p-S444peptides (fig. S4, E and F). Both phospho-S442and total phospho-4T/S signals were induced byvirus infection (Fig. 3B).

Targeted mass spectrometry revealsphosphorylation of mouse STING atSer365 in DNA-stimulated cells

We generated L929 cells stably expressing aC-terminal Flag-tagged mouse STING (humanSTING C-terminus lacks arginine or lysine thatcan be cleaved by trypsin) in place of endogenousSTING, which was depleted by shRNA. In thesecells, HT-DNA transfection resulted in strongIRF3 activation, which was completely blockedby treatment with the TBK1 inhibitor, BX795 (fig.S8F). Tryptic digestion of purified mSTING-Flagfrom the DNA-transfected cells led to the iden-tification of a peptide (“LLIS365GMDQPLPLR”)covering the S365 site (S366 in human STING)by nano-LC-MS. Following virus infection, we ob-served substantially increased abundance of anion with m/z of 766.90 (z = 2+), which was con-firmed to be phosphorylated S365 peptides byHCD fragmentation. Furthermore, by using tar-geted quantification (table S1), we observed ro-bust induction of the phosphorylated S365 peptideonly from the DNA-transfected sample (Fig. 5C).

Kinase inhibitors for cell-free assay orcell-based experiments

The kinase inhibitors were dissolved in dimethylsulfoxide (DMSO) and used both in vitro and incells at the following final concentrations: TBK1inhibitor (BX795, Selleckchem), 4 mM; IKK inhib-itor (TPCA-1, Sigma), 20 mM; and PLK1 inhibitor(BI2356, Selleckchem), 20 mM. For cell-free assays,the kinase inhibitors were incubated with the re-action mixtures at 30°C, and the final DMSOconcentrationwas kept below 5% of total reactionvolume. In cell-based experiments, the inhibitorswere added 1 hour before viral infection, HT-DNA,or TRIF transfection.

Immunofluorescence

BJ cells with or without HT-DNA transfectionwere fixed and incubated with the STING p-S366–specific antibody, followedby a secondary antibody

aaa2630-12 13 MARCH 2015 • VOL 347 ISSUE 6227 sciencemag.org SCIENCE

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

conjugated with Alexa Fluor 488. Green fluores-cent signals were imaged by a Zeiss LSM710 con-focalmicroscope (Carl Zeiss). Nuclei were stainedwith 4′,6-diamidino-2-phenylindole (DAPI) in themounting medium (Vector Labs).

Proximity ligation assay (PLA)

Proximity ligation assays to detect STING phos-phorylation and STING-IRF3 interaction wereconducted using the Duolink In Situ Red KitMouse/Rabbit (Sigma-Aldrich). L929 cells stablyexpressing huSTING-Flag and HA-IRF3 2A withor without HT-DNA transfection were fixed andincubatedwithprimaryantibodiesagainst twopro-teins as indicated; secondary antibodies conju-gatedwith plus andminus PLAprobeswere thenadded. After ligation, rolling circle amplification,and hybridization with fluorescently labeledoligonucleotides, red fluorescent dots (indicatingclose proximity of proteins recognized by twodifferent primary antibodies) were imaged by aZeiss LSM710 confocal microscope. Cell shapeswere indicated by phalloidin-labeled actin fila-ments, and nuclei were stained with DAPI in themounting medium.

Computational biology

Combining the experimental data and sequenceprofiles of IRF3, MAVS, STING, and TRIF (fig.S11), we derived that an IRF3 binding motif har-bors the following sequence features: pLxIS (p,hydrophilic; x, nonaromatic; S, phosphorylated)and another serine surrounded by a hydrophobicresidue within the upstream 15 amino acids. Start-ing from all of the 99,459 alternatively splicedisoforms of human proteins in the Ensembldatabase (38), we identified 4229 motifs withthis pattern. Requiring thismotif to be conservedamong the 43mammalian species (table S3A), wenarrowed this list down to 403 motifs from 399isoforms. We required 90% of sequences to followthe sequence pattern for the most conservedpositions (the second and the last two positions),whereas for those more variable positions, weallowed 20% of sequences to be amino acids thatwere not suggested as permissible in experi-mental data and sequence alignments of IRF3,MAVS, STING, and TRIF.The observation that this motif tends to be in

the flexible linkers (usually predicted to be dis-ordered) between domains helped us to furtherreduce the number of candidates. We detecteddomains in these proteins using HMMER (39)against Pfam domains (40) and predicted thedisordered regions using the EspritzWeb server(41) with parameters trained on disordered pro-teins in the Disprot database (42). We identified72 candidate motifs from 72 protein isoformsthat were largely in the linkers between domainsand that were predicted to be largely disordered.All of these candidate protein isoforms were fur-ther mapped to the Uniprot entries (43), result-ing in the final candidate list of 21 protein-codinggenes (table S3B).We extracted the gene ontology (GO) terms

(44) in the category of “biological process” asso-ciatedwith these candidate proteins fromUniprot.

The enriched GO terms associated with theseproteins were detected with a binomial test, andthemost significantly enrichedGO terms are listedin table S3C. In addition, we used the number ofshared GO terms to represent the functional sim-ilarity between proteins and clustered all of thecandidates in CLANS (45) based on similarity infunction; the result is shown in fig. S12. We fur-ther assigned a confidence level of these motifsbased on: (i) how well they overlap with domainlinkers and disordered regions and (ii) additionalsequence features we observed in IRF3, MAVS,STING, and TRIF (i.e., the presence of multipleserine residues upstream of the motif and thepreference of negatively charged versus positivelycharged residues in the motif). The confidencelevel of these motifs is reflected by the color codein fig. S12.The largest cluster of proteins containing the

consensus motif and other features describedabove is that related to innate immune responses(fig. S12). Statistical tests also support significantenrichment in several GO terms related to innateimmune response. In addition to MAVS, STING,TRIF, and IRF3, the proteins identified from thisanalysis include IRF5, which is important forproinflammatory cytokine induction by multiplepathways; brain-specific angiogenesis inhibitor1–associatedprotein2 (BAIP2_HUMANinUniprot);and dual-specificity mitogen-activated protein ki-nase kinase 4 (MP2K4_HUMAN in Uniprot). Theroles and regulations of these and other proteinsidentified from our computational analysis re-quire further studies.

REFERENCES AND NOTES

1. S. Akira, S. Uematsu, O. Takeuchi, Pathogen recognition andinnate immunity. Cell 124, 783–801 (2006). doi: 10.1016/j.cell.2006.02.015; pmid: 16497588

2. M. Yoneyama, T. Fujita, RNA recognition and signaltransduction by RIG-I-like receptors. Immunol. Rev. 227,54–65 (2009). doi: 10.1111/j.1600-065X.2008.00727.x;pmid: 19120475

3. J. Wu et al., Cyclic GMP-AMP is an endogenous secondmessenger in innate immune signaling by cytosolic DNA.Science 339, 826–830 (2013). doi: 10.1126/science.1229963;pmid: 23258412

4. L. Sun, J. Wu, F. Du, X. Chen, Z. J. Chen, Cyclic GMP-AMPsynthase is a cytosolic DNA sensor that activates the type Iinterferon pathway. Science 339, 786–791 (2013).pmid: 23258413

5. A. Pichlmair, C. Reis e Sousa, Innate recognition of viruses.Immunity 27, 370–383 (2007). doi: 10.1016/j.immuni.2007.08.012; pmid: 17892846

6. D. B. Stetson, R. Medzhitov, Recognition of cytosolic DNAactivates an IRF3-dependent innate immune response.Immunity 24, 93–103 (2006). doi: 10.1016/j.immuni.2005.12.003; pmid: 16413926

7. X. Jiang et al., Ubiquitin-induced oligomerization of the RNAsensors RIG-I and MDA5 activates antiviral innate immuneresponse. Immunity 36, 959–973 (2012). doi: 10.1016/j.immuni.2012.03.022; pmid: 22705106

8. F. Hou et al., MAVS forms functional prion-like aggregatesto activate and propagate antiviral innate immune response.Cell 146, 448–461 (2011). doi: 10.1016/j.cell.2011.06.041;pmid: 21782231

9. S. Liu et al., MAVS recruits multiple ubiquitin E3 ligases toactivate antiviral signaling cascades. eLife 2, e00785 (2013).doi: 10.7554/eLife.00785; pmid: 23951545

10. W. Zeng et al., Reconstitution of the RIG-I pathwayreveals a signaling role of unanchored polyubiquitin chainsin innate immunity. Cell 141, 315–330 (2010). doi: 10.1016/j.cell.2010.03.029; pmid: 20403326

11. W. Zeng, M. Xu, S. Liu, L. Sun, Z. J. J. Chen, Key role ofUbc5 and lysine-63 polyubiquitination in viral activation ofIRF3. Mol. Cell 36, 315–325 (2009). doi: 10.1016/j.molcel.2009.09.037; pmid: 19854139

12. S. Liu, Z. J. Chen, Expanding role of ubiquitination in NF-kBsignaling. Cell Res. 21, 6–21 (2011). doi: 10.1038/cr.2010.170;pmid: 21135871

13. K. Clark, L. Plater, M. Peggie, P. Cohen, Use of thepharmacological inhibitor BX795 to study the regulation andphysiological roles of TBK1 and IkappaB kinase epsilon: Adistinct upstream kinase mediates Ser-172 phosphorylationand activation. J. Biol. Chem. 284, 14136–14146 (2009).doi: 10.1074/jbc.M109.000414; pmid: 19307177

14. L. Deng et al., Activation of the IkappaB kinase complex byTRAF6 requires a dimeric ubiquitin-conjugating enzyme complexand a unique polyubiquitin chain. Cell 103, 351–361 (2000).doi: 10.1016/S0092-8674(00)00126-4; pmid: 11057907

15. Single-letter abbreviations for the amino acid residues areas follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly;H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln;R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.

16. R. Lin, C. Heylbroeck, P. M. Pitha, J. Hiscott, Virus-dependentphosphorylation of the IRF-3 transcription factor regulatesnuclear translocation, transactivation potential, andproteasome-mediated degradation. Mol. Cell. Biol. 18,2986–2996 (1998). pmid: 9566918

17. M. Yoneyama et al., Direct triggering of the type I interferonsystem by virus infection: Activation of a transcription factorcomplex containing IRF-3 and CBP/p300. EMBO J. 17, 1087–1095(1998). doi: 10.1093/emboj/17.4.1087; pmid: 9463386

18. W. Suhara et al., Analyses of virus-induced homomeric andheteromeric protein associations between IRF-3 andcoactivator CBP/p300. J. Biochem. 128, 301–307 (2000).doi: 10.1093/oxfordjournals.jbchem.a022753; pmid: 10920266

19. R. Lin, Y. Mamane, J. Hiscott, Structural and functional analysisof interferon regulatory factor 3: Localization of thetransactivation and autoinhibitory domains. Mol. Cell. Biol. 19,2465–2474 (1999). pmid: 10082512

20. B. Y. Qin et al., Crystal structure of IRF-3 reveals mechanismof autoinhibition and virus-induced phosphoactivation.Nat. Struct. Biol. 10, 913–921 (2003). doi: 10.1038/nsb1002;pmid: 14555996

21. K. Takahasi et al., X-ray crystal structure of IRF-3 and itsfunctional implications. Nat. Struct. Biol. 10, 922–927 (2003).doi: 10.1038/nsb1001; pmid: 14555995

22. Y. Shi, J. Massagué, Mechanisms of TGF-beta signaling fromcell membrane to the nucleus. Cell 113, 685–700 (2003).doi: 10.1016/S0092-8674(03)00432-X; pmid: 12809600

23. H. Ishikawa, G. N. Barber, STING is an endoplasmic reticulumadaptor that facilitates innate immune signalling. Nature 455,674–678 (2008). doi: 10.1038/nature07317; pmid: 18724357

24. Y. Tanaka, Z. J. Chen, STING specifies IRF3 phosphorylation byTBK1 in the cytosolic DNA signaling pathway. Sci. Signal. 5,ra20 (2012). doi: 10.1126/scisignal.2002521; pmid: 22394562

25. H. Konno, K. Konno, G. N. Barber, Cyclic dinucleotidestrigger ULK1 (ATG1) phosphorylation of STING to preventsustained innate immune signaling. Cell 155, 688–698 (2013).doi: 10.1016/j.cell.2013.09.049; pmid: 24119841

26. H. Ishikawa, Z. Ma, G. N. Barber, STING regulates intracellularDNA-mediated, type I interferon-dependent innate immunity.Nature 461, 788–792 (2009). doi: 10.1038/nature08476;pmid: 19776740

27. M. Yamamoto et al., Role of adaptor TRIF in the MyD88-independent toll-like receptor signaling pathway. Science301, 640–643 (2003). doi: 10.1126/science.1087262;pmid: 12855817

28. K. J. Han et al., Mechanisms of the TRIF-induced interferon-stimulated response element and NF-kappaB activation andapoptosis pathways. J. Biol. Chem. 279, 15652–15661 (2004).doi: 10.1074/jbc.M311629200; pmid: 14739303

29. M. Yamamoto et al., Cutting edge: A novel Toll/IL-1 receptordomain-containing adapter that preferentially activates theIFN-beta promoter in the Toll-like receptor signaling.J. Immunol. 169, 6668–6672 (2002). doi: 10.4049/jimmunol.169.12.6668; pmid: 12471095

30. J. W. Wu et al., Crystal structure of a phosphorylatedSmad2. Recognition of phosphoserine by the MH2 domain andinsights on Smad function in TGF-beta signaling. Mol. Cell 8,1277–1289 (2001). doi: 10.1016/S1097-2765(01)00421-X;pmid: 11779503

31. B. Chen et al., The WAVE regulatory complex links diversereceptors to the actin cytoskeleton. Cell 156, 195–207 (2014).doi: 10.1016/j.cell.2013.11.048; pmid: 24439376

SCIENCE sciencemag.org 13 MARCH 2015 • VOL 347 ISSUE 6227 aaa2630-13

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

32. J. Pei, B. H. Kim, N. V. Grishin, PROMALS3D: A tool formultiple protein sequence and structure alignments. NucleicAcids Res. 36, 2295–2300 (2008). doi: 10.1093/nar/gkn072;pmid: 18287115

33. M. Lopez-Pelaez et al., Protein kinase IKKb-catalyzedphosphorylation of IRF5 at Ser462 induces its dimerizationand nuclear translocation in myeloid cells. Proc. Natl. Acad.Sci. U.S.A. 111, 17432–17437 (2014). doi: 10.1073/pnas.1418399111; pmid: 25326418

34. J. Ren, X. Chen, Z. J. Chen, IKKb is an IRF5 kinase thatinstigates inflammation. Proc. Natl. Acad. Sci. U.S.A. 111,17438–17443 (2014). doi: 10.1073/pnas.1418516111;pmid: 25326420

35. R. B. Seth, L. Sun, C. K. Ea, Z. J. Chen, Identification andcharacterization of MAVS, a mitochondrial antiviralsignaling protein that activates NF-kappaB and IRF 3. Cell122, 669–682 (2005). doi: 10.1016/j.cell.2005.08.012;pmid: 16125763

36. J. J. Woodward, A. T. Iavarone, D. A. Portnoy, c-di-AMPsecreted by intracellular Listeria monocytogenesactivates a host type I interferon response. Science 328,1703–1705 (2010). doi: 10.1126/science.1189801;pmid: 20508090

37. O. Shalem et al., Genome-scale CRISPR-Cas9 knockoutscreening in human cells. Science 343, 84–87 (2014).doi: 10.1126/science.1247005; pmid: 24336571

38. P. Flicek et al., Ensembl 2014. Nucleic Acids Res. 42,D749–D755 (2014). pmid: 24316576

39. R. D. Finn, J. Clements, S. R. Eddy, HMMER web server:Interactive sequence similarity searching. Nucleic Acids Res.39 (suppl. 2), W29–W37 (2011). doi: 10.1093/nar/gkr367;pmid: 21593126

40. R. D. Finn et al., Pfam: The protein families database. NucleicAcids Res. 42, D222–D230 (2014). doi: 10.1093/nar/gkt1223;pmid: 24288371

41. I. Walsh, A. J. Martin, T. Di Domenico, S. C. Tosatto, ESpritz:Accurate and fast prediction of protein disorder. Bioinformatics28, 503–509 (2012). doi: 10.1093/bioinformatics/btr682;pmid: 22190692

42. M. Sickmeier et al., DisProt: The database of disorderedproteins. Nucleic Acids Res. 35 (suppl. 1), D786–D793 (2007).doi: 10.1093/nar/gkl893; pmid: 17145717

43. UniProt Consortium, Activities at the Universal ProteinResource (UniProt). Nucleic Acids Res. 42, D191–D198 (2014).doi: 10.1093/nar/gkt1140; pmid: 24253303

44. Gene Ontology Consortium, Gene Ontology annotations andresources. Nucleic Acids Res. 41, D530–D535 (2013).doi: 10.1093/nar/gks1050; pmid: 23161678

45. T. Frickey, A. Lupas, CLANS: A Java application for visualizingprotein families based on pairwise similarity. Bioinformatics 20,3702–3704 (2004). doi: 10.1093/bioinformatics/bth444;pmid: 15284097

46. J. Pei, N. V. Grishin, AL2CO: Calculation of positional conservationin a protein sequence alignment. Bioinformatics 17, 700–712(2001). doi: 10.1093/bioinformatics/17.8.700; pmid: 11524371

ACKNOWLEDGMENTS

We thank Y. Yu, H. Mirzae, and C. Long for advice on massspectrometry; L. Jia for advice on generating phospho-specificantibodies; and L. Sun, N. Varnado, and M. Xu for critically readingour manuscript. The data presented in this manuscript aretabulated in the main paper and in the supplementary materials.This work was supported by grants from the NIH (AI-93967 andGM-63692 to Z.J.C. and GM-094575 to N.G.) and the WelchFoundation (I-1389 to Z.J.C. and I-1505 to N.G.). X.Cai,J.W., and Q.C. were supported by International Student Fellowshipsfrom HHMI. Z.J.C. is an HHMI Investigator.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/347/6227/aaa2630/suppl/DC1Figs. S1 to S12Tables S1 to S4References (47–49)

9 November 2014; accepted 16 January 2015Published online 29 January 2015;10.1126/science.aaa2630

aaa2630-14 13 MARCH 2015 • VOL 347 ISSUE 6227 sciencemag.org SCIENCE

RESEARCH | RESEARCH ARTICLEon M

arch 15, 2020

http://science.sciencemag.org/

Dow

nloaded from

activationPhosphorylation of innate immune adaptor proteins MAVS, STING, and TRIF induces IRF3

Zhijian J. ChenSiqi Liu, Xin Cai, Jiaxi Wu, Qian Cong, Xiang Chen, Tuo Li, Fenghe Du, Junyao Ren, You-Tong Wu, Nick V. Grishin and

originally published online January 29, 2015DOI: 10.1126/science.aaa2630 (6227), aaa2630.347Science 

, this issue 10.1126/science.aaa2630Scienceproduction carefully to avoid inflammation and autoimmunity.

IFNreceptor-adaptor protein pairs to activate IRF3 and generate IFNs. This is important because cells must regulate their now report a common signaling mechanism used by all three types of innate immuneet al.host defense. Liu

protein kinase TBK1 and the transcription factor IRF3, which tells cells to secrete interferon proteins (IFNs) important forimmune system to an infection. Each receptor type signals through a different adapter protein. These signals activate the

Innate immune receptors such as RIG-I, cGAS, and Toll-like receptors bind microbial fragments and alert theInnate immune receptor signaling, united

ARTICLE TOOLS http://science.sciencemag.org/content/347/6227/aaa2630

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2015/01/28/science.aaa2630.DC1

CONTENTRELATED

http://stke.sciencemag.org/content/sigtrans/10/488/eaah5054.fullhttp://stke.sciencemag.org/content/sigtrans/9/442/ra85.fullhttp://stke.sciencemag.org/content/sigtrans/8/388/ra78.fullhttp://stke.sciencemag.org/content/sigtrans/9/456/ra115.fullhttp://stke.sciencemag.org/content/sigtrans/10/477/eaan5400.fullhttp://stke.sciencemag.org/content/sigtrans/10/476/eaah4248.fullhttp://stke.sciencemag.org/content/sigtrans/10/460/eaae0435.fullhttp://stke.sciencemag.org/content/sigtrans/9/454/ec267.abstracthttp://stke.sciencemag.org/content/sigtrans/9/429/ec121.abstracthttp://stke.sciencemag.org/content/sigtrans/8/390/ec229.abstracthttp://stke.sciencemag.org/content/sigtrans/8/385/ra69.fullhttp://stke.sciencemag.org/content/sigtrans/7/307/ra3.fullhttp://stke.sciencemag.org/content/sigtrans/8/368/ec64.abstracthttp://stke.sciencemag.org/content/sigtrans/4/187/pe39.fullhttp://stke.sciencemag.org/content/sigtrans/8/366/ra25.fullhttp://stke.sciencemag.org/content/sigtrans/5/214/ra20.full

REFERENCES

http://science.sciencemag.org/content/347/6227/aaa2630#BIBLThis article cites 48 articles, 15 of which you can access for free

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Copyright © 2015, American Association for the Advancement of Science

on March 15, 2020

http://science.sciencem

ag.org/D

ownloaded from

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Copyright © 2015, American Association for the Advancement of Science

on March 15, 2020

http://science.sciencem

ag.org/D

ownloaded from