igem poster regionals final draft2012.igem.org/files/poster/berkeley.pdf · 2012. 10. 14. · all...

1
Abstract MiCode Design Automation Motivation Orthogonal Interaction Parts MiCoding a Library Conclusion Constructed a proof of principle protein-protein interaction assay The more generalized equation for calculating the possible set of MiCodes. MiCodes Many applications in synthetic biology demand precise control over subcellular localization, cell morphology, motility, and other such phenotypes that are only observable via microscopy. At present, engineering these properties is challenging due in large part to the inherent throughput limitation imposed by microscopy. We have developed a strategy that enables high-throughput library screening with microscopy by coupling a unique fluorescence signature with each genotype present in a library. These MiCodes (microscopy barcodes) are generated by targeting combina- tions of fluorophores to several organelles within yeast, and they eliminate the need to isolate and observe clonal populations separately. MiCodes can potentially scale to library sizes of 10^6 or more, and their analysis can be largely automated using existing image processing software. As a proof of principle, we applied MiCodes to the problem of finding unique pairs of protein-protein interaction parts. (#States) (#Channels) (#Organelles) = # unique MiCodes Undergraduates: Robert Chen, Celia Cheung, Thomas Chow, Austin Jones, Harneet Rishi, Masaki Yamada, Vincent Yeh Advisors: Will DeLoache, Terry Johnson, John Dueber MiCodes UC BERKELEY By localizing fluorescent proteins to specific organelles, each cell can be given a “microscopic barcode,” or MiCode. Each member of a library will be assigned a unique MiCode, distinguishing it from the rest of the library and tying the MiCode phenotype to a specific genotype. To do this, we needed to decide which organelles to target and how to target them. We also had to optimize our choice of promoters so that no signal was too strong or too weak in comparison to the rest. ( ) =4,096 2 3 4 ( ) =1,048,576 2 4 5 Synthetic biologists often use engineered protein-protein interactions when designing their systems. In particu- lar, interaction pairs have been applied to both scaffolding and signaling applications with impressive results. In such systems, orthogonality of protein pair interactions is desired to reduce cross-talk between system components and allow for more precise control. However, the limited number and diversity of well-characterized, orthogonal protein interaction pairs restricts the complexity of systems that can be designed. Acknowledgements Characterized current registry yeast promoters pSTE5, pCYC1, pADH1, pTDH3 Successfully designed a new automatable technique to screen libraries by linking genotypes to phenotypes visible through microscopy Developed a quick, modular way to construct MiCode sets of over 10^6 variants In the figures above we can see a single sample MiCode of red nucleus, red vaculolar membrane, blue plasma membrane and green actin. By adding one more organelle and one more fluorescent protein, the available MiCode set size increases by 3 orders of magnitude! Promoter Characterization Results: Our software proved capable of successfully identifying MiCoded nuclei 95% of the time when processing our sample image size (100 cells). We can also differentiate between fluorescent membranes and punc- tate structures in a single channel. In the future, with a large, already-constructed dataset (a complete MiCode library), we would likely employ machine learning tech- niques to improve the automated identification. General workflow: 1) Separate fluorescent channels. 2) Convert to grayscale image. 3) Detect high intensity pixel blobs. 4) Convert to pseudo color image, determine geometric properties of each blob. 5) Match properties with measurements we determined for each organelle and construct MiCode. Wrote software to identify MiCoded nuclei with a 95% success rate We chose to use Golden Gate Assembly because it allowed us to create libraries of multiple parts in a single cloning step. In our MiCode Assembly (MCA), we utilized two type IIs enzymes, BsaI and BsmBI. By alternating the usage of these two enzymes in each round of assembly and reintroducing the sites when necessary, we can combine many parts quickly and efficiently. We designed MCA such that parts could be interchanged and many MiCodes could be made in parallel. In the final one-pot round of assembly, we combine several multigene cassettes together to form MiCoded zipper pairs. We combine the bait and prey libraries together in a large single-pot reaction. Each zipper is on a multigene cassette along with known MiCode components. When combined, each pairing of bait and prey comes together to produce a full, unique MiCode. Micode set size can be expanded by varying: States: A general binary system consists of fluorescent pro- teins being either be on or off. Additionally, UV-activated fluorescent proteins form a ternary system (on, off , or photo- activatable) Channels: We currently utilize three channels: red, green and blue. Other non-overlapping fluorophores are also avail- able. Organelles: We optimized the use of four organelles (actin, plasma membrane, nucleus and vacuolar membrane) for our project but additional organelles can be utilized as well. We chose these four locations from fourteen potential candidates based on the following criteria: 1. Minimal cell-to-cell variability of the localized signal. 2. Localized signal distinguishable from the other chosen locations. In order to design these MiCodes we had to determine which organelles we could easily and distinctly target in order to produce our MiCodes. We referenced the YeastGFP protein localization database from the O'Shea and Weissman labs at UCSF. We found proteins that localized to various organelles, PCR'd them off of the genome, and tagged them with a fluorescent protein. "Prey": The prey construct has a leucine zipper fused to mKate, a monomeric red fluorescent protein (RFP). "Bait": The bait construct has a different leucine zipper fused to a peroxisome targeting signal sequence (PTS1). The two constructs shown in the figure above are combined into a plasmid and integrated into yeast to create unique strains. Once expressed in the cytosol, the bait zipper will be recruited to the per- oxisome by the PTS1. A range of possible phenotypes can be observed: We wrote software to quantitate the affinity of three previously characterized leucine zipper pairs (Keating, MIT). Our metric for gauging zipper-interaction was amount of mKate in peroxisome divided by amount of mKate in cytosol. To test the MiCode application for the leucine zipper assay we created two MiCodes- one for the known strong binding pair (20+2) and one for a known weak binding pair (20+13). MiCode 1 identifies cells with the strongly binding leucine zipper pair via a green nucleus and blue plasma membrane. Micode 2 identifies cells with the weakly binding leucine zipper pair via a blue nucleus and green plasma membrane. Above are the resulting images from this proof of concept demonstration. “Prey” “Bait” Zipper B Zipper A Weak Strong To accommodate large libraries we need to automate image acquisition and MiCode identification. Image acquisition can be automated through the use of microscopes with automated stages. To automate MiCode iden- tification, we programmed scripts in Matlab and Cell Profiler. Cell segmentation allows for cell-by-cell analysis. We wrote scripts in Matlab and achieved an identification rate of 96 percent for 400 sample images. Microscopy is a versatile imaging technology that allows us to observe various phenotypes such as subcellular localization, cell morphology and motility that are not easily visualized using other techniques. Systems engineered using these cellular features could benefit from genetic libraries, which allow sampling of large parameter spaces to elucidate optimal design parameters. However, library screening by microscopy is currently limited because it does not preserve a link between member genotypes among a mixed population. To screen mixed genotype library members at the same time, we need some method of linking member genotype to an observable phenotype on the microscope. We propose barcodes that couple a fluorescent signature with each library member’s genotype. Strong Binding: The prey zipper will be recruited to the peroxisome along with the bait, causing the yeast cell to have RFP concentrated in the peroxisome. No Binding: The prey zipper will remain in the cytosol producing diffuse RFP the cytosol. Weak Binding: Both punctate red in the peroxisomes and diffuse red fluorescent protein in the cytosol should be observed. Microscopes such as the Zeiss fluorescent micro- scope are commonplace in research laboratories How many unique MiCodes can we make? 5 Plasma Characterizing an orthogonal set of protein-protein interactions requires measuring the affinity of every member in the set with every other member. Traditional screening methods for protein-protein inter- actions (such as transcription-driven reporter expression in yeast 2- hybrid systems) are not optimized to gather data across a range of affini- ties or do not have high enough throughput to analyze large library sizes. With the utilization of MiCodes and the ability of microscopy to record spatial information, we were able to design an assay to directly observe and screen for orthogonal protein-protein interactions. 1 2 Actin Plasma Membrane Plasma membrane targeting sequence mKate-CIIC Registry Part:BBa_K900005 Nucleus Vacuolar Membrane Vacuolar membrane localizing protein ZRC1-mTurquoise Registry Part: BBa_K900003 Actin localizing protein ABP1-Venus Registry Part:BBa_K900004 Nucleus localizing protein H2A2-Venus Registry Part: BBa_K900002 Bait LZ Prey LZ - RFP PM - CFP Actin- RFP VM - RFP Nuc - GFP Prey Half MiCode Bait Half MiCode Bait LZ Prey LZ - RFP Prey Half MiCode Bait Half MiCode PTS1 We found that different organelles exhibited varying levels of targeted fluorescence intensity due to their geometric shape. To optimize the expres- sion, we characterized several potential promoters already present in the registry. We cloned each pro- moter in front of two different fluorescent proteins to account for variability due to downstream se- quence. The device was cloned on a backbone with a Leu2 marker and a Cen6 origin of replication and transformed into S228C S. cerevisiae . Ideally, we wanted the calibration curve to span several orders of magnitude and exhibit a linear relationship, which was true for all the promoters we character- ized except for pCYC1. Calibration curve of fluorescence intensity. Error bars are +/- one standard deviation. YFP-yellow fluorescent protein, RFP- red fluorescent protein All microscopy images have been enhanced for ease of visualization on poster. 10 100 1000 10000 100000 10 100 1000 10000 100000 (YFP uorescence)/OD (RFP uorescence)/OD Yeast Promoter Characterizaon pTDH3 pADH1 pCYC1 pSTE5 control Keating Lab, MIT

Upload: others

Post on 13-Sep-2020

6 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: iGEM poster regionals final draft2012.igem.org/files/poster/Berkeley.pdf · 2012. 10. 14. · All microscopy images have been enhanced for ease of visualization on poster. 10 100

Abstract MiCode Design

Automation

Motivation

Orthogonal Interaction Parts

MiCoding a Library

Conclusion

Constructed a proof of principle protein-protein interaction assay

The more generalized equation for calculating the possible set of MiCodes.

MiCodes

Many applications in synthetic biology demand precise control over subcellular localization, cell morphology, motility, and other such phenotypes that are only observable via microscopy. At present, engineering these properties is challenging due in large part to the inherent throughput limitation imposed by microscopy. We have developed a strategy that enables high-throughput library screening with microscopy by coupling a unique fluorescence signature with each genotype present in a library. These MiCodes (microscopy barcodes) are generated by targeting combina-tions of fluorophores to several organelles within yeast, and they eliminate the need to isolate and observe clonal populations separately. MiCodes can potentially scale to library sizes of 10^6 or more, and their analysis can be largely automated using existing image processing software. As a proof of principle, we applied MiCodes to the problem of finding unique pairs of protein-protein interaction parts.

(#States)(#Channels)

(#Organelles)

= # unique MiCodes

Undergraduates:Robert Chen, Celia Cheung, Thomas Chow, Austin Jones, Harneet Rishi, Masaki Yamada, Vincent Yeh

Advisors:Will DeLoache, Terry Johnson, John DueberMiCodes

UC BERKELEY

By localizing fluorescent proteins to specific organelles, each cell can be given a “microscopic barcode,” or MiCode. Each member of a library will be assigned a unique MiCode, distinguishing it from the rest of the library and tying the MiCode phenotype to a specific genotype. To do this, we needed to decide which organelles to target and how to target them. We also had to optimize our choice of promoters so that no signal was too strong or too weak in comparison to the rest.

( ) =4,09623 4

( ) =1,048,57624 5

Synthetic biologists often use engineered protein-protein interactions when designing their systems. In particu-lar, interaction pairs have been applied to both scaffolding and signaling applications with impressive results. In such systems, orthogonality of protein pair interactions is desired to reduce cross-talk between system components and allow for more precise control. However, the limited number and diversity of well-characterized, orthogonal protein interaction pairs restricts the complexity of systems that can be designed.

Acknowledgements

Characterized current registry yeast promoters pSTE5, pCYC1, pADH1, pTDH3

Successfully designed a new automatable technique to screen libraries by linking genotypes to phenotypes visible through microscopy

Developed a quick, modular way to construct MiCode sets of over 10^6 variants

In the figures above we can see a single sample MiCode of red nucleus, red vaculolar membrane, blue plasma membrane and green actin.

By adding one more organelle and one more fluorescent protein, the available MiCode set size increases by 3 orders of magnitude!

Promoter Characterization

Results: Our software proved capable of successfully identifying MiCoded nuclei 95% of the time when processing our sample image size (100 cells). We can also differentiate between fluorescent membranes and punc-tate structures in a single channel. In the future, with a large, already-constructed dataset (a complete MiCode library), we would likely employ machine learning tech-niques to improve the automated identification.

General workflow: 1) Separate fluorescent channels. 2) Convert to grayscale image. 3) Detect high intensity pixel blobs. 4) Convert to pseudo color image, determine geometric properties of each blob. 5) Match properties with measurements we determined for each organelle and construct MiCode.

Wrote software to identify MiCoded nuclei with a 95% success rate

We chose to use Golden Gate Assembly because it allowed us to create libraries of multiple parts in a single cloning step. In our MiCode Assembly (MCA), we utilized two type IIs enzymes, BsaI and BsmBI. By alternating the usage of these two enzymes in each round of assembly and reintroducing the sites when necessary, we can combine many parts quickly and efficiently. We designed MCA such that parts could be interchanged and many MiCodes could be made in parallel.

In the final one-pot round of assembly, we combine several multigene cassettes together to form MiCoded zipper pairs. We combine the bait and prey libraries together in a large single-pot reaction. Each zipper is on a multigene cassette along with known MiCode components. When combined, each pairing of bait and prey comes together to produce a full, unique MiCode.

Micode set size can be expanded by varying:

States: A general binary system consists of fluorescent pro-teins being either be on or off. Additionally, UV-activated fluorescent proteins form a ternary system (on, off , or photo-activatable)Channels: We currently utilize three channels: red, green and blue. Other non-overlapping fluorophores are also avail-able.Organelles: We optimized the use of four organelles (actin, plasma membrane, nucleus and vacuolar membrane) for our project but additional organelles can be utilized as well.

We chose these four locations from fourteen potential candidates based on the following criteria:

1. Minimal cell-to-cell variability of the localized signal.2. Localized signal distinguishable from the other chosen locations.

In order to design these MiCodes we had to determine which organelles we could easily and distinctly target in order to produce our MiCodes. We referenced the YeastGFP protein localization database from the O'Shea and Weissman labs at UCSF. We found proteins that localized to various organelles, PCR'd them off of the genome, and tagged them with a fluorescent protein.

"Prey": The prey construct has a leucine zipper fused to mKate, a monomeric red fluorescent protein (RFP).

"Bait": The bait construct has a different leucine zipper fused to a peroxisome targeting signal sequence (PTS1).

The two constructs shown in the figure above are combined into a plasmid and integrated into yeast to create unique strains. Once expressed in the cytosol, the bait zipper will be recruited to the per-oxisome by the PTS1. A range of possible phenotypes can be observed:

We wrote software to quantitate the affinity of three previously characterized leucine zipper pairs (Keating, MIT). Our metric for gauging zipper-interaction was amount of mKate in peroxisome divided by amount of mKate in cytosol.

To test the MiCode application for the leucine zipper assay we created two MiCodes- one for the known strong binding pair (20+2) and one for a known weak binding pair (20+13). MiCode 1 identifies cells with the strongly binding leucine zipper pair via a green nucleus and blue plasma membrane. Micode 2 identifies cells with the weakly binding leucine zipper pair via a blue nucleus and green plasma membrane. Above are the resulting images from this proof of concept demonstration.

“Prey”

“Bait”Zipper B

Zipper A

Weak Strong

To accommodate large libraries we need to automate image acquisition and MiCode identification. Image acquisition can be automated through the use of microscopes with automated stages. To automate MiCode iden-tification, we programmed scripts in Matlab and Cell Profiler.

Cell segmentation allows for cell-by-cell analysis. We wrote scripts in Matlab and achieved an identification rate of 96 percent for 400 sample images.

Microscopy is a versatile imaging technology that allows us to observe various phenotypes such as subcellular localization, cell morphology and motility that are not easily visualized using other techniques. Systems engineered using these cellular features could benefit from genetic libraries, which allow sampling of large parameter spaces to elucidate optimal design parameters. However, library screening by microscopy is currently limited because it does not preserve a link between member genotypes among a mixed population. To screen mixed genotype library members at the same time, we need some method of linking member genotype to an observable phenotype on the microscope. We propose barcodes that couple a fluorescent signature with each library member’s genotype.

Strong Binding: The prey zipper will be recruited to the peroxisome along with the bait, causing the yeast cell to have RFP concentrated in the peroxisome.

No Binding: The prey zipper will remain in the cytosol producing diffuse RFP the cytosol.

Weak Binding: Both punctate red in the peroxisomes and diffuse red fluorescent protein in the cytosol should be observed.

Microscopes such as the Zeiss fluorescent micro-scope are commonplace in research laboratories

How many unique MiCodes can we make?

5Plasma

Characterizing an orthogonal set of protein-protein interactions requires measuring the affinity of every member in the set with every other member. Traditional screening methods for protein-protein inter-actions (such as transcription-driven reporter expression in yeast 2-hybrid systems) are not optimized to gather data across a range of affini-ties or do not have high enough throughput to analyze large library sizes. With the utilization of MiCodes and the ability of microscopy to record spatial information, we were able to design an assay to directly observe and screen for orthogonal protein-protein interactions.

1 2

ActinPlasma Membrane

Plasma membrane targeting sequence

mKate-CIICRegistry Part:BBa_K900005

Nucleus Vacuolar Membrane

Vacuolar membrane localizing protein

ZRC1-mTurquoise Registry Part: BBa_K900003

Actin localizing protein

ABP1-VenusRegistry Part:BBa_K900004

Nucleus localizing protein

H2A2-VenusRegistry Part: BBa_K900002

Bait LZPrey LZ - RFP

PM - CFP Actin- RFPVM - RFP Nuc - GFP

Prey Half MiCode Bait Half MiCode

Bait LZPrey LZ - RFPPrey Half MiCode Bait Half MiCode

PTS1

We found that different organelles exhibited varying levels of targeted fluorescence intensity due to their geometric shape. To optimize the expres-sion, we characterized several potential promoters already present in the registry. We cloned each pro-moter in front of two different fluorescent proteins to account for variability due to downstream se-quence. The device was cloned on a backbone with a Leu2 marker and a Cen6 origin of replication and transformed into S228C S. cerevisiae. Ideally, we wanted the calibration curve to span several orders of magnitude and exhibit a linear relationship, which was true for all the promoters we character-ized except for pCYC1. Calibration curve of fluorescence intensity.

Error bars are +/- one standard deviation.

YFP-yellow fluorescent protein, RFP- red fluorescent protein

All microscopy images have been enhanced for ease of visualization on poster.

10

100

1000

10000

100000

10 100 1000 10000 100000

(YFP

fluor

esce

nce)

/OD

(RFP fluorescence)/OD

Yeast Promoter Characterization

pTDH3pADH1pCYC1pSTE5control

Keating Lab, MIT