Dynamic Proteomics of Individual Cancer Cells in Response to a Drug A. A. Cohen,1*��� N. Geva-Zatorsky,1* E. Eden,1* M. Frenkel-Morgenstern,1 I. Issaeva,1 A. Sigal,2 R. Milo,3 C. Cohen-Saidon,1 Y. Liron,1 Z. Kam,1 L. Cohen,1 T. Danon,1 N. Perzov,1 U. Alon1 Why do seemingly identical cells respond differently to a drug? To address this, we studied the dynamics and variability of the protein response of human cancer cells to a chemotherapy drug, camptothecin. We present a dynamic-proteomics approach that measures the levels and locations of nearly 1000 different endogenously tagged proteins in individual living cells at high temporal resolution. All cells show rapid translocation of proteins specific to the drug mechanism, including the drug target (topoisomerase-1), and slower, wide-ranging temporal waves of protein degradation and accumulation. However, the cells differ in the behavior of a subset of proteins. We identify proteins whose dynamics differ widely between cells, in a way that corresponds to the outcomes���cell death or survival. This opens the way to understanding molecular responses to drugs in individual cells. Tandstate he of a cell is largely determined by the levels of thousands of proteins in space time (1���4). To affect the cell state, drugs are used (5���7), but little is known about the detailed effects of drugs on the dynamics of proteins in individual human cells. Here, we ask how a drug affects the dynamics of the proteome and how these dynamics differ for individual cells. To address this, our model system was hu- man cancer cells responding to an anticancer drug with a well-characterized target and mech- anism of action, camptothecin (CPT). This drug is a topoisomerase-1 (TOP1) poison with no other known targets. CPT locks TOP1 in a complex with the DNA, causing DNA breaks and inhibiting tran- scription, which eventually causes cell death (8). To follow the response to the drug, we en- deavored to accurately measure the level and localization of about 1000 proteins in individual cells over time. We found a diverse protein re- sponse to the drug, with rapid localization changes of proteins specific to the drug mecha- nism of action, followed by slower wide-ranging temporal changes of protein levels. Furthermore, we found that the drug target TOP1 is among the very first to respond both in protein level and localization. For most proteins, the response to the drug shows moderate cell-cell variability. De- viating from this norm is a set of proteins for which responses are widely different between individual cells. Some of these proteins are involved in cell fate decisions and at least two proteins show cell-to-cell differences that are correlated with the fate of the cell. Thus, examining spatiotemporal proteome dynamics in individual cells offers clues about what is special about the subpopulation of cells that escapes the drug action. Dynamic proteomics system. We used a retrovirus-based approach, called ���CD tagging,��� in human H1299 lung carcinoma cells (9���14). We constructed a library of over 1200 cell clones, each expressing a different fluorescently tagged, full-length protein from its endogenous chromo- somal location. Time-lapse fluorescence micros- copy was used to obtain movies of the proteins over several days of growth (15). Obtaining quantitative information from time- lapse fluorescent movies is known to be challeng- ing, because it is difficult to automatically detect the cell boundaries and track them over time (16���20). Here, we overcame this problem by a tagging strategy that made cells more easily iden- tifiable by image-analysis software. We used two rounds of CD tagging with the red fluorescent protein mCherry to obtain a cell clone with red fluorescence in the cytoplasm and stronger red fluorescence in the nucleus (Fig. 1A). Custom software (15) used the red fluorescence pattern to automatically distinguish the cell from its back- ground and to differentiate the nucleus from the cytoplasm (Fig. 1B). The algorithms in the soft- ware can also automatically detect morphological correlates of cell states [e.g., cell death and mitosis (15)]. We then used this clone (H1299-cherry) as a basis for our tagged protein library. We introduced an enhanced yellow fluorescent protein (eYFP or Venus) into the red-tagged cells by an additional round of CD tagging, expanded the yellow-tagged cells into clones, and identified the yellow-tagged proteins (13). Thus, the red tagging is the same in all cells of the library and is independent of the second yellow tag on the protein of interest. The library includes over 1260 different tagged proteins, of which about 80% are character- ized proteins and about 20% are not charac- terized (for a list of tagged proteins see www. dynamicproteomics.net). We excluded the pro- teins whose localization did not match previous reports (about one-sixth of the proteins) and studied the remaining 1020 proteins. These include diverse functional categories and localization patterns in- cluding membrane, nuclear, nucleolar, cytoskel- eton, Golgi, endoplasmic reticulum, and other cell locations (fig. S1). The CD tagging method we used tends to preserve protein functionality (13, 21, 22). Note, however, that our use of the library does not re- quire proteins to be functional, but merely to act as reliable reporters for the dynamics and location of the endogenous proteins. To test this, we mea- sured the dynamics of endogenous proteins using immunoblots with specific antibodies to 20 dif- ferent proteins in the parental H1299-cherry cells. In 80% of the cases (16 out of 20), the immu- noblot dynamics agreed with the fluorescence dynamics from the movies (Pearson correlation R 0.5, P 10-4) (fig. S2). Immunoblots of tagged cell clones with antibodies against green fluorescent protein (GFP) further indicated that the tagged proteins are full-length fusions (table S1). As in many high-throughput methods, we recommend that, when using the library to study specific proteins, protein functionality should be tested by other means. Most proteins named below were reannotated and tested as indicated. Assay of proteomic response to drug. Cells were grown in 12-well optical plates in an auto- mated fluorescence microscope with autofocus and control of temperature, CO2, and humidity. Each well contained cells tagged for a different protein. After 24 hours of growth, the drug CPT was added (10 mM), and cells were tracked for another 48 hours (Fig. 1C). Images in phase, red and yellow were taken every 20 min at four positions in each well. The resulting time-lapse movies had over 200 consecutive frames per protein studied, where each frame contained 10 to 40 different cells. Movies were stored and analyzed automatically (15), resulting in traces of protein level and location in each cell over time (see supporting material for sample movies). The cells showed vigorous divisions during the 24 hours before drug addition, with a cell cycle time of ~20 hours. When the drug was added, cells showed loss of motility and growth arrest after ~10 hours and began to show cell rounding and blebbing (morphological correlates of cell death), which reached about 15% of the cells after 36 hours (fig. S3). Cell cycle stage at the time the drug was added did not seem to influence the response to the drug, as assayed by automatic identification of cell division and cell death events (detailed in SOM text and figs. S4 and S5). In experiments in which the drug was washed away after 48 hours, a small fraction of the cells (about 10-4) survived to divide and form colonies after several weeks of incubation. Day-to-day repeats starting from frozen cells showed a mean error in the eYFP fluorescent RESEARCH ARTICLES 1Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot 76100, Israel. 2Division of Biology, California Institute of Technology, Pasadena, CA 91125, USA. 3Depart- ment of Plant Sciences, Weizmann Institute of Science, Rehovot 76100, Israel. *These authors contributed equally to this work. ���To whom correspondence should be addressed. E-mail: ariel.cohen@weizmann.ac.il www.sciencemag.org SCIENCE VOL 322 5 DECEMBER 2008 1511

signals of up to 15% (fig. S6). Thus, dynamic changes as small as 20 to 30% in tagged protein intensity typically can be accurately detected by using the present assay in individual cells. Temporal profiles of protein concentration. We begin with an account of the average pop- ulation level of the fluorescent intensity of each protein and then describe the individual cell behavior. We found that most (76%) proteins showed a significant decrease in fluorescence intensity in response to the drug, on diverse time scales. A subset of proteins (7%) showed a sig- nificant increase in intensity. The median dynam- ic range of this response was a 1.3-fold change in fluorescence, and the largest changes were about fivefold. Proteins showed several classes of dy- namical profiles (Fig. 2, A and B, and fig. S7) (15). The present data include dynamics of about 150 uncharacterized proteins (table S4) found throughout all profiles (Fig. 2B). Groups of functionally related proteins tended to show similar dynamics and protein localization profiles. For example, ribosomal proteins showed rapid, highly correlated degradation (Fig. 2C), which was confirmed by immunoblots (fig. S2). Proteins with slower apparent degradation include cytoskeleton components and metabolic enzymes. The timing of degradation of most cytoskeleton- associated proteins correlated with the timing of the loss of cell motility, as measured by tracking cell position over time (Fig. 2D). Proteins that rose late in the response include helicases implicated in DNA damage repair apoptosis-related proteins, such as the Bcl2-associated proteins BAG2 and BAG3 and the programmed cell death protein PDCD5. The drug target is among the first to respond. The total cellular fluorescence levels of the tagged drug target TOP1 decreased on a time scale of 1 hour, preceding almost all other responses in the present study (TOP1 is among the first 1% of responding proteins) (Fig. 2, arrow). The rate of TOP1 fluorescence decrease was CPT dose��� dependent (fig. S8, A and C). Immunoblots con- firmed the rapid degradation of both eYFP-tagged Fig. 1. Library of tagged proteins and its image analysis. (A) The cell clone library was generated in two steps: First, a red fluorescent tag (mCherry) flanked by splice signals was intro- duced on a retrovirus into the genome of H1299 cells, which resulted in cells that express proteins with an internal mCherry exon. After two rounds of tagging, a cell clone was selected with a red labeling pattern that is suitable for image analysis, bright in the nucleus and weaker in the cytoplasm. This clone formed the basis for an additional round of tagging, with a yellow fluorescent tag (eYFP or Venus) as an internal exon. Individual yellow- tagged cells were sorted then ex- panded into clones, and the tagged protein in each clone was identified. (B) Image analysis used the red flu- orescent images to automatically de- tect cell and nuclear boundaries and to quantify the YFP intensity at each time point. (C) Cells were grown in an incubated microscope for 24 hours under normal conditions and then for an additional 48 hours in the presence of 10 mM CPT. Cells were imaged every 20 min, and fluorescent intensity in each cell was automatically tracked. Cell divisions and morphological changes associated with cell death were automatically detected (15). Shown is a schematic of two daughter cells of the cell in the top panel. The cell labeled with the blue track shows blebbing and fragmentation typical of apoptosis. Scale bar, 45 mm. MLV, murine leukemia virus. LTR, long terminal repeat. 5 DECEMBER 2008 VOL 322 SCIENCE www.sciencemag.org 1512 RESEARCH ARTICLES