Fibroblasts from metastatic sites induce broad-spectrum drug desensitization via modulation of mitochondrial priming

Due to tumor heterogeneity, most believe that effective treatments should be tailored to the features of an individual tumor or tumor subclass. It is still unclear what information should be considered for optimal disease stratification, and most prior work focuses on tumor genomics. Here, we focus on the tumor micro-environment. Using a large-scale co-culture assay optimized to measure drug-induced cell death, we identify tumor-stroma interactions that modulate drug sensitivity. Our data show that the chemo-insensitivity typically associated with aggressive subtypes of breast cancer is not cell intrinsic, but rather a product of tumor-fibroblast interactions. Additionally, we find that fibroblast cells influence tumor drug response in two distinct and divergent manners, which were predicable based on the anatomical origin from which the fibroblasts were harvested. These divergent phenotypes result from modulation of “mitochondrial priming” of tumor cells, caused by secretion of inflammatory cytokines, such as IL6 and IL8, from stromal cells.


INTRODUCTION
associated with poor outcomes (Kalluri and Zeisberg, 2006). Moreover, fibroblasts are well known 118 to supply growth factors and matrix proteins, which may alter how cells respond to DNA damage 119 (Lee et al., 2012).

120
To identify interactions between tumor cells and stromal fibroblasts that alter drug response, 121 we used an in vitro co-culture system that was successfully used to study tumor-stroma interactions 122 in other contexts (Straussman et al., 2012). In this experimental platform cancer cells are 123 genetically modified to express GFP, which facilitates rapid, quantitative, high-throughput, and 124 cancer cell specific measurement of drug response dynamics. We piloted this study using  125 TNBC cells, either grown in mono-culture or in co-culture with HADF, a primary non-immortalized 126 human fibroblast cell harvested from the adrenal gland. We determined α−smooth muscle actin 127 (SMA) expression, a marker of the "activated fibroblast" expression state, which is commonly 128 observed in fibroblasts associated with tumors (also called Cancer Associated Fibroblasts, or CAFs, 129 and myofibroblasts). Immunofluorescence microscopy experiments confirmed that SMA expression 8 in HADF cells is generally low and variable when these cells are grown in mono-culture. SMA 131 expression is increased in HADF cell co-cultured with BT-20 TNBC cells, and a similar pattern was 132 observed for other primary fibroblasts (Supplemental Figure 2). For our pilot drug screen, these 133 cells were exposed to one of two drugs: erlotinib, a small molecule EGFR inhibitor, or camptothecin, 134 a potent Topo I inhibitor (Supplemental Figure 3A). Erlotinib does not kill BT-20 cells but does 135 induce a transient growth arrest (i.e. cytostasis), whereas camptothecin potently kills BT-20 cells 136 (i.e. cytotoxicity) (Lee et al., 2012). Total well fluorescence measured using a fluorescence plate 137 reader revealed that co-culture with HADF enhanced the proliferation rate of BT-20 cells to a small 138 extent. Furthermore, we found that co-culture with HADF potently blocked erlotinib-mediated 139 cytostasis of BT-20 cells, but had no effect on camptothecin sensitivity. Notably, well-based 140 fluorescence failed to capture the potent death that we observe by other methods following 141 camptothecin exposure, with all measurements in the time course recording higher values than the 142 initial pre-drug measurement (Supplemental Figure 3A and 1A). We were concerned that this 143 reflected a poor sensitivity of this assay, particularly with respect to quantifying the degree of 144 cytotoxicity rather than proliferation. Thus, 96 hours after drug exposure, we collected images of 145 these wells using fluorescence microscopy and quantified cell numbers from these images using a 146 CellProfiler-based automated analysis pipeline (Lamprecht et al., 2007). Our quantitative image 147 analysis confirmed the growth rate increase induced by HADF, as well as the loss of erlotinib-148 induced cytostasis in HADF co-culture (Supplemental Figure 3B). Importantly, however, image 149 analysis revealed a strong stromal interaction that was not observed by well-based fluorescence 150 measurements. Less than 1% of BT-20 cells survived chronic camptothecin exposure if grown in 151 mono-culture, but roughly 25% of these cells survived in the presence of HADF (Supplemental 152 Figure 3C). Taken together, these data indicate that well-based measurement of GFP fluorescence is appropriate for quantifying changes to proliferation, but not sufficient for quantifying the degree of 154 cell death in a population of cells. 9

Co-culture screen optimized to monitor cytotoxicity reveals widespread stromal influence on
Because we were primarily focused on the study of cytotoxic DNA damaging agents, we apoptotic cell death (Figure 2A and B) (Montero et al., 2015). At low concentrations, this dye exists 162 as a monomer and yields green fluorescence, however, when accumulated at high concentrations 163 within mitochondria, this dye forms aggregates, which yield orange/red fluorescence. Thus, the red 164 fluorescence of the JC-1 dye reports cellular mitochondrial integrity, which is lost when cells 165 activate apoptosis. To assess the suitability of JC-1 to quantify modulation in the degree of cell 166 death, we again piloted this assay on BT-20 cells treated with camptothecin in the presence or 167 absence of HADF. Images of these cells taken prior to drug exposure confirm punctate red 168 fluorescence in BT-20, but not HADF, confirming that the dye is not exchanged between cells in co-169 culture ( Figure 2B). 96 hours after exposure to camptothecin, the majority of BT-20 cells had 170 significantly reduced JC-1 red fluorescence, suggesting that mitochondrial integrity has been 171 compromised ( Figure 2C). Importantly, JC-1 red fluorescence measured using a fluorescence plate 172 reader was sufficiently sensitive for observing both the potent cell death of BT-20 cells in mono-173 culture, and the protective effect of HADF cells in co-culture ( Figure 2D).

174
To evaluate the role of stromal fibroblasts in DNA damage sensitivity, we selected six TNBC 175 cell lines (3 BL and 3 ML) that have relatively similar levels of sensitivity to DNA damage. These 176 JC-1 labeled TNBC cells were grown in mono-culture or in co-culture with each of a panel of 16 177 primary human fibroblasts. Each culture was exposed to increasing doses of 42 anticancer drugs 178 (at least one drug per class for all current FDA approved breast cancer drugs, Supplemental Tables found a strong overall correlation among biological replicates, indicating that the stromal influences observed were not due to measurement noise (Supplemental Figure 4). To identify TNBC-fibroblast at early times (i.e. 8 hours), low doses (0.1 µM), and responses to anti-estrogen drugs 187 (Supplemental Table 4). Non-response to anti-estrogen compounds is expected as TNBCs do not 188 express estrogen or progesterone receptors.

189
The majority of TNBC cell-fibroblast interactions did not alter drug sensitivity (Supplemental 190 Figure 4A-B). Nonetheless, our screen revealed many striking phenotypes, which strongly altered 191 drug sensitivity in both positive and negative directions. To determine the reliability of these 192 measurements, we selected both strong and moderate phenotypes to validate by flow cytometry.

193
For example, our screen identified that palbociclib killed more than 80% of HCC-1143 cells, a

194
Basal-Like TNBC, if applied to these cells in mono-culture. However, this drug was rendered 195 ineffective when HCC-1143 cells were co-cultured with the fibroblast cell, HCPF, resulting in only a 196 20 -40% decrease in cell viability (orange dots in Figure 2E). A flow cytometry based analysis of 197 cell death recapitulated this drug desensitization phenotype ( Figure 2F and G). Additionally, our co-198 culture screen identified instances in which the efficacy of etoposide is improved in co-culture 199 conditions. For example, etoposide was ineffective in killing Mesenchymal-Like Hs578T cells in 200 mono-culture, but killed more than 50% of these cells grown in co-culture with skin fibroblast cells, 201 WS1 (purple dots in Figure 2E). This phenotype was interesting because our prior studies have 202 found that etoposide, a Topo II inhibitor, is minimally active in mono-culture, which was surprising 203 given the clinical utility of this compound (Lee et al., 2012). Flow cytometry based analysis of cell 204 death confirmed that etoposide induced cell death in Hs578T is significantly enhanced by co-culture Prior studies that have interrogated fibroblast-tumor cell-drug interactions have found that 210 these interactions generally result in drug resistance, with rare instances in which stromal cell 211 interactions lead to drug sensitization (Mcmillin et al., 2010;Straussman et al., 2012). Mechanisms 212 that contribute to this directional variability have not been identified, likely because few drug 213 sensitizing phenotypes had been previously found. In contrast, our screen reveals that fibroblasts 214 sensitize and de-sensitize TNBC drug response with similar frequencies (Supplemental Figure 4B). the observed directional variation. We sought to determine if certain TNBC cells, fibroblast cells, or 217 drugs were intrinsically more likely to be involved in sensitizing or de-sensitizing interactions 218 (Supplemental Figures 5-7). Indeed, we found that some TNBC cells (e.g. MDA-MB-468), drugs 219 (e.g. sunitinib), or fibroblasts (e.g. WS1) appear to be involved in directionally biased interactions.

220
However, we wanted to integrate these insights to determine the relative importance of each of 221 these features in our dataset.

222
We performed principal component analysis (PCA) on our screening data. PCA uses the 223 correlation structure of the data to reduce data dimensionality to a smaller number of "principal 224 components" which maximally capture the variance of the dataset (Janes and Yaffe, 2006). PCA 225 can be used to generate a simplified description of the observed data, and here, we were 226 particularly interested in using PCA to quantify the relative contribution of each measured influence 227 (e.g. specific tumor cells, fibroblasts, drugs, or unique combinations of each) to the overall observed 228 pattern of data.

229
PCA identified 10 principal components, with the first two components capturing 53% of the 230 overall variation in the data. The projection of our data onto PC1 and PC2 revealed a clear 231 separation of BL and ML cells, revealing that TNBC subtype dependent responses account for a 232 significant portion of the observed pattern ( Figure 3A and C). Notably, this expected pattern was not 12 visible in drug response data collected on these TNBC cells grown in mono-culture ( Figure 1B and between BL and ML subtypes of TNBC are not a cell intrinsic property, but rather a product of 236 interactions between TNBC cells and stromal components such as fibroblasts.

237
Another important insight derived from PCA of TNBCs grown in co-culture was that the 238 variation associated with BL vs. ML subtypes was captured exclusively on PC2, rather than PC1 239 ( Figure 3C). By definition, PC1 captures information that is unrelated to PC2, and in this dataset,  fibroblasts alter the magnitude of TNBC drug response (correlated X-Y variance), or alternatively, 258 the degree to which TNBC drug response was altered by co-culture (uncorrelated X-Y variance). To 13 inspect this further, we arrayed all data clustering each unique cancer-fibroblast-drug combination by dose and time, in order to subsequently grouped by stromal location and drug, and a map was created for each TNBC cell line 263 to facilitate visual inspection of the relative influences induced by each fibroblast line. To test 264 whether fibroblast origin was associated with differences in the magnitude of drug response, we 265 generated maps using the percent viability in co-culture (i.e. y-axis data from Figure 2E). From 266 these maps, differences between fibroblast lines were not apparent, suggesting that fibroblast origin 267 does not alter drug sensitivity magnitude (Supplemental Table 3). Next, to test if fibroblast origin 268 was associated with the degree to which drug responses were altered in co-culture, we generated 269 maps using the co-culture:mono-culture response ratio ( Figure 4B and Supplemental Figure 9).

270
From this analysis, a clear difference between fibroblast lines was visible, with fibroblasts derived 271 from common sites of metastasis generally enhancing survival of TNBCs, whereas those derived 272 from other organs generally were neutral or enhanced TNBC cell death. Interestingly, these 273 patterns were observed across nearly all drugs, revealing that the location specific trends were 274 more robust than drug specific responses. To determine if these visual trends were statistically 275 robust, we calculated the mean response ratio for each unique cancer-fibroblast-drug combination, 276 and separated these data by metastatic location ( Figure 4C). These data confirmed a statistically 277 significant difference in the directionality of influence between fibroblasts derived from organs that 278 are common or uncommon metastatic sites. Additionally, we also performed statistical analyses 279 including only the 5039 drug responses, which were the largest and most significantly changed 280 phenotypes (Supplemental Figure 4D). This analysis of extreme "outliers" revealed that, while the 281 total number of "hits" were similar between metastatic and non-metastatic sites, the directionality of 282 these hits was significantly different between these groups, and consistent with the insights generated using average response ratio ( Figure 4D). Taken together, these results are consistent 284 with those from PCA, and further reveal that fibroblasts induced fundamentally opposing influences on the drug response of TNBC cells, largely dependent on the anatomical origin from which the 286 cells were harvested.

288
Fibroblasts alter TNBC drug response through modulation of the mitochondrial priming 289 state of TNBC cells 290 Next, we aimed to determine the mechanism by which fibroblasts interact with TNBC cells to 291 produce divergent and largely drug-independent modulation of drug response. The simplest 292 mechanism that is consistent with our observations would be that these fibroblasts cause a direct 293 TNBC cell growth or survival defect, independent of the drugs added ( Figure 5A, example i). To test 294 this, we used GFP-tagged TNBC cells to monitor TNBC specific growth/survival phenotypes. We 295 found that most fibroblasts cells either did not alter the growth rate of TNBC cells or induced a 296 modest growth rate increase of TNBC cells grown in co-culture ( Figure 5B). Furthermore, for 297 fibroblasts that consistently sensitized drug response rates in all TNBC cell lines (WS1, C12385, or 298 HUF), co-culture did not significantly alter growth or survival, suggesting that a fitness or survival 299 defect does not account for the broad-spectrum drug sensitization seen in co-culture with these 300 cells. In rare instances co-culture conditions did result in a significant TNBC cell growth rate 301 decrease, such as seen with MDA-MB-231 cells grown with H6013, a fibroblast derived from lung 302 tissue ( Figure 5B). Notably, the M231-H6013 interaction induced broad-spectrum drug 303 desensitization (i.e. enhanced survival, see Supplemental Figure 9). Thus, even in the rare 304 instances in which fibroblast cells mediated fitness defects, growth rate or survival modulation does 305 not appear to account for the observed pattern of influences on TNBC drug response.

306
A second mechanism by which fibroblast cells could enhance drug efficacy could be by 307 metabolizing the drugs, creating a more potent or more bioavailable compound ( Figure 5A, example 308 ii). This mechanism was recently reported to explain a microbiome-drug interaction that modulates 309 toxicity of the chemotherapeutic 5-FU (García-González et al., 2017). To test the role of fibroblast-310 mediated drug metabolism, we focused on drugs that activate death through induction of DNA damage. For this set of compounds, drug potency should be proportional to level of γ−H2AX, which 312 marks sites of DNA double stranded breaks. We quantified γ−H2AX nuclear intensity in the 313 presence and absence of fibroblast co-culture. These measurements were made at 4 time points 314 following exposure to teniposide, a Topo II inhibitor similar to doxorubicin, which is used clinically in 315 the treatment of TNBC (doxorubicin fluorescence limits use of this compound). We used GFP-316 labeled TNBC cells to identify TNBC cell nuclei and images were quantified using automated image 317 analysis (CellProfiler, (Lamprecht et al., 2007)). Overall, we found many cases where fibroblasts 318 modulated γ−H2AX levels ( Figure 5D). Importantly, however, the degree to which γ−H2AX was 319 modulated by fibroblast cells was poorly correlated with the phenotypic influence of these 320 fibroblasts ( Figure 5D). Furthermore, we inspected the most strongly sensitizing and de-sensitizing 321 co-culture environments to determine if these extreme cases could be explained by differences in 322 the apparent drug potency. TNBC nuclear γ−H2AX intensity was similar in BT-20 cells co-cultured 323 with C12385 and Hs27A, fibroblasts that strongly sensitized and desensitized drug responses, 324 respectively. Thus, it does not appear that fibroblast influences on drug sensitivity generally occur 325 through modulation of the drugs themselves.

326
The insights gained from γ−H2AX intensity are also consistent with our general observation 327 that fibroblast cells influence drug sensitivity in similar ways across diverse classes of drugs. In 328 other words, it does not appear that the mechanisms by which fibroblast cells influence the drug 329 responses in TNBC cells are specific to the drug compounds themselves or drug-specific responses 330 of TNBC cells. Drug-induced cell death is the product of at least two independent influences: the 331 drug-specific cell response (i.e. the ability of a drug to change a cell from a healthy to a dead state) 332 and the degree to which the cell is "primed" for death (i.e. how "close" the healthy cell is to dying) 333 (Chonghaile et al., 2011;Montero et al., 2015). Thus, a third mechanism that we tested was 334 whether fibroblasts alter the degree of mitochondrial apoptotic priming. We used the BH3 profiling mitochondria. We selected five fibroblast cells that produced the strongest and most consistent 338 modulation of drug sensitivity. The mitochondrial response to BIM peptide was quantified by 339 monitoring cytochrome c retention by flow cytometry ( Figure 5F). BH3 profiling revealed that 340 fibroblast co-culture conditions significantly altered the mitochondrial priming state of BT-20 and 341 MDA-MB-231 cells. Furthermore, the degree to which mitochondrial priming was increased or 342 decreased was also highly correlated with relative drug sensitivity observed in our co-culture screen 343 ( Figure 5G).

345
Fibroblast-secreted IL8 induces hyper-sensitivity to DNA damaging chemotherapy 346 WS1 (skin), C12385 (uterine), and HUF (uterine) cells, sensitized all six TNBC cell lines to 347 nearly every drug tested. We prioritized understanding mechanisms by which these cells induce 348 broad-spectrum drug sensitization, as these may be therapeutically relevant insights. To begin, we 349 determined whether conditioned media from these fibroblast cells also induced drug sensitization.

350
Conditioned media was collected following 48 hours of culture with fibroblasts and added to TNBC 351 cells prior to addition of teniposide. To measure the rate of TNBC cell death, we used Sytox green, 352 a cell impermeant dye that is fluorescent only when bound to DNA. We found that conditioned 353 media from WS1, HUF, and C12385 sensitized all six TNBC cell lines tested to teniposide ( Figure   354 6A). In contrast, conditioned media from Hs27A, a bone fibroblast cell, generally did not alter TNBC 355 drug response, although these data were more variable across TNBC cells.

356
To identify secreted factors that are responsible for these phenotypes, we profiled 357 conditioned media for the presence of 45 common cytokine, chemokine, and growth factors ( Figure   358 6B). We reasoned that the relative secretion profile for factors that induce drug sensitization should 359 be inversely correlated with the relative drug sensitivity observed in our co-culture screen. Of the 45 360 cytokines profiled, we observed strong negative correlation only for IL8 ( Figure 6C and D). sensitivity and secretion of IL6, a cytokine that is already known to induce resistance to DNA whereas recombinant IL6 decreased the rate of cell death by approximately 2-fold ( Figure 6E). To 367 further determine the role of IL8 secretion in the drug sensitization phenotype observed in 368 conditioned media, we tested IL8 neutralizing antibodies with conditioned media from WS1, HUF, or 369 C12385 fibroblasts. In all three cases, IL8 neutralizing antibodies significantly inhibited the drug 370 sensitization induced by conditioned media ( Figure 6F). Notably, IL8 neutralizing antibodies failed to 371 restore drug sensitivity to the levels observed in the absence of fibroblast conditioned media,

372
suggesting that other secreted factors also contribute to the drug sensitization phenotype.

375
In this study, we explored interactions between tumor cells and stromal cells to identify 376 those that modulate sensitivity to commonly used chemotherapeutics. We found that fibroblasts can 377 alter drug sensitivity of tumor cells, and that the responses were highly variable, both in magnitude 378 and in direction. Our statistical analysis clarified that the directional variability in fibroblast influence 379 is predominantly associated with the anatomical organ from which the fibroblast cells were 380 harvested. Specifically, fibroblasts from common sites of metastasis typically desensitize tumor cell 381 drug responses. This interaction was fundamentally different than what was observed with 382 fibroblasts from organs that do not typically accommodate metastatic growth, which typically 383 sensitized tumor cell drug responses. Somewhat surprisingly, these influences were consistently 384 observed, regardless of which drug was applied, which was caused by fibroblast-dependent 385 modulation of mitochondrial priming within cancer cells. Our analysis of a small set of common The contribution of cancer associated fibroblasts to a variety of tumor phenotypes has been compared to prior studies, our screen revealed a greater proportion of drug sensitizing interactions 392 between stromal and cancer cells. This difference could have resulted from the depth of our screen, 393 as strong drug sensitizing phenotypes are also rare in our data. Alternatively, it is also possible 394 fibroblast mediated drug sensitization is a more common phenotype in TNBC cells, as this cancer 395 subtype was not deeply profiled in prior studies. Another possibility is that our screening 396 methodology, which was designed to exclusively monitor drug induced cell death, contributed to the 397 enhanced resolution of cell death sensitization. In fact, this feature may be likely to play a part given

403
Nonetheless, the findings from our study have important implications that should be 404 considered in the context of "personalized" or "precision" medicine. These concepts are explored 405 typically using genomic analyses of tumor cells. Our data suggest that interactions between tumor 406 and stromal cells often alter drug sensitivity, and moreover that these interactions are a potentially 407 more potent source of variation in drug sensitivity than tumor cell gene expression state. Thus, our 408 data suggest that personalized treatment regimens will ultimately need to consider micro-    and cells were exposed for 1, 6, and 18 hours before being fixed with 4% formaldehyde for 10 504 minutes at room temperature. Cells were washed twice in PBS, then permeabilized with 0.5% Triton 505 X100 for 10 minutes at room temperature. Cells were washed twice with PBS; blocked in 10% goat 506 serum (Thermofisher cat# 16210064) for one hour; stained with the p-Histone H2A.X (Ser139) 507 antibody (Cell Signaling Technologies #9718S) in 1% goat serum in PBS overnight at 4C; stained 508 with Alexa-647 antibody (1:250 dilution, Thermofisher A21244) in 1% goat serum in PBS for 2 hours 509 at room temperature. Imaging was performed using an IXM-XL high throughput automated 510 microscope. Analysis was performed using a custom CellProfiler pipeline (available upon request).

512
For α-SMA staining, fibroblast cells were stained with 5 µM CellTrace Far Red dye (Thermofisher 513 cat# C34564) and cancer cells were stained with 5uM CellTrace Violet dye, each as described 514 above. Each cell type was plated at a 1:1 ratio. Cell fixation, permeabilization, and staining were 515 performed as above for H2AX. Cells were stained with the α-SMA antibody (Cell Signaling       are shown relative to the Log2 survival ratio (data are from co-culture drug screen as in Figure 2E). (E and F) Cell death measured using Sytox green following exposure to 5 µM camptothecin (Cam), in the presence or absence of recombinant IL6 or IL8 (E), or conditioned media (CM) and/or IL8 neutralizing antibody (IL8AB). IL6 and IL8 were used at 50 ng/µL, and IL8AB was used at 100 µg/ mL. Eotaxin  FGF2  IFNg  GMCSF  GROa  HGF  IFNa  IL1a  IL1b  IL10  IL12  IL13  IL15  IL17a  IL18  IL1ra  IL2  IL21  IL22  IL23  IL27  IL31  IL4  IL5  IL6  IL7  IL8  IL9  IP10  LIF Figure 1A. (B) Cell viability measured as in (A) for 10 Topo I/II inhibitors. Data are z scored max death at 72 hours (Emax). Dendograms from hierarchical clustering shown for drugs and for cells (Basal A cells highlighted with blue bar; Claudin-low cells highlighted with red bar).  Supplemental Figure 2: Primary fibroblasts stain positive for markers of activation in coculture with TNBC cells. Primary fibroblasts grown in mono-culture or in co-culture with BT20 cells were stained for expression of α-smooth muscle actin (SMA), a marker for activated fibroblasts (a.k.a. "Cancer Associated Fibroblasts", CAFs, or myofibroblasts). SMA in green. FarRed dye (CellTrace) used to stain fibroblasts and Blue dye (CellTrace) used to stain BT20. BT20 cells grown in mono-culture did not stain positive for SMA expression (data not shown). Data are from biological quadruplicate experiments. Mono-culture is BT20 cells labeled with GFP; co-culture is GFP-BT20 + HADF, seeded at a 1:1 ratio. Data are GFP fluorescence per well in untreated cells, or following exposure to 10 µM erlotinib or 500 nM camptothecin. (B) Representative images from experiment in panel (A). Images collected using fluorescence microscopy following the 96 hour measurement on a plate reader. GFP-BT20 are green; HADF are stained blue using a whole cell stain. (C) Cell viability based on quantitative image analysis. Wells from experiment in panel (A) were imaged at 96 hours and cell numbers were quantified using CellProfiler. % Survival is relative to untreated cells grown in mono-culture. At least 300 cells were counted in every image, except BT20 mono-cultures treated with camptothecin, where the average number of cells per image was 20 (8 images collected). Well-based fluorescence measurements taken using a plate reader accurately capture proliferation phenotypes but fail to capture differences in cell death.  Figure 2E, with drug responses of TNBC cells grown in mono-culture on the x-axis, and responses in co-culture with CAFs on the y-axis. In each plot, the overall dataset is shown in blue circles and the data for each TNBC cell line is shown in orange. Data plotted as in Figure 2E. Drugs organized by class, with cytotoxic/DNA damaging agents on top (first 4 rows) and targeted therapies below (bottom 3 rows). Data for each drug are highlighted in orange.