Research ArticleCancer therapy

High-throughput dynamic BH3 profiling may quickly and accurately predict effective therapies in solid tumors

See allHide authors and affiliations

Science Signaling  16 Jun 2020:
Vol. 13, Issue 636, eaay1451
DOI: 10.1126/scisignal.aay1451

Speedy screen for tumor therapies

Although cell death screens using patient biopsies could be used to identify effective, personalized treatments, it takes several days to obtain results, meaning that the cells ultimately guiding treatment decisions may become molecularly different than those in the patient. Bhola et al. developed a high-throughput method (called HT-DBP) that identifies, within 24 hours, drugs that initiate cell death programs in tumor cells from freshly isolated patient biopsies. HT-DBP identified single agents and combinations that shrunk breast and colon tumors in mice but that would not have seemed as promising when screened after the usual multiday period of culture outgrowth. The findings may enable a fast and broadly applicable tumor screening technique to guide effective treatment decisions for patients.

Abstract

Despite decades of effort, the sensitivity of patient tumors to individual drugs is often not predictable on the basis of molecular markers alone. Therefore, unbiased, high-throughput approaches to match patient tumors to effective drugs, without requiring a priori molecular hypotheses, are critically needed. Here, we improved upon a method that we previously reported and developed called high-throughput dynamic BH3 profiling (HT-DBP). HT-DBP is a microscopy-based, single-cell resolution assay that enables chemical screens of hundreds to thousands of candidate drugs on freshly isolated tumor cells. The method identifies chemical inducers of mitochondrial apoptotic signaling, a mechanism of cell death. HT-DBP requires only 24 hours of ex vivo culture, which enables a more immediate study of fresh primary tumor cells and minimizes adaptive changes that occur with prolonged ex vivo culture. Effective compounds identified by HT-DBP induced tumor regression in genetically engineered and patient-derived xenograft (PDX) models of breast cancer. We additionally found that chemical vulnerabilities changed as cancer cells expanded ex vivo. Furthermore, using PDX models of colon cancer and resected tumors from colon cancer patients, our data demonstrated that HT-DBP could be used to generate personalized pharmacotypes. Thus, HT-DBP appears to be an ex vivo functional method with sufficient scale to simultaneously function as a companion diagnostic, therapeutic personalization, and discovery tool.

INTRODUCTION

Clinically actionable knowledge can be obtained from the direct study of primary patient tumors. Current research practice on primary samples typically focuses on obtaining large amounts of information about the molecular constituents of tumors (DNA, RNA, proteins, metabolites, and others). These analyses have yielded important successes that have directly improved cancer treatment (13). Nonetheless, they necessarily lose information about dynamic interactions of these constituents because they are static measurements of initial conditions (4). Unfortunately, static approaches do not identify active drugs for enough cancer patients, and alternative diagnostic approaches are needed.

To determine how a patient tumor will respond to a drug, it is hard to imagine a more practical way to do this than to put the living cancer cell in contact with that drug. However, inadequate accuracy of prior attempts at ex vivo chemosensitivity determination limited enthusiasm about this approach (5). These older strategies often tested very small numbers of drugs, often in long-term cultures deleterious to maintenance of cancer phenotype and used rudimentary readouts incapable of single-cell resolution. The tools were simply not good enough to drive clinical decision-making, and enthusiasm waned (59). We now have many more potential therapeutics to test, as well as vastly improved cell biological knowledge and better cellular analytics. The potential quality of the information to be garnered from exposing patient cancer cells to drugs has led to a recent reexamination of its application.

A compelling approach for identifying novel or personalized therapies is to perturb living cancer cells with drugs ex vivo and measure clinically relevant cellular phenotypes (4, 1018). The ideal therapeutic perturbation strategy minimizes ex vivo culture time, maximizes the testable number of therapeutics, and measures effects that are directly relevant to in vivo response. Here, we report a novel strategy called high-throughput dynamic BH3 profiling (HT-DBP) that fulfills these criteria. We previously developed BH3 profiling to measure the proximity of a cell to the apoptotic threshold, a property called “apoptotic priming” (fig. S1A) (19, 20). Apoptotic priming is inferred from the degree of mitochondrial outer membrane permeabilization induced by standardized concentrations of synthetic peptides derived from the BH3 domains of proapoptotic B-cell lymphoma 2 (BCL-2) family proteins. We subsequently developed dynamic BH3 profiling (DBP), which measures how short, 4- to 24-hour ex vivo drug treatments increase apoptotic priming in patient tumor cells (fig. S1, B and C). When a cell exhibits a statistically significant increase in apoptotic priming after drug treatment, that cell is likely to respond to that drug both in vitro and in vivo (11, 21, 22).

A key limitation of DBP is that only a few compounds could be tested on a primary sample at a time (11). This is a small fraction of clinically applicable cancer therapeutics and an even smaller fraction of preclinical or tool compounds (23). With a limitation of 10 to 20 compounds per sample, we could only perform hypothesis-driven drug testing and could not perform unbiased screening of chemical libraries to discover new therapies or new indications for existing therapies. Therefore, we sought to expand the number of drugs that could be tested with DBP to facilitate unbiased chemical screening of freshly isolated tumor cells. Early technological iterations of DBP required manual transfer of cancer cells from drug treatment vessels into fluorescence-activated cell sorting (FACS) vessels where apoptotic priming is measured (fig. S1D) (11). We therefore miniaturized and automated DBP to facilitate identification of the maximal number of apoptotic sensitizing drugs for any given tumor biopsy specimen.

RESULTS

Our previously reported low-throughput, hypothesis-based use of DBP relied on flow cytometry or well-based fluorimetry (11, 22). To perform unbiased, high-throughput measurements of early induction of apoptotic cell death at single-cell resolution, we implemented automated immunofluorescence microscopy and miniaturized the DBP assay (outlined in Fig. 1A). Single-cell suspensions of dissociated tumors were plated in 384-well plates and treated with chemical libraries using pin transfer tools. Twenty-four hours after drug treatment, cells were washed in phosphate-buffered saline (PBS), and the BH3 profiling buffers and synthetic BH3 peptides were added. To tune the assay for maximal sensitivity to drug-induced apoptotic signaling, 4 hours before completion of drug exposures, a single peptide concentration was chosen for each sample such that in the absence of drug, 10% of cells lose cytochrome c (fig. S2). Tumor cells were discriminated from stroma using tumor-discriminating antigens, such as epithelial cell adhesion molecule (EpCam) for epithelial tumors. Drugs that induce apoptotic signaling cause heightened BH3 peptide-induced cytochrome c loss compared to dimethyl sulfoxide (DMSO)–treated control wells, measured on a per cell basis using immunofluorescence microscopy. Drug-potentiated peptide-induced loss of cytochrome c was quantified as the difference in percentage of cytochrome c–positive cells between DMSO-treated and drug-treated wells (Fig. 1A), a parameter called “delta priming” (11). In all cases, a drug concentration of 1 μM was used in initial chemical screens, which was based on experiments showing good correlation of priming activity at drug concentrations below 1.1 μM (fig. S3).

Fig. 1 HT-DBP screen of 1650 compounds identifies chemicals that sensitize freshly isolated tumor cells, and not healthy cells, for apoptosis.

(A) Schematic showing the workflow of the HT-DBP screening platform. Compounds with the largest delta priming cause the largest increase in apoptotic sensitivity. (B) Cytochrome c staining in response to Bim peptide dose in MMTV-PyMT tumors. Scale bars, 100 μm. Images are representative of two independent experiments. (C) Quantification of the dose-response curve from imaging described in (B). The arrow indicates the peptide concentration chosen for use in the screening. Data are means ± SD of six replicates, representative of two independent experiments. (D) HT-DBP on MMTV-PyMT tumors to identify compounds that increase apoptotic sensitivity. DMSO-treated wells are shown in blue; compound-treated wells are shown in black. Data are means of two independent experiments. (E) Images of selected wells from the chemical screen and the DMSO. Wells were treated with the indicated for 24 hours and subsequently treated with 0.39 μM synthetic peptide. Cytochrome c immunofluorescence is shown in green, and Hoechst 33342 staining is shown in blue. Images are representative of two independent screens. (F) Comparison of a screen on freshly isolated adult mouse hepatocytes and freshly isolated MMTV-PyMT tumor cells. Data represent means of two independent experiments. (G) Chemical annotation of drug targets from the HT-DBP screen on MMTV-PyMT tumors. Each dot represents a single compound. Asterisks indicate instances where compounds with a similar target increase apoptotic priming (P < 0.0001, one-way ANOVA). Data represent means of two independent experiments. MOMP, mitochondrial outer membrane permeabilization; AChR, acetylcholine receptor; GluR, glutamate receptor; IKK, IκB kinase; 5-HT, 5-hydroxytryptamine (serotonin); PDK-1, phosphoinositide-dependent kinase 1.

We first verified that HT-DBP could identify compounds that induced apoptotic signaling in the MDA-MB-231 breast cancer cell line. Specifically, we identified the peptide concentration where about 10% of cytochrome c was lost from cells (fig. S4, A and B) and measured the relative increase in apoptotic signaling (fig. S4C). The change in priming observed in HT-DBP correlated with cytotoxicity 48 hours later (fig. S4, D and E, and data files S1 and S2). We next compared an earlier low-throughput FACS-based method of DBP with our new high-throughput method of DBP (fig. S5, A and B) and found that these produced similar rankings of chemical sensitivity (fig. S5C).

A rigorous test of our approach is to compare DBP results to treatment response in an in vivo model. Murine models offer the possibility to treat the same tumor in vivo many different ways. Therefore, we next asked whether we could assign active therapies to tumors from the mouse mammary tumor virus-polyoma middle tumor-antigen (MMTV-PyMT) mouse autochthonous breast cancer model (24). Single-cell suspensions made from freshly isolated tumors were distributed to 384-well plates and treated with a library of 1650 bioactive drugs (described in table S1) at a concentration of 1 μM for 24 hours (further detailed in Materials and Methods) and then subjected to HT-DBP (Fig. 1, B to D, and data file S3). We evaluated drug-induced apoptotic priming only in cells that stained for mouse EpCam (fig. S6). Cells were identified using the multiwavelength cell-scoring module in MetaMorph (fig. S7). The screen produced adequate high-throughput screening metrics (fig. S8, A and B) and high correlation of replicates in these screens (fig. S8, C and D, and data files S4 and S5). Several chemicals increased apoptotic priming above levels found in DMSO-treated wells (Fig. 1, D and E). These results indicate that HT-DBP can identify apoptosis-sensitizing compounds on freshly isolated tumor cells.

We next prioritized compounds with therapeutic potential by excluding broadly toxic compounds. We performed HT-DBP counterscreens on nontumor cells: adult mouse hepatocytes (fig. S9, A to C, and data file S6) (25) and human foreskin fibroblasts (19) (fig. S9, D and E, and data file S7). We next compared compounds that induced apoptotic priming in freshly isolated MMTV-PyMT tumor cells (Fig. 1F and data file S8). Compounds that sensitized nontumor cells for apoptosis at a concentration of 1 μM were deprioritized for further analysis. Using the multiplicity of unique compounds in the bioactive chemical library with the same nominal target, we sought to infer druggable pathway dependencies of tumor cells and not healthy cells. We determined that several compounds known to target heat shock protein 90 (HSP90), mammalian target of rapamycin complex 1/2 (mTORC1/2), Src, mitogen-activated protein kinase kinase (MEK), or Akt preferentially provoked apoptotic signaling in tumor cells (Fig. 1G and data file S9). Note that the three compounds showing the highest delta priming in the phosphatidylinositol 3-kinase (PI3K) column all target mTORC1/2, too. These results indicate that HT-DBP can identify compounds and pathways that selectively prime tumors, and not healthy cells, for apoptosis.

To validate DBP as a predictive tool, we next asked whether drugs identified by HT-DBP cause tumor regression in vivo. We selected six drugs for in vivo testing; three were predicted to be active, and three were predicted to be inactive. We prioritized drugs that were inexpensive, readily available, and had published dosing regimens. The dosing regimens of the drugs that we chose were previously demonstrated to have on-target in vivo pharmacodynamic effect and were tolerable to the mice (2631). Treatments based on assay hits with known pharmacokinetic properties [dasatinib (26), 17-dimethylaminoethylamino-17-demethoxygeldanamycin (17-DMAG) (27), AZD2014 (28), and a combination of dasatinib and 17-DMAG] showed decreased tumor volume compared to vehicle-treated mice at day 14 (Fig. 2, A and B). Drugs that did not score in our assay [navitoclax, lapatinib, and sunitinib (2931)] did not induce tumor regression (Fig. 2B). The magnitude of ex vivo apoptotic sensitization correlated with tumor response in vivo (Fig. 2C). Active drugs 17-DMAG and AZD2014 did not significantly alter cell numbers relative to DMSO-treated wells at 24 hours (Fig. 1D) and would not have been identified by standard screening methods that typically depend on completion of cell death to register a chemical hit.

Fig. 2 HT-DBP predicts in vivo response in breast cancer models.

(A) Change in tumor volume relative to the start of treatment over time for select drug treatment in MMTV-PyMT tumors. Data are means ± SEM of at least seven mice per group. (B) Fold change in tumor volume at day 14 relative to the start of treatment for MMTV-PyMT tumors. Each dot represents a single mouse; data are means ± SEM of at least seven mice per group. Dasatinib [10 mg/kg, intraperitoneally (ip)], 17-DMAG (10 mg/kg, ip), AZD2014 (15 mg/kg, orally), lapatinib (50 mg/kg, orally), or sunitinib (50 mg/kg, orally) was dosed daily, 5 days a week for 2 weeks. Navitoclax (100 mg/kg, orally) was dosed daily for 2 weeks. (C) Correlation between DBP and MMTV-PyMT tumor response in vivo (R2 = 0.83, P = 0.004; Pearson). Delta priming data (horizontal) are means ± SD of n = 3 independent experiments at drug concentration of 1 μM. Tumor volume data (vertical) are means ± SEM from 7 to 15 mice per group. (D) HT-DBP of select drug combinations in the DF-BM355 breast cancer PDX model. Each point represents an independent experiment. Lines represent means of n = 2 experiments. (E) Correlation between DBP and median survival of DF-BM355 mice treated with compounds. n = 5 to 9 mice per group (R2 = 0.82, P = 0.005; Pearson).

We next asked whether we could perform HT-DBP on small-volume samples to correlate ex vivo drug sensitivity with known in vivo responses. Using 3 × 105 viable cells from the breast cancer brain metastasis patient-derived xenograft (PDX) model DF-BM355 (32), we used HT-DBP to rank the predicted activity of drug combinations (Fig. 2D and fig. S10, A and B). We found good correlation between HT-DBP and previously reported median survival of drug-treated animals (Fig. 2E) (32). The correlation between ex vivo HT-DBP measurements and in vivo response for MMTV-PyMT and DF-BM355 breast cancer models indicate that HT-DBP can potentially assign effective in vivo therapies.

The ability to rapidly pharmacotype human tumors using HT-DBP could enable the use of patient-specific ex vivo drug sensitivity data to guide clinical decision-making and drug discovery at a scale heretofore not possible. To investigate the feasibility of this approach, we performed HT-DBP using 1650 compounds at a concentration of 1 μM on seven different human colorectal tumors isolated from PDX models (33). Freshly isolated tumor cells were identified on the basis of EpCam staining (fig. S11A) and were screened at a single synthetic Bim peptide concentration (fig. S11, B to D). To facilitate comparisons between chemical screens on different tumors, we normalized delta priming between the SD of DMSO-treated wells and the delta priming value of the compound that caused the maximum apoptotic sensitization (fig. S12, A and B). This resulted in a map of chemical sensitivities of the seven PDX models and two nontumor cells (fig. S13 and data file S10). We examined the top 35 hits that did not affect healthy cells from the hepatocyte or human foreskin fibroblast screens (Fig. 3A) and identified chemical vulnerabilities observed across all tumors (such as to navitoclax and abexinostat) and more private vulnerabilities observed in a fraction of tumors (such as to vorinostat and milciclib). On the basis of the screen, we treated the COCA9 PDX model with navitoclax or the pan-CDK (cyclin-dependent kinase) inhibitor AT7519 or both and found that although single agents did not have an effect, the combination treatment delayed tumor growth (Fig. 3B). Ultimately, a resulting matrix of drug sensitivities could form the foundation for development of effective patient-specific combination therapy strategies.

Fig. 3 Identification of apoptotic sensitizing compounds and drug targets in PDXs of colorectal cancer.

(A) Delta priming measurements of the top 35 hits from HT-DBP on seven PDX models of colorectal cancer and healthy cells. Red indicates compounds that cause the highest increase in apoptotic priming. Dark blue indicates compounds that are less than three times the SD of DMSO-treated wells. Data represent means of two replicates. HFF, human foreskin fibroblast. (B) Tumor volume after 21 days of in vivo treatment of COCA9 with navitoclax (100 mg/kg, orally, daily), AT7519 (15 mg/kg, ip, daily), or a combination of navitoclax and AT7519. The asterisk indicates a significant difference in tumor volume relative to vehicle-treated cells (Mann-Whitney, P = 0.03). Each point represents a single mouse; n = 4 to 5 mice per treatment arm. (C) Delta priming of the different PDX models based on nominal drug targets. Nominal targets include epidermal growth factor receptor (EGFR) (31 compounds), pan-CDK (15 compounds), pan-HDAC (20 compounds), MEK (12 compounds), and HSP90 (5 compounds). Data represent means of two replicates. (D) Comparison of delta priming in COCA74P and COCA74M. Data represent means of two replicates. (E) Delta priming of 17-DMAG for COCA74P and COCA74M. Each point represents an independent experiment. Lines represent means of n = 2 experiments. (F) Delta priming of abexinostat for COCA74P and COCA74M. Each point represents an independent experiment. Lines represent means of n = 2 experiments.

Next, using HT-DBP, we asked whether we could use chemical sensitivity results to infer pathway dependencies present in individual colon cancers. Although functional genetic tools are frequently used to identify vulnerabilities in cell lines, use of these tools in primary tissue is difficult. By using annotations of nominal small-molecule targets of a known bioactive library with sufficient protein target coverage, we asked whether we could identify pathway vulnerabilities of each tumor without extended ex vivo culture (Fig. 3C, fig. S14, and data file S11) (34). In deconvoluting protein targets of chemical compounds, we found that broad CDK and not pan-HDAC (histone deacetylase) inhibition was an effective strategy for sensitizing the COCA39 PDX model for apoptosis (fig. S14). Conversely, the COCA74M PDX model was sensitive to pan-HDAC inhibition but not pan-CDK inhibition (fig. S14). Thus, HT-DBP permits chemical biology approaches to identifying mechanism of action in primary tumor cells without extended ex vivo culture.

In comparing colon cancer PDX models derived from the primary and a metastatic site from the same patient (COCA74P and COCA74M), we identified both common and different chemical vulnerabilities (Fig. 3D). For example, several pan-HDAC inhibitors sensitize both primary and metastatic tumors for apoptosis; however, several HSP90 inhibitors sensitize only the primary tumor for apoptosis. Specifically, the HSP90 inhibitor 17-DMAG shows a preferential sensitization effect on COCA74P relative to COCA74M (Fig. 3E), whereas the pan-HDAC inhibitor abexinostat sensitizes both tumors equivalently for apoptosis (Fig. 3F). These data imply that intrapatient heterogeneity in drug response may exist between primary and metastatic sites.

Historically, established cancer cell lines have proven useful but imprecise tools for assigning in vivo therapy (35), driving our concern about ex vivo methods that require prolonged ex vivo culture. We hypothesized that adaptation to long-term ex vivo culture might alter chemical sensitivities compared to the fresh tumor. This hypothesis has never, to our knowledge, been directly tested. We realized that because tumors from the MMTV-PyMT model readily form cell lines, we have a unique opportunity with HT-DBP to directly test this hypothesis across a broad panel of compounds. We therefore performed HT-DBP on early-passage cancer cell lines, expanded in two-dimensional cell culture for over 1 month, derived from the same pool of primary tumor cells tested above in Fig. 1D (Fig. 4A; fig. S15, A and B; and data file S12). Evaluating compounds that did not prime healthy cells from the hepatocyte or human foreskin fibroblast screen at a concentration of 1 μM, several compounds sensitized both the tumor and the derived cell line for apoptosis, like HSP90 inhibitors. We also identified compounds that primed the cell line but not the tumor and the tumor but not the cell line (Fig. 4, A and B, and data file S13). Differences between freshly isolated tumor cells and cultured cancer cell lines were greater than assay noise measured in technical replicates or biological replicates (fig. S8, C and D). We used HT-DBP to evaluate drug dose responses of compounds that scored in the screen and for specific drugs measured both similar and differential activity in cell lines compared to cells from freshly isolated tumors (Fig. 4, C to E; fig. S15C; and data file S14). This includes AZD2014, an mTOR inhibitor that preferentially primed the fresh tumor cells for apoptosis relative to the cell line (Fig. 4D), and navitoclax, which preferentially primed the cell line relative to the freshly isolated tumor cells for apoptosis (Fig. 4E). Navitoclax does not noticeably alter tumor growth in vivo, but AZD2014 delays tumor growth in vivo (Fig. 2, B and C) and may have been missed or deprioritized by performing chemical screens on cancer cell lines.

Fig. 4 Evolution of apoptotic chemical vulnerabilities in cell culture conditions measured by HT-DBP.

(A) Comparison of freshly isolated MMTV-PyMT tumors with cells cultured ex vivo for 1 month. Compounds that inhibit HSP90 are colored blue. Compounds that inhibit mTOR are colored green. Gray dashed lines indicate 3 SDs of DMSO-treated wells. R2 = 0.38 by Pearson analysis. Data are means of two independent screens. (B) Identity of compounds that primed the freshly isolated tumor only, the cell line only, or both. This analysis only evaluated compounds that did not prime healthy cells for apoptosis. (C to E) Comparison of delta priming by (C) 17-DMAG, (D) AZD2014, and (E) navitoclax on freshly isolated MMTV-PyMT tumor cells cultured ex vivo for one month. Each point represents an independent experiment. Lines represent means of n = 2 experiments. (F and G) Comparison of drug-induced delta priming in (F) COCA74P and (G) COCA61 cells at days 1 and 8 after tumor extraction. Data are means of triplicates. AUC, area under the curve.

To further evaluate changes in drug-induced apoptotic sensitivity during ex vivo culture, we performed HT-DBP on freshly isolated tumor cells from the COCA74P and COCA61 colorectal cancer PDX models and on cells from each model that were grown ex vivo for 8 days. In COCA74P, we observed that there was enhanced sensitivity to ABT263 and A1331852 after 8 days in culture (Fig. 4F, fig. S16, and data file S15). Conversely, in COCA61, we observed minimal differences in sensitivity to ABT263 and A1331852 after 8 days in culture; however, we found that the MEK inhibitors trametinib and AZD8330 induced greater apoptotic priming in the day 8 cultures relative to the day 1 culture (Fig. 4G and fig. S16). Several chemical dependencies were preserved as cells remain cultured in vitro (Fig. 4, F and G). Ultimately, changes in chemical sensitivity with prolonged ex vivo culture may represent one reason for the mixed translational track record of cancer cell line chemical screens.

Last, we sought to determine apoptotic chemical sensitivities on human colon tumors directly obtained from patient tumors without intervening ex vivo culture or propagation in mice. HT-DBP was performed on colon tumors resected from the primary site and using a limited panel of drugs (Fig. 5, A and B, and data file S16). Similar to results in PDX models, we found that Bcl-XL inhibition using A1331852 increased apoptotic signaling in both tumors (Fig. 5, B and C). However, MEK inhibition using trametinib increased apoptotic signaling in one tumor but not the other (Fig. 5, B and D). Last, abexinostat (a pan-HDAC inhibitor) and AT7519 (a pan-CDK inhibitor), which increased apoptotic signaling in colon PDX models, also increased apoptotic signaling in primary human colon tumors (Fig. 5B). Thus, HT-DBP enables chemical screening on patient tumors with limited time ex vivo and can identify patient-to-patient differences in chemical sensitivities.

Fig. 5 Identification of apoptotic sensitizers in primary colon cancer using HT-DBP.

(A) Representative image of Hoechst 33342 (blue), EpCam immunofluorescence (red), and cytochrome c immunofluorescence (green) in a primary colon tumor after 24 hours of ex vivo DMSO treatment and after a BH3 profile. Scale bar, 75 μm. (B) HT-DBP on colon tumors (n = 3) using a limited panel of compounds at concentrations of 100, 250, and 500 nM. (C) Images of cytochrome c loss in colon tumors using DBP of the Bcl-XL inhibitor A1331852. Scale bars, 50 μm. (D) Images of cytochrome c loss in Colon-01 and cytochrome c retention in Colon-02 tumors using DBP of the MEK inhibitor trametinib. Scale bars, 50 μm. Images are representative of three replicates.

DISCUSSION

There has not previously been a method available to directly and immediately assess chemical vulnerabilities of a patient’s fresh primary tumor across a large panel of compounds. HT-DBP provides valuable chemical vulnerability information on the actual tumor without intervening model creation. Performing accurate preclinical drug screening or determining effective personalized therapy regimens using patient-derived models of cancer requires high pharmacologic fidelity of the models to the original tumor. Although high genomic fidelity is often observed in patient-derived models of cancer (36), an evaluation of pharmacologic fidelity is not typically performed.

Cancer cell lines have been a workhorse of cancer biology for many decades. However, identification of chemical vulnerabilities in cancer cell lines has a mixed record for effective translation to useful clinical treatments. Even drug screens on nominally identical cell lines cultured by different laboratories return divergent results (37, 38). Here, by developing a high-throughput method to evaluate drug sensitivity of tumors within 24 hours of excision, we provide data that the very process of prolonged ex vivo propagation rapidly alters chemical vulnerabilities compared to the primary tumor. Specifically, dissimilarities that we observed in drug sensitivities of the MMTV-PyMT mouse model of breast cancer at 24 hours ex vivo and after 30 days of ex vivo cell culture demonstrate that drug sensitivity evolves with prolonged ex vivo culture. AZD2014 showed minimal drug sensitivity with extended ex vivo culture; however, this drug delayed tumor growth in vivo, indicating the potential to miss promising therapies by screening on cancer cell lines. Applying HT-DBP to other tumor types will help identify drug sensitivity artifacts of extended ex vivo culture. Moreover, HT-DBP could screen for specific culture conditions that minimize ex vivo artifacts of chemical sensitivity.

Whereas the direct identification of clinically relevant apoptotic sensitizing chemicals represents a direct translational application of HT-DBP, identifying signal pathway vulnerabilities in primary tumors could motivate the development of targeted inhibitors (34). Here, using compound libraries with sufficient protein target coverage, we used chemical biology strategies and HT-DBP to evaluate signaling pathways that represent vulnerabilities in freshly isolated tumors from the MMTV-PyMT mouse model and PDX models of colorectal cancer. This was enabled by a 100-fold increase in the number of compounds screened as potential apoptotic sensitizers compared to the original DBP method. The identification of signal pathway vulnerabilities using genetic perturbagens with primary samples requires weeks of ex vivo culture. We anticipate that applying clinically relevant compounds and tool compounds in HT-DBP screens will enable the application of chemical biology strategies to identify signal pathway vulnerabilities in primary patient tumors without the need for prolonged ex vivo culture.

A potential use of HT-DBP will be the assembly of combination regimens for typically chemorefractory tumors. Evaluating combination therapies requires a factorial increase in the number of screening wells, which is not amenable to freshly isolated tumors, or primary tumors. One of the theoretical advantages of DBP over conventional measures of the cell death is the ability to identify compounds that sensitize cells for apoptosis but may not induce frank cell death, rendering them invisible by most other techniques. These compounds that sensitize cells for apoptosis, but do not kill cells, could be effective in combination.

Precision medicine in oncology has relied almost exclusively on genomic information. However, most patients do not benefit from such methods, and alternatives must be identified (4). Technologies like HT-DBP that can perform unbiased screening of chemical libraries on human tumors and function as patient biomarkers have theoretical advantages in drug discovery. Identification of effective biomarkers after drug development is nontrivial, and in the absence of accurate biomarkers, drugs that are effective in a small fraction of patients are generally not approved or are not widely adopted (39). We anticipate that the adaptation of functional biomarkers such as HT-DBP into high-throughput screening will identify new therapeutics. In addition, biomarkers like HT-DBP can then be deployed as companion diagnostics to ensure enrollment of appropriate patient populations in clinical trials and clinical practice. This approach would likely decrease the failure rates of drug development programs and provide greater clinical benefit than sequential drug discovery and biomarker identification.

MATERIALS AND METHODS

Cell culture

MDA-MB-231 and SU86.86 cells were cultured in RPMI (Gibco) with 10% fetal bovine serum (FBS) (Gibco) and penicillin-streptomycin (Gibco). When extracted, MMTV-PyMT tumor cells were cultured in RPMI with 10% FBS and 1× penicillin-streptomycin. MMTV-PyMT cell lines were derived by culture in the same media for 1 month. Cells were passaged about every 3 to 4 days. Freshly isolated murine hepatocytes were cultured in advanced Dulbecco’s modified Eagle’s medium (DMEM)/F12 (Gibco) and 10% FBS (Gibco). Human foreskin fibroblasts were cultured in DMEM/F12 (Gibco) with 10% FBS (Gibco) and penicillin-streptomycin (Gibco). Panc8902 cells were cultured in DMEM with 10% FBS and penicillin-streptomycin.

Assessment of cell death

Annexin V staining was performed by staining cells in annexin V binding buffer for 15 min, fixing cells with paraformaldehyde and glutaraldehyde for 10 min, and neutralizing fixatives with tris and glycine. Cells were washed and imaged using an IXM XLS high-content microscope (Molecular Devices), and the fraction of annexin V–positive cells was quantified using the multiwavelength cell scoring module in MetaMorph.

Animal models

For the MMTV-PyMT model, animals were maintained at the Dana-Farber Cancer Institute (Protocol 10-067). For primary screens, tumors were collected from different mammary fat pads and pooled to get appropriate numbers of cells. For animal treatments, MMTV-PyMT cells that were extracted from mice and not cultured in vivo were injected along with Matrigel (Corning) into the mammary fat pad of syngenic FVB/NJ mice as described previously (40). Seven to 15 mice were used for each in vivo treatment arm. Dasatinib (ApexBio) and 17-DMAG (ApexBio) were intraperitoneally dosed at 10 mg/kg. AZD2014 (ApexBio) was dosed via oral gavage once daily at 15 mg/kg. Lapatinib (SelleckChem) was dosed via oral gavage at 50 mg/kg. Sunitinib (ApexBio) was dosed via oral gavage at 50 mg/kg. Navitoclax (MedChemExpress) was dosed via oral gavage at 100 mg/kg. All treatments were performed daily, 5 days a week for 2 weeks.

With regard to breast PDX tumors, DF-BM355 tumors were previously treated with compounds in vivo (32) and were expanded as previously described (32). For colon PDX tumors, tissues to establish PDX models were obtained according to Institutional Review Board (IRB)–approved research protocols (14-030). Fresh primary colorectal cancer biopsies were first incubated in an antibiotic cocktail of penicillin/streptomycin/amphotericin B/ciprofloxacin for 1 to 2 hours and implanted into the flanks of 5-weeks-old female nude mice (Nu/Nu; Taconic). When the xenografts reached ~200 mm, three mice were sacrificed, and the tumors were harvested and serially passaged as subcutaneous implants of tumor fragments about 2 to 3 mm in diameter.

Human samples

Colon tumors were resected from the primary site of patients as a part of standard of care. Patient consent and tumors were obtained according to IRB research protocols (03-189). Tumors were dissociated using collagenase 4 and hyaluronidase for 30 min to 1 hour. Cells were plated and drugged with indicated compounds and were BH3-profiled 24 hours after drug treatment.

Tumor and tissue dissociation

MMTV-PyMT tumors were dissociated using gentleMACS and the Miltenyi mouse dissociation kits (Miltenyi). Colon cancer PDX models were dissociated using gentleMACS using a mixture of collagenase IV and hyaluronidase (Sigma).

Drug treatments

HT-DBP and annexin V drug treatments on MDA-MB-231 cells were performed using the Selleck Bioactive library at the Institute of Chemistry and Cell Biology (ICCB) at Harvard Medical School, consisting of 1902 bioactive compounds. Compounds were screened at a concentration of 1 μM. Screens on freshly isolated tissue from the MMTV-PyMT tumors, murine hepatocytes, and human foreskin fibroblasts were performed at the Broad Institute using a Selleck bioactive library consisting of 1650 compounds all at a concentration of 1 μM. Drug screening libraries are not randomized between technical replicates. Most of the bioactive compounds have been tested in clinical or preclinical models. Permutations of five drugs (JQ1, BKM120, everolimus, lapatinib, and MEK162) were evaluated in the DF-BM355 breast cancer model on the basis of drug treatments in the initial publication. For DF-BM355 treatments, drugs were added using the D300e (Hewlett-Packard/Tecan) digital dispenser at a concentration of 1 μM. All screens were at least performed in technical replicate. Screens on MDA-MB-231, MMTV-PyMT primary tissue and cell lines, human foreskin fibroblasts, adult mouse hepatocytes, and DF-BM355 tumors represent the average of biological replicates. Screens on colon PDX models represent the average of technical replicates. Screens on human colon primary samples represent the average of technical triplicates. Compounds from the MMTV-PyMT screen that were subsequently evaluated in vivo were first evaluated ex vivo using the same stocks of newly purchased compounds that were also used in vivo. These drugs were added using the D300e digital drug dispenser. Drug treatments where there was a drug dose response were randomized. Cells were plated in either 3712, 3764BC, or 3542 384-well plates (Corning).

High-throughput dynamic BH3 profiling

Upon drug treatment, cells were incubated at 37°C. Subsequently, medium was washed from plates using the BioTek 406EL plate washer (BioTek). Medium was replaced with PBS. A 2× concentrated BH3 profiling buffer was added to cells with appropriate levels of digitonin (0.001% for mouse cells and 0.002% for human cells) and the appropriate levels of peptide. For peptide titrations, 2× BH3 profiling buffer was added manually. For BH3 profiling of drug-treated plates at a single peptide concentration, the 2× buffer was added using the Thermo Multidrop Combi (Thermo Fisher Scientific). Cells were fixed in paraformaldehyde. Fixatives were neutralized using a tris/glycine buffer, and cells were subsequently stained with antibodies in either a saponin or Tween 20 permeabilizing solution. Cells were stained overnight, and before imaging, stain solution was washed using the BioTek plate washer. A detailed protocol for HT-DBP is provided in the Supplementary Materials (text S1).

With the exception of adult mouse hepatocytes, HT-DBP is performed using the synthetic BH3 peptide modeled after the Bim protein. The synthetic Bim BH3 peptide did not induce cytochrome c release from adult mouse hepatocytes, consistent with prior studies (41). Hepatocytes exclusively express the proapoptotic effector protein BAK, and not BAX, and more readily undergo mitochondrial outer membrane permeabilization in the presence of the synthetic Bid BH3 peptide. This is consistent with the preferential activation of BAX by Bim and the preferential activation of BAK by BID. We therefore used the synthetic BH3 peptide modeled after the Bid BH3 protein to perform HT-DBP with adult mouse hepatocytes.

Antibodies

Nuclei were stained with Hoechst 33342 (Invitrogen). Cytochrome c was measured using the Cytochrome c–Alexa Fluor 647 antibody (BioLegend). Mouse tumor cells were identified with mouse EpCam-FITC (fluorescein isothiocyanate) (BioLegend). Colon tumor cells were identified using human EpCam–Alexa Fluor 488 (BioLegend).

Imaging

All imaging was performed on the IXM XLS high-content widefield microscope (Molecular Devices; at the ICCB at Harvard Medical School or the Broad Institute). A 10× objective was used to perform all imaging. We typically used a 4′,6-diamidino-2-phenylindole (DAPI) filter cube to measure Hoechst 33342 staining, FITC to measure Alexa Fluor 488–EpCam staining, and a Cy5 cube to measure Cytochrome c–Alexa Fluor 647 antibody staining.

Image and data analysis

Image analysis was performed in MetaMorph using the multiwavelength cell-scoring module and the adaptive background correction module to segment cells on the basis of an intensity above local background. This results in an approximate single-cell segmentation (for example, shown in fig. S7) and the area of cytochrome c intensity. Cells are scored as being positive or negative on the basis of the area. All subsequent data analysis was performed in Excel or GraphPad Prism. All statistical analysis was performed in GraphPad Prism.

Statistical analysis

All chemical screens were performed in technical replicates. Correlation of technical replicates and biological replicates of HT-DBP or annexin staining were evaluated using Pearson two-tailed tests, as indicated in the figure legends. Correlation of ex vivo and in vivo responses of breast cancer models was performed using Pearson two-tailed tests. Identification of nominal protein/pathway targets that increase apoptotic sensitivity was performed using one-way analysis of variance (ANOVA), as indicated in the text. Correlation of HT-DBP with annexin staining in breast cancer cells was performed using Pearson two-tailed tests.

SUPPLEMENTARY MATERIALS

stke.sciencemag.org/cgi/content/full/13/636/eaay1451/DC1

Text S1. Detailed HT-DBP protocol.

Fig. S1. Schematic of DBP.

Fig. S2. Selection of informative peptide screening concentrations.

Fig. S3. Identification of a drug screening concentration.

Fig. S4. HT-DBP enables identification of compounds that sensitize cancer cells for apoptosis.

Fig. S5. Similarity between FACS and microscopy dynamic BH3 profiles.

Fig. S6. Images of stained MMTV-PyMT tumor cells.

Fig. S7. Example of cell masks.

Fig. S8. Screening data for freshly isolated MMTV-PyMT tumors.

Fig. S9. Counterscreens in healthy cells.

Fig. S10. Supplemental data for in vivo validation experiments for HT-DBP.

Fig. S11. Identification of compounds that sensitize the COCA9 colon cancer PDX for apoptosis.

Fig. S12. Quantification of apoptotic chemical vulnerabilities in colon cancer PDX models.

Fig. S13. Heat map of apoptotic chemical vulnerabilities in colon cancer PDX models.

Fig. S14. Comparison of apoptotic chemical vulnerabilities in colon cancer PDX models.

Fig. S15. Supplemental data for comparison of chemical vulnerabilities of freshly isolated and cultured cancer cells.

Fig. S16. Supplemental data for comparison of chemical vulnerabilities of freshly isolated and cultured cancer cells.

Table S1. List of compounds used in the chemical screen.

Data file S1. Cell count and delta priming of MDA-MB-231 line.

Data file S2. Apoptosis and delta priming of MDA-MB-231 line.

Data file S3. Cell count and delta priming of MMTV-PyMT tumor.

Data file S4. Technical replicate of MMTV-PyMT delta priming.

Data file S5. Biological replicate of MMTV-PyMT delta priming.

Data file S6. Cell count and delta priming of primary mouse hepatocytes.

Data file S7. Cell count and delta priming of human foreskin fibroblasts.

Data file S8. Delta priming in adult mouse hepatocytes and MMTV-PyMT tumors.

Data file S9. Delta priming of freshly isolated MMTV-PyMT tumors by nominal drug target.

Data file S10. Normalized delta priming of colorectal PDX models.

Data file S11. Normalized delta priming of colorectal PDX models by nominal target.

Data file S12. Delta priming of MMTV-PyMT–derived cell line.

Data file S13. Correlation between MMTV-PyMT tumor and derived cell line.

Data file S14. Dose response by delta priming in MMTV-PyMT tumor and derived cell line.

Data file S15. Delta priming in colorectal PDX models.

Data file S16. Delta priming in primary human CRC cells ex vivo.

REFERENCES AND NOTES

Acknowledgments: We acknowledge the support from the cDOT group (Broad Institute). We thank P. Sorger, L. Mailszewski, C. Shamu, and J. Smith for support at the LSP, ICCB, and HiTS at Harvard Medical School. We acknowledge D. Tuveson (Cold Spring Harbor Laboratory) for useful discussions and for coining the term “pharmacotype.” Funding: A.J.A. acknowledges support from the Doris Duke Charitable Foundation, the Pancreatic Cancer Action Network, and National Cancer Institute K08 CA218420-01 and P50CA127003. A.L. acknowledges support from R01 CA205967, R35 CA242427, Ludwig Cancer Research at Harvard, and the Starr Cancer Consortium. P.D.B. acknowledges support from the Barr Foundation. K.N. acknowledges support from P50 CA127003, R01 CA205406, and the Project P Fund. J.J.Z. acknowledges support from R35 CA210057, P50 CA168504, DoD W81XWH-18-1-0491, and the Breast Cancer Research Foundation. Author contributions: P.D.B. and A.L. designed the study and wrote the manuscript. P.D.B., E.A., E. Su, J.L.G., A.S., J.N., O.C., T.H., E.L., K.M., and M.S.P. performed experiments. A.J.A., D.C., E. Sicinska, J.M.C., and J.R. provided technical assistance or assisted in the design of some experiments. J.L.G., J.N., K.N., and J.J.Z. assisted in the design of in vivo experiments. Competing interests: A.L. discloses consulting and sponsored research agreements with AbbVie, Novartis, and Astra-Zeneca. He is an equity-holding founder of Flash Therapeutics and is on the SAB of Dialectic Therapeutics. The following are U.S. Patents regarding BH3 profiling owned by Dana-Farber: 10,393,733; 9,902,759; 9,856,303; 9,540,674; 8,221,966; and 7,868,133. A.L., J.R., and P.D.B. are inventors on patent applications US20180128813A1 and US20180120297A1 held/submitted by the Dana-Farber Cancer Institute that covers high-throughput BH3 profiling. A.J.A. is a consultant for Oncorus Inc. K.N. discloses sponsored research agreements with Pharmavite, Genentech, Gilead Sciences, Celgene, Trovagene, Tarrex Biopharma, Revolution Medicines, and Evergrande Group. K.N. discloses serving on advisory boards for Genentech, Lilly, Bayer, Seattle Genetics, and Array BioPharma. K.N. has served as a paid consultant for Tarrexx Biopharma. J.N. has a consulting relationship with Geode Therapeutics. J.J.Z. is a founder and board director of Crimson Biotech and Geode Therapeutics. J.L.G. is a consultant for GlaxoSmithKline (GSK) and Array BioPharma and receives sponsored research support from GSK and Eli Lilly. All other authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.

Stay Connected to Science Signaling

Navigate This Article