Personalized In Vitro and In Vivo Cancer Models to Guide Precision Medicine

Abstract

Precision medicine is an approach that takes into account the influence of individuals' genes, environment, and lifestyle exposures to tailor interventions. Here, we describe the development of a robust precision cancer care platform that integrates whole-exome sequencing with a living biobank that enables high-throughput drug screens on patient-derived tumor organoids. To date, 56 tumor-derived organoid cultures and 19 patient-derived xenograft (PDX) models have been established from the 769 patients enrolled in an Institutional Review Board-approved clinical trial. Because genomics alone was insufficient to identify therapeutic options for the majority of patients with advanced disease, we used high-throughput drug screening to discover effective treatment strategies. Analysis of tumor-derived cells from four cases, two uterine malignancies and two colon cancers, identified effective drugs and drug combinations that were subsequently validated using 3-D cultures and PDX models. This platform thereby promotes the discovery of novel therapeutic approaches that can be assessed in clinical trials and provides personalized therapeutic options for individual patients where standard clinical options have been exhausted.Significance: Integration of genomic data with drug screening from personalized in vitro and in vivo cancer models guides precision cancer care and fuels next-generation research. Cancer Discov; 7(5); 462-77. ©2017 AACR.See related commentary by Picco and Garnett, p. 456This article is highlighted in the In This Issue feature, p. 443.

Conflict of interest statement

The authors declare no potential conflicts of interest.

©2017 American Association for Cancer Research.

Figures

Figure 1

Whole exome sequencing detects a limited number of clinically targetable alterations in patients with advanced cancer. A, Overview of the sites of specimen origin collected from the patients and run through the EXaCT-1 test. The majority of these samples were taken from metastatic sites of patients with advanced disease. B, Whole Exome Sequencing has been performed on a total of 769 specimens. Data presented here includes also large scale deletions (>50 genes) and each gene is individually included in the analysis. In 85.8% (660/769) of the cases sequenced somatic alterations in currently not targetable cancer genes were detected. In total three cases (0.4%), two gastrointestinal stromal tumors (GISTs) with an activating KIT mutation and a clear cell renal cell carcinoma with a BRAF mutation, have FDA approved drugs available. In 9.6% (71/737) of these, we have somatic alterations in cancer genes that could be clinically actionable by off – label use of approved drugs, however clinical efficacy has not been proven. In 4.2% (32/737) of the cases we did not detect any somatic alterations in known cancer genes. C, Bar graph showing, above the x-axis the number of alterations in cancer genes detected in each case (gray). Below the x-axis we show in green (*) the three cases that have FDA approved drugs available and in orange the cases that have clinically actionable gene alterations by potential off-label use of FDA approved drugs. D, List of the 20 most relevant SNVs/indels in cancer genes detected in our cohort. Cancer genes (red) that have FDA approved drugs available for NSCLC (crizotinib, erlotinib, gefitinib) and for ovarian cancer (Olaparib). E, The 20 most common SCNA detected, genes that have FDA approved drugs available (green) for ovarian cancer (olaparib) and for chronic lymphocytic leukemia (venetoclax).

Figure 2

Personalized models to guide precision medicine in advanced cancer. Illustration of our precision medicine program depicting the workflow, beginning with sequencing using the EXaCT-1 whole exome sequencing test (top row), continuing with the establishment of PDTOs, which are compared to the primary tumor sample through histology and sequencing before they are subjected to drug screening (middle row), and utilized to generate PDXs where potential drugs are validated in mice (bottom row). The sequencing data is available in our internal cbio-portal and reported back to the referring physician. Tumor organoid cultures are prepared from fresh patient tumor samples as personalized in vitro models. After the initial characterization, targeted or high throughput single and combination drug screens can be performed in an iterative process in order to nominate therapeutic strategies that are further evaluated in personalized in vivo models.

Figure 3

Development of pre-clinical models for the guidance of precision medicine. A, Of the 769 samples that have been run through the sequencing program, 50.9% were FFPE specimens, 44.5% were freshly collected tissue specimens, and a minority of 4.6% were from patients with hematologic malignancies sent to us as DNA specimens. B, 152/342 of the freshly collected specimens had a tissue biopsy or resection specimen to attempt the development of tumor organoids. Of these 152, 56 (36.8%) patient derived organoids were successfully initiated from numerous tissues including prostate (10/52), bladder/ureter (8/24), kidney (6/10), breast (4/6), colon/rectum (8/10), esophagus (1/6), soft tissue (3/6), brain (5/9), pancreas (5/7), lung (1/2), small intestine (2/3), ovary (1/1) and uterus (2/2). C, 22 of these patient derived organoid models were subsequently injected into mice of which 19 organoid lines from colorectal cancer (CRC) (7), pancreatic ductal adenocarcinoma (PDAC) (3), uterine cancers (2), neuroendocrine prostate cancer (NEPC) (2), renal cell carcinoma (RCC) (2), urothelial cancer (1), lung adenocarcinoma (1), and sarcoma (1) successfully engrafted. D, Histology of primary tumor samples, tumor organoids and xenografts from six different solid tumors; endometrial adenocarcinoma, pancreatic ductal adenocarcinoma, colorectal cancer, uterine carcinosarcoma, urothelial carcinoma and renal cell carcinoma. Tumor organoid gross morphology (row 2) shows tumor type specific structures such as the formation of lumina as seen for pancreas and colon. Tumor organoid and PDX histology shows conservation of the histopathological features of the native tumors. H&E stain from native tumor tissue, scale bar 200μM, tumor organoids bright field view in vitro, scale bar 10μM, H&E stain from tumor organoids, scale bar 20μM (urothelial carcinoma 200μM), H&E stain from PDXs, scale bar 20μM.

Figure 4

Allele specific copy number heat map showing the genomic characterization of patient derived in vitro and in vivo models. Allele-specific copy number of 1,062 putative cancer genes (columns) derived from 57 whole exome tumor tissues samples (rows) from 15 patients. Six copy number states are represented: homozygous (Homo del) and hemizygous deletions (Hemi del), Wild type, low-level gain (Gain), and high-level amplification (Amp), loss of heterozygosity (Loss of het). Relevant biomarkers are highlighted at the bottom. Left column annotation reports the tissue type (N: native tumor, O: tumor organoid and P: organoid derived PDX). The different numbers for T denote different locations of the native tumor samples sequenced. Different O numbers indicate tumor organoids in culture over time (O1-passage 5, O2 – passage 10, O3- passage 15, O4 – passage 20, O5- passage 30). P1 is the initial xenograft derived from tumor organoid passage < 20 and P2 is an orthotopic xenograft (WCMC601 intrauterine). Right column annotation reports the tumor type of the patient.

Figure 5

High throughput drug screen and validation to nominate patient specific therapeutic options. Once organoids have been established and validated the cells can be utilized to identify patient specific responses to therapeutic agents through the use of selected or high throughput drug screens. A, D, G, J, Heat maps of the drug screen results depicting the relative sensitivity of the patients' tumor cells from most resistant (red) to most sensitive (blue). Black dots indicate agents that were selected for validation and further studies. B, E, H, K, Graphs of the response of Patient's tumor cells to each compound in the library as a Z-score (AUC) compared to other primary cells screened with the same library (N(SEngine) = 43, N(NCI) = 10). C, F, I, L, Depict the in vitro validation of selected drugs in the 3D system. Patient A: The patient's tumor cells were generally resistant to many agents. Selective enrichment was seen for targeted agent classes such as PI3K (AZD8482, buparlisib, GDC-0980, idelalisib, taselisib, PIK-75, NVP-BGT226) and HDAC inhibitors (vorinostat, belinostat). Drug sensitivity was validated using the 3D matrigel system and compared to the patient's actual treatment paclitaxel and carboplatin as single agents. Patient B: The high throughput drug screen demonstrated that cells from Patient B were responsive to a broad array of chemotherapeutic drugs including antimetabolites methotrexate and fludarabine phosphate. Mitoxantron, and paclitaxel as part of the patient's actual treatment and topotecan, were also effective in these cells. The cells showed sensitivity to several classes of targeted agents including inhibitors of PI3K (AZD8482, buparlisib, GDC-0980, idelalisib, taselisib, PIK-75,) and HDAC (vorinostat, belinostat). Patient C: Tumor cells showed resistance to most chemotherapeutics and targeted agents (as indicated in the heat map), high sensitivity was seen for the targeted agent trametinib, a MEK inhibitor. Drug sensitivity was validated in our 3D matrigel system using also oxaliplatin and 5-FU as comparison to what the patient initially received. Patient D: Patient D harbors an APC mutation and a frameshift deletion. The tumor cells were sensitive to a small number of drugs, including EGFR inhibitors, particularly for afatinib and showed sensitivity to neratinib. Drug sensitivity was validated in our 3D matrigel system using oxaliplatin and 5-FU as comparison to what the patient initially was treated with.

Figure 6

High throughput combination drug screening to nominate potent drug combinations. Using a combination of the genomics (EXaCT-1) and drug sensitivities from the primary drug screen, secondary drug screens were performed on the patient derived tumor organoids from the same four cases. A, D, G, J, Heat maps of the single therapy (left) and combination drug (right) screens depicting the relative sensitivity of the patients' tumor cells from most resistant (red) to most sensitive (blue). Black dots indicate agents that were selected for validation. B, E, H, K, Depict graphs of the response of patient's tumor cells to each compound in the library as a Z-score in the presence of the combination treatments (y-axis) and the fold change of the IC50 identified with each compound in combination as compared to the agent as a mono-therapy. C, F, I, L, Depict the in vitro validation of selected drugs in our 3D matrigel system. Patient A: A sensitizing drug screen using a PI3K inhibitor (buparlisib), as the investigational most advanced selective inhibitor of p110α/β/δ/γ did show an enhanced drug effect for HDAC inhibitors (vorinostat, belinostat) (Supplementary Table 3). However, a high throughput combination drug screen using vorinostat (at its IC30) showed that many of the investigational PI3K/AKT pathway drugs were enhanced compared to monotherapeutic use (Supplementary Table 4) A significant difference using vorinostat (IC30) with buparlisib in combination was seen compared to buparlisib alone (2 Way ANOVA: ****, p-value < 0.0001). Vorinostat also enhanced the effects of EGFR inhibitors in this patient's tumor cells (Supplementary Table S2). Patient B: Sensitizing screen with buparlisib for Patient B enhanced the drug effects of drugs such as HDAC inhibitors and Olaparib (Supplementary Table 5). For the validation of the olaparib and buparlisib combination we extended the drug assay up to 6 days and noticed a significant difference compared to olaparib as monotherapy (2 Way ANOVA: ****, p-value < 0.0001). Patient C: The combination drug screen using trametinib has increased the efficacy of several compounds (Supplementary Table 6). As an example celecoxib, which did not have an effect on cell survival as a single compound activity but shows a significant effect in combination (2 Way ANOVA: ****, p-value < 0.0001). Patient D: Afatinib sensitizing drug screen for Patient D showed enhanced effects for HDAC and IGF-1R inhibitors, which were confirmed in our 3D matrigel system (Supplementary Table 7). A significant difference using vorinostat with afatinib (IC30) in combination was seen compared to vorinostat alone (2 Way ANOVA: ****, p-value < 0.0001).

Figure 7

In vivo validation of drug screens. A, Patient B xenografts were treated with carbotaxol (patient's actual treatment) or vehicle, or compounds that were identified through the single compound or combination drug screens. Tumor volumes of these mice are shown on the left and their tumor mass after 10 days of treatment are shown on the right. A significant effect upon tumor growth was observed with combinations of buparlisib and olaparib (2 Way ANOVA: ** p-value = 0.0071) as well as buparlisib and vorinostat (2 Way ANOVA: **, p-values = 0.0059) when compared to carbotaxol. Similarly when the mass of the tumors at endpoint were compared to the combinations of buparlisib and olaparib (Student's t-test: * p-value 0.0205) and buparlisib and vorinostat (Student's t-test: *, p-value 0.0369) showed significant improvement over crabotaxol. Error bars indicate the standard deviation. B, Representative (H&E) stained section and IHC against the proliferation marker Ki67 of representative tumors from each of the treatment groups, scale bar 300 μM. carbotaxol and monotherapy treatments showed identical KI67 positivity as seen in the vehicle treated group. In both combination treatments we only noticed few more individual cells that were not proliferating compared to the vehicle and other treatments, but in general there is no relevant difference noticeable. C, Patient D xenografts were treated with FOLFOX (the patient's actual treatment) or vehicle, or compounds that were identified through the single compound or combination drug screens. Tumor volumes of these mice are shown on the left and their tumor mass after 14 days of treatment are shown on right. As compared to FOLFOX both afatinib alone (2 Way ANOVA: *, p-value 0.025) and afatinib with vorinostat (2 Way ANOVA: *, p-value 0.0232) showed more tumor regression than FOLFOX. The difference was also significant in the tumor mass at harvest between afatinib alone (Student's t-test: *, p-value 0.0362) and afatinib with vorinostat compared to FOLFOX (Student's t-test: *, p-value 0.0368). Error bars indicate the standard deviation. D, Representative H&E stained section and IHC against the proliferation marker Ki67 of representative tumors from each of the treatment groups, scale bar 300μM. The xenografts that have been treated with afatinib alone and afatinib in combination with trametinib and vorinostat showed clear reduction in the Ki67 positive cells when compared to the vehicle, FOLFOX and vorinostat as monotherapy treated groups.