Probabilistic modeling of personalized drug combinations from integrated chemical screen and molecular data in sarcoma

Abstract

Background: Cancer patients with advanced disease routinely exhaust available clinical regimens and lack actionable genomic medicine results, leaving a large patient population without effective treatments options when their disease inevitably progresses. To address the unmet clinical need for evidence-based therapy assignment when standard clinical approaches have failed, we have developed a probabilistic computational modeling approach which integrates molecular sequencing data with functional assay data to develop patient-specific combination cancer treatments.

Methods: Tissue taken from a murine model of alveolar rhabdomyosarcoma was used to perform single agent drug screening and DNA/RNA sequencing experiments; results integrated via our computational modeling approach identified a synergistic personalized two-drug combination. Cells derived from the primary murine tumor were allografted into mouse models and used to validate the personalized two-drug combination. Computational modeling of single agent drug screening and RNA sequencing of multiple heterogenous sites from a single patient's epithelioid sarcoma identified a personalized two-drug combination effective across all tumor regions. The heterogeneity-consensus combination was validated in a xenograft model derived from the patient's primary tumor. Cell cultures derived from human and canine undifferentiated pleomorphic sarcoma were assayed by drug screen; computational modeling identified a resistance-abrogating two-drug combination common to both cell cultures. This combination was validated in vitro via a cell regrowth assay.

Results: Our computational modeling approach addresses three major challenges in personalized cancer therapy: synergistic drug combination predictions (validated in vitro and in vivo in a genetically engineered murine cancer model), identification of unifying therapeutic targets to overcome intra-tumor heterogeneity (validated in vivo in a human cancer xenograft), and mitigation of cancer cell resistance and rewiring mechanisms (validated in vitro in a human and canine cancer model).

Conclusions: These proof-of-concept studies support the use of an integrative functional approach to personalized combination therapy prediction for the population of high-risk cancer patients lacking viable clinical options and without actionable DNA sequencing-based therapy.

Keywords: Artificial intelligence and machine learning; Combination therapy; Computational modeling; Drug screening; High-throughput sequencing; Pediatric cancer; Personalized therapy; Sarcoma.

Fig. 1

Schematic representation of experimental and computational approach to personalized combination targeted therapy predictions. Following tumor extraction and culture establishment, biological data is generated (e.g., chemical screening, transcriptome sequencing, exome sequencing, siRNA interference screening and phosphoproteomic analysis) and used as input for PTIM modeling. To briefly explain the graphical model representation, targets A and B denote two independent single points of failure. Targets C and D denote parallel targets, which independently are not predicted to be effective, but together will be synergistic and lead to significant cell growth inhibition. Targets A, B, and the C-D parallel block are in series and may target independent pathways. Series blocks, when inhibited together, may abrogate cancer resistance mechanisms by knockdown of independent pathways. Model sensitivity scores for gene target combinations are used to design and rank follow-up in vitro validation and in vivo validation experiments. The “Exome-Seq” representative images was adapted from an image on the Wikipedia Exome sequencing article originally created by user SarahKusala and available under Creative Commons 3.0 license. An unaltered portion of the image was used. The mouse image used is public domain and accessed through Bing image search at the following weblink: http://img.res.meizu.com/img/download/uc/27/83/20/60/00/2783206/w100h100

Fig. 2

Probabilistic Target Inhibition Maps (PTIMs) and experimental in vitro and in vivo results for U23674 alveolar rhabdomyosarcoma (aRMS) drug combinations. Targets with adjacent asterisks indicate targets selected for in vitro validation. Values in the center of PTIM blocks represent expected scaled sensitivity following inhibition of associated block targets. a Abbreviated baseline chemical screen-informed PTIM. b Abbreviated chemical screen RNA-seq + informed PTIM. c Abbreviated chemical screen + exome-seq informed PTIM. The values within the target blocks indicate scaled drug sensitivity for the given target combination [16] when the targets are inhibited via one or more chemical compounds. More information can be found in prior publications [16, 18]. In (d-e), results are based on n = 3 technical replicates with n = 4 replicates per treatment condition. d Dose response curve for OSI-906 varied dosage + GDC-0941 fixed dosage. The response for GDC-0941 at varied dosages is included. e Dose response curve for GDC-0941 varied dosage + OSI-906 fixed dosage. The response for OSI-906 at varied dosages is included. f Schematic representation of in vivo experiment design. g Kaplan-Meier survival curves for in vivo orthotropic mouse experiment. Mice were treated with vehicle (n = 8 mice, black line), 50 mg/kg OSI-906 (n = 8 mice, blue line), 150 mg/kg GDC-0941 (n = 7 mice, red line), or combination 50 mg/kg OSI-906 + 150 mg/kg GDC-0941 (n = 8 mice, purple line). The medicine bottle image is public domain, provided by user Kim via clker.com (http://www.clker.com/clipart-blank-pill-bottle-3.html)

Fig. 3

New cell cultures and patient-derived xenograft model of EPS with chemical space characterization. a PCB490 biopsy sample divided into distinct regions to create different primary tumor cell cultures for study. b Western blot demonstrating loss of INI1 in multiple primary tumor sites and in published EPS cell lines. c Histology of surgical biopsy of PCB490. d Immunohistochemical staining of PCB490 for INI1 shows absence in tumor cells (black arrow) but presence in co-mingled non-cancerous cells. e Histology of PCB490 patient-derived xenograft. f INI1 absence (black arrow) in immunohistochemical staining of PCB490 patient-derived xenograft. g Drug Screen V3 results from primary EPS cell cultures, published EPS cell lines, and a normal myoblast cell line. The heat values indicate drug sensitivity as IC50 values, scaled between 10 nM (red) and 10 μM (white, representing no IC₅₀ achieved) h Heatmap of Pearson correlation coefficients of 60-agent drug screen results between a normal myoblast cell line (SkMC), three EPS cell lines (ESX, VA-ES-BJ, FU-EPS-1), three sites from PCB490 (PCB490–3, PCB490–4, PCB490–5), and an additional EPS patient-derived culture (PCB495). The heat values correspond to Pearson correlation coefficients between drug sensitivities of different cell models

Fig. 4

Probabilistic Target Inhibition Maps (PTIMs) of Drug Screen V3 and Roche screen results for spatially-distinct epithelioid sarcoma tumor regions. Values in the center of PTIM blocks represent expected scaled sensitivity following inhibition of associated block targets. a-c PTIMs informed by Roche Orphan Kinase Library and V3 screens. Targets of sunitinib are highlighted red, targets of BEZ235 are highlighted blue. a Abbreviated PTIM for PCB490–3. b Abbreviated PTIM for PCB490–4. c Abbreviated PTIM from PCB490–5 with integrated RNA-seq data. d Results from PCB490–5 patient-derived xenograft in vivo validation studies presented as group-wide tumor volumes following vehicle treatment (n = 3 mice, green line), treatment by 30.0 mg/kg sunitinib (n = 3 mice, red line), treatment by 25.0 mg/kg BEZ235 (n = 3 mice, blue line), and treatment by 25.0 mg/kg BEZ235 + 30.0 mg/kg sunitinib (n = 3 mice, purple line)

Fig. 5

Undifferentiated pleomorphic sarcoma (UPS) Probabilistic Target Inhibition Map (PTIM)-guided resistance abrogation experiments. Values in the center of PTIM blocks represent expected scaled sensitivity following inhibition of associated block targets. a Histology of PCB197 human UPS sample (20x magnification). b Histology of S1–12 canine UPS sample (20x magnification). c Abbreviated PTIM model for the pediatric preclinical testing initiative (PPTI) screen of PCB197 human UPS sample. d Abbreviated PTIM model built from the PPTI screen of S1–12 canine UPS sample. e Schematic of experimental design for resistance abrogation experiments. f Cellular regrowth of PCB197 human UPS sample over 100 days following treatment by single and multi-agent compounds in sequence and in combination. g Cellular regrowth of S1–12 canine UPS sample over 100 days following treatment by single and multi-agent compounds in sequence and in combination. Data in (f-g) is based on n = 4 replicate experiments