Background
Colorectal cancer (CRC) continues to be a major public health problem, both in the United States and worldwide; it is the third most common cancer in the United States with approximately 150,000 new cases per year [
1,
2]. Metastatic disease currently remains predominantly incurable, and treatment is primarily for palliation of symptoms and disease control. In general, 5-fluorouracil (5-FU)-containing regimens have formed the backbone of chemotherapy to treat CRC for the last several decades. Recently, additional compounds have proven to be effective as treatment in first, second, and third line metastatic disease. These include both traditional chemotherapeutic agents along with targeted biologic agents [
1‐
3]. Although there have been great strides made to improve the survival of patients with metastatic CRC, the median survival for patients still remains at a mere 30 months [
1‐
3].
Over the past decade, targeting molecular pathways of tumor growth/proliferation has become a major focus of anti-cancer treatments to develop new and novel drugs in CRC. For example, agents like bevacizumab, which targets the vascular endothelial growth factor (VEGF) pathway, or cetuximab and panitumumab, which target the epidermal growth factor receptor (EGFR) pathway, have become standard-of-care therapies. However, once patients have completed treatment or become resistant to these currently-available treatments, there are no effective options left for patients. Unfortunately, new drugs for the treatment of metastatic CRC have been limited, and over the past few years, only two drugs, regorafenib and lonsurf, have been approved in the refractory setting for the treatment of metastatic CRC.
Like most other cancers, the failure rate for new cancer drugs is more than 80% in Phase II and 50% in Phase III [
4,
5], and failure rates for both Phase II and Phase III oncology clinical trials have been rising since 2001. Part of the high failure rate results from a relative lack of models that faithfully recapitulate the disease state. To address this lack of models, researchers have turned to patient-derived models of cancer, such as cell lines, organoids, and patient-derived xenografts (PDXs), which are increasingly being accepted as “standard” preclinical models to facilitate the identification and development of new therapeutics. For example, large-scale drug screens of cancer cell line panels have been used to identify sensitivity to a large number of potential therapeutics [
6]. Similarly, tumor organoid cultures from CRC specimens have also been used to perform drug screens [
7], and PDXs of CRC are also being used to predict drug response [
8] and to identify novel drug combinations [
9]. Finally, combinations of patient-derived models are currently being explored to develop precision medicine strategies for cancer care [
10].
In the current study, we developed a precision medicine strategy for patients with metastatic CRC. Specifically, we developed a series of patient-matched cell lines and PDXs. The cell lines were first used to perform high throughput drug screens to identify potential therapeutic targets, and the matched PDXs were then used to validate these findings. Using this approach, we observed patient-specific heterogeneity in response to both standard-of-care agents and targeted therapies. Among the targeted therapies, ponatinib and trametinib were the most efficacious for different patient-derived models. Further mechanistic studies of ponatinib’s downstream targets demonstrated potential antitumor activity by co-targeting the fibroblast growth factor receptor (FGFR), SRC, platelet derived growth factor receptor (PDGFR) or vascular endothelial growth factor receptor (VEGFR) signaling. Consistent with these observations mining of next-generation RNA sequencing (RNA-Seq) data identified mutations in these pathways as potential molecular predictors of response. Together, our results support the use of a precision medicine pipeline to identify personalized therapies and predictive biomarkers for the treatment of metastatic CRC.
Methods
Generation of patient-derived Xenograft models and matched PDX cell lines
Patient derived CRC tumor tissue samples were collected under a Duke Institutional Review Board (IRB) approved protocol (Pro00002435). All participants provided written informed consent to participate in the study. PDX models of CRC were then generated as described previously [
11,
12], and all in vivo mouse experiments were performed in accordance with the animal guidelines and with the approval of the Institutional Animal Care and Use committee (IACUC) at the Duke University Medical Center. Briefly, to generate PDXs, tissue samples were washed in phosphate buffered saline (PBS), dissected into small pieces (< 2 mm), and injected into the flanks of 8–10-week-old JAX NOD.CB17-PrkdcSCID-J mice. Mice were purchased from the Duke University Rodent Genetic and Breeding Core and housed in IVC cages containing corn cob using the day to night pattern (7 am- 7 pm) lightening control.
Matched PDX cell lines were generated from the PDXs as follows. Once the PDX tumors reached a size of > 1000 mm3, tumors were harvested, homogenized and grown in 10 cm2 tissue culture-treated dishes in cell culture media (DMEM media, 10% fetal bovine serum (FBS), 10 U/ml penicillin and streptomycin) at 37 °C and 5% CO2. Clonal populations of each cell line were then obtained by isolating a single clone using trypsinization of the clone sealed off from the dish by an O ring. The following matched CRC PDXs and cell lines were generated and used in this study; CRC119, CRC057, CRC240, CRC247, 16–159 and 15–496. Cell lines were authenticated using the Duke University DNA Analysis Facility Human cell line authentication (CLA) service by analyzing DNA samples from each individual cell line for polymorphic short tandem repeat (STR) markers using the GenePrint 10 kit from Promega (Madison, WI, USA).
High-throughput screening
Automated liquid handling was provided by the Echo Acoustic Dispenser (Labcyte) for drug addition or Well mate (Thermo Fisher) for cell plating, and assays were performed using a Clarioscan plate reader (BMG Labtech). Immediately prior to cell plating, 384 well plates were stamped with 119 FDA-approved drug compounds at a final concentration of 1 uM. The compound library (Approved Oncology Set VI) was provided by the NCI Developmental Therapeutics Program (
https://dtp.cancer.gov/). CRC119, CRC057, CRC240, CRC247, 16–159 and 15–496 cell lines were plated in these drug pre-coated plates in a range of 500 and 1000 cells/well. Cell viabilities were assessed via CellTiter-Glo Luminescent Cell Viability Assay (Promega, USA) 72 h after cell plating. Percent killing was quantified using the following formula: 100*[1-(the value average CellTiterGlo
drug/average CellTiterGlo
DMSO)].
In vitro drug sensitivity assays
CRC119, CRC057 and CRC240 cell lines were cultured in DMEM + 10% FBS + 1% Penicillin/Streptomycin and plated in drug-free medium at concentrations between 3000 and 6000 cells/well in 96 well plate. Ponatinib (AP24534) was purchased from Selleck Chemicals (Houston, TX) and was solubilized in DMSO to a final concentration of 50 mM. Five replicates were used for each drug concentration. Each cell line was exposed to a series of seven different drug concentration (1.6 nM – 25 μM) after 24 h of incubation at 37 °C. Cell viability was measured 72 h following the addition of DMSO or drug via CellTiter-Glo Luminescent Cell Viability Assays (Promega, USA), and IC50 values were calculated for each cell line using GraphPad Prism software (La Jolla, CA, USA).
In vivo drug sensitivity assays
To test the sensitivity of CRC119, CRC057 and CRC240 PDXs to ponatinib, oxaliplatin and irinotecan, 150 μl of homogenized PDX tissue-PBS suspensions at 150 mg/ml concentration were subcutaneously injected into the right flanks of 5 female and 5 male mice (JAX NOD.CB17- PrkdcSCID-J, 10 weeks old, ~ 25 g). Following injection, mice were randomized into control and treatment groups. 5 times a week in the morning oral dosing of ponatinib (30 mg/kg); 5 times a week in the morning intraperitoneal dosing of oxaliplatin (10 mg/kg) and irinotecan (10 mg/kg) were initiated when tumor volumes reached approximately 150 mm3. Tumor volume measurements were performed every other day using calipers, and the following formula was used to calculate tumor size: (length x (width)2)/2. The mice were euthanized by bilateral thoracotomy under CO2 induced anesthesia at the end of the study.
Western blotting analysis
Western blots were performed pre- and 24-h post-treatment with vehicle (DMSO) or ponatinib at IC50 doses of each cell line. A total of 100,000 cells were lysed in radioimmunoprecipitation assay (RIPA) lysis buffer supplemented with protease and phosphatase inhibitor cocktail (company), and a total of 50 μg of RIPA lysate was electrophoretically separated at 200 V on 4–20% sodium dodecyl sulfate polyacrylamide gels using a BioRad MiniProtean Tetra system. Subsequent to transfer onto nitrocellulose membranes at 50 V for 2 h, membranes were blocked in StartingBlock T20 (ThermoFisher) for 1 h at room temperature, followed by incubation in primary antibody diluted in StartingBlock T20 overnight at 4 °C with rocking. Membranes were washed three times for 5 min each in PBS + 0.05% Tween-20, incubated in corresponding Horse Radish Peroxidase (HRP) conjugated secondary antibodies according to the specifications of the manufacturer’s protocols. The Odyssey Infrared Imaging System (LI-COR Biosciences) was used for membrane imaging. The following primary antibodies and dilutions were used: FGFR1 (#9740), FGFR2 (#11835), pFGFR (#3471), pSRC (#2105), pVEGFR (#12599), pERK (#4377), pAkt (#4060), pSTAT5 (#4322), pSTAT3(#9131), β-Actin (#4970) antibodies (Cell Signaling Technology Inc., USA); pPDGFR (ab5460) antibody (Abcam, Cambridge, MA, USA); and pABL (sc-293,130), FGFR3 (#sc-390,423), FGFR4 (#sc-136,988) (Santa Cruz Biotechnology, Santa Cruz, CA, USA). All antibodies were used at 1:1000 dilutions.
Data analysis and statistics
GraphPad Prism 6 software (La Jolla, CA, USA) was used for in vitro and in vivo data recording and statistical analysis. 2-way ANOVA analysis was used to compare the tumor size between control groups and treatment groups in vivo and drug sensitivity in vitro. A p-value < 0.05 was considered statistically significant.
RNA-Seq analysis
The RNA-seq libraries were prepared and sequenced in Illumina HiSeq 4000 with 150 bp paired-end reads. The reads were aligned to human genome hg19. In variant calling analysis, pipeline of GATK [
13] developed by Broad Institute is followed (
https://software.broadinstitute.org/gatk/). One hundred fifty bp PE reads were first aligned using STAR-2pass method with default parameters. The output SAM files were processed by using Picard (
http://broadinstitute.github.io/picard/) subsequently to add read group, sort, mark duplicates and index. GATK tool was used for variant calling, and SnpEff was used to annotated the identified variants. In fusion analysis, STAR-Fusion (
https://www.biorxiv.org/content/early/2017/03/24/120295) package developed by the Broad Institute was applied to detect fusion reads in the paired-end RNA-seq data with default parameters.
Discussion
Patient derived models of cancer are accepted as efficient tools for the development of the cancer therapeutics [
21]. Specifically, morphological and molecular mimicry between these models and the original patient tumors facilitate the evaluation of anticancer drug responses and resistance [
22]. Recently, high-throughput drug screens of patient-derived organoids followed by validation of drug candidates in patient derived xenograft (PDX) models has been coupled with genomic analysis to develop personalized medicine platforms in various types of cancer [
10]. Specifically, in colorectal cancer (CRC), matched PDX and cell line platform has been used as a preclinicaltool for functional gene validation and proof-of-concept studies to identify novel druggable vulnerabilities [
23]. Additional studies such as using tumor organoid cultures from CRC specimens to perform drug screens [
7] or patient derived xenografts of CRC to predict drug response [
8] have paved the way for the use these patient derived models of cancer to identify and develop new therapeutics.
In this study, we have developed our own precision medicine strategy for patients with metastatic CRC using matched cell lines and PDX platform coupled with high throughput drug screens and genomic analyses to identify novel targets and potential predictive biomarkers. The similar responses of the matched cell lines and PDX tumors to standard-of-care CRC treatment agents, including oxaliplatin and irinotecan, suggest that our strategy is a reliable means to identify effective therapies. Interesting enough, our one patient who received neoadjuvant thearpy prior to resection of their cancer (CRC119) was found to be responsive to an irinotecan therapy and simlary her cell and PDX were also found to be sensitive to irinotecan (Fig.
2) In addition to our evaluation of standard-of-care agents, our high-throughput drug screening using matched cell lines allowed us to discover several pathways of interest, including FGFR, PDGFR, and VEGFR, all of which may contribute to CRC growth. Ponatinib, a multi-kinase inhibitor of these pathways, significantly inhibited cell growth in vitro and PDX tumor growth in vivo in our CRC models. Ponatinib was initially designed to inhibit BCR-ABL [
15], and provided to the patients who were resistant to dasatinib or nilotinib treatment. In addition to its effective inhibition of both wild type and several mutant forms of BCR-ABL kinases [
24], further studies have demonstrated ponatinib’s ability to target several other tyrosine kinases [
25]. However, our screening data suggested that there was no antitumor activity with the single kinase inhibitors of the ponatinib’s other targets. This can be explained by 1) synergistic effect from co-targeting these receptors as Lee et al. previously reported effective tumor inhibition in in vivo colon cancer models with CHIR-258, which is EGFR, FGFR, PDGFR and VEGFR inhibitor [
26], 2) overlapping downstream pathways of these receptors, that might allow cancer cells to develop resistance mechanisms using alternative receptor tyrosine kinases. Ellis et al. points out this resistance mechanisms and specifically compensatory activation of FGFR pathway after VEGFR inhibition [
27]. Consistent with these studies, we also demonstrated that co-targeting these receptors can effectively inhibited tumorigenesis. Interestingly, we observed downstream activation of the RAS/RAF/MEK/ERK pathway in CRC240 and CRC057 and the PI3K/AKT/mTOR pathway in CRC119 after ponatinib treatment, which may indicate these pathways as potential resistance mechanisms.
Deregulation of the FGFR signaling pathway plays an important role in carcinogenesis [
28]. Genomic alterations in the FGFR genes that enhance FGFR signaling are mediated by either receptor amplification, mutations or chromosomal translocation [
29]. Specifically, FGFR amplification has been found in lung and breast cancer, and response to FGFR inhibition has been found in amplified FGFR tumors [
30‐
32]. In addition to somatic activating mutations, germline single-nucleotide polymorphisms (SNPs) in FGFR have been found to activate the FGFR pathway [
33]. Finally, activating gene fusions of FGFR have been discovered in a number of different cancers [
34,
35]. In colorectal cancer, genomic alterations in FGFR such as gene amplifications [
36] are not as common as fusion in FGFR3 [
37] or gene copy number gain in FGFR1 [
38]. Along these lines, we used RNA-Seq data to potential chromosomal translocations or mutations to identify predictors of response. While no fusions were found (Supplementary Figure
1), mutations were found in FGFR1, 2 and 4, with the most common mutation being P136L in FGFR4 in all six samples. Although these are potentially interesting findings, we realize that this is a limited analysis and further studies will need to be performed to validate these findings, but the incoproation of genomic profiling to complement functional studies remain a critical component of any precision medicine pipeline.
Simarily, we do realize the limitations of our current precisioin medicine pipeline. As this was our intial proof of concept developoment of our pipeline, our drug screen contained only 119 drugs and the majority of the targeted agents in our screen targeted multiple pathways suggesting that in our screen, combinatorial therapy may be critical to find the optimal therapy in CRC. In addition, our screen which used cell lines limits our in vitro studies that involves anticancer compounds that target the microenvironment. Therefore, this limitation may cause underpredicted in vitro cell line response to these compounds, as in CRC240, which was found to be moderately sensitive to ponatinib in vitro, but quite sensitive in vivo. Despite these challenges, in vitro cell line models have still been widely used for initial pharmacogenomic studies as they allow for simple and low cost biological research but future work will determine if using patient derived organoids which overcome the challenges of cell lines can be used in our precision medicine pipeline.
The development of precision medicine strategies for cancer faces numerous challenges, including accessing patient samples, establishing reliable models for testing, and the genetic and non-genetic diversity inherent within the ever-evolving cancer. Here we propose a pipeline and workflow to address several of these challenges. By establishing patient-matched cell lines and PDXs we are able to leverage the speed and flexibility of in vitro systems while simultaneously providing a robust system for in vivo validations that takes into account, at least in part, tumor heterogeneity and contributions of the tumor microenvironment. Indeed, we have previously shown that PDXs faithfully recapitulate patient tumor histology and preserve tumor-associated stroma and coupled with high-throughput screens and genomics, this pipeline represents a useful paradigm to identify and validate new treatments for CRC that can be expanded to other solid tumors.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.