Evaluating Patient-Derived Colorectal Cancer Xenografts as Preclinical Models by Comparison with Patient Clinical Data

Colorectal cancer remains a major cause of mortality worldwide and colorectal cancer patient death is generally attributable to metastasis development. Comprehensive molecular characterization of colorectal cancer has identified key gene and pathway alterations important for initiation and progression of colorectal cancer, including alterations in the PI3K and ERBB–RAS pathways (1, 2). Some genetic anomalies have been also shown to predict response to specific therapies, such as activating mutations in KRAS, which predict resistance to anti-EGFR monoclonal antibodies (MAb; ref. 3). For efficient development of new therapies and companion biomarkers, preclinical models mimicking the molecular epidemiology and drug sensitivity of human tumors are needed.

In colorectal cancer, tumor-specific patient-derived xenograft (PDX) models have shown to retain the intratumoral clonal heterogeneity, chromosomal instability, and histology of the parent tumor through passages in mice (4–7). To extend these observations, we investigated here a collection of 52 colorectal PDXs (6), composed of 48 microsatellite stability (MSS) and four microsatellite instability (MSI) tumors, for the presence and prevalence of molecular features reported in large colorectal cancer patient cohorts (1, 2, 8). In particular, we studied key alterations in IGF2–PI3K and ERBB–RAS pathways and the role of these alterations in predicting response to cetuximab.

Tumor xenografts were established directly from patient tumors (6) and were routinely passaged by subcutaneous engraftment in immunodeficient CB17-SCID mice (Charles River Laboratories). Xenografts, passage P6-P9, were harvested from 3 mice for each model, when they reached around 150 to 300 mm³ in size for RNA and DNA extraction. For in vivo pharmacological studies, cetuximab (Imclone) was given at 12.5 mg/kg/adm, (Q3D×2) ×2 i.p.), mice bearing 100 to 200 mm³ tumors at start of therapy (n = 8–10 per group) as already described (6). All experimental procedures were approved by Sanofi Laboratory Animal Care and Use committee.

MSI testing was performed according to the National Cancer Institute guidelines using a five-microsatellite consensus panel (6).

Next-generation sequencing and mutation calling were performed at Beijing Genomics Institute (BGI). Library preparation was performed using exome capture Agilent SureSelect All Exon 50M. Libraries were sequenced using the Illumina HiSeq platform. Quality single-nucleotide polymorphism (SNP) calling criteria have been applied: SNP quality is equal or greater than 20; the minimum sequencing depth is 4× and the mean is 100×, with 99% of coverage target region. To evaluate and eliminate the false-positive SNPs calls generated by cross hybridization with mouse DNA, we have detected and filtered out reads aligned to mouse reference sequences before doing human whole-exome sequencing analysis; by this way, only specific human calling are considerate. Besides, KRAS, BRAF, and PIK3CA mutations were validated by Sanger method in a different tumor sample.

Evaluation of genome-wide, gene copy number was performed using the 250k and 400k oligonucleotide CGH array Agilent technology using two biologic duplicates and two independent experiments. Oligonucleotide array CGH processing was performed as detailed in the manufacturer's protocol (version 6.2 October 2009; http://www.agilent.com). The log₂ ratio and segmentation were generated using Array Studio software. Array Studio, Array Viewer, Array Server, and all other Omicsoft products or service names are registered trademarks or trademarks of Omicsoft Corporation.

The analysis of gene expression was done using U133 Plus Affymetrix microarrays with biologic triplicate (three tumor tissues removed from three distinct mice for each model, passage P6-P9).

Affymetrix data of candidate genes were confirmed by qRT-PCR using previously described methodology (9).

PTEN and INPP4B expression were determined on 4-μm-thick AFA-fixed paraffin-embedded sections. Antigen retrieval was done by incubating tissue sections in an 850-Watt microwave oven for 36 minutes in Tris–EDTA or in citrate buffer for INPP4B and PTEN staining, respectively. Tissue sections were then incubated for 1 hour at room temperature with primary antibodies (anti-INPP4B, clone EPR3108Y, dilution 1:50, rabbit mAb, LSBio; anti-PTEN, clone SP218, dilution 1:50, rabbit mAb, Spring). Staining was revealed by using OmniMap HRP anti-Rabbit (Ventana Medical Systems) and diaminobenzidine (Dako) as chromogen.

Comprehensive molecular characterization of tumor samples from colorectal cancer patients has identified a handful of recurrent mutated genes within critical pathways (1, 2, 10). Among these, the PI3K and ERBB–RAS signaling, accurately dissected by the Cancer Genome Atlas Network (TCGA; ref. 2), provide promising therapeutic targets.

To gain more insight into the genomic abnormalities within the PI3K and ERBB–RAS signaling pathways, a large cohort of 52 colorectal PDXs established by the CReMEC consortium (48 MSS and four MSI tumors; ref. 6) was analyzed.

We first examined six genes identified as key upstream elements in the PI3K pathway (2): IGF2, IRS2, PIK3CA, PIK3R1, PTEN, and INPP4B (Fig. 1A, Table 1). Several lines of evidence underline the importance of IGF2 in colorectal cancer. IGF2 is the single most overexpressed gene in colorectal neoplasia relative to normal colorectal mucosa (11) and loss of imprinting of IGF2, one mechanism for its frequent overexpression, is also a risk factor for colorectal cancer (12). More recently, TCGA revealed IGF2 as an important node in the PI3K pathway with mutual exclusion between IGF2, IRS2, PIK3CA, PIK3R1, and PTEN genomic alterations. Here, gene expression analyses identified IGF2 overexpression in seven PDXs. As reported in patients (1, 2), IGF2 overexpression in PDXs (5 out of 7) is mainly due to focal IGF2 amplification (Fig. 1A and B).

The binding of IGF2 to IGF1R activates the intrinsic tyrosine kinase activity of IGF1R, which results in the phosphorylation of the insulin receptor substrates (IRS), leading to PI3K activation. Gene expression analysis of IRS1 and IRS2 revealed no alterations in IRS1. However, overexpression of IRS2 (n = 5) was detected in mutually exclusive pattern with IGF2 amplification or overexpression (Fig. 1A). All PI3KCA aberrations (n = 12) were oncogenic mutations, affecting all functional domains of the enzyme but with preferential mutation hotspots within exons 9 and 20, as previously described in colorectal cancer (13). PIK3R1 mutations have been rarely reported in colorectal cancer (2) and none were detected in the present PDX collection.

Two PDXs showed PTEN homozygous deletion associated with loss of protein expression, whereas no PTEN mutation was detected (Fig. 1C). Recently, another lipid phosphatase, inositol polyphosphate 4-phosphatase type II (INPP4B), has emerged as a potential tumor suppressor in prostate, breast, and ovarian cancers (14). Downregulation of INPP4B gene expression was detected here in two PDXs, with concomitant loss of protein expression. Immunohistochemical analyses confirmed mutual exclusion between PTEN and INPP4B downexpression (Fig. 1C).

Interestingly, a pattern of mutual exclusion in the PI3K pathway also exists between IGF2, IRS2, PIK3CA, PTEN, and INPP4B alterations. These data imply that therapeutic targeting of the IGF2 pathway could inhibit PI3K activity and suggest INPP4B as a tumor suppressor gene in colorectal cancer.

Mutations or gene amplification of candidate genes in the ERBB–RAS pathway was then analyzed. EGFR displayed no mutations but gene amplification associated with gene overexpression in two MSS PDXs. No mutation was identified in ERBB2, but one PDX showed ERBB2 amplification, accompanied by overexpression. Two PDXs displayed a T389I ERBB3 mutation, probably damaging (PolyPhen prediction software). No gene alteration was present in ERBB4.

We found that 69% (33 of 48) of MSS tumors and 100% (4 of 4) of MSI tumors have oncogenic alterations in KRAS, NRAS, or BRAF with a significant pattern of mutual exclusion (Fig. 1A). In accordance with published data, KRAS missense mutations in codons 12, 13, and 61 were the most frequent KRAS mutations (observed mutated in 18, 5 and two PDX models, respectively). Two additional PDXs showed two oncogenic KRAS mutations, K117N and A146T, previously reported in colorectal cancer with similar low frequencies (1, 15). Six PDXs displayed NRAS mutations, with all mutations occurring in codon 61. Six PDXs displayed BRAF mutations: four of these were the frequent hot-spot V600E mutation and two were less frequent mutations, D594N and G469A, already reported in colorectal cancer (16). BRAF V600E mutations were associated with MSI, as this mutation was present in 75% (3 of 4) of MSI tumors compared with 2% (1 of 48) of MSS tumors, (P < 0.0001, Yates χ² test). BRAF V600E mutations were mutually exclusive from KRAS and NRAS mutations as usually described (16).

Finally, we observed no significant association of alterations in the RAS and PI3K pathways, suggesting that simultaneous inhibition of the RAS and PI3K pathways might be necessary for successful therapy in the subgroup displaying cooccurrence of these molecular alterations.

These genomic analyses enable an assessment of the diversity and the frequency of genomic changes altering these two major signaling pathways in our colorectal cancer PDX models and comparison with TCGA data (Fig. 2). TCGA has reported that 77% (23 of 30) of hypermutated tumors are MSI tumors (2). As the present PDX collection displays a low frequency of MSI tumors (4 of 52, 8%) close to that of patient tumors (23 of 224, 10%; ref. 2), we focused on MSS PDX and patient tumors. The frequency of studied molecular epidemiology data from these two groups showed remarkable concurrence, suggesting that the PDX bank represents a useful set of preclinical models for testing new therapies and emphasizing the potential therapeutic value of targeting IGF2 in colorectal cancer. In the same way, recent analyses by the Bodmer laboratory have shown that in vitro colorectal cancer cell lines provide useful preclinical tools because of well represented genetic diversity of patient tumors in cell lines (17, 18). It led us to an analysis of gene abnormalities specifically within IGF2–PI3K pathway in a large panel of 62 human colorectal cancer cell lines using the Broad–Novartis Cancer Cell Line Encyclopedia data (http://www.broadinstitute.org/ccle/home). We carefully separated MSS and MSI cell lines because of overpresentation of MSI cell lines, which could interfere with mutation frequencies. IGF2–PI3K pathway alterations appear almost mutually exclusive within the 36 MSS cell lines with some redundancy between PIK3CA activating mutations and IRS2 overexpression (Supplementary Fig. S1). Whereas PIK3CA, PTEN, and PIK3R1 aberration profiles recapitulate the patient tumor observation, the frequencies of IGF2 and IRS2 upregulation in MSS colorectal cancer cell lines are under- and overrepresented, respectively.

In addition to maintaining the genomic and histologic heterogeneity, translationally relevant preclinical models need to reproduce drug response observed in patients. Although KRAS mutations had been identified as a strong predictive biomarker of resistance to cetuximab and panitumumab (3), only a subset of KRAS wild-type (WT) patients respond to anti-EGFR MAbs, underlining that additional predictive biomarkers exist within KRAS WT tumors. The characterization of alterations occurring in additional candidate genes (NRAS, BRAF, PIK3CA, PTEN) increased indeed the negative predictive value up to 70%, but it is not sufficient to identify all resistant cases (19).

To assess drug response prediction in our PDX models, cetuximab response was analyzed in the PDX panel. To be consistent with clinical criteria, we considered responders the PDXs displaying partial or complete response and nonresponders the PDXs displaying growth stabilization or progression (Supplementary Fig. S2). With these scoring criteria, eight out of 52 PDXs (15%) were responders to cetuximab, with complete tumor disappearance in three PDX models. The all eight responders were WT KRAS tumors (Figs. 1A and 3), outlining the requirement of WT KRAS genotype for clinical benefit. This low proportion of PDX responders in an unselected population (15%) is highly concordant with patient data (8) and PDX data from an independent metastatic colorectal cancer xenograft series (7). Noteworthy, cetuximab has been shown to be active against KRAS-mutated xenografts (6), leading to a statistically significant reduced tumor growth but with no tumor shrinkage. This kind of PDX response means nevertheless progressive disease from a clinical point of view. Therefore, assessment parameters have to be carefully analyzed to avoid over- or misinterpretation of drug efficacy.

The majority of nonresponder PDXs (27 of 44) showed canonical activating mutations in KRAS (codons 12, 13, 61, 117, and 146). The Trusolino group reported similar results involving KRAS mutations in codons 61, 117, and 146 in primary resistance to cetuximab in a large and independent colorectal PDX series (7). While most clinical studies limited KRAS mutation assessment to codons 12–13, KRAS codon 61 and 146 mutations, in addition to NRAS and BRAF mutations, have also been shown in retrospective studies to predict resistance to cetuximab or panitumumab in WT KRAS codon 12 and 13 metastatic colorectal cancer (19, 20). The European Medicines Agency recently updates and restricts the indication for cetuximab to WT RAS metastatic colorectal cancer (not only WT KRAS codon 12–13).

Likewise, none of the four BRAF V600E–mutated, four NRAS-mutated, and four KRAS (codon 61, 117 or 146) mutated PDXs responded to cetuximab with tumor shrinkage. Therefore, exclusion of these mutations enables an improved selection of PDXs likely to respond to cetuximab, increasing the response rate from 28% (8 out of 29 12–13 codons KRAS WT) to 53% (8 of 15) in PDXs that are fully wild-type for all three genes, as reported retrospectively in patients (19, 20). This study functionally cross validates a recent clinical stratification based on combination of predictive biomarkers obtained retrospectively in patients (19). These data support the utility of our PDX panel for identifying predictors of drug response in metastatic colorectal cancer patients.

As for PTEN and PIK3CA impact, clinical data are more conflicting (8, 13, 19). In the present preclinical work, only PTEN homozygous deletion, leading to absolute PTEN inactivation, has been taken into account. This PTEN loss occurred within KRAS-mutated xenografts, displaying lack of response to cetuximab. Individual contribution of PIK3CA mutations to the absence of response is difficult to assess because of the PIK3CA mutation diversity in different protein domains and coexistence of these mutations with KRAS and BRAF mutations (13). Moreover, one PDX with IGF2 activation and two other PDXs with PIK3CA mutation respond to cetuximab. Among the five IGF2-overexpressed KRAS WT tumors, only one responds to cetuximab. Taken together, these data suggest that IGF2–PI3K components are not biomarkers of resistance to anti-EGFR therapies and underline the interest to combine anti-IGF2 and anti-EGFR treatment.

Within the group of 15 KRAS/BRAF/NRAS wild-type PDXs, further investigation has been performed for additional putative predictive biomarkers of resistance to anti-EGFR (8, 21): EGFR gene amplification, overexpression and mutation; gene overexpression of two EGFR ligands (epiregulin and amphiregulin), MET gene amplification and overexpression, KRAS gene amplification and HRAS mutation (Supplementary Table S1). Noteworthy, none of the patients with colorectal cancer, from whom the triple wild-type PDXs were derived, had been exposed to anti-EGFR therapy before surgery, ruling out the possibility of acquired resistance in the pretreated PDXs. The analysis of these parameters did not allow to statistically discriminating between the responder and nonresponder groups (Fisher exact test, P > 0.05). Nevertheless, it is noteworthy that cetuximab treatment was ineffective in mice engrafted with the three PDX models carrying KRAS amplification.

Collectively, the present data demonstrate the relevance of colorectal PDXs as models for preclinical drug development. The PDX models remarkably fit the molecular epidemiology and the cetuximab drug response profiles of colorectal cancer patient populations, justifying the growing use of mouse clinical trials in cancer drug development and decision making (5). More importantly, these data support the identification of KRAS (exon 2, 3, and 4)/NRAS/BRAF wild-type patients for treatment with cetuximab, and IGF2 as an attractive novel cancer drug target in a large subset of colorectal cancer patients.

P. Vrignaud is an employee of and is a stock holder of Sanofi. J. Watters is Head, Translational Medicine Oncology, at Sanofi. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Nunes, I. Bièche, V. Dangles-Marie

Development of methodology: M. Nunes, S. Vacher, S. Richon, C. Dib, I. Bièche

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Vrignaud, S. Vacher, S. Richon, W. Cacheux, L.-B. Weiswald, S. Chateau-Joubert, C. Dib, J. Watters, I. Bièche

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Nunes, P. Vrignaud, S. Vacher, A. Lièvre, W. Cacheux, G. Massonnet, S. Chateau-Joubert, W. Zhang, D. Bergstrom, I. Bièche, V. Dangles-Marie

Writing, review, and/or revision of the manuscript: M. Nunes, P. Vrignaud, S. Richon, A. Lièvre, W. Cacheux, L.-B. Weiswald, J. Watters, D. Bergstrom, S. Roman-Roman, I. Bièche, V. Dangles-Marie

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Nicolas, V. Dangles-Marie

Study supervision: M. Nunes, P. Vrignaud, V. Dangles-Marie

We thank CReMEC consortium and especially Ludovic Lacroix, Ludovic Bigot, Fariba Nemati, Cyril Berthet and Olivier Duchamp for PDX tissues and nucleic acid managing and valuable discussions. Additional acknowledgments go to all the additional scientific collaborators from the “Institut Curie Digestive Group” led by Sylvie Robine. We thank Didier Meseure and Jean-Jacques Fontaine for help in histological analyses.

This work was supported by the Comité départemental des Hauts-de-Seine de la Ligue Nationale Contre le Cancer, the Conseil régional d'Ile-de-France, and the Cancéropôle Ile-de-France and the Association pour la recherche en cancérologie de Saint-Cloud (ARCS), Genevieve and Jean-Paul Driot Transformative Research Grant, Philippe and Laurent Bloch Cancer Research Grant, Hassan Hachem Translational Medicine Grant, and Sally Paget-Brown Translational Research Grant.