Alvarez et al. reply — Quantitative, objective metrics to assess the fidelity of cancer models— such as cell lines, organoids or patient-derived xenografts—remain elusive, and histological criteria or the presence of specific mutations is often used as a surrogate. Hofving et al.1 have reported that two cell lines (H-STS and KRJ-1) recently used as a model to prioritize treatment for gastro-enteropancreatic neuroendocrine tumors (GEP-NETs)2 are not appropriate models for this disease because, as also previously reported3, they are derived from Epstein–Barr virus–immortalized lymphoblastoid cells. Yet, what constitutes an appropriate model to study a specific tumor remains a matter of considerable debate. Indeed, histology and mutational status may negatively bias model selection. For instance, we have recently assessed the suitability of 16 neuroblastoma cell lines bearing MYCN amplifications as high-fidelity models for the study of the MYCN-dependent subtype of this tumor4. Unexpectedly, only four of these cell ines effectively recapitulated key tumor dependencies and associated mechanisms that have been shown to be highly conserved across virtually all MYCN-subtype patients in two independent cohorts.
The results of that study motivated us to seek more quantitative, molecular-level criteria to assess the fidelity of a given model to a tumor of interest, by integrating two independent measurements: (i) the conservation of regulatory networks inferred exclusively from patient-derived samples in the tumor-model and (ii) the overlap of patient-specific master-regulator proteins—that is, proteins representing mechanistic regulatory determinants of tumor state and tumor dependencies5—with model-specific master regulators. Master-regulator proteins are identified on the basis of the enrichment of their positively regulated and repressed targets in a phenotype-specific signature, by using the VIPER algorithm, which has been extensively validated6. Such unbiased criteria may be especially valuable for tumors that, like GEP-NETs, lack effective in vitro or in vivo models.
In the GEP-NET study2, we used these criteria to evaluate the suitability of 923 cell lines, including those in the Cancer Cell Line Encyclopedia7, as well as H-STS and KRJ-1, which were previously reported as small-bowel NET patient-derived cell lines8. For VIPER analysis of patient-derived tumors, we used a metastatic progression signature based on the genes differentially expressed between each hepatic metastasis and a set of cluster-matched primary tumors2. For cell lines, differential expression was computed against P-STS cells, representing a bona fide primary small-bowel GEP-NET model, as confirmed by mutational analysis9. H-STS and KRJ-1 emerged as the fourth and sixth best available models (Supplementary Fig. 4 in ref.2), respectively, and were further selected because of their established ability to grow as xenografts and their GEP-NET-patient derived nature.
We have now reported several additional data and analyses that address the criticisms raised by Hofving et al.1, by validating the use of H-STS and KRJ-1 as appropriate models for GEP-NET-related drug mechanism of action (MoA) analysis10. Below, we summarize these results in response to the Correspondence from Hofving et al.
First, we showed that, with the same criteria for cell-line prioritization, other normal and transformed B cells emerged as very poor matches for GEP-NETs. The median average rank was 591, and the top-ranked cell line—ST486, an Epstein–Barr virus–negative Burkitt’s lymphoma—was in the 22nd position. Thus, H-STS and KRJ-1 are unusual B cell–derived lines in terms of recapitulating the regulatory network and master-regulator activity of GEP-NET patient-derived samples.
In addition, we showed that drug screens in primary cells derived from a patient with rectal NET (P0NETCL) and in GEP-NET patient-derived explants confirmed the H-STS- and KRJ-1-based drug MoA and drug-sensitivity predictions10. Of the 126 drugs screened in H-STS cells, 95 were also profiled in P0NETCL cells, thus supporting direct comparison of MoA inference, on the basis of the overlap of proteins whose activity was significantly affected by the drug10. The MoA conservation was highly significant, and 60 of the 95 assessed drugs (63%) showed highly significant MoA similarity in P0NETCL and H-STS (P <10−10, Bonferroni corrected, one-tailed aREA test6,10) and suggesting that the latter are a useful model for GEP-NET-related drug studies (Fig. 2a in ref.10). Indeed, MoA conservation was at least as good as positive controls represented by three tumor-matched cell-line pairs for which compound perturbational profiles were also available—including U87 and HF2597 (glioblastoma multiforme), AsPC-1 and PANC-1 (pancreatic ductal adenocarcinoma) and LNCaP and DU-145 (prostate adenocarcinoma). In sharp contrast, negative controls represented by the same six cell lines—individually identified as poor GEP-NET models, on the basis of our criteria—showed poor MoA conservation with P0NETCL cells (Fig. 2 of ref.10). The MoA similarity for entinostat in P0NETCL and H-STS was highly significant (P <10−11, Bonferroni corrected, one-tailed aREA test6,10; red dot in Fig. 2a of ref.10).
We then evaluated the reproducibility of OncoTreat-predicted drug sensitivity(P <10−5, Bonferroni corrected, one-tailed aREA test2, 6) in 69 GEP-NET hepatic metastases, by using perturbational profiles from either H-STS or P0NETCL cells (Fig. 2b of ref.10). OncoTreat drug-sensitivity prediction overlap was significant for P0NETCL and H-STS cells (P <10−130, one-tailed Fisher’s exact test; odds ratio = 6.7), and significantly stronger than that for P0NETCL cells and six low-scoring cell lines, which were considered negative controls10. These results further confirm that H-STS cells are a high-fidelity model for GEP-NET drug-activity inference, on the basis of both MoA conservation and OncoTreat sensitivity.
We additionally obtained perturbational profiles from drug-treated GEP-NET hepatic-metastasis-derived explants. Because explants needed to be treated within 24 h, OncoTreat-predicted drugs were pre-prioritized on the basis of the tumor primary site, including pancreatic, small-bowel and rectal NETs. Each explant was treated with as many drugs as allowed by tissue availability (Supplementary Table 1 of ref.10). Overlap of H-STS-based OncoTreat predictions was compared with master-regulator activity reversal measured in each explant, at a conservative statistical-significance threshold (P ≤ 10−5, Bonferroni corrected, one-tailed aREA test2,6). Of 40 tested drugs, 13 were correctly predicted to reverse master-regulator activity (true positives), 14 were correctly predicted to have no master-regulator-reversal activity (true negatives), 5 were incorrectly predicted to have activity (false positives), and 8 were incorrectly predicted to have no activity (false negatives) (68% prediction accuracy, P <0.05, one-tailed Fisher’s exact test) (Fig. 4 in ref.10). These findings indicate that—despite their histological differences, and contrary to the conclusions drawnby Hofving et al.—H-STS cells are an effective model to predict drug-mediated master-regulator activity reversal in bona fide patient-derived GEP-NET metastases. Interestingly, the rectal-NET explant predicted to not be entinostat sensitive was correctly predicted to be belinostat sensitive, whereas a third histone deacetylase inhibitor, vorinostat, was correctly predicted to be inactive10. These results suggest that H-STS-based OncoTreat analysis effectively discriminates between drugs with closely related MoA. In agreement with these results, an independent study in bona fide NET cells has reported higher sensitivity to histone deacetylase inhibitors than that in control cells3.
Together, these data suggest that although H-STS and KRJ-1 cells may not represent histology-matched GEP-NET models, they effectively recapitulate GEP-NET-specific drug MoA, as assessed in bona fide GEPNET-derived cells as well as critical GEP-NET regulatory and tumor-dependency features. As a result, objective molecular criteria based on mechanism conservation, as proposed by Alvarez et al.2, may provide a valuable metric for model selection, even when histology-matched models may not be available.
Acknowledgements
The results in this manuscript were produced by using generous support from the NET Research Foundation to A.C., D.L.R.-L. and M.H.K. for the GEP-NET explant component of the study. We would also like to acknowledge the National Institutes of Health 1R35CA197745 (Outstanding Investigator Award) and U01CA217858 (Cancer Target Discovery and Development network) to A.C., and the NCI instrumentation grants S10OD012351 and S10OD021764 to A.C., which were instrumental for the data analysis. All RNA-seq libraries and sequencing were performed in the JP Sulzberger Columbia Genome Center, with support from a P30 Cancer Center Support Grant (P30CA013696).
Footnotes
Competing interests
M.J.A. is Chief Scientific Officer and an equity holder at DarwinHealth, Inc., a company that has licensed some of the algorithms used in this manuscript from Columbia University. A.C. is a founder, equity holder, consultant and director of DarwinHealth, Inc. Columbia University is also an equity holder in DarwinHealth, Inc.
References
- 1.Hofving T, Karlsson J, Nilsson O & Nilsson JA Nat Genet. 10.1038/s41588-019-0490-z (2019). [DOI] [PubMed] [Google Scholar]
- 2.Alvarez MJ et al. Nat. Genet 50, 979–989 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Hofving T et al. Endocr. Relat. Cancer 25, X1–X2 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Rajbhandari P et al. Cancer Discov. 8, 582–599 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Califano A & Alvarez MJ Nat. Rev. Cancer 17, 116–130 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Alvarez MJ et al. Nat. Genet 48, 838–847 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Barretina J et al. Nature 483, 603–607 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Pfragner R et al. Anticancer Res. 29, 1951–1961 (2009). [PubMed] [Google Scholar]
- 9.Rinner B et al. BMC Genomics 13, 594 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Alvarez MJ et al. Preprint at bioRxiv 10.1101/677435 (2019). [DOI] [Google Scholar]