Karyotypic divergence confounds cellular phenotypes in large pharmacogenomic studies

Rene Quevedo, Nehmé El-Hachem, Petr Smirnov, Zhaleh Safikhani, Trevor J. Pugh, Benjamin Haibe‐KainsbioRxiv 2019 · DOI

Abstract

Somatic copy-number alterations (SCNAs) affecting large genomic regions are a primary source of genomic diversity in cancer and significantly impact cellular phenotypes. Clonal heterogeneity within cancer cell lines can influence phenotypic presentation, including drug response. In this study, we aggregated and analyzed SNP and copy-number profiles from six pharmacogenomic datasets encompassing 1,691 cell lines screened for 13 molecules. To identify sources of genotype and karyotype discordances, we compared SNP genotypes and segmental copy-ratios across 5 kb genomic bins. We further assessed gene expression and drug sensitivity data to compare discordant and concordant lines. Our results revealed that while outright misidentification is rare (0.4%), karyotypic discordance is more prevalent (7.0%). We highlight specific cell lines, such as REH, NCI-H23, and PSN1, where drug response discordances correlate with divergent copy-number profiles. These findings emphasize the necessity of verifying cell line copy-number profiles to accurately interpret drug sensitivity data in biomedical research.

Introduction

Cancer cell lines serve as the primary models for understanding tumor biology and predicting therapeutic responses. However, the utility of these models has been challenged by a "reproducibility crisis" in large-scale pharmacogenomic databases, such as the Cancer Cell Line Encyclopedia (CCLE) and the Genomics of Drug Sensitivity in Cancer (GDSC). While researchers have historically attributed discrepancies in drug sensitivity to cell line misidentification or contamination, this study investigates a more subtle and pervasive factor: genomic instability leading to karyotypic drift.

Cancer cells are inherently unstable, accumulating mutations and structural changes over time as they are cultured across different laboratories. This study explores how changes in the number of copies of genomic regions—karyotypic divergence—alter the cellular phenotype and, consequently, the pharmacological profile of these models.

Methods

The researchers conducted a comprehensive meta-analysis of genomic data across several high-profile datasets, including GDSC, CCLE, gCSI (Genentech Cell Line Screening Initiative), and others. The methodology focused on three main pillars:

  1. Genomic Comparison: The team used the Jaccard index to compare genotypes (SNP-based "DNA fingerprints") and a variance-weighted Pearson correlation to compare karyotypes (chromosomal structure) across 5kb genomic bins.
  2. Phenotype Mapping: Genomic differences were cross-referenced with transcriptomic data (gene expression) and pharmacological data, specifically the area above the drug-response curve (AAC).
  3. Predictive Modeling: The researchers employed LOBICO (Logic Optimization for Binary Input to Continuous Output) to determine if specific copy-number changes could mechanistically explain observed differences in drug sensitivity.

Results

The Spectrum of Identity

The study found that cell line identity is not binary but exists on a spectrum. While current authentication standards like STR profiling have successfully reduced outright misidentification—found in only 0.4% of the 1,378 cell lines profiled in multiple studies—karyotypic discordance is significantly more common, affecting 7.0% of the samples. This suggests that a cell line can pass a standard identity test while remaining functionally distinct due to chromosomal drift.

Predictable Phenotypic Shifts

Karyotypic drift is not merely stochastic noise; it results in predictable biological consequences. The researchers demonstrated that specific chromosomal losses or gains correlate with shifts in drug response. For instance, in the NCI-H23 cell line, the loss of a specific region on chromosome 4q was directly linked to increased resistance to CDK4/6 inhibitors. Similarly, the cell lines REH and PSN1 were identified as exemplars where drug response discordances between CCLE and GDSC could be traced back to divergent copy-number profiles.

Transcriptome vs. Drug Response

An interesting finding was that gene expression profiles tended to remain more stable than drug responses across drifted lines. This suggests that while a cell line might appear transcriptomically similar to its original state under baseline conditions, its underlying genomic changes become manifest primarily when the cell is subjected to pharmacological stress.

Discussion

The findings highlight that a one-time authentication of a cell line is insufficient for long-term research reliability. The proclivity of cancer cell lines to acquire somatic copy-number variants results in biological effects that can confound the interpretation of large-scale pharmacogenomic studies.

Limitations

There are inherent challenges in detecting these changes, as current methods are more adept at identifying large-scale chromosomal shifts than focal copy-number variants. Furthermore, the high level of inherent noise in drug sensitivity assays can sometimes make it difficult to distinguish between experimental error and true biological drift.

Conclusion

This study underscores the need for continuous genomic monitoring in biomedical research. To ensure the reproducibility of science, researchers must move beyond simple identity authentication and regularly verify the copy-number profiles of their cell lines. Accounting for the inevitable "evolution" that occurs in vitro is essential for the accurate interpretation of drug sensitivity data and the successful translation of preclinical findings to the clinic.

Summary updated

Abstract

ABSTRACT Background Somatic copy-number alterations that affect large genomic regions are a major source of genomic diversity in cancer and can impact cellular phenotypes. Clonal heterogeneity within cancer cell lines can affect phenotypic presentation, including drug response. Methods We aggregated and analyzed SNP and copy number profiles from six pharmacogenomic datasets encompassing 1,691 cell lines screened for 13 molecules. To look for sources of genotype and karyotype discordances, we compared SNP genotypes and segmental copy-ratios across 5 kb genomic bins. To assess the impact of genomic discordances on pharmacogenomic studies, we assessed gene expression and drug sensitivity data for compared discordant and concordant lines. Results We found 6/1,378 (0.4%) cell lines profiled in two studies to be discordant in both genotypic and karyotypic identity, 51 (3.7%) discordant in genotype, 97 (7.0%) discordant in karyotype, and 125 (9.1%) potential misidentifications. We highlight cell lines REH, NCI-H23 and PSN1 as having drug response discordances that may hinge on divergent copy-number q Conclusions Our study highlights the low level of misidentification as evidence of effective cell line authentication standards in recent pharmacogenomic studies. However, the proclivity of cell lines to acquire somatic copy-number variants can alter the cellular phenotype, resulting in a biological and predictable effects on drug sensitivity. These findings highlight the need for verification of cell line copy number profiles to inform interpretation of drug sensitivity data in biomedical studies.
BibTeX
@misc{quevedo2019karyotypic,
  author = {Rene Quevedo and Nehmé El-Hachem and Petr Smirnov and Zhaleh Safikhani and Trevor J. Pugh and Benjamin Haibe‐Kains},
  title = {{Karyotypic divergence confounds cellular phenotypes in large pharmacogenomic studies}},
  year = {2019},
  doi = {10.1101/574350},
  howpublished = {bioRxiv},
}
Figure