Pan-Cancer Analysis Reveals Technical Artifacts in TCGA Germline Variant Calls

Abstract

The degree to which germline variation drives cancer development and shapes tumor phenotypes remains largely unexplored, possibly due to a lack of large scale publicly available germline data for a cancer cohort. Here we called germline variants on 9,618 cases from The Cancer Genome Atlas (TCGA) database representing 31 cancer types. We identified batch effects affecting loss of function (LOF) variant calls that can be traced back to differences in the way the sequence data were generated both within and across cancer types. Overall, LOF indel calls were more sensitive to technical artifacts than LOF Single Nucleotide Variant (SNV) calls. In particular, whole genome amplification of DNA prior to sequencing led to an artificially increased burden of LOF indel calls, which confounded association analyses relating germline variants to tumor type despite stringent indel filtering strategies. Due to the inherent noise we chose to remove all 614 amplified DNA samples, including all acute myeloid leukemia and virtually all ovarian cancer samples, from the final dataset. This study demonstrates how insufficient quality control can lead to false positive germlinetumor type associations and draws attention to the need to be sensitive to problems associated with a lack of uniformity in data generation in TCGA data. There are substantial differences in the way exome sequence data was generated both across and within cancer types in TCGA. We observe that differences in sequence data generation introduced batch effects, or variation that is due to technical factors not true biological variation, in our variant data. Most notably, we observe that amplification of DNA prior to sequencing resulted in an excess of predicted damaging indel variants. We show how these batch effects can confound germline association analyses if not properly addressed. Our study highlights the difficulties of working with large public genomic datasets like TCGA where samples are collected over time and across data centers, and particularly cautions the use of amplified DNA samples for genetic association analyses.

Identifying and mitigating batch effects in whole genome sequencing data

“…It is likely that PCR-free versus PCR library preparation and sequencing center played a key role in creating this detectable batch effect, similar to [36], as we found clear separation in PCA visualizations of quality metrics by these variables (Additional file 1: Figure S1). We found the two groups were best explained using year of sequencing so designated samples sequenced in years 2010, 2011, and 2012 as group 1 ( N  = 918 samples) and samples sequenced in years 2013 and 2014 as group 2 ( N  = 313 samples).…”

Challenges and Opportunities in Studying the Epidemiology of Ovarian Cancer Subtypes

“…Hofree et al defined genetic subtypes of HGSC by performing network-based clustering on germ-line and somatic variant data from TCGA, but these subtypes are not concordant with the gene expression subtypes [75]. A recent analysis of genotyping accuracy has raised questions about the quality of sequencing-based variant calls in TCGA’s HGSC samples [76] which may affect the findings in Hofree et al [75]. Given the exploratory nature of molecular clustering and limitations of the approaches used, more research is needed about how many underlying molecular subtypes exist [77–79] and if they are consistent across populations [80].…”

Epithelial ovarian cancer risk: A review of the current genetic landscape

“…10,39,40 However, TCGA did not find any germline variants in likely significant genes RAD51C or RAD51D, and have been criticized for inaccurate results due to technical artefacts, particularly affecting the ovarian cancer cases. 41 Homologous recombination deficient (HRD) ovarian cancers have greater sensitivity to DNA-damaging agents that crosslink DNA such as cisplatin as HR is required for the repair of these lesions, and improved OS. 36,42,43 Being able to identify women with HRD cancers has clear clinical implications in terms of chemotherapy regime planning and development and use of targeted therapies.…”

“…The Cancer Genome Atlas (TCGA) found HR to be defective in approximately half of 489 women with stage II to IV HGSOCs, attributed to germline variants in BRCA1 (in 9% of tumours) or BRCA2 (8%), somatic variants in BRCA1 or BRCA2 (3%), epigenetic silencing of BRCA1 (11%), amplification of EMSY (8%), PTEN deletion/mutation (7%), hypermethylation of RAD51C (3%), ATM or ATR pathogenic variants (2%) and variants of other HR genes (5%) . However, TCGA did not find any germline variants in likely significant genes RAD51C or RAD51D , and have been criticized for inaccurate results due to technical artefacts, particularly affecting the ovarian cancer cases . Homologous recombination deficient (HRD) ovarian cancers have greater sensitivity to DNA‐damaging agents that crosslink DNA such as cisplatin as HR is required for the repair of these lesions, and improved OS .…”