The Challenges & Considerations of Reporting Somatic Variants in Cancer
Posted 30th April 2018 by Jane Williams
Industrial and “omic” scale mutational profiling of solid tumours are possible as a result of massively parallel sequencing technology. However, these clinical genomic testing capabilities have also brought myriad new biological, technical and operational challenges not encountered in monogenic disorder testing. All of this impacts the interpretation and reporting of somatic variants. As a testament to the nascency of this field, the first published consensus guidelines for interpretation of variants in cancer was only released in early 2017. While these guidelines are a good start towards standardization of interpretation and classification methods, much work is still required by the testing lab to establish reliable and robust interpretation protocols capable of producing consistent, meaningful test result reports. There is also more work for inter-laboratory consistency.
Cancers specimens used for genomic testing are a mixture of cells (tumour heterogeneity), consisting of mosaic genomes derived from healthy cells in addition to clonal populations of an evolving tumour. Instead of discrete mutational states of 0, .5, 1 observed in germline testing, this mixture of genomes produce mutational fractions distributed as a continuous interval between 0 and 1.
The consequences of this are:
- Intra-sample variability influencing variant detection limits and a minimum depth of coverage
- The need to differentiate between germline and somatic alterations
- The inability to know the composition of subclonal populations of cells
- Additional considerations that impact reliability, accuracy and quality of the test results
At the molecular level, tumourigenesis is a multi-step process involving dysregulation of oncogenes (turned on) and tumour suppressors (turned off), making genetic heterogeneity the status quo in cancer sequencing. Advanced tumours are a polygenic condition, therefore, labs must have the capacity to interpret and report numerous variants per patient, rather than 1-2 mutations driving pathogenesis of inherited monogenic disorders.
Furthermore, because the primary purpose of testing cancer patients is to assess variants as biomarkers for potential targeted therapeutic avenues rather than diagnostic assignments, additional comprehensive layers of interpretation to evaluate gene-drug response and potential resistance mechanisms must be in place.
Many cancer genes are complicit in numerous tumour/histologic-types and this phenotypic heterogeneity necessitates repeating the process to classify variants in the context of specific tumours. For example, KRAS hotspot mutations are classified Tier I when detected in colorectal cancers due to clinical practice guidelines recommending against the use of anti-EGFR monoclonal antibody therapies such as cetuximab. However, the same mutation has been classified as Tier II when detected in many other cancers.
Databases (DB) and knowledgebases (KB) remain an excellent resource for interpretation of somatic variants in the omics-era, but with so many such resources available today, vetting them for viability within the clinical workflow (strengths, limitations, variable methods, etc) can be an overwhelming task. Databases provide relatively objective, often quantitative information that is highly structured and easily integrated into pipelines. Knowledgebases provide richer information, but are less structured, subjective and employ some form of interpretive curation. Although these provide a degree of utility to the clinical laboratory, none provide the standalone reliability to serve as filtering parameters.
While individual knowledgebases may not stand up to clinical lab scrutiny, an aggregate of KBs would deliver much greater coverage, however, both the subjective and variable methodologies between sources pose challenges for collectively integrating them into a computational pipeline for aggregate use. Lack of standardization across the broader biomedical landscape also impacts laboratory efficiency and burdens resources. For example, clinicaltrials.gov has minimally structured data and no standards for genomic criteria. The result is enormous trial-to-trial variability that impedes computational mining, limiting most protocols to brute force manual curation methods.
Additionally, the massive volume of biomedical studies on cancers, cancer genes and cancer therapies constitute a monumental effort for literature review process. This is compounded by the polygenic nature of tumours (requiring review of 3-8 variants per patient on average) and the lack of a comprehensive gene/mutation-centric literature resource (Griffith et al. Nature Genetics, 2017. PMID: 28138153).
Turnaround time on the oncology side of the industry is typically in the 2-4 week range, which requires significant investment in automation and software infrastructure to meet such intense consumer demands. To put this in perspective, the Mendelian genetics testing industry has TATs in the 2-4 month range for a large portion of tests and interpret/report on monogenic disorders almost exclusively at the level of pathogenicity. However, in cancer, one needs to interpret/report on average 5 variants per report by:
- Determining their molecular function/pathogenicity
- Assessing actionability by evaluating the mutations as a therapeutic marker of FDA-approved drugs
- Review the clinical practice guideline recommendations
- Investigate drugs and clinical trials, as well potential resistance mechanisms reciprocally between reported variants.
In some cases, these variants are also reported out with additional prognostic and/or diagnostic evidence and/or classified by their prognostic or diagnostic significance.
The cumulative effort involved in ensuring quality results in this dynamic and complex testing environment demand integration of scalable software infrastructure into the analysis and reporting workflow. This can be a huge barrier for smaller labs and start-ups getting involved in tumour testing. There is little way around this considering a typical 100-500 genes panel will readily produce 5 or more potentially targetable, disease-causing mutations that are assessed in the context of molecular function/disorder before being databased for later use.
Andrew Hesse, Manager of Clinical Data Analytics & Reporting at the Jackson Laboratory, will be speaking at the Precision Medicine & Biomarkers Leaders Summit on liquid biopsies in the clinic and the validation of a pan cancer panel.
Take a look at the agenda for the Precision Medicine & Biomarkers Leaders Summit: Europe, as well as the full list of speakers and presentation titles.
Leave a Reply