Prevalence of Clonal Hematopoiesis Mutations in Tumor-Only Clinical Genomic Profiling of Solid Tumors

Key Points

Question  To what extent can clonal hematopoiesis (CH) lead to misclassification of blood-derived somatic mutations as tumor-derived somatic mutations in clinical next-generation sequencing performed in the absence of a matched blood control?

Findings  In this gene-sequencing study including a cohort of 17 469 patients with advanced cancer, paired next-generation sequencing results show that 5% of the patients would have at least 1 CH-associated mutation misattributed as tumor derived in the absence of matched blood sequencing.

Meaning  As a subset of CH-derived alterations involve actionable cancer genes, failure to recognize such mutations as blood derived may lead to erroneous treatment recommendations; sequencing matched blood can be used to distinguish frequent CH somatic mutations from those in the solid-tumor cells and perhaps lead to more accurate precision therapy.

Importance  Although clonal hematopoiesis (CH) is well described in aging healthy populations, few studies have addressed the practical clinical implications of these alterations in solid-tumor sequencing.

Objective  To identify and quantify CH-related mutations in patients with solid tumors using matched tumor-blood sequencing, and to establish the proportion that would be misattributed to the tumor based on tumor-only sequencing (unmatched analysis).

Design, Setting, and Participants  Retrospective analysis of samples from 17 469 patients with solid cancers who underwent prospective clinical sequencing of DNA isolated from tumor tissue and matched peripheral blood using the MSK-IMPACT assay between January 2014 and August 2017.

Main Outcomes and Measures  We identified the presence of CH-related mutations in each patient’s blood leukocytes and quantified the fraction of DNA molecules harboring the mutation in the corresponding matched tumor sample.

Results  The mean age of the 17 469 patients with cancer at sample collection was 59.2 years (range, 0.3-98.9 years); 53.6% were female. We identified 7608 CH-associated mutations in the blood of 4628 (26.5%) patients. A total of 1075 (14.1%) CH-associated mutations were also detectable in the matched tumor above established thresholds for calling somatic mutations. Overall, 912 (5.2%) patients would have had at least 1 CH-associated mutation erroneously called as tumor derived in the absence of matched blood sequencing. A total of 1061 (98.7%) of these mutations were absent from population scale databases of germline polymorphisms and therefore would have been challenging to filter informatically. Annotating variants with OncoKB classified 534 (49.7%) as oncogenic or likely oncogenic.

Conclusions and Relevance  This study demonstrates how CH-derived mutations could lead to erroneous reporting and treatment recommendations when tumor-only sequencing is used.

Clinical next-generation sequencing (NGS) of tumor samples is increasingly used to identify targetable genomic alterations and guide treatment selection in patients with cancer.^1-3 While patient-matched blood control samples are required to distinguish between somatic and inherited variants, practical challenges in the implementation of clinical NGS have led to the frequent use of assays that analyze unmatched tumors (“tumor-only”).^2,4 While this is generally sufficient to identify cancer mutational hot spots,⁴ for more comprehensive assays, the identification and filtering of germline variants can be challenging. Strategies to filter presumptive germline variants include the removal of common variants based on population frequencies and the retention of actionable variants.⁵

Clonal hematopoiesis (CH) is the somatic acquisition of genomic alterations in hematopoietic stem and/or progenitor cells, leading to clonal expansion.⁶ In patients with cancer, CH is a common occurrence, associated with aging, smoking, and radiation therapy.⁷ Clonal hematopoiesis is associated with increased risk of therapy-related hematologic malignant neoplasms, and genes frequently mutated in CH are also commonly altered in hematological malignant neoplasms.^7,8 Recent reports of JAK2 V617F, a common leukemia mutation, in solid tumors^9,10 prompted us to examine the degree to which high prevalence of CH in patients with cancer could confound the results of tumor-only genomic profiling due to the presence of infiltrating leukocytes.¹⁰ Herein, we studied whether CH mutations may be misattributed as somatic tumor variants in patients when a matched normal sample is not sequenced and that some are actionable alterations, which could result in erroneous treatment recommendations.¹¹

We analyzed deep-coverage targeted NGS data of paired blood and tumor samples, as described previously,⁷ from 17 469 patients with advanced solid malignant neoplasms sequenced using the MSK-IMPACT platform between January 2014 and August 2017. This study was approved by the Memorial Sloan Kettering Cancer Center Institutional Review Board (IRB), and written consent was obtained from most participants. An IRB waiver was obtained to include patients who did not sign the consent form.

To identify the prevalence of CH alterations detected in solid-tumor specimens, we considered variants passing our detection thresholds for somatic mutation calling in tumor NGS data.¹² We retained CH alterations with a read depth of at least 20, alternate allele depth of at least 8, and a variant allele fraction (VAF) of at least 0.02 in the tumor sample. Alterations were annotated with population frequencies from 1000 Genomes, ExAC, and gnomAD databases to identify variants that would be potentially filtered out by tumor-only sequencing.

Clonal hematopoiesis alterations detected in solid tumors were annotated with OncoKB, a knowledge base that identifies oncogenic alterations and ranks potentially actionable alterations based on the level of evidence supporting that alteration as a predictive biomarker of drug sensitivity to US Food and Drug Administration (FDA)-approved or investigational therapies for a defined cancer type.¹¹

A generalized linear model was used to correlate patient age with the presence of CH mutations in solid tumors (CH-ST), correcting for sex and cancer type. P ≤ .05 was considered significant. Fisher exact test followed by Benjamini-Hochberg false discovery rate assessment was used to look for associations between tumor type and presence of CH-ST variants. Tumor types with a false discovery rate of 0.05 or less were considered significant.

We first examined MSK-IMPACT sequencing data of matched tumor and blood samples of 17 469 patients with cancer with 69 cancer types (eTable in Supplement 1). The median coverage was 723× for tumor samples and 486× for blood samples. The mean patient age at sample collection was 59.2 years (range, 0.3-98.9 years). Mutational analysis of hematopoietic cells identified 7608 presumptive somatic, nonsilent mutations in 396 genes from 4628 patients (26.5%). This led to identification of 1075 mostly nonrecurrent CH mutations (14.1%) (eSpreadsheet in Supplement 2) in solid tumors (CH-ST) from 912 patients (5.2%) with a median VAF of 0.04 (range, 0.02-0.21) in the solid-tumor samples and 0.16 (range, 0.04-0.53) in the matched blood samples. The incidence of CH-ST mutations increased with age (P < .001) as with CH (Figure 1A) and were most commonly observed in patients with skin cancer and non–small cell lung cancer (Figure 1B). The most commonly altered genes were DNMT3A, TET2, and PPM1D (Figure 2A).

Next, we sought to determine what proportion of these variants would have been retained by a tumor-only variant calling pipeline. Comparison of population frequencies in ExAC and gnomAD indicated that 1061 (98.7%) of these mutations were absent from the general population. Annotating CH-ST variants with OncoKB classified 534 (49.7%) CH variants as oncogenic or likely oncogenic (Figure 2B). Seventeen of 534 (3.2%) CH-ST mutations were associated with approved or investigational therapies (Table). Several of these are mutations common to both hematologic and solid cancers, such as IDH1/2, and can be used as biomarkers of targeted therapy. One example is a KRAS G12R mutation observed in a 77-year-old woman with colorectal cancer, which we identified as arising from CH based on higher VAF in blood than in tumor (blood VAF = 0.087; tumor VAF = 0.041) (Table and eFigure 1 in Supplement 1). Given that FDA-approved therapies require wild-type status for KRAS, misattribution of this mutation in the tumor without a matched-blood sample could lead to the withholding of treatment that could benefit the patient.

Another example of a targetable alteration is a presumed CH-associated BRCA2 Q3156* mutation observed in an 84-year-old man with melanoma and a history of pancreatic cancer at 0.118 VAF in the tumor (eFigure 2A and B in Supplement 1); this might have incorrectly prompted consideration of PARP (poly adenosine diphosphate ribose polymerase) inhibitor therapy. However, given the patient’s history of multiple cancers and the blood VAF of 0.344, we considered the possibility that the BRCA2 mutation was a pathogenic germline variant with loss of heterozygosity in the tumor. Allele-specific copy number analysis by FACETS¹³ revealed retention of both alleles, ruling out loss of heterozygosity in the tumor cells (eFigure 2C in Supplement 1). Sanger sequencing of the patient’s blood, tumor sample, saliva, buccal swap, and normal colon tissues confirmed that the variant was acquired in the hematopoietic compartment (eFigure 2D in Supplement 1). Taken together, these results demonstrate the perils of both tumor-only sequencing platforms and blood-only germline sequencing,¹⁴ where CH-ST mutations can confound results and lead to misguided clinical management.

As the application of NGS technologies continues to expand in clinical settings, it is important to identify sources of potential discrepancies and misleading results. While the use of population scale databases strengthened by cancer databases is a common method of identifying presumptive somatic mutations with tumor-only sequencing, it should be noted that different methodologies can differentially affect reporting and clinical decisions. Even then, correctly attributing hot spot mutations common to both solid tumors and hematological malignant neoplasms to the correct cell of origin, such as mutations in IDH1/2 and KRAS, is difficult in the absence of paired tumor-blood sequencing. This is further exemplified in a recent report¹⁰ in which activating JAK2 V617F mutations were described in the context of lung cancer based on unmatched tumor sequencing. Because JAK2 mutations have not been identified by other large-scale sequencing studies, the finding of JAK2 V617K mutations in that study likely represents an artifact resulting from detection of CH-ST mutations in tumor due to leukocyte infiltration.

One limitation of this study is that laboratories could establish different criteria for their variant filters, such as reporting only hot-spot mutations or using internal databases to identify recurrent likely germline mutations. While these could help with identifying CH-ST mutations, neither would eliminate them completely.

To our knowledge, this is the largest report on CH-ST based on a population of patients with advanced cancer. While a small fraction of patients are affected overall, we show that the prevalence of CH-ST is higher in older patients and that tumor-only sequencing results should be evaluated carefully, especially if treatment decisions are based on the variant results.

Accepted for Publication: April 24, 2018.

Published Online: June 5, 2018. doi:10.1001/jamaoncol.2018.2297

Correction: This article was corrected on October 26, 2018, to correct inaccurate conflict of interest disclosures.

Corresponding Author: Ahmet Zehir, PhD, Department of Pathology, Memorial Sloan Kettering Cancer Center, 1275 York Ave, PO Box 36, New York, NY 10065 (zehira@mskcc.org).

Author Contributions: Dr Zehir had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Ptashkin, Coombs, Hyman, Solit, Baselga, Levine, Berger, Zehir.

Acquisition, analysis, or interpretation of data: Ptashkin, Mandelker, Coombs, Bolton, Yelskaya, Solit, Baselga, Arcila, Ladanyi, Zhang, Levine, Berger, Zehir.

Drafting of the manuscript: Ptashkin, Solit, Berger, Zehir.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Ptashkin, Zehir.

Obtained funding: Solit, Levine.

Administrative, technical, or material support: Ptashkin, Mandelker, Yelskaya, Hyman, Solit, Arcila, Zhang.

Study supervision: Mandelker, Solit, Ladanyi, Berger, Zehir.

Conflict of Interest Disclosures: . Dr Baselga reports nonfinancial support and reasonable reimbursement for travel from Roche/Genentech; receiving fees from and stock ownership in the following companies: Aura Biosciences (including serving on the board of directors from 2013-2017), Infinity Pharmaceuticals (including serving on the board of directors from 2013-2017), PMV Pharma Biotechnologies (including serving on the scientific advisory board from 2014-present), Juno Therapeutics (acquired by Celgene) (including serving on the scientific advisory board from 2014-2017), Grail (including serving as member or chair of the scientific advisory board from 2016-2018), Varian Medical Systems (including serving on the board of directors from 2017-2018), Bristol-Myers Squibb (including serving on the board of directors from March to September 2018), Seragon (acquired by Roche) (including serving on the scientific advisory board of directors from 2013-2014); stock ownership in ApoGen Biotechnologies (including serving on the scientific advisory board from 2014-present), and Foghorn Therapeutics (including serving on the board from 2017-present); serving as cofounder, receiving fees from, and stock ownership in Tango (formerly Synthetic Lethal) from 2016-present and Northern Biologics (formerly Mosaic Biomedicals) (including serving on the scientific advisory board from 2013-present); receiving consulting and travel fees from Novartis and Eli Lilly; serving as cofounder of Venthera; and serving as investigator on a patent licensed to Memorial Sloan Kettering for use of phosphoinositide 3-kinase inhibitors for treatment of vascular malformations and serving as investigator for patents pending assignment to Memorial Sloan Kettering for combination therapy using PDK1 and PI3K inhibitors and inhibition of KMT2D for the treatment of breast cancer. No other disclosures were reported.

Funding/Support: This study was funded by the National Cancer Institute under the MSK Cancer Center Support Grant/Core Grant (P30 CA008748), the Starr Cancer Consortium, and Cycle for Survival.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Meeting Presentation: This work was presented at the American Society of Clinical Oncology Annual Meeting; June 5, 2018; Chicago, Illinois.

Additional Contributions: We gratefully acknowledge the members of the Molecular Diagnostics Service in the Memorial Sloan Kettering Cancer Center Department of Pathology and the Marie-Josée and Henry R. Kravis Center for Molecular Oncology. We also acknowledge the patients and their families for their participation.