Large-Scale Pharmacogenomics Analysis of Patients With Cancer Within the 100,000 Genomes Project Combining Whole-Genome Sequencing and Medical Records to Inform Clinical Practice

Abstract

Purpose: As part of the 100,000 Genomes Project, we set out to assess the potential viability and clinical impact of reporting genetic variants associated with drug-induced toxicity for patients with cancer recruited for whole-genome sequencing (WGS) as part of a genomic medicine service.

Methods: Germline WGS from 76,805 participants was analyzed for pharmacogenetic (PGx) variants in four genes (DPYD, NUDT15, TPMT, UGT1A1) associated with toxicity induced by five drugs used in cancer treatment (capecitabine, fluorouracil, mercaptopurine, thioguanine, irinotecan). Linking genomic data with prescribing and hospital incidence records, a phenome-wide association study (PheWAS) was performed to identify whether phenotypes indicative of adverse drug reactions (ADRs) were enriched in drug-exposed individuals with the relevant PGx variants. In a subset of 7,081 patients with cancer, DPYD variants were reported back to clinicians and outcomes were collected.

Results: We identified clinically relevant PGx variants across the four genes in 62.7% of participants in our cohort. Extending this to annual prescription numbers in England for the drugs affected by these PGx variants, approximately 14,540 patients per year could potentially benefit from a reduced dose or alternative drug to reduce the risk of ADRs. Validating PGx associations in a real-world data set, we found a significant association between PGx variants in DPYD and toxicity-related phenotypes in patients treated with capecitabine or fluorouracil. Reported DPYD variants were deemed informative for clinical decision making in a majority of cases.

Conclusion: Reporting PGx variants from germline WGS relevant to patients with cancer alongside primary findings related to their cancer can be clinically informative, informing prescribing to reduce the risk of ADRs. Extending the range of actionable variants to those found in patients of non-European ancestry is important and will extend the potential clinical impact.

FIG 1.

Analysis framework. The gene-drug pairs were defined by the National Health Service of England (NHSE)-Genomics England PGx Working Group. The genetic cohort comprised 76,805 individuals from various inferred ancestries. The size of the unrelated cohort (n = 54,831) is represented by the numbers in brackets within the cohort breakdown textbox. The genomic analysis consisted of two branches. One branch represents the genomic landscape of the 14 PGx variants, linkage disequilibrium analysis, and allele frequency comparison against gnomAD in the unrelated cohort, followed by a PheWAS. The PheWAS tests the association of the PGx variants with phenotypes including those indicative of possible adverse drug reactions, and it was performed by integrating the Systematic Anti-Cancer Therapy data set and HES records with participant PGx genotypes. A PheWAS baseline analysis was performed including all records before any treatment. The treatment PheWAS analysis included all records up to 16 weeks of treatment. The phecodes were obtained from the ICD-10 codes in the data sets. Relevant data were available for three PheWASs: DPYD and fluorouracil, DPYD and capecitabine, and UGT1A1 and irinotecan. The second branch of this study—DPYD cohort (n = 7,081)—is the translation of the genomic information into genotype-predicted drug responses. ^aIt should be noted that 224 participants in the DPYD cohort are not part of the 100KGP analysis cohort (Data Supplement). gnomAD, the Genome Aggregation Database; HES, Hospital Episode Statistics; ICD, International Classification of Diseases; PGx, pharmacogenetics; PheWAS, Phenome Wide Association Study; SACT, systematic anticancer therapy.

FIG 2.

Proportion of PGx variants for DPYD, UGT1A1, TPMT, and NUDT15 in the genetic cohort, with ethnicity breakdown. All DPYD variants were present in participants of European ancestry (<3.0%-4.4% of European cohort) and completely absent in East Asian ancestry. Moreover, DPYD*7 (c.295_298delTCAT; rs72549309) was only present in Europeans (<3.0%). DPYD*2A (c.1905+1G>A; rs3918290) and *13 (c.1679T>G; rs55886062) are absent in Africans and admixed Americans, and *13 is absent in South Asians. UGT1A1*28 is more common than *6 (211G>A; rs4148323) in all ancestry groups (>50% compared with <5%, respectively), except in East Asians where both alleles are of similar frequency (23.3% and 25.1%, respectively). All TPMT variants are present in Europeans (<3.0%-9.4%), and *3C (719A>G; rs1142345) is the most common TPMT variant in all ancestry groups (3.5%-9.4%). TPMT*4 (626-1G >A; rs1800584) is absent in ancestries except for Europeans, and *2 (238G>C; rs1800462) is absent in East Asians and South Asians. Both NUDT15 variants are present in all ancestries and are present in admixed Americans, East and South Asians more than Africans, Europeans, and mixed ancestries. For ancestry, any proportion that is <3.0% has been disguised to prevent reidentification of participants in the study. PGx, pharmacogenetic.

FIG 3.

Translation of the actionable diplotypes to treatment recommendations on the basis of CPIC guidelines and primary and secondary care prescribing data in England in 2018-2019. The percentage of patients in the genetic cohort (n approximately 76k) who would be recommended a normal dose, reduced dose, or reduced dose/alternative drug on the basis of their genotype for each drug affected by the PGx variants independent of clinical indication. A total of 68,539 participants with variants related to thiopurines (mercaptopurine, thioguanine, and azathioprine) would remain on the normal dose; 8,017 would require a dose reduction, and 249 would require a dose reduction or alternate drug; 4,679 participants would require a dose reduction for variants related to capecitabine and fluorouracil, and 32 would require an alternate drug. For irinotecan, 8,153 would require a dose reduction. The table on the right shows the total number of patients who were prescribed these drugs in primary and secondary care (using numbers for 2018 to 2019) and presents the number of patients per year in England who would be predicted to require a reduced dose or an alternative drug. CPIC, Clinical Pharmacogenetics Implementation Consortium; NHSE, National Health Service of England; PGx, pharmacogenetic.

FIG 4.

(A) Distribution of different tumor types in present in the DPYD cohort. The graph is grouped into tumor types that are treated with fluoropyrimidine and those that are not. (B) (Left) Distribution of DPYD variants. (Middle) Distribution of different ancestries for each variant. (Right) Proportion of participants who have or have not had a particular DPYD variant added to the outcomes questionnaires. The HapB3 likely causative (c.1129-5923 C>G; rs75017182) and HapB3 proxy (c.1236 G>A; rs56038477) variants are the most common, and *13 is the least common variant. Note that bars with low participant numbers (<5 or <10%) have disguised numbers to prevent reidentification of participants.