This was a descriptive cross-sectional pharmacogenetic variant prevalence study of a population of 41 patients. The sample was selected based on the availability of genomic sequencing data obtained using the NGS platform developed by the A Coruña University Hospital Complex. The studied population corresponded to the total number of patients recruited by the Hospital within the framework of a project geared towards validating pharmacokinetic and pharmacogenetic biomarkers related with the risk of developing neuropathy following administration of taxanes in the context of the neoadjuvant breast cancer therapy.

The genomic regions of clinical interest were captured using a personalized capture probe library (SureSelect Target Enrichment Kit for the Illumina paired-end multiplex sequencing method; Agilent Technologies, Santa Clara, California, USA) and sequenced on the HiSeq 1500 platform (Illumina, San Diego, California) following Illumina protocols14,15. The read depth (number of times a base was sequenced by independent reads) of every nucleotide of genes from the defined genomic regions of interest was >30× (mean: 250×-400×). Analytical validation of this platform has been previously described13. The capture probe library allows sequencing of a total of 433,000 bases. The genes and regions of interest evaluated in this study correspond to a subset of all the genomic regions included in the capture probe library.

A group of genomic regions was selected from 18 pharmacogenomics-related genes that were considered clinically actionable (CACNA1S, CYP2B6, CYP2C19, CYP2C9, CYP2D6, CYP3A5, CYP4F2, DPYD, F5, G6PD, HLA-A, HLA-B, NUDT15, RYR1, SLCO1B1, TPMT, UGT1A1 and VKORC1). These genes have been described in several clinical guidelines, including CPIC (Clinical Pharmacogenetics Implementation Consortium)16, DPWG (Dutch Pharmacogenomics Working Group)17 and CPNDS (Canadian Pharmacogenomics Network for Drug Safety)18. A mixed research strategy was developed, which consisted of: a) the development of a specific allele-variant database that allowed an automatic evaluation of the genetic variants and the pharmacogenetic alleles described in the literature; this database comprised 1,027 variants and was developed based on the PharmVar19 and PharmGKB20 databases, and on the GeT-RM pharmacogenomic projects21–23; b) an analysis of the candidate functional variants in the coding regions of genes CACNA1S, CYP2B6, CYP2C19, CYP2C9, CYP2D6, DPYD, NUDT15, RYR1, SLCO1B1, TPMT, UGT1A1 and VKORC1.

The sequence analysis was carried out using a purpose-developed bioinformatic algorithm that included the demultiplexing of the samples as well as all the steps needed to obtain a validated report of the annotated variants, together with their coverage and quality parameters13. Haplotypes were assigned following a purpose-designed algorithm that used variant-allele translation tables developed together with the variant files (vcf format) and the coverage data (cov format) obtained from each sample13.

The analysis of the copy number variants (CNVs) and the structural variants of CYP2D6 was carried out using a previously-described and validated comparative coverage depth strategy13,24.

It was done using the genotype-to-phenotype prediction classification system described in pharmacogenomic prescription guidelines and recommendations. These standards are summarized below. Phenotypes were determined by genotyping sets (haplotypes) of genetic variants known as star alleles “*”. Every patient has two star alleles that are collectively referred to as a diplotype or genotype (e.g., *1/*2). Each star allele was then assigned a function (i.e., no, decreased, normal or increased function) and a corresponding numerical activity level based on the evidence available on databases and in leading publications such as PharmVar. The activity levels of the two alleles in each individual were combined and translated into a phenotype (poor, intermediate, normal, rapid, ultrarapid) that was then linked to a selection of specific drugs and a dosing recommendation22,25.

It was determined based on the prescription recommendations described in the CPIC, DPWG and CPNDS clinical guidelines16–18. Three different categories were established: “non actionable”, “conditional” and “actionable”. Table 2 in the Annex includes a detailed description of this classification.

In the first place, a clinically actionable allele prevalence study was conducted; alleles were grouped by patient and by gene. Secondly, an analysis was carried out of the prevalence of the different pharmacogenetic phenotypes obtained from the genotype interpretation process. Thirdly, the clinical actionability of the pharmacogenetic phenotypes identified foe each of the drugs described in pharmacogenetic clinical guidelines was established. Lastly, a bioinformatic algorithm was used to select the potentially deleterious candidate variables using the following filtering criteria: they had to be rare variants (whose gnomAD population frequency was below 1%) located in coding regions (gene coding exons), which could bring about changes in the protein sequence (nonsense, missense) and with a phred score above 20 for the CADD bioinformatic predictor (the phred score is used to select the most deleterious 1% of all possible variants of the gene).

The present study was approved by the Drug Research Ethics Committee of Galicia (CEIm-G ID 2017/437). All the patients included gave their informed consent to participate in the study.