Background: A genomic resource for rare loss-of-function variants

The Pakistan Genomic Resource (PGR) team analyzed exomes and whole genomes from 173,303 people and identified predicted homozygous loss-of-function variants in 6,476 genes. This means that both copies of a gene, one inherited from each parent, carry the same loss-of-function variant. Naturally occurring variants of this kind offer an opportunity to study changes in the human body when protein function is reduced or lost.

Loss-of-function variants are useful for assessing targets for drugs that inhibit proteins, but they are generally rare, making it difficult to find enough carriers. South Asia, including Pakistan, has been underrepresented in large-scale genomic studies. The researchers investigated variants and gene functions not captured by existing resources in a population with genetic diversity and a high degree of relatedness. Rather than experimentally deleting genes, they examined health information from people with naturally occurring variants.

Key findings: From variant catalogs to human phenotypes

PGR is a nested case–control cohort that recruited patients with various diseases and controls at hospitals in 23 cities across Pakistan. The analysis included exomes from 166,625 people and whole genomes from 6,678 people. The researchers identified 6,625,995 variants in the protein-coding regions of 19,021 genes. Of these, 30% were absent from the full gnomAD v4.1 comparison dataset, which includes South Asian data; the proportion reached 47% when compared with non-South Asian data alone. This means the variants were absent from those comparison datasets, not that they were all discovered for the first time in humans.

A total of 34,364 participants, about 20% of the cohort, were predicted to carry homozygous loss-of-function variants in at least one gene. The variants involved 6,476 genes, about one-third of protein-coding genes. These predictions were obtained through computational analyses and filtering and must be distinguished from experimental confirmation of functional loss for every variant. The team analyzed population structure and runs of homozygosity inherited from common ancestors. They also examined associations of gene-level predicted loss-of-function variants and predicted damaging missense variants with 46 biomarkers.

To assess therapeutic targets, the researchers conducted recall-by-genotype studies, inviting participants and consenting family members back for testing based on their genotypes. For LRRK2, a Parkinson’s disease therapeutic target, kidney function abnormalities were observed in all three homozygous loss-of-function carriers: two from the original cohort and one identified through recall. The researchers interpreted this as a reason to monitor kidney function during long-term LRRK2 inhibition. However, observations in three people cannot establish drug-induced kidney toxicity or causality.

In the recall study of TRPM8, a gene related to cold sensation, pain responses were assessed by immersing a hand in water at 4°C for up to 150 seconds. Loss-of-function variants showed nominally significant associations with longer times to pain onset and longer durations of pain tolerance. There were only three homozygous carriers, and the analysis did not apply a multiple-testing correction. Migraine-related phenotypes were not collected, so the findings cannot be considered evidence that a migraine treatment is effective.

Implications and outlook: Evidence for target assessment and access conditions

PGR expanded the reference resources available for interpreting clinical variants and studying disease in the Pakistani population. Linking genomic information with clinical measurements and family recall testing creates an opportunity to compare target benefits or safety concerns raised in animal studies with human data. However, because recruitment was hospital-based, the cohort should not be interpreted as a random sample representative of the country’s general population. Findings from small numbers of carriers require independent replication, and unmeasured confounding may have contributed to the observed associations.

Data access conditions also need to be distinguished. Variant catalogs and biomarker summary statistics are public, but individual-level genomic and phenotypic data are not freely downloadable. Academic researchers must submit a research plan and a privacy-protection plan for access review. The initial review response time of 45 business days stated in the paper is not a deadline for approval or completion of data delivery. The study design, findings, and access conditions are described in the full paper.