Genotyping arrays are extremely unreliable for detecting very rare variants in human genetic studies: an example from a recent study of MC4R

Loss of function mutations in MC4R are the commonest cause of monogenic obesity. Lotta et al. recently published a study that used data from the UK Biobank and in vitro functional studies to conclude that beta-arrestin recruitment explains almost all (R²=88%) of the impact of MC4R rare variants on BMI (Lotta et al., 2019). The authors concluded that “MC4R signalling through beta-arrestin is critical for its role in the regulation of body weight” and that “Harnessing b-arrestin-biased MC4R signaling may represent an effective strategy for weight loss and the treatment of obesity-related cardiometabolic diseases.” (Lotta et al., 2019).

We question the conclusions of Lotta et al. because the majority of the variants in their study are not accurately genotyped. Lotta et al. based their analysis on 61 low frequency and rare variants in MC4R. The variants included single nucleotide variants and insertion-deletions and ranged in frequency from 0.0001% to 2%, with 48 having a frequency <0.005%. We have previously shown that rare variant genotypes from the UK Biobank and other genotyping arrays have a high rate of false positives because of the difficulty of genotype clustering of rare alleles (Weedon et al., 2021; Wright et al., 2019). Here we used exome sequence data from 49,960 of the ~500,000 UK Biobank individuals to test genotype accuracy for the MC4R variants analysed in Lotta et al.

Fifty-nine of the 61 variants reported by Lotta et al. were directly genotyped on the UK Biobank array. Based on the genotype array genotypes, 45 of these variants should have had at least one heterozygote in the 49,960 exome sequenced individuals. For 26 of these 45 variants, 100% of the heterozygous genotypes called by the genotyping array were false positives (i.e. they were homozygous reference in the exome sequencing data, Supplementary Data S1). For an additional 9 variants >10% of genotype array heterozygous calls were false positives. We confirmed the quality of the exome sequencing by manual inspection of IGV plots for all genotyping array variant genotypes. The remaining 14 variants, where no putative heterozygous genotypes occurred in individuals with exome sequence data available, are also likely to contain substantial genotyping errors because 12 have a minor allele frequency <0.001% (Weedon et al., 2021).

In addition, one variant that Lotta et al. suggests is strongly associated with increased BMI and loss of beta-arrestin recruitment (C271F) is incorrectly annotated as a C>A variant in the genotyping array data. All three individuals with a variant at this position in the exome sequence data have a C>T change resulting in a different missense variant from the one for which functional data was assayed in the study (C271Y). Similarly, the A244E variant is reported as a G>T variant in 9 individuals from the SNP array data. However, in the exome sequencing data only 1 person was heterozygous for this variant, but 7 of the remaining individuals were heterozygous for a different G>A change at this position that results in a A244V change.

In summary, we have shown that there are a large number of genotyping errors in the Lotta et al. study. This is due to the limitation of genotyping arrays to accurately call rare variants. In light of this, we think the analyses of Lotta et al. need to be re-examined.