Transitioning from gnomAD V2 to V4: Impact on genetic variant classification

The latest version of gnomAD , the database that most clinical genomics labs rely on to obtain information about variant allele frequencies in ostensibly healthy people, was recently updated to version 4 (V4). GnomAD allele frequencies are critical to determining whether observed variants represent benign variation or are potentially pathogenic.

?

GnomAD V4 includes variants from exomes and genomes and is based on reference genome GRCh38. Importantly, the number of individuals represented in the database has grown from ~140,000 in version 2 (V2) to over 800,000 in V4. V4 also expands diversity to include new ancestral groups previously underrepresented in V2.

?

Most labs will transition over from gnomAD V2 to V4, but there are some important things to consider before doing so.

?

Overall allele frequency differences between gnomAD V2 and V4

gnomAD is not a representative population-based sample. It is a pool of multiple convenience samples of various sizes of different ethnic groups. Because of this, the overall population allele frequencies are subject to change whenever a new cohort is added to gnomAD.

?

For example, V4 includes a new Middle Eastern cohort of ~6000 individuals. The next version (V5) will likely include a large number of the multi-ethnic All of Us study participants.

?

Because of this, expect to see some differences in allele frequencies between V2 and V4, but because there is still a large European bias in gnomAD, the differences in overall allele frequency may not be that profound.

?

Availability of GroupMax Filtering Allele Frequency in gnomAD V4

V4 includes a new field, GroupMax Filtering Allele Frequency. This replaces the old PopMax from V2 and represents the highest ancestry group-specific filtering allele frequency. Technically, it is the lower bound of the 95% confidence interval of the allele frequency in the ancestry group. It is calculated for all non-bottlenecked ancestry groups.

?

GroupMax FAF is important because it is the guideline-recommended frequency to use for identifying benign/likely benign variants. It features prominently in the forthcoming ACMG/ClinGen V4 guidelines (preview - not yet released).

?

In theory, GroupMax FAF values should be more stable across versions of gnomAD than overall allele frequency, because GroupMax is only calculated on non-bottleneck, continental subpopulation. However, in V4.1, gnomAD GroupMax values now include variants from the Middle Eastern population, so even GroupMax values may fluctuate between versions of gnomAD.

?

With GroupMax available, labs may wonder whether they should use this allele frequency to make decisions about ACMG rules BA1/BS1, as opposed to using overall gnomAD allele frequency.

?

Notwithstanding that it is advised by ACMG guidelines to use GroupMax, the question is really whether GroupMax and overall AF differ significantly.

Comparison of GroupMax and Overall allele frequencies

We compared gnomAD V4 GroupMax and overall allele frequency (AF) for all protein-altering variants in BRCA2 as an example. The difference between the two AF measurements was less than an order of magnitude for 95% of variants. Among the 5% of variants with an order of magnitude difference or greater, the vast majority (4.8%) were variants where GroupMax values were higher than the overall AF. These tend to represent variants that are common in non-European populations.

Impact of using gnomAD V4 frequencies with older cutoffs for BA1/BS1

On a more practical level, we evaluated how many gnomAD variants in BRCA2 would be assigned BA1 (stand alone Benign criteria) or BS1 (strong Benign criteria) if gnomAD V4 GroupMax or overall AF were used.

?

The ClinGen Variant Curation Expert Panel (VCEP) for BRCA2 sets the following thresholds: BA1 >0.001; BS1 >0.0001. Using these thresholds, we found that using GroupMax FAF resulted in 3 times more variants being classified as Benign (based on BA1), than if overall AF were used, and nearly twice as many variants with strong evidence supporting Benign classification (based on BS1).

?

The 75 variants attributed BA1 using GroupMax FAF included 13 variants with conflicting evidence of pathogenicity in ClinVar, and the remaining were classified as B/LB in ClinVar. Among the 25 variants attributed BA1 using overall AF, only 1 was conflicting in ClinVar, with the remaining B/LB.

Importantly, the BS1 threshold captures several variants classified as P/LP in ClinVar. Using GroupMax values, three well-established pathogenic variants would be attributed BS1, strong evidence of being benign.

?We next asked whether this same phenomenon would have been observed using gnomAD V2. Using gnomAD V2, BS1 was invoked for only 1 P/LP variant (ClinVar 38260) using Exome GroupMax FAF. One of the other two (38221) had a lower frequency in V2, while the other (38076) was new to V4.

?

To determine if this phenomenon was unique to BRCA2, we considered 14 other genes from the ACMG SF3.2 gene list where there were VCEP-recommended BA1/BS1 cutoffs. Among them we found that using GroupMax FAF identified one ClinVar 2-star Pathogenic variant in HNF1A surpassing the VCEP-recommended BS1 threshold (CV 1328238) and two PALB2 variants surpassing the BS1 threshold (CV 126711 - 3 star Pathogenic and CV 136132 - 2 star Pathogenic).

?Conclusions

These analyses suggest that, while gnomAD V4 GroupMax FAF may be the most appropriate allele frequency to use when classifying variants going forward, labs should re-evaluate how their thresholds for BA1 and BS1 were established to ensure that when using GroupMax FAF, pathogenic variants aren’t misclassified as B/LB.

?

ClinGen has recommended an approach for setting BA1/BS1 thresholds based on empirical data that would minimize misclassifications. We have highlighted this method in a previous article . More information can be found in this ClinGen webinar as well.

?