ABCA4 ClinVar Pathogenic vs Benign Variant Frequencies in 1000 Genomes

We tested whether ClinVar-classified pathogenic variants in the ABCA4 gene have lower population allele frequencies than benign variants, using 1000 Genomes phase-3 super-population data. After merging 299 ClinVar-classified ABCA4 variants (176 pathogenic, 123 benign) with 1000 Genomes allele frequencies via a ±1bp positional join, we obtained 28 overlapping variants (16 pathogenic, 12 benign). A Mann-Whitney U test on allele frequencies yielded p = 5.08×10⁻¹ (not significant), with identical median allele frequencies of 7.99×10⁻⁴ in both groups (direction: pathogenic not rarer). This null result is substantially underpowered — only 9.4% of ClinVar variants overlapped with 1000G positions — and should not be interpreted as evidence that pathogenic and benign ABCA4 variants have equal population frequencies.

Introduction

ABCA4 (ATP-binding cassette subfamily A member 4) is a photoreceptor-specific transporter implicated in Stargardt disease and other inherited retinal dystrophies. The textbook expectation is that pathogenic variants should be rarer in the population than benign ones, as natural selection purges deleterious alleles. We tested this expectation for ABCA4 using ClinVar clinical significance classifications and 1000 Genomes phase-3 population allele frequencies.

Methods

Data Sources

We retrieved 500 ClinVar records for ABCA4 from NCBI E-utilities (retmax=500), of which 299 were classified as pathogenic/likely-pathogenic (n=176) or benign/likely-benign (n=123) after excluding uncertain, conflicting, and unclassified records. Clinical significance was parsed from the germline_classification field; compound calls containing "conflicting" were excluded.

We fetched 1000 Genomes phase-3 variant data for the ABCA4 genomic region (chr1:93,990,000–94,460,000) via tabix byte-range requests, yielding 12,807 variants with per-super-population allele frequencies (AFR, AMR, EAS, EUR, SAS). The 1000G data uses GRCh37 coordinates.

Join Strategy

ClinVar records from the E-utilities esummary endpoint lack ref/alt alleles and dbSNP rsIDs, preventing precise variant matching. We performed a positional join with ±1 base-pair tolerance, matching each ClinVar position to the nearest 1000G variant within 1bp. This fuzzy join yielded 28 overlapping variants (16 pathogenic, 12 benign) — a substantial reduction from the full ClinVar set. The ±1bp tolerance handles minor coordinate discrepancies but does not perform a systematic cross-assembly lift; ClinVar reports GRCh38 positions while 1000G uses GRCh37, so some mismatches at variant boundaries are expected.

Statistical Analysis

We applied a two-sided Mann-Whitney U test to compare allele frequencies between pathogenic and benign groups. We also computed a 2×2×5 chi-square test of independence across (pathogenic/benign) × (alt/ref) × (5 super-populations) on aggregated allele counts; this test is confounded by the non-random distribution of variants across populations and should not be interpreted as a direct test of pathogenicity effects. Analyses used the population_allele_freq reference template (adapted) with seed=1234.

Results

The Mann-Whitney U test did not detect a significant difference in allele frequency between pathogenic and benign ABCA4 variants (U = 110.5, p = 0.508). The median allele frequency was identical in both groups (7.99×10⁻⁴). The direction of effect was "pathogenic not rarer" — if anything, pathogenic variants had a non-significantly higher mean allele frequency in AFR (0.130 vs 0.049) but lower in other populations.

The chi-square test on aggregated allele counts across all five super-populations was highly significant (χ² = 1425.8, df = 12, p ≈ 0), but this reflects the confounded structure of the comparison — the pathogenic and benign variant sets differ in their population-level allele frequency distributions in ways that are not attributable to pathogenicity alone.

Limitations

1. Severely underpowered: Only 28 of 299 ClinVar variants (9.4%) overlapped with 1000G positions, limiting statistical power. The low overlap is driven by ClinVar's lack of ref/alt alleles and the GRCh37/GRCh38 coordinate mismatch.

2. Coordinate system mismatch: ClinVar reports GRCh38 positions while 1000G phase-3 uses GRCh37. The ±1bp join may introduce mismatches at variants near indels or structural variants, and cannot resolve systematic cross-assembly differences.

3. Retmax cap: The ClinVar fetch was capped at 500 records; the full ABCA4 ClinVar dataset likely contains more variants, potentially including more with 1000G overlap.

4. 1000G allele frequencies are aggregate: The per-super-population AFs are summary statistics, not individual genotype data, limiting the granularity of population-specific comparisons.

5. Template mismatch: The population_allele_freq template was designed for comparing allele frequencies across super-populations, not for comparing pathogenicity classes. We adapted it for this question, but the chi-square results should be interpreted with caution as they conflated population structure with pathogenicity effects.

Provenance

• ClinVar: NCBI E-utilities, accession ABCA4[gene], 500 records (retmax=500), 85,860 bytes

  • URL: https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esummary.fcgi
  • Region: ABCA4 gene

• 1000 Genomes phase 3: EBI FTP, chr1:93,990,000–94,460,000 (pysam tabix byte-range), 12,807 variants, 616,433 bytes

  • URL: https://ftp.1000genomes.ebi.ac.uk/vol1/ftp/release/20130502/

• Analysis: populationallelefreq reference template (adapted), seed=1234

  • Script: analysis.py
  • Join: ClinVar × 1000G via ±1bp positional tolerance
  • Total download: 702,293 bytes