Ancestry-stratified variant classification in monogenic diabetes genes: annotation coverage and differential curation burden

This study reveals that while non-European patients with monogenic diabetes face a significant 70% annotation gap in ClinVar and gnomAD, the primary equity issue is not a simple excess of variants of uncertain significance but rather a distinct curation deficit and reclassification lag compared to European populations, highlighting the urgent need for ancestry-stratified evaluation of variant curation standards.

The Gist

The Big Picture: A Library with Missing Books

Imagine the world of genetic medicine as a massive library where doctors go to look up "books" (genetic variants) to figure out if a patient's DNA is causing their diabetes.

Two main libraries hold these books:

1. ClinVar: The "Medical Dictionary." It tells doctors if a specific genetic change is "Bad" (Pathogenic), "Good" (Benign), or "We Don't Know Yet" (VUS).
2. gnomAD: The "Population Census." It tells doctors how common a genetic change is in different groups of people.

The Problem: This library was built almost entirely by people of European descent. It's like a library in London that has millions of books about British history but very few books about African, Asian, or South American history.

This paper asks: Does this imbalance hurt patients who aren't European?

The Shocking Discovery: 70% of the Books Are Missing

The researchers looked at 17 genes known to cause a specific type of diabetes called Monogenic Diabetes (a rare, single-gene form of the disease). They checked the "Census" (gnomAD) to see what genetic changes exist in people around the world, and then checked the "Dictionary" (ClinVar) to see if those changes had been explained.

The Result:

• 70% of the genetic variants found in the global population have NO entry in the Medical Dictionary.
• Imagine a doctor finding a strange code in a patient's DNA. They open the dictionary, but the page is blank. They have no idea if it's dangerous or harmless.

Why this matters:
If you are a patient of European ancestry, your genetic code is likely in the dictionary. If you are of African, South Asian, or Middle Eastern ancestry, your specific genetic code is likely missing entirely. You don't get a "Maybe" answer; you get no answer at all.

The "Uncertain" Trap (The VUS Problem)

Usually, when geneticists talk about health inequality, they say: "Non-European patients get more 'Uncertain' results (VUS)."

This paper found something more complex:

1. The European "Backlog": Because so many European people have been tested, there is a huge pile of European genetic changes in the dictionary. However, many of them are marked as "Uncertain" because scientists haven't done the extra homework to prove they are safe or dangerous. It's a curation backlog.
2. The Non-European "Blackout": For non-European patients, the problem isn't just that they get "Uncertain" results. It's that their variants are missing from the dictionary entirely.
   • Analogy: If you ask a librarian for a book about a specific village in Africa, and the library doesn't even have a card for that village, you can't even start your research.

The "GCK" Twist: When the Rules Backfire

There is one specific gene, GCK, that shows a weird twist.

• In Europe: Scientists have studied this gene for years. They found that many changes people thought were "Bad" are actually "Safe." They have updated the dictionary to say "Benign."
• In Non-European Populations: The dictionary hasn't been updated yet. Because the studies proving these changes are safe were mostly done on European families, the dictionary still says "Uncertain" for the exact same changes in African or Asian patients.

The Result: Non-European patients are more likely to be told they have an "Uncertain" result for GCK, while European patients get a clear "Safe" answer. This isn't because the DNA is different; it's because the research data is missing for non-European groups.

The Real-World Cost: A 10-Year Delay

Why does a missing book matter?

• Monogenic Diabetes is treatable.
  • If you have Type A, you can stop taking insulin and take a simple pill instead.
  • If you have Type B, you don't need any medicine at all.
• The Consequence: If the doctor can't read the genetic code (because the book is missing or the page is blank), they treat the patient for regular diabetes.
  • They might give insulin to someone who doesn't need it.
  • They might give a pill to someone who needs insulin.
  • The paper notes that non-European patients wait an average of 10 years longer to get the right diagnosis and the right treatment.

The Solution: How Do We Fix the Library?

The author suggests three steps to fix this inequality:

1. Submit More Data: Hospitals serving non-European populations need to send their genetic findings to the "Dictionary" (ClinVar). We need to fill in the blank pages.
2. Use Existing Data Smarter: We already have the "Census" data (gnomAD) showing how common these variants are in different groups. We just need to use that data to update the "Dictionary" and re-classify old results.
3. Check the Rules: The rules scientists use to decide if a gene is "Bad" or "Good" need to be tested to make sure they work equally well for all skin colors and ancestries, not just Europeans.

The Bottom Line

The biggest problem isn't that non-European patients get too many "Uncertain" answers. The biggest problem is that for 70% of the genetic variants found in the world, there is no answer at all.

Until we fill in the missing pages of the genetic dictionary for all human populations, millions of patients will continue to wait a decade longer for the right treatment, simply because their genetic story hasn't been written down yet.

Technical Summary

1. Problem Statement

The study addresses a critical equity gap in clinical genomics: the skewed population composition of major variant databases (ClinVar and gnomAD), which are heavily biased toward European ancestry. While it is well-documented that non-European patients often receive more "Variants of Uncertain Significance" (VUS) results, this paper investigates whether this disparity holds true for monogenic diabetes (MODY) genes and identifies the root causes.

The authors hypothesize that the primary issue is not merely a higher rate of VUS for non-European patients, but a fundamental annotation gap (variants lacking any clinical classification) and a curation deficit (lack of functional/segregation data to resolve classifications) that disproportionately affects non-European populations. This leads to delayed diagnoses and inappropriate treatments (e.g., unnecessary insulin therapy) for non-European patients.

2. Methodology

The study employed a large-scale, cross-referential analysis of public genomic databases:

• Data Sources:
  • ClinVar: Downloaded variant_summary.txt.gz (GRCh38, April 2026 release; ~4.42 million variants).
  • gnomAD: Accessed v4.0 genome dataset (76,215 whole genomes) via GraphQL API.
• Gene Selection: Focused on 17 genes associated with monogenic diabetes (e.g., HNF1A, HNF4A, GCK, KCNJ11, PDX1). KLF11 was included but flagged as refuted by ClinGen.
• Ancestry Stratification: Variants were stratified into eight genetic ancestry groups: African/African-American (AFR), Latino/Admixed American (AMR), East Asian (EAS), South Asian (SAS), Middle Eastern (MID), non-Finnish European (NFE), Finnish (FIN), and Ashkenazi Jewish (ASJ).
• Cross-Referencing: Variants present in both databases were matched by chromosome, position, reference, and alternate alleles.
  • Note: A variant was counted in every ancestry group where it had a non-zero allele count, meaning totals are not mutually exclusive.
• Metrics Calculated:
  • Annotation Coverage: The percentage of gnomAD variants for a given gene that have any entry in ClinVar.
  • Classification Rates: Proportions of Pathogenic (P), Likely Pathogenic (LP), VUS, Likely Benign (LB), and Benign (B) within the classified subset.
  • Statistical Analysis: Chi-square tests with Bonferroni correction for VUS rate comparisons; Pearson correlation for coverage vs. divergence.

3. Key Results

A. The Annotation Gap (Primary Finding)

• 70.3% of variants lack classification: Of the 14,691 variants identified in gnomAD across the 17 genes, only 29.7% had a corresponding ClinVar entry.
• Asymmetric Impact: While the gap exists for all groups, it disproportionately affects non-European patients. Variants specific to African, South Asian, or Southeast Asian populations are more likely to be absent from ClinVar entirely. Consequently, these variants often result in "no result" or are filtered out by frequency filters, rather than being reported as VUS.
• Gene Variability: Coverage ranged from 12.7% (APPL1) to 61.3% (KCNJ11). Even for the most common MODY gene, HNF1A, 60% of gnomAD variants remained unannotated.

B. VUS Rates and Ancestry

Contrary to the general trend in clinical genetics where non-European groups have higher VUS rates, this study found a more complex picture:

• NFE (Non-Finnish European) had the highest VUS rate (32.1%): This is attributed to a "curation backlog"—high volumes of rare variants submitted from European cohorts without sufficient functional follow-up to resolve them.
• AFR (African) VUS rate (29.2%) was statistically indistinguishable from NFE: Unlike other non-European groups (which had significantly lower VUS rates), AFR variants showed a high VUS burden. However, the mechanism differs: it is driven by a deficit in functional and segregation data rather than a surplus of uncurated submissions.
• Other Non-European Groups: EAS, SAS, AMR, and MID groups had significantly lower VUS rates than NFE (all p < 0.001), largely because many of their variants are unannotated (in the 70% gap) rather than classified as VUS.

C. The "Pattern Inversion" in GCK

A critical finding was observed in the GCK gene (associated with MODY2):

• Non-European VUS rate (18.5%) > European VUS rate (15.0%): This inversion suggests that while European variants are being progressively reclassified as Benign/Likely Benign based on accumulated functional data (via ClinGen MDEP rules), equivalent variants in non-European populations remain VUS due to a lack of population-specific data.
• Implication: Gene-specific curation rules calibrated on European data may inadvertently perpetuate disparities if not validated across ancestries.

4. Key Contributions

1. Reframing the Equity Problem: The study argues that the primary barrier to equity is not just the rate of VUS, but the 70% annotation gap. Variants absent from ClinVar cannot be classified, leading to diagnostic failure rather than just uncertainty.
2. Mechanism Differentiation: It distinguishes between two distinct mechanisms driving disparities:
   • European: A "curation backlog" (high submission volume, low functional resolution).
   • Non-European: An "annotation and data deficit" (lack of submissions and lack of functional/population frequency evidence).
3. Gene-Specific Nuance: It demonstrates that disparities are not uniform across genes; some (like GCK) show an inversion where non-European variants are more likely to be VUS due to the specific application of ancestry-biased curation rules.
4. Critique of Current Criteria: It highlights that ClinGen Variant Curation Expert Panel (VCEP) criteria, which rely heavily on allele frequency (PM2) and functional data (PS3) derived from European populations, may systematically misclassify variants in non-European contexts.

5. Significance and Recommendations

• Clinical Impact: The "10-year delay" in diagnosis for non-European patients is directly linked to database limitations. Patients receive incorrect treatments (e.g., insulin for GCK or HNF1A cases that don't need it) because the genetic evidence cannot be interpreted.
• Proposed Solutions:
  1. Systematic Submissions: Clinical labs serving non-European populations must submit variants to ClinVar to close the annotation gap.
  2. Reclassification using Existing Data: Apply diverse population allele frequencies (from gnomAD) to re-evaluate existing ClinVar entries, potentially reclassifying "Pathogenic" variants as Benign (as seen in cardiomyopathy studies) or resolving VUS.
  3. Ancestry-Stratified Validation: ClinGen VCEPs must validate gene-specific rules across ancestry groups before implementation to ensure they do not perpetuate bias.

Conclusion: The paper concludes that the genomics community possesses the tools to address these disparities. The bottleneck is not a lack of sequencing, but a lack of diverse data curation and the systematic application of existing diverse population data to clinical interpretation.