Performance of a Type 1 Diabetes Genetic Risk Score in a Multi-centric Study from India and its Implications in Clinical Practice

This multi-centric study validates that a 67-SNP Type 1 Diabetes Genetic Risk Score performs consistently across diverse Indian populations but requires ancestry-specific threshold optimization and shows variable efficacy based on autoantibody status and age of onset to ensure accurate clinical diagnosis.

The Gist

The Big Picture: A Genetic "Weather Forecast" for Diabetes

Imagine you have a genetic risk score (GRS). Think of this like a weather forecast for Type 1 Diabetes (T1D). For years, scientists have built these forecasts using data from people of European ancestry. They created a "storm warning" system that tells you how likely you are to get hit by a diabetes "storm."

However, this paper asks a crucial question: Does this European weather forecast work accurately for people in India?

The answer is: Yes, it predicts the storm, but the "alarm settings" are all wrong. If you use the European settings in India, you might miss the storm entirely.

---

The Story of the Study

1. The Problem: One Size Does Not Fit All

Type 1 Diabetes is an autoimmune disease where the body's immune system accidentally attacks the pancreas. In Europe, scientists found 67 specific genetic "switches" (SNPs) that act like tripwires for this disease. They built a score based on these tripwires.

But India is genetically different from Europe. It's like trying to use a map of London to navigate Mumbai. The streets (genes) might look similar, but the traffic patterns (how the genes behave) are different. Also, in India, many people with Type 1 Diabetes don't show the usual "smoke signals" (autoantibodies) that doctors look for, making diagnosis tricky.

2. The Experiment: Testing the Map

The researchers gathered a massive team from four different cities in India (Pune, Hyderabad, Mysore, and Mumbai). They looked at:

• 597 patients with Type 1 Diabetes.
• 3,347 healthy people (controls).

They checked the genetic "tripwires" in everyone's DNA to see if the European score could tell the difference between a patient and a healthy person.

3. The Findings: The Score Works, But Needs Tuning

A. The Score is a Good Detective
The genetic score was surprisingly good at spotting Type 1 Diabetes in Indians. It correctly identified the disease about 84% of the time. This is great news because it means we can use this tool across different regions and languages in India. It's a unified tool for the whole country.

B. The "Smoke Signal" Matters
The score worked best when patients had "smoke signals" (autoantibodies) in their blood.

• With smoke signals: The score was a sharp detective (85% accuracy).
• Without smoke signals: The score got a bit foggy (75% accuracy).
• Analogy: It's like a metal detector. It works great on a beach with lots of coins, but if the beach is empty, it's harder to find the one hidden coin.

C. Kids vs. Adults: The Age Factor
The score was much better at identifying children with diabetes than adults.

• Why? Children usually have a very specific, high-risk genetic combination (like a "super-charged" storm). Adults often have a "lighter" genetic load or different genetic combinations that look more like Type 2 Diabetes.
• Analogy: The score is like a net designed to catch big fish (kids with strong genetics). It lets the smaller, slippery fish (adults with milder genetics) slip through the holes.

D. The "Alarm Threshold" is Too High
This is the most important discovery. The European scientists set the "alarm" to go off at a high score (e.g., 13.0).

• In Europe, 50% of people with diabetes have a score above 13.0.
• In India, only 30% of people with diabetes have a score above 13.0.
• The Result: If Indian doctors use the European alarm, they will miss 70% of the patients! The genetic "storm" in India is real, but it's slightly less intense than in Europe, so the alarm needs to be set lower to catch it.

4. The Solution: A New "India-Specific" Setting

The researchers calculated a new "alarm setting" specifically for India.

• Old European Setting: 13.0 (Misses many Indian patients).
• New Indian Setting: 11.85 (Catches far more patients).

By lowering the threshold, the test becomes much more sensitive. It's like turning down the volume on a smoke detector so it goes off when there's a little bit of smoke, rather than waiting for a full-blown fire.

---

Why This Matters for You (The "So What?")

1. Better Diagnosis: In India, doctors often struggle to tell the difference between Type 1 and Type 2 diabetes, especially in adults. This genetic test, with the new Indian settings, acts as a second pair of eyes to help make the right call faster.
2. No More "One-Size-Fits-All": This study proves that we cannot just copy-paste medical tools from Europe to India. We need ancestry-aware tools. What works for London doesn't work for Mumbai without adjustment.
3. Helping the "Invisible" Patients: There are many Indians with Type 1 diabetes who don't have the usual antibodies. This genetic score can help identify them, ensuring they get the insulin they need instead of being misdiagnosed with Type 2.

The Bottom Line

Think of this study as calibrating a scale. The European scale was accurate for Europeans, but if you put an Indian on it, it would give the wrong weight. The researchers didn't throw the scale away; they just re-calibrated the numbers so it works perfectly for the Indian population.

This is a huge step forward for precision medicine in India, ensuring that genetic tools are fair, accurate, and life-saving for everyone, regardless of where they were born.

Technical Summary

1. Problem Statement

Type 1 Diabetes (T1D) is an autoimmune disorder with high heritability (~80%), largely driven by the Human Leukocyte Antigen (HLA) region. While Genetic Risk Scores (GRS) based on 67 SNPs have been developed and validated primarily in European populations to distinguish T1D from Type 2 Diabetes (T2D) and non-diabetic controls, their applicability in non-European populations remains limited.

• Specific Gaps in India:
  • Indian populations differ genetically and clinically from Europeans (e.g., different HLA architecture, higher prevalence of islet cell autoantibody-negative T1D, and distinct age-of-onset patterns).
  • Previous validation in India was limited to a single cohort (Western India, Indo-European linguistic group).
  • There is a lack of systematic evaluation of T1D GRS performance across diverse Indian linguistic groups (Indo-European and Dravidian) and clinical settings.
  • It is unknown whether European-derived GRS thresholds accurately classify genetic risk in Indians or if population-specific optimization is required.

2. Methodology

The study employed a multi-centric, cross-sectional design involving rigorous genotyping, imputation, and statistical analysis.

• Study Cohorts:
  • Cases: 597 clinically diagnosed T1D patients from four tertiary care centers across India:
    • Chellaram Diabetes Institute (CDI), Pune (Western India).
    • Tapadia Diagnostic Research Centre (TDRC) & Osmania General Hospital (OSM), Hyderabad (Southern India).
    • JSS Academy of Higher Education & Research, Mysuru (Southern India).
  • Controls: 3,347 non-diabetic controls from two prospective birth cohorts:
    • Mysore Parthenon Cohort (MPC), Mysore (Dravidian linguistic group).
    • Mumbai Maternal Nutrition Project (MMNP), Mumbai (Indo-European linguistic group).
• Phenotyping:
  • Classification into Childhood-onset (<18 years) and **Adult-onset** (>18 years).
  • Measurement of three islet cell autoantibodies (IA): GAD65, IA2, and ZnT8.
  • Random C-peptide measurement to assess beta-cell function.
• Genotyping & QC:
  • DNA extracted from peripheral blood.
  • Genotyped using Illumina Infinium Global Screening Array (GSA) v3.0MD (and v1.0 for some controls).
  • Strict Quality Control (QC): Sample call rate >95%, Hardy-Weinberg equilibrium p < 10⁻⁶, MAF ≥ 0.5%.
  • Imputation: Performed using the NIH TOPMED R3 reference panel. HLA alleles (DQA1/DQB1) were imputed to two-field resolution using HIBAG.
  • GRS Construction: A 67-SNP T1D GRS was calculated, incorporating HLA-DQ haplotypes and non-HLA loci. Proxy SNPs were used where original SNPs were missing or had low imputation quality in South Asian data.
• Statistical Analysis:
  • Discriminative Performance: Assessed via Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC).
  • Threshold Optimization: Optimal cut-offs derived using Youden's Index (maximizing sensitivity + specificity - 1).
  • Comparisons: Wilcoxon rank-sum tests for GRS distribution; Multivariable logistic regression for HLA-diplotype associations (adjusted for sex).
  • Ancestry Analysis: Principal Component Analysis (PCA) to assess population substructure and ensure homogeneity.

3. Key Contributions

1. Multi-Regional Validation: First large-scale validation of the 67-SNP T1D GRS across diverse Indian linguistic groups (Indo-European and Dravidian), confirming its utility as a unified diagnostic tool.
2. Stratified Performance Analysis: Demonstrated that GRS performance is significantly influenced by islet autoantibody status and age of onset, providing genetic insights into T1D heterogeneity in India.
3. Ancestry-Specific Thresholds: Proved that European-derived GRS thresholds systematically under-classify genetic risk in Indians and established India-specific optimal cut-offs.
4. HLA Architecture Insights: Elucidated the distinct roles of HLA-DQ25 and HLA-DQ81 haplotypes in childhood vs. adult-onset T1D within the Indian population.

4. Key Results

• Overall Discriminative Power:

  • The 67-SNP GRS showed consistent performance across all Indian cohorts with a combined ROC-AUC of 0.84 (range: 0.78–0.87).
  • Performance was comparable to the authors' previous single-cohort study, validating the GRS across the country's genetic gradient.

• Impact of Autoantibody Status and Age:

  • IA Status: The GRS performed significantly better in IA-positive patients (AUC 0.85) compared to IA-negative patients (AUC 0.75). The HLA component drove this difference; non-HLA markers performed similarly regardless of antibody status.
  • Age of Onset: The GRS had higher discriminative ability in childhood-onset T1D (AUC 0.84) compared to adult-onset T1D (AUC 0.74).
  • Mean GRS: Adult-onset patients had significantly lower mean GRS scores (12.15) than childhood-onset patients (13.25).

• HLA-DQ Haplotype Analysis:

  • Risk Haplotypes: DQ25 (DQA105:01~DQB102:01) and DQ81 (DQA103:01~DQB103:02) were the most common risk haplotypes.
  • Diplotype Associations:
    • Childhood-onset: Strong enrichment of DQ81-containing diplotypes (e.g., DQ81/DQ81, DQ81/X).
    • Adult-onset: Enrichment of protective diplotypes and lower frequency of DQ81 combinations.
    • Effect Sizes: The compound heterozygous DQ25/DQ81 conferred the highest risk (OR ~72 in childhood, ~17 in adults). DQ25/X had a higher effect size than DQ81/X.

• Threshold Optimization (European vs. Indian):

  • Distribution Shift: The Indian T1D GRS distribution showed a clear leftward shift compared to Europeans.
  • European Cut-off Failure: Using the European 50th centile (GRS = 14.60) classified only 30% of Indian T1D patients as high risk (vs. expected 50%). The European 90th centile (GRS = 16.54) captured only 4% of Indian patients.
  • Optimal Indian Cut-off:
    • Using Youden's index, the optimal threshold for the general Indian cohort was GRS = 11.85, achieving 74.7% sensitivity (vs. 56.4% with the European cut-off of 13.00).
    • For the stringent subgroup (Childhood-onset + IA-positive), the optimal cut-off was 12.24, yielding 75.7% sensitivity (vs. 61.8% with European cut-off).

5. Significance and Clinical Implications

• Diagnostic Utility: The study validates the T1D GRS as a robust tool for distinguishing T1D from T2D and non-diabetic controls in India, particularly in resource-limited settings where autoantibody testing may be unavailable or yield negative results.
• Equity in Genomics: It highlights the critical necessity of ancestry-aware optimization. Applying European thresholds to Indian populations leads to significant under-diagnosis of genetic risk.
• Subtype Differentiation: The findings suggest that adult-onset T1D in India may have a distinct genetic architecture (enriched protective haplotypes) compared to childhood-onset, complicating its differentiation from T2D. The GRS helps stratify these risks.
• Future Directions: The study advocates for the implementation of population-specific GRS thresholds in clinical practice in India and calls for larger studies to identify population-specific risk variants to further refine these tools.

Conclusion: The 67-SNP T1D GRS is a valuable, unified diagnostic metric for India, but its clinical application requires India-specific cut-offs to avoid under-classification. The performance varies by autoantibody status and age, reflecting underlying HLA architectural differences between childhood and adult-onset disease.