Genetic Profiling of Autoimmune Diseases and Exploring Clusters Through Polygenic Risk Score Analysis Using Cohort Data from the UK Biobank

This study leverages UK Biobank and TriNetX cohort data to characterize the genetic overlap, shared biological pathways, and comorbidity patterns across autoimmune diseases through polygenic risk score analysis and pathway mapping, revealing both common and distinct genetic architectures that support the concept of poly-autoimmunity.

The Gist

The Big Picture: A Genetic Detective Story

Imagine the human immune system as a highly trained security guard. Its job is to spot and stop intruders (like viruses or bacteria). In autoimmune diseases, this security guard gets confused and starts attacking the building itself (your own healthy tissues).

This paper is like a massive detective investigation. The researchers used two giant databases—one called the UK Biobank (a library of health and DNA data from half a million people in the UK) and another called TriNetX (a global network of hospital records)—to figure out three main things:

1. Why do some people get multiple autoimmune diseases at once?
2. How are the genetic "blueprints" for these diseases similar or different?
3. Do different databases tell the same story?

1. The "Party" of Diseases (Comorbidity)

The researchers noticed that autoimmune diseases often throw a party together. If you have one, you are statistically more likely to have another. They call this polyautoimmunity.

• The Analogy: Think of autoimmune diseases like different flavors of ice cream. Usually, people stick to one flavor. But this study found that many people are eating a "sundae" with multiple flavors at once.
• The Findings: They looked at 15 different diseases (like Rheumatoid Arthritis, Psoriasis, and Lupus). They found that some pairs are best friends. For example, Crohn's disease and Ulcerative Colitis (both gut issues) almost always show up together, like peanut butter and jelly. Multiple Sclerosis and Lupus also hang out together often.
• The Twist: However, the "flavor" of the party depends on where you look. When they compared the UK data with the global hospital data, some pairs that seemed like best friends in the UK looked like strangers in the global data. This suggests that how we define and record these diseases in different hospitals can change the story.

2. The Genetic "Risk Score" (PGS)

To understand why these diseases cluster, the researchers looked at people's DNA. They created a Polygenic Risk Score (PGS).

• The Analogy: Imagine your DNA is a deck of cards. Some cards are "bad" for your immune system. A PGS is like a scorecard that counts how many "bad" cards you have.
  • A high score means you have a lot of "bad" cards and are at higher risk.
  • A low score means you have fewer "bad" cards.
• The Findings:
  • Shared Risk: People with high scores for one disease often had high scores for another. This explains why diseases cluster; they share some of the same "bad cards" in the deck.
  • Unique Risk: But, the scores weren't identical. Some people had high scores for Psoriasis but low scores for Lupus. This means the diseases have their own unique "bad cards" too.
  • The "Opposites" Effect: Interestingly, for some pairs (like Multiple Sclerosis and Psoriasis), having a high risk for one actually seemed to lower the risk for the other. It's like having a genetic "shield" against one disease that accidentally makes you more vulnerable to another.

3. The "HLA" Neighborhood (The Genetic Hotspot)

The researchers had to be very careful about a specific area of the DNA called the HLA region.

• The Analogy: Think of the HLA region as a very crowded, noisy city block where everyone knows everyone. It's so complex that it's hard to tell which specific house (gene) is causing the problem.
• The Strategy: To get a clear view, the researchers temporarily "closed off" this city block in their analysis.
• The Result: When they looked at the rest of the DNA, they found that some diseases (like Lupus and Rheumatoid Arthritis) relied heavily on that crowded city block. Others (like Psoriasis and Type 1 Diabetes) had strong signals even outside that block. This tells us that different diseases have different "genetic engines."

4. Finding the "Suspects" (Genes and Pathways)

The team used a computer tool to filter through millions of DNA variations to find the specific "suspects" (genes) responsible.

• The Common Suspects: They found 14 genes that appeared across multiple diseases. Some were famous suspects already known to science (like PTPN22 and IL23R).
• The New Suspects: They also found some new names on the list (like ZNF322 and BTN1A1) that hadn't been strongly linked to autoimmune diseases before.
• The Network: They didn't just look at single genes; they looked at how these genes talk to each other. They found that in some diseases, the immune system is "over-activated" (like a fire alarm going off constantly), while in others, it's "under-activated" or suppressed.

5. The "Two Libraries" Problem (UKB vs. TriNetX)

Finally, the researchers compared their findings from the UK Biobank against the global TriNetX database.

• The Analogy: Imagine two librarians describing the same book. One librarian (UKB) is very detailed and strict about how they categorize books. The other (TriNetX) has a much larger collection but uses slightly different labels.
• The Conflict: Sometimes, the librarians agreed perfectly (e.g., Crohn's and Colitis are always linked). But sometimes, they disagreed. For example, a disease pair might look like a strong match in the UK library but a weak match in the global library.
• The Lesson: This doesn't mean one is wrong; it means that how we collect data matters. Differences in how patients are diagnosed, recorded, or even the demographics of the people in the database can change the results.

Summary

This paper is a massive map of the genetic landscape of autoimmune diseases. It confirms that:

1. These diseases are often related and share genetic "bad cards."
2. Some diseases are more similar to each other than others.
3. We need to be careful about how we group and record these diseases, because different databases can tell slightly different stories.
4. There are both known and brand-new genetic clues that help explain why our immune systems sometimes turn on us.

The study stops at mapping these connections and identifying the genes; it does not claim to have found a cure or a new treatment, but rather provides a clearer picture of the puzzle pieces involved.

Technical Summary

Technical Summary: Genetic Profiling of Autoimmune Diseases and Exploring Clusters Through Polygenic Risk Score Analysis Using Cohort Data from the UK Biobank

Problem Statement
Autoimmune diseases (ADs) arise from immune responses against self-antigens, characterized by significant phenotypic diversity and a high rate of comorbidity (polyautoimmunity). While Genome-Wide Association Studies (GWAS) have identified susceptibility loci, challenges remain in linking these variants to biological mechanisms, understanding the shared genetic architecture across different ADs, and accounting for the contributions of both common and rare variants. Furthermore, existing Polygenic Risk Score (PGS) models often lack comprehensive frameworks that integrate allele frequency with functional annotations, and cross-cohort comparisons of comorbidity patterns are hindered by inconsistent data harmonization between large biobanks and real-world electronic health record (EHR) networks. This study addresses the need to characterize disease overlap, genetic heterogeneity, and shared biological mechanisms across autoimmune conditions using integrated phenotypic and genetic data.

Methodology
The study employed a multi-faceted analytical framework integrating data from the UK Biobank (UKB) and the TriNetX (TNX) network:

1. Phenotypic and Comorbidity Analysis:

   • Datasets: Phenotypic data from 502,371 UKB participants and EHR data from 71,069,654 White individuals in TNX were utilized.
   • Disease Scope: Fifteen ADs were analyzed, including Crohn's disease (CD), multiple sclerosis (MS), psoriasis (PSO), rheumatoid arthritis (RA), systemic lupus erythematosus (SLE), ulcerative colitis (UC), and others.
   • Statistical Approach: Odds ratios (ORs) were calculated to assess diagnostic overlap and comorbidity patterns. Fisher's exact test was used for p-value determination. Propensity score matching was applied in TNX to control for covariates (sex, age, ethnicity).

2. Polygenic Risk Score (PGS) Correlation:

   • Cohorts: Two PGS datasets were analyzed: a standard dataset (485,989 samples, 8 ADs) and an enhanced dataset (104,544 samples, 6 ADs).
   • Analysis: Pearson's correlation coefficients were computed to evaluate shared genetic architectures and pleiotropy between paired ADs. Two-dimensional scatter plots visualized risk score distributions and clustering patterns.

3. Genetic Variant Identification and Functional Annotation:

   • Variant Filtering: Using unrelated White British UKB participants, significant variants were identified with a genome-wide threshold ($P < 5 \times 10^{-8}$). The HLA region (chr6: 27.4–34.4 Mb) was excluded to reduce complexity.
   • Annotation: Variants were mapped to GRCh38 and annotated using the Variant Effect Predictor (VEP).
   • Stratification: Variants were filtered by:
     • Allele Frequency (AF): Common (AF $\ge$ 0.05) vs. Rare (AF < 0.05).
     • Functional Impact: High/Moderate impact (e.g., missense, splice site) vs. Low/Modifier impact.
     • Locus Grouping: Variants within 500 kb were grouped to define loci.
   • Gene Selection: Genes associated with variants across these categories were cross-referenced to identify shared genetic drivers.

4. Network Propagation and Pathway Analysis:

   • Method: A network propagation strategy using the STRING v12 interaction network (edges $\ge$ 700) and a regularized Laplacian kernel was applied to genes shared across ADs.
   • Pathway Mapping: Results were mapped to Hallmark gene sets from the Molecular Signatures Database (MSigDB) to identify enriched biological pathways (e.g., immune response, metabolism).

Key Results

• Comorbidity and Phenotypic Overlap:

  • Significant sample overlap was observed, particularly between PSO and RA (993 shared samples in UKB) and UC and CD.
  • Strong positive associations (OR > 1) were found for several pairs, such as SLE and RA, and CD and UC.
  • Cross-cohort comparison (UKB vs. TNX) revealed consistent trends for strong associations (e.g., MS and other demyelinating diseases) but notable discrepancies in effect sizes and direction for specific pairs (e.g., PSO and MS showed a positive OR in UKB but a negative/inverse association in TNX).

• PGS Distribution and Genetic Architecture:

  • PGS analysis revealed both shared and distinct genetic architectures.
  • MS vs. PSO/RA: MS showed a distinct distribution with a positive trend on the y-axis, while PSO and RA trended positively on the x-axis, indicating partial genetic separation.
  • SLE vs. MS: A positive linear correlation was observed, suggesting shared genetic susceptibility.
  • SLE vs. PSO: These diseases exhibited near-orthogonal mean PGS locations, indicating minimal genetic similarity despite some clinical overlap.
  • Comorbid Cases: Individuals with dual diagnoses (e.g., PSO and RA) generally clustered in the upper-right quadrant, indicating elevated genetic risk for both conditions, supporting the concept of polyautoimmunity.

• Variant and Gene Discovery:

  • After HLA exclusion, 14 shared genes were identified across the analyzed ADs.
  • Impact-Based Filtering: 11 of the 14 shared genes were identified via high/moderate impact variants, highlighting the importance of functional consequence over frequency alone.
  • Novel Associations: The study identified novel gene-disease associations, including ZNF391, POM121L2, and LINC02356, which were not previously linked to ADs in the GWAS Catalog.
  • Known Genes: Re-identification of established genes (IL23R, PTPN22, SH2B3, MAGI3) validated the methodology.
  • Butyrophilin Family: Multiple BTN family genes (BTN1A1, BTN2A1, BTN3A1, BTN3A2) were found across CED, SLE, and T1D, suggesting a broader immunomodulatory role.

• Pathway Enrichment:

  • Network propagation identified distinct pathway profiles. The IL6-JAK-STAT3 signaling pathway was upregulated in CED, CD, T1D, and UC but downregulated in PSO.
  • PSO showed negative enrichment in immune and metabolic pathways, suggesting pathway suppression, whereas CED and UC showed broad activation of interconnected pathways.

Significance and Claims
The paper claims to provide a harmonized, data-driven resource that links genetic architecture, comorbidity patterns, and methodological robustness. Key contributions include:

1. Validation of Polyautoimmunity: The study supports the existence of partial genetic overlap rather than complete convergence among ADs, demonstrating that comorbidity often arises from moderate overlapping susceptibility rather than extreme risk in a single condition.
2. Methodological Integration: By integrating common, rare, and functionally impactful variants, the study identifies both known and novel gene associations, demonstrating that impact-based filtering captures biologically meaningful signals often missed by frequency-based GWAS alone.
3. Cross-Dataset Insights: The comparison between UKB and TNX highlights the critical impact of dataset-specific factors (e.g., coding inconsistencies, cohort composition) on comorbidity patterns, emphasizing the need for standardized approaches in autoimmune research.
4. Sex as a Biological Variable: The results reinforce sex as a critical covariate, with consistent female predominance in most ADs (e.g., SLE, RA, MS) and male predominance in ankylosing spondylitis, likely driven by hormonal and X-linked genetic factors.
5. Functional Context: The network propagation analysis adds a functional dimension to genetic associations, distinguishing between systemic dysregulation (e.g., in CED/UC) and localized or repressive network effects (e.g., in PSO).

The authors conclude that while their work delineates shared and distinct features of ADs, future progress requires diverse, longitudinal cohorts, standardized diagnostic definitions, and experimental validation to fully translate these genetic insights into precision medicine.