The impact of low-frequency genetic variants on serum protein levels

By integrating low-frequency genetic variants with serum protein measurements in a large Icelandic cohort, this study reveals that these variants significantly expand the detection of cis-pQTLs, uncovering widespread allelic heterogeneity and distinct biological characteristics—such as enrichment in coding regions and essential pathways—that are often missed when focusing solely on common variants.

The Gist

Imagine your body is a massive, bustling city. Inside this city, there are millions of tiny workers called proteins. These workers build structures, fix roads, deliver packages, and keep the city running smoothly. The instructions for hiring, firing, and managing these workers are written in a giant library called your DNA.

For a long time, scientists have been studying this library to understand why some people get sick and others stay healthy. However, they mostly looked at the most popular, easy-to-read books in the library—the common genetic variants. These are like the bestsellers that almost everyone has.

This new study is like a detective deciding to look at the rare, dusty books on the top shelves of the library. These are the low-frequency variants. They are much harder to find and fewer people have them, but the researchers suspected they might hold the keys to some of the city's most critical, hidden problems.

Here is what they discovered, broken down into simple concepts:

1. The "7,596" Workers

The researchers used a high-tech scanner (called SomaScan) to take a census of 7,596 different protein workers in the blood of over 5,000 Icelanders. Think of this as taking a headcount of almost every type of worker in the city at once.

2. Finding the Hidden Managers (The Low-Frequency Variants)

In the past, scientists mostly found the managers (genetic variants) that controlled the common workers. But this study asked: What about the managers for the rare, specialized workers?

By looking at the "rare books" (low-frequency variants), they found hundreds of new managers they had never seen before.

• The Analogy: Imagine you only knew the managers of the street sweepers and mail carriers (common variants). This study found the managers for the specialized bomb disposal units and deep-sea divers (rare variants). You didn't know these managers existed until you looked harder.

3. The "Secondary Signals" (The Backup Plans)

The researchers found that for many proteins, there isn't just one manager; there are several.

• The Common Manager: Usually, the main manager is a "common variant" that everyone has. This manager handles the day-to-day stuff.
• The Rare Backup: But often, there are secondary managers (low-frequency variants) that step in when things get complicated. These are like the "special ops" teams that only show up for specific, high-stakes emergencies.
• The Discovery: The study found that these "special ops" managers are often located far away from the protein's main office (the gene), acting like remote control switches that are hard to find unless you have a very detailed map.

4. Who Do These Rare Managers Control?

This is the most fascinating part. The researchers noticed a pattern in which workers these rare managers controlled:

• Common Managers tend to control the "public-facing" workers—like the ones you see on the street (secreted proteins). These are easier to study and often vary a lot between people.
• Rare Managers tend to control the "essential infrastructure" workers—like the power plant operators or the city's core security. These are proteins that are vital for life. If these proteins break, the whole city (the body) is in big trouble. Because they are so important, nature doesn't let them change easily, which is why the genetic variants controlling them are so rare.

5. Why This Matters for Medicine

Why should you care about these dusty books and rare managers?

• Better Drug Targets: If you want to fix a broken city, you need to know who the real decision-makers are. By finding these rare managers, scientists can identify new drug targets that were previously invisible.
• Understanding Disease: Some diseases happen because of a glitch in these "essential infrastructure" proteins. If we only look at the common variants, we miss the root cause of these serious conditions.
• The "Missing Link": The study found that about 155 proteins were only controlled by these rare managers. Without this study, we would have thought these proteins had no genetic control at all. Now we know they are being managed, just by a very exclusive, rare club.

The Bottom Line

Think of this study as upgrading the city's map. Before, the map only showed the main highways (common variants). Now, the researchers have added the secret tunnels, the back alleys, and the emergency routes (low-frequency variants).

This new map shows that the city is much more complex and interconnected than we thought. It reveals that the most critical parts of our biology are guarded by rare genetic switches. By understanding these switches, we can build better medicines and understand diseases in a way we never could before.

Technical Summary

1. Problem Statement

While Genome-Wide Association Studies (GWAS) have successfully identified thousands of common genetic variants associated with disease risk, the functional interpretation of these variants remains challenging, particularly as most reside in non-coding regions. Furthermore, the majority of proteogenomic studies focusing on protein Quantitative Trait Loci (pQTLs) have historically prioritized common variants (Minor Allele Frequency [MAF] > 1%).

This focus leaves a significant gap in understanding the role of low-frequency (LF) variants (MAF 0.1–1%) and rare variants. LF variants often occur in evolutionarily constrained regions and may have larger effect sizes on protein function or regulation. Previous studies suggested that common variants primarily affect peripheral proteins in molecular networks, while "hub" proteins (central to biological pathways) might be regulated by rarer variants. Consequently, a substantial portion of the serum proteome (approx. one-third of measured proteins) lacked detectable common variant cis-pQTLs, limiting the ability to perform Mendelian Randomization (MR) and drug target validation for these critical proteins.

2. Methodology

The study utilized a large-scale, population-based cohort to map the genetic architecture of the circulating proteome with an expanded allele-frequency spectrum.

• Cohort & Data:

  • Population: 5,291 individuals from the AGES-Reykjavik Study (Icelandic elderly cohort).
  • Proteomics: Serum levels of 7,596 proteins (SOMAmers) measured using the SomaScan v4.1 (7k) platform. This expanded upon a previous study using a 5k platform.
  • Genomics: Genotypes were imputed to the TOPMed reference panel (r2), yielding 11.9 million variants (MAF > 0.1%). This represents a significant improvement over previous Haplotype Reference Consortium (HRC) imputation, particularly for LF and rare variants.
  • Validation: Imputation quality was validated against directly genotyped HumanExome BeadChip data, showing near-perfect concordance (~0.999) for LF variants.

• Statistical Analysis:

  • GWAS: Linear models were fitted for each SOMAmer against 11.9M variants, adjusting for age, sex, genotyping platform, and 10 genetic principal components.
  • Significance Thresholding: A permutation-based empirical thresholding approach (using QTLtools) was applied within a ±500 kb cis-window to determine significance for each protein, correcting for multiple testing.
  • Conditional Analysis: Stepwise conditional analysis (forward-backward regression) was performed to identify independent cis-signals, distinguishing primary signals from secondary signals and separating the effects of common vs. LF variants.
  • Replication: Results were compared against external cohorts (deCODE, ARIC-EA, ARIC-AA) and previous AGES analyses to assess replication rates and effect size concordance.
  • Functional Enrichment: Proteins were categorized based on their pQTL status (Common-only, LF-only, or None) and analyzed for enrichment in tissue specificity, secretion, protein-protein interaction (PPI) connectivity, and Loss-of-Function (LoF) intolerance.

3. Key Contributions

• Expanded Proteomic Coverage: Identified cis-pQTLs for 693 novel SOMAmers not previously reported in major proteogenomic studies (deCODE, ARIC), largely due to the expanded SomaScan 7k panel.
• Inclusion of Low-Frequency Variants: Demonstrated that incorporating LF variants (MAF 0.1–1%) significantly increases the number of detected independent genetic signals per protein, revealing widespread allelic heterogeneity.
• Discovery of LF-Exclusive Signals: Identified 155 proteins that are exclusively regulated by independent LF cis-pQTLs, which would have been missed in common-variant-only analyses.
• Characterization of Regulatory Architecture: Provided a detailed map of how variant frequency correlates with genomic location (distal vs. proximal) and functional consequence (coding vs. non-coding).

4. Key Results

• Signal Discovery & Allelic Heterogeneity:

  • The study identified 7,953 independent cis-signals across 2,166 SOMAmers (1,893 unique proteins).
  • 27.4% of these independent signals were driven by LF variants.
  • 49% of proteins with cis-pQTLs harbored at least one independent LF signal.
  • 155 proteins were affected only by LF signals. Among proteins appearing to have a single common variant signal, 28% were reclassified as having multiple independent signals once LF variants were included.

• Effect Sizes and Variance:

  • LF variants exhibited larger median absolute effect sizes (|β| = 1.10) compared to common variants (|β| = 0.24).
  • While common variants generally explained more total variance (median $R^2$ = 0.029 vs. 0.0095 for LF), LF variants contributed more variance than common variants for 339 proteins (15.7%), including metabolic proteins like CEL, SERPINA1, and APOA1.

• Genomic Location and Function:

  • Distal Signals: LF signals were significantly more likely to be distal (median 89 kb from gene boundary) compared to common signals (5 kb).
  • Functional Consequences: Within gene boundaries, LF variants were strongly enriched for coding consequences (e.g., missense, stop-gain) and splice-related annotations. In contrast, common intragenic variants were predominantly non-coding (intronic, UTR).
  • Secondary Signals: LF variants were increasingly represented among higher-ranked secondary signals, suggesting they capture regulatory layers hidden by linkage disequilibrium (LD) with primary common variants.

• Protein Characteristics:

  • Common Variant Proteins: Tended to be secreted, exhibit tissue-specific expression, and be located at the periphery of molecular networks.
  • LF-Only Proteins: Were functionally distinct; they were less tissue-specific, showed higher connectivity in PPI networks (acting as "hubs"), and were less tolerant to Loss-of-Function (LoF) mutations. This suggests they are part of essential, tightly regulated biological pathways.
  • Proteins with secondary signals (often driven by LF variants) were enriched for disease-associated genes but depleted in LoF-intolerant genes.

• Replication:

  • Replication rates for external cohorts (deCODE, ARIC) ranged from 59% to 66%, with >99% concordance in effect direction, confirming the robustness of the findings despite differences in ancestry and sample size.

5. Significance

• Drug Target Prioritization: The study highlights that proteins regulated by LF variants are often central to essential biological pathways and disease mechanisms. Since drug programs targeting rare variants with large effects have higher regulatory approval success rates, this catalog of LF-pQTLs provides a high-value resource for identifying novel therapeutic targets.
• Refining Causal Inference: By expanding the set of genetic instruments (pQTLs) to include LF variants, the study improves the power and precision of Mendelian Randomization studies, allowing for causal inference on proteins previously considered "genetically unexplained."
• Understanding Genetic Architecture: The findings reveal a complex, heterogeneous cis-regulatory architecture where common and low-frequency variants regulate distinct subsets of proteins with different biological properties. This challenges the notion that common variants alone can fully explain the genetic regulation of the proteome.
• Data Resource: The authors have made full genome-wide association summary statistics publicly available, facilitating future proteogenomic research and colocalization analyses.

In conclusion, this study demonstrates that expanding the allele-frequency spectrum beyond common variants is essential for a complete understanding of the genetic regulation of the human proteome, uncovering regulatory mechanisms and biological targets that are invisible to standard GWAS approaches.