Rare variant associations with plasma protein levels in the UK Biobank

The influence of rare variants on the plasma proteome is largely unknown because previous rare variant proteogenomic studies have been limited in scale.

“However, their influence on the plasma proteome is largely unknown because previous rare variant proteogenomic studies have been limited in scale9,10.”

• cites [9] Kierczak, M. et al. Contribution of rare whole-genome sequencing variants to plasma protein levels and the missing heritability. Nat. Commun. 13, 2532 (2022). sourceSupported
• cites [10] Katz, D. H. et al. Whole genome sequence analysis of the plasma proteome in Black adults provides novel insights into cardiovascular disease. Circulation 145, 357–370 (2022). sourcePartly supported

  The Introduction (page unavailable) says earlier proteogenomic findings came nearly entirely from European populations using genotyping arrays, and the Discussion (page unavailable) describes this study as the largest of its kind in a Black population. This supports a gap in knowledge about genetic determinants of the plasma proteome, but the paper does not say that previous rare-variant proteogenomic studies were limited in scale or identify that as the reason rare variants’ influence was largely unknown. (LLM, conf=0.77)

  Experimental — accuracy not yet measured

Most pQTLs to date have been discovered via genome-wide association studies, which predominantly focus on common variants.

“To date, most pQTLs have been discovered via genome-wide association studies (GWASs), which predominantly focus on common variants3,5–8.”

• cites [3] Pietzner, M. et al. Mapping the proteo-genomic convergence of human diseases. Science 374, eabj1541 (2021). sourcePartly supported

  The paper describes its own genome–proteome-wide association study as identifying pQTLs (Abstract) and says it concentrated on common genetic variation (Discussion; page unavailable). It does not establish the broader claim that most pQTLs discovered to date came from GWASs. (LLM, conf=0.84)

  Experimental — accuracy not yet measured
• cites [5] Sun, B. B. et al. Plasma proteomic associations with genetics and health in the UK Biobank. Nature https://doi.org/10.1038/s41586-023-06592-6 (2023). sourceSupported
• cites [6] Sun, B. B. et al. Genomic atlas of the human plasma proteome. Nature 558, 73–79 (2018). sourceSupported
• cites [7] Folkersen, L. et al. Genomic and drug target evaluation of 90 cardiovascular proteins in 30,931 individuals. Nat. Metab. 2, 1135–1148 (2020). sourcePartly supported

  The Results section, “Genome-wide meta-analysis of 90 proteins” (page unavailable), reports discovering hundreds of pQTLs through a GWAS meta-analysis, supporting GWAS as a pQTL discovery method. The paper does not establish that most pQTLs discovered to date came from GWAS or that GWAS predominantly focus on common variants; its Online Methods instead describe filtering by a minimum minor-allele count. (LLM, conf=0.79)

  Experimental — accuracy not yet measured
• cites [8] Ferkingstad, E. et al. Large-scale integration of the plasma proteome with genetics and disease. Nat. Genet. 53, 1712–1721 (2021). sourceNot checked — no full text of the source

In a separate array-based GWAS on the UKB-PPP cohort, testing common variants (MAF > 0.1%) and 2,922 protein assays resulted in 14,287 primary genetic associations.

“In a separate array-based GWAS on the UKB-PPP cohort, we tested for the association between common variants (MAF > 0.1%) and 2,922 protein assays, resulting in 14,287 primary genetic associations5.”

• cites [5] Sun, B. B. et al. Plasma proteomic associations with genetics and health in the UK Biobank. Nature https://doi.org/10.1038/s41586-023-06592-6 (2023). sourcePartly supported

  The 'Discovery of pQTLs' section reports testing 2,922 proteins and finding 14,287 primary associations, but those results came from a European-ancestry discovery subset of 34,557 participants, not the full UKB-PPP cohort. The 'Replication of pQTLs' section reports 23,588 primary associations for the full cohort. The methods describe imputed variants and a minor-allele-count filter, not the claimed MAF > 0.1% cutoff. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

Using an n-of-one permutation analysis, P ≤ 1 × 10−8 was identified as an appropriate P value threshold.

“Using an n-of-one permutation analysis as previously described11, we identified P ≤ 1 × 10−8 as an appropriate P value threshold (Methods and Supplementary Table 3).”

• cites [11] Wang, Q. et al. Rare variant contribution to human disease in 281,104 UK Biobank exomes. Nature 597, 527–532 (2021). sourceSupported

The effect sizes of nominally significant ExWAS pQTLs (P < 1 × 10−4) strongly correlated with GWAS-derived pQTLs (r2 = 0.96).

“The effect sizes (β) of nominally significant ExWAS pQTLs (P < 1 × 10−4) strongly correlated with the GWAS-derived pQTLs5 (r2 = 0.96; Supplementary Fig. 2).”

• cites [5] Sun, B. B. et al. Plasma proteomic associations with genetics and health in the UK Biobank. Nature https://doi.org/10.1038/s41586-023-06592-6 (2023). sourceSupported

ASGR1 also emerged as a hotspot in the UKB-PPP GWAS and other large pQTL studies.

“ASGR1, which encodes a subunit of the asialoglycoprotein receptor, also emerged as a hotspot in the UKB-PPP GWAS and other large pQTL studies5,6,8.”

• cites [5] Sun, B. B. et al. Plasma proteomic associations with genetics and health in the UK Biobank. Nature https://doi.org/10.1038/s41586-023-06592-6 (2023). sourceSupported
• cites [6] Sun, B. B. et al. Genomic atlas of the human plasma proteome. Nature 558, 73–79 (2018). sourceCould not verify

  The ‘trans pQTLs identify pathways to disease’ section describes twelve hotspot regions and names examples, but does not identify ASGR1 as one. Searches for ASGR1 and the asialoglycoprotein receptor found no passage in our copy establishing the claimed ASGR1 hotspot. (LLM, conf=0.72)

  Experimental — accuracy not yet measured
• cites [8] Ferkingstad, E. et al. Large-scale integration of the plasma proteome with genetics and disease. Nat. Genet. 53, 1712–1721 (2021). sourceNot checked — no full text of the source

The HBB-SLC16A1 association likely reflects the enrichment of HBB PTVs in individuals of South Asian ancestries.

“This likely reflects the enrichment of HBB PTVs in individuals of South Asian ancestries, as observed in our UKB PheWAS11.”

• cites [11] Wang, Q. et al. Rare variant contribution to human disease in 281,104 UK Biobank exomes. Nature 597, 527–532 (2021). sourcePartly supported

  The Pan-ancestry collapsing analysis reports a significant association between HBB protein-truncating variants and thalassaemia in participants of South Asian ancestry. It does not report that HBB PTVs are enriched in that group relative to other ancestries or connect them to an HBB–SLC16A1 association, so it supports only part of the attributed explanation. (LLM, conf=0.83)

  Experimental — accuracy not yet measured

There was a significant enrichment of ClinVar pathogenic and likely pathogenic variants among the 1,484 significant rare missense and in-frame indel cis-CDS pQTLs (observed: 3.5%; expected: 0.54%; two-tail binomial P = 2.5 × 10−25).

“We found a significant enrichment of ClinVar15 pathogenic and likely pathogenic variants among the 1,484 significant rare missense and in-frame indel cis-CDS pQTLs (observed: 3.5%; expected: 0.54%; two-tail binomial P = 2.5 × 10−25; Extended Data Fig. 3a).”

• cites [15] Landrum, M. J. et al. ClinVar: improving access to variant interpretations and supporting evidence. Nucleic Acids Res. 46, D1062–D1067 (2018). sourceSupported

PTVs in EPOR were associated with increased erythropoietin (ptv model; P = 3.5 × 10−24; β = 1.78; 95% CI, 1.43 to 2.12).

“For example, PTVs in EPOR, encoding the erythropoietin receptor, were associated with increased erythropoietin, an example of compensatory upregulation (‘ptv’ model; P = 3.5 × 10−24; β = 1.78; 95% confidence interval (CI), 1.43 to 2.12)28.”

• cites [28] Harlow, C. E. et al. Identification and single-base gene-editing functional validation of a cis-EPO variant as a genetic predictor for EPO-increasing therapies. Am. J. Hum. Genet. 109, 1638–1652 (2022). sourceCould not verify

  The Results (Table 1; page unavailable) report an association between a cis-EPO variant and circulating EPO levels, not an analysis of protein-truncating variants in EPOR. Searches of our copy for EPOR, protein-truncating variants, compensatory upregulation, and the stated P value found no passage reporting the claimed association or statistics. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

Ten of the 20 PTVs associated with increased protein abundances (50%) were predicted to escape nonsense-mediated decay because they occurred within the last exon, the penultimate exon within 50 bp of the 3′ junction or the first exon within the first 200 nucleotides of the coding sequence.

“Ten of the 20 PTVs associated with increased protein abundances (50%) were predicted to escape nonsense-mediated decay because they occurred within the last exon, the penultimate exon within 50 base pairs (bp) of the 3′ junction or the first exon within the first 200 nucleotides of the coding sequence12.”

• cites [12] Coban-Akdemir, Z. et al. Identifying genes whose mutant transcripts cause dominant disease traits by potential gain-of-function alleles. Am. J. Hum. Genet. 103, 171–187 (2018). sourceCould not verify

  The supplied passage has no section or page label and contains only links, not findings about protein abundance or nonsense-mediated decay. Searches for the 10-of-20 PTV count and the stated exon-position criteria returned the same links, so our copy does not verify the claim. (LLM, conf=0.75)

  Experimental — accuracy not yet measured

There were 22 trans-CDS associations between known ligand-receptor pairs.

“There were also 22 trans-CDS associations between known ligand–receptor pairs37 (Supplementary Table 11).”

• cites [37] Ramilowski, J. A. et al. A draft network of ligand-receptor-mediated multicellular signalling in human. Nat. Commun. 6, 7866 (2015). sourceSupported

There were only three significant gene-level cis-CDS pQTLs in the synonymous collapsing model.

“Moreover, there were only three significant gene-level cis-CDS pQTLs in the synonymous (syn) collapsing model19,20 (full definitions of collapsing models can be found in Extended Data Table 1).”

• cites [19] Shen, X., Song, S., Li, C. & Zhang, J. Synonymous mutations in representative yeast genes are mostly strongly non-neutral. Nature 606, 725–731 (2022). sourceCould not verify

  The Extended Data passages discuss effects of synonymous mutations on yeast fitness and mRNA levels, not gene-level cis-CDS pQTLs. Searches for “cis-CDS pQTLs,” “synonymous collapsing model,” and “pQTL” found no passage reporting three significant associations. (LLM, conf=0.76)

  Experimental — accuracy not yet measured
• cites [20] Dhindsa, R. S. et al. A minimal role for synonymous variation in human disease. Am. J. Hum. Genet. 109, 2105–2109 (2022). sourcePartly supported

  The Methods section describes a synonymous-variant gene-based collapsing model, supporting the model cited in the sentence. But the reported results concern UK Biobank clinical phenotypes, not cis-CDS pQTLs; the paper does not support the count of three significant pQTLs. (LLM, conf=0.82)

  Experimental — accuracy not yet measured

The 36,466 missense variants and in-frame indels that were not associated with cis changes in protein abundances (P > 1 × 10−4) were significantly depleted of ClinVar pathogenic/likely pathogenic variants (observed: 0.4%; expected: 0.54%; two-tail binomial P = 1.1 × 10−4).

“By contrast, the 36,466 missense variants and in-frame indels that were not associated with cis changes in protein abundances (P > 1 × 10−4) were significantly depleted of ClinVar pathogenic/likely pathogenic variants (observed: 0.4%; expected: 0.54%; two-tail binomial P = 1.1 × 10−4).”

• cites [15] Landrum, M. J. et al. ClinVar: improving access to variant interpretations and supporting evidence. Nucleic Acids Res. 46, D1062–D1067 (2018). sourcePartly supported

  Split into 2 atoms — 0 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

Most rare variant pQTLs (MAF ≤ 0.1%) were not detected in the common variant analysis on the same cohort.

“most rare variant pQTLs (MAF ≤ 0.1%) were not detected in the common variant analysis on the same cohort5.”

• cites [5] Sun, B. B. et al. Plasma proteomic associations with genetics and health in the UK Biobank. Nature https://doi.org/10.1038/s41586-023-06592-6 (2023). sourceCould not verify

  The Discovery of pQTLs section reports associations with MAF below 1% and 0.5%, but does not compare pQTLs at MAF ≤ 0.1% with a common-variant analysis of the same cohort. I searched for that threshold and comparison; the Non-European pQTL mapping section discusses a different comparison, across ancestry groups. (LLM, conf=0.72)

  Experimental — accuracy not yet measured

1,487 (51%) distinct genes carried at least one of these 5,025 missense variants.

“In total, 1,487 (51%) distinct genes carried at least one of these 5,025 missense variants11.”

• cites [11] Wang, Q. et al. Rare variant contribution to human disease in 281,104 UK Biobank exomes. Nature 597, 527–532 (2021). sourceCould not verify

  The Variant-level associations section reports 326 significant missense variants, and the Methods section describes aggregating variants by gene, but neither gives the claimed count of 5,025 variants or 1,487 genes (51%). Searches for both numbers and for the gene-level proportion found no passage substantiating the claim in our copy. (LLM, conf=0.75)

  Experimental — accuracy not yet measured

Many of the downregulated proteins in TET2-CH are markers of other haematopoietic lineages, such as KIR2DL2 and KIR2DL3, and PRG2 and PRG3.

“Many of the downregulated proteins are markers of other haematopoietic lineages, such as KIR2DL2 and KIR2DL3 (natural killer cell activation)52, and PRG2 and PRG3 (constituents of eosinophil granules also involved in basophil stimulation)53,54.”

• cites [52] Colonna, M. & Samaridis, J. Cloning of immunoglobulin-superfamily members associated with HLA-C and HLA-B recognition by human natural killer cells. Science 268, 405–408 (1995). sourceNot checked — no full text of the source
• cites [53] O’Donnell, M. C., Ackerman, S. J., Gleich, G. J. & Thomas, L. L. Activation of basophil and mast cell histamine release by eosinophil granule major basic protein. J. Exp. Med. 157, 1981–1991 (1983). sourcePartly supported

  The introduction (p. 1981) identifies major basic protein as a constituent of eosinophil granules, and the discussion (p. 1988–1989) reports its activation of human basophils. The cited paper does not identify PRG2 or PRG3 by those names or establish the stated role for both proteins; the downregulation finding belongs to the citing study, not this citation. (LLM, conf=0.84)

  Experimental — accuracy not yet measured
• cites [54] Plager, D. A. et al. Comparative structure, proximal promoter elements, and chromosome location of the human eosinophil major basic protein genes. Genomics 71, 271–281 (2001). sourceNot checked — no full text of the source

Three of the JAK2 trans-CDS pQTLs are proteins involved in the integrin β2 pathway: FCGR2A, GP1BA and ICAM2.

“Three of the JAK2 trans-CDS pQTLs are proteins involved in the integrin β2 pathway: FCGR2A, GP1BA and ICAM2 (ref. 45) (Fig. 3a).”

• cites [45] Schaefer, C. F. et al. PID: the Pathway Interaction Database. Nucleic Acids Res. 37, D674–679 (2009). sourceCould not verify

  The Introduction describes PID as a database of human signaling and regulatory pathways, and the Web Interface and Application section explains how to retrieve a pathway’s participating molecules (page numbers unavailable). Our copy does not identify FCGR2A, GP1BA or ICAM2 as members of an integrin β2 pathway, or discuss JAK2 trans-CDS pQTLs. Searches for those protein names, integrin β2 and JAK2 did not find a passage establishing the claim. (LLM, conf=0.74)

  Experimental — accuracy not yet measured

The three other proteins increased in abundance in TET2-CH were CD1C, CLEC4C and CD86.

“The three other proteins increased in abundance in TET2-CH were CD1C (a marker of conventional dendritic cells), CLEC4C (a marker of plasmacytoid dendritic cells) and CD86 (a marker of dendritic cells, monocytes and other antigen-presenting cells)51.”

• cites [51] Collin, M. & Bigley, V. Human dendritic cell subsets: an update. Immunology 154, 3–20 (2018). sourcePartly supported

  The cDC2 Phenotype and distribution section identifies CD1c as expressed by conventional dendritic cells, and the plasmacytoid dendritic cell Phenotype and distribution section explicitly lists CLEC4C as a marker. The copy does not substantiate the CD86 marker description (searched: CD86). The citation appears to support the marker descriptions, not the citing authors’ TET2-CH abundance finding. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

There were no significant associations between these three NLRC4 variants and clinically relevant phenotypes in the published PheWAS of 470,000 UKB exomes.

“Interestingly, there were no significant associations between these three variants and clinically relevant phenotypes in our published PheWAS of 470,000 UKB exomes (https://azphewas.com)11, suggesting that pharmacologic inhibition of NLRC4 may be safe.”

• cites [11] Wang, Q. et al. Rare variant contribution to human disease in 281,104 UK Biobank exomes. Nature 597, 527–532 (2021). sourcePartly supported

  The Main section describes a UK Biobank exome PheWAS, and the Discussion says its results are available at azphewas.com. But the paper reports 281,104 participants, not 470,000, and our copy does not report whether the three NLRC4 variants had significant associations with clinically relevant phenotypes (searched: NLRC4). (LLM, conf=0.86)

  Experimental — accuracy not yet measured

29% of genes with at least one trans-CDS association in the collapsing analyses (n = 73 of 254) are targets of currently approved drugs listed in DrugBank (expected: 14%; two-tailed binomial P = 8.7 × 10−10).

“For example, 29% of genes with at least one trans-CDS association in the collapsing analyses (n = 73 of 254) are targets of currently approved drugs listed in DrugBank (expected: 14%; two-tailed binomial P = 8.7 × 10−10)41.”

• cites [41] Wishart, D. S. et al. DrugBank: a comprehensive resource for in silico drug discovery and exploration. Nucleic Acids Res. 34, D668–672 (2006). sourceSupported

The effect sizes of the URmtr-derived cis-CDS pQTLs (mean |β| = 1.62) were significantly larger than the UR ones (mean |β| = 1.45; Wilcoxon P = 1.2 × 10−5).

“The effect sizes of the ‘URmtr’-derived cis-CDS pQTLs (mean |β| = 1.62) were significantly larger than the ‘UR’ ones (mean |β| = 1.45; Wilcoxon P = 1.2 × 10−5) (Fig. 2c).”

There was strong directional concordance between ExWAS-derived pQTLs and pQTLs detected in an independent Icelandic population.

“There was also strong directional concordance between the ExWAS-derived pQTLs and pQTLs detected in an independent Icelandic population8, which had increased resolution for rarer variants (MAF > 0.01%; Supplementary Note).”

• cites [8] Ferkingstad, E. et al. Large-scale integration of the plasma proteome with genetics and disease. Nat. Genet. 53, 1712–1721 (2021). sourceNot checked — no full text of the source

Four of the 20 PTVs associated with increased protein abundances were annotated as loss of splice donor or acceptor sites.

“Four were annotated as loss of splice donor or acceptor sites, which can have more variable effects than frameshift and nonsense variants.”

20 of the 77 (26%) proteins associated with STAB1 variants were also associated with STAB2 variants.

“Interestingly, 20 of the 77 (26%) proteins associated with STAB1 variants were also associated with STAB2 variants.”

One missense variant in GOT1 had a suggestive association with reduced aspartate aminotransferase in the proteogenomic ExWAS.

“One missense variant in GOT1 had a suggestive association with reduced aspartate aminotransferase in the proteogenomic ExWAS.”

99% (n = 687 of 691) of the cis signals from the ptv model were associated with decreased protein levels.

“As expected, 99% (n = 687 of 691) of the cis signals from the ptv model were associated with decreased protein levels (Fig. 2c).”

Only 19% of the rare (MAF ≤ 0.1%) associations from our ExWAS were significant in the UKB-PPP GWAS.

“As expected, given the constraints of GWAS, only 19% of the rare (MAF ≤ 0.1%) associations from our ExWAS were significant in the UKB-PPP GWAS.”

The rare significant (P ≤ 1 × 10−8) and suggestive (1 × 10−8 < P ≤ 1 × 10−4) missense cis-CDS pQTLs were more likely to be predicted damaging by REVEL.

“Moreover, the rare significant (P ≤ 1 × 10−8) and suggestive (1 × 10−8 < P ≤ 1 × 10−4) missense cis-CDS pQTLs were more likely to be predicted damaging by the in silico predictor, REVEL16 (Extended Data Fig. 3b).”

• cites [16] Ioannidis, N. M. et al. REVEL: an ensemble method for predicting the pathogenicity of rare missense variants. Am. J. Hum. Genet. 99, 877–885 (2016). sourceNot checked — no full text of the source

For 47 of the 50 genes with at least five rare missense cis-CDS pQTLs (94%), at least 75% of the significant missense pQTLs were associated with decreased protein abundance.

“For 47 of these 50 genes (94%), at least 75% of the significant missense pQTLs were associated with decreased protein abundance.”

All six PCSK9 cis-CDS pQTLs (P < 1 × 10−4) were also significantly (P ≤ 1 × 10−8) associated with low-density lipoprotein levels in the expected direction.

“All six variants were also significantly (P ≤ 1 × 10−8) associated with low-density lipoprotein levels in the expected direction (Supplementary Table 7).”

There was a nearly significant association between PTVs in PSAP and increased cathepsin D (P = 9.0 × 10−8; β = 1.61; 95% CI, 1.02 to 2.20).

“For example, there were significant associations between PTVs in PSAP and increased progranulin (GRN; P = 5.5 × 10−17; β = 2.60; 95% CI, 1.99 to 3.21) and cathepsin B (P = 1.2 × 10−11; β = 2.10; 95% CI, 1.49 to 2.70), and a nearly significant association with increased cathepsin D (P = 9.0 × 10−8; β = 1.61; 95% CI, 1.02 to 2.20) (Supplementary Table 8).”

Among 1,084 gene-protein associations specifically arising from the ptv model, 667 (62%) were not significant in the ExWAS.

“Among 1,084 gene–protein associations specifically arising from the ptv model, 667 (62%) were not significant in the ExWAS.”

Mutations in FLT3 were associated with increased levels of the FLT3 ligand (FLT3LG) (ptvraredmg model; P = 1.2 × 10−22; β = 0.85; 95% CI, 0.68 to 1.02).

“Likewise, we identified a trans association between mutations in FLT3, encoding the fms-related tyrosine kinase 3, and increased levels of the FLT3 ligand (FLT3LG; ‘ptvraredmg’ model; P = 1.2 × 10−22; β = 0.85; 95% CI, 0.68 to 1.02).”

501 (25.5%) of the 1,962 gene-protein abundance signals identified in the collapsing analysis did not reach significance in the ExWAS.

“Notably, 501 (25.5%) of the 1,962 gene–protein abundance signals identified in the collapsing analysis did not reach significance in the ExWAS (Extended Data Fig. 4a).”

There were significant associations between PTVs in PSAP and increased progranulin (GRN; P = 5.5 × 10−17; β = 2.60; 95% CI, 1.99 to 3.21) and cathepsin B (P = 1.2 × 10−11; β = 2.10; 95% CI, 1.49 to 2.70).

“For example, there were significant associations between PTVs in PSAP and increased progranulin (GRN; P = 5.5 × 10−17; β = 2.60; 95% CI, 1.99 to 3.21) and cathepsin B (P = 1.2 × 10−11; β = 2.10; 95% CI, 1.49 to 2.70), and a nearly significant association with increased cathepsin D (P = 9.0 × 10−8; β = 1.61; 95% CI, 1.02 to 2.20) (Supplementary Table 8).”

Among common cis-CDS pQTLs, the proportions of PTVs and missense, synonymous and non-coding variants closely matched the proportions observed for the total variants included in the ExWAS.

“Moreover, among the common cis-CDS pQTLs, the proportions of PTVs and missense, synonymous and non-coding variants closely matched the proportions observed for the total variants included in the ExWAS (that is, the expected null distribution; Fig. 1c).”

Non-synonymous variants in TSHR were associated with increased thyroid stimulating hormone (TSHB) (flexdmg model; P = 1.6 × 10−31; β = 0.66; 95% CI, 0.55 to 0.76).

“For example, there was a significant association between non-synonymous variants in TSHR, encoding the thyroid stimulating hormone receptor, and increased thyroid stimulating hormone (TSHB) (‘flexdmg’ model; P = 1.6 × 10−31; β = 0.66; 95% CI, 0.55 to 0.76).”

The percentage of protein-lowering missense variants in the remaining three genes ranged from 14% to 50%.

“The percentage of protein-lowering missense variants in the remaining three genes ranged from 14% to 50% (Supplementary Table 2).”

In total, there were 7,412 significant associations across the nine non-synonymous QV models.

“In total, there were 7,412 significant associations across the nine non-synonymous QV models (Fig. 2a).”

PTVs in ATM were associated with increased levels of alpha-fetoprotein (P = 1.8 × 10−9, β = 0.25, 95% CI, 0.17 to 0.33).

“Another example that only became significant in the pan-ancestry analysis was the trans association between PTVs in ATM, associated with ataxia telangiectasia and several cancers, and increased levels of alpha-fetoprotein (P = 1.8 × 10−9, β = 0.25, 95% CI, 0.17 to 0.33)31.”

• cites [31] Waldmann, T. A. & McIntire, K. R. Serum-α-fetoprotein levels in patients with ataxia-telangiectasia. Lancet 2, 1112–1115 (1972). sourceNot checked — no full text of the source

PTVs and missense variants encompassed a significantly larger percentage of rare (MAF ≤ 0.1%) and ultra-rare (MAF ≤ 0.01%) cis-CDS pQTLs.

“In comparison, PTVs and missense variants encompassed a significantly larger percentage of rare (MAF ≤ 0.1%) and ultra-rare (MAF ≤ 0.01%) cis-CDS pQTLs (Fig. 1c and Supplementary Table 5).”

Rare cis associations had larger magnitudes of effect (median |β| = 1.86) than rare trans associations (median |β| = 1.26).

“Rare cis associations also had larger magnitudes of effect (median |β| = 1.86) than rare trans associations (median |β| = 1.26; Wilcoxon P < 10−300) (Extended Data Fig. 2b).”

Most pQTLs identified in the collapsing analysis were associated with changes in the abundance of a single protein.

“Most pQTLs identified in the collapsing analysis were associated with changes in the abundance of a single protein (Fig. 2b).”

The models that included PTVs alongside putatively damaging missense variants encompassed most of the 4,528 cis-CDS signals.

“The models that included PTVs alongside putatively damaging missense variants encompassed most of the 4,528 cis-CDS signals.”

Only 112 of the rare PTV signals from the ExWAS did not achieve significance in the PTV collapsing model.

“Meanwhile, only 112 of the rare PTV signals from the ExWAS did not achieve significance in the PTV collapsing model (Extended Data Fig. 4b).”

Three rare protein-coding variants in NLRC4 were associated with significant changes in plasma levels of IL-18 in the ExWAS.

“For example, three rare protein-coding variants in NLRC4 were associated with significant changes in plasma levels of the proinflammatory cytokine IL-18 in the ExWAS (Extended Data Table 2 and Supplementary Table 2).”

The median absolute effect size (β) of rare cis-CDS pQTLs was 1.85, compared with 0.3 for common cis-CDS pQTLs.

“As expected, the median absolute effect size (β) of rare cis-CDS pQTLs was 1.85, compared with 0.3 for common cis-CDS pQTLs (Wilcoxon P < 10−300).”

Among the 254 genes with trans-CDS associations, 87% were associated with three or fewer proteins.

“Among the 254 genes with trans-CDS associations, 87% were associated with three or fewer proteins.”

Of the 2,227 rare cis-CDS pQTLs, 538 (24.2%) were PTVs.

“Of the 2,227 rare cis-CDS pQTLs, 538 (24.2%) were PTVs (Fig. 1d and Supplementary Table 5).”

The absolute effect sizes of rare trans-CDS pQTLs (median |β| = 1.26) were significantly larger than those of common trans-CDS pQTLs (median |β| = 0.08).

“Similarly, the absolute effect sizes of rare trans-CDS pQTLs (median |β| = 1.26) were significantly larger than those of common trans-CDS pQTLs (median |β| = 0.08; Wilcoxon P < 10−300) (Extended Data Fig. 2a).”

Most often, rare missense variants within the same gene had a similar effect on protein abundance.

“Most often, rare missense variants within the same gene had a similar effect on protein abundance.”

In the pan-ancestry collapsing analysis, there were 752 unique study-wide significant gene-protein abundance associations that were not significant in the European-only analyses.

“In this combined analysis, there were 752 unique study-wide significant gene–protein abundance associations that were not significant in the European-only analyses.”

228 associations that were significant in the European-only analyses did not reach study-wide significance in the pan-ancestry analysis.

“On the other hand, 228 associations that were significant in the European-only analyses did not reach study-wide significance in the pan-ancestry analysis (Supplementary Tables 8 and 10).”

PTVs in HBB were associated with increased levels of SLC16A1 (β = 1.85; 95% CI, 1.33 to 2.37; P = 3.2 × 10−12).

“One newly significant example was the trans association between PTVs in HBB and increased levels of the monocarboxylic acid transporter SLC16A1 (β = 1.85; 95% CI, 1.33 to 2.37; P = 3.2 × 10−12).”

Certain genes appeared to be pQTL hotspots, associated with over 20 different protein abundances, including ASGR1 (n = 186), STAB1 (n = 77), STAB2 (n = 41) and GNPTAB (n = 37).

“However, certain genes appeared to be pQTL ‘hotspots’, associated with over 20 different protein abundances, including ASGR1 (n = 186), STAB1 (n = 77), STAB2 (n = 41) and GNPTAB (n = 37) (Fig. 2a,b).”

90% of the more common study-wide associations (MAF > 0.1%) in our study were also significant in the GWAS.

“Furthermore, 90% of the more common study-wide associations (MAF > 0.1%) in our study were also significant in the GWAS (Fig. 1b).”

Only 77 (21%) of the 372 significant trans signals from the ptv model were associated with decreased protein levels.

“Meanwhile, only 77 (21%) of the 372 significant trans signals from the ptv model were associated with decreased protein levels.”

Nearly all the PTVs were associated with decreased protein abundances (n = 518 of 538; 96%).

“As expected, nearly all the PTVs were associated with decreased protein abundances (n = 518 of 538; 96%).”

Of the associations that only reached significance in the collapsing analysis, 101 (20.2%) were cis-CDS.

“Of the associations that only reached significance in the collapsing analysis, 101 (20.2%) were cis-CDS (Supplementary Table 8).”

The signals from the remaining nine significant genes in the PTV-only model were attenuated in the plasma pQTL-informed missense models.

“The signals from the remaining nine significant genes in the PTV-only model were attenuated in the plasma pQTL-informed missense models (Supplementary Table 15).”

There was complete concordance across Olink panels for suggestive and significant ExWAS cis-CDS pQTLs (P < 1 × 10−4) corresponding to proteins measured in multiple panels (CXCL8, TNF, IDO1 and LMOD1).

“Encouragingly, there was complete concordance across panels (Supplementary Table 18).”

The HSD17B13 splice variant rs72613567 was associated with altered levels of SMPD1 (P = 2.7 × 10−11; β = −0.05; 95% CI, −0.03 to −0.06).

“In the ExWAS, this variant was associated with altered levels of ASS1 (P = 1.4 × 10−12; β = −0.05; 95% CI, −0.04 to −0.06), CES3 (P = 2.9 × 10−12; β = 0.07; 95% CI, 0.08 to 0.05), FUOM (P = 1.7 × 10−10; β = −0.05; 95% CI, −0.03 to −0.06), HYAL1 (P = 9.5 × 10−9; β = −0.04; 95% CI, −0.03 to −0.06), SHBG (P = 1.9 × 1−13; β = −0.05; 95% CI, −0.04 to −0.06) and SMPD1 (P = 2.7 × 10−11; β = −0.05; 95% CI”

There was a newly significant association between TCN1 and vitamin B deficiency.

“There was also a newly significant association between TCN1, encoding a vitamin B12 binding protein, and vitamin B deficiency.”

The HSD17B13 splice variant rs72613567 was associated with altered levels of SHBG (P = 1.9 × 1−13; β = −0.05; 95% CI, −0.04 to −0.06).

“In the ExWAS, this variant was associated with altered levels of ASS1 (P = 1.4 × 10−12; β = −0.05; 95% CI, −0.04 to −0.06), CES3 (P = 2.9 × 10−12; β = 0.07; 95% CI, 0.08 to 0.05), FUOM (P = 1.7 × 10−10; β = −0.05; 95% CI, −0.03 to −0.06), HYAL1 (P = 9.5 × 10−9; β = −0.04; 95% CI, −0.03 to −0.06), SHBG (P = 1.9 × 1−13; β = −0.05; 95% CI, −0.04 to −0.06) and SMPD1 (P = 2.7 × 10−11; β = −0.05; 95% CI”

At the P ≤ 1 × 10−8 threshold, the expected number of collapsing pQTL false positives is 4 out of the 7,412 (0.05%) observed significant associations.

“At this P ≤ 1 × 10−8 threshold, the expected number of collapsing pQTL false positives is 4 out of the 7,412 (0.05%) observed significant associations.”

The P value of the association between PCSK9 and dyslipidaemia improved from P = 4.02 × 10−17 (OR = 0.35; 95% CI, 0.27 to 0.46) to P = 7.69 × 10−112 (OR = 0.63; 95% CI, 0.60 to 0.65) in the ptvolink2pcnt model.

“The P value of the association between PCSK9 and dyslipidaemia markedly improved from P = 4.02 × 10−17 (odds ratio (OR) = 0.35; 95% CI, 0.27 to 0.46) to P = 7.69 × 10−112 (OR = 0.63; 95% CI, 0.60 to 0.65) in the ptvolink2pcnt model.”

At the P ≤ 1 × 10−8 threshold, the expected number of ExWAS pQTL false positives is 24 out of the 328,975 observed significant associations (0.007%).

“At this P ≤ 1 × 10−8 threshold, the expected number of ExWAS pQTL false positives is 24 out of the 328,975 observed significant associations (0.007%).”

The HSD17B13 splice variant rs72613567 was associated with altered levels of HYAL1 (P = 9.5 × 10−9; β = −0.04; 95% CI, −0.03 to −0.06).

“In the ExWAS, this variant was associated with altered levels of ASS1 (P = 1.4 × 10−12; β = −0.05; 95% CI, −0.04 to −0.06), CES3 (P = 2.9 × 10−12; β = 0.07; 95% CI, 0.08 to 0.05), FUOM (P = 1.7 × 10−10; β = −0.05; 95% CI, −0.03 to −0.06), HYAL1 (P = 9.5 × 10−9; β = −0.04; 95% CI, −0.03 to −0.06), SHBG (P = 1.9 × 1−13; β = −0.05; 95% CI, −0.04 to −0.06) and SMPD1 (P = 2.7 × 10−11; β = −0.05; 95% CI”

cis-CDS pQTLs corresponding to PTVs were nearly always associated with decreased protein levels.

“cis-CDS pQTLs corresponding to PTVs were nearly always associated with decreased protein levels”

In total, 24 of 5.4 billion permuted tests had P ≤ 1 × 10−8.

“In total, 24 of 5.4 billion permuted tests had P ≤ 1 × 10−8 (Supplementary Table 3).”

They identified 5,025 missense variants with cis-acting negative betas (P < 0.0001) among the genes encoding the 2,923 Olink protein analytes.

“We identified 5,025 missense variants with cis-acting negative betas (P < 0.0001) among the genes encoding the 2,923 Olink protein analytes.”

The associations of PCSK9, ANGPTL4, LPL and ANGPTL3 with lipid-related traits all improved under the ptvolink and ptvolink2pcnt models.

“Consistent with related binary phenotypes, the associations of PCSK9, ANGPTL4, LPL and ANGPTL3 with lipid-related traits all improved under the ptvolink and ptvolink2pcnt models.”

Rare cis- and trans-CDS pQTLs conferred significantly larger effect sizes than common variant pQTLs.

“rare cis- and trans-CDS pQTLs conferred significantly larger effect sizes than common variant pQTLs.”

The HSD17B13 splice variant rs72613567 was associated with altered levels of FUOM (P = 1.7 × 10−10; β = −0.05; 95% CI, −0.03 to −0.06).

“In the ExWAS, this variant was associated with altered levels of ASS1 (P = 1.4 × 10−12; β = −0.05; 95% CI, −0.04 to −0.06), CES3 (P = 2.9 × 10−12; β = 0.07; 95% CI, 0.08 to 0.05), FUOM (P = 1.7 × 10−10; β = −0.05; 95% CI, −0.03 to −0.06), HYAL1 (P = 9.5 × 10−9; β = −0.04; 95% CI, −0.03 to −0.06), SHBG (P = 1.9 × 1−13; β = −0.05; 95% CI, −0.04 to −0.06) and SMPD1 (P = 2.7 × 10−11; β = −0.05; 95% CI”

There were two associations where the flexnonsyn model outperformed the ptvolink model: PROC with thrombophlebitis and LCAT with HDL cholesterol.

“There were two associations where the flexnonsyn model outperformed the ptvolink model (PROC with thrombophlebitis and LCAT with HDL cholesterol; Fig. 4b and Extended Data Fig. 5).”

The rare missense cis-CDS pQTLs are enriched for pathogenic and predicted damaging missense variants.

“the rare missense cis-CDS pQTLs are enriched for pathogenic and predicted damaging missense variants.”

The P values of the UKB-PPP excluded models and full models were highly correlated (R > 0.99).

“The P values of the UKB-PPP excluded models and full models were highly correlated (R > 0.99; Supplementary Fig. 4).”

15 genes that were not significant in the standard ptv model became significant in at least one of the pQTL-informed models.

“Impressively, 15 genes that were not significant in the standard ptv model became significant in at least one of the pQTL-informed models (Fig. 4b and Supplementary Table 15).”

There was an association between EPO and reduced haematocrit that was only significant in the ptvolink2pcnt model (P = 1.8 × 10−83; β = −0.24; 95% CI, −0.27 to −0.22).

“There was also an association between EPO and reduced haematocrit that was only significant in the ptvolink2pcnt model (P = 1.8 × 10−83; β = −0.24; 95% CI, −0.27 to −0.22).”

Most of the significant (P ≤ 1 × 10−8) associations in the CH analysis arose with a VAF ≥ 10% cut-off.

“Most of the significant (P ≤ 1 × 10−8) associations arose with a VAF ≥ 10% cut-off (Supplementary Table 13).”

The HSD17B13 splice variant rs72613567 was associated with altered levels of ASS1 (P = 1.4 × 10−12; β = −0.05; 95% CI, −0.04 to −0.06).

“In the ExWAS, this variant was associated with altered levels of ASS1 (P = 1.4 × 10−12; β = −0.05; 95% CI, −0.04 to −0.06), CES3 (P = 2.9 × 10−12; β = 0.07; 95% CI, 0.08 to 0.05), FUOM (P = 1.7 × 10−10; β = −0.05; 95% CI, −0.03 to −0.06), HYAL1 (P = 9.5 × 10−9; β = −0.04; 95% CI, −0.03 to −0.06), SHBG (P = 1.9 × 1−13; β = −0.05; 95% CI, −0.04 to −0.06) and SMPD1 (P = 2.7 × 10−11; β = −0.05; 95% CI”

Epitope effects do not seem to systematically confound these coding pQTLs.

“epitope effects do not seem to systematically confound these coding pQTLs.”

Several quantitative trait associations became more significant in the pQTL-informed models.

“Several quantitative trait associations also became more significant in these models (Extended Data Fig. 6 and Supplementary Table 16).”

The HSD17B13 splice variant rs72613567 was associated with altered levels of CES3 (P = 2.9 × 10−12; β = 0.07; 95% CI, 0.08 to 0.05).

“In the ExWAS, this variant was associated with altered levels of ASS1 (P = 1.4 × 10−12; β = −0.05; 95% CI, −0.04 to −0.06), CES3 (P = 2.9 × 10−12; β = 0.07; 95% CI, 0.08 to 0.05), FUOM (P = 1.7 × 10−10; β = −0.05; 95% CI, −0.03 to −0.06), HYAL1 (P = 9.5 × 10−9; β = −0.04; 95% CI, −0.03 to −0.06), SHBG (P = 1.9 × 1−13; β = −0.05; 95% CI, −0.04 to −0.06) and SMPD1 (P = 2.7 × 10−11; β = −0.05; 95% CI”

They ran the synonymous (empirical null) collapsing model and found only seven events achieved a signal below the P ≤ 1 × 10−8 threshold.

“we ran the synonymous (empirical null) collapsing model and found only seven events achieved a signal below this threshold.”

The ANGPTL3-dyslipidaemia signal improved from P = 8.73 × 10−9 (OR = 0.58; 95% CI, 0.48 to 0.70) to P = 9.62 × 10−17 (OR = 0.57; 95% CI, 0.50 to 0.66) in the ptvolink model.

“Likewise, the ANGPTL3-dyslipidaemia signal improved from P = 8.73 × 10−9 (OR = 0.58; 95% CI, 0.48 to 0.70) to P = 9.62 × 10−17 (OR = 0.57; 95% CI, 0.50 to 0.66) in the ptvolink model.”

In the standard ptv model, 11 genes with at least one qualifying cis-CDS missense pQTL were significantly associated with at least one phenotype.

“In the standard ptv model, 11 genes with at least one qualifying cis-CDS missense pQTL were significantly associated with at least one phenotype (Supplementary Table 15).”

Only 4 of 499.9 million permuted tests had P ≤ 1 × 10−8.

“Only 4 of 499.9 million permuted tests had P ≤ 1 × 10−8 (Supplementary Table 3).”

They identified an additional 382 cautionary variants on top of the original 8,365 previously described.

“we identified an additional 382 cautionary variants (Supplementary Table 17) on top of the original 8,365 previously described.”

Genomic inflation was well-controlled (median λGC = 1.03; 95% range, 1.01-1.05).

“Genomic inflation was well-controlled (median λGC = 1.03; 95% range, 1.01–1.05; Supplementary Fig. 1 and Supplementary Table 4).”

After sample-level quality control, 49,736 (92%) multi-ancestry samples with exomes remained for downstream analyses.

“We performed rigorous sample-level quality control (Methods), leaving 49,736 (92%) multi-ancestry samples with exomes for downstream analyses.”

The relatively low percentage of cis-position, trans-CDS associations compared with the ExWAS (5% versus 52%) highlights the strength of collapsing analysis in mitigating contamination due to linkage disequilibrium.

“This relatively low percentage of cis-position, trans-CDS associations compared with the ExWAS (5% versus 52%) highlights the strength of collapsing analysis in mitigating contamination due to linkage disequilibrium.”

Collapsing analyses increase statistical power for discovering rare variant-driven associations.

“These data demonstrate how collapsing analyses increase statistical power for discovering rare variant-driven associations.”

The cis-CDS pQTLs are enriched for biologically relevant signals.

“These results collectively demonstrate that the cis-CDS pQTLs are enriched for biologically relevant signals.”

The n-of-1 permutation analysis and the synonymous QV model converged on P ≤ 1 × 10−8 as an appropriately conservative P value threshold.

“The n-of-1 permutation analysis and the synonymous QV model converged on P ≤ 1 × 10−8 as an appropriately conservative P value threshold (Supplementary Table 3 and Methods).”

Linkage disequilibrium contaminates common pQTLs, making it challenging to ascribe causality to these variants without fine-mapping.

“These results illustrate how linkage disequilibrium contaminates common pQTLs, making it challenging to ascribe causality to these variants without fine-mapping.”

Population genetics-based methods can prioritize functional missense variants, offering complementary information to in silico pathogenicity predictors.

“This underscores the ability of population genetics-based methods to prioritize functional missense variants, offering complementary information to in silico pathogenicity predictors.”

Including cis-CDS missense pQTLs can enhance conventional loss-of-function gene collapsing analyses.

“These results demonstrate how including cis-CDS missense pQTLs can enhance conventional loss-of-function gene collapsing analyses.”

There was very little overlap between the protein abundances associated with each of the five CH genes, suggesting distinct downstream effects.

“Strikingly, there was very little overlap between the protein abundances associated with each of these five genes, suggesting distinct downstream effects of the somatic events detected in each.”

Some rare, putative gain-of-function mutations in NLRC4 may not be sufficient to cause an observable phenotype.

“Moreover, these data show that some rare, putative gain-of-function mutations in this gene may not be sufficient to cause an observable phenotype, highlighting the value of this resource in clinical diagnostic settings.”

They pruned the cohort for sample duplicates and first-degree genetic relatedness, resulting in 49,736 (95%) participants.

“We then pruned this cohort for sample duplicates and first-degree genetic relatedness (no pair with a kinship coefficient exceeding 0.1769, n = 462), resulting in 49,736 (95%) participants available for the multi-ancestry analyses performed in this paper.”

They identified and excluded 46 genes enriched for exome sequencing batch effects.

“we had previously created dummy phenotypes to correspond to each of the five exome sequence delivery batches to identify and exclude from analyses 46 genes that were enriched for exome sequencing batch effects”