Cross-ancestry comparison of aptamer and antibody protein measures

PAVs may result in pQTL associations that reflect differential probe binding to alternate isoforms rather than differential protein abundances.

“Nevertheless, interpreting cis- and trans-pQTL remains complicated, as PAVs may result in pQTL associations that reflect differential probe binding to alternate isoforms12,17–19 rather than differential protein abundances.”

• cites [12] Katz, D. H. et al. Proteomic profiling platforms head to head: leveraging genetics and clinical traits to compare aptamer- and antibody-based methods. Sci. Adv. 8, eabm5164 (2022). sourceCould not verify

  The Results section on platform correlation and cis pQTLs describes associations between genetic variants and protein measurements, while the Discussion treats cis pQTLs as a possible indicator of assay specificity. Our copy does not address whether protein-altering variants produce associations through differential probe binding to alternate isoforms rather than differences in protein abundance (searched: PAVs, isoforms, differential binding). (LLM, conf=0.72)

  Experimental — accuracy not yet measured
• cites [17] Eldjarn, G. H. et al. Large-scale plasma proteomics comparisons through genetics and disease associations. Nature 622, 348–358 (2023). sourcePartly supported

  The ‘Relationship of pQTLs between platforms’ section says PAV-linked cis-pQTLs are likely to reflect epitope effects rather than variation in protein levels. The Discussion also says the platforms may target different proteoforms, but it does not specifically link PAV-driven binding differences to alternate isoforms. Page numbers were not available in our copy. (LLM, conf=0.83)

  Experimental — accuracy not yet measured
• cites [18] Suhre, K. et al. A genome-wide association study of mass spectrometry proteomics using a nanoparticle enrichment platform. Nat. Genet. 57, 2987–2996 (2025). sourcePartly supported

  The Discussion states that PAV-containing peptides can produce an epitope-effect pQTL that does not fully reflect genotype-dependent protein abundance. The Results (Validity of the MSPA score) likewise identify coding variants that could alter epitopes. The paper discusses affinity binders targeting different isoforms as another possible explanation, but does not specifically establish that PAVs cause differential probe binding to alternate isoforms. (LLM, conf=0.84)

  Experimental — accuracy not yet measured
• cites [19] Pietzner, M. et al. Mapping the proteo-genomic convergence of human diseases. Science 374, eabj1541 (2021). sourcePartly supported

  The Results section on cis and trans genetic architecture says PAVs can affect a protein’s binding epitope and that conformation, rather than plasma abundance, may explain some genetic effects. A separate Results section on gene expression and splicing suggests that alternative protein isoforms can have different affinities for the assay reagent. The paper does not link PAVs specifically to differential binding to alternate isoforms. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

The detection of a cis-pQTL for a given protein measure is often used as a metric to confirm on-target affinity-probe binding.

“The detection of a cis-pQTL for a given protein measure is often used as a metric to confirm on-target affinity-probe binding12,15,16; existing studies comparing SomaScan and Olink measures have established that the detection of a cis-pQTL is a strong predictor of inter-platform correlation coefficients for a given protein12,15,16.”

• cites [12] Katz, D. H. et al. Proteomic profiling platforms head to head: leveraging genetics and clinical traits to compare aptamer- and antibody-based methods. Sci. Adv. 8, eabm5164 (2022). sourceSupported
• cites [15] Raffield, L. M. et al. Comparison of proteomic assessment methods in multiple cohort studies. Proteomics 20, e1900278 (2020). sourcePartly supported

  The Discussion says cis-pQTL presence can provide clues to assay specificity and suggests that a probe has at least some binding to its target protein. It does not treat a cis-pQTL as confirmation of on-target binding: it cautions that binding differences can produce spurious pQTLs and that a cis-pQTL does not exclude off-target binding. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites [16] Pietzner, M. et al. Synergistic insights into human health from aptamer- and antibody-based proteomic profiling. Nat. Commun. 12, 6822 (2021). sourceCould not verify

  The Methods (Annotation of pQTLs; page unavailable) describe checking cis-pQTLs for protein-altering variants, but do not say their detection is used to confirm on-target probe binding. The Discussion (page unavailable) says identifying the protein species bound to an affinity reagent may require mass spectrometry. Searches for cis-pQTLs as a binding-specificity or validation metric found no passage establishing the claimed practice. (LLM, conf=0.73)

  Experimental — accuracy not yet measured

As genetic variants may vary in frequency across populations, such variants could drive systematic differences in protein measures.

“As these genetic variants may vary in frequency across populations, such variants could drive systematic differences in protein measures20.”

• cites [20] Olson, N. C. et al. Soluble urokinase plasminogen activator receptor: genetic variation and cardiovascular disease risk in black adults. Circ. Genom. Precis. Med. 14, e003421 (2021). sourceSupported

Several cis-pQTL signals driven by PAVs-most commonly missense variants-were observed on each platform.

“While the majority of cis-pQTL variants were non-coding on both platforms (Supplementary Fig. 8), we observe several cis-pQTL signals driven by protein altering variants (PAVs—most commonly missense variants) on each platform, raising concerns as to whether the signal captures abundances or differences in isoform, as has been shown in other studies18,24,25.”

• cites [18] Suhre, K. et al. A genome-wide association study of mass spectrometry proteomics using a nanoparticle enrichment platform. Nat. Genet. 57, 2987–2996 (2025). sourcePartly supported

  The Results section, ‘Validity of the MSPA score’ (page not provided), reports candidate epitope-changing coding variants for cis-pQTLs on both SOMAscan and Olink, including detected PAV-containing peptides. This supports the presence of PAV-related signals on both platforms, but the passage does not establish that missense variants were the most common PAVs or that every candidate variant drove its signal. (LLM, conf=0.84)

  Experimental — accuracy not yet measured
• cites [24] Pietzner, M. et al. Nanoparticle enriched mass spectrometry proteomics in British South Asians identifies novel variant-protein-disease mechanisms. Preprint at medRxiv, https://doi.org/10.1101/2025.08.12.25333522 (2025). sourcePartly supported

  The section “Protein altering variants and overlap with tissue gene expression loci explain pQTLs” (page unavailable) reports cis-pQTLs associated with protein-altering variants. “Cross-platform concordance of pQTLs” (page unavailable) also reports PAV-associated cis-pQTLs in comparisons with both Olink HT and SomaLogic 11k. The passages do not establish that missense variants were the most common PAVs on each platform, and they describe associations rather than showing that the PAVs drove those signals. (LLM, conf=0.84)

  Experimental — accuracy not yet measured
• cites [25] Wang, Q. S. et al. Platform-dependent effects of genetic variants on plasma APOL1. Preprint at BioRxiv, https://doi.org/10.1101/2025.01.30.635763 (2025). sourcePartly supported

  The Introduction reports that APOL1 G1/G2 mutations have pQTL effects on both Olink and SomaLogic, and the Discussion describes a similar effect for V1. The paper supports protein-altering variants affecting measurements on both platforms, but its APOL1-focused analysis does not establish that missense variants are the most common drivers of cis-pQTL signals generally. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

Existing studies comparing SomaScan and Olink measures have established that the detection of a cis-pQTL is a strong predictor of inter-platform correlation coefficients for a given protein.

“The detection of a cis-pQTL for a given protein measure is often used as a metric to confirm on-target affinity-probe binding12,15,16; existing studies comparing SomaScan and Olink measures have established that the detection of a cis-pQTL is a strong predictor of inter-platform correlation coefficients for a given protein12,15,16.”

• cites [12] Katz, D. H. et al. Proteomic profiling platforms head to head: leveraging genetics and clinical traits to compare aptamer- and antibody-based methods. Sci. Adv. 8, eabm5164 (2022). sourcePartly supported

  In Results, ‘Accuracy: Platform correlation and cis pQTLs’ (page unavailable), the paper reports that highly correlated Olink–SomaScan reagent pairs were likely to show cis-pQTLs on both platforms. It also reports that lower-correlation pairs were more likely to show a cis-pQTL on Olink only. The evidence therefore supports an association with detection on both platforms, but not the unqualified claim that detecting a cis-pQTL for a protein measure is a strong predictor of its inter-platform correlation coefficient. (LLM, conf=0.84)

  Experimental — accuracy not yet measured
• cites [15] Raffield, L. M. et al. Comparison of proteomic assessment methods in multiple cohort studies. Proteomics 20, e1900278 (2020). sourcePartly supported

  The Results section reports that proteins with a SOMAscan cis-pQTL were enriched among the more highly correlated SOMAscan–Olink measures. But the Discussion says the Olink–SOMAscan dataset was too small for cis-pQTL analysis, and cautions that a cis-pQTL does not guarantee high inter-assay correlation. The paper therefore supports an association, not the claim that cis-pQTL detection was established as a strong predictor of correlation coefficients. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites [16] Pietzner, M. et al. Synergistic insights into human health from aptamer- and antibody-based proteomic profiling. Nat. Commun. 12, 6822 (2021). sourcePartly supported

  The Results section (‘Shared genetic effects are target-dependent’; page unavailable) reports that lower observational correlation between SomaScan and Olink measures was associated with a higher likelihood of a platform-specific pQTL. That supports a relationship between cross-platform correlation and pQTL agreement, but does not establish that detecting a cis-pQTL is a strong predictor of the correlation coefficient for a protein. (LLM, conf=0.77)

  Experimental — accuracy not yet measured

Comparing affinity-based pQTL to pQTL identified in MS studies may facilitate signal interpretation.

“While MS-based pQTL studies are relatively limited in size and number (with some of the broadest studies, used for pQTL comparison here, currently including 1260 Americans and 1980 proteins18, or 325 participants and ~2800 proteins10), comparing affinity-based pQTL to pQTL identified in MS studies may facilitate signal interpretation.”

• cites [18] Suhre, K. et al. A genome-wide association study of mass spectrometry proteomics using a nanoparticle enrichment platform. Nat. Genet. 57, 2987–2996 (2025). sourceSupported
• cites [10] Suhre, K. et al. Nanoparticle enrichment mass-spectrometry proteomics identifies protein-altering variants for precise pQTL mapping. Nat. Commun. 15, 989 (2024). sourceSupported

PILRA showed strong platform-discordant measures (Pearson's r = -0.35) across the present and prior studies.

“This protein showed strong platform-discordant measures (Pearson’s r = -0.35) across the present and prior studies16,17 (Fig. 4A).”

• cites [16] Pietzner, M. et al. Synergistic insights into human health from aptamer- and antibody-based proteomic profiling. Nat. Commun. 12, 6822 (2021). sourcePartly supported

  In Results (‘Directionally discordant associations at an Alzheimer’s disease locus’; page unavailable), the paper reports opposite genetic-association directions for PILRA measured by Olink and by two SomaScan reagents. That supports platform discordance, but our copy does not report a PILRA Pearson correlation of −0.35 or establish that value across the present and prior studies. (LLM, conf=0.85)

  Experimental — accuracy not yet measured
• cites [17] Eldjarn, G. H. et al. Large-scale plasma proteomics comparisons through genetics and disease associations. Nature 622, 348–358 (2023). sourceCould not verify

  The Inter-platform correlation of levels section reports correlations across proteins, and Methods > Correlation of assays across platforms says the paper used Spearman correlation. Searches for PILRA and the claimed Pearson r = −0.35 found no passage in our copy reporting that protein-specific result or the comparison across the present and prior studies. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

PAV-driven cis-pQTL signals raise concerns as to whether the signal captures protein abundances or differences in isoform.

“While the majority of cis-pQTL variants were non-coding on both platforms (Supplementary Fig. 8), we observe several cis-pQTL signals driven by protein altering variants (PAVs—most commonly missense variants) on each platform, raising concerns as to whether the signal captures abundances or differences in isoform, as has been shown in other studies18,24,25.”

• cites [18] Suhre, K. et al. A genome-wide association study of mass spectrometry proteomics using a nanoparticle enrichment platform. Nat. Genet. 57, 2987–2996 (2025). sourcePartly supported

  The Discussion says PAV-containing peptides can produce a pQTL signal that does not fully reflect genotype-dependent protein abundance. The Results section on protein isoforms separately shows that a SPINK5 pQTL can reflect differences among isoforms, but it does not identify that example as PAV-driven. The source therefore supports both interpretive concerns, but not the specific link between PAV-driven signals and isoform differences. (LLM, conf=0.83)

  Experimental — accuracy not yet measured
• cites [24] Pietzner, M. et al. Nanoparticle enriched mass spectrometry proteomics in British South Asians identifies novel variant-protein-disease mechanisms. Preprint at medRxiv, https://doi.org/10.1101/2025.08.12.25333522 (2025). sourceSupported
• cites [25] Wang, Q. S. et al. Platform-dependent effects of genetic variants on plasma APOL1. Preprint at BioRxiv, https://doi.org/10.1101/2025.01.30.635763 (2025). sourceSupported

The 644 proteins with a cis-pQTL detected on both platforms had higher inter-platform correlation (mean Pearson r = 0.58, SD = 0.20) than proteins with no significant cis-pQTL (mean r = 0.21, SD = 0.30).

“Similar to prior studies12,17, the 644 proteins with a cis-pQTL detected on both platforms had higher inter-platform correlation (mean Pearson’s r = 0.58, SD = 0.20) than proteins with no significant cis-pQTL (mean Pearson’s r = 0.21, SD = 0.30)(Fig. 3C and Supplementary Data 7).”

• cites [12] Katz, D. H. et al. Proteomic profiling platforms head to head: leveraging genetics and clinical traits to compare aptamer- and antibody-based methods. Sci. Adv. 8, eabm5164 (2022). sourceSupported
• cites [17] Eldjarn, G. H. et al. Large-scale plasma proteomics comparisons through genetics and disease associations. Nature 622, 348–358 (2023). sourceSupported

Platform-concordant cis-pQTL signals were most likely to contain a GTEx v8 eQTL lead variant for the protein-encoding gene in the credible set, compared to platform-specific or platform-discordant signals.

“Consistent with prior reports17,19, we found that platform-concordant cis-pQTL signals were most likely to contain a GTEx v8 eQTL lead variant for the protein-encoding gene in the credible set, compared to platform-specific or platform-discordant signals.”

• cites [17] Eldjarn, G. H. et al. Large-scale plasma proteomics comparisons through genetics and disease associations. Nature 622, 348–358 (2023). sourcePartly supported

  The ‘Relationship of pQTLs between platforms’ section reports that some cis-pQTL variants are in high LD with cis-eQTLs for the targeted protein’s gene. The Methods also say the authors checked correspondence with top cis-eQTLs from sources including GTEx. Our copy does not report that platform-concordant signals are more likely than platform-specific or discordant signals to contain a GTEx v8 eQTL lead variant in a credible set. (LLM, conf=0.78)

  Experimental — accuracy not yet measured
• cites [19] Pietzner, M. et al. Mapping the proteo-genomic convergence of human diseases. Science 374, eabj1541 (2021). sourcePartly supported

  The Abstract reports that 64% of pQTLs were consistent across the SomaScan and Olink platforms, and the Results section ‘Shared genetic architecture with gene expression and splicing’ reports GTEx v8 eQTL colocalization for cis-pQTLs (page numbers unavailable). Neither passage compares platform-concordant, platform-specific, and platform-discordant signals for the likelihood that a credible set contains the protein-encoding gene’s eQTL lead variant. The cited paper supports the broader context, but not the specific comparison attributed to it. (LLM, conf=0.82)

  Experimental — accuracy not yet measured

Platform-discordant signals for 2 proteins, APOL1 and PGLYRP2, had evidence of impacting specific peptides but not overall protein abundance in an MS study.

“Platform-discordant signals for 2 proteins, APOL1 and PGLYRP2, had evidence of impacting specific peptides but not overall protein abundance in an MS study10 (Supplementary Data 4, 5, and 8);”

• cites [10] Suhre, K. et al. Nanoparticle enrichment mass-spectrometry proteomics identifies protein-altering variants for precise pQTL mapping. Nat. Commun. 15, 989 (2024). sourcePartly supported

  The Results section on robust pQTLs (page not available) shows that a variant can affect a detected peptide without changing overall protein abundance, using F5 as an example. Our copy does not establish that this occurred for the two named proteins, APOL1 and PGLYRP2; searches for both names found no supporting passage. (LLM, conf=0.78)

  Experimental — accuracy not yet measured

Studies incorporating comparisons to MS protein measures, measures of gene transcript abundance, and/or functional validation have established cases where platform-discordant PAV cis-pQTL reflect differences in probe binding to genetically encoded isoforms, including for uromodulin and apolipoprotein-1.

“While PAVs may certainly alter a protein’s overall abundance through impacts on protein stability and other mechanisms, studies incorporating comparisons to MS protein measures, measures of gene transcript abundance, and/or functional validation have established cases where platform-discordant PAV cis-pQTL reflect differences in probe binding to genetically encoded isoforms, including for uromodul”

• cites [31] Li, Y. et al. Genome-wide studies reveal factors associated with circulating uromodulin and its relationships to complex diseases. JCI Insight 7, e157035 (2022). sourcePartly supported

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

  Experimental — accuracy not yet measured
• cites [25] Wang, Q. S. et al. Platform-dependent effects of genetic variants on plasma APOL1. Preprint at BioRxiv, https://doi.org/10.1101/2025.01.30.635763 (2025). sourcePartly supported

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

  Experimental — accuracy not yet measured

We and others observed a platform-discordant PAV cis-pQTL for PILRA.

“We and others observed a platform-discordant PAV cis-pQTL for this protein16.”

• cites [16] Pietzner, M. et al. Synergistic insights into human health from aptamer- and antibody-based proteomic profiling. Nat. Commun. 12, 6822 (2021). sourceSupported

Several putatively impacted proteins are potential biomarkers or drug targets (for example, suPAR and PILRA).

“It remains challenging to establish the pervasiveness of assay-interference effects; however, we note that several putatively impacted proteins are potential biomarkers or drug targets (for example, suPAR32,33 and PILRA27).”

• cites [32] Reisinger, A. C. et al. Soluble urokinase plasminogen activator receptor (suPAR) predicts critical illness and kidney failure in patients admitted to the intensive care unit. Sci. Rep. 11, 17476 (2021). sourceSupported
• cites [33] Jehn, U. et al. Soluble urokinase-type plasminogen activator receptor (suPAR) is a risk indicator for eGFR loss in kidney transplant recipients. Sci. Rep. 11, 3713 (2021). sourceSupported
• cites [27] Zhan, H., Cammann, D., Cummings, J. L., Dong, X. & Chen, J. Biomarker identification for Alzheimer’s disease through integration of comprehensive Mendelian randomization and proteomics data. J. Transl. Med. 23, 278 (2025). sourceSupported

Nearly twice as many credible sets were observed on Olink compared to SomaScan.

“More proteins showed a cis-pQTL on Olink compared to SomaScan (48% vs. 39%, respectively); additionally, we observed nearly twice as many credible sets on Olink compared to SomaScan (although we note this is only among the 2157 proteins measured on both platforms - we did not consider proteins measured on only one platform).”

The platforms followed similar patterns of agreement within each ancestry, with median Pearson r = 0.30 for AFR, 0.28 for AMR, 0.31 for EAS, and 0.32 for EUR.

“Overall, platforms followed similar patterns of agreement within each ancestry, evidenced by similar median correlations and distributions of Pearson’s correlation coefficients observed within each ancestry [median Pearson’s r AFR = 0.30, AMR = 0.28, EAS = 0.31, EUR = 0.32] (Supplementary Fig. 3 and Supplementary Data 2).”

Among the 80 proteins with ancestry-heterogeneous inter-platform correlation estimates, 69 (86%) had a cis-pQTL.

“Bonferroni significance (p < 0.05/2708) | 80 | 69(86%) | 31 (39%) | 45(56%) | 13(16%) | 29(36%) | 16(20%) | 24(30%) | 10(13%)”

Among the 80 proteins with ancestry-heterogeneous inter-platform correlation estimates, 31 (39%) had a PAV cis-pQTL.

“Bonferroni significance (p < 0.05/2708) | 80 | 69(86%) | 31 (39%) | 45(56%) | 13(16%) | 29(36%) | 16(20%) | 24(30%) | 10(13%)”

Among all 2157 proteins, 119 (5.5%) had an ancestry-differentiated PAV cis-pQTL.

“All proteins | All proteins | 2157 | 1239 (57%) | 425 (20%) | 563(26%) | 119(5.5%) | 605(28%) | 190(8.8%) | 348(16%) | 58(2.7%)”

Among all 2157 proteins, 1239 (57%) had a cis-pQTL on either platform.

“All proteins | All proteins | 2157 | 1239 (57%) | 425 (20%) | 563(26%) | 119(5.5%) | 605(28%) | 190(8.8%) | 348(16%) | 58(2.7%)”

Among the 80 proteins with ancestry-heterogeneous inter-platform correlation estimates, 13 (16%) had an ancestry-differentiated PAV cis-pQTL.

“Bonferroni significance (p < 0.05/2708) | 80 | 69(86%) | 31 (39%) | 45(56%) | 13(16%) | 29(36%) | 16(20%) | 24(30%) | 10(13%)”

Among the 80 proteins with ancestry-heterogeneous inter-platform correlation estimates, 45 (56%) had an ancestry-differentiated cis-pQTL.

“Bonferroni significance (p < 0.05/2708) | 80 | 69(86%) | 31 (39%) | 45(56%) | 13(16%) | 29(36%) | 16(20%) | 24(30%) | 10(13%)”

Significant cis-pQTL credible sets (containing at least one variant p < 5 × 10−8) were observed for 840 proteins on SomaScan.

“Here, we observed significant credible sets (containing at least one variant p < 5 × 10−8) for 840 proteins on SomaScan (Supplementary Data 4) and 1036 proteins on Olink (see “Methods”, Table 3, Supplementary Data 5 and Fig. 3A).”

Among all 2157 proteins, 425 (20%) had a cis-pQTL containing a PAV.

“All proteins | All proteins | 2157 | 1239 (57%) | 425 (20%) | 563(26%) | 119(5.5%) | 605(28%) | 190(8.8%) | 348(16%) | 58(2.7%)”

Significant cis-pQTL credible sets were observed for 1036 proteins on Olink.

“Here, we observed significant credible sets (containing at least one variant p < 5 × 10−8) for 840 proteins on SomaScan (Supplementary Data 4) and 1036 proteins on Olink (see “Methods”, Table 3, Supplementary Data 5 and Fig. 3A).”

80 protein targets exhibited significant differences in inter-platform correlation estimates across ancestries (Cochran's Q P < 1.85 × 10−5, Bonferroni adjustment for 2708 probe pairs).

“While the majority of individual proteins showed similar cross-platform agreement across ancestries, 80 protein targets exhibited significant differences in assay agreement across ancestries (Cochran’s Q P < 1.85 × 10−5, Bonferroni adjustment for the total number of probe pairs (2708)) (Tables 1 and 2; Supplementary Data 3), suggesting systematic differences in measurement accuracy across ancestr”

More proteins showed a cis-pQTL on Olink compared to SomaScan (48% vs. 39%, respectively) among the 2157 proteins measured on both platforms.

“More proteins showed a cis-pQTL on Olink compared to SomaScan (48% vs. 39%, respectively); additionally, we observed nearly twice as many credible sets on Olink compared to SomaScan (although we note this is only among the 2157 proteins measured on both platforms - we did not consider proteins measured on only one platform).”

A higher proportion of secreted proteins had a significant cis-pQTL association compared to intracellular and membrane proteins.

“A higher proportion of secreted proteins had a significant cis-pQTL association compared to intracellular and membrane proteins (Supplementary Fig. 7).”

Similar proportions of SomaScan and Olink protein measures were associated with age, sex, BMI, and type 2 diabetes (T2D) case/control status at the Bonferroni significance threshold (P < 1.8 × 10−5).

“Here, similar proportions of SomaScan and Olink protein measures were associated with age, sex, BMI, and type 2 diabetes (T2D) case/control status at the Bonferroni significance threshold (P < 1.8 × 10−5, significant at α = 0.05, with correction for 2708 probe pairs tested) (Supplementary Data 17 and 18).”

Adjusting proteins associated with platform-specific pleiotropic trans-pQTL loci had little effect on protein measure correlation.

“Similarly, adjusting proteins associated with platform-specific pleiotropic trans-pQTL loci had little effect on protein measure correlation (Supplementary Note 5, Supplementary Fig. 54).”

Proteins with Pearson's correlation r > 0.6 were most likely to show a platform-concordant cis-pQTL signal.

“Similarly, proteins with Pearson’s correlation r > 0.6 were most likely to show a platform-concordant cis-pQTL signal (Supplementary Fig. 11).”

Adjustment for protein-altering variants with opposite directions of effect by platform improves inter-platform protein measure correlation.

“Furthermore, adjustment for protein-altering variants with opposite directions of effect by platform improves inter-platform protein measure correlation and results in more concordant genetic and phenotypic associations.”

Most participants exhibited >0.5 genetic similarity to African (AFR, n = 439), Admixed American (AMR, n = 231), East Asian (EAS, n = 277), or European (EUR, n = 939) 1000 Genomes Project reference populations, based on global ancestry proportion estimates.

“Most participants exhibited >0.5 genetic similarity to African (AFR, n = 439), Admixed American (AMR, n = 231), East Asian (EAS, n = 277), or European (EUR, n = 939) 1000 Genomes Project reference populations, based on global ancestry proportion estimates (Supplementary Data 1), enabling stratification by genetic similarity for some analyses.”

A small proportion of affinity-identified pQTL (0.4% SomaScan, 0.6% Olink) harbor a previously reported MS-pQTL lead variant in the credible set.

“A small proportion of affinity-identified pQTL (0.4% SomaScan, 0.6% Olink) harbor a previously reported MS-pQTL lead variant in the credible set (Table 3; Supplementary Data 8, 4, and 5).”

Adjustment for protein-altering variants with opposite directions of effect by platform results in more concordant genetic and phenotypic associations.

“Furthermore, adjustment for protein-altering variants with opposite directions of effect by platform improves inter-platform protein measure correlation and results in more concordant genetic and phenotypic associations.”

For proteins with a platform-discordant PAV, an improved correlation of effect sizes among protein-phenotype associations was observed when using PAV-adjusted measures as input.

“For proteins with a platform-discordant PAV, we observed an improved correlation of effect sizes observed among protein-phenotype associations when using PAV-adjusted measures as input (Supplementary Data 6 and Supplementary Fig. 33).”

Adjusting for the PILRA PAV in the cis-pQTL mapping model unmasks and strengthens a second, independent association with the Olink measure of PILRA, driven by non-coding variant chr7:100480786:C:G.

“Finally, adjusting for this PAV in the cis-pQTL mapping model unmasks and strengthens a second, independent association with the Olink measure of PILRA, driven by non-coding variant chr7:100480786:C:G (Supplementary Fig. 15).”

suPAR is associated with an ancestry-differentiated, platform-discordant PAV.

“Soluble urokinase plasminogen activator receptor (suPAR) is another protein associated with an ancestry-differentiated, platform-discordant PAV.”

Incorporating PAV-adjusted PILRA measures into epidemiological models improved the strength and concordance of downstream PILRA-age associations.

“Furthermore, incorporating PAV-adjusted PILRA measures into epidemiological models improved the strength and concordance of downstream PILRA-age associations (Fig. 4E).”

19 proteins had platform-overlapping cis-pQTL credible sets for which lead variants had opposite effect-directions (platform-discordant).

“While most of these signals were led by variants with concordant directions of association across platforms (referred to here as “platform-concordant” signals), we identified 19 proteins with platform-overlapping cis-pQTL credible sets for which lead variants had opposite effect-directions (“platform-discordant”) (Supplementary Fig. 10).”

Upon adjusting for copies of the PAV allele, the correlation of residual PILRA measures improved to 0.27.

“Upon adjusting for copies of the PAV allele, the correlation of residual PILRA measures improved to 0.27 (Fig. 4A).”

Proteins with a platform-concordant, significant cis-pQTL had generally higher inter-platform correlation (median r = 0.61, SD = 0.18) than proteins with a platform-discordant cis-pQTL (median r = 0.40, SD = 0.20) or a cis-pQTL on one platform only (median r = 0.12, SD = 0.28).

“Furthermore, the correlation of proteins with a platform-concordant, significant cis-pQTL was generally higher (median Pearson’s r = 0.61, SD = 0.18) than the correlation of proteins with a platform-discordant cis-pQTL (median Pearson’s r = 0.40, SD = 0.20) or a cis-pQTL on one platform only (median Pearson’s r = 0.12, SD = 0.28) (Fig. 3B, C; Supplementary Data 7 and Supplementary Fig. 11).”

Nine trans-pQTL regions were pleiotropic for SomaScan protein measures only (significant trans-pQTL associations with ≥5 proteins on SomaScan, 0 or 1 on Olink), including APOE, F5, and HPX.

“Nine trans-pQTL regions were pleiotropic for SomaScan protein measures only (significant trans-pQTL associations with ≥5 proteins on SomaScan, 0 or 1 on Olink), including APOE, F5, and HPX.”

72 proteins on SomaScan and 84 proteins on Olink were associated with cis-pQTL containing highly ancestry-differentiated PAVs (Chi-Square >202, the 75th X2 percentile among variants tested in MESA).

“Overall, 72 proteins on SomaScan and 84 proteins on Olink were associated with cis-pQTL containing highly ancestry-differentiated PAVs (Chi-Square >202, the 75th X2 percentile among variants tested in MESA).”

Among each phenotype, proteins with a Bonferroni significant phenotype association on both platforms were more likely to associate with a cis-pQTL on both platforms.

“Among each phenotype, proteins with a Bonferroni significant phenotype association on both platforms were more likely to associate with a cis-pQTL on both platforms (Supplementary Data 17).”

For age, sex, and BMI, a small number of proteins (6 to 25) had platform-discordant, Bonferroni significant associations on both platforms.

“For age, sex, and BMI, we observed a small number of proteins with platform-discordant, Bonferroni significant associations on both platforms (6 to 25, Supplementary Data 17), suggesting that differences in protein measurements may indeed result in differences in epidemiological associations.”

18 platform-discordant PAV associations were identified when considering all probes per overlapping protein.

“The 18 platform-discordant PAV associations identified in analyses considering all probes per overlapping protein are displayed in Table 4.”

Across all proteins, effect sizes for each phenotype were modestly correlated (Pearson's r of approximately 0.6).

“Across all proteins, effect sizes for each phenotype were modestly correlated (Pearson’s r of approximately 0.6) (Supplementary Fig. 32).”

Inter-platform PILRA measure correlation improved most among AMR and EAS participants after PAV adjustment.

“Inter-platform protein measure correlation improved most among AMR and EAS participants (Fig. 4D and Supplementary Fig. 28).”

Among the 392 proteins with a cis-pQTL association detected with Olink measures only, 30% were associated with at least one PAV.

“Among the 196 and 392 proteins with a cis-pQTL association detected with SomaScan or Olink measures only, similar proportions were associated with at least one PAV (25% SomaScan, 30% Olink) (Fig. 3C).”

Among the 80 proteins with Bonferroni-significant heterogeneity in inter-platform correlation estimates between ancestries, the rate of ancestry-differentiated cis-pQTL containing PAVs was higher (16%) than among proteins overall (5.5%, 118/2157).

“Indeed, we observed a higher rate of ancestry-differentiated cis-pQTL containing PAVs among the 80 proteins with Bonferroni-significant heterogeneity in inter-platform correlation estimates between ancestries (Cochran’s Q P < 1.8 × 10−5, significant at α = 0.05, with correction for 2708 probe pairs tested) (16% compared to 5.5% (118/2157) of proteins overall) (Table 1).”

10 pleiotropic trans-pQTL regions were associated with at least 5 proteins on each platform, including ABO, FUT2, and F12.

“We identified 10 pleiotropic trans-pQTL regions that were associated with at least 5 proteins on each platform (although not necessarily the same proteins on both platforms), including well-characterized ABO, FUT2, and F12.”

Approximately 25-30% of the signals on each platform are likely driven by protein-altering variants.

“We evaluate protein measure associations with genetic variants, and find approximately 25-30% of the signals on each platform are likely driven by protein-altering variants.”

The majority of cis-pQTL credible sets identified on each platform contained a GWAS catalog lead variant.

“The majority of cis-pQTL credible sets identified on each platform contained a GWAS catalog lead variant (Table 3; Supplementary Data 4 and 5).”

Among the 196 proteins with a cis-pQTL association detected with SomaScan measures only, 25% were associated with at least one PAV.

“Among the 196 and 392 proteins with a cis-pQTL association detected with SomaScan or Olink measures only, similar proportions were associated with at least one PAV (25% SomaScan, 30% Olink) (Fig. 3C).”

Adjustment for platform-discordant, ancestry-differentiated PAVs generally reduced cross-ancestry heterogeneity in inter-platform correlation estimates.

“Adjustment for platform-discordant, ancestry-differentiated PAVs generally reduced cross-ancestry heterogeneity in inter-platform correlation estimates.”

Similar proportions of proteins tested in MR on each platform showed significant (p < 0.05) associations with AD.

“Similar proportions of proteins tested in MR on each platform showed significant (p < 0.05) associations with AD.”

Proteins with cross-ancestry differences in platform agreement were enriched for association with PAVs with ancestry-differentiated allele frequencies.

“Additionally, we established that proteins with cross-ancestry differences in platform agreement were enriched for association with PAVs with ancestry-differentiated allele frequencies, supporting epitope-binding effects as one driver of cross-ancestry differences in platform agreement.”

Proteins associated with concordant cis-genetic signals exhibited the highest average inter-platform correlation.

“Proteins associated with concordant cis-genetic signals exhibited the highest average inter-platform correlation, increasing confidence that the affinity probes for these proteins captured the same entities and may have high potential for cross-platform meta-analysis.”

Among the 2157 proteins with SomaScan 7k and Olink Explore 3072 measures compared here, there is a wide range of inter-platform correlations for given protein measures.

“Among the 2157 proteins with SomaScan 7k and Olink Explore 3072 measures compared here, we observe a wide range of inter-platform correlations for given protein measures, as in prior studies.”

Five proteins in SomaScan MR (CHIT1, GDF-15, HGFAC, HRC, qSOX1) and 5 in Olink MR analyses (OSCAR, LTA, SPINT1, TMPRSS5, TNFRSF13C) were significant in models excluding platform-discordant and platform-specific PAV instruments but were not significant in the model including these instruments as PAVs.

“Furthermore, we observed 5 proteins in SomaScan MR (CHIT1, GDF-15, HGFAC, HRC, qSOX1), and 5 in Olink MR analyses (OSCAR, LTA, SPINT1, TMPRSS5, TNFRSF13C), which were significant in models excluding platform-discordant and platform-specific PAV instruments but were not significant in the model including these instruments as PAVs.”

All 5 proteins with opposite directions of AD-MR effect between platforms associated with a platform-discordant or platform-specific PAV on at least one platform.

“Notably, all 5 of these proteins associated with a platform-discordant or platform-specific PAV on at least one platform.”

The concordance of significant MR effect estimates increased when excluding first platform-discordant, then platform-discordant and platform-specific PAV instruments from MR analyses (from Pearson's r 0.48, 0.51, to 0.63).

“While we note that less proteins could be tested across analyses excluding PAVs (as these were the only instruments available for some proteins), we observed the concordance of significant MR effect estimates increase (from Pearson’s r 0.48, 0.51, to 0.63, Fig. 6).”

Adjustment for protein principal components improves power for cis-pQTL discovery on both the SomaScan and Olink platforms.

“Here, we show that adjustment for protein PCs improves power for cis-pQTL discovery on both platforms (Supplementary Fig. 5).”

Several later protein principal components likely capture impacts of trans-pQTL on both platforms.

“we performed a GWAS of the first 100 protein PCs on each platform to show that several later PCs likely capture impacts of trans-pQTL (Supplementary Fig. 6)”

Adjustment for 10 protein principal components does not maximize cis-pQTL detection.

“While adjustment for 10 protein PCs does not maximize cis-pQTL detection”

The 80 proteins with ancestry-heterogeneous inter-platform correlation estimates suggest systematic differences in measurement accuracy across ancestries, on one or both platforms.

“While the majority of individual proteins showed similar cross-platform agreement across ancestries, 80 protein targets exhibited significant differences in assay agreement across ancestries (Cochran’s Q P < 1.85 × 10−5, Bonferroni adjustment for the total number of probe pairs (2708)) (Tables 1 and 2; Supplementary Data 3), suggesting systematic differences in measurement accuracy across ancestr”

trans-pQTL associated with several protein measures may reflect non-specific affinity-probe binding or alterations to proteins involved in depositing PTMs that may interfere with affinity-probe binding.

“trans-pQTL associated with several protein measures may, among many other potential mechanisms, reflect non-specific affinity-probe binding or alterations to proteins involved in depositing PTMs that may interfere with affinity-probe binding.”

Adjustment for PAVs may improve concordance of protein measures and downstream associations.

“In summary, adjustment for PAVs may improve concordance of protein measures and downstream associations.”

cis-pQTL presence/absence may be a strong predictor of inter-platform concordance.

“Although we note that some platform-specific cis-pQTL were observed even for proteins with highly-correlated measures (Fig. 3C and Supplementary Fig. 11) (which may be due to factors such as differences in intra-assay variability, or differing sensitivities for assays to environmental factors interacting with genetic variation to regulate protein abundance), overall, we propose that cis-pQTL pres”

The LOD for an assay may impact protein quantification precision or pQTL detection, particularly for low-abundance proteins.

“The LOD for an assay may impact protein quantification precision or pQTL detection, particularly for low-abundance proteins.”

Concordant eQTL evidence may support that pQTL associations detected on one platform only capture true differences in abundance.

“Concordant eQTL evidence may support that pQTL associations detected on one platform only capture true differences in abundance.”

Adjustment for putative assay interference effects may strengthen the potential to detect true associations with protein abundances, for a subset of proteins.

“demonstrating that adjustment for putative assay interference effects may strengthen the potential to detect true associations with protein abundances, for a subset of proteins.”

Differences in protein measurements may result in differences in epidemiological associations.

“For age, sex, and BMI, we observed a small number of proteins with platform-discordant, Bonferroni significant associations on both platforms (6 to 25, Supplementary Data 17), suggesting that differences in protein measurements may indeed result in differences in epidemiological associations.”

PAV-driven cis-pQTL associations may reflect assay interference due to different affinities for encoded isoforms, particularly when the association is detected on only one platform or detected on both with different effect directions between platforms.

“As PAVs can impact affinity-probe binding, cis-pQTL associations driven by PAVs may reflect assay interference due to different affinities for encoded isoforms, particularly when the association is detected on only one platform or detected on both with different effect directions between platforms.”

80 proteins correlate differently across ancestry groups likely in part due to differing protein-altering variant frequencies by ancestry.

“We highlight 80 proteins that correlate differently across ancestry groups likely in part due to differing protein-altering variant frequencies by ancestry.”

MS pQTL findings may be used to better disentangle true biological abundance signals from signals possibly capturing isoform abundance differences, or other effects.

“Hence, MS pQTL findings may be used to better disentangle true biological abundance signals from signals possibly capturing isoform abundance differences, or other effects.”

PAVs may be one driver of inter-ancestry heterogeneity in cross-platform correlation.

“As PAVs that drive putative epitope effects may differ in frequency between ancestry groups, PAVs may be one driver of inter-ancestry heterogeneity in cross-platform correlation (Fig. 1).”

Using protein measures that do not necessarily capture overall protein abundances may mask some associations.

“Proteomics measurements are increasingly used to understand how proteins influence traits and diseases; however, using protein measures that do not necessarily capture overall protein abundances may mask some associations.”

If proteomics platforms are employed clinically, failure to account for genetic variation that does not capture overall protein abundance measures may result in systematically different measures among carriers of such genetic variants, thereby creating challenges for downstream applications in disease diagnosis/prognosis.

“Similarly, if proteomics platforms are employed clinically, failure to account for genetic variation that does not capture overall protein abundance measures may result in systematically different measures among carriers of such genetic variants, thereby creating challenges for downstream applications in disease diagnosis/prognosis.”

• AssumptionThe authors state this as an assumption: When pQTL are applied as genetic instruments in MR, they capture the effects of overall protein abundances.

Epitope-binding effects are one driver of cross-ancestry differences in platform agreement.

“Additionally, we established that proteins with cross-ancestry differences in platform agreement were enriched for association with PAVs with ancestry-differentiated allele frequencies, supporting epitope-binding effects as one driver of cross-ancestry differences in platform agreement.”

Genetic variants with different frequencies across ancestries may contribute to cross-ancestry measurement differences.

“We further demonstrate that genetic variants with different frequencies across ancestries may contribute to these measurement differences.”

Excluding these PAVs as instruments may reveal potentially causal relationships between overall protein abundances and disease outcomes.

“Hence, excluding these PAVs as instruments may reveal potentially causal relationships between overall protein abundances and disease outcomes.”

• AssumptionThe authors state this as an assumption: When pQTL are applied as genetic instruments in MR, they capture the effects of overall protein abundances.

cis-pQTL driven by coding genetic variation, such as PAVs, may in some cases reflect impacts to the target protein structure that affect protein measurements, particularly for cis-pQTL with platform-discordant directions of effect or platform-specific significance.

“However, cis-pQTL driven by coding genetic variation, such as protein altering variants (PAVs) may, in some cases, reflect impacts to the target protein structure that affect protein measurements, particularly for cis-pQTL with platform-discordant directions of effect or platform-specific significance.”

Failure to account for assay-interfering PAVs in studies of circulating proteins may hinder the detection of disease-relevant associations or putative biomarkers.

“Failure to account for assay-interfering PAVs in studies of circulating proteins may hinder the detection of disease-relevant associations or putative biomarkers.”

Environmental or other genetic effects apart from PAVs may contribute to ancestry-heterogeneity in inter-platform correlation.

“However, the improvement in inter-platform correlation across ancestry groups was not linear, and several proteins with ancestry-heterogeneity were also associated with ancestry-differentiated non-coding cis-pQTL or trans-pQTL, further suggesting the interplay of environmental or other genetic effects apart from PAVs.”

Due to sample size constraints, ancestry clusters are derived at the continental level in this study.

“Due to sample size constraints, ancestry clusters are also derived at the continental level in this study; future studies should consider heterogeneity within continents for cross-platform protein correlation and protein measurement impacting pQTL.”

Adjusting for likely assay-interference variants can, in some cases, unmask previously undetected biology.

“Moreover, we demonstrate that adjusting for likely assay-interference variants can, in some cases, unmask previously undetected biology.”

Including pQTL instruments that do not necessarily capture overall abundance effects, but rather isoform differences, may, in some cases, bias MR results towards the null.

“This suggests that including pQTL instruments that do not necessarily capture overall abundance effects, but rather isoform differences, may, in some cases, bias MR results towards the null.”

• AssumptionThe authors state this as an assumption: When pQTL are applied as genetic instruments in MR, they capture the effects of overall protein abundances.

The platform-specific PAV may, in some cases, be masking other signals (for example, from non-coding or regulatory variants) that do reflect differences in overall protein abundance.

“This is particularly surprising, as, in theory, these instruments explain less of total trait variance, supporting that the platform-specific PAV may, in some cases, be masking other signals (for example, from non-coding or regulatory variants) that do reflect differences in overall protein abundance.”

• AssumptionThe authors state this as an assumption: When pQTL are applied as genetic instruments in MR, they capture the effects of overall protein abundances.

Excluding platform-specific PAV instruments from MR models may increase power to detect significant associations for a subset of proteins.

“Furthermore, we establish that excluding platform-specific PAV instruments from MR models may increase power to detect significant associations for a subset of proteins.”

• AssumptionThe authors state this as an assumption: When pQTL are applied as genetic instruments in MR, they capture the effects of overall protein abundances.

The median coefficient of variation for all SomaScan ANML-normalized protein measures in MESA is 2.12 (IQR 1.74, 2.78), based on 1292 samples (646 pairs of split duplicates).

“The median coefficient of variation for all SomaScan ANML-normalized protein measures in MESA is 2.12 (IQR 1.74, 2.78).”