Abstract
Obesity is a major risk factor for Coronavirus disease (COVID-19) severity; however, the mechanisms underlying this relationship are not fully understood. As obesity influences the plasma proteome, we sought to identify circulating proteins mediating the effects of obesity on COVID-19 severity in humans. Here, we screened 4,907 plasma proteins to identify proteins influenced by body mass index using Mendelian randomization. This yielded 1,216 proteins, whose effect on COVID-19 severity was assessed, again using Mendelian randomization. We found that an s.d. increase in nephronectin (NPNT) was associated with increased odds of critically ill COVID-19 (OR = 1.71, P = 1.63 × 10-10). The effect was driven by an NPNT splice isoform. Mediation analyses supported NPNT as a mediator. In single-cell RNA-sequencing, NPNT was expressed in alveolar cells and fibroblasts of the lung in individuals who died of COVID-19. Finally, decreasing body fat mass and increasing fat-free mass were found to lower NPNT levels. These findings provide actionable insights into how obesity influences COVID-19 severity.
Similar content being viewed by others
Main
COVID-19 has claimed more than 6 million lives globally since the beginning of the pandemic1 and obesity increases the risk of severe COVID-19 (refs. 2,3). To date, multiple pathways have been explored as mechanisms linking obesity to COVID-19 severity, including metabolic abnormalities, systemic inflammation and respiratory dysfunction (for example, impaired gas exchange)3,4,5; however, the underlying mediators whereby obesity influences COVID-19 outcomes are not fully understood. One strategy to disentangle this relationship is to identify circulating proteins mediating the effects of obesity on COVID-19 outcomes. As circulating proteins can be measured, and in some cases modulated, the identification of mediator proteins may provide insights into the pathways whereby obesity increases the risk of severe COVID-19 and offer potential targets for therapeutic interventions.
A previous large cross-sectional study showed that body mass index (BMI) is significantly associated with changes in the plasma levels of 1,576 proteins6 and another study supported the considerable influence of obesity on the plasma proteome7; however, as proteins are intricately involved in complex biological processes, observational studies may be biased by unmeasured confounding factors and reverse causation. Considering that COVID-19 has been associated with substantial changes in the levels of circulating proteins8, the effects of circulating proteins on COVID-19 severity are also subject to such biases.
Mendelian randomization (MR) can help to protect against such biases. MR is a method that uses genetic variants as instrumental variables to evaluate the causal effects of exposures (risk factors) on outcomes. As genetic variants are randomly allocated at conception, they are largely independent of confounders, thereby decreasing the risk of confounding. Additionally, they are not subject to reverse causation because the allocation of genetic variants always precedes the onset of diseases9,10.
From the beginning of the pandemic, MR has played a critical role in providing evidence of modifiable risk factors for COVID-19 (refs. 2,11,12). For example, multiple MR studies have identified potential therapeutic targets for COVID-19, including OAS1, ABO, IFNAR2, interleukin-6, ELF5 and FAS12,13,14,15,16,17,18. Indeed, some MR findings have been validated by randomized controlled trials, thereby demonstrating the utility of MR19,20,21,22,23,24.
Nevertheless, MR is based on several instrumental variable assumptions9,10: (1) the genetic variants used as instrumental variables are associated with the exposure; (2) they are not associated with factors that confound the relationship between the exposure and the outcome and; (3) they influence the outcome only through exposure (also known as exclusion restriction). The most problematic of these assumptions is the last, as the violation of exclusion restriction can bias the causal estimate through directional horizontal pleiotropy. Careful assessment of directional horizontal pleiotropy is, therefore, required; however, with a proper selection of instrumental variables and sensitivity analyses, MR can serve as a powerful tool to help understand causal mechanisms for disease in humans9,10 and in the case of obesity and COVID-19, can be used to screen thousands of proteins that may help to explain this relationship.
Here, we integrated proteome-wide MR using the large-scale aptamer-based plasma protein measurements, multiple sensitivity analyses, colocalization, fine-mapping, single-cell RNA-sequencing (scRNA-seq) analysis and mediation analysis to identify plasma proteins mediating the effect of obesity on COVID-19 severity.
Results
Study overview and summary
The study was conducted as shown in Fig. 1.
We identified circulating proteins mediating the effect of obesity on COVID-19 severity using a two-step MR approach: First, we estimated the effect of BMI on 4,907 plasma proteins using MR, which yielded 1,216 BMI-driven proteins (step 1 MR). Second, we estimated the effect of the BMI-driven proteins on COVID-19 severity outcomes, again using MR (step 2 MR). This was followed by multiple validity assessments and follow-up analyses.
Step 1 MR
First, we estimated the effect of BMI on circulating protein levels using two-sample MR on a proteome-wide scale. Two-sample MR can estimate the causal effect of the exposure on the outcome using summary statistics of genome-wide association studies (GWAS). For this, we used a GWAS of BMI in 694,649 individuals from GIANT and the UK Biobank by Yengo et al.25 and plasma proteome GWAS from the deCODE study26 that measured plasma protein abundances using 4,907 aptamers in 35,559 individuals. Hereafter, ‘protein’ is used when referring to an aptamer targeting a protein26. After two-sample MR and sensitivity analyses, 1,216 proteins were estimated to be influenced by BMI, including nephronectin (NPNT) and hydroxysteroid 17-β dehydrogenase 14 (HSD17B14).
Step 2 MR
Next, we performed two-sample MR to estimate the causal effects of the above-identified proteins (BMI-driven proteins) on critically ill COVID-19 and COVID-19 hospitalization outcomes (collectively referred to as COVID-19 severity outcomes). For this analysis, we used cis-acting protein quantitative loci (cis-pQTLs) from the deCODE study26, thereby minimizing the risk of directional horizontal pleiotropy. For COVID-19 severity outcomes, we used GWAS data from the COVID-19 Host Genetics Initiative2. This step 2 MR identified NPNT and HSD17B14 as putatively causal proteins for the COVID-19 severity outcomes.
Validation analyses for NPNT and HSD17B14
We performed multiple validation analyses for step 1 and step 2 MR as described below.
Step 1 MR validation. To validate whether NPNT and HSD17B14 were influenced by obesity, we performed separate MR analyses using body fat percentage (another proxy measure for obesity) as the exposure and plasma protein levels as the outcomes. The MR analysis found that both NPNT and HSD17B14 were increased by body fat percentage, consistent with the step 1 MR using BMI.
We also checked whether NPNT and HSD17B14 were observationally associated with BMI using a published observational association study from INTERVAL (n = 2,729) and found that NPNT, but not HSD17B14, was positively associated with BMI.
These validation analyses collectively supported the causal effect of obesity on NPNT, but not on HSD17B14.
Step 2 MR validation. To ensure that the findings in step 2 MR were not biased by linkage disequilibrium (LD), which can reintroduce confounding, we used colocalization analyses to evaluate whether the cis-pQTLs of the identified proteins and COVID-19 severity outcomes shared a single causal variant. We found that NPNT, but not HSD17B14, shared a single causal variant with COVID-19 severity outcomes, indicating that the MR estimates for HSD17B14 in step 2 MR could have been biased by LD.
In addition, we repeated MR using cis-pQTLs from two independent studies (the FENLAND study27 and the AGES Reykjavik study28), which showed that NPNT, but not HSD17B14, influenced the COVID-19 severity outcomes. Given the lack of colocalization and replication for HSD17B14, we excluded HSD17B14 from further analyses.
For NPNT, we further investigated the consistency between MR findings and observational associations between NPNT and COVID-19 severity outcomes using the BQC19 cohort, which showed the same direction of effect as the MR analyses.
Collectively, these findings showed a causal effect of plasma NPNT levels on COVID-19 severity outcomes.
Follow-up analyses for NPNT
Colocalization of NPNT’s cis-pQTL with eQTL and sQTL. The observed differences in aptamer-measured NPNT levels may be predominantly explained by a particular isoform, rather than total NPNT levels. Thus, we used colocalization to evaluate whether the cis-pQTL shared the same causal variant with either its expression QTL (eQTL; genetic variants that explain the total RNA expression levels of NPNT) or its splicing QTL (sQTL; genetic variants that explain a specific isoform level of NPNT). We found that the cis-pQTL shared the same causal variant with the sQTL but not with the eQTL. These findings demonstrate that an NPNT splice isoform is likely measured by the aptamer targeting NPNT.
Single-cell RNA-sequencing data of SARS-CoV-2-infected lungs. Next, to gain insights into the biological role of NPNT in SARS-CoV-2-infected lungs, we analyzed scRNA-seq data of lung autopsy samples from patients who died due to COVID-19. We found that NPNT is significantly expressed in alveolar cells and fibroblasts of the lung, indicating its role in air exchange and fibrosis.
Mediation analysis. Further, we performed an MR mediation analysis to quantify the extent to which the total effect of obesity on COVID-19 severity outcomes was mediated by plasma NPNT levels. We found that NPNT partially mediated the total effect of BMI and consistent results were found for body fat percentage.
Multivariable MR analyses of body fat and fat-free mass. Finally, we estimated the independent causal effect of body fat and fat-free mass on plasma NPNT levels and COVID-19 severity outcomes using multivariable MR. We found that body fat mass increased plasma NPNT levels and the risk of COVID-19 severity outcomes, whereas fat-free mass decreased plasma NPNT levels and the risk of COVID-19 severity outcomes. These findings demonstrate that decreasing body fat and increasing body fat mass (for example, through actions such as appropriate exercise and diet) can decrease plasma NPNT levels and thus, may reduce COVID-19 severity outcomes, thereby suggesting NPNT as an actionable target.
Each of these steps is described in more detail below.
BMI to plasma proteins (step 1 MR)
To estimate the causal effect of BMI on plasma protein levels on a proteome-wide scale, we performed a two-sample MR using BMI as the exposure and 4,907 plasma protein levels as the outcomes. For two-sample MR, we used an inverse variance weighted method with a random-effects model (Methods and Supplementary Table 1). The F-statistic, which is a measure of the strength of the association between genetic variants and BMI, was 94.2 and thus did not indicate weak instrument bias (suspected when F-statistic < 10)29 (Supplementary Tables 2). Out of the 4,907 proteins screened, 1,304 were estimated to be influenced by BMI, using a Bonferroni-adjusted threshold of P < 1.0 × 10−5 (0.05 / 4,907), highlighting the substantial influence of BMI on plasma protein levels (Fig. 2 and Supplementary Tables 3 and 4).
a, Flow diagram of the step 1 MR analyses. b, Volcano plot illustrating the effect of BMI on each plasma protein from the MR analyses using inverse variance weighted method. Red and gray horizontal lines represent P = 1.0 × 10−5 (Bonferroni correction for 4,907 proteins: 0.05 / 4,907) and 0.05, respectively. A proteins’ shape denotes whether the protein passed all sensitivity tests (heterogeneity, directional pleiotropy and reverse-causation assessment) (circle) or failed any of them (triangle). c, MR scatter-plot for the effect of BMI on plasma NPNT levels. d, MR scatter-plot for the effect of BMI on plasma HSD17B14 levels. A genetically predicted increase in BMI by one s.d. was associated with increased levels of NPNT (β = 0.145, 95% CI 0.084–0.206, P = 3.03 × 10−6) and HSD17B14 (β = 0.144, 95% CI 0.085–0.202, P = 1.71 × 10−6) using the inverse variance weighted method. MR-Egger, weighted median and weighted mode methods yielded directionally consistent results with the inverse variance weighted method.
In sensitivity analyses, we tested the robustness of the MR findings with a heterogeneity test, directional pleiotropy test and reverse-causation test (Methods) to retain only the proteins that were robustly influenced by BMI. We did not find significant heterogeneity for 1,304 Bonferroni-significant proteins (I2 < 50% for all). Of 1,304 proteins, 1,229 showed no apparent sign of directional horizontal pleiotropy with the MR-Egger intercept test30 (PEgger intercept > 0.05). MR analyses with the weighted median, weighted mode and MR-Egger slope methods also showed directionally consistent results with the inverse variance weighted method for NPNT and HSD17B14 (Fig. 2 and Supplementary Table 4). To assess potential reverse causation, whereby the proteins influenced BMI, we performed bidirectional MR that used protein levels as the exposures and BMI as the outcome (Methods). Thirteen proteins exhibited bidirectional effects (Supplementary Table 5) and were excluded from further analyses. Hence, a total of 1,216 protein levels were identified as BMI-driven proteins, which were estimated to be influenced by BMI with no apparent heterogeneity, directional pleiotropy or reverse causation. We proceeded to the second step of our study (step 2 MR) with these 1,216 proteins, including NPNT and HSD17B14.
BMI-driven proteins to COVID-19 severity (step 2 MR)
Next, we evaluated the causal effects of the above-obtained BMI-driven proteins on COVID-19 outcomes, again using two-sample MR. We used cis-pQTLs for these proteins as instrumental variables and GWASs from the COVID-19 Host Genetics Initiative (release 7)2 as outcomes (Supplementary Table 1). We used cis-pQTLs (pQTLs that reside within ±1 Mb region around a transcription start site of a protein-coding gene) as the exposures to protect against bias from directional horizontal pleiotropy. This is because cis-pQTLs reside near the transcription start site of the protein-coding gene and are more likely to directly influence the protein levels than the trans-pQTLs31,32. As cis-pQTLs are likely to directly influence the transcription or translation of their associated gene, the risk of directional horizontal pleiotropy would be greatly reduced.
We searched for the cis-pQTLs for 1,216 BMI-driven proteins using the deCODE study26. Following the cis-pQTL search and data harmonization, 358 and 352 proteins were tested in MR for their estimated causal effects on critically ill COVID-19 and COVID-19 hospitalization, respectively. The F-statistics for the tested proteins were all >10, substantially reducing the risk of weak instrument bias29. F-statistics for NPNT and HSD17B14 were 252.5 and 66.3, respectively (Supplementary Table 2).
Critically ill COVID-19 and COVID-19 hospitalization outcomes are collectively referred to as COVID-19 severity outcomes. Throughout the study, we focused on these two outcomes from the COVID-19 Host Genetics Initiative and did not include COVID-19 susceptibility outcome (reported COVID-19 infection). We did so because the determinants of COVID-19 susceptibility may reflect local testing strategy and resource allocation, which pose difficulties in the interpretation of genetic findings.
Based on a Bonferroni-adjusted threshold of P < 1.40 × 10−4, MR revealed that a one s.d. increase in genetically predicted NPNT levels was associated with increased odds of critically ill COVID-19 (OR = 1.71, 95% CI 1.45–2.02, P = 1.63 × 10−10) and COVID-19 hospitalization (OR = 1.36, 95% CI 1.22–1.53, P = 4.52 × 10−8) (Fig. 3 and Supplementary Tables 6 and 7). Similarly, a one s.d. increase in genetically predicted HSD17B14 levels was associated with increased odds of critically ill COVID-19 (OR = 1.92, 95% CI 1.39–2.65, P = 6.85 × 10−5).
a, Flow diagram of the step 2 MR analyses. b,c, Volcano plot illustrating the effect of BMI-driven proteins on critically ill COVID-19 (b) and COVID-19 hospitalization (c) from the MR analyses using the inverse variance weighted method or Wald ratio when only one SNP was available as an instrumental variable. Red and blue horizontal lines represent P = 1.4 × 10−4 (Bonferroni correction for 358proteins: 0.05 of 358) and 0.05, respectively. A proteins’ shape denotes whether the protein passed (circle) all sensitivity tests (heterogeneity, directional pleiotropy and reverse causation assessment) or failed any of them (triangle). d, Forest plot of the MR results for NPNT and HSD17B14, showing the OR per one s.d. increased in plasma levels of NPNT and HSD17B14 for critically ill COVID-19 and hospitalization outcomes.
To verify the assumption of a lack of directional pleiotropy, which can reintroduce confounding, we checked whether the cis-pQTLs for NPNT and HSD17B14 were associated with any traits or diseases using PhenoScanner (http://www.phenoscanner.medschl.cam.ac.uk/)33 and Open Target Genetics (https://genetics.opentargets.org/) databases at the genome-wide significant threshold of P = 5 × 10−8. The lead cis-pQTL for NPNT (rs34712979) from the deCODE study was associated with lung-related traits (Supplementary Table 8); however, as NPNT has an established role as an extracellular matrix protein and fibrosis in the lungs34,35,36, it is possible that the NPNT cis-pQTL affects such traits by altering NPNT levels and thus should not violate the assumption of no directional pleiotropy. Indeed, MR showed that NPNT levels were estimated to influence the forced expiratory volume in 1 s (FEV1) / forced vital capacity (FVC) ratio (a key lung function index used for the definition of chronic obstructive lung disease (COPD)37) and asthma (Supplementary Table 9). This suggests that these findings may be a case of vertical pleiotropy, which does not bias MR interpretation38,39,40. Notably, the NPNT-increasing rs1662979 G allele, which increases the risk of COVID-19 severity, was found to improve lung function (higher FEV1 / FVC ratio) and decrease the risk of COPD (Supplementary Tables 8 and 9). A similar phenomenon has been reported for ELF5 and MUCB5; the COVID-19 severity risk-increasing alleles of ELF5 and MUCB5 were observed to improve lung function and decrease the risk of idiopathic pulmonary fibrosis18. This suggests that COVID-19 has distinct underlying mechanisms that influence the severity risk. No other cis-pQTL for NPNT or HSD17B14 was associated with any trait or disease, thereby reducing the possibility of directional pleiotropy. Next, to assess potential bias from reverse causation, we performed the MR-Steiger test, which supported a causal direction of plasma NPNT and HSD17B14 levels influencing COVID-19 severity outcomes (Supplementary Table 10).
Validation analyses for NPNT and HSD17B14
Step 1 MR validation
MR analyses using body fat percentage. BMI is an easy-to-measure, widely used proxy for obesity, which can offer clinically relevant information; however, another proxy for obesity, body fat percentage, can more directly measure body fat accumulation. Thus, to evaluate whether circulating NPNT and HSD17B14 levels were influenced by body fat accumulation, we repeated step 1 MR using body fat percentage as the exposure instead of BMI. We used the same 4,907 plasma protein levels GWAS from the deCODE study26 as the outcomes. The F-statistic for body fat percentage for these analyses was 61.1, which indicated no evidence of weak instrument bias.
We found that one s.d. increase in body fat percentage was associated with increased levels of NPNT (β = 0.14, 95% CI 0.07–0.22, P = 1.23 × 10−4 and HSD17B14 (β = 0.17, 95% CI 0.10–0.24, P = 3.61 × 10−6; Methods and Supplementary Table 11), consistent with the step 1 MR findings with BMI.
Comparison with observational studies from INTERVAL. One way to assess potential biases is to test the same hypothesis using a different study design. As each study design has its own inherent potential biases, similar results across designs can serve to strengthen causal inference through a triangulation of results41. Therefore, in supplementary analyses, we compared our MR findings for the effect of BMI on NPNT and HSD17B14 with published results from the INTERVAL study, which evaluated the associations between BMI and 3,622 plasma protein levels among 2,729 individuals6.
In the INTERVAL study, a one s.d. increase in BMI was cross-sectionally associated with increased NPNT (β = 0.13, 95% CI 0.09–0.17, P = 4.52 × 10−10), which was directionally consistent with the MR findings. On the contrary, HSD17B14 showed inconsistent results, wherein a one s.d. increase in BMI was not associated with HSD17B14 (β = -0.02, 95% CI −0.05–0.02, P = 0.415).
These validation analyses collectively supported the causal effect of obesity on NPNT, but not on HSD17B14.
Step 2 MR validation
Colocalization of cis-pQTLs with COVID-19 severity outcomes. Considering that the MR analyses from step 1 and step 2 implicated NPNT and HSD17B14 as candidate proteins mediating the effect of BMI on COVID-19 severity, we performed colocalization to assess whether the cis-pQTLs for NPNT and HSD17B14 shared the same single causal variant with critically ill COVID-19 and hospitalization. This can test whether MR analyses were biased by LD.
The colocalization analyses revealed that NPNT had a high posterior probability of colocalization with critically ill COVID-19 and hospitalization (posterior probability (PPshared) > 99.9% for both) (Fig. 4). We confirmed the robustness of the colocalization results using different priors (Methods and Supplementary Table 12), which consistently showed that the cis-pQTL for NPNT colocalized with COVID-19 severity outcomes. Moreover, using combined annotation-dependent depletion (CADD)-scores42 as priors for fine-mapping, we confirmed that rs34712979, the lead cis-pQTL for NPNT, was a causal variant for critically ill COVID-19 and hospitalization with a posterior inclusion probability of >0.99 (Supplementary Tables 13 and 14).
a,b, We evaluated whether the cis-pQTL for NPNT (a) and HSD17B14 (b) shared the same causal variant with critically ill COVID-19 or COVID-19 hospitalization outcomes using colocalization.
On the contrary, cis-pQTLs for HSD17B14 did not colocalize with either of the COVID-19 severity outcomes (Fig. 4), indicating that the MR findings for HSD17B14 were likely to be biased by LD.
MR analyses using cis-pQTLs from different studies
To further test whether plasma levels of NPNT or HSD17B14 were causal for critically ill COVID-19, we performed additional sets of MR analyses using cis-pQTLs from different cohorts. The cis-pQTLs for NPNT and HSD17B14 were identified using data from the FENLAND study by Pietzner et al. (n = 10,708)27 and the AGES Reykjavik study by Emilsson et al. (n = 3,200)28. We assessed the PhenoScanner and Open Target Genetics databases for additional associations for these cis-pQTLs with other phenotypes and found none.
MR analyses using these cis-pQTLs estimated a consistent causal effect of NPNT on critically ill COVID-19 (the FENLAND study, OR = 1.89, 95% CI 1.56–2.29, P = 1.21 × 10−10; the AGES Reykjavik study, OR = 1.25, 95% CI 1.06–1.48, P = 8.26 × 10−3) and COVID-19 hospitalization (the FENLAND study, OR = 1.45, 95% CI 1.28–1.66, P = 2.17× 10−8; the AGES Reykjavik study, OR = 1.17, 95% CI 1.04–1.31, P = 7.30 × 10−3).
In contrast, the estimated causal effects of HSD17B14 on these COVID-19 outcomes were not supported (Supplementary Table 15). Given the lack of colocalization and replication, we concluded that the initial finding indicating a causal effect of HSD17B14 on COVID-19 severity outcomes using a cis-pQTL from the deCODE study was likely biased by a difference in LD structure. Therefore, we excluded HSD17B14 from further evaluation.
Comparing with observational associations using BQC19
Considering that the MR analyses (step 2 MR) used plasma protein levels in a non-infectious state from the deCODE study as the exposures, in supplementary analyses, we observationally assessed whether the plasma levels of NPNT in a non-infectious state were associated with the risk of COVID-19 using logistic regression in 293 individuals from the BQC19 cohort. We restricted the analyses to individuals who met our criteria of critically ill COVID-19, COVID-19 hospitalization or COVID-19-negative controls (Methods).
Logistic regression analysis adjusting for age, sex and sample collection batch showed that increased non-infectious plasma levels of NPNT were associated with the increased risk of critically ill COVID-19 (OR = 1.47, 95% CI 1.05–2.06, P = 2.60 × 10−2) and COVID-19 hospitalization (OR = 1.49, 95% CI 1.11–2.02, P = 9.87 × 10−3), which were directionally consistent with the step 2 MR findings (Supplementary Table 16).
Follow-up analyses for the putatively causal protein
Colocalization of NPNT’s cis-pQTL with eQTL and sQTL
The observed differences in NPNT levels measured by the SomaScan assay may be explained by levels of a particular isoform, rather than total NPNT levels. Thus, we performed colocalization analyses to evaluate whether the cis-pQTL share the same causal variant with either the total RNA expression level or a specific isoform level of NPNT. Given that the lung is a primary target organ in the context of COVID-19 severity, NPNT is highly expressed in the lung, NPNT splice isoforms have previously been implicated as a risk factor for COPD and other lung outcomes43,44,45 and single-nucleotide polymorphisms (SNPs) influencing a specific isoform level would be sQTLs, we performed colocalization analysis of the cis-pQTL with eQTLs and sQTLs for NPNT in lung tissue from GTEx46.
Colocalization analysis of NPNT pQTL and sQTL within a 1-Mb region (±500 kb) surrounding the lead cis-pQTL (rs34712979) revealed that there was a high probability of pQTL and sQTL sharing a single causal variant (posterior probability for a shared causal signal (PPshared) = 100.0%); however, this was not true for the eQTL (PPshared < 0.01%; Fig. 5). We confirmed the robustness of the colocalization results using different priors (Methods and Supplementary Table 17).
We evaluated whether the cis-pQTL for NPNT shared the same causal variant with eQTL or sQTL of NPNT in the lung using colocalization.
Given that the thyroid, lung and arteries are the top three NPNT-expressing tissues according to the GTEx46, we performed the same colocalization analyses in the thyroid and arteries (aorta, tibial and coronary) and found that the pQTL of NPNT also colocalized with sQTL in these tissues (PPshared = 100.0% for all). Furthermore, it colocalized with the eQTL in the aorta and tibial artery (PPshared = 100.0% for both), but not in the thyroid or coronary artery (PPshared = 11.9% and 5.8%, respectively).
Notably, the lead cis-pQTL from the deCODE study, rs34712979, was also identified by the FENLAND study27 and has been reported to create a cryptic splice acceptor site, which inserts a three-nucleotide sequence coding a serine residue at the 5′-splice site of exon 2, resulting in perturbations of the α-helix motif43. Further, this lead cis-pQTL (rs34712979) and another cis-pQTL (rs78213340) from the AGES Reykjavik study—that was not in high LD with rs34712979 (r2 = 0.234)—were both associated with the same exon-skipping splicing in the lung in GTEx (Supplementary Table 18). Hence, two different cis-pQTLs of NPNT from three different studies (all using SomaScan) were associated with the same splicing pattern, suggesting that the SomaScan assay measures a specific isoform of the NPNT protein.
Collectively, these findings indicate that the SomaScan NPNT-targeting aptamer measures specific isoform levels and the specific isoform of NPNT with a serine insertion at the N terminus, influences the effect of NPNT on COVID-19 severity.
Single-cell RNA-sequencing data of SARS-CoV-2-infected lungs
To gain insights into the biological role of NPNT in SARS-CoV-2-infected lungs, we explored the lung cell types that significantly expressed the NPNT gene by analyzing scRNA-seq data from lung autopsy samples (106,792 cells) of 16 patients who died of COVID-19 (ref. 47) (Single-Cell Portal of the Broad Institute, accession ID SCP1052).
We found that NPNT is widely expressed in lung cell types, including epithelial cells (type 1 and type 2 alveolar cells) and fibroblasts, highlighting its role in air exchange and fibrosis. Furthermore, we conducted a subgroup analysis of SARS-CoV-2-positive and -negative cells and found that NPNT was significantly expressed in SARS-CoV-2-positive alveolar cells and fibroblasts (permutation test P < 0.001; Methods and Fig. 6).
a, NPNT expression levels of each cell type at single-cell resolution in the 16 lung donors with COVID-19. UMAP, Uniform Manifold and Projection. b, Thirty-nine annotated cell types of the lung. EC, endothelial cell; Treg, regulatory T; NK, natural killer; RBC, red blood cell. c, NPNT expression status in 106,449 SARS-CoV-2 non-infected cells (viral infection−, top) or 343 SARS-CoV-2-infected cells (viral infection+, bottom) in 16 lung donors. NPNT expression levels of 39 cell types in the two groups are shown in a box plot. In each box, the horizontal line denotes a median value of the expression levels and the asterisk inside each box denotes the mean value. Each box extends from the 25th to the 75th percentile of each group. Whiskers extend 1.5 × interquartile range from the top and bottom of the box. Log (TP10K + 1) was calculated by normalizing original gene counts by total unique molecular identifier counts, multiplying by 10,000 (TP10K) and then taking the natural logarithm.
Mediation analysis
As BMI likely has thousands of effects upon human physiology and creates an important perturbation of the proteome, we aimed to estimate the proportion of the effect of BMI that was mediated only through plasma NPNT levels. To do so, we performed a mediation analysis using network MR with the product of coefficients method to understand the extent to which plasma NPNT levels mediate the association between BMI and critically ill COVID-19 and COVID-19 hospitalization.
For critically ill COVID-19, we estimated the effect of BMI on critically ill COVID-19 mediated by plasma NPNT levels (Fig. 7). We first estimated the effect of BMI on plasma NPNT levels and then multiplied this estimate by the effect of plasma NPNT levels on critically ill COVID-19 (Methods). The ratio of the effect mediated by NPNT was calculated by dividing the NPNT-mediated effect estimate by the total effect estimate of BMI on critically ill COVID-19. We repeated the same process for COVID-19 hospitalization.
The dark blue arrow represents the total effect of BMI on critically ill COVID-19. The red arrow represents the effect of BMI on critically ill COVID-19 mediated by NPNT. For the effect of BMI on critically ill COVID-19 mediated by NPNT, the product of coefficients method calculates the proportion mediated by multiplying βBMI-to-NPNT and βNPNT-to-severity and subsequently dividing it by βBMI-to-severity, where βBMI-to-NPNT is the effect of BMI on NPNT, βNPNT-to-severity is the effect of NPNT on critically ill COVID-19, and βBMI-to-severity is the total effect of BMI on critically ill COVID-19. We evaluated the proportion mediated for the effect of obesity-related exposures (BMI, body fat percentage and body fat mass) on COVID-19 severity outcomes (critically ill COVID-19 and COVID-19 hospitalization).
We found that plasma NPNT levels partially mediated the total effect of BMI on critically ill COVID-19 (proportion mediated = 13.9%, 95% CI 6.1–21.6%, P = 4.52 × 10−4) and COVID-19 hospitalization (proportion mediated = 10.6%, 95% CI 4.4–16.7%, P = 8.02 × 10−4) (Supplementary Table 19).
In supplementary analyses, we evaluated whether NPNT mediates the total effect of body fat percentage on the COVID-19 severity outcomes. We found that plasma NPNT levels mediated the total effect of body fat percentage on critically ill COVID-19 (proportion mediated = 9.5%, 95% CI 3.6–15.4%, P = 1.59 × 10−3) and COVID-19 hospitalization (proportion mediated = 7.7%, 95% CI 2.7–12.7%, P = 2.57 × 10−3). We also found consistent results for body fat mass; plasma NPNT levels mediated the total effect of body fat mass on critically ill COVID-19 (proportion mediated = 13.4%, 95% CI 6.1–20.6%, P = 2.85 × 10−4) and COVID-19 hospitalization (proportion mediated = 9.9%, 95% CI 4.3–15.6%, P = 5.61 × 10−4) (Supplementary Table 19). All of these results consistently suggested that plasma NPNT levels partially mediated the effect of obesity, measured by BMI, body fat percentage or body fat mass, on COVID-19 severity outcomes.
Throughout the above analyses, we did not adjust for the exposure while estimating the effect of the mediator on the outcome (βNPNT-to-severity; Fig. 7) to avoid weak instrument bias (Methods). This approach was also used in previous studies48,49,50. In supplementary mediation analyses, we found that adjusting for the exposure when estimating βNPNT-to-severity (Sobel test) also support the role of NPNT as a mediator for the effect of obesity-related exposures (BMI, body fat percentage and body fat mass) on COVID-19 severity outcomes; however, the estimated proportion mediated was modest, likely due to weak instrument bias (Supplementary Table 20).
Multivariable MR analyses of body fat and fat-free mass
Given the consistent evidence that BMI influenced NPNT levels, which in turn influence COVID-19 severity, we aimed to identify a way of modulating plasma NPNT levels by gaining a better understanding of how the estimated causal effect of BMI on plasma NPNT levels was influenced by fat or fat-free mass in humans. In step 1 MR, we showed that increased BMI is estimated to increase plasma levels of NPNT; however, BMI is a function of height and weight and does not take into account body compositions, such as body fat and fat-free mass. Thus, we specifically assessed their independent effects on plasma NPNT levels. For this, we performed multivariable MR using body fat and fat-free mass as the exposures and plasma NPNT levels as the outcomes (Methods). Conditional F-statistics for body fat mass and fat-free mass were 38.8 and 59.3, respectively, which did not indicate weak instrument bias (which is suspected when F-statistics is <10) (Supplementary Table 21)51.
Multivariable MR using inverse variance weighted method found that a one s.d. increase in body fat mass was associated with increased plasma levels of NPNT (β = 0.21, 95% CI 0.14–0.28, P = 2.74 × 10−9), whereas a one s.d. increase in body fat-free mass was associated with decreased plasma levels of NPNT (β = −0.13, 95% CI −0.22, −0.05, P = 2.98 × 10−3) (Fig. 8). In sensitivity analyses, Q-statistics for instrumental validity did not suggest evidence of pleiotropy (Q-statistics = 940.8, PQ-statistics = 0.09). Multivariable MR-Egger also showed directionally consistent results and no evidence of directional pleiotropy was observed with the MR-Egger intercept test (Supplementary Table 21).
We performed multivariable MR with the inverse variance weighted method using body fat and fat-free mass as the exposures and plasma NPNT levels as the outcome.
Further, to test the influence of body fat and fat-free mass on COVID-19 severity outcomes, we conducted multivariable MR analyses using body fat mass and fat-free mass as the exposures and either critically ill COVID-19 or COVID-19 hospitalization as the outcome. We found that a one s.d. increase in body fat mass was associated with increased odds of critically ill COVID-19 (OR = 1.89, 95% CI 1.65–2.16, P = 4.83 × 10−20) and COVID-19 hospitalization (OR = 1.59, 95% CI 1.45–1.75, P = 4.28 × 10−22), whereas a one s.d. increase in body fat-free mass was associated with a decreased risk of critically ill COVID-19 (OR = 0.77, 95% CI 0.65–0.91, P = 1.94 × 10−3) and COVID-19 hospitalization (OR = 0.87, 95% CI 0.77–0.97, P = 1.57 × 10−2). Q-statistics for instrumental validity and the MR-Egger intercept test did not show evidence of pleiotropy (Supplementary Table 21).
These findings suggest that decreasing body fat mass and increasing fat-free mass (for example, through actions such as appropriate diet and exercise) can reduce plasma NPNT levels and thus reduce the risk of critically ill COVID-19, thereby indicating NPNT as an actionable target.
Discussion
In the present study, we conducted MR analyses and found that NPNT likely mediates an important proportion of the effect of obesity on COVID-19 severity. Considering that BMI is a highly polygenic trait with more than 530 associated loci25 and influences more than 1,200 circulating proteins, as shown in the present and previous studies6,7, it is remarkable that a single protein explains a reasonably large proportion of the effect of BMI and other obesity-related traits on COVID-19 severity outcomes. Additionally, colocalization analyses provided evidence that the NPNT cis-pQTL is shared with the lung sQTL for NPNT, which leads to a splice isoform with a serine insertion at the N terminus of NPNT43. These results suggest that NPNT mediates a proportion of obesity’s effect on critically ill COVID-19 and that this effect may be conferred by alternative splicing of NPNT which introduces a serine residue at its N terminus.
NPNT, or nephronectin, is an extracellular matrix protein that controls integrin binding activity. It is known as a functional ligand of integrin α8/β-1 in kidney development and is also associated with the development, remodeling and survival of various tissues through the binding of integrins52,53. NPNT has also been implicated in inflammation and autoimmunity54,55, which may align with the suggested role of obesity-induced inflammation in COVID-19 severity56. In addition, NPNT is expressed in lung alveolar cells and fibroblasts57,58 and recent GWASs have identified the same NPNT splice variant, rs34712979, to be associated with lung function-related traits (for example, FEV1, FVC and FEV1/FVC ratio) and COPD44,45. Our scRNA-seq analysis on COVID-19 lung autopsy samples also found that NPNT was expressed in SARS-CoV-2-infected alveolar cells and fibroblasts, suggesting its role in air exchange and fibrosis in SARS-CoV-2-affected lungs. These findings collectively suggest that NPNT alternative splicing, which results in an isoform with a serine insertion in the N terminus of the protein, increases the risk of deterioration of lung function and lung diseases such as COPD and COVID-19. Future studies are required to investigate the functional properties of this alternative splice isoform.
Our findings may have important clinical implications. The multivariable MR approach showed that decreasing body fat mass and increasing body fat-free mass, which can be achieved through non-pharmacological interventions such as exercise and appropriate diet59, can reduce plasma NPNT levels and the risk of COVID-19 severity outcomes. Moreover, recent trials have shown GLP-1/GIP co-agonist tirzepatide60,61 and GLP-1 receptor agonists including semaglutide62,63 and liraglutide64,65 can reduce body fat mass, while preserving body fat-free mass. We believe that these findings are notable because they offer the possibility to potentially modulate plasma NPNT levels, using available therapies. Such hypotheses require further investigations in clinical trials.
This study has both strengths and weaknesses. MR and colocalization analyses robustly implicated NPNT as a causal mediator of the relationship between BMI and COVID-19 severity. The robustness of the MR findings was enhanced by the large sample sizes used to derive the findings. This study identifies a mediator of obesity on COVID-19 employing a two-step MR approach; a previous study using a similar framework with limited statistical power earlier in the pandemic could not identify a strong protein signal66. Furthermore, our MR findings withstood multiple sensitivity analyses and were supported by colocalization, fine-mapping, replication MR, observational evaluation and RNA-seq studies in COVID-19 lung samples. Notably, multivariable MR showed that plasma NPNT levels were increased by body fat mass but decreased by fat-free mass, indicating that the effect of BMI on plasma NPNT levels was driven by body fat mass and partially counteracted by fat-free mass. These findings suggest that NPNT could be a potential intervention target in individuals with obesity to prevent critically ill COVID-19 and highlight one of the possible mechanisms by which appropriate diet and exercise can confer risk reduction of COVID-19 severity through the reduction in plasma NPNT levels.
This study also has important limitations. First, MR and colocalization analyses were restricted to individuals of European ancestry to avoid confounding by population stratification and heterogeneity of genetic associations across different ancestries. Whether NPNT mediates the effect of BMI on COVID-19 severity in populations of non-European ancestries requires further investigation. Second, we did not perform sex-stratified analysis due to the unavailability of sex-specific datasets. Third, as there were no genome-wide significant genetic instrumental variables for BMI in the 1-Mb region around the cis-region of NPNT (1-Mb around the transcription start site), we could not pinpoint a single genetic variant that directly links BMI to NPNT. Hence, the effect of BMI on NPNT is likely mediated by a combination of multiple trans-effects, which was also indicated by the scatter-plot in the MR analysis for NPNT; however, this is not surprising considering that BMI is a highly polygenic, complex trait and multiple pathways could confer the effect of BMI on other diseases67. Fourth, we do not know the molecular mechanism by which BMI influences the splice isoform. Previous studies have suggested that obesity is associated with alternative splicing68,69,70 and extracellular matrix protein remodeling71, which may align with our findings. However, further investigation is required to clarify the specific molecular mechanisms involved. Last, we do not rule out the possibility that total levels of NPNT (all isoforms) measured by the NPNT-targeting aptamer mediate the effect of obesity on COVID-19 severity; however, given the evidence provided by the MR, colocalization, fine-mapping and a biological understanding of lead cis-pQTL (the variant causes alternative splicing resulting in perturbations of the α-helix motif in the lung), it is likely that the specific isoform measured by the aptamer is driving the effect. Nevertheless, isoform-specific measurements (for example, mass spectrometry) will be required to confirm these findings.
In conclusion, we integrated a two-step MR approach, sensitivity analyses, colocalization, fine-mapping, scRNA-seq and mediation analyses to identify NPNT as a noteworthy mediator of the effect of obesity on COVID-19 severity outcomes. We also showed that decreasing body fat mass and increasing fat-free mass (for example, by actions such as exercise and appropriate diet) can lower NPNT levels and thus may improve COVID-19 severity outcomes. These findings provide actionable insights into how obesity influences COVID-19 severity.
Methods
Step 1: BMI to plasma proteins
Two-sample MR
For BMI GWAS, we used the meta-analysis with the largest sample size, consisting of 693,529 individuals of European ancestry from the GIANT consortium and the UK Biobank25. The consortium details are provided in Supplementary Table 1 and post-hoc power calculation is provided in Supplementary Table 22.
For GWAS of plasma protein levels, we used the largest proteomic GWAS available26, which measured 4,907 proteins in 35,559 individuals of European ancestry using the SomaScan assay v.4 (SomaLogic).
For two-sample MR, the effect of BMI on plasma protein levels was assessed using the inverse variance weighted method with a random-effects model in TwoSampleMR v.0.5.6 (ref. 72) (https://mrcieu.github.io/TwoSampleMR/). The instrumental variables for the exposure were defined as genome-wide significant and independent SNPs (P < 5 × 10−8; r2 < 0.001, with a clumping window of 10 Mb). SNPs in the human major histocompatibility complex (MHC) region at chromosome 6: 28,477,797–33,448,354 (GRCh37) were excluded considering its complex LD structure. We used PLINK v.1.9 (ref. 73) (http://pngu.mgh.harvard.edu/purcell/plink/) to obtain instrumental variables by clumping SNPs using the 1000 Genomes Project European reference panel74 and applying an LD threshold of r2 < 0.001. The genome-wide significant independent SNPs with the lowest P value were selected from each LD block. If instrumental variable SNPs were not present in an outcome GWAS, we used proxy SNPs (r2 > 0.8 with the original SNP). Proxy SNPs were identified using snappy v.1.0 (https://gitlab.com/richards-lab/vince.forgetta/snappy) with 1000 Genomes Project’s European reference panel74. Data harmonization and MR analyses were conducted using TwoSampleMR v.0.5.6. Specifically, data harmonization was performed with the ‘harmonise_data()’ function using default settings, including the removal of palindromic SNPs that have minor allele frequency >0.42. MR was performed using the ‘mr()’ function. A Bonferroni correction was used to set a statistical significance threshold by dividing 0.05 by the number of proteins (P = 1.0 × 10−5 (0.05 / 4,907)). We note that this correction is overly conservative as many proteins are non-independent. We did so to safeguard against false positive findings. In a heterogeneity test, we calculated I2 statistics using ‘Isq()’ function and heterogeneity P value with ‘mr_heterogeneity()’ function; results with an I2 > 50% and heterogeneity P value (Q_pval) < 0.05 were considered to be heterogeneous (substantial heterogeneity)75. For evaluating directional pleiotropy, we used the MR-Egger intercept test, which was performed using the ‘mr_pleiotropy_test()’ function; where directional pleiotropy was considered to be present when the MR-Egger intercept differed from the null (P < 0.05). We note that even in the presence of moderate heterogeneity, balanced horizontal pleiotropic effects would not violate the MR assumption of a lack of directional pleiotropy76,77. For NPNT and HSD17B14, we used MR-Egger, weighted median and weighted mode methods as additional sensitivity analyses to evaluate the directional consistency of β coefficients with the inverse variance weighted method.
For reverse MR, whereby the effects of plasma protein levels on BMI were assessed, we used cis-pQTLs from the deCODE study as the exposures and BMI GWAS as the outcome, deploying the inverse variance weighted method or Wald ratio method when only one SNP was available as an instrumental variable. After the cis-pQTL search, proxy and data harmonization, 357 proteins were tested in MR for their estimated reverse effects. Results with P < 1.4 × 10−4 (0.05 / 357; Bonferroni correction) were considered statistically significant.
We used BMI GWAS from the UK Biobank (not the meta-analysis GWAS) because multiple variants, including cis-pQTL for NPNT from the deCODE study (rs34712979), were dropped during the stringent quality control process of the meta-analysis (for example, rs34712979 was not present in the GIANT GWAS and thus dropped during the meta-analysis process).
To assess statistical power, F-statistics were calculated as previously described78 using the following formula: \(F = \frac{{R^2(n - 2 - k)}}{{(1 - R^2)k}}\) where: R2 = proportion of variance in the exposure trait and k = number of instrumental variables (Supplementary Table 2). We also performed post-hoc power calculations using an online power calculator for MR (https://sb452.shinyapps.io/power/) (Supplementary Table 22).
Step 2: BMI-driven proteins to COVID-19 severity outcomes
Two-sample MR
We identified cis-pQTLs from the GWAS of 4,907 proteins in 35,559 individuals of European ancestry from the deCODE study26. cis-pQTLs were defined as pQTLs located within 1 Mb around a transcription start site of a protein-coding gene. Further details of the dataset are provided in Supplementary Table 1. Genetic coordinates of transcription start sites of each gene used to define cis-pQTLs are provided in Supplementary Table 2 of the deCODE study26.
For COVID-19 outcomes, we used a GWAS meta-analysis from the COVID-19 Host Genetics Initiatives data release 7 (https://www.covid19hg.org/). The outcomes included critically ill COVID-19 (13,769 cases and 1,072,442 controls) and COVID-19 hospitalization (32,519 cases and 2,062,805 controls). These two outcomes were collectively referred to as COVID-19 severity outcomes. Critically ill COVID-19 was defined as a requirement for respiratory support among hospitalized individuals with laboratory-confirmed SARS-CoV-2 infection or death due to COVID-19. COVID-19 hospitalization was defined as a laboratory-confirmed SARS-CoV-2 infection that required hospitalization.
Using the data above, we carried out two-sample MR for the effects of plasma protein levels on COVID-19 severity outcomes. We used the inverse variance weighted method for proteins with two or more instrumental variables and the Wald ratio method for proteins with a single instrumental variable. When an instrumental variable SNP was not found in an outcome GWAS, we searched and used a proxy SNP (r2 < 0.8 with the original SNP) using snappy v.1.0 (https://gitlab.com/richards-lab/vince.forgetta/snappy) with a 1000 Genomes Project European reference panel74. Results with a P < 1.4 × 10−4 (0.05 / 358; Bonferroni correction with the number of proteins tested in the step 2 MR) were considered statistically significant. We used the MR-Egger intercept test to assess lack of directional pleiotropy for proteins with three or more genetic instrumental variables; however, we could not use this test for NPNT because it had fewer than three genetic instrumental variables. Therefore, we used the PhenoScanner (http://www.phenoscanner.medschl.cam.ac.uk/) and Open Targets Genetics (https://genetics.opentargets.org/) databases to test whether variants had potential pleiotropic associations with other diseases or traits. Associations with P < 5 × 10−8 were considered statistically significant. We also assessed reverse causation, wherein the effect of COVID-19 severity may influence plasma protein levels with the MR-Steiger test using ‘directionality_test()’ function from TwoSampleMR v.0.5.6 (ref. 72).
Validation analyses for proteins prioritized by step 1 and step 2 MR (NPNT and HSD17B14)
Step 1 MR validation
MR analysis using body fat percentage. We repeated the step 1 MR (described above) for NPNT and HSD17B14 using body fat percentage as the exposure instead of BMI. For this, we used body fat percentage GWAS in 454,633 individuals of European ancestry from the UK Biobank, obtained from the IEU OpenGWAS project (https://gwas.mrcieu.ac.uk/). The accession ID was ukb-b-8909.
Comparison of the MR findings with the published observational association study from INTERVAL. The INTERVAL study was a prospective cohort study conducted in England. The study recruited ~50,000 participants of primarily European ancestry without a self-reported history of any major disease. The recruitment took place between 2012 and 2014, before the COVID-19 pandemic. The study measured 3,622 plasma protein levels in 2,737 individuals using the SomaScan assay. The study performed a linear regression to evaluate associations between BMI and plasma protein levels, adjusting for age and sex6. The derived β estimates represented a normalized s.d. unit difference in each protein level per one s.d. (4.8 kg m−2) increase in BMI. Further details of the study have been described in depth previously6.
Step 2 MR validation
Colocalization of cis-pQTLs with COVID-19 severity outcomes. To identify potential bias caused by confounding owing to LD, we performed colocalization to evaluate whether cis-pQTLs shared a single causal variant between the three COVID-19 outcomes for the causal proteins and NPNT or HSD17B14. We used the coloc R package v.5.1.0 (ref. 79) (https://chr1swallace.github.io/coloc/) to evaluate all SNPs within the 1-Mb region around the top cis-pQTL for the NPNT (defined as the cis-pQTL with the smallest P value). The posterior probability of hypothesis 4 (H4) or PPshared (two traits sharing a single causal variant) > 0.8 was considered to indicate strong evidence of colocalization. Colocalization analyses were conducted using the coloc R package v.5.1.0 (ref. 79), using default priors of p1 = p2 = 10−4 and p12 = 10−5, where p1 is a prior probability that only trait 1 has a genetic association in the region, p2 is a prior probability that only trait 2 has a genetic association in the region and p12 is a prior probability that both trait 1 and trait 2 share the same genetic association in the region. To test the robustness of the results, we evaluated different combinations of priors: p1 = c(10−4, 10−5, 10−6), p2 = c(10−4, 10−5, 10−6), p12 = c(10−5, 5 × 10−6, 10−6), as performed previously18.
Fine-mapping of the NPNT region in the COVID-19 severity GWAS. The COVID-19 Host Genetics Initiatives release 7 data for those of European ancestry with critically ill COVID-19 and hospitalization was inputted into FINEMAP v.1.4 (ref. 80) (http://www.christianbenner.com/) using default options. We assumed a maximum of one causal SNP, given a single GWAS significant SNP in the NPNT region. Summary statistics were filtered for SNPs present in the European ancestry, as performed by Huffman et al.81. We set prior probabilities using scaled CADD-PHRED scores42 (https://cadd.gs.washington.edu/). The scaled CADD-PHRED scores were normalized so that their sum equaled one. The variance of effect size was set such that the maximum OR a variant can have with 95% probability is two, then scaled using the case–control ratio as described previously81.
MR analyses using cis-pQTLs for NPNT and HSD17B14 from different cohorts. We also obtained cis-pQTLs and their estimates for plasma protein levels of NPNT and HSD17B14 from the FENLAND study and AGES Reykjavik study (Supplementary Table 10). We repeated the two-sample MR using these cis-pQTLs as instrumental variables and the COVID-19 severity outcomes as described above.
Comparing with observational associations using BQC19. BQC19 (Biobanque Québécoise de la COVID-19) is a province-wide biobank that provides global access to important biological and clinical data from patients with COVID-19 and control subjects in Quebec, Canada (https://www.bqc19.ca/). Blood samples were collected in acid citrate dextrose tubes from 264 SARS-CoV-2 infectious and 463 non-infectious patients of European ancestry (Supplementary Information for the definition of infectious or non-infectious state). Detailed description of sample processing can be found in the Supplementary Information. Briefly, the samples underwent proteomic profiling on the SomaScan v.4 assay and 4,907 aptamers were used for analysis, consistent with the deCODE study26. For quality control, protein levels were natural log-transformed and then batch-corrected using the ComBat function implemented in the sva R package v.3.44.0 (ref. 82).
For logistic regression analysis in BQC19, given that the MR analyses used plasma protein levels in a non-infectious state as the exposures, we observationally assessed an association between plasma NPNT in a non-infectious state and the risk of COVID-19 severity outcomes in the BQC19 cohort. We defined the COVID-19 severity outcomes in accordance with those for the GWAS from COVID-19 Host Genetics Initiative (Supplementary Information). We performed logistic regression analysis using one of the COVID-19 outcomes as the dependent variable and standardized plasma NPNT levels as the independent variable while adjusting for sex, age and batch number. We did not adjust for clinical risk factors such as smoking and socioeconomic status.
Follow-up analyses for the putatively causal protein (NPNT)
Colocalization of NPNT’s cis-pQTL with eQTL and sQTL
We used sQTL and eQTL GWASs derived from the lung and those of thyroid and arteries (the top three NPNT-expressing tissues) in the individuals of European ancestry from the GTEx Portal v.8 dataset46 (https://gtexportal.org/).
To evaluate whether cis-pQTL is more affected by sQTL, rather than total expression, we also carried out colocalization of the cis-pQTL with eQTLs and sQTLs of NPNT using the GTEx v.8 dataset. Colocalization analyses were performed in the 1-Mb region around the cis-pQTL for NPNT (rs34712979) using ‘coloc.abf()’ function from the coloc v.5.1.0 (ref. 79) with default priors of p1 = p2 = 10−4 and p12 = 10−5 (the definitions of p1, p2 and p12 can be found above). PPshared > 0.8 was considered to indicate strong evidence of colocalization. To test the robustness of the results, we evaluated 27 different combinations of priors: p1 = c(10−4, 10−5, 10−6), p2 = c(10−4, 10−5, 10−6), p12 = c(10−5, 5 × 10−6, 10−6), as performed previously18. We also evaluated whether the lead cis-pQTL from the deCODE study (rs34712979), the FENLAND study (rs34712979) and the AGES Reykjavik study (rs78213340) were associated with the same splice pattern in the lung using GTEx. For eQTL and sQTL, we considered associations with P < 1 × 10−7 statistically significant, as did the deCODE study26.
Single-cell RNA-seq data of SARS-CoV-2-infected lungs
To understand the NPNT expression pattern in the lungs of SARS-CoV-2-infected individuals, we obtained single-cell transcriptomic data of SARS-CoV-2-infected lungs by Delorey et al.47 from the Single-Cell Portal of the Broad Institute (https://singlecell.broadinstitute.org/single_cell/) (accession ID SCP1052). The data contained 106,792 single cells from the lungs of 16 autopsy donors aged 30 to older than 89 years who died due to COVID-19.
We reanalyzed the data, focusing on NPNT expression status. We analyzed the gene expression matrix and the associated metadata using R v.4.1.2 and Seurat R package v.4.1.1 (ref. 83) (https://satijalab.org/seurat/). For visualization, NPNT expression levels were represented on a log transcript per 10 thousand + 1 (TP10K + 1) scale. To cluster the cells, we adopted the clustering annotation from the original study47. To test whether NPNT expression was enriched in specific cell types, we calculated the proportion of NPNT-expression cells in this cell type. Subsequently, we permuted the cell type labels 1,000 times and obtained the frequency (permutation P value) of the same cell type containing the same or a larger proportion of NPNT-expression cells. Additionally, we compared the NPNT expression level between a target cell type and the other cell types using Wilcoxon rank-sum test. This enrichment analysis was performed separately in SARS-CoV-2-infected and uninfected cells.
Mediation analysis
We undertook mediation analysis to calculate the proportion of the effect of BMI on critically ill COVID-19 mediated by NPNT using network MR (Fig. 7). We used the product of coefficient method to estimate the NPNT-mediated effect (the effect of BMI on critically ill COVID-19 that was accounted for by NPNT) and the same instrumental variables and outcome GWASs from step 1 and step 2 MR.
First, we estimated the effect of BMI on NPNT, then multiplied this by the effect of NPNT on critically ill COVID-19. Subsequently, the proportion of the total effect of BMI on COVID-19 mediated by NPNT was estimated by dividing the NPNT-mediated effect (βNPNT-to-severity) by the total effect (βBMI-to-severity), as described previously49,84. Additionally, we evaluated whether NPNT mediated the effect of body fat percentage on COVID-19 hospitalization and the effect of body fat mass on critically ill COVID-19 and COVID-19 hospitalization.
We used the product of coefficients method without adjusting for the exposure (BMI or body fat percentage) when estimating the effect of the mediator on the outcome (βNPNT-to-severity) to avoid weak instrument bias. This approach was also used in previous studies48,49,50. We did so because the above-mentioned exposure adjustment requires multivariable MR using NPNT and BMI (or body fat percentage; fat mass) as exposures; however, there are only two instrumental variables for NPNT (cis-pQTL), but hundreds of instrumental variables for BMI. Thus, when using multivariable MR, which includes instrumental variables from both exposures in the model, the association between plasma NPNT levels and instrumental variables would be substantially weakened (the large number of instrumental variables of BMI would decrease the strength of the association between plasma NPNT and the genetic variants). Nevertheless, in sensitivity analyses, we performed mediation analyses with adjustment for the exposure when estimating βNPNT-to-severity (Sobel test) (Supplementary Table 20). For multivariable MR, we performed data harmonization using the ‘mv_harmonise_data()’ function from TwoSampleMR v.0.5.6 (ref. 72), followed by multivariable MR causal estimation using the ‘mv_multiple()’ function from MVMR v.0.3 (https://github.com/WSpiller/MVMR)51. We calculated conditional F-statistics and heterogeneity Q-statistics using the ‘strength_mvmr()’ and ‘pleiotropy_mvmr()’ functions, respectively, again from MVMR v.0.3. To the best of our knowledge, the sample dataset of the exposure did not overlap (one was the plasma NPNT levels from the deCODE study, which is an Icelandic cohort and another was from the obesity-related traits from the UK Biobank); thus, we set ‘gencov’ to be zero when calculating these measures following the instruction by the package85, as performed previously86. We also quantified directional pleiotropy with the MR-Egger intercept test with the ‘mr_mvegger()’ function of MendelianRandomization v.0.6.0 (ref. 85) (https://github.com/cran/MendelianRandomization).
Multivariable MR of body fat and fat-free mass
We obtained GWAS of body fat and fat-free mass from the UK Biobank using the IEU OpenGWAS project (https://gwas.mrcieu.ac.uk/). The accession ID for each GWAS was ukb-b-19393 and ukb-b-13354, respectively.
For GWAS of plasma NPNT levels, we used the same GWAS from the deCODE study26 as the one used in step 1 MR, which measured plasma NPNT levels in 35,559 individuals of European ancestry by using the SomaLogic SomaScan assay v.4.
For COVID-19 severity outcomes, we used the same GWASs of critically ill COVID-19 and COVID-19 hospitalization from the COVID-19 Host Genetics Initiatives data release 7 (https://www.covid19hg.org/).
As body fat mass and fat-free mass are genetically correlated with each other (r = 0.64)87, we performed multivariable MR to estimate the independent effect of body fat and fat-free mass on plasma NPNT levels or COVID-19 severity outcomes. For instrumental variables, we identified genome-wide significant and independent SNPs for body fat and fat-free mass using the same criteria as in step 1 MR (P < 5 × 10−8 for significance and r2 < 0.001 with a clumping window of 10 Mb for independence).
SNPs in the MHC region were excluded. After data harmonization, we undertook multivariable MR with the inverse variance weighted method using a random-effects model. We used body fat and fat-free mass as the exposures and plasma NPNT levels as the outcome. Results with P < 0.025 (0.05 / 2; Bonferroni correction) were considered statistically significant. LD clumping was performed using PLINK v.1.9 (ref. 73). We performed data harmonization using the ‘mv_harmonise_data()’ function from TwoSampleMR v.0.5.6 (ref. 72), followed by multivariable MR causal estimation using the ‘mv_multiple()’ function from MVMR v.0.3 (https://github.com/WSpiller/MVMR)51. For sensitivity analyses, we first calculated genetic covariance matrix for exposures (body fat mass and fat-free mass) using the ‘phenocov_mvmr()’ function and then used ‘strength_mvmr()’ and ‘pleiotropy_mvmr()’ functions from MVMR v.0.3 (ref. 51) to calculate conditional F-statistics and Q-statistics, respectively. To calculate a phenotypic correlation matrix used in the ‘phenocov_mvmr()’ function, we used metaCCA v.1.22.0 (https://github.com/acichonska/metaCCA)88, as previously performed by Vabistsevits et al.86. Last, to perform multivariable MR-Egger analysis, we used the ‘mr_mvegger()’ function from MendelianRandomization v.0.6.0 (ref. 85) (https://github.com/cran/MendelianRandomization).
Ethical approval
For summary-level data, all contributing cohorts obtained ethical approval from their intuitional ethics review boards. The contributing cohorts include: UK Biobank, GIANT consortium, deCODE study, FENLAND study, AGES Reykjavik study, INTERVAL study, COVID-19 Host Genetics Initiative and BQC19. For individual-level data in BQC19, BQC19 received ethical approval from the Jewish General Hospital research ethics board (2020–2137) and the Centre Hospitalier de l’Université de Montréal institutional ethics board (MP-02-2020-8929, 19.389). All participants provided informed consent.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
GWAS summary statistics for each trait are available as follows:
BMI (https://portals.broadinstitute.org/collaboration/giant/).
Plasma proteome from the deCODE study (https://www.decode.com/summarydata/).
COVID-19 outcomes (https://www.covid19hg.org/results/r7/).
GTEx Portal v.8 (https://gtexportal.org/home/datasets/).
Plasma proteome from BQC19 (https://www.mcgill.ca/genepi/mcg-covid-19-biobank). Access to the data of BQC19 can be obtained upon approval of requests via bqc19.ca.
Cis-pQTLs of each study are available in the corresponding publications’ supplementary materials26,27,28.
Body fat percentage, body fat mass and fat-free mass GWASs are available at IEU OpenGWAS project with accession IDs ukb-b-8909, ukb-b-19393 and ukb-b-13354, respectively (https://gwas.mrcieu.ac.uk/).
scRNA-seq data of COVID-19 lung autopsy samples are available at the Single-Cell Portal under accession ID SCP1052 (https://singlecell.broadinstitute.org/single_cell/). CADD-scores v.1.6 can be accessed at https://cadd.gs.washington.edu/score.
Genotype data from the 1000 Genomes Project is available at https://www.internationalgenome.org/data.
Source data are provided with this paper.
Code availability
We used R v.4.1.2 (https://www.r-project.org/), TwoSampleMR v.0.5.6 (https://mrcieu.github.io/TwoSampleMR/), snappy v.1.0 (https://gitlab.com/richards-lab/vince.forgetta/snappy), coloc v.5.1.0 (https://chr1swallace.github.io/coloc/), FINEMAP R package v.1.4, Seurat v.4.0.6 (https://satijalab.org/seurat/), PLINK v.1.9 (http://pngu.mgh.harvard.edu/purcell/plink/) and GCTA fastGWA v.1.93.3 (https://yanglab.westlake.edu.cn/software/gcta/). Custom codes are available on GitHub (https://github.com/satoshi-yoshiji/TwostepMR_obesity_COVID/).
References
Johns Hopkins University. COVID-19 Global Map https://coronavirus.jhu.edu/map.html (2022).
COVID-19 Host Genetics Initiative. Mapping the human genetic architecture of COVID-19. Nature 600, 472–477 (2021).
Stefan, N., Birkenfeld, A. L., Schulze, M. B. & Ludwig, D. S. Obesity and impaired metabolic health in patients with COVID-19. Nat. Rev. Endocrinol. 16, 341–342 (2020).
Foulkes, A. S. et al. Understanding the link between obesity and severe COVID-19 outcomes: causal mediation by systemic inflammatory response. J. Clin. Endocrinol. Metab. 107, e698–e707 (2021).
Zickler, M. et al. Replication of SARS-CoV-2 in adipose tissue determines organ and systemic lipid metabolism in hamsters and humans. Cell. Metab. 34, 1–2 (2022).
Goudswaard, L. J. et al. Effects of adiposity on the human plasma proteome: observational and Mendelian randomisation estimates. Int. J. Obes. 45, 2221–2229 (2021).
Zaghlool, S. B. et al. Revealing the role of the human blood plasma proteome in obesity using genetic drivers. Nat. Commun. 12, 1279 (2021).
Filbin, M. R. et al. Longitudinal proteomic analysis of severe COVID-19 reveals survival-associated signatures, tissue-specific cell death, and cell–cell interactions. Cell. Rep. Med. 2, 100287 (2021).
Skrivankova, V. W. et al. Strengthening the reporting of observational studies in epidemiology using Mendelian randomization. JAMA 326, 1614 (2021).
Skrivankova, V. W. et al. Strengthening the reporting of observational studies in epidemiology using Mendelian randomisation (STROBE-MR): explanation and elaboration. BMJ 375, n2233 (2021).
Ponsford, M. J. et al. Cardiometabolic traits, sepsis, and severe COVID-19: a Mendelian randomization investigation. Circulation 142, 1791–1793 (2020).
Luo, S., Liang, Y., Wong, T. H. T., Schooling, C. M. & Au Yeung, S. L. Identifying factors contributing to increased susceptibility to COVID-19 risk: a systematic review of Mendelian randomization studies. Int. J. Epidemiol. 51, 1088–1105 (2022).
Zhou, S. et al. A Neanderthal OAS1 isoform protects individuals of European ancestry against COVID-19 susceptibility and severity. Nat. Med. 27, 659–667 (2021).
Gaziano, L. et al. Actionable druggable genome-wide Mendelian randomization identifies repurposing opportunities for COVID-19. Nat. Med. 27, 668–676 (2021).
Bovijn, J., Lindgren, C. M. & Holmes, M. V. Genetic variants mimicking therapeutic inhibition of IL-6 receptor signaling and risk of COVID-19. Lancet Rheumatol. 2, e658–e659 (2020).
Klaric, L. et al. Mendelian randomisation identifies alternative splicing of the FAS death receptor as a mediator of severe COVID-19. Preprint at medRxiv https://doi.org/10.1101/2021.04.01.21254789 (2021).
Niemi, M. E. K., Daly, M. J. & Ganna, A. The human genetic epidemiology of COVID-19. Nat. Rev. Genet. 23, 533–546 (2022).
Pietzner, M. et al. ELF5 is a potential respiratory epithelial cell-specific risk gene for severe COVID-19. Nat. Commun. 13, 4484 (2022).
RECOVERY Collaborative Group. Tocilizumab in patients admitted to hospital with COVID-19 (RECOVERY): a randomised, controlled, open-label, platform trial. Lancet 397, 1637–1645 (2021).
Holmes, M. V., Richardson, T. G., Ference, B. A., Davies, N. M. & Davey Smith, G. Integrating genomics with biomarkers and therapeutic targets to invigorate cardiovascular drug development. Nat. Rev. Cardiol. 18, 435–453 (2021).
Manousaki, D., Mokry, L. E., Ross, S., Goltzman, D. & Richards, J. B. Mendelian Randomization Studies Do Not Support a Role for Vitamin D in Coronary Artery Disease. Circ. Cardiovasc. Genet. 9, 349–356 (2016).
Jiang, X. et al. Circulating vitamin D concentrations and risk of breast and prostate cancer: a Mendelian randomization study. Int. J. Epidemiol. 48, 1416–1424 (2019).
Meng, X. et al. Phenome-wide Mendelian-randomization study of genetically determined vitamin D on multiple health outcomes using the UK Biobank study. Int. J. Epidemiol. 48, 1425–1434 (2019).
Manson, J. E. et al. Vitamin D Supplements and Prevention of Cancer and Cardiovascular Disease. N. Engl. J. Med. 380, 33–44 (2019).
Yengo, L. et al. Meta-analysis of genome-wide association studies for height and body mass index in ∼700000 individuals of European ancestry. Hum. Mol. Genet. 27, 3641–3649 (2018).
Ferkingstad, E. et al. DECODE: Large-scale integration of the plasma proteome with genetics and disease. Nat. Genet. 53, 1712–1721 (2021).
Pietzner, M. et al. Synergistic insights into human health from aptamer- and antibody-based proteomic profiling. Nat. Commun. 12, 6822 (2021).
Emilsson, V. et al. Co-regulatory networks of human serum proteins link genetics to disease. Science 361, 769–773 (2018).
Lawlor, D. A., Harbord, R. M., Sterne, J. A. C., Timpson, N. & Davey Smith, G. Mendelian randomization: Using genes as instruments for making causal inferences in epidemiology. Stat. Med. 27, 1133–1163 (2008).
Bowden, J., Davey Smith, G. & Burgess, S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int. J. Epidemiol. 44, 512–525 (2015).
Swerdlow, D. I. et al. Selecting instruments for Mendelian randomization in the wake of genome-wide association studies. Int. J. Epidemiol. 45, 1600–1616 (2016).
Zhang, J. et al. Plasma proteome analyses in individuals of European and African ancestry identify cis-pQTLs and models for proteome-wide association studies. Nat. Genet. 54, 593–602 (2022).
Staley, J. R. et al. PhenoScanner: a database of human genotype-phenotype associations. Bioinformatics 32, 3207–3209 (2016).
Tsukui, T. et al. Collagen-producing lung cell atlas identifies multiple subsets with distinct localization and relevance to fibrosis. Nat. Commun. 11, 1920 (2020).
Xie, T. et al. Single-cell deconvolution of fibroblast heterogeneity in mouse pulmonary fibrosis. Cell Rep. 22, 3625–3640 (2018).
Obeidat, M. E. et al. Molecular mechanisms underlying variations in lung function: a systems genetics analysis. Lancet Respir. Med. 3, 782–795 (2015).
Bhatt, S. P. et al. Discriminative accuracy of FEV1:FVC thresholds for COPD-related hospitalization and mortality. JAMA 321, 2438–2447 (2019).
Davies, N. M., Holmes, M. V. & Davey Smith, G. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. BMJ 362, k601 (2018).
Holmes, M. V., Ala-Korpela, M. & Smith, G. D. Mendelian randomization in cardiometabolic disease: challenges in evaluating causality. Nat. Rev. Cardiol. 14, 577–590 (2017).
Au Yeung, S. L., Li, A. M., He, B., Kwok, K. O. & Schooling, C. M. Association of smoking, lung function and COPD in COVID-19 risk: a two-step Mendelian randomization study. Addiction 117, 2027–2036 (2022).
Lawlor, D. A., Tilling, K. & Davey Smith, G. Triangulation in aetiological epidemiology. Int. J. Epidemiol. 45, 1866–1886 (2016).
Kircher, M. et al. A general framework for estimating the relative pathogenicity of human genetic variants. Nat. Genet. 46, 310–315 (2014).
Saferali, A. et al. Characterization of a COPD-associated NPNT functional splicing genetic variant in human lung tissue via long-read sequencing. Preprint at medRxiv https://doi.org/10.1101/2020.10.20.20203927 (2020).
Shrine, N. et al. New genetic signals for lung function highlight pathways and chronic obstructive pulmonary disease associations across multiple ancestries. Nat. Genet. 51, 481–493 (2019).
Sakornsakolpat, P. et al. Genetic landscape of chronic obstructive pulmonary disease identifies heterogeneous cell-type and phenotype associations. Nat. Genet. 51, 494–505 (2019).
GTEx Consortium. The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science 369, 1318–1330 (2020).
Delorey, T. M. et al. COVID-19 tissue atlases reveal SARS-CoV-2 pathology and cellular targets. Nature 595, 107–113 (2021).
Woolf, B., Zagkos, L. & Gill, D. TwoStepCisMR: a novel method and R package for attenuating bias in cis-Mendelian randomization analyses. Genes 13, 1541 (2022).
Burgess, S., Daniel, R. M., Butterworth, A. S., Thompson, S. G. & Consortium, E. P.-I. Network Mendelian randomization: using genetic variants as instrumental variables to investigate mediation in causal pathways. Int. J. Epidemiol. 44, 484–495 (2015).
Relton, C. L. & Davey Smith, G. Two-step epigenetic Mendelian randomization: a strategy for establishing the causal role of epigenetic processes in pathways to disease. Int. J. Epidemiol. 41, 161–176 (2012).
Sanderson, E., Spiller, W. & Bowden, J. Testing and correcting for weak and pleiotropic instruments in two-sample multivariable Mendelian randomization. Stat. Med. 40, 5434–5452 (2021).
Brandenberger, R. et al. Identification and characterization of a novel extracellular matrix protein nephronectin that is associated with integrin α8β1 in the embryonic kidney. J. Cell Biol. 154, 447–458 (2001).
Morimura, N. et al. Molecular cloning of POEM: a novel adhesion molecule that interacts with α8β1 integrin. J. Biol. Chem. 276, 42172–42181 (2001).
Inagaki, F. F. et al. Nephronectin is upregulated in acute and chronic hepatitis and aggravates liver injury by recruiting CD4-positive cells. Biochem. Biophys. Res. Commun. 430, 751–756 (2013).
Kon, S., Honda, M., Ishikawa, K., Maeda, M. & Segawa, T. Antibodies against nephronectin ameliorate anti-type II collagen-induced arthritis in mice. FEBS Open Bio. 10, 107–117 (2020).
O’Rourke, R. W. & Lumeng, C. N. Pathways to severe COVID-19 for people with obesity. Obesity 29, 645–653 (2021).
Strunz, M. et al. Alveolar regeneration through a Krt8+ transitional stem cell state that persists in human lung fibrosis. Nat. Commun. 11, 3559 (2020).
Xie, T. et al. Mesenchymal growth hormone receptor deficiency leads to failure of alveolar progenitor cell function and severe pulmonary fibrosis. Sci. Adv. 7, eabg6005 (2021).
Stiegler, P. & Cunliffe, A. The role of diet and exercise for the maintenance of fat-free mass and resting metabolic rate during weight loss. Sports Med. 36, 239–262 (2006).
Jastreboff, A. M. et al. Tirzepatide once weekly for the treatment of obesity. N. Engl. J. Med. 387, 205–216 (2022).
Heise, T. et al. Effects of subcutaneous tirzepatide versus placebo or semaglutide on pancreatic islet function and insulin sensitivity in adults with type 2 diabetes: a multicentre, randomised, double-blind, parallel-arm, phase 1 clinical trial. Lancet Diabetes Endocrinol. 10, 418–429 (2022).
Garvey, W. T. et al. Two-year effects of semaglutide in adults with overweight or obesity: the STEP 5 trial. Nat. Med. 28, 2083–2091 (2022).
Blundell, J. et al. Effects of once-weekly semaglutide on appetite, energy intake, control of eating, food preference and body weight in subjects with obesity. Diabetes Obes. Metab. 19, 1242–1251 (2017).
Pi-Sunyer, X. et al. A randomized, controlled trial of 3.0 mg of liraglutide in weight management. N. Engl. J. Med. 373, 11–22 (2015).
Grannell, A. et al. Liraglutide does not adversely impact fat-free mass loss. Obesity 29, 529–534 (2021).
Richardson, T. G., Fang, S., Mitchell, R. E., Holmes, M. V. & Davey Smith, G. Evaluating the effects of cardiometabolic exposures on circulating proteins which may contribute to severe SARS-CoV-2. eBioMedicine 64, 103228 (2021).
Locke, A. E. et al. Genetic studies of body mass index yield new insights for obesity biology. Nature 518, 197–206 (2015).
Pihlajamaki, J. et al. Expression of the splicing factor gene SFRS10 is reduced in human obesity and contributes to enhanced lipogenesis. Cell Metab. 14, 208–218 (2011).
Zhao, X. et al. FTO-dependent demethylation of N6-methyladenosine regulates mRNA splicing and is required for adipogenesis. Cell Res. 24, 1403–1419 (2014).
Roundtree, I. A., Evans, M. E., Pan, T. & He, C. Dynamic RNA modifications in gene expression regulation. Cell 169, 1187–1200 (2017).
Kim, M., Lee, C. & Park, J. Extracellular matrix remodeling facilitates obesity-associated cancer progression. Trends Cell Biol. 32, 825–834 (2022).
Hemani, G. et al. The MR-Base platform supports systematic causal inference across the human phenome. eLife 7, e34408 (2018).
Purcell, S. et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am. J. Hum. Genet. 81, 559–575 (2007).
The 1000 Genomes Project Consortium. A global reference for human genetic variation. Nature 526, 68–74 (2015).
Deeks, J. J. et al. in Cochrane Handbook for Systematic Reviews of Interventions 2nd edn (eds Higgins, J. P. T. et al.) Ch. 10 (The Cochrane Collaboration, 2019).
Hemani, G., Bowden, J. & Davey Smith, G. Evaluating the potential role of pleiotropy in Mendelian randomization studies. Hum. Mol. Genet. 27, R195–R208 (2018).
Burgess, S., Bowden, J., Fall, T., Ingelsson, E. & Thompson, S. G. Sensitivity analyses for robust causal inference from mendelian randomization analyses with multiple genetic variants. Epidemiology 28, 30–42 (2017).
Pierce, B. L., Ahsan, H. & Vanderweele, T. J. Power and instrument strength requirements for Mendelian randomization studies using multiple genetic variants. Int. J. Epidemiol. 40, 740–752 (2011).
Giambartolomei, C. et al. Bayesian test for colocalisation between pairs of genetic association studies using summary statistics. PLoS Genet. 10, e1004383 (2014).
Benner, C. et al. FINEMAP: efficient variable selection using summary data from genome-wide association studies. Bioinformatics 32, 1493–1501 (2016).
Huffman, J. E. et al. Multi-ancestry fine mapping implicates OAS1 splicing in risk of severe COVID-19. Nat. Genet. 54, 125–127 (2022).
Leek, J. T., Johnson, W. E., Parker, H. S., Jaffe, A. E. & Storey, J. D. The sva package for removing batch effects and other unwanted variation in high-throughput experiments. Bioinformatics 28, 882–883 (2012).
Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587 (2021).
Carter, A. R. et al. Understanding the consequences of education inequality on cardiovascular disease: Mendelian randomisation study. BMJ 365, l1855 (2019).
Grant, A. J. & Burgess, S. Pleiotropy robust methods for multivariable Mendelian randomization. Stat. Med. 40, 5813–5830 (2021).
Vabistsevits, M. et al. Deciphering how early life adiposity influences breast cancer risk using Mendelian randomization. Commun. Biol. 5, 337 (2022).
Yoshiji, S. et al. Causal associations between body fat accumulation and COVID-19 severity: a Mendelian randomization study. Front. Endocrinol. 13, 899625 (2022).
Cichonska, A. et al. metaCCA: summary statistics-based multivariate meta-analysis of genome-wide association studies using canonical correlation analysis. Bioinformatics 32, 1981–1989 (2016).
Acknowledgements
We thank the COVID-19 Host Genetics Initiative for providing the latest summary statistics for COVID-19 outcomes. We acknowledge S. Wang, T. Khosroheidari, L. Cuddeback, W. Schwarzmann and D. Denmark at SomaLogic for constructive discussions. We acknowledge Biorender (biorender.com) for providing materials used to create the illustrative diagram. The Richards research group is supported by the Canadian Institutes of Health Research (grants 365825, 409511, 100558 and 169303), the McGill Interdisciplinary Initiative in Infection and Immunity (MI4), the Lady Davis Institute of the Jewish General Hospital, the Jewish General Hospital Foundation, the Canadian Foundation for Innovation, the National Institutes of Health Foundation, Genome Quebec, the Public Health Agency of Canada, McGill University, Cancer Research UK (grant no. C18281/A29019) and the Fonds de Recherche Quebec Santé (FRQS). J.B.R. is supported by an FRQS Mérite Clinical Research Scholarship. The support from Calcul Quebec and Compute Canada is acknowledged. TwinsUK is funded by the Welcome Trust, the Medical Research Council, the European Union, the National Institute for Health Research-funded BioResource and the Clinical Research Facility and Biomedical Research Centre based at Guy’s and St Thomas’ NHS Foundation Trust in partnership with King’s College London. S.Y. is supported by the Japan Society for the Promotion of Science. T.L. is supported by a Vanier Canada Graduate Scholarship, an FRQS doctoral training fellowship and a McGill University Faculty of Medicine Studentship. G.B.-L. is supported by scholarships from the FRQS, the Canadian Institutes of Health Research and Quebec’s Ministry of Health and Social Services. Y.C. is supported by an FRQS doctoral training fellowship and the Lady Davis Institute/TD Bank Studentship Award. M.H. is supported by grants from the SciLifeLab/Knut and Alice Wallenberg national COVID-19 research program (KAW 2020.0182 and KAW 2020.0241), the Swedish Heart-Lung Foundation (20210089, 20190639 and 20190637) and the Swedish Society of Medicine (SLS-938101). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Author information
Authors and Affiliations
Contributions
Conception and design was the responsibility of S.Y. and J.B.R. Methodology was overseen by S.Y., T.L. and J.B.R. Data analysis was performed by S.Y., T.L., J.D.S.W., C.-Y.S. and J.B.R. Visualization was conducted by S.Y. and T.L. Writing of the original draft was carried out by S.Y. Review and editing was carried out by S.Y., G.B.-L., T.L., J.D.S.W., C.-Y.S., T.N., D.R.M., Y.C., K.L., M.H., Y.I., Z.A., S.L., N.D., C.D., M.V., C.T., X.X., M.B., F.S., L.L., H.M.M., M.A., J.A., V.M., N.J.T., H.Z., S.Z., V.F., Y.F. and J.B.R.
Corresponding author
Ethics declarations
Competing interests
J.B.R. has served as an advisor to GlaxoSmithKline and Deerfield Capital. The institution of J.B.R. has received investigator-initiated grant funding from Eli Lilly, GlaxoSmithKline and Biogen for projects unrelated to this research. J.B.R. is the CEO of 5 Prime Sciences (www.5primesciences.com), which provides research services for biotech, pharma and venture capital companies for projects unrelated to this research. T.L. and V.F. are employees of 5 Prime Sciences. T.N. has received speaking fees from Boehringer Ingelheim and AstraZeneca regarding the projects unrelated to this research. The other authors declare no competing interests.
Peer review
Peer review information
Nature Metabolism thanks Shiu Lun Ryan Au Yeung and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Yanina-Yasmin Pesch, in collaboration with the Nature Metabolism team.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Supplementary Information
Supplementary Information.
Supplementary Tables
Supplementary Tables 1–22.
Source data
Source Data Fig. 2
Statistical source data.
Source Data Fig. 3
Statistical source data.
Source Data Fig. 4
Statistical source data.
Source Data Fig. 5
Statistical source data.
Source Data Fig. 6
Statistical source data.
Source Data Fig. 8
Statistical source data.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Yoshiji, S., Butler-Laporte, G., Lu, T. et al. Proteome-wide Mendelian randomization implicates nephronectin as an actionable mediator of the effect of obesity on COVID-19 severity. Nat Metab 5, 248–264 (2023). https://doi.org/10.1038/s42255-023-00742-w
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s42255-023-00742-w
This article is cited by
-
Proteomic insights into the associations between obesity, lifestyle factors, and coronary artery disease
BMC Medicine (2023)
-
From target discovery to clinical drug development with human genetics
Nature (2023)
-
Actionable targets to reduce COVID-19 severity
Nature Metabolism (2023)
-
Alternative splicing in lung influences COVID-19 severity and respiratory diseases
Nature Communications (2023)
-
Colocalization of expression transcripts with COVID-19 outcomes is rare across cell states, cell types and organs
Human Genetics (2023)
