Proteogenomics in cerebrospinal fluid and plasma reveals new biological fingerprint of cerebral small vessel disease

Abstract

Cerebral small vessel disease (cSVD) is a leading cause of stroke and dementia with no specific treatment, of which molecular mechanisms remain poorly understood. To identify potential biomarkers and therapeutic targets, we applied Mendelian randomization to examine over 2,500 proteins measured in plasma and, uniquely, cerebrospinal fluid, in relation to magnetic resonance imaging (MRI) markers of cSVD in more than 40,000 individuals. Here we show that 49 proteins are associated with MRI markers of cSVD, most prominently in cerebrospinal fluid. We highlight associations that are consistent across platforms and ancestries, and supported by complementary observational analyses, and we explore differences between fluids. The proteins are enriched in pathways related to the extracellular matrix, immune response and microglial activity. Many also associate with stroke and dementia, and several correspond to existing drug targets. Together, these findings reveal a robust biological fingerprint of cSVD and highlight opportunities for biomarker and drug discovery and repositioning.

Fig. 1. Summary of the analysis plan.

‘Summary level’ and ‘individual level’ correspond to analyses conducted using summary-level-based and individual-level-based datasets, respectively. Created with BioRender.com.

Fig. 2. Discovery protein–cSVD associations in CSF and plasma using cis-pQTL MR.

a, Volcano plots of proteins associated with WMHs using cis-pQTL MR in CSF. b, Volcano plots of proteins associated with PVS burden using cis-pQTL MR in CSF. c, Volcano plots of proteins associated with WMHs using cis-pQTL MR in plasma. d, Volcano plots of proteins associated with PVSs using cis-pQTL MR in plasma. Each dot represents the MR results for proteins using inverse-variance weighted (IVW) analysis when multiple instrumental variables available, or the Wald ratio when only one instrument was available. Benjamini–Hochberg P_FDR values are represented in this graph. Represented proteins are significantly associated with MRI marker at P_FDR (Benjamini–Hochberg FDR threshold) < 0.05. The dashed line in each volcano plot represents the corrected threshold after additionally correcting for the number of phenotypes tested (P < 0.0125). e, Venn diagram of identified causal proteins associated with MRI-cSVD. An asterisk denotes proteins identified in plasma; a dagger symbol denotes proteins associated in both plasma and CSF; other proteins are associated in CSF.

Fig. 3. Summary of protein–cSVD associations in discovery and follow-up (lifespan, cross-fluid, cross-platform and cross-ancestry) analyses.

a, Heatmap of protein–cSVD associations using CSF-based MR analyses as discovery. b, Heatmap of protein–cSVD associations using plasma-based MR analyses as discovery. 1. Discovery MR using cis-pQTLs from CSF (a) and plasma (b) and testing their association with WMH volume and PVS burden in the largest meta-analysis of GWAS. 2. Lifespan follow-up MR using cis-pQTLs from CSF (a) and plasma (b) and testing their association with WMHs and PVSs in young adults (i-Share study). 3. Cross-fluid follow-up MR using cis-pQTLs from plasma (a) and CSF (b) and testing their association with WMHs and PVSs in the other fluid than the discovery findings. 4. Cross-platform follow-up using plasma individual-level proteomic data measured with the Olink platform in independent European-ancestry samples (3C-Dijon and UK Biobank studies). 5. Cross-ancestry follow-up using plasma individual-level proteomic data measured with the SomaScan platform in an independent East Asian-ancestry sample (Nagahama study). Columns 1, 2 and 3 represent the direction of effect size from IVW or Wald ratio MR depending on the number of instrumental variables. Columns 4 and 5 represent the direction of effect size from two-sided linear regression analyses adjusted for age, sex, batch, total intracranial volume for WMHs and the first four principal components of population stratification. Dark squares correspond to significant results after FDR correction (P_FDR< 0.05). The asterisk corresponds to significant associations after additional correction for the four phenotypes tested (*P_FDR< 0.0125). Hatched squares correspond to nominal associations (P < 0.05). Orange squares correspond to a positive association (higher protein levels being associated with higher cSVD burden) and green squares correspond to a negative association. FU, follow-up. The hash symbol denotes results of the 3C-Dijon analysis only. Proteins in bold are those showing at least one nominal association (P < 0.05) in the same direction in follow-up analyses. The exact P values are reported in Supplementary Tables 2 and 9 (a) and 4 and 10 (b).

Fig. 4. Clinical significance of protein–cSVD findings in CSF and plasma.

a, Forest plot of protein–cSVD associations with stroke (N = 73,652/1,234,808) and its subtypes (ischemic stroke, N = 62,100/1,234,808; small vessel stroke, N = 6,811/1,234,808; and intracerebral hemorrhage, N = 1,545/1,481) using IVW or Wald ratio MR. b, Forest plot of protein–cSVD association with Alzheimer’s disease (N = 71,880/383,378) using IVW or Wald ratio MR. c, Forest plot of protein–cSVD association with stroke and dementia using IVW meta-analysis of two-sided cause-specific Cox models adjusting for age, sex, self-reported ancestry and educational attainment (for incident dementia) of 3C-Dijon and UK Biobank studies (N = 54,108; 1,440/1,555 incident stroke and dementia cases). All proteins associated with MRI-cSVD identified in the discovery analysis in CSF and plasma were used for this analysis. Full lines represent proteins measured in CSF. Dashed lines represent proteins measured in plasma. Proteins associated at least at P < 0.05 for at least one of the outcomes tested are represented (for stroke, associations with all (sub)types are represented when one or more was significant). Asterisks denote results that are significant after multiple-testing correction (P_FDR< 0.05). Dots correspond to effect estimates (odds ratio (a and b) and hazard ratio (c)), and errors bars correspond to 95% confidence intervals.

Fig. 5. Proteomics-driven drug discovery.

a, Drug-discovery analysis conducted using CSF protein–cSVD MR IVW or Wald ratio estimates for WMH and PVS findings. b, Drug-discovery analysis conducted using plasma protein–cSVD MR IVW or Wald ratio estimates for WMHs. Proteins in yellow correspond to proteins associated with the MRI-cSVD marker in CSF and in red in plasma, in discovery analyses. An asterisk denotes proteins with associations in at least one of the follow-up modalities (at P < 0.05). Red arrows correspond to a protective effect of a protein on MRI-cSVD (reducing cSVD burden) or an inhibitory effect of a drug on the cSVD-associated protein; blue arrows correspond to a deleterious effect of a protein on MRI-cSVD (promoting cSVD burden) or an analog effect of a drug on the cSVD-associated protein. Drugs in orange cross the BBB. mAb, monoclonal antibody.

Fig. 6. Integrated summary of our findings.

Proteins associations with WMH, PVS or both are represented in the middle. For each MRI marker, the left side corresponds to CSF findings and the right side to plasma findings. An asterisk denotes proteins with cross-ancestry association. A hash symbol denotes proteins with lifespan association. Associations with stroke, dementia or both (P_FDR< 0.05) in either MR or observational analysis are represented on the left of the figure. Subtypes of stroke are as follows: AS, any stroke; IS, ischemic stroke. Minus and plus signs correspond to the direction of association referring to higher level of the protein. Blue plus or minus signs correspond to findings in CSF and pink in plasma. Empty plus or minus signs correspond to a situation where opposite directions were observed in the same tissue using MR and observational study. Drug repositioning is represented on the right of the figure. (i) Proteins associated with the same MRI-cSVD marker in cross-fluid follow-up (for cSVD-associated proteins identified in CSF discovery: showing significant association in plasma follow-up; for cSVD-associated proteins identified in plasma discovery: showing significant association in CSF follow-up); (ii) CSF-specific proteins (showing no significant association in plasma follow-up); (iii) plasma-specific proteins (showing a nonsignificant association in CSF follow-up); (iv) no follow-up available. AD, Alzheimer’s disease. Created with BioRender.com.

Extended Data Fig. 1. Discovery protein-cSVD associations in CSF and plasma using cis-pQTL mendelian randomization.

A. String plot of proteins associated with WMH. B. String plot of proteins associated with PVS (WM, BG and HIP). Network nodes represent proteins: colored nodes query proteins and first shell of interactors. Edges represent protein-protein associations. Cyan and pink edges are known interactions, cyan: from curated databases, and pink: experimentally determined. Green and blue edges correspond to predicted interactions. Green: gene neighborhood, and blue: gene co-occurrence. Purple corresponds to protein homology, yellow to text mining and black to co-expression.

Extended Data Fig. 2. Multivariable Mendelian randomization (MVMR) exploring the modifying effect of systolic blood pressure on the association of A. CSF proteins and B plasma proteins with MRI-cSVD.

Only proteins with >1 pQTL available for analysis are represented. Dots correspond to effect estimates (beta) and error bars to 95% confidence intervals. Forest plots correspond respectively to protein associations with WMH, WM-PVS, HIP-PVS and BG-PVS. Red lines correspond to primary MR results; blue lines correspond to multivariable MR (MVMR).

Extended Data Fig. 3. Genetic correlation between proteins associated with MRI-cSVD using protein quantitative trait loci.

A. Genetic correlation across pQTLs for 24 CSF protein levels associated with MRI-cSVD. B. Genetic correlation across pQTLs for 21 CSF or plasma proteins significantly associated with MRI-cSVD and available in both plasma and CSF (x axis: pQTL for CSF protein levels; y axis: pQTL for plasma protein levels). C. Genetic correlation across pQTLs for 9 plasma protein levels associated with MRI-cSVD. Genetic correlations were estimated using LD score regression. P-values are based on two-sided Wald tests of the null hypothesis, without adjustment for multiple comparisons. * p < 0.05, ** p < 0.01, *** p < 0.001. Only proteins that converged for genetic correlation analyses are displayed.

Extended Data Fig. 4. Correlation of protein levels measured in the UKB across the 26 cSVD-associated proteins.

Correlations were estimated using Spearman correlation (two-sided). Unadjusted pvalues are displayed. *p < 7.7×10-3 (Bonferroni corrected threshold).

Extended Data Fig. 5. Association of plasma protein levels with MRI-cSVD stratified on hypertension status.

Forest plots correspond to protein associations with WMH and HIP -PVS. Results corresponds to meta-analyses of association results in 3C-Dijon and UK Biobank (N = 6,581; 2,088 hypertensive/3,406 non-hypertensive) for WMH; and 3C-Dijon only for HIP-PVS (N = 1,087; 235 hypertensive/852 non-hypertensive), stratified on hypertension status: hypertensive (HTN), non-hypertensive (Non-HTN), all combined (Overall) and adjusted on systolic blood pressure (HTN adjusted). Dots represent effect estimates (beta for WMH and odds ratio for HIP-PVS) and error bars correspond to 95% confidence intervals.

Extended Data Fig. 6. Protein-cSVD associations with dementia subtypes (vascular and Alzheimer’s disease): meta-analysis of 3C-Dijon and UK Biobank.

Results of cause-specific Cox models (Methods). N-vascular dementia=385; N-Alzheimer=1,107). Dots represent hazard ratios and errors bars corresponds to 95% confidence intervals.

Extended Data Fig. 7. Comparison of effect size estimates for MR associations of cSVD-associated proteins with ischemic stroke and small vessel stroke between Europeans (EUR) and East-Asians (EAS).

A. Ischemic Stroke. B. Small vessel Stroke.

Extended Data Fig. 8. Cell-type enrichment in single cell RNA-seq databases using STEAP.

Upset plot displays the number of significant enrichment results by protein (pQTL) horizontally and by cell-type vertically. CSF pQTLs are in black and plasma pQTLs are in blue. Details are displayed in Supplementary Table 25. Human and mouse single-cell databases are used in this analysis (Methods).

Extended Data Fig. 9. Single-nucleus gene expression/enrichment analyses.

A. Single-nucleus cerebrovascular gene expression data of cSVD-associated protein coding genes in dorsolateral prefrontal cortex (ROS-MAP study). B. Enrichment analyses of cSVD-associated protein coding genes in microglial states and vascular cells using in silico vascular enrichment (two-sided). Unadjusted pvalues are displayed. Dotted line corresponds to pval<0.05.

Extended Data Fig. 10. Distribution of MRI markers of cSVD in 3C-Dijon and UK-Biobank.

Histogram of white matter hyperintensities (WMH) and perivascular spaces (PVS) distribution after normal inverse transformation in 3C-Dijon (A- C) and the UK Biobank (D-F).