The genetic regulation of protein expression in cerebrospinal fluid

Abstract

Studies of the genetic regulation of cerebrospinal fluid (CSF) proteins may reveal pathways for treatment of neurological diseases. 398 proteins in CSF were measured in 1,591 participants from the BioFINDER study. Protein quantitative trait loci (pQTL) were identified as associations between genetic variants and proteins, with 176 pQTLs for 145 CSF proteins (P < 1.25 × 10^-10 , 117 cis-pQTLs and 59 trans-pQTLs). Ventricular volume (measured with brain magnetic resonance imaging) was a confounder for several pQTLs. pQTLs for CSF and plasma proteins were overall correlated, but CSF-specific pQTLs were also observed. Mendelian randomization analyses suggested causal roles for several proteins, for example, ApoE, CD33, and GRN in Alzheimer's disease, MMP-10 in preclinical Alzheimer's disease, SIGLEC9 in amyotrophic lateral sclerosis, and CD38, GPNMB, and ADAM15 in Parkinson's disease. CSF levels of GRN, MMP-10, and GPNMB were altered in Alzheimer's disease, preclinical Alzheimer's disease, and Parkinson's disease, respectively. These findings point to pathways to be explored for novel therapies. The novel finding that ventricular volume confounded pQTLs has implications for design of future studies of the genetic regulation of the CSF proteome.

Keywords: Mendelian randomization; biomarkers; cerebrospinal fluid; genetic regulation; pQTL.

Figure 1. Study flowchart

A schematic overview of the study design.

Figure 2. CSF pQTL genomic map

Each point represents a significant pQTL for a CSF protein. The sizes of the bubbles are proportional to the β‐coefficient of the effects. An interactive version of the plot is available as source data for Fig EV1.

Figure 3. CSF pQTL mapping

1. Relationships between pQTL effect (β) and minor allele frequency (MAF) for CSF pQTLs. cis‐pQTLs and trans‐pQTLs are indicated (only pQTLs significant after Bonferroni correction are included).

2. For all Bonferroni significant CSF cis‐pQTLs, the degree of significance is shown by the distance from transcription start site (TSS). The interquartile range was −37.3 to 0.12 Kb. The most significant pQTLs are annotated.

3. Functional annotation of genetic variants for CSF pQTLs, generated by FUMA, for CSF pQTLs that were significant after Bonferroni correction. Enrichment of functional consequences of SNPs was tested against the European 1,000 genome reference panel. Asterixes indicate significant differences between proportions for significant CSF pQTLs versus the reference panel (***P < 0.001; *P < 0.05).

Figure EV1. Interactive pQTL genomic map for CSF pQTLs

This is an interactive version of main Fig 2. Full interactive functionality is provided in the Source Data file “Fig EV1 CSF_pQTL_interactive.html”, which is available online. Source data are available online for this figure.

Figure EV2. CSF biomarkers measured with orthogonal methods

1. A–D

   Between‐assay correlations for CSF biomarkers measured with both OLINK methods (proximity extension assay) and orthogonal methods, for four proteins where this data was available (panel A: CHI3L1, panel B: CCL2, panel C: CCL4, panel D: NFL). pQTLs were identified for CHI3L1, CCL2 and CCL4, as described in the main manuscript. pQTLs identified by the orthogonal methods are included in Dataset EV2 (using alternative protein labels for the alternative assays: MCP1 for CCL2, MIP1b for CCL4, and YKL‐40 for CHI3L1, rows marked yellow). For CCL2, the same genetic variant was identified (rs2228467, trans‐pQTL) with both assays. For CCL4, one trans‐pQTLs identified by proximity extension assay was validated (rs113341849) and one cis‐pQTL (rs879571071) was identified, which was in LD with a cis‐pQTL identified for the proximity extension assay (rs8064426, R² = 0.200, D′ = 0.687). For CHI3L1, one cis‐pQTL was also identified (rs4950928), which was in high LD with a cis‐pQTL identified for the proximity extension assay (rs946262, R² = 0.902, D′ = 1.0). For all these cases, the effect sizes of the pQTLs were similar, with stable direction of effects.

Figure EV3. pQTLs in CSF and plasma

The figure shows the general relationship between pQTLs in CSF and plasma, for all pQTLs that were at least genome‐wide significant (P < 5e‐8) in CSF. The only pQTL which was significant after Bonferroni correction in CSF and genome‐wide significant in plasma, where the direction of effect differed between the tissues, was a cis‐pQTL for CXCL1, where the variant was associated with lower levels in CSF and higher in plasma.

Figure 4. Overview of CSF pQTL architecture

1. Number of pQTLs per CSF protein (panel A), and number of CSF proteins per top genetic variant among the identified pQTLs (panel B).

Figure EV4. Replication of CSF pQTLs from Yang et al (2021)

This figure shows a comparison between pQTLs that were identified both in the current study and in another recent publication on CSF pQTLs (Yang et al, 2021). Full interactive functionality is provided in the Source Data file “Fig EV4 Replication CSF pQTLs.html”, which is available online. Source data are available online for this figure.

Figure 5. eQTLs and CSF pQTLs for corresponding genetic variants and proteins

The plot includes significant CSF pQTLs and eQTLs, from the meta‐analysis by Sieberts et al, . Most of the significant variants had concordant directions as eQTLs and pQTLs. Co‐localization is indicated by color of text and shapes (probability 0.85 or higher shown in blue with dark shapes, probability below 0.85 shown in red with gray shapes).

Figure 6. Associations between CSF proteins and ventricle volume

1. Associations for all CSF proteins with volumes of lateral ventricles. Associations significant after multiple comparisons are indicated in blue. Test assumptions were verified by visual inspection of diagnostic plots. Selected significant proteins are annotated.

2. Enrichment analysis for CSF proteins associated with ventricles.

3. Enrichment analysis for CSF proteins not associated with ventricles.

Figure EV5. Effects on CSF pQTL from adjustment for ventricle volume

This figure shows the change in CSF pQTL β‐coefficients with and without additional adjustment for ventricle volume and intracranial volume. pQTLs involving genetic variants in the GMNC‐OSTN region on chromosome 3 are highlighted in blue. Note that these analyses were done on a subset of participants with MRI data. The plot (and corresponding Dataset EV14) is restricted to CSF protein‐pQTL pairs that were at least genome‐wide significant in the full cohort and at least significant with P < 5 × 10⁻⁶ in the MRI subcohort. Full interactive functionality is provided in the Source Data file “Fig EV5 Percent_Change_beta_ICV_VV.html”, which is available online. Source data are available online for this figure.