CSF proteome Mendelian randomization: colocalization gaps for ADHD drug targets

Why CSF proteome Mendelian randomization is reshaping psychiatric drug target search

Psychiatric drug development faces high late-stage failure rates partly because preclinical target hypotheses often lack genetic support [1]. Mendelian randomization using protein quantitative trait loci offers a way to test whether genetically proxied protein levels causally affect disease risk, and therapies supported by genetic evidence have higher approval probabilities [1]. Earlier CSF proteome QTL mapping established that cerebrospinal fluid protein levels are under genetic control and can be mapped genome-wide [4], while drug-target MR frameworks in multiple sclerosis demonstrated that combining cis-pQTL instruments with Bayesian colocalization can prioritize proteins with shared causal variants [5]. The new study applies this integrated approach to six mental disorders using 713 CSF proteins and GWAS summary statistics with sample sizes from 72,517 to 500,199 [1]. It reports that 17, 13, 16, 24, 36, and 27 proteins had potential causal effects on anorexia nervosa, PTSD, ADHD, major depressive disorder, schizophrenia, and bipolar disorder respectively at P < 0.05 [1].

The interpretation is that CSF proteins are plausible drug targets because they reflect central nervous system biology more directly than plasma proteins, though the study design cannot distinguish whether CSF protein changes are causes or consequences of disease without colocalization support [1]. The boundary is that MR estimates rely on instrumental variable assumptions, and the Wald ratio method used here typically applies to single-SNP instruments, limiting sensitivity analyses for pleiotropy and heterogeneity [1][5].

Colocalization separates shared genetic etiology from linkage artifacts

Bayesian colocalization tests whether two traits share the same causal variant rather than merely correlated signals due to linkage disequilibrium [1][5]. In the new study, colocalization with posterior probability > 0.70 identified PGD with anorexia nervosa, LGALS9 with PTSD, SERPING1 with major depressive disorder, AGRP with schizophrenia, and FOLH1 and HGFAC with bipolar disorder [1]. These six protein-disorder pairs represent the strongest candidates because they have both MR significance and shared genetic etiology evidence [1]. By contrast, the 16 proteins significant in MR for ADHD showed no colocalization signal, meaning none reached the posterior probability threshold for shared causal variant evidence [1]. This pattern echoes the multiple sclerosis MR study, where five of six MR-significant proteins also colocalized, and the one that did not was treated as less reliable [5].

The interpretation is that colocalization failure does not prove absence of causal effect but does weaken target prioritization because the MR signal could arise from linkage disequilibrium with a nearby causal variant affecting a different gene [1][5]. The boundary is that colocalization power depends on variant density and sample size, and the CSF pQTL dataset from 835 individuals is smaller than typical GWAS, so false negatives are possible [1].

The ADHD colocalization gap and what disorder comparisons reveal

ADHD stands out in this study not for a positive finding but for the absence of colocalization despite 16 proteins reaching MR significance [1]. The most significant ADHD protein in MR was growth hormone receptor (GHR), which the authors note has been linked to reduced exercise-induced growth hormone secretion in children with ADHD and to growth impairment [1]. However, without colocalization, GHR and the other 15 proteins remain MR-only signals that could be explained by linkage disequilibrium or pleiotropy [1]. Comparing across disorders, schizophrenia yielded the largest number of MR-significant proteins at 36, followed by bipolar disorder at 27 and major depressive disorder at 24, yet only one protein per disorder reached colocalization for schizophrenia, major depression, and anorexia nervosa, and two for bipolar disorder [1]. This suggests that MR significance counts alone overstate the number of robust targets, and colocalization acts as a stringent filter [1][5].

The interpretation is that ADHD may genuinely lack common genetic etiology between CSF proteins and disease risk in the studied sample, or the available pQTL instruments may be insufficient to detect colocalization [1]. The boundary is that the ADHD GWAS from Demontis et al. included 27 risk loci, and the CSF pQTL data came from older adults with and without Alzheimer's disease, so developmental and age-related differences could contribute to the null colocalization result [1].

DrugBank compounds and the limits of genetic target prioritization

The study identified 18 drug compounds in DrugBank that target the prioritized proteins, suggesting repurposing opportunities [1]. For anorexia nervosa, PGD catalyzes a step in the pentose phosphate pathway generating NADPH, an antioxidant protecting neurons from oxidative stress, and drugs targeting PGD exist in DrugBank [1]. For PTSD, LGALS9 is a galectin family protein involved in neuroinflammation and neuroimmune responses, with evidence of neuroprotective roles from serum studies in bipolar disorder [1]. For major depressive disorder, SERPING1 is a complement system regulator implicated in immune-inflammatory pathways and neurodegeneration [1]. For schizophrenia, AGRP neurons in the hypothalamus regulate appetite and energy expenditure and may influence hypothalamic-pituitary-adrenal axis activity [1]. For bipolar disorder, HGFAC is expressed in white matter astrocytes and may support synaptic plasticity, while FOLH1 is an enzyme linked to NAAG levels and working memory circuitry [1].

The interpretation is that genetic evidence alone cannot establish clinical efficacy, and the DrugBank matches are hypotheses for repurposing rather than proven treatments [1]. The boundary is that the study did not perform external validation in independent cohorts, unlike the multiple sclerosis MR study that replicated findings in UK Biobank and FinnGen [5], so the prioritized targets require experimental and clinical validation before translation [1].

Evidence boundaries: European ancestry, CSF sampling, and unresolved causality

All GWAS and pQTL data came from European-ancestry participants, and the CSF pQTL dataset included 835 individuals with a mean age of 69.4 years, including Alzheimer's disease cases and cognitively normal controls [1]. The study states that associations between genetic variants and protein levels were not disease-specific and not age-related, but the older adult composition limits generalization to younger populations where many mental disorders first manifest [1]. The MR approach assumes that genetic variants affect disease only through the protein of interest, and the Wald ratio method with single instruments precludes standard pleiotropy tests like MR-Egger [1][5]. Colocalization used a posterior probability threshold of 0.70, which is less stringent than the 0.80 threshold used in the multiple sclerosis study, potentially increasing false positive colocalization claims [1][5].

The interpretation is that the ADHD null colocalization result is the most clinically provocative finding because it challenges the assumption that MR-significant CSF proteins are automatically drug targets [1]. The open question is whether larger CSF pQTL datasets, trans-pQTL instruments, or multi-ancestry GWAS would reveal colocalizing proteins for ADHD or confirm the current negative result [1][4]. Ethical considerations in translating genetic causal inference to clinical contexts also warrant caution, as noted in broader MR discussions [2].

About These Sources

This research page is built on 5 peer-reviewed studies — published from 2017 to 2026, 3 from 2024 or later, collectively cited 346 times — selected as the most relevant from 9 studies that passed quality screening, drawn from 68 papers retrieved from a database of over 500 million.

Sources used in this answer