Mendelian randomization analysis of the brain, cerebrospinal fluid, and plasma proteome identifies potential drug targets for attention deficit hyperactivity disorder

Summary

Background

The need for new therapeutics for attention deficit hyperactivity disorder (ADHD) is evident. Brain, cerebrospinal fluid (CSF), and plasma protein biomarkers with causal genetic evidence could represent potential drug targets. However, a comprehensive screen of the proteome has not yet been conducted.

Methods

We employed a three-pronged approach using Mendelian Randomization (MR) and Bayesian colocalization analysis. Firstly, we studied 608 brains, 214 CSF, and 612 plasma proteins as potential causal mediators of ADHD using MR analysis. Secondly, we analysed the consistency of the discovered biomarkers across three distinct subtypes of ADHD: childhood, persistent, and late-diagnosed ADHD. Finally, we extended our analysis to examine the correlation between identified biomarkers and Tourette syndrome and pervasive autism spectrum disorder (ASD), conditions often linked with ADHD. To validate the MR findings, we conducted sensitivity analysis. Additionally, we performed cell type analysis on the human brain to identify risk genes that are notably enriched in various brain cell types.

Findings

After applying Bonferroni correction, we found that the risk of ADHD was increased by brain proteins GMPPB, NAA80, HYI, CISD2, and HYI, TIE1 in CSF and plasma. Proteins GMPPB, NAA80, ICA1L, CISD2, TIE1, and RMDN1 showed overlapped loci with ADHD risk through Bayesian colocalization. Overexpression of GMPPB protein was linked to an increase in the risk for all three ADHD subtypes. While ICA1L provided protection against both ASD and ADHD, CISD2 increased the probability of both disorders. Cell-specific studies revealed that GMPPB, NAA80, ICA1L, and CISD2 were predominantly present on the surface of excitatory-inhibitory neurons.

Interpretation

Our comprehensive MR investigation of the brain, CSF, and plasma proteomes revealed seven proteins with causal connections to ADHD. Particularly, GMPPB and TIE1 emerged as intriguing targets for potential ADHD therapy.

Funding

This work was partly funded by the Key R & D Program of Zhejiang (T.L. 2022C03096); the National Natural Science Foundation of China Project (C.Z. 82001413); Postdoctoral Foundation of West China Hospital (C.Z. 2020HXBH163).

Research in context.

Evidence before this study

Prior to this study, the search for ADHD biomarkers and potential therapeutic targets was limited in its exploration of the proteomic landscape. There was a notable gap in understanding the molecular mediators at the proteomic level that could be causal in the pathogenesis of ADHD and its subtypes.

Added value of this study

Utilizing a three-pronged approach involving Mendelian Randomization (MR) and Bayesian colocalization, our research has identified seven key proteins with causal connections to ADHD, including GMPPB and TIE1, which are potential targets for therapeutic intervention. The analysis of proteins across different biological mediums (brain, CSF, and plasma) and the consistency of these biomarkers across three ADHD subtypes represent a novel contribution. Furthermore, the study extends our understanding of the proteomic relationships between ADHD and comorbid conditions such as ASD and Tourette syndrome, offering insights into shared pathophysiological mechanisms.

Implications of all the available evidence

The identification of specific proteins linked to ADHD offers new avenues for the development of targeted therapies, moving beyond the current focus on symptom management. Overall, this study lays the groundwork for a new era in ADHD therapeutics, where treatments are more personalized and based on a deeper molecular understanding of the disorder.

Introduction

Attention deficit hyperactivity disorder (ADHD) is one of the most common neuropsychiatric disorders, affecting 5%–6% of children,¹ with an adult prevalence rate of 2.5%–4.9%.¹ A significant proportion of affected individuals continue to experience symptoms into adulthood.¹ ADHD is characterised by symptoms such as impulsivity, hyperactivity, and inattention inappropriate for a person's age.² Genetic studies, including twin and adoption studies, have determined that ADHD is approximately 76% heritable.³ Despite current guidelines recommending first-line treatment with methylphenidate, atomoxetine, or amphetamine for children and adolescents, these treatments fail in up to 50% of paediatric patients with ADHD due to insufficient symptom improvement or adverse side effects.^4,5 Consequently, the development of new pharmacological targets is imperative. Previous studies have identified several proteins linked to ADHD risk as potential therapeutic targets.^6,7 However, due to the high attrition rate in drug development, there is an ongoing need to investigate proteins implicated in ADHD pathogenesis before proceeding to clinical trials.⁸ Nelson et al. demonstrated that drug targets with disease associations backed by genetic evidence have a two-fold higher likelihood of achieving market approval.⁹ Additionally, recent technological advancements in high-throughput protein quantification have allowed genome-wide association studies (GWAS) to concurrently identify genetic determinants of various brain, cerebrospinal fluid, and blood proteins.10, 11, 12 Therefore, integrating genetic and proteomic data through Mendelian randomization (MR) analysis could offer a powerful approach to discovering new, effective drug targets for ADHD.¹³

MR is a statistical genetic framework designed to assess the causal relationship between an exposure (in this case, a protein) and an outcome (i.e., ADHD).¹⁴ MR has been used to evaluate the causality of selected plasma proteins for ADHD, including beta-mannosidase (MANBA)¹⁵ and the fragile X mental retardation protein (FMRP).¹⁶ Yet, no comprehensive scans of the brain, CSF, and plasma proteome have been performed to identify novel mediators of ADHD. In this study, we employed MR to (1) systematically screen 608 brain, 214 CSF, and 612 plasma proteins to identify novel targets for ADHD; (2) explore whether identified risk proteins share genetic factors associated with the onset of ADHD and their relationship with autism spectrum disorder (ASD) and Tourette syndrome; and (3) verify the MR findings via reverse causality detection and Bayesian colocalization analysis, and conduct cell-type specific analysis to further explore potential drug targets.

Methods

pQTL for brain, CSF, and plasma

Fig. S1 presents a schematic description of the research design. For the primary analysis, we utilized brain pQTL data from a study conducted by Wingo et al.¹⁷ The included pQTLs met the following criteria: 1) demonstrated genome-wide significant association (P < 5 × 10⁻⁸); 2) showed independent association (linkage disequilibrium clumping r² < 0.001 within a 10,000 kb window); 3) featured robust SNPs as determined by F-statistics ≥10. In order to satisfy the requirement that the genetic variant not be directly associated with the outcome, we searched through the GRASP v2.0 database for each SNP to determine if it was significantly associated with the outcome (P < 5 × 10⁻⁸).¹⁸ As our two-sample MR were conducted on two independent datasets (i.e., no overlapping sample between exposure and outcome), it is valid to assume confounding was minimized.¹⁹ Ultimately, 616 cis-pQTLs were identified for 608 proteins. The CSF pQTL data was retrieved from the study by Yang et al.,²⁰ adhering to the screening criteria utilized for the brain pQTL dataset, which resulted in 233 cis-acting SNPs for 214 proteins. Lastly, the plasma pQTL data was obtained from the study by Zheng et al.²¹ that incorporated five previously published GWASs. The primary analysis included 616 cis-acting SNPs for 612 proteins (Table S1).

GWAS summary statistics

Our primary outcomes utilized GWAS summary statistics for ADHD from the iPSYCH + deCODE + PGC data repository. They employed a fixed-effects model via METAL software, analysing data from the largest available ADHD GWAS to date, which included 38,691 cases and 186,843 controls.²² GWAS summary statistics for ADHD subtypes: childhood (N_case = 14,878), persistent (N_case = 1473) and late-diagnosed ADHD (N_case = 6961), alongside 38,303 controls, were accessed through the iPSYCH.²³ We also incorporated GWAS summary statistics for ASD (N_case = 18,381, N_control = 27,969)²⁴ and Tourette syndrome (TS, N_case = 4819, N_control = 9488),²⁵ conditions potentially genetically overlapped with ADHD.

More detailed information about covariate adjustments of GWAS datasets is summarized in the Supplementary Materials of the original study.22, 23, 24, 25

Statistical analysis

Mendelian randomization (MR) analysis of ADHD

We used SNPs related to brain, CSF, and plasma proteins as exposures and the target disorders as the outcome to conduct MR using “TwoSampleMR” (https://github.com/MRCIEU/TwoSampleMR). The Wald ratio approach was employed when a protein only had one pQTL, otherwise the inverse variance weighted (IVW) analysis method was applied.²⁶ Due to the limited number of IVs, post-MR analyses such as Cochran's Q test for heterogeneity, MR-Egger intercept test for horizontal pleiotropy, MRPRESSO test for the outlier detection, and I²GX test for “no measurement error” were not feasible.^27,28 For the primary analysis, we adjusted for multiple testing using Bonferroni correction and set a threshold of 0.05/number of proteins analysed to prioritize the results for further analysis. The corrected thresholds were set at P < 8.22 × 10⁻⁵ in the brain, P < 2.34 × 10⁻⁴ in CSF, and P < 8.17 × 10⁻⁵ in plasma.

Targeted MR analysis of ADHD subtypes and two related disorders

Given the possibility that risk proteins may vary or remain consistent depending on when ADHD initially manifests (e.g., childhood, persistent, and late-diagnosed cases),^23,29 we examined the potential causal role of risk proteins identified in the main ADHD dataset in relation to the risk of childhood, persistent, and late-diagnosed ADHD. Considering that both TS and ASD share clinical and genetic risk factors with ADHD,³⁰ we also employed MR to assess the causal influence of ADHD biomarkers on these two related disorders. The SNP-biomarker effect was calculated using the risk proteins identified in the primary ADHD dataset, and the SNP-outcome effect was derived using the GWAS summary datasets for ADHD subtypes, ASD, and TS.

Reverse causality detection

To verify the causal direction between identified proteins and ADHD, Steiger filtering was employed.^10,31 A P-value <0.05 was considered statistically significant. The above steps were conducted using the "TwoSampleMR" R package (github.com/MRCIEU/TwoSampleMR).

Bayesian colocalization analysis

Bayesian colocalization analyses were carried out using the coloc package (https://chr1swallace.github.io/coloc/index.html) to assess the probability that two traits share the same cause variation. As previously mentioned, Bayesian colocalization provides the posterior probability for five hypotheses on whether a single variable is shared across two features. We evaluated the posterior probability of hypothesis 4 (PPH4), wherein the protein and ADHD were associated with the region through shared variants. A gene was classified as having evidence of colocalization based on gene-based PPH4>80%,³² as evaluated by at least one algorithm and the coloc. abf approach.

External validation and consistency comparison

In order to further validate our identified brain proteins that cause ADHD, we performed a proteome-wide association study (PWAS) of the genetic effects of ADHD using protein weights of postmortem brain tissue collected in the dorsolateral prefrontal cortex from the Religious Orders Study/Memory and Aging Project (Rosmap) and the Banner Sun Health Research Institute (Banner). Specifically, FUSION was used to determine the impact of SNPs on protein abundance for brain proteins with significant heritability (P < 0.01) using a variety of prediction models, including as top1, BLUP, LASSO, ENET, and BSLMM. The weights that were employed in this study were from Wingo et al.¹⁷ The genetic effects of ADHD (ADHD GWAS z-score) could be integrated with brain protein mass using FUSION, which needed a linear combination of z-score × weight of independent SNPs at each locus. After that, by applying a summary-data-based Mendelian Randomization (SMR)³³ to pQTL and ADHD, the risk protein results found by two-sample MR were further verified. A careful unadjusted P < 0.05 from HEIDI was utilized to demonstrate the possible impact of linkage on significant SMR outcomes.

Cell-type specificity analysis

We investigated cell type-specific expression of the cause genes in the brain34, 35, 36 with a human brain single-cell RNA sequencing (RNA-seq) dataset.³⁷ As described on the website, individual layers of the cortex were dissected, and nuclei were dissociated and sorted from human brain tissues using the neuronal marker NeuN. SMART-Seq v4 or 10X Genomics Chromium Single Cell 3′ v3 RNA-seq was used to characterize expression in postmortem and neurosurgical donor brain nuclei. We employed CELLEX (CELL-type EXpression-specificity), a method for generating cell-type expression specificity (ES) profiles, to derive gene expression specificity values.³⁷ The CELLEX methodology inherently accounts for statistical measures of specificity, where a higher ES score indicates more significant enrichment in a particular cell type, compared to a null distribution of expression profiles across cell types.

Ethics statement

Ethical approval was granted for each of the original studies.^17,20, 21, 22, 23, 24, 25^,37 In addition, no individual-level data were used in this study. Therefore, no new ethical review board approval was required.

Role of funders

This work was supported in part by grants from Key R & D Program of Zhejiang, National Natural Science Foundation of China Project and Postdoctoral Foundation of West China Hospital. They had no role in the study design, data collection, data analyses, interpretation or writing of this report.

Results

Screening the proteome for ADHD causal proteins

MR analysis, at Bonferroni significance, revealed 7 protein-ADHD pairs (Table 1, Table 2 and Fig. 1a–c). This includes Mannose-1-phosphate guanyltransferase beta (GMPPB), Islet cell autoantigen 1-like protein (ICA1L), N-alpha-acetyltransferase 80 (NAA80), Putative hydroxypyruvate isomerase (HYI), and CDGSH iron-sulfur domain-containing protein 2(CISD2) in the brain. Tyrosine-protein kinase receptor Tie-1 (TIE1) in the CSF and TIE1 and Regulator of microtubule dynamics protein 1 (RMDN1) in the plasma were identified. Specifically, up-regulated GMPPB (OR = 2.06; 95% CI, 1.54–2.76, P = 1.04 × 10⁻⁶), NAA80 (OR = 2.25; 95% CI = 1.60–3.15; P = 2.46 × 10⁻⁶), HYI (OR = 1.77; 95% CI = 1.37–2.28; P = 1.07 × 10⁻⁵), CISD2 (OR = 3.53, 95% CI = 1.98–6.28; P = 1.87 × 10⁻⁵), TIE1_CSF (OR = 4.75, 95% CI = 2.49–9.05, P = 2.14 × 10⁻⁶), and TIE1_Plasma (OR = 1.22; 95% CI = 1.13–1.31; P = 1.36 × 10⁻⁷) increased the risk of ADHD, while ICA1L (OR = 0.34; 95% CI = 0.22–0.53; P = 2.09 × 10⁻⁶) and RMDN1 (OR = 0.89; 95% CI = 0.84–0.94; P = 6.64 × 10⁻⁵) decreased the risk of ADHD (Table 1, Table 2).

Table 1.

Summary of significant brain proteins representing causal mediators for ADHD.

Tissue	Protein	Two-sample MR analysis					Bayesian colocalization analysis	Steiger filtering analysis		SMR P	HEIDI P	PWAS1		PWAS2
		Method	SNP	Effect allele	OR (95% CI)	P	PPH4	Passed	P			Z	P	Z	P
Brain	GMPPB	Wald ratio	rs6809879	G	2.06 (1.54–2.76)	1.04E-06	97.60%	TRUE	4.02E-17	2.17E-05	5.93E-02	4.57	4.86E-06	4.9	9.68E-07
Brain	ICA1L	Wald ratio	rs7582720	C	0.34 (0.22–0.53)	2.09E-06	98.70%	TRUE	3.87E-13	6.57E-05	4.62E-01	−4.75	2.01E-06	−4.49	7.18E-06
Brain	NAA80	Wald ratio	rs13100173	A	2.25 (1.60–3.15)	2.46E-06	97.90%	TRUE	1.19E-07	4.12E-04	7.08E-01	4.72	2.38E-06	–	–
Brain	HYI	Wald ratio	rs6954	T	1.77 (1.37–2.28)	1.07E-05	0.40%	TRUE	6.00E-14	1.33E-04	2.54E-02	5.02	5.12E-07	3.8	1.45E-04
Brain	CISD2	Wald ratio	rs223452	T	3.53 (1.98–6.28)	1.87E-05	88.80%	TRUE	9.23E-11	3.16E-04	1.73E-01	4.26	2.05E-05	4.34	1.41E-05

All displayed results surpassed correction for multiple hypotheses testing (P < 8.22 × 10⁻⁵). Odds ratios are expressed in terms of risk per 1–standard deviation increase in biomarker levels–Protein not profiled in the confirmation proteomic dataset. MR, Mendelian randomization; GMPPB, Mannose-1-phosphate guanyltransferase beta; ICA1L, Islet cell autoantigen 1-like protein; NAA80, N-alpha-acetyltransferase 80; HYI, Putative hydroxypyruvate isomerase; CISD2, CDGSH iron-sulfur domain-containing protein 2. PWAS¹ and PWAS² are proteome-wide association studies that made use of brain protein weights from the Religious Orders Study/Memory and Aging Project and the Banner Sun Health Research Institute.

Table 2.

Summary of significant plasma and CSF proteins representing causal mediators for ADHD.

Tissue	Protein	Two-sample MR analysis					SMR P	HEIDI P	Bayesian colocalization analysis	Steiger filtering analysis
		Method	SNP	Effect allele	OR (95% CI)	P			PPH4	Passed	P
CSF	TIE1	Wald ratio	rs3768046	A	4.75 (2.49–9.05)	2.14E-06	2.96E-04	6.19E-01	91%	TRUE	3.46E-08
Plasma	TIE1	Wald ratio	rs2275180	A	1.22 (1.13–1.31)	1.36E-07	2.95E-06	NA	91.30%	TRUE	2.63E-24
Plasma	RMDN1	Wald ratio	rs11781016	A	0.89 (0.84–0.94)	6.64E-05	1.37E-04	NA	85.70%	TRUE	3.31E-39

The results of MR in CSF (P < 2.34 × 10−4) and plasma (P < 8.17 × 10−5) were adjusted by Bonferroni correction.MR, Mendelian randomization; CSF, cerebrospinal fluid; TIE1, Tyrosine-protein kinase receptor Tie-1; RMDN1, Regulator of microtubule dynamics protein 1. NA (not applicable) indicates an undetermined result because the number of pQTL SNPs were too small for HEIDI to test.

Fig. 1.

MR results for brain, CSF and plasma proteins and the risk of ADHD. Volcano plots of the MR results for (a) 608 brain (b) 214 CSF and (c) 612 plasma proteins on the risk of ADHD. a, b and c show MR analysis with Wald ratio method on brain, CSF and plasma proteins on the risk of ADHD, respectively. OR for increased risk of ADHD were expressed as per SD increase in plasma protein levels and 10-fold increase in CSF protein levels. Dashed horizontal black line corresponded to P < 8.22 × 10⁻⁵ in brain, P < 2.34 × 10⁻⁴ in CSF and P < 8.17 × 10⁻⁵ in plasma.

Characterizing the effects of identified ADHD proteins on subtypes and ASD

We then investigated whether the risk proteins found in ADHD were present in childhood, persistent, and late-diagnosed ADHD. We discovered that four brain risk proteins (GMPPB, CISD2, HYI, and ICA1L) and one plasma protein, RMDN1, had similar causal effects in childhood and adulthood-diagnosed ADHD (Fig. 2 and Table S2). Genetically determined levels of GMPPB (OR = 1.9; 95% CI, 1.23–2.93, P = 3.8 × 10⁻³), CISD2 (OR = 8.88; 95% CI, 3.67–21.47, P = 1.27 × 10⁻⁶), HYI (OR = 1.62; 95% CI, 1.11–2.38, P = 1.35 × 10⁻²), TIE1_Plasma (OR = 1.31; 95% CI, 1.17–1.46, P = 3.48 × 10⁻⁶), and TIE1_CSF (OR = 8.67; 95% CI, 3.24–23.21, P = 1.69 × 10⁻⁵) increased the risk of childhood-diagnosed ADHD, whereas ICA1L (OR = 0.36; 95% CI, 0.19–0.7, P = 2.67 × 10⁻³) and RMDN1(OR = 0.83; 95% CI, 0.76–0.91, P = 4.79 × 10⁻⁵) decreased the risk. No significant proteins were detected in persistent ADHD except for GMPPB (Table S2). Next, we investigated whether identified proteins mediated the risk of ASD and TS. Significant associations were found for CISD2 and ICA1L, with genetically determined levels of CISD2 increasing the risk of ASD and ICA1L decreasing the risks. No significant associations were observed for TS (Fig. S2 and Table S3).

Fig. 2.

Association between identified biomarkers and risk for ADHD subtypes. Associations right the black midline represent risk-conferring effects, and those left the black midline represents protective effects. Four brain risk proteins (GMPPB, CISD2, HYI and ICA1L) and one plasma protein RMDN1 had similar causal effects in childhood and adulthood-diagnosed ADHD. ∗Nominally significant (P < 0.05).

Sensitivity analysis for causal proteins

Steiger Filtering further confirmed directionality (Table 1, Table 2). Bayesian colocalization analysis strongly suggested that GMPPB (PPH4 = 0.976), NAA80 (PPH4 = 0.979), ICA1L (PPH4 = 0.987), CISD2 (PPH4 = 0.888), TIE1 (PPH4 in CSF = 0.91, PPH4 in plasma = 0.913), and RMDN1 (PPH4 = 0.857) shared the same variant with ADHD (Table 1, Table 2). Upon phenotype scanning with GRASP V2.0, the identified SNP-related risk proteins were not associated with ADHD (Table S4).

Comparison of analyzed proteins and external validation

We further sought external validation for the proteins initially identified in the brain through a proteome-wide association study (PWAS) using a p-value threshold of 0.05. The integrative analysis applied to different datasets revealed that GMPPB (P_Rosmap = 4.86 × 10⁻⁶, P_Banner = 9.68 × 10⁻⁷), ICA1L (P_Rosmap = 2.01 × 10⁻⁶, P_Banner = 7.18 × 10⁻⁶), HYI (P_Rosmap = 5.12 × 10⁻⁷, P_Banner = 1.45 × 10⁻⁴), and CISD2 (P_Rosmap = 2.05 × 10⁻⁵, P_Banner = 1.41 × 10⁻⁵) are associated with ADHD (Tables S5 and S6). We also used SMR to externally validate the two-sample MR-identified proteins, and comprehensive SMR analysis verified that seven risk proteins in the brain, cerebrospinal fluid, and plasma are significantly associated with ADHD (Table 1, Table 2).

Cell-type specificity of risk gene in the human brain

We then evaluated the cell-type specificity of these proteins within the brain. An enrichment analysis using human single-cell RNA-seq data from the Cell Types database revealed a cell-type specificity in expressing the four risk genes. Specifically, CISD2 demonstrated a proteome-wide significant association with ADHD and was predominantly enriched in inhibitory neurons. On the other hand, GMPPB, ICA1L, and NAA80 showed expression in excitatory neurons (Fig. 3).

Fig. 3.

Single-cell-type expression of the potential risk genes. Bar graph of single-cell-type enrichment for risk genes. The diagram depicts CELL-type EXpression specificity (y-axis) for each gene (x-axis), with evidence of substantial enrichment within a specific brain cell type (histogram of the bar).

Discussion

In our investigation, we leveraged proteomic data from the brain, CSF, and plasma to explore potential causative proteins for ADHD. We identify seven proteins as potential targets for ADHD treatment. Among these, TIE1 was detected in the CSF, RMDN1 in circulation, and GMPPB, NAA80, HYI, CISD2, and ICA1L were found within the brain tissue. Notably, our subtype analysis of ADHD—including childhood-diagnosed, persistent, and late-diagnosed forms—consistently highlighted abnormalities in the GMPPB protein across all subtypes. Additionally, the detection of the TIE1 protein in both CSF and plasma was linked to an elevated risk of ADHD. These identified proteins, predominantly located on the surface of neurons that mediate excitatory and inhibitory signals, appear to play a vital role in the neurological mechanisms underlying ADHD.

Our study emphasizes the critical value of integrating three different tissue sources to unveil the molecular mechanisms of ADHD, paving the way for developing targeted therapeutic interventions. Specifically, in brain tissue, our findings suggest that the upregulation of GMPPB, NAA80, HYI, and CISD2 exacerbates the risk of developing ADHD. GMPPB encodes GDP-mannose pyrophosphorylase B, crucial for converting mannose-1-phosphate and GTP to GDP-mannose.³⁸ It has been observed that individuals with ADHD exhibit abnormalities in energy metabolism.³⁹ Our research indicates that GMPPB is overexpressed in the brains of patients with ADHD during the later stages of the condition, underscoring the potential role of mitochondrial dysfunction in ADHD. This is further supported by a meta-analysis showing a significant increase in mitochondrial dysfunction in individuals with ADHD.⁴⁰ The precise mechanisms linking GMPPB overexpression to ADHD require further investigation. NAA80, known as actin N-terminal acetyltransferase, has been implicated in developmental delays, muscular weakness, and high-frequency hearing loss due to allelic mismatch variations.⁴¹ Recent transcriptome-wide association studies (TWAS) have linked NAA80 transcriptional dysregulation to ADHD, suggesting its involvement in the modulation of excitatory synapses and synaptic actin rearrangement seen in neurodevelopmental disorders.⁴² Our findings propose NAA80 as a potential therapeutic target for ADHD, given its association with increased GMPPB expression and ADHD risk. The specific mechanism of action of the NAA80 risk protein in ADHD still needs further exploration. The CISD2 gene encodes the CDGSH iron-sulfur domain-containing protein 2, which plays a role in autophagy regulation, calcium buffering, the unfolded protein response, and longevity in mammals.⁴³ CISD2 deficiency is associated with an accelerated ageing phenotype and defective mitochondrial accumulation.⁴⁴ Our research indicates that higher CISD2 protein levels increase the risk of ASD and ADHD, suggesting a novel target for cross-disease treatment. Conversely, increased ICA1L protein abundance, previously shown to protect against loneliness,⁴⁵ was also found to potentially reduce the likelihood of ASD and ADHD in our study.

For cerebrospinal fluid and plasma, TIE1, encoding the orphan receptor protein TIE1 of the receptor tyrosine kinase family, has been linked to ADHD through unique expression cis-regulatory polymorphisms identified in Han Chinese children.⁴⁶ Additionally, the rs3768046 polymorphism at the TIE1 locus is associated with the severity of ADHD symptoms.⁴⁷ Our research suggests a consistent correlation between ADHD risk and elevated TIE1 levels in plasma and cerebrospinal fluid, potentially due to chronic inflammation affecting the blood–brain barrier in ADHD.⁴⁸ These findings highlight the potential of plasma and cerebrospinal fluid as valuable resources for identifying proteins associated with ADHD.

For plasma, previous research has not established a connection between Regulator of Microtubule Dynamics protein 1 (RMDN1) and ADHD. However, an increase in the expression of microtubule-associated proteins was observed in the striatum of rat models of ADHD induced by maternal separation.⁴⁹ Our study suggests that elevated levels of the RMDN1 protein in the plasma of patients may reduce the risk of ADHD. More research is needed to determine the specific processes relating RMDN1 overexpression to ADHD. These preliminary findings suggest that proteins in brain, blood and CSF could serve as promising therapeutic targets for ADHD, though further clinical and mechanistic research is needed to confirm these results.

This research presents multiple strengths. Firstly, the continuous advancement and detailed investigation of protein markers in plasma, the brain, and CSF hold the promise of unveiling novel therapeutic targets. Through our analysis, we have pinpointed potential causative proteins for ADHD and explored existing medications that interact with these proteins, utilizing the Drugbank database for reference. Notably, we identified two drugs—Natalizumab, an antibody targeting ICA1L, and Fostamatinib, an inhibitor of TIE1—that may impact the disease pathway, as detailed in Table S7. Proteins implicated in the brain could be directly involved in ADHD's pathogenesis, offering insights into the disorder's underlying mechanisms. Conversely, proteins identified in plasma and CSF may serve as valuable diagnostic and prognostic indicators. Secondly, the pathogenic proteins identified are prevalent across various ADHD subtypes, highlighting their significant potential as targets for novel therapeutic interventions. Lastly, the robustness of our conclusions is further bolstered by employing a diverse array of analytical methods, including Bayesian colocalization analyses, SMR and PWAS. These methods have consistently demonstrated the directional concordance of the majority of identified risk genes. Future therapeutic strategies for ADHD may well focus on adjusting risk proteins, offering a promising avenue for intervention.

The interpretation and generalizability of our findings are subject to several limitations. First, the specificity of protein quantification may be compromised by genetic variation and qualitative changes in proteins (e.g., amino acid substitutions, post-translational modifications, changes in splicing). Second, due to the constraints in data availability, a majority of the participants in the current analysis are of European ancestry. Consequently, future biomarker GWAS involving in non-European populations is necessary to facilitate cross-ethnic MR analysis. Moreover, heterogeneity tests, and pleiotropy tests, was limited since all prioritized proteins had just one cis-acting SNP and no trans-pQTLs. This method contrasts with traditional Mendelian randomization in GWAS studies, which usually involves numerous SNP instrumental variables and necessitates comprehensive heterogeneity and pleiotropy testing. Future studies on Mendelian randomization of pharmaceutical targets may need to improve the depth of proteome sequencing and incorporate more pQTL-associated SNPs in order to perform heterogeneity and pleiotropy tests. Finally, although we identified some interactions between ADHD causal proteins and existing clinical drug targets, these results are suggestive and should not be considered definitive. Confirming the involvement of risk proteins in the pathophysiology of ADHD requires more investigation.

Conclusions

Our systematic MR analysis of the brain, cerebrospinal fluid, and circulating proteomes has identified seven proteins causally linked to ADHD. GMPP8 emerged as an especially compelling drug target due to its abnormal levels across different age stages of ADHD onset. Moreover, an increased abundance of TIE1 in both CSF and plasma consistently correlates with ADHD risk. Two newly identified protective biomarkers are brain ICA1L and plasma RMDN1, providing potential avenues for risk mitigation. Intriguingly, ICA1L and CISD2 have been found to concurrently influence the risk of ASD, underscoring their broader neurological implications. These findings, while promising, are the first step in a longer journey. Subsequent studies are imperative to evaluate the feasibility of these protein biomarkers as effective drug targets for ADHD treatment, with the potential to significantly advance therapeutic strategies for this condition.

Contributors

TL and CCZ designed the experiment. CCZ, LQJ, and XJL collected the data and performed the MR, PWAS, and cell-type specificity analysis. TL, LQJ, WJG and WD verified the data. WJG, WD and XH reviewed and edited the manuscript. Each author read, approved, and made changes to the manuscript's final draft.

Data sharing statement

The data used in the study can be accessed and downloaded from original studies.^17,20, 21, 22, 23, 24, 25^,37

Declaration of interests

The authors declare no competing interests.

Appendix A. Supplementary data

Supplementary Figure S2

Association between identified biomarkers and risk for ASD and TS. Associations right the black midline represent risk-conferring effects, and those left the black midline represents protective effects. Genetically determined levels of CISD2 increased risk of ASD, and ICA1L decreased risks of ASD. Significant associations were not observed for TS. ∗Nominally significant (P < 0.05).

Supplementary Figure S1.

Graphical representation of the research design. PWAS, proteome-wide association study. SMR, summary-data-based Mendelian Randomization.