Mapping cerebrospinal fluid biomarkers to dementia risk through Mendelian randomization

Abstract

Cerebrospinal fluid (CSF) biomarkers can directly reflect physiological and pathological changes within the central nervous system, serving as a critical element in the preliminary diagnosis and disease monitoring of central nervous system disorders. Specific CSF biomarker patterns may be associated with different types of dementia. The goal of this analysis is to scrutinize the causal correlation between 338 CSF biomarkers and the danger of 4 dementia categories, namely Alzheimer disease, vascular dementia, frontotemporal dementia, and dementia with Lewy bodies, using Mendelian randomization (MR) analysis. Forward and reverse MR were employed to examine this causal relationship. Inverse variance weighting was used as the primary statistical method, with weighted median, MR-Egger, simple mode, and weighted mode methods as supplementary approaches. Concurrently, scrutiny of horizontal pleiotropy, heterogeneity, and sensitivity was undertaken to substantiate the inferences made from the MR study. We identified 9 positive and 12 negative causal relationships between genetic predisposition to CSF biomarkers and dementia. Notably, a bidirectional causal relationship was found between 3-methoxytyrosine levels and Alzheimer disease. The findings of this MR study support a causal association between CSF biomarkers and the risk of 4 distinct types of dementia.

1. Introduction

Dementia is a neurological syndrome characterized primarily by significant decline in cognition and memory, with major classifications including Alzheimer disease (AD), frontotemporal dementia (FTD), dementia with Lewy bodies (DLB), and vascular dementia (VD).^[1] According to an authoritative report from The Lancet, it is estimated that by 2050, the global number of individuals with dementia could exceed 100 million.^[2] Identifying early, sensitive indicators associated with dementia is crucial for effective management of patients with this condition.

Cerebrospinal fluid (CSF) biomarkers can directly reflect physiological and pathological changes within the CNS and hold substantial importance in the preliminary diagnosis and monitoring of CNS disorders. Among the various types of dementia, CSF is a robust biomarker for disease onset and progression. In AD, CSF biomarkers show signature alterations including decreased amyloid-beta 42 (Aβ42), abnormal Aβ42/Aβ40 ratios, elevated phosphorylated tau, and increased total tau. These core biomarkers collectively enable in vivo detection of AD pathology, while concomitant elevation of neurofilament light chain reflects associated neurodegenerative burden.^[3] FTD exhibits markedly elevated CSF neurofilament light chain concentrations, which correlate with disease severity and predict survival outcomes.^[4] DLB demonstrates significantly reduced CSF α-synuclein levels, a hallmark of underlying synucleinopathy.^[5] Additionally, lipocalin-2 has shown potential as a biochemical marker for differentiating VD.^[6] However, the sensitivity and specificity of CSF biomarkers in dementia research still pose challenges. Thus, it is likely that specific patterns of CSF biomarkers remain to be discovered for different types of dementia.

Adopting genome-wide association study (GWAS) and Mendelian randomization (MR) methodologies enables a more thorough and systematic evaluation of the causal connection between cerebrospinal fluid (CSF) biomarkers and the multifaceted manifestations of dementia. Genetic variants serve as instrumental variables (IVs) in the MR analytical approach, enabling evaluation of the causal effect of an exposure on an outcome. Employing genetic variants as instrumental proxies, the MR analytical framework can alleviate the challenges posed by confounding factors and reverse causation inherent in observational studies. In this study, we conducted a 2-sample MR analysis to explore the causal relationship between CSF biomarkers and multiple types of dementia, including AD, VD, FTD, and DLB. Additionally, through reverse causal analysis, we examined whether the genetic susceptibility to dementia influences CSF biomarkers.

2. Materials and methods

2.1. Data sources

The GWAS data for CSF metabolite levels were derived from the study published by Daniel J. Panyard et al,^[7] which analyzed 338 CSF metabolites through a comprehensive metabolome-wide and GWAS. Data for DLB in this study were sourced from the IEU Open GWAS project (https://gwas.mrcieu.ac.uk/), with the dataset ID ebi-a-GCST90001390. This dataset comprised 6618 samples, including 2591 DLB cases and 4027 controls. The GWAS summary data for other types of dementia were obtained from the FinnGen database (https://www.finngen.fi/en), with the VD dataset consisting of 1256 VD cases and 392,463 controls. The FTD dataset included 129 cases and 392,463 controls, while the AD dataset comprised 6145 cases and 388,560 controls (Table 1). All summary data utilized in this study are publicly available and can be freely downloaded. The original studies involved in this research had already obtained ethical approval, and no additional informed consent or ethical approval was required.

Table 1.

Sources of dementia data.

GWAS ID	Trait	Data source	Cases	Controls	Population
F5_ALZHDEMENT	Dementia in Alzheimer disease	FINNGEN	6145	388560	European
FTD	Frontotemporal dementia	FINNGEN	129	392463	European
ebi-a-GCST90001390	Dementia with Lewy bodies	IEU Open GWAS	2591	4027	European
VD_U	Vascular dementia	FINNGEN	1256	392463	European

2.2. SNP selection criteria

A forward MR analysis was carried out to evaluate the causal impacts of CSF levels on distinct forms of dementia. To establish a robust association between the chosen IVs and the exposure, the single nucleotide polymorphisms (SNPs) were required to meet a rigorous statistical significance threshold, typically set at P < 5 × 10^-8. Initially, SNPs were filtered at this threshold; however, only a limited number of inflammatory factors had more than 2 independent SNPs under these conditions. Consequently, the threshold was expanded to P < 1 × 10^-6 to select suitable IVs and obtain more comprehensive results. To uphold the independence of the IVs and fulfill the independence assumption, linkage disequilibrium among the selected SNPs was evaluated, with thresholds set at r² < 0.001, and a distance of 10,000 kb; to avoid bias from weak IVs in the MR analysis, calculating the F-statistic was the approach used to gauge the strength of the IVs, and only SNPs with an F-statistic over 10 were retained to prevent the issue of weak instrument bias.^[8]

In the reverse MR study, the focus was on appraising the causal impact of multiple dementia types on CSF biomarker levels. In this study, SNPs significantly associated with dementia were identified with a threshold of P < 5 × 10^-8. To eliminate linkage disequilibrium among these SNPs, we set the correlation coefficient threshold to <0.001 and the genetic distance to 10,000 kb, similar to the forward MR analysis. The reverse MR analyses utilized the same GWAS dataset that was employed in the forward MR investigation (Fig. 1). Our study protocol adheres to the STROBE-MR guidelines for MR analyses, with all methods conducted in accordance with these guidelines and standards.^[9]

Figure 1.

Flowchart illustrating the framework for Mendelian randomization (MR) analysis concerning cerebrospinal fluid biomarkers and dementia.

2.3. Statistical analysis

Five different MR analysis techniques were utilized in this study, with the inverse variance weighted (IVW) method^[10] being the principal approach. To validate the IVW results, the study incorporated the weighted median,^[11] MR-Egger,^[12] simple mode, and weighted mode MR analysis methods as additional techniques. Heterogeneity and pleiotropy were assessed using both the IVW method and MR-Egger regression. Heterogeneity was quantified using Cochran Q test statistic, which calculates the degree of variability among SNPs. Horizontal pleiotropy was evaluated with the MR-Egger intercept method; if P < .05, pleiotropy was considered present, indicating that outlier SNPs might need to be excluded for a reassessment of the MR analysis’s robustness. Additionally, the MR-PRESSO method was employed to identify and remove potential outliers, followed by re-analysis. To examine the impact of individual SNPs that notably influenced the results, a sensitivity analysis was conducted using the leave-one-out procedure.^[13] All statistical analyses were performed using the “Two Sample MR” package in R software (version 4.3; R Foundation for Statistical Computing, Vienna, Austria). To accurately interpret the relationship between CSF biomarkers and dementia and to minimize the possibility of false positives, the significance threshold for the P-value was set at .01.

3. Results

3.1. Causal effect of cerebrospinal fluid biomarkers on dementia

When the IVW approach was used as the primary analysis method for MR, results with a P-value that was smaller than .01 were considered to have statistical significance. The analysis indicated a causal relationship between certain CSF biomarkers and increased risk of AD. Specifically, higher levels of 3-methoxytyrosine (OR = 1.081, CI: 1.022–1.144, P = .006), leucine (OR = 1.575, CI: 1.227–2.021, P < .001), theophylline (OR = 1.065, CI: 1.021–1.112, P = .003), and methionine sulfoxide (OR = 0.816, CI: 0.72–0.92, P = .001) were associated with an increased risk of AD. Conversely, higher levels of creatine (OR = 0.585, CI: 0.415–0.825, P = .002) and N1-methyl-2-pyridone-5-carboxamide (OR = 0.943, CI: 0.914–0.972, P = .0002) were associated with a reduced risk of AD.

For VD, lower levels of inosine (OR = 0.721, CI: 0.567–0.916, P = .007) and 5-methylthioadenosine (MTA) (OR = 0.859, CI: 0.766–0.962, P = .008) were associated with a reduced risk, while higher levels of 1-stearoyl-2-linoleoyl-GPC (18:0/18:2) (OR = 1.053, CI: 1.015–1.093, P = .005) showed a causal relationship with increased VD risk. Furthermore, higher levels of creatinine (OR = 0.047, CI: 0.004–0.477, P = .009) and symmetric dimethylarginine and asymmetric dimethylarginine (SDMA + ADMA) (OR = 0.37, CI: 0.17–0.77, P = .008) exhibited a negative association with FTD risk.

Additionally, higher levels of 3-hydroxyisobutyrate (OR = 0.50, CI: 0.34–0.73, P = .0004), methionine sulfoxide (OR = 0.816, CI: 0.72–0.92, P = .001), and X-12101 (OR = 0.781, CI: 0.665–0.916, P = .002) were negatively associated with the risk of DLB (see Table 2 and Fig. 2 for details).

Table 2.

Causal effects of cerebrospinal fluid on 4 dementia subtypes.

Exposure	Outcome	MR-Egger		Weighted median		Inverse variance weighted		Simple mode		Weighted mode
		OR (95% CI)	P-val	OR (95% CI)	P-val	OR (95% CI)	P-val	OR (95% CI)	P-val	OR (95% CI)	P-val
3-methoxytyrosine levels	AD	1.131 (0.989–1.292)	.082	1.126 (1.040–1.129)	.003**	1.081 (1.022–1.144)	.006**	1.162 (1.010–1.337)	.044*	1.151 (1.019–1.299)	.031*
Creatine levels	AD	0.968 (0.425–2.204)	.940	0.536 (0.331–0.868)	.011*	0.585 (0.415–0.825)	.002**	0.460 (0.187–1.132)	.105	0.460 (0.210–1.004)	.064
Leucine levels	AD	1.250 (0.683–2.289)	.476	1.598 (1.135–2.250)	.007**	1.575 (1.227–2.021)	<.001**	1.759 (0.918–3.373)	.102	1.692 (0.826–3.468)	.164
N1-methyl-2-pyridone-5-carboxamide levels	AD	0.964 (0.908–1.024)	.241	0.957 (0.914–1.002)	.065	0.943 (0.914–0.972)	.0002**	0.964 (0.871–1.066)	.481	0.955 (0.0.895–1.020)	.176
Theophylline levels	AD	1.040(0.938–1.153)	.456	1.039(0.977–1.104)	.217	1.065(1.021–1.112)	.003**	1.002(0.880–1.140)	.971	1.014(0.9021.140)	.812
Theophylline levels	VD	0.914 (0.615–1.356)	.658	0.671 (0.475–0.948)	.023*	0.721 (0.567–0.916)	.007**	0.689 (0.378–1.256)	.232	0.642 (0.425–0.969)	.041*
Inosine levels	VD	1.025 (0.954–1.101)	.488	1.049 (0.988–1.114)	.113	1.053 (1.015–1.093)	.005**	1.056 (0.922–1.209)	.431	1.089 (0.996–1.190)	.060
1-Stearoyl-2-linoleoyl-GPC (18:0/18:2) levels	VD	0.951 (0.779–1.160)	.622	0.914 (0.766–1.090)	.320	0.859 (0.0.766–0.962)	.008**	0.939 (0.652–1.353)	.738	0.999 (0.791–1.262)	.995
5-Methylthioadenosine (MTA) levels	FTD	0.014 (0.00009–2.274)	.114	0.011 (0.0004–0.310)	.007**	0.047 (0.004–0.477)	.009**	0.023 (0.00009–6.131)	.199	0.0125 (0.00006–2.511)	.118
Creatinine levels	FTD	0.894 (0.227–3.581)	.874	0.553 (0.178–1.174)	.305	0.373-(0.179–0.777)	.008**	0.134 (0.017–1.033)	.059	0.502 (0.106–2.375)	.389
Dimethylarginine (SDMA + ADMA) levels	DLB	0.463 (0.214–1.001)	.061	0.495 (0.292–0.838)	.008**	0.500 (0.340–0.736)	.0004**	0.519 (0.194–1.384)	.201	0.499 (0.188–1.324)	.174
3-Hydroxyisobutyrate levels	DLB	0.904 (0.700–1.169)	.448	0.830 (0.696–0.989)	.038*	0.816 (0.720–0.924)	.001**	0.821 (0.565–1.195)	.309	0.817 (0.566–1.178)	.284
Methionine sulfoxide levels	DLB	0.820 (0.596–1.128)	.234	0.824 (0.648–1.046)	.113	0.781 (0.665–0.916)	.002**	0.854 (0.525–1.388)	.529	0.879 (0.585–1.320)	.539

AD = Alzheimer disease, ADMA = asymmetric dimethylarginine, DLB = dementia with Lewy bodies, FTD = frontotemporal dementia, SDMA = symmetric dimethylarginine, VD = vascular dementia.

^*P < .05.

^**P < .01.

Figure 2.

(A–M) Forest plots illustrating the association between specific cerebrospinal fluid biomarkers and dementia risks.

The results across various methods (including IVW, MR-Egger, weighted median, and weighted mode) showed consistent OR directionality. MR-Egger analysis did not reveal any potential horizontal pleiotropy. Additionally, Cochran Q heterogeneity test yielded P > .05, indicating no heterogeneity in the results (Table 3). As an additional measure to verify the robustness of the MR findings, a leave-one-out sensitivity analysis was undertaken, which revealed that the omission of any single SNP did not markedly alter the overall causal relationship, as presented in Figure 3.

Table 3.

Study on pleiotropy and heterogeneity in the relationship between cerebrospinal fluid and dementia.

Exposure	Outcome	MR-Egger intercept		MR-Egger			Inverse variance weighted
		Egger intercept	P-value	Q	Q_df	Q_pval	Q	Q_df	Q_pval
3-Methoxytyrosine levels	AD	-0.009	.476	25.593	25.000	.429	26.130	26.000	.455
Creatine levels	AD	-0.018	.201	12.382	21.000	.928	14.128	22.000	.896
Leucine levels	AD	0.013	.420	23.422	21.000	.321	24.176	22.000	.338
N1-methyl-2-pyridone-5-carboxamide levels	AD	-0.008	.398	129.401	118.000	.222	130.190	119.000	.227
Theophylline levels	AD	0.009	.617	30.606	29.000	.384	30.875	30.000	.421
Inosine levels	VD	-0.021	.150	44.834	38.000	.206	47.375	39.000	.167
1-Stearoyl-2-linoleoyl-gpc (18:0/18:2) levels	VD	0.010	.379	243.009	252.000	.646	243.785	253.000	.649
5-Methylthioadenosine (MTA) levels	VD	-0.015	.223	86.062	93.000	.681	87.563	94.000	.667
Creatinine levels	FTD	0.047	.616	20.931	23.000	.585	21.189	24.000	.627
Dimethylarginine (SDMA + ADMA) levels	FTD	-0.116	.146	47.667	48.000	.485	49.852	49.000	.439
3-Hydroxyisobutyrate levels	DLB	0.006	.821	20.493	26.000	.767	20.545	27.000	.807
Methionine sulfoxide levels	DLB	-0.025	.369	47.595	61.000	.895	48.411	62.000	.896
X-12101 levels	DLB	-0.007	.728	26.723	27.000	.478	26.846	28.000	.526

AD = Alzheimer disease, ADMA = asymmetric dimethylarginine, DLB = dementia with Lewy bodies, FTD = frontotemporal dementia, SDMA = symmetric dimethylarginine, VD = vascular dementia.

Figure 3.

(A–M) Sensitivity analysis plots for the association between cerebrospinal fluid biomarkers and dementia.

3.2. Causal effect of dementia on cerebrospinal fluid biomarkers

When employing the IVW technique as the principal method for the MR analysis, and considering dementia as the exposure factor, the findings revealed that AD was associated with an elevated risk of several CSF biomarkers, including 1,5-anhydroglucitol (1,5-AG) levels (OR = 1.057, CI: 1.014–1.101, P = .008), 3-methoxytyrosine levels (OR = 1.092, CI: 1.022–1.166, P = .008), 4-methyl-2-oxopentanoate levels (OR = 1.024, CI: 1.005–1.043, P = .009), choline phosphate levels (OR = 1.071, CI: 1.019–1.126, P = .006), dimethylarginine (SDMA + ADMA) levels (OR = 1.037, CI: 1.013–1.062, P = .002), ethylmalonate levels (OR = 1.039, CI: 1.009–1.069, P = .009), and lysine levels (OR = 1.027, CI: 1.006–1.047, P = .008) (see Table 4 for details).

Table 4.

Causal influence of 4 dementia subtypes on cerebrospinal fluid.

Exposure	Outcome	MR-Egger		Weighted median		Inverse variance weighted		Simple mode		Weighted mode
		OR (95% CI)	P-val	OR (95% CI)	P-val	OR (95% CI)	P-val	OR (95% CI)	P-val		OR (95% CI)	P-val
AD	Represents 3-methoxytyrosine levels	1.125 (1.025–1.235)	.047*	1.107 (1.032–1.189)	.004**	1.092 (1.022–1.166)	.008**	1.144 (0.911–1.437)	.282		1.110 (1.038–1.187)	.018*
AD	1,5-Anhydroglucitol (1,5-AG) levels	1.056 (0.996–1.119)	.113	1.054 (1.009–1.102)	.017*	1.057 (1.014–1.101)	.008**	1.085 (0.966–1.219)	.210		1.053 (1.006–1.103)	.059
AD	4-Methyl-2-oxopentanoate levels	1.020 (0.993–1.047)	.188	1.021 (1.002–1.041)	.029*	1.024 (1.005–1.043)	.012*	1.014 (0.953–1.078)	.102		1.021 (1.0004–1.042)	.085
AD	Choline phosphate levels	1.115 (1.040–1.196)	.022*	1.0861.029–1.145)	.002**	1.071 (1.019–1.126)	.006**	1.042 (0.892–1.216)	.616		1.087 (1.033–1.145)	.014*
AD	Dimethylarginine (SDMA + ADMA) levels	1.059 (1.025–1.093)	.012*	1.044 (1.019–1.0709)	.0004**	1.037 (1.013–1.062)	.002**	1.025 (0.937–1.121)	.594		1.044 (1.018–1.0703)	.011*
AD	Ethylmalonate levels	1.024 (0.983–1.067)	.294	1.035 (1.005–1.066)	.021*	1.039 (1.009–1.069)	.009**	1.059 (0.970–1.157)	.236		1.036 (1.003–1.070)	.068
VD	Lysine levels	1.020 (0.984–1.058)	.298	1.024 (0.999–1.049)	.050	1.027 (1.006–1.047)	.008**	1.043 (0.992–1.096)	.125		1.019 (0.994–1.045)	.157
DLB	1-Ribosyl-imidazoleacetate levels	1.066 (1.006–1.131)	.095	1.060 (1.020–1.101)	.002**	1.052 (1.017–1.088)	.003**	1.058 (0.992–1.128)	.143		1.061 (1.018–1.106)	.035*

AD = Alzheimer disease, ADMA = asymmetric dimethylarginine, DLB = dementia with Lewy bodies, SDMA = symmetric dimethylarginine, VD = vascular dementia.

^*P < .05.

^**P < .01.

4. Discussion

In this study, using MR analysis based on large-scale genetic consortia, we systematically explored, for the first time, the causal relationships between 338 different CSF biomarkers and dementia risk using genetic instruments. By integrating the results of both forward and reverse MR analyses, we identified a positive correlation between elevated levels of 3-methoxytyrosine, leucine, and theophylline with an increased risk of AD. Conversely, elevated levels of creatine and N1-methyl-2-pyridone-5-carboxamide was found to be associated with a decreased risk of AD. Additionally, the findings indicated that AD was a risk factor associated with heightened concentrations of 3-methoxytyrosine, 1,5-AG, 4-methyl-2-oxopentanoate, choline phosphate, dimethylarginine (SDMA + ADMA), and ethylmalonate.

4.1. AD

AD is a neurodegenerative disorder that progressively worsens, and is defined by the pathological features of amyloid plaque buildup, tau protein phosphorylation, as well as various metabolic abnormalities detected in the CSF.^[3] The findings of this study reveal a bidirectional causal relationship between 3-methoxytyrosine, a metabolite in the aromatic amino acid metabolic pathway, and AD. Previous studies have shown that in an AD mouse model, urinary levels of 3-methoxytyrosine were elevated compared to those in wild-type mice, suggesting a potential deficiency in the activity of aromatic L-amino acid decarboxylase (AADC).^[14] The enzyme AADC is critical for the conversion of dopamine precursors to dopamine, and its activity level is strongly associated with the pathological mechanisms involved in the development of AD. Changes in AADC activity can affect the levels of dopamine-related metabolites; for example, an increase in 3-methoxytyrosine is often used as an indirect indicator of decreased AADC activity. Further research has shown that, in the treatment group of AD models, the level of 3-methoxytyrosine decreases,^[15] suggesting that treatment may help maintain dopamine levels in the brain by enhancing AADC activity or its associated metabolic pathways. These results align with the findings from our present investigation, which uncovered a correlation between heightened levels of 3-methoxytyrosine and an elevated risk of AD, indicating a complex bidirectional influence between the 2.

Leucine levels are positively correlated with an increased risk of AD. As a critical amino acid for protein synthesis, reduced leucine levels may affect protein synthesis and neuronal function and are involved in energy metabolism.^[16,17] Contrarily, investigations comparing the levels of 18 amino acids in the CSF and plasma between AD patients and healthy volunteers have shown that the concentration of leucine is considerably lower in the CSF of those with AD.^[18] Notably, the concentration of leucine has been found to be significantly linked to the deposition of tau protein and the generation and elimination of Aβ.^[19,20] Considering that the irregular accumulation of Aβ and abnormalities in tau protein are key contributors to the pathology of AD, changes in leucine levels may be related to the pathological processes of AD. Although current data support the association between reduced leucine levels and AD, elevated leucine levels might also represent an alternative form of metabolic imbalance, potentially impacting protein synthesis and energy metabolism, thus posing a potential risk factor for AD development.

Theophylline, a downstream metabolite of methylxanthines found in coffee, tea, and cocoa (chocolate), possesses some lipophilicity, allowing it to cross the blood-brain barrier and enter brain tissue and CSF. Recent studies suggest that theophylline may have a positive impact on brain health and may help in reducing the likelihood of age-related neurodegenerative conditions, including AD.^[21] However, the results of this investigation suggest that theophylline may in fact be a risk factor for AD, as high doses of theophylline might induce central nervous system stimulation symptoms.

Our results suggest that creatine levels may reduce the risk of AD. Creatine is an essential substance in brain energy metabolism, and its concentration changes can reflect variations in neuronal energy status or cell density. Impaired brain energy metabolism is a hallmark of AD and often occurs before symptoms manifest. The relationship between creatine transporter deficiency and AD biomarkers has been documented, with studies showing that creatine transporter deficiency patients exhibit significantly increased levels of total tau protein, Aβ40, and Aβ42 in CSF.^[22] Additionally, creatine supplementation has been shown to improve brain energy metabolism in AD animal models, along with enhancing AD biomarkers and cognitive function, suggesting that creatine could be a potential intervention target to address energy metabolism issues in AD patients.^[23] This evidence highlights the pivotal significance of creatine in regulating neuronal energy metabolism and demonstrates the profound effect that variations in its concentration can have on the pathological cascades associated with neurodegenerative conditions.

N1-methyl-2-pyridone-5-carboxamide (2PY) levels, a metabolite of niacin, were found in this study to be connected to a lessened susceptibility to AD when elevated. Nevertheless, there is currently a lack of direct investigative research exploring the association between 2PY and AD. Existing literature has implied that elevated 2PY concentrations are linked to elevated risks of cardiovascular disease and enteritis, suggesting that abnormalities in niacin metabolism may affect energy metabolism, antioxidant stress responses, and inflammatory states.^[24,25] In the context of AD, however, elevated 2PY levels may have a protective effect. Continued research efforts are essential to thoroughly elucidate the precise linkages between 2PY and AD, to affirm its viability as a prospective target for AD deterrence or treatment.

Reverse MR analysis revealed an association between AD and elevated levels of 1,5-AG, a biomarker of glycemic control that reflects short-term blood glucose fluctuations. In individuals with well-controlled blood glucose, 1,5-AG levels in the blood tend to be higher. Diabetic patients, however, exhibit lower 1,5-AG levels (indicating higher blood glucose peaks), which has been associated with an elevated risk of dementia and hastened cognitive deterioration.^[26] Nevertheless, no significant association between 1,5-AG and AD risk has been observed in prospective cohort studies.^[27] Despite this, the results of this study suggest that elevated 1,5-AG levels in AD patients may reflect a physiological state or mechanism distinct from that observed in diabetic individuals.

AD is also a risk factor for elevated choline phosphate levels and dimethylarginine (SDMA + ADMA) levels. Choline phosphate is a crucial precursor in the synthesis of phosphatidylcholine, a major component of cell membranes essential for maintaining normal neuronal function. Studies have indicated that the pathological process of AD involves an increase in choline levels.^[28] As choline phosphate is an intermediate in the conversion of choline to phosphatidylcholine, an increase in choline levels could imply a corresponding increase in choline phosphate levels; however, this hypothesis requires specific experimental data for confirmation.

Additionally, the study results indicate that, compared to the control group, levels of ADMA and SDMA are significantly elevated in AD patients.^[29] Importantly, increased quantities of ADMA and SDMA have been correlated with the extent of cognitive impairment seen in AD. This suggests that ADMA and SDMA levels are elevated in AD patients and that this elevation correlates with the degree of cognitive dysfunction.^[30] The 4-methyl-2-oxopentanoate and ethylmalonate are organic acids and metabolites found in CSF. Although many metabolites have been studied, current data do not provide direct evidence of a clear association between 4-methyl-2-oxopentanoate levels or ethylmalonate levels and AD.

4.2. VD

Inosine levels and MTA levels are associated with a reduced risk of VD, while increased levels of 1-stearoyl-2-linoleoyl-GPC (18:0/18:2) elevate the risk of VD. VD is also a risk factor for elevated lysine levels.

Elevated inosine levels are linked to a reduced risk of VD. Due to its antioxidant properties, inosine effectively combats cellular damage caused by oxidative stress.^[31] Studies have shown that inosine can inhibit oxidative stress markers (such as malondialdehyde and nitric oxide) and pro-inflammatory cytokines (such as TNF-α and IL-1β), while replenishing glutathione levels.^[32,33] In models of Huntington disease, inosine has demonstrated neuroprotective effects and positive impacts on dementia symptoms.^[33] Furthermore, in mice injected with Aβ and treated with donepezil, inosine levels in brain tissue were significantly higher than in mice injected with Aβ alone, suggesting that elevated inosine levels might be an indicator of cognitive improvement.^[34] Therefore, elevated inosine levels may not only provide antioxidant and anti-inflammatory effects but also serve as a marker of cognitive improvement, thereby reducing the risk of VD.

MTA is a metabolite in the polyamine and methionine metabolic pathways. It plays a role in tumor suppression by inhibiting tumor cell proliferation and invasion, inducing apoptosis, and regulating the inflammatory microenvironment within tumor tissues.^[35] However, current studies do not provide direct evidence of a specific relationship between MTA levels and VD. The potential role of MTA in VD may be related to its regulatory effects on metabolism and inflammation, and further research is required to explore this connection.

Increased levels of 1-stearoyl-2-linoleoyl-GPC (18:0/18:2) are associated with a higher risk of VD. Changes in 1-stearoyl-2-linoleoyl-GPC (18:0/18:2) levels may be related to blood pressure regulation and have potential implications for cardiovascular health.^[36] Although there is no direct evidence regarding its role in VD, it is hypothesized that metabolic abnormalities in high levels of 18:0/18:2 phospholipid components could lead to lipid deposition, disrupt normal blood flow, and adversely impact vascular health, thereby increasing the risk of VD.

Reverse MR revealed that VD is a risk factor for elevated lysine levels. Modifications such as propionylation and butyrylation occur on lysine residues. Studies have shown that donepezil can improve cognitive function in rat models of VD and reduce the levels of certain post-translational modifications (e.g., propionylation and butyrylation) in the hippocampus.^[37] Lysine serves as the site for these modifications, and changes in these modifications may impact protein function, thereby affecting cognitive abilities.^[37] Consequently, modifications on lysine may be associated with the pathological processes of VD.

4.3. FTD

Amplified creatinine amounts and elevated SDMA + ADMA amounts have been linked to a diminished probability of developing FTD.

FTD is a neurodegenerative disorder marked by gradual deterioration in behavioral, executive, and linguistic capacities. Gas chromatography-mass spectrometry studies comparing plasma metabolite profiles of FTD patients and cognitively healthy individuals have shown that creatinine metabolite levels are lower in FTD patients compared to the control group.^[38] To date, there is a dearth of direct empirical studies exploring the linkage between SDMA + ADMA levels and the probability of developing FTD. A systematic review and meta-analysis indicated that, in patients with AD and VD, circulating concentrations of arginine metabolites ADMA and SDMA, which are related to nitric oxide synthesis, are significantly higher than in the control group.^[29] Another study assessed changes in participants’ cognitive abilities using the longitudinal Raven Progressive Matrices and found that elevated plasma ADMA concentrations predicted cognitive deterioration in middle-aged and older adults approximately 4 years later.^[39] Although higher levels of ADMA and SDMA may be associated with more pronounced cognitive decline, it remains unclear whether changes in these metabolite levels apply equally to patients with FTD. Further research is required to explore the specific mechanisms of these metabolites in FTD and their relationship to disease progression, aiming to better understand the potential link between dimethylarginine (SDMA + ADMA) levels and FTD risk.

4.4. DLB

The data gathered in this research indicates that elevated measurements of 3-hydroxyisobutyrate, methionine sulfoxide, and the chemical X-12101 are predictive of a diminished risk of being afflicted with DLB; however, elevated 1-ribosyl-imidazoleacetate levels are a risk factor for DLB.

3-Hydroxyisobutyrate is a metabolite in the valine degradation pathway. There is limited research on its relationship with DLB; prior studies have revealed that among patients diagnosed with AD and mild cognitive decline, CSF levels of valine are decreased, while levels of 3-hydroxyisobutyrate, a metabolite in the valine degradation pathway, are elevated.^[40] This suggests that 3-hydroxyisobutyrate levels may be associated with dementia-related metabolic pathways; however, its potential role in reducing DLB risk requires further investigation.

Currently, there is no direct evidence in the literature indicating that methionine sulfoxide (MetO) levels and X-12101 levels correlate with a lessened probability of being diagnosed with DLB. MetO is an oxidation product of methionine and act as a signifier of oxidative stress.^[41] Elevated MetO levels are closely associated with AD, where MetO–clusterin levels are significantly increased in the brains of AD patients.^[42] Additionally, methionine oxidation reduces clusterin’s ability to bind with Aβ, thereby impairing its chaperone function.^[42] In the context of DLB, higher methionine sulfoxide levels may reflect an effective antioxidant defense mechanism, potentially contributing to a reduced risk of DLB. Elevated X-12101 levels may indicate the activation of specific metabolic or antioxidant pathways that exert a protective effect on neurons, thereby reducing the incidence of DLB.

Furthermore, research on 1-ribosyl-imidazoleacetate is limited, though a MR study reported a positive correlation between 1-ribosyl-imidazoleacetate and the risk of ischemic stroke.^[43] The results of this investigation imply that DLB could also be a risk factor linked to increased levels of 1-ribosyl-imidazoleacetate. DLB patients may experience metabolic abnormalities or states of chronic inflammation and oxidative stress,^[44,45] which could lead to increased levels of 1-ribosyl-imidazoleacetate, thereby indicating that DLB could be a risk factor linked to elevated 1-ribosyl-imidazoleacetate. However, these hypotheses require further experimental validation.

Currently, early detection methods for dementia remain limited, with a lack of objective and effective biomarker tests. Dementia itself may affect CSF composition. As a result, we implemented a comprehensive MR approach to assess the causal linkages between CSF markers and 4 subtypes of dementia, a strategy that can help identify and study early predictors of dementia progression. Inevitably, our investigation possesses certain constraints and limitations. Firstly, the restricted number of participants for specific dementia subgroups could have impaired our potential to uncover certain causal links, especially FTD and DLB, was limited. Second, as our study sample primarily consists of European populations, the generalizability of the results may be constrained. Additionally, while MR provides a method to reduce the impact of confounding factors, it cannot entirely exclude the possibility of horizontal pleiotropy or gene–environment interactions.

5. Conclusion

This study explored the causal relationships between CSF measures and dementia subtypes, revealing 9 positive and 12 negative causal associations between genetic predispositions for CSF biomarkers and dementia. Additionally, a bidirectional relationship between 3-methoxytyrosine and AD was identified. These findings could help inform future dementia prevention and treatment strategies. However, the study’s conclusions are limited by sample size and potential biases in the data. Moreover, the genetic variants examined may not fully capture the complex environmental factors influencing dementia. The analysis also primarily focused on CSF biomarkers, overlooking other important factors that could provide a more comprehensive understanding of dementia risk.

Acknowledgments

We would like to thank all the participants and investigators involved in this research.

Author contributions

Conceptualization: Xiaomin Zhu, Wei Chen, Lin Wu.

Data curation: Yulan Fu, Yingrui Huang, Ying Zhang, Haosen Liao.

Formal analysis: Xiaomin Zhu, Guifeng Zhuo.

Funding acquisition: Wei Chen, Lin Wu.

Methodology: Xiaomin Zhu, Yulan Fu.

Resources: Xiaomin Zhu.

Software: Xiaomin Zhu, Haosen Liao.

Supervision: Xiaomin Zhu.

Visualization: Xiaomin Zhu, Chengqing Yang, Yulan Fu, Yingrui Huang, Ying Zhang, Haosen Liao.

Writing – original draft: Xiaomin Zhu, Guifeng Zhuo, Chengqing Yang, Yulan Fu.

Writing – review & editing: Wei Chen, Lin Wu.