Multi-omics-based decoding of circulating biomarkers in amyotrophic lateral sclerosis and risks in environmental toxins

Abstract

Background

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease characterized by the interplay of genetic and environmental factors, and currently, there there is a lack of effective diagnostic or therapeutic strategies available. This study aims to identify circulating biomarkers for ALS and investigate their interactions with environmental toxins.

Methods

This research utilizes plasma proteomic genome-wide association study (GWAS) data and whole blood transcriptomic data from ALS patients to screen for potential circulating biomarkers through Mendelian randomization (MR). Subsequently, functional enrichment analysis and immune infiltration analysis were performed. An integrated machine learning approach will be used to construct a diagnostic model, with hub genes selected based on SHAP values. The model’s performance will be validated using receiver operating characteristic (ROC) curves, nomogram, and decision curve analysis (DCA). Finally, reverse network toxicology will be used to explore the interaction mechanisms between hub genes and environmental toxins.

Results

Based on a MR analysis of plasma proteomics, we identified 68 plasma proteins significantly associated with the risk of ALS. By integrating differentially expressed genes (DEGs) from whole blood transcriptomics (1,116 DEGs), we selected four potential circulating biomarkers: FCRL3, HTATIP2, RNASE6, and SF3B4. Functional enrichment analysis indicated that the pathogenesis of ALS is closely related to autophagy, apoptosis, the endoplasmic reticulum unfolded protein response, and the NF-κB signaling pathway. Immune infiltration analysis revealed a disruption of the immune microenvironment mediated by T cells/myeloid cells in ALS patients. Validation through 113 machine learning algorithms showed that the random forest model exhibited the best diagnostic performance (AUC = 0.786), while SHAP analysis confirmed the contribution ranking of hub biomarkers: RNASE6 > FCRL3 > HTATIP2 > SF3B4. Further validation of their diagnostic value was performed using ROC curves, nomograms, and DCA. Environmental toxins analysis revealed that substances such as benzo(a)pyrene exhibit significant neurotoxicity, and molecular docking confirmed that they can interfere with the function of hub biomarkers through strong binding (∆G < -5 kcal·mol⁻¹), suggesting potential environmental pathogenic mechanisms in ALS.

Conclusions

This study not only highlights the value of FCRL3, HTATIP2, RNASE6, and SF3B4 as potential diagnostic biomarkers and therapeutic targets for ALS but also provides new evidence for the involvement of environmental toxins, particularly benzo(a)pyrene, in the pathogenesis of ALS through gene-environment interactions.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40360-025-01024-9.

Introduction

Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disease characterized by the gradual degeneration of motor neurons in the cerebral cortex, brainstem, and spinal cord, ultimately leading to muscle atrophy, paralysis, respiratory failure, and death [1, 2]. As the most representative and clinically severe form of motor neuron disease, ALS has a global prevalence of approximately 2 per 100,000 individuals, with a median survival time post-diagnosis typically ranging from 3 to 5 years [3, 4].

The pathogenesis of ALS is complex and multifactorial, involving interactions between genetic susceptibility and environmental factors, as well as dysregulation of various cellular and molecular pathways [5, 6]. On the genetic level, multiple gene mutations have been associated with the disease, including KIF5A, annexin A11, C9orf72, and ATXN2. These mutations may contribute to disease progression by disrupting intracellular transport, protein homeostasis, or RNA metabolism [7–10]. Other key mechanisms implicated in ALS include glutamate-mediated excitotoxicity, excessive inflammatory responses, autophagy dysfunction, oxidative stress, and mitochondrial dysfunction. These processes may synergistically lead to neuronal damage and non-cell-autonomous neurodegeneration [11–13]. Furthermore, the abnormal accumulation of TAR DNA-binding protein 43 (TDP-43) in the nucleus, along with associated disruptions in RNA metabolism and stress granule regulation, has been observed in 97% of ALS cases [14]. Current understanding suggests that the disease progression of ALS can be divided into an early neuroprotective phase and a late neurodegenerative phase. However, the precise mechanisms underlying these stages remain incompletely elucidated. Ongoing research focuses on genetic heterogeneity, the interplay of environmental factors, and the multifactorial nature of disease pathogenesis, which are critical areas for developing therapeutic strategies [5, 15, 16].

Environmental toxins are widely recognized as significant risk factors for ALS, particularly in sporadic cases, which account for 90% of occurrences. Environmental exposure increases the risk of developing ALS and influences disease progression through neurotoxic mechanisms, primarily involving categories such as pesticides, heavy metals, cyanobacterial toxins, persistent organic pollutants (POPs), and mycotoxins [17–21]. Epidemiological evidence indicates a significant association between occupational exposure and environmental pollutants with ALS incidence. For instance, a nationwide case-control study from the Centers for Disease Control and Prevention (CDC) in the United States confirmed that occupational lead exposure (odds ratio [OR] = 1.77) and residues of organochlorine pesticides (such as alpha-endosulfan, oxychlordane, and heptachlor) in serum were significantly related to ALS risk, with the highest OR reaching 3.57 [17, 18, 22]. A study within the Italian population found a positive correlation between exposure to the cyanobacterial toxin L-BMAA and the development of ALS [19, 21], while exposure to heavy metals like lead exacerbated the degeneration of spinal motor neurons through oxidative stress and mitochondrial dysfunction [23, 24]. Laboratory studies further confirm that environmental toxins can induce excitotoxicity due to glutamate and neuroinflammatory responses, thereby promoting the pathological progression of ALS [25]. For example, spinal motor neurons exhibit heightened susceptibility to environmental neurotoxins such as methylmercury, which may accelerate neuronal apoptosis [25]. Although existing epidemiological and experimental evidence supports the role of environmental toxins as significant risk factors for ALS, and their pathogenic effects may be mediated through gene-environment interactions, the specific pathogenic mechanisms require further elucidation [26–28].

ALS currently lacks effective treatment options, with diagnosis primarily relying on the exclusion of clinical symptoms. Consequently, research on circulating biomarkers is of paramount importance [29–31]. The exploration of circulating biomarkers can aid in early diagnosis, prognostic prediction, disease progression monitoring, and treatment strategy guidance [32]. Current studies predominantly focus on the analysis of circulating molecules such as proteins, non-coding RNAs, and cytokines [30, 33]. The composition of the circulating proteome reflects the body’s inflammatory and metabolic status; notably, the circulating protein aggregates in ALS patients are enriched with axonal proteins like neurofilament heavy chain, which are associated with the formation of aggregates in the brain and have been identified as biomarkers for ALS [34]. However, due to the high heterogeneity of ALS, integrating circulating proteome data to explore biomarkers holds promise for enhancing early diagnosis and prognostic assessment capabilities [35].

In this study, we utilized summary data from a plasma proteomics genome-wide association study (GWAS) [36] and whole blood transcriptomic data from ALS patients to identify potential circulating biomarkers for ALS through Mendelian randomization (MR). We conducted functional enrichment analyses and employed single sample gene set enrichment analysis (ssGSEA) to evaluate immune microenvironment characteristics. Various machine learning algorithms were utilized to construct diagnostic models, selecting hub genes based on SHAP values, and predicting interactions between hub genes and environmental pollutants by comparing with the Comparative Toxicogenomics Database (CTD) and molecular docking studies. In summary, this research employs multi-omics approaches to identify potential circulating biomarkers for ALS and investigates their associations with environmental toxins, aiming to elucidate the gene-environment interactions in the pathogenesis of ALS and provide a basis for early diagnosis and personalized treatment.

Materials and methods

MR analysis design

This study strictly adheres to the STROBE-MR guidelines for Mendelian randomization (MR) analysis, which relies on the validity of instrumental variables (IVs) meeting three core assumptions: First, IVs must have a strong genetic association with the exposure (plasma proteins) (relevance assumption); second, IVs should be independent of confounding factors related to both exposure and outcome (independence assumption); finally, the effect of IVs on the outcome must be entirely mediated through the exposure (exclusivity assumption).

Data sources for MR analysis

In this study, exposure data were obtained from the deCODE team’s summary statistics of a GWAS (https://www.decode.com/summarydata/) based on plasma proteomics [36], which conducted genetic association analyses on 4,719 plasma proteins in individuals of Icelandic descent (n = 35,559). Outcome data were sourced from the FinnGen Consortium’s 12th release (R12) [37], utilizing the ALS-GWAS (https://storage.googleapis.com/finngen-public-data-r12/summary_stats/release/finngen_R12_G6_ALS.gz) dataset (FinnGen R12 G6_ALS), which included ALS patients (n = 667) and controls (n = 216,760) (diagnostic criteria: ICD-10 G12.2).

Instrumental variable selection

The selection process comprised four critical steps: First, single nucleotide polymorphisms (SNPs) were initially screened based on a genome-wide significance threshold (p < 5 × 10⁻⁸). Subsequently, linkage disequilibrium clustering was performed using the 1000 Genomes European reference panel (r² < 0.001, window = 10,000 kb). Next, the effect allele directions for both exposure and outcome were harmonized, and SNPs with palindromic structures were excluded. Finally, the strength of the instruments was verified using the F-statistic (F = (β_exposure/SE_exposure)²), with weak instruments (F < 10) being discarded.

MR statistical analysis methods

The analysis was conducted using the “TwoSampleMR” package in R (version 4.5.0). The effect size was assessed using odds ratios (OR) and their 95% confidence intervals (CI), with statistical significance set at p < 0.05. When only one SNP was available for analysis, the Wald ratio method was employed for MR estimation; when multiple SNPs were present, the inverse variance weighted (IVW) method with random effects was utilized for MR analysis. The key sensitivity analysis aims to validate the effectiveness of the IVs. We employed Cochran’s Q test to assess heterogeneity. Additionally, we utilized the MR-Egger intercept test to exclude systematic horizontal pleiotropy, requiring a p-value for the intercept greater than 0.05. Furthermore, we identified and corrected for pleiotropic outlier SNPs using the MR-PRESSO method, thereby ensuring the robustness and reliability of the causal inference results.

Differential gene and circulating biomarker selection

Gene expression microarray datasets related to ALS were retrieved from the GEO database (https://www.ncbi.nlm.nih.gov/geo/). The dataset GSE112680 includes whole blood gene expression data from ALS patients (n = 164) and healthy controls (n = 137). The GSE112676 dataset was utilized as an external validation cohort, comprising whole blood gene expression data from ALS patients (n = 233) and controls (n = 508). Differentially expressed genes (DEGs) were identified using the R package ‘limma’, with thresholds set at |log2FC| >0.1 and adjusted p < 0.05 to ensure a sufficient number of DEGs for subsequent bioinformatics analyses. A Venn diagram was employed to identify the intersection between MR causal proteins and DEGs, and these intersecting proteins were defined as potential circulating biomarkers for ALS.

Functional enrichment analysis

Functional enrichment analyses for the candidate ALS circulating biomarkers were conducted using the DAVID database (https://davidbioinformatics.nih.gov/tools.jsp), focusing on Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (species: Homo sapiens, significance threshold p < 0.05). The top five entries for GO biological processes (BP), cellular components (CC), and molecular functions (MF) were presented using bar plots, while a bubble chart displayed the top ten KEGG pathways.

Immune microenvironment analysis

The infiltration levels of 28 immune cell subpopulations were quantified using the single-sample gene set enrichment analysis (ssGSEA) method, implemented via the GSVA package in R [38]. The Wilcoxon rank-sum test was employed to compare immune infiltration differences between the normal control group and the ALS group (significance threshold p < 0.05). Furthermore, Spearman rank correlation analysis was conducted to assess the relationship between the expression levels of candidate ALS circulating biomarkers and the degree of immune cell infiltration, aiming to elucidate potential immune regulatory mechanisms.

Identification and validation of hub genes

We developed a diagnostic model comprising 113 algorithm combinations, which integrates 12 machine learning algorithms (such as random forest and LASSO regression) to reduce the risk of overfitting. This model evaluates the diagnostic performance and contribution ranking of preselected hub genes, obtained by intersecting MR with DEGs [39]. The model development and variable selection were based on an internal training dataset (GSE12680), and the model’s performance was validated using an external testing dataset (GSE12676). The model with the highest average area under the curve (AUC) in both the training and validation sets was designated as the optimal model, which was then used to filter hub genes [40]. Leveraging the concept of Shapley values from game theory, we employed SHAP (SHapley Additive exPlanations) to interpret the predictions of the machine learning model. This approach assigns importance values to each feature of the model, thereby quantifying the relative contributions of the selected genes [41]. To assess the diagnostic value of the hub genes, we utilized the “timeROC” package to plot receiver operating characteristic (ROC) curves and calculate the area under the curve (AUC). Subsequently, we constructed a nomogram using the “rms” package to visualize risk probabilities and plotted calibration curves to evaluate predictive accuracy. Additionally, decision curve analysis (DCA) was performed to assess the clinical utility of the nomogram.

Prediction of environmental toxins and toxicological analysis

Based on the identified hub genes, we predicted potential related environmental toxins using the Comparative Toxicogenomics Database (CTD, http://ctdbase.org/). ADMETlab 3.0 (https://admetlab3.scbdd.com/) is an online platform designed for the systematic assessment of compounds’ absorption, distribution, metabolism, excretion, and toxicity parameters. In conjunction with reverse virtual screening of the ALS hub genes, we employed ADMETlab 3.0 for the toxicity prediction and analysis of environmental toxins.

Molecular docking

This study utilized molecular docking techniques to investigate the interaction mechanisms between candidate environmental toxins and hub proteins. Initially, the three-dimensional structures of hub proteins were obtained from the Protein Data Bank (PDB), and the molecular structures of environmental toxins were downloaded from the PubChem database. Hydrogen atoms were added to the proteins and small molecules using AutoDock Tools 1.5.6. Subsequently, semi-flexible docking was performed using AutoDock Vina 1.2.0, with the exhaustiveness parameter set to 25 and the Lamarckian genetic algorithm applied, ultimately calculating the binding free energy.

Results

Potential biomarkers based on plasma proteomics

In this study, we employed the inverse variance weighted (IVW) method as the primary analytical model to evaluate the potential causal relationship between plasma protein levels and the risk of ALS. The analysis revealed that, under the support of instrumental variables meeting core assumptions and stringent selection criteria, a total of 68 plasma proteins exhibited statistically significant associations with ALS risk (p < 0.05). To ensure the scientific rigor and reliability of these associations, we conducted a comprehensive sensitivity analysis on the 68 proteins, which included heterogeneity testing, MR-Egger intercept testing, and MR-PRESSO outlier detection. Following this stringent screening process based on the core assumptions of IVs, only 39 plasma proteins exhibited causal associations that were confirmed to be robust and reliable, successfully passing all core assumption validations. Among these, 21 plasma proteins were positively correlated with an increased risk of ALS, while the remaining 18 plasma proteins were negatively correlated with a reduced risk of ALS (Fig. 1A). The results of the MR analysis are presented in Supplementary Tables 1S–4S.

Fig. 1.

Potential circulating biomarkers based on plasma proteomics and whole blood transcriptomics. (A) Forest plot of Mendelian randomization analysis of proteomics. (B) Volcano plot of DEGs analysis. (C) Selection of circulating biomarkers

Circulating biomarkers

Based on the gene expression data from whole blood of 301 patients in the GSE112680 dataset, we identified a total of 1,116 DEGs, comprising 606 upregulated genes and 510 downregulated genes (Fig. 1B). By constructing a Venn diagram, we identified four potential circulating biomarkers for ALS (FCRL3, HTATIP2, RNASE6, SF3B4) derived from both the plasma proteomics and whole blood transcriptomics (Fig. 1C).

Functional enrichment analysis

The results indicated that the GO terms, including “phagocytosis,” “apoptotic signaling pathway,” and “unfolded protein response,” may be involved in neuroinflammation and cellular death mechanisms (Fig. 2A). The KEGG pathway analysis highlighted the enrichment of pathways such as “autophagy,” “apoptosis,” and “NF-κB signaling pathway,” suggesting that these biological processes play a significant role in the pathogenesis of ALS (Fig. 2B).

Fig. 2.

Functional enrichment analysis and immune infiltration analysis. (A) GO enrichment analysis. (B) KEGG enrichment analysis. (C) Immune infiltration analysis. (D) Candidate biomarker immune infiltration landscape

Immune infiltration analysis

The results of the immune infiltration analysis indicate a significant disruption of the immune microenvironment in patients with ALS (p < 0.05) (Fig. 2C). The primary characteristics of this disruption include an imbalance between adaptive and innate immunity. Notably, there are significant differences in activated CD4⁺/CD8⁺ T cells, pro-inflammatory Th1/Th17 cells, and regulatory T cells (Tregs) among the groups. Additionally, myeloid immune cells, including monocytes, neutrophils, and activated dendritic cells, also show significant intergroup differences. This overactivation of T cells, along with increased infiltration of myeloid cells, collectively drives the neuroinflammatory response associated with ALS. Subsequently, we constructed an immune landscape of potential biomarkers for ALS, revealing that the identified circulating biomarkers are associated with various immune cell types (Fig. 2D).

Selection and validation of hub biomarkers

This study utilized an internal training dataset (GSE12680) and an external independent validation dataset (GSE12676) to systematically screen 113 combinations of machine learning algorithms for the construction of a diagnostic model for ALS. The objective was to evaluate the diagnostic performance and contribution of four pre-selected potential biomarkers (FCRL3, HTATIP2, RNASE6, SF3B4). The results indicated that the Random Forest (RF) algorithm exhibited the best performance, with an area under the curve (AUC) of 0.786, and was therefore selected as the optimal diagnostic model (Fig. 3A). This model was employed to validate the diagnostic efficacy of these genes (Fig. 3B). Furthermore, we employed the SHAP method within an interpretable machine learning framework to quantitatively assess the relative contributions of these hub genes, yielding the following SHAP values: RNASE6 > FCRL3 > HTATIP2 > SF3B4 (Fig. 3C). The diagnostic efficacy of the hub genes was validated through ROC curve analysis, revealing an AUC of 0.985 for the internal validation set and 0.588 for the external validation set (Fig. 3D). A subsequent nomogram constructed using a logistic regression model demonstrated a significant association between the expression levels of the hub genes and the risk of ALS (Fig. 3E). Calibration curves indicated a high consistency between the predicted and actual risks (Fig. 3F). DCA showed that the hub gene prediction model significantly enhanced clinical net benefits across a risk threshold range of 20% to 95%, outperforming both “intervene in all” and “no intervention” strategies (Fig. 3G). In summary, FCRL3, HTATIP2, RNASE6, and SF3B4 may serve as potential biomarkers for ALS. The machine learning framework offers significant value in performance validation and contribution ranking.

Fig. 3.

Selection and Validation of Hub Genes. (A) Construction of diagnostic models. (B) Random forest model. (C) SHAP interpretation of the Random forest model. (D) ROC validation curves for hub genes in internal and external ALS datasets. (E) Logistic regression model nomogram. (F) Calibration curve. (G) Decision curve analysis

Toxicity analysis of environmental toxins

Based on reverse virtual screening of hub targets, we identified three environmental toxins: Benzo(a)pyrene, Bisphenol A, and Triphenyl phosphate. Predictions from ADMETlab 3.0 indicated that all three environmental toxins exhibited significant neurotoxicity, with Benzo(a)pyrene demonstrating the strongest neurotoxic effects and a broader range of multi-organ toxicity (Fig. 4).

Fig. 4.

Toxicity predictions of environmental toxins

Molecular docking results

Benzo(a)pyrene, Bisphenol A, and Triphenyl phosphate formed multiple hydrogen bonds with the hub proteins FCRL3, HTATIP2, RNASE6, and SF3B4, with binding energies all below − 5 kcal·mol⁻¹ (Fig. 5A and B), indicating a strong binding affinity between these compounds and the hub proteins. Notably, Benzo(a)pyrene exhibited significantly lower binding energies with all four proteins, with the HTATIP2-Benzo(a)pyrene interaction showing the lowest binding energy (∆G = -9.4 kcal·mol⁻¹). This suggests that Benzo(a)pyrene may alter the conformation and function of the HTATIP2 protein through specific binding, potentially mediating neurotoxic effects.

Fig. 5.

Molecular docking. (A) Free energy of binding between environmental toxins and hub proteins. (B) Visualization of the docking results between environmental toxins and hub proteins

Discussion

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease currently lacking specific diagnostic biomarkers and treatment options. Environmental toxins are significant risk factors for ALS and may contribute to the disease’s onset and progression through gene-environment interactions; however, the precise molecular mechanisms remain unclear. Given the critical value of circulating biomarkers in early disease diagnosis and dynamic monitoring, this study integrates multi-omics data and employs Mendelian randomization and reverse network toxicology to identify potential circulating biomarkers for ALS, while also exploring their interactions with environmental pollutants. The aim is to provide new insights into the pathogenesis of ALS and to develop corresponding diagnostic and therapeutic strategies.

Through Mendelian randomization analysis, we identified 68 plasma proteins significantly associated with ALS risk (33 positively correlated and 35 negatively correlated). Combining these findings with whole blood transcriptomic data, we selected four potential circulating biomarkers: FCRL3, HTATIP2, RNASE6, and SF3B4. To date, no studies have reported an association between these biomarkers and ALS; however, machine learning models constructed using these circulating biomarkers demonstrated promising diagnostic performance, offering important clues for the diagnosis and treatment of ALS. Functional enrichment analysis revealed correlations between biological processes such as “autophagy,” “apoptosis,” “unfolded protein response,” and the “NF-kappa B signaling pathway” with ALS pathogenesis. Additionally, immune infiltration analysis indicated a significant disruption of the immune microenvironment in ALS patients.

Autophagy is a critical pathway for the clearance of abnormal protein aggregates. In ALS, impaired autophagy leads to the accumulation of toxic proteins, such as SOD1 aggregates, which accelerates the degeneration of motor neurons [42, 43]. Furthermore, mutations in autophagy-related genes, such as p62/SQSTM1, are closely associated with the pathology of ALS, providing further evidence that the imbalance between autophagy and protein homeostasis is a significant pathogenic mechanism in this disease [44]. In ALS patients, there is persistent activation of endoplasmic reticulum (ER) stress and an imbalance in the unfolded protein response (UPR) within motor neurons [45–47]. ER stress can activate the IKKβ-NF-κB pathway, exacerbating neuroinflammatory responses [48]. While the UPR is an adaptive response intended to restore protein homeostasis, in ALS, it shifts to a state of chronic activation, which can trigger apoptosis through pathways such as JNK-AP1 and CHOP [49]. Moreover, the UPR and autophagy work together to maintain protein homeostasis, and their dysregulation can lead to the accumulation of proteins like p62 [44]. The NF-κB signaling pathway contributes to the pathological progression of ALS by activating microglia and astrocytes, which release inflammatory factors, thereby exacerbating motor neuron damage [50]. Studies have shown that inhibiting NF-κB activation can effectively reduce inflammatory responses and improve pathological changes associated with mutant hSOD1 by promoting autophagy and inhibiting oxidative stress [50]. Consequently, in the pathogenesis of ALS, autophagy, apoptosis, UPR, and NF-κB-related pathways form a multidimensional interactive network. Targeting the regulation of these pathways—such as enhancing autophagic clearance, inhibiting ER stress, or reducing NF-κB-mediated inflammation—may represent a potential therapeutic direction for ALS.

The pathogenesis of ALS is closely linked to dysregulation of the immune system, particularly the alterations in peripheral immune cells that play a critical role in disease progression [51, 52]. T cells represent the predominant population of peripheral cells infiltrating the central nervous system in ALS patients. Research indicates that although CD4⁺ T cells possess neuroprotective properties, their numbers are reduced and their regulatory functions are impaired in individuals with ALS [53]. Conversely, cytotoxic CD8⁺ T cells selectively induce the death of motor neurons through interactions with MHC-I class molecules and self-antigen complexes [54]. The quantity and functionality of regulatory T cells (Tregs) are negatively correlated with disease progression, suggesting they may serve as a novel therapeutic target [55, 56]. Further studies have revealed a shift in T cells within the peripheral blood of ALS patients towards pro-inflammatory Th1 and Th17 cell subsets, accompanied by a decrease in anti-inflammatory Th2 and Treg cells. Notably, inflammatory cytokines such as IL-1β, IL-6, and IFN-γ are significantly elevated, indicating that Th1/Th17 cell-mediated pro-inflammatory responses play a crucial role in ALS [57]. Additionally, changes in the proportions of peripheral blood monocytes and neutrophil subpopulations are closely associated with disease severity and respiratory dysfunction [58]. In summary, the imbalance of these immune cells is interrelated; for instance, the reduction of Tregs leads to uncontrolled inflammation driven by CD4⁺/CD8⁺ T cells and monocytes, creating a neurotoxic feedback loop. This highlights the immune system as a key target in the pathogenesis of ALS [51].

Through reverse virtual screening of hub targets, we identified three environmental toxins with significant neurotoxic effects: benzo[a]pyrene (BaP), bisphenol A (BPA), and triphenyl phosphate (TPHP). Among these, BaP exhibits the strongest neurotoxicity. BaP, a representative polycyclic aromatic hydrocarbon (PAH), can induce the expression and abnormal aggregation of TDP-43, a core pathological protein associated with ALS, by activating the aryl hydrocarbon receptor (AhR) signaling pathway. This process accelerates the degeneration of motor neurons. However, the direct association between BaP and ALS requires further epidemiological and mechanistic studies for confirmation [20, 59]. Exposure to BaP can trigger microglial activation and mitochondrial dysfunction, aligning with the mechanisms of neuroinflammation and oxidative stress observed in ALS [60]. In animal models, BaP-induced brain tissue damage is associated with motor function impairment, suggesting its potential role in promoting neurodegeneration [60]. Furthermore, epidemiological evidence indicates the presence of PAHs in the cerebrospinal fluid of ALS patients, and occupational exposure (such as diesel exhaust and industrial emissions) is linked to an increased risk of ALS [61–64]. BPA exposure has been shown to cause degeneration of motor neurons in zebrafish embryos, accompanied by impaired motor function, providing experimental evidence for ALS pathological phenotypes; however, direct evidence of its impact on ALS patients remains limited [65]. TPHP, as an organophosphate flame retardant, has an unclear neurotoxic mechanism and lacks direct evidence linking it to ALS. In summary, future research should focus on elucidating the underlying mechanisms and conducting prospective cohort studies on mixed exposure.

This study has the following limitations: First, Mendelian randomization analysis relies on genetic instrumental variables. Although sensitivity analyses have confirmed the robustness of the findings, residual confounding factors may still affect causal inferences. Second, the interactions between environmental toxins and hub proteins are based solely on computational simulations, and their biological functional impacts require further experimental validation. Therefore, future work will involve using animal models to verify the regulatory effects of environmental toxins on target proteins and elucidate the mechanisms of neurotoxicity.

Conclusions

This study integrates plasma proteomics from deCODE with whole blood transcriptomics data from patients with ALS. Through MR and bioinformatics analyses, we identified four potential circulating biomarkers for ALS: FCRL3, HTATIP2, RNASE6, and SF3B4. A machine learning model constructed based on these hub biomarkers demonstrated strong diagnostic performance, providing important insights for the diagnosis and treatment of ALS. Functional enrichment analysis revealed correlations between biological processes such as “autophagy,” “apoptosis,” “endoplasmic reticulum unfolded protein response,” and the “NF-κB signaling pathway” with the pathogenesis of ALS. Additionally, immune infiltration analysis indicated significant T cell hyperactivation and increased myeloid cell infiltration in ALS patients. Furthermore, the study explored the role of environmental factors, identifying that environmental toxins such as benzo(a)pyrene may exert neurotoxic effects through specific binding to hub target proteins. This research offers new insights and potential targets for understanding the pathogenesis of ALS, early diagnosis, and personalized treatment strategies from a gene-environment interaction perspective.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

MR

Mendelian randomization

ALS

Amyotrophic lateral sclerosis

SHAP

SHapley Additive exPlanations

CTD

Comparative Toxicogenomics Database

GWAS

Genome-wide association study

GEO

Gene expression omnibus

BaP

Benzo(a)pyrene

BPA

Bisphenol A

TPHP

Triphenyl phosphate

DEGs

Differentially expressed genes

RF

Random Forest

GO

Gene Ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

ROC

Receiver Operating Characteristic

AUC

Area Under the Curve

DCA

Decision Curve Analysis

Author contributions

Methodology, writing- original draft preparation: L.X. and B.H; Software, data curation: Y.Z; Writing- review and editing, project administration: X.L. and T.C; Conceptualization: H.H. All authors have read and agreed to the published version of the manuscript.

Funding

This work was financially supported by the Scientific Research Foundation of Yunnan Provincial Education Department (2024Y392; 2024Y389), the Natural Science Foundation of Hunan Province (2024JJ7319; 2025JJ70465; 2025JJ70442), the doctoral research project initiation fund at Hunan University of Medicine (202412), the international Cooperative Project of Traditional Chinese Medicine (2541STC72898), the Reform Project of Hunan Provincial Education Department (202401001789).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.