Comprehensive analysis of autoimmune-related genes in amyotrophic lateral sclerosis from the perspective of 3P medicine

Abstract

Background

Although growing evidence suggests close correlations between autoimmunity and amyotrophic lateral sclerosis (ALS), no studies have reported on autoimmune-related genes (ARGs) from the perspective of the prognostic assessment of ALS. The purpose of this study was to investigate whether the circulating ARD signature could be identified as a reliable biomarker for ALS survival for predictive, preventive, and personalized medicine.

Methods

The whole blood transcriptional profiles and clinical characteristics of 454 ALS patients were downloaded from the Gene Expression Omnibus (GEO) database. A total of 4371 ARGs were obtained from GAAD and DisGeNET databases. Wilcoxon test and multivariate Cox regression were applied to identify the differentially expressed and prognostic ARGs. Then, unsupervised clustering was performed to classify patients into two distinct autoimmune-related clusters. PCA method was used to calculate the autoimmune index. LASSO and multivariate Cox regression was performed to establish risk model to predict overall survival for ALS patients. A ceRNA regulatory network was then constructed for regulating the model genes. Finally, we performed single-cell analysis to explore the expression of model genes in mutant SOD1 mice and methylation analysis in ALS patients.

Results

Based on the expressions of 85 prognostic ARGs, two autoimmune-related clusters with various biological features, immune characteristics, and survival outcome were determined. Cluster 1 with a worsen prognosis was more active in immune-related biological pathways and immune infiltration than Cluster 2. A higher autoimmune index was associated with a better prognosis than a lower autoimmune index, and there were significant adverse correlations between the autoimmune index and immune infiltrating cells and immune responses. Nine model genes (KIF17, CD248, ENG, BTNL2, CLEC5A, ADORA3, PRDX5, AIM2, and XKR8) were selected to construct prognostic risk signature, indicating potent potential for survival prediction in ALS. Nomogram integrating risk model and clinical characteristics could predict the prognosis more accurately than other clinicopathological features. We constructed a ceRNA regulatory network for the model genes, including five lncRNAs, four miRNAs, and five mRNAs.

Conclusion

Expression of ARGs is correlated with immune characteristics of ALS, and seven ARG signatures may have practical application as an independent prognostic factor in patients with ALS, which may serve as target for the future prognostic assessment, targeted prevention, patient stratification, and personalization of medical services in ALS.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13167-022-00299-w.

Introduction

Amyotrophic lateral sclerosis (ALS), an adult-onset neurodegenerative disorder, is the most frequent entity that affects motor neurons (MNs). ALS is characterized by the degeneration of upper and lower MNs, leading to death within 3 years after symptom onset, usually as a consequence of respiratory failure [1]. The incidence of ALS in Europe and North America is reported to be between 1.5 and 3 per 100,000 people per year. It is estimated that more than 15 million people who are alive at present will succumb to ALS in the USA and the UK, a statistic that suggests that more than 1 in 500 deaths in adults are caused by ALS. The clinical types of the disease are heterogeneous based on the involvement of a different set of MNs or a different region. For example, in accordance with the location of the main pathology, patients with ALS might develop weakness with flaccidity and atrophy of the limbs, hyperreflexia, spasticity with increased limb tone but little muscle loss, and an atrophied tongue with thick speech and swallowing difficulties [1]. ALS has been classically divided into familial ALS (fALS) and sporadic ALS (sALS), which accounts for approximately 10% of cases. However, there are no effective treatments available for ALS to date. The white paper of the “European Association for Predictive, Preventive and Personalized Medicine (PPPM)” (EPMA) mentions that there is a need of reliable markers for the identification of high-risk individual with neurodegenerative diseases [2]. From the viewpoint of PPPM, understanding the risk factors and identifying biomarkers of clinical progression for ALS including any pathogenic pathways will improve the chances of developing effective therapeutics.

Although the etiology and pathogenesis of the disease remain unknown, growing evidence supports the role for autoimmune mechanisms in ALS. The association of ALS with autoimmune disorders (AIDs) has gained increased attention during recent years under the PPPM paradigm. For example, Turner et al. conducted a case–control study and found that the risk of ALS among patients with previous AIDs, such as a prior diagnosis of asthma, celiac disease, younger-onset diabetes (younger than 30 years), multiple sclerosis, and myasthenia gravis, was significantly higher than expected [3]. Similar results were also demonstrated by a recent study, which showed that patients with ALS had a 47% higher risk than control subjects of being previously diagnosed with AIDs [4]. In addition, the preexistence of AIDs also affects the clinical features and prognosis of patients with ALS. Li et al. recently found that patients with coexisting ALS and AIDS had older onset ages and worse respiratory function but similar overall survival than those with pure ALS [5]. There are some genetic causal relationships between AIDs and ALS. A Mendelian randomization study using genome-wide association study (GWAS) summary statistics demonstrated the causal neuroprotective role of type 2 diabetes on ALS in the European population and provided empirically suggestive evidence of an increased risk of type 2 diabetes on ALS in the East Asian population [6]. A recent study examined the genetic relation between ALS and 10 AIDs and found that there was a significant positive genetic correlation between ALS and celiac disease, multiple sclerosis, rheumatoid arthritis (RA), and systemic lupus erythematosus (SLE) as well as a significant negative genetic correlation between ALS and Crohn’s disease, inflammatory bowel disease, and ulcerative colitis [7].

Repeat expansions in C9orf72 have revolutionized our understanding of ALS [8]. Mutations in C9orf72 have been reported to be the most prevalent genetic variant in both fALS and sALS cases [9]. Patients with C9orf72-related ALS have been reported to have an unexpected increase in AIDs prevalence [10]. C9orf72 − / − knockout mice exhibit immune dysregulation and develop features of autoimmunity [11]. The inflammation induced by the stimulator of interferon genes can be suppressed by C9orf72 [12]. C9orf72 could also be a genetic risk factor for autoimmune conditions. Both the number of patients with intermediate expansions and the overall number of intermediate alleles of C9orf72 in the SLE and SLE + RA cohorts were significantly higher than in controls [13].

Although growing evidence suggests close correlations between autoimmunity and ALS, no studies have reported on autoimmune-related genes (ARGs) from the perspective of the prognostic assessment of ALS. Bioinformatics strategies are one of the essential components of the conversion of traditional medicine to PPPM [14]. In our study, ARGs were collected from the Gene and Autoimmune Disease Association Database (GAAD) [15] and DisGeNET databases [16]. The development of molecular biology and next-generation sequencing technologies has enabled researchers to study disease mechanisms at the genetic and mRNA levels. The purpose of this study was to investigate whether peripheral ARGs could be identified as a prognostic factor for patients with ALS. In this study, we first screened out differentially expressed ARGs (DEARGs) based on the transcriptional profiles of whole blood of patients with ALS from the Gene Expression Omnibus (GEO) database. Prognostic DEARGs were further identified using univariate Cox regression. Two autoimmune clusters were identified by performing an unsupervised cluster analysis using the R package “ConsensusClusterPlus.” We constructed a nine-ARG signature for predicting survival by integrating least absolute shrinkage and selection operator (LASSO) regression and multivariate Cox regression. It was estimated using a receiver operating characteristic (ROC) curve and survival analysis. Then, a nomogram was built by combining the risk score and clinical parameters to predict the overall prognostic value for patients with ALS. The concordance index (C-index), ROC curve analysis, calibration curve analysis, and decision curve analysis (DCA) were performed to assess the nomogram. The biological functions and immune characteristics between different autoimmune clusters and risk groups were evaluated with a comprehensive bioinformatics analysis. In addition, competitive endogenous RNA (ceRNA), where long non-coding RNAs (lncRNAs) compete with protein-coding mRNAs for binding to miRNAs, represents a novel layer of gene regulation that controls both physiological and pathological processes. A ceRNA network for the nine model genes was constructed in our analysis. We also performed single-cell sequence (scRNA-seq) analysis to explore the expression of model genes in mutant SOD1 mice. Finally, the methylated positions of model genes between ALS patients and controls were also explored. Collectively, our findings highlight the functional role of ARGs in predicting the prognosis of ALS and provide new insights to elucidate the mechanisms of immune regulation in patients with ALS. The exploration of ARGs as biomarkers of ALS could inform high-risk populations of how to moderate prognosis through personalized medicine from the viewpoint of PPPM.

Methods and materials

Datasets and preprocessing

Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/) is a public functional genomics data repository of high-throughput gene expression data, chips, and microarrays. A total of six ALS patient datasets (GSE112676, GSE112680, GSE148097, GSE106443, GSE163560, and GSE89472) were obtained from the NCBI GEO database. The GSE112676 dataset containing 233 patients with ALS and 508 controls was collected using the platform GPL6947 Illumina HumanHT-12 V3.0 Expression BeadChip. The GSE112680 dataset was based on an Illumina HumanHT-12 V4.0 Expression BeadChip platform (GPL10558) and included 164 ALS and 137 control samples. The GSE148097 dataset is a miRNA profile including 13 ALS and six control samples that was measured using the platform GPL16791 Illumina HiSeq 2500. The GSE106443 (platform GPL18573 Illumina NextSeq 500) is composed of peripheral blood collected from 17 patients with ALS and three healthy controls. The GSE106443 dataset includes lncRNA and mRNA profiles, while only the lncRNA profile was extracted in this work. From the GSE163560 database (platform GPL23126 Affymetrix Human Clariom D Assay), which includes 36 patients who were randomly assigned to three treatment arms (1MIU IL-2, 2MIU IL-2, and placebo), we selected data from the 2MIU IL-2 group on days 1, 8, 64, and 85 after injection of IL-2. GSE89472 dataset is a methylation dataset, including five monozygotic twin pairs discordant for ALS (e.g., five ALS patients and five healthy controls). The detailed information about the included datasets is listed in Table 1.

Table 1.

Descriptions of datasets included in the present study

Accession	Platform	Type	Sample
GSE112676	GPL6947	mRNA profile	ALS:HC = 233:508
GSE112680	GPL10558	mRNA profile	ALS:HC = 164:137
GSE148097	GPL16791	miRNA profile	ALS:HC = 13:6
GSE106443	GPL18573	lncRNA profile	ALS:HC = 17:3
GSE163560	GPL23126	mRNA profile	Day1:day8:day64:day85 = 12:11:24:12*
GSE178693	GPL21103	scRNA-seq	SOD1(G93A) mutant:wild-type mice = 2:2
GSE89472	GPL13534	Methylation profiling	ALS:HC = 5:5

ALS, amyotrophic lateral sclerosis; HC, healthy controls. * Treated with 2MIU-IL-2

The GSE112676 and GSE112680 datasets were based on the Illumina platform. Both dataset have the same clinicopathological information of ALS patients, including the age at onset (the time of symptom onset), the site of onset (bulbar region or the spinal), the gender (male or female), the follow-up period, and the survive status (live or death). The background-corrected GSE112676 and GSE112680 expression matrix was quantile normalized using the normalizeBetweenArrays function of the “limma” R package [17]. If multiple probes were matched with one gene, the probe with the maximal median values of expression was annotated into the homologous gene symbol based on the platform’s annotation information. We used the “ComBat” algorithm to remove batch effects in the two microarray datasets [18]. Principle component analysis (PCA) before and after normalization and batch corrections from the microarray data was used to show the removal of batch effects in the raw data. Our study design is briefly described in the flow chart (Fig. 1).

Fig. 1.

Experimental technical roadmap of the present study. ALS, amyotrophic lateral sclerosis; ARGs, autoimmune-related genes; DEARGs, differentially expressed ARGs; LASSO, least absolute shrinkage, and selection operator; PCA, principal component analysis; ROC, receiver operating characteristic

Biological pathways enriched in patients with ALS patients

The gene sets of “c5.go.bp.v7.5.1.symbols,” “h.all.v7.5.1.symbols,” and “c2.cp.kegg.v7.5.1.symbols” were downloaded from the MSigDB database [19] to enrich the biological processes (BPs) of gene ontology (GO), HALLMARK pathways, and KEGG pathways. The “fgsea” package in R was used to display the enrichment results of gene set enrichment analysis (GSEA) for GO:BP. The GSVA algorithm was used to calculate the HALLMARK and KEGG pathway activation score, which was applied using the R package “GSVA” [20]. The R package “limma” was used to compare the differences in pathway activation scores between patients with ALS and controls.

Identification of DEARGs

ARGs were collected from the GAAD and DisGeNET databases. A total of 4371 ARGs were obtained after deleting duplicate genes. To identify differentially expressed genes (DEGs) between controls and patients with ALS, we performed a Wilcoxon test. The genes with a p value < 0.05 were regarded as DEGs, which were further intersected with the ARGs. The overlapping genes were designated as DEARGs. The KEGG pathways of DEARGs were enriched using the “clusterProfiler” package.

Unsupervised cluster analysis of prognostic DEARGs in ALS

Univariate Cox regression analysis was performed for the DEARGs in patients with ALS. Those with p < 0.05 were considered to be significant. The protein–protein interaction (PPI) network of the genes was constructed in the STRING online tool [21]. Then, the prognostic DEARGs in ALS were identified. Based on the expression profile of the prognostic DEARGs, we performed an unsupervised cluster analysis to identify distinct autoimmune subtypes using the R package “ConsensusClusterPlus” [22]. A Kaplan–Meier (KM) curve was used to show the relationship between two clusters and clinical outcome, and the log-rank test was used to evaluate differences using the “survival” and “survminer” packages.

Biological functions and immune characteristics between the two clusters

The abovementioned GSVA algorithm was used to calculate the HALLMARK and KEGG pathway activation score between the two clusters using the R package “GSVA.”

xCell [23], a novel gene signature-based method, was used to estimate the immune scores, stromal scores, and microenvironment scores between two clusters. A single-sample GSEA (ssGSEA) algorithm was introduced to quantify the relative infiltration of 28 immune cell types and 17 immune activities. The expression levels of 11 checkpoint-related genes (PDCD1, CD274, CTLA4, ICOS, HAVCR2, CD80, CD47, BTLA, TIGIT, SIRPA, and VTCN1) were also compared between the two clusters.

Autoimmune index generation and correlations with immune characteristics in ALS

To quantify the autoimmune level per individual, we established an evaluation index called the autoimmune index. PCA was conducted based on prognostic DEARGs using the prcomp function to assess the distinguishable ability for identified subtypes. PC1 and PC2 were extracted to form signature scores. We then applied a method similar to GGI to construct the m6A score [24]:

$$\mathrm{Autoimmune}\mathrm{index}=\sum (PC1i+PC2i)$$

where i indicates the expression of prognostic DEARGs.

The relationships between the autoimmune index and immune characteristics (immune cells, immune activities, and immune checkpoints) were also assessed using Pearson’s correlation analysis.

Establishment and verification of a prognostic model based on prognostic DEARGs

A total of 397 patients with ALS were divided into a training cohort and a test cohort in a ratio of 5:5. Least absolute shrinkage and selection operator (LASSO) analysis was conducted to downsize the prognostic DEARGs previously filtered using the “glmnet” R package. We chose the minimum lambda as the optimal value. The hub model genes used to establish the risk model were determined by multivariate Cox regression analysis. The risk score for each sample was calculated as follows: Risk scores = ∑(coefficienti × expression of signature genei). The patients were divided into high-risk and low-risk groups according to the medium value. The prognostic evaluation ability of the risk scores was evaluated by plotting the KM survival curve and ROC curve using the “timeROC” package. The test cohort and overall cohort were used to validate the prognostic model. The relationships between risk groups and the clinicopathological features are depicted as histograms of the frequency distribution and Sankey plots.

Biological functions and immune characteristics between the two risk groups in the overall cohort

The GSEA method was used to identify the different KEGG pathways between the two risk groups in the entire cohort. The xCell algorithm was applied to assess the stromal and immune scores between the high- and low-risk groups. As before, we utilized the ssGSEA method to assess the abundance of different immune cells and responses.

Nomogram construction

To improve the prognostic risk stratification of patients with ALS and to assist in clinical diagnosis and treatment, we constructed a nomogram model with the clinicopathological features and risk scores using the “rms” package. It was used as a quantitative tool to predict the prognosis of patients with ALS. Calibration plots of the nomogram were used to depict the predictive value between the predicted 1-, 3-, and 5-year survival events and the virtually observed outcomes. Time-dependent ROC curves for the 1-, 3-, and 5-year survival rates were used to assess the nomogram. DCA was performed to estimate the clinical utility of the nomogram by calculating the net benefits at different threshold probabilities.

Construction of a ceRNA network and gene expression responses to IL-2

The GSE148097 and GSE106443 datasets were used to identify the differentially expressed miRNA (DEmiRNA) and lncRNA (DElncRNA) using the “limma” package with the criteria of p < 0.05. We then predicted the miRNA of hub model genes using multiMiR [25], a new miRNA-target interaction R package and database. The predicted miRNA intersected with the DEmiRNA. Then, we also predicted the targets of the overlapping miRNA that intersected with DElncRNA to find the overlapping lncRNA. Finally, a lncRNA-miRNA-mRNA ceRNA network for ALS was constructed.

Low-dose IL-2 has been proposed as an immune-modulatory strategy to increase regulatory T cells (Tregs) in patients with ALS and decrease neuroinflammation [26]. In our study, we evaluated the gene expression responses to IL-2 of hub model genes in the GSE163560 dataset.

Expression of model genes in mutant SOD1 mice using scRNA-seq analysis

Superoxide dismutase 1 gene (SOD1) was the first ALS-associated gene and linked to 15% of fALS cases [27], leading to the breakthrough with the first appearance of transgenic model of SOD1-G93A mice [28]. GSE178693 dataset is a single-cell transcriptome dataset containing samples from the brainstem of two mutant SOD1 mice and two age-matched non-carrier wild-type mice. The analysis of single-cell data follows the “Seurat” procedure [29]. Samples with less than 200 genes, more than 20% mitochondria genes, and more than 1% ribosomal genes were filtered out. “FindVariableFeatures” function was used to find the top 2000 variable gene. “FindIntegrationAnchors” and “IntegrateData” function was used to integrate multiple samples. Uniform manifold approximation and projection (UMAP) with a resolution of 0.3 was used to show main cell clusters. Clusters were annotated to identify cell types by using “SingleR” package.

Methylation analysis of model genes

The GSE89472 dataset (detected in GPL13534 Illumina HumanMethylation450 BeadChip) contains array-based gene methylation profiles of blood samples from five ALS patients and 5 healthy donors. The ChAMP package in R was used to process the data [30]. Differences in methylation levels between ALS patients and healthy controls were calculated using the “champ.DMP” function. The methylation patterns of positions of model genes were visualized using “ComplexHeatmap” package in R.

Statistical analysis

All statistical analyses were performed using R software (version R-4.1.0). The Wilcoxon test was used for statistical analysis between the two groups, and the Kruskal–Wallis test was used to compare more than two groups. Categorical variables were compared using chi-square analysis. Unsupervised cluster analysis was performed using the R package “ConsensusClusterPlus.” The KM curve plotted the relationship between score and clinical outcome, and the log-rank test was used to evaluate differences using the “survival” and “survminer” packages. Univariate and multivariate Cox regression analyses were performed with the “survival” package. The relationships of genes with genes and those of genes with immune cells were constructed using Pearson’s correlation method. LASSO regression analysis was performed using the “glmnet” package. A nomogram model was constructed using the “rms” package. p < 0.05 was considered to indicate statistical significance. The significance level is denoted as follows: *p < 0.05, **p < 0.01, and ***p < 0.001.

Results

Functional analysis and screening of DEARGs

We merged the GSE112676 and GSE112680 datasets into an entire cohort for further analysis. The PCA plots showed that the batch effect was removed (Fig. 2A and B). We used the GSEA method to uncover different GO:BP between patients with ALS and controls. The top up- and downregulated processes were depicted, indicating that the positive regulation of IL-10 production and the amyloid precursor protein catabolic process, axonal fasciculation, and modulation of age-related behavioral decline were most upregulated in patients with ALS (Fig. 2C). We also utilized a GSVA method to find different hallmark pathways between patients with ALS and controls. The results showed that immune-related pathways, such as inflammatory responses, complement responses, allograft rejection, and interferon responses were more enriched in patients with ALS (Fig. 2D).

Fig. 2.

Differences of biological functions between ALS and controls and identification of DEARGs. A PCA plot before (A) and after (B) removal of batch effects. The points of the scatter plots represented each samples based on the top two principal components (PC1 and PC2) of gene expression profiles. C The top up- and downregulated biological processes of ALS compared with the controls. D A heat map showing the differences of Hallmark pathways between ALS and controls. E A Venn plot by overlapping the DEGs and ARGs to identify DEARGs. F The KEGG enrichment results of DEARGs

We performed a Wilcoxon test to identify the DEGs between patients with ALS and controls. A total of 7607 DEGs were identified and 1612 DEARGs were further screened by intersecting with the ARGs (Fig. 2E). The expression patterns of these genes were visualized with a heat map (Fig. S1). The KEGG results showed that these DEARGs were involved in NOD-like receptor signaling pathway, natural killer cell-mediated cytotoxicity, and Th17, Th1, and Th2 cell differentiation (Fig. 2F).

Molecular subgroups clustered by prognostic DEARGs

A total of 85 prognostic DEARGs were screened using univariate Cox regression analysis (Table S1). The PPI network of these 85 genes is depicted with an interaction score of 0.15 (Fig. S2A). A consensus clustering approach was conducted to divide the patients with ALS in the entire cohort into subgroups based on these 85 prognostic genes. The optimal clustering stability was identified when K = 2 (Fig. 3A and Fig. S3). The 397 patients with ALS were divided into two subtypes: Cluster 1 (n = 271) and Cluster 2 (n = 126). A heat map showed the expression patterns of the 85 genes between the two clusters (Fig. S2B). We compared the survival of the two clusters and found that Cluster 2 showed better overall survival (HR = 0.567, p < 0.001) (Fig. 3C). A histogram of the frequency distribution revealed that Cluster 1 had a larger proportion of deaths and age at onset ≥ 65 years than Cluster 2 (Fig. 3D and E) (Chi-square test, p = 0.009 and 0.012, respectively), while there was no difference in sex or the site of onset distribution between the two clusters (Fig. 3F and G) (Chi-square test, p = 0.683 and 0.413, respectively).

Fig. 3.

Consensus cluster based on the expression matrix of DEARGs. A Consensus clustering matrix for k = 2. B Consensus clustering of cumulative distribution function (CDF) for k = 2–10. C Survival curve of the patients in the two clusters. The histogram of the frequency distribution of survival status (D), age at onset (E), sex (F), and site of onset (G) in two clusters

Biological functions and immune characteristics between the two clusters

GSVA and GSEA were performed to explore the biological behaviors of the two clusters. Compared with Cluster 1, Cluster 2 showed enriched hallmark pathways related to inflammation and apoptosis, such as allograft rejection, TNF-α signaling via NF-κB, and IL6/JAK/STAT3 signaling (Fig. 4A). These results were consistent with the KEGG results determined by GSVA (Fig. 4B). In our GSEA results, we also found that the TNF signaling pathway, inflammatory bowel disease, and ferroptosis were involved in Cluster 1, which included patients with worse survival (Fig. S4).

Fig. 4.

Biological enrichment analysis and immune infiltration for distinct autoimmune phenotypes. A and B revealed the Hallmark and KEGG pathways in two different autoimmune phenotypes. C The immune scores and stromal scores were calculated by the xCell method. D Box diagram of the proportion of 28 types of immune cells using ssGSEA algorithm. E Box diagram of the proportion of 17 types of immune responses using ssGSEA algorithm. F The expression levels of 11 immune checkpoint-related genes. Significance level was denoted by *p value < .05, **p value < .01, ***p value < .001

To investigate the relationship between subtypes and immune characteristics in the whole blood of patients with ALS, we performed xCell analysis to calculate the immune scores and used the ssGSEA method to evaluate the immune cell infiltration, immune responses, and expression of immune checkpoints. The xCell results indicated that Cluster 2 had a lower immune score than Cluster 1 (Fig. 4C). Consistent with the xCell results, the immune cell infiltration analysis showed that most immune cells were more significantly enriched in Cluster 1, such as activated CD8 + T cells, activated CD4 + T cells, type 1 T helper cells, Tregs, activated B cells, eosinophils, and neutrophils, whereas the abundance of natural killer cells and myeloid-derived suppressor cells was enhanced in Cluster 2 (Fig. 4D). We further explored the immune responses in the subtypes and found that the BCR and TCR signaling pathway, interferon receptors, and TNF family member receptors were increased in Cluster 1, whereas chemokines, cytokines, cytokine receptors, and TGF-β family members were more enriched in Cluster 2 (Fig. 4E). The expression levels of immune checkpoints, such as CD274, CD47, and SIRPA, were downregulated in Cluster 2 (Fig. 4F).

Generation of the autoimmune index

Based on the expression levels of 85 prognostic DEARGs, we developed a PCA algorithm and established a scoring system called the autoimmune index to comprehensively quantify the autoimmune patterns of patients with ALS. The PCA plot showed that the two identified clusters were clearly distinguishable (Fig. 5A). Subsequently, the KM curve analysis and the log-rank test demonstrated that a higher autoimmune index was associated with a better prognosis than a lower autoimmune index (Fig. 5B). Further analysis revealed a higher autoimmune index, younger age at onset, and more alive patients than dead in Cluster 2 than in Cluster 1 (Fig. 5C–E), whereas no difference was observed between the two clusters in the site of onset or sex (Fig. 5F–G).

Fig. 5.

Generation of the autoimmune index. A Principal component analysis between two clusters based on the expression patterns of DEARGs. B Survive analysis of high- and low-autoimmune index. The comparison of an autoimmune index between two clusters (C), survival status (D), age at onset (E), sex (F), and site of onset (G). H The relationship between the autoimmune index and immune checkpoint genes. F The relationship between the autoimmune index with the immune cells and responses

In addition, we explored the associations between the autoimmune index and immune characteristics. The results showed that the autoimmune index was significantly correlated with most immune checkpoint genes (Fig. 5H). In particular, SIRPA was strongly negatively correlated with the autoimmune index (r = -0.598, p < 0.001). Furthermore, there were significant adverse correlations between the autoimmune index and immune infiltrating cells and immune responses (Fig. 5I). In particular, neutrophils and interferon receptors were most negatively correlated with the autoimmune index (abs (Cor) > 0.6).

Construction of a prognostic model based on the prognostic DEARGs

A risk signature model was constructed to assess the prognostic prediction value of ARGs in patients with ALS. The patients with ALS were randomly divided into a training cohort (n = 199) and a validation cohort (n = 198) at a ratio of 5:5. A summary of the clinicopathological features of patients in the training and validation cohorts is presented in Table 2. A lambda value of 0.0869 was used as the optimal lambda value for the LASSO algorithm in the training cohort to identify the potential prognostic genes in ARGs, and 14 genes with optimal lambda values were identified (Fig. 6A and B). Based on the genes generated from the LASSO analysis, multivariate Cox analysis using stepwise multivariate regression identified nine candidate genes to construct the risk model. The detailed information on the model genes is provided in Table 3. The correlation plot showed close relationships among these nine genes (Fig. S5A). The prognostic ability of each gene in the univariate Cox regression analysis was visualized via a forest plot and KM curve (Fig. S5B–K). In accordance with the constructed prognostic model, each patient was assigned with a risk score as follows: risk score = (− 4.053) × KIF17 + (− 0.965) × CD248 + (− 1.00) × ENG + (− 1.125) × BTNL2 + (− 1.26) × CLEC5A + (− 0.576) × ADORA3 + (− 0.314) × PRDX5 + (0.469) × AIM2 + (0.806) × XKR8. The coefficients of the nine genes are presented in Fig. 6C. The established risk model successfully classified the training cohort into high-risk and low-risk groups depending on the median value. The expression patterns of the nine model genes and distribution of survival status and risk scores of patients in the high- and low-risk groups are visualized (Fig. 6D). Survival analysis revealed that patients in the high-risk group had a poorer prognosis (p < 0.001) (Fig. 6E). Time-dependent ROC analysis was applied to further evaluate the prediction efficiency of the risk scores, with the area under the curve (AUC) values of 1, 3, and 5 years of 0.73, 0.75, and 0.78, respectively, in the training cohort (Fig. 6F). The Sanky plot indicated the relationships of risk groups with the clinicopathological features (Fig. 6G). The high-risk group had a higher proportion of Cluster 1 (Fig. 6H) and a death rate (Fig. 6I). The age at onset of the low-risk group was younger than that in the high-risk group (Fig. 6J) and more male patients was in low-risk group (Fig. 6K), but no significant difference in the site of onset was observed (Fig. 6L).

Table 2.

Characteristics of patients in training and validation cohorts

Variables		Overall	Training cohort	Test cohort	p value
No. of patients		397	199	198
Age of onset (mean (SD))		62.17 (11.95)	63.13 (12.49)	61.21 (11.33)	0.108
Sex (%)	Female	158 (39.80)	84 (42.21)	74 (37.37)	0.378
	Male	239 (60.20)	115 (57.79)	124 (62.63)
Site of onset (%)	Bulbar	146 (36.78)	75 (37.69)	71 (35.86)	0.784
	Spinal	251 (63.22)	124 (62.31)	127 (64.14)
Survival time (years; mean (SD))		2.88 (1.93)	2.89 (1.97)	2.87 (1.89)	0.938
Event (%)	Alive	55 (13.85)	30 (15.08)	25 (12.63)	0.575
	Death	342 (86.15)	169 (84.92)	173 (87.37)

SD, standard deviation

Fig. 6.

Construction of a risk model in the training cohort. A Feature selection by LASSO regression and B the coefficients change of different genes with different lambda. C The coefficients of selected nine model genes in multivariate Cox regression. D Distribution of risk score (up) and survival status (middle) of atherosclerotic patients in the high- and low-risk groups and heat map (down) illustrating the expression patterns of the nine candidate genes in the two groups. E Survival curve of the ALS patients in the two groups. F Time-dependent ROC curve of the risk model. G A Sankey diagram showing the distribution of risk groups and clinical features. The relative proportion of different clusters (H), survival status (I), age at onset (J), sex (K), and site of onset (L) in two risk groups

Table 3.

Detailed information of model genes

Gene	Description	Chr	HR*	HR 95%CI (lower)	HR 95%CI (upper)	p value
KIF17	Kinesin family member 17	1	0.144	0.041	0.506	0.003
CD248	CD248 molecule	11	0.590	0.417	0.836	0.003
ENG	Endoglin	9	0.419	0.221	0.794	0.008
BTNL2	Butyrophilin like 2	6	0.403	0.168	0.968	0.042
CLEC5A	C-type lectin domain-containing 5A	7	0.481	0.241	0.958	0.037
ADORA3	Adenosine A3 receptor	1	0.564	0.356	0.893	0.015
PRDX5	Peroxiredoxin 5	11	0.816	0.684	0.975	0.025
AIM2	Absent in melanoma 2	1	1.292	1.034	1.614	0.024
XKR8	XK-related 8	1	1.413	1.050	1.902	0.023

Chr, chromosome. * HR, hazard ratio of univariate Cox regression

Risk score as an independent risk factor validated by the test and overall cohorts

Univariate/multivariate Cox regression analyses revealed that the constructed risk model was an independent predictive marker of the prognosis of patients with ALS (Fig. 7A and B, Tables S2 and S3). These results demonstrated that the constructed risk model had excellent independence in predicting the prognosis in patients with ALS.

Fig. 7.

Independence of risk scores and validation in test and entire cohort. A Univariate and B multivariate Cox regression of risk scores and clinical characteristics in the training cohort. C, D, and E Distribution of risk score (up) and survival status (down) of ALS patients, survival curve, and time-dependent ROC curve of the risk model in the high- and low-risk groups in the test cohort, respectively. F, G, and H Distribution of risk score (up) and survival status (down) of ALS patients, survival curve, and time-dependent ROC curve of the risk model in the high- and low-risk groups in the entire cohort, respectively

To further test the stability of the risk scores, the predictive value was validated in the test cohort and the entire cohort. The test cohort was classified into low- and high-risk groups, and the risk score distributions and survival status are presented in Fig. 7C. The KM survival analysis showed that patients with high-risk scores demonstrated a prominent poor prognosis (log-rank test, p = 0.009; HR = 1.496) (Fig. 7D). However, the ROC curve showed that risk scores in the test cohort exhibited an unsatisfactory predictive value considering 1-, 3-, and 5-year AUC values, which were 0.57, 0.60, and 0.62, respectively (Fig. 7E). The relationships of the risk groups with the clinicopathological features in the test cohort were the same as those in the training cohort (Fig. S6A and B). Apart from the site of onset, cluster, survival status, age at onset, and sex showed a significant difference between the low- and high-risk groups.

The risk score distributions and survival status in the overall cohort are depicted in Fig. 7F. KM analysis indicated that patients with a high-risk score had a poorer prognosis than patients with a low-risk score (log-rank test, p < 0.001; HR = 2.063) (Fig. 7G). The 1-, 3-, and 5-year AUC values for the ROC analysis were 0.67, 0.67, and 0.71, respectively, in the entire cohort (Fig. 7H). Similar to those of the training and test cohorts, the relationships of the risk groups with the clinicopathological features in the overall cohort showed significant differences in cluster, survival status, age at onset, and sex between the low- and high-risk groups (Fig. S6C and D) (chi-square test, p < 0.001).

Clinical correlation analysis and stratification analysis of the risk score

The clinicopathological analysis in the entire cohort revealed higher risk scores in patients with an age at onset of 65 years old than in patients ≤ 65 years old (Fig. 8A), and female patients had a higher risk score than males (Fig. 8B). However, no significant difference was observed in the risk scores between spinal and bulbar sites of onset (Fig. 8C). We further compared the prognosis associated with different clinicopathological features in the two risk groups. We found that the patients ≤ 65 years old at onset and a spinal site of onset had a better prognosis than patients > 65 years old at onset with a bulbar site of onset in both the low- and high-risk groups, whereas no statistical difference was detected for sex differences (Fig. 8D). To further explore the relationships of risk scores with clinical characteristics, when the patients were regrouped according to age at onset, sex, and site of onset, the risk score still exhibited potent predictive performance, and those patients with lower risk scores had a better prognosis (Fig. 8E). As a result, this risk score was demonstrated to be a highly independent predictor of prognosis in patients with ALS.

Fig. 8.

Association of risk score and clinical characteristics in the entire cohort. Higher risk scores in patients with older age at onset than in younger patients (A) and in the female than in males (B). No significant difference was identified in patients with a different site of onset (C). D Differences in survival risk by age at onset, sex, and site at onset t are not affected by high- or low-risk group patients. E The survival curve of patients regrouped according to age at onset, sex, and site at the onset

Biological pathways and immune infiltration between the high- and low-risk groups

To elucidate the potential regulatory mechanisms leading to prognostic differences between the high- and low-risk groups in the entire cohort, we applied GSEA algorithms to enrich the significant KEGG pathways between the two groups. The results revealed that the complement and coagulation cascades, cytokine–cytokine receptor interactions, RA, and the TNF signaling pathway were predominantly enriched in the high-risk group (Fig. 9A), whereas metabolism-related pathways, such as arginine and proline metabolism and carbon metabolism, and nucleotide acid repairmen, such as mismatch repair and nucleotide excision repair, were significantly enriched in the low-risk group (Fig. 9B).

Fig. 9.

The biological pathways and immune microenvironment between high- and low-risk groups. A and B The up- and downregulated KEGG pathways enriched in the high-risk group compared to the low-risk group with GSEA analysis. C The comparison of 28 immune cells in two groups with ssGSEA analysis. D The correlations between nine model genes with immune cells. E The comparison of 17 immune responses in two groups with ssGSEA analysis. Significance level was denoted by *p value < .05, **p value < .01, ***p value < .001

To investigate the relationship between the different risk groups and immune features in the whole blood of patients with ALS, we conducted ssGSEA to uncover the different immune cell infiltration and immune responses between risk groups. The abundance of central memory CD8 + T cells, type 2 T helper cells, and neutrophils was significantly higher in the blood of high-risk patients with ALS, whereas Tregs, activated B cells, and CD56 (dim) natural killer cells were significantly lower (Fig. 9C). To investigate the relationships between the nine model genes and immune cells, the correlations between them were calculated using Pearson’s correlation method and were visualized with a heat map (Fig. 9D). The immune responses revealed that chemokines, cytokine receptors, cytokines, interferon receptors, interleukin receptors, and TGF-β family members were more active in the high-risk group than in the low-risk group (Fig. 9E).

Development of a nomogram based on the risk score and clinical features

Considering the inconvenient clinical utility of the risk score in predicting survival in patients with ALS, a nomogram incorporating the risk score and clinicopathological parameters was established to predict the 1-, 3-, and 5-year survival in the overall cohort (Fig. 10A). Calibration curves assessing the performance of the nomogram demonstrated a satisfactory match between the actual and nomogram predicted 1-, 3-, and 5-year survival probabilities (Fig. 10B). The C-index of the nomogram was superior to the risk score and clinicopathological parameters, suggesting that the nomogram presented better discrimination (Fig. 10C). In particular, the AUC values of the ROC curve for the 1-, 3- or 5-year survival of the nomogram were 0.677, 0.727, and 0.723, respectively (Fig. 10D–F). Moreover, we also discovered that the combined nomogram showed some net benefit for predicting survival probabilities in the DCA curves, demonstrating that the nomogram had a high potential for clinical utility (Fig. 10G–I).

Fig. 10.

Construction of a predictive nomogram. A The nomogram for predicting the overall survival of patients with ALS at 1, 3, and 5 years. B Calibration of the nomogram at 1, 3, and 5 years. C C-index of the nomogram, the risk score, and clinicopathological parameters. D–F ROC curve for 1-, 3-, and 5-year survival of the nomogram, the risk score, and clinicopathological parameters, respectively. G–I DCA for 1-, 3-, and 5-year survival of the nomogram, the risk score, and clinicopathological parameters, respectively

Construction of a ceRNA network and gene expression responses to IL-2

A ceRNA network describes a regulatory network that integrates mutually interacting mRNAs and non-coding transcripts (e.g., lncRNAs, miRNAs). The GSE148097 dataset was used to identify the DEmiRNA between patients with ALS and healthy controls. A total of 26 DEmiRNAs (including 11 upregulated and 15 downregulated miRNAs) were screened (Fig. 11A). After intersecting the DEmiRNA and predicted miRNA of the nine model genes using the “multiMiR” R package, we identified 17 potential DEmiRNAs (Table S4). We also further predicted the lncRNA of the 17 potential DEmiRNAs using the “multiMiR” R package. We further detected the DElncRNA between patients with ALS and controls using the GSE106443 dataset. The volcano plots showed that a total of 194 lncRNAs were upregulated and 214 lncRNAs were downregulated (Fig. 11B). After overlapping the predicted lncRNAs and DElncRNAs, five potential lncRNAs were finally screened (Table S5). The ceRNA network of lncRNA-miRNA-mRNA was visualized with a Sankey diagram (Fig. 11C).

Fig. 11.

Construction of a ceRNA network (A–C). A A volcano plot showing the differentially expressed miRNA in the GSE148097 dataset. B A volcano plot showing the differentially expressed lncRNA in the GSE106443 dataset. C A Sankey plot indicating the ceRNA regulatory network for hub genes. Gene expression of nine model genes responses to IL-2 (D–E). D The expression levels of nine model genes between ALS and controls. E The expression levels of nine model genes in ALS patients after IL-2 treatment

Previous studies have demonstrated the potential effectiveness of IL-2 treatment for patients with ALS. In our study, we also evaluated the gene expression responses to IL-2 of the nine model genes in the GSE163560 dataset. The expression levels of the nine genes were compared between patients with ALS and controls (Fig. 11D). With the administration of low-dose IL-2, we found that the expression levels of CD248, ENG, CLEC5A, PRDX5, and XKR8 were significantly reduced (Fig. 11E). Given the upregulation of three genes (CLEC5A, PRDX5, and XKR8) in ALS, these genes may serve as the therapeutic targets of IL-2 treatment.

Expression of model genes in SOD1 mutant mice by using scRNA-seq analysis

To explore the expression patterns of the nine models genes in the brain, we performed a scRNA-seq analysis in brainstem of SOD1 mutant mice. After clustering and visualization, the samples can be grouped into seven cell subpopulations in UMAP plot (Fig. 12A). The heatmap displayed the expression patterns of the top five marker genes of each cell subpopulations (Fig. 12B). We then conducted feature plots to visualize the expression and distribution of our model genes in these cell clusters (Fig. 12C). We can observe that Aim2, Adora3, and Clec5a were mainly expressed in Microglia cluster; Cd248 and Eng were mainly expressed in endothelial cells; Xkr8 and Kif17 were weakly expressed in Astrocytes cluster; and Prdx5 was strongly expressed in most cell subpopulations. Ptnl2 gene was not detected in this dataset. We also compared the expression difference of model genes between mutant SOD1 and wild-type mice in selected cell types (Fig. 12D). However, only the expression of Aim2 and Eng was significantly different between two groups, implying the different microenvironments of the peripheral blood and central nervous system.

Fig. 12.

Expression of models genes in mutant SOD1 mice by using scRNA-seq analysis. A UMAP visualization of seven cell clusters. Each dot represents a single cell, and each color represents a cell type. B The top 5 marker genes of each cell cluster are displayed with a heatmap. C Feature plot visualize the expression and distributions of models genes. D The expression comparison of model genes between mutant SOD1 and wild-type mice in selected cell types

Methylation analysis of model genes

To identify the model genes whose methylation may be modified in the ALS disease, we explore the methylation positions and levels of nine model genes in GSE89472 dataset. In our analysis, we found that the methylation levels of most of positions between patients with ALS and controls were not significantly different (Fig. S7A and Table S6). Only four differentially methylated positions (DMPs) (cg07340708, cg10254167, cg13138832, and cg20105233) of model genes were identified. Among them, the methylation levels of BTNL2 (cg07340708, cg10254167, andcg13138832) (Fig. S7B–D) and PRDX5 (cg20105233) (Fig. S7E) were lower in ALS than control. Combined with the transcriptional levels of BTNL2 and PRDX5, we can speculate that the higher expression levels of BTNL2 and PRDX5 were due to the lower methylated positions of the two genes.

Discussion

ALS is among the most rapidly and devastating progressive forms of neurodegenerative disease with a median survival of 3 to 5 years from symptom onset. Several studies have suggested autoimmunity in the pathogenesis of ALS or as a mediator of heterogeneity [31, 32]. The genetic correlations between ARDs and ALS have been identified in the past several years. However, the interaction between ARGs and ALS from the perspective of transcriptomics, particularly in the prognosis of ALS, remains unknown. This study focused on the roles of ARGs from the peripheral blood transcriptome profile in the progression of ALS as prognostic factors to predict survival from the viewpoint of PPPM.

In this study, we collected the transcriptional dataset of peripheral blood from patients with ALS from GEO databases. Two datasets with the available prognostic information and clinical features were first incorporated after removing the batch effects. The DEGs between patients with ALS and controls were identified using the Wilcoxon test. The function annotation analysis revealed that immune-related pathways, such as inflammatory responses and complement responses, were enhanced in the whole blood of patients with ALS. After intersection with the ARGs, a total of 1612 DEARGs overlapped, out of which 85 prognostic DEARGs were screened using univariate Cox regression analysis. Then, we identified two different autoimmune clusters based on the expression patterns of prognostic DEARGs and compared the differences in prognosis and immune characteristics between the clusters. We found that Cluster 2 had better outcomes and presented a lower peripheral immune status. We proposed a scoring system called the “autoimmune index” to comprehensively quantify the autoimmune pattern of patients with ALS, and the group with a higher index exhibited a good prognosis, and the index was negatively correlated with the immune characteristics. We further constructed nine-ARG prognostic signature from the peripheral blood of patients with ALS using the training cohort that showed excellent prognostic prediction efficiency and was well-validated in the test and entire cohorts. Subsequent subgroup analysis and construction of a prognostic nomogram, ROC curve analysis, calibration curve analysis, and DCA were performed, which is in line with EPMA. A ceRNA for regulation of model genes was constructed, and gene expression responses to IL-2 were also assessed. Finally, the expressions of model genes in mutant SOD1 mice were determined in single-cell levels, and the DMPs of model genes were identified using methylation analysis.

In our enrichment analysis, we discovered that immune-related pathways were more active in patients with ALS, in Cluster 1 with a poor prognosis, and in the high-risk group, implying that inflammation in peripheral blood of patients with ALS promotes the initiation and progression of ALS. Specifically, complement responses, allograft rejection, interferon responses, TNF-α signaling via NF-κB, and IL-6/JAK/STAT3 signaling were enriched among these biological functions. Lu et al. discovered the altered production of the systemic inflammatory response in ALS and found that the levels of creatine kinase, ferritin, TNF-α, IL-1β, IL-2, IL-8, IL-12p70, IL-4, IL-5, IL-10, and IL-13 were significantly higher in plasma samples from patients with ALS patients than in controls [33]. High ferritin, creatine kinase, and IL-2 levels were predictors of poor survival, suggesting a role of the peripheral immune system in ALS progression [33, 34]. The levels of circulating cytokines are also abnormal in ALS animal models. Compared with controls, the levels of 6Ckine, ALK-1, CD30 L, eotaxin-1, galectin-1, GITR, IL-2, IL-6, IL-10, IL-13, IL-17B R, MIP-1α, MIP-3β, RANKL, TROY, and VEGF-D were found to be dysregulated in transgenic SOD1G93A mice at an asymptomatic stage [35]. In addition, the increased levels of several cytokines, such as eotaxin-1 and galectin-1, were associated with a shorter survival time. These results were consistent with the results of our immune response evaluation that revealed that chemokines, cytokine receptors, cytokines, interferon receptors, interleukin receptors, and TGF-β family members were more active in the high-risk group with poor outcomes. However, the alterations of the immune response results between the two autoimmune subtypes were surprisingly partially contrasted. Chemokines, cytokines, cytokine receptors, and TGF-β family members were more active in Cluster 2, which had a better prognosis. The interpretation of studies profiling immune markers in the peripheral blood can be challenging due to recent environmental exposures, a complex mixture of disease states, and genetic backgrounds. In addition, the combined effects of these inflammatory responses may shape the immune environment and promote the progression of ALS. Compared with Cluster 1, the activity of TNF family member receptors was reduced in Cluster 2. A previous study demonstrated that plasma TNF-α and TNF-related apoptosis-inducing ligand with ALS-causing SOD1 mutations (mSOD1 ALS) was correlated negatively with survival. Despite these complexities, a comprehensive understanding of the systemic immune response in ALS is important not only for unraveling disease mechanisms but also for predicting the development of ALS.

In addition to these inflammatory responses involved in the genesis and prognosis of ALS, immune cell populations also play a crucial role in the pathogenesis of ALS. The total leukocyte counts in patients with ALS were significantly higher than those in controls [36]. In particular, the levels of neutrophils, CD16 monocytes, and natural killer cells were increased in patients with ALS, with CD16 monocyte levels correlating with disease severity [36]. In our study, we also explored the potential correlation of peripheral blood immune cells with prognosis in ALS. We found that natural killer cells were significantly upregulated and that neutrophils were downregulated in both Cluster 2 and the low-risk group, which both exhibited good outcomes. Natural killer cells and neutrophils were correlated with disease progression in a sex- and age-dependent manner [37, 38]. Cui et al. discovered that natural killer cells were negatively associated with the risk of death from ALS in a longitudinal analysis of blood cell counts in an ALS cohort [39]. Neutrophil numbers also had a significant correlation with disease progression [36]. We also observed that activated CD8 + T cells, activated CD4 + T cells, type 1 T helper cells, activated B cells, and BCR and TCR signaling pathways were enhanced in Cluster 1, which was a poorer prognostic subtype. The correlations between activated CD8 + T cells and activated CD4 + T cells as risk factors for disease progression were demonstrated by previous studies [36, 39]. The increase in activated B cell and BCR signaling pathways was related to the increased serum immunoglobulins of patients with ALS and autoimmune status [40]. A recent study showed a positive polygenic association between serum immunoglobulins and ALS [41]. Elevated serum autoantibodies, such as serum autoantibodies against HMGB1 [42] and ACTB [43], were found in more severe disease status. Tregs are a key cell type responsible for suppressing the immune response and maintaining immune tolerance. Given the inhibitory roles of Tregs, increased or activated populations of these cells may have therapeutic benefits in patients with ALS [44]. Several clinical trials have demonstrated the safety and potential benefit of Tregs in influencing disease progression rates [26, 45]. A meta-analysis of preclinical studies revealed that Tregs significantly prolonged survival in mouse models of ALS [44]. In our risk stratification analysis, the abundance of Tregs in the low-risk group was higher than that in the high-risk group, which was consistent with the previous findings. However, in our subtype analysis, Cluster 2 with a good prognosis had less Treg infiltration. Considering the simultaneously enhancement of myeloid-derived suppressor cells in Cluster 2, we assumed that immune suppression of myeloid-derived suppressor cells may prominently contribute to the anti-inflammatory response in Cluster 2. Low-dose IL-2 can enhance Treg function in auto-inflammatory conditions. The immunologically effective and neuroprotective effect of IL-2 was demonstrated in patients with ALS and ALS animal models [26, 46]. Blood-based biomarkers can also be used to screen drug responses in human. To explore the gene responses to IL-2 from the perspective of PPPM, we found that expression levels of five (CD248, ENG, CLEC5A, PRDX5, and XKR8) out of nine model genes were significantly downregulated in the blood of patients with ALS following low-dose IL-2 administration. Three genes (CLEC5A, PRDX5, and XKR8) exhibited higher levels in patients with ALS than in controls; therefore, these genes may serve as the therapeutic targets of IL-2 treatment.

Recently, the construction of risk model in the framework of PPPM has attracted extensive attention, which emphasizes early assessment of risk to predict the prognosis of sever disease and prevent the progression via individualized treatment. This concept is also applicable to the medical management of ALS patients. In our study, we constructed a nine-ARG (KIF17, CD248, ENG, BTNL2, CLEC5A, ADORA3, PRDX5, AIM2, and XKR8) signature. The independence of the signature was well demonstrated through univariate Cox regression analysis, multivariate Cox regression analysis, and subgroup analysis. CLEC5A is a spleen tyrosine kinase (Syk)-coupled C-type lectin that is highly expressed on monocytes, macrophages, neutrophils, and dendritic cells. Previous studies have demonstrated the critical role of CLEC5A in acute viral infections [47, 48]. CLEC5A is involved in different proinflammatory responses, such as neutrophil extracellular trap formation and the production of reactive oxygen species and proinflammatory cytokines [47]. In the present study, we found increased expression levels of CLEC5A in patients with ALS, which were significantly suppressed by IL-2 treatment, suggesting the potential role of CLEC5A as a therapeutic target. The peroxidase PRDX5, also known as thioredoxin peroxidase, is an antioxidant enzyme that has been extensively studied for its antioxidant properties and protective role in neurological and cardiovascular conditions [49]. However, the transcript expression of PRDX5 was significantly higher in the whole blood of patients with ALS in our analysis, implying a different environment between peripheral blood and the central nervous system. Xk-related protein 8 (XKR8) promotes the phagocytosis of dying cells by altering the phospholipid distribution in the plasma membrane during apoptosis [50]. As a newly discovered risk gene, higher expression of XKR8 was identified in the whole blood of patients with ALS than in controls. In addition, the higher expression of XKR8 was positively correlated with the death rate in the survival analysis. The CD248 receptor can help to promote quiescence of naive CD8( +) human T cells [51]. In our analysis, we observed that the expression of CD248 was downregulated in patients with ALS and that higher levels of CD248 were associated with a better prognosis, suggesting the protective roles of CD248 in patients with ALS. Endoglin (CD 105, TGF-β receptor III, encoded by ENG) is a homodimeric transmembrane glycoprotein that is essential for TGF-β signaling. Endoglin is responsible for vascular dysfunction-associated diseases such as hereditary hemorrhagic telangiectasia-1, pre-eclampsia, and intrauterine growth restriction [52]. The levels of serum endoglin in patients with ALS were decreased [53], consistent with our transcriptome analysis. In our survival analysis, we observed that lower expression of ENG was associated with a worse prognosis.

In addition, we found that the age at onset was also an independent factor affecting the prognosis of patients with ALS. The histogram of the frequency distribution revealed that Cluster 1 and the high-risk group with a poorer prognosis had a larger proportion of patients with an older age at onset. Older age at onset was associated with a lower autoimmune index, which was a protective factor for the prognosis of patients with ALS. These results were consistent with previous findings of a worse prognosis in older patients with ALS [54, 55].

To further enhance the accuracy of prognostic prediction, we constructed a nomogram by integrating the risk score and clinical characteristics of patients with ALS including sex, age at onset, and site of onset. The results indicated that risk score and age at onset are two independent risk factors in the prognosis of ALS. The constructed nomogram was assessed by evaluating the C-index, ROC curve, calibration curve, and DCA, which further demonstrated the prediction efficiency of the prognostic model for ALS by integrating genetic and individual information in the context of PPPM [56].

Considering that less than 2% of transcripts code for proteins, the altered expression profiles of non-coding RNAs (ncRNAs) including miRNAs and lncRNAs has attracted more and more attention in ALS [57, 58]. We constructed a ceRNA regulatory network for the model genes, including five lncRNAs, four miRNAs, and five mRNAs, which may help us to understand the complex pathophysiological changes of ALS and provide better PPPM. Among the five mRNAs, three genes (CLEC5A, PRDX5, and XKR8) that respond to IL-2 treatment were also identified. miRNA-484 was conspicuous because it targets four hub model genes, including CLEC55A and XKR8. However, the researches about miRNA-484 are very deficient. Early study discovered the diagnosis ability and staging of hepatic cirrhosis and fibrosis and breast cancer [59, 60]. In our analysis, miRNA-484 was differentially expressed in ALS compared to controls and targets the hub ARGs, which may provide a novel direction for future study.

There are several strengths in this study. First, we identified two molecular clusters of patients with ALS based on the prognostic DEARG expression matrix, and the two clusters showed different prognoses and immune statuses. Second, we calculated the autoimmune index for each individual and found that patients with a higher index presented a better prognosis. Third, comprehensive and in-depth analyses including univariate/multivariate Cox regression and LASSO regression were performed to construct a prognostic model that was further validated by the test cohort and overall cohort. Finally, a nomogram was further established for quantitative calculation, which is conducive to clinical promotion and application. Nevertheless, several limitations still exist in our study. For example, external validation with a prospective dataset is necessary to validate our risk model, all the results were acquired by bioinformatics analysis, and the lack of experimental validation limited the evidence level of this study. These findings need further been confirmed in vitro and in vivo experiments, to clarify the pathological mechanism of ARGs underlying ALS onset, which may help ARGs to apply in the field of PPPM for ALS.

Conclusions and expert recommendations

In this study, we identified two molecular subgroups with different prognoses and immune statuses. We constructed a nine-ARG prognostic signature that is closely related to the immune environment and can better predict survival in patients with ALS after clinical classifications have been made. Then, a nomogram was built by combining the risk score and clinical parameters to predict the overall prognostic value for patients with ALS. The constructed nomogram was further evaluated by using C-index, ROC curve analysis, calibration curve analysis, and DCA, indicating the potential clinical value.

Predictive medical approach: In present work, a total of 1,612 DEARGs were identified between ALS patients and controls, and 85 prognostic DEARGs were further screened using univariate Cox regression analysis. A nine ARGs signature selected by LASSO regression was constructed using multivariate Cox regression analysis. The risk score calculated by the prognostic model has been demonstrated to be an important predictor for individual outcome, which help to identify high-risk populations with ALS. A nomogram by combining risk score with other clinicopathological features has great clinical application value for risk stratification and prognostic assessment in patients with ALS. Patients with ALS may benefit from these results in developing PPPM.

Targeted prevention

It may be a cost-effective way for the targeted prevention of ALS if we can early identify ALS patients with bad clinical outcomes. In the present work, we quantified the autoimmune index and risk score per individual and found the patients with low-autoimmune index or high-risk score had a worse prognosis. Thus, our autoimmune index and risk score based on ARGs could stratify the patients with ALS into different characteristic subgroups and help us identify high-risk groups for targeted prevention. By integrating the patient’s expression profile (recommending measured in hospital periods) and clinical parameters, the constructed nomogram of the present work may be served as a simple, economical, and widely practicable tool to assess the individual risk of patients with ALS in clinical practice.

Personalized medicine

Although growing evidence suggests close correlations between autoimmunity and ALS, no studies have reported on ARGs from the perspective of the prognostic assessment of ALS. We recommend focusing the study on ARGs in patients with ALS. The nine ARGs of our result can help clinicians evaluate the clinical outcomes of patients with ALS and provide targeted therapeutic strategies through intervention into their functions. Moreover, the results of the constructed ceRNA regulatory network, gene response to IL-2, scRNA-seq analysis, and methylation analysis in a PPPM system in ALS would provide a unique benefit to a clinically intelligent and novel approach in the treatment of patients with ALS. Besides, we also found distinct immune infiltration between high- and low-risk group, which might improve the application of personalized management and the development of immunochemotherapy in ALS treatment from the perspectives of PPPM.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

Conceptualization, SL and QZ; methodology, SL; software, SL; validation, QZ; formal analysis, SL; investigation, SL; writing—original draft preparation, SL; writing—review and editing, LW; visualization, LW; and supervision, JL.

Funding

This research was supported by the National Natural Science Foundation of China (No. 81902551) and the Hunan Provincial Natural Science Foundation of China (No. 2021JJ31072).

Data availability

The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.

Code availability

Code in the present work can be obtained for reasonable requirements.

Declarations

Ethics approval and consent to participate

Ethics approval and consent to participate GEO belong to public databases. The patients involved in the database have obtained ethical approval. Users can download relevant data for free for research and publish relevant articles.

Consent for publication

All authors have read and agreed to the published version of the manuscript.

Competing interests

The authors declare no competing interests.