Exploration of biomarkers for systemic lupus erythematosus by machine-learning analysis

doi:10.21203/rs.3.rs-3147806/v1

Download PDF

Research Article

Exploration of biomarkers for systemic lupus erythematosus by machine-learning analysis

https://doi.org/10.21203/rs.3.rs-3147806/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 10 Nov, 2023

Read the published version in BMC Immunology →

You are reading this latest preprint version

Background

In recent years, research on the pathogenesis of systemic lupus erythematosus (SLE) has made great progress. However, the prognosis of the disease remains poor, and high sensitivity and accurate biomarkers are particularly important for the early diagnosis of SLE.

Methods

SLE patient information was acquired from three Gene Expression Omnibus (GEO) databases and used for differential gene expression analysis, such as weighted gene coexpression network (WGCNA) and functional enrichment analysis. Subsequently, three algorithms, random forest (RF), support vector machine-recursive feature elimination (SVM-REF) and least absolute shrinkage and selection operation (LASSO), were used to analyze the above key genes. Furthermore, the expression levels of the final core genes in peripheral blood from SLE patients were confirmed by real-time polymerase chain reaction (PCR) assay.

Results

Five core genes (ABCB1, CD247, DSC1, KIR2DL3 and MX2) were found in this study. Moreover, the nomogram model showed that the five optimal key genes had good reliability and validity, which were further confirmed by clinical samples from SLE patients. The receiver operating characteristic (ROC) curves of the five genes also revealed that they had critical roles in the pathogenesis of SLE.

Conclusion

Overall, five key genes were obtained and validated through machine-learning analysis of the databases, which might offer a new perspective for the molecular mechanism and potential therapeutic targets for SLE.

Systemic lupus erythematosus

Machine learning

Biomarker

Genes

Databases

Systemic lupus erythematosus (SLE) is a highly heterogeneous autoimmune disease with systemic multisystem involvement[1]. The pathogenesis is complex and it occurs mostly in young women. The abnormal immune system mediated by B cells and T cells is a key link in the occurrence and development of SLE[2–4]. Clinical manifestations of SLE are widespread, including kidney, skin, blood and joint symptoms[5]. Currently, the immunological diagnostic criteria of SLE mainly include anti-dsDNA antibodies, anti-nuclear antibodies and anti-SM antibodies[6]. However, their specificity and sensitivity are low, and the diagnosis of SLE is still not satisfactory. Hence, exploring the characteristic genes related to the pathogenesis and development of SLE is necessary.

With the update of bioinformatics, various prediction methods for SLE continue to emerge[7, 8]. However, most predictive methods have low accuracy and efficiency. At present, machine learning algorithms are widely used to screen markers with more diagnostic value and build more accurate clinical prediction models.

Therefore, in this study, we processed DEGs and WGCNA analyses to identify candidate genes associated with SLE. Then, a variety of machine learning algorithms including LASSO, RF and SVM-REF, were combined to obtain the five optimal key genes. Furthermore, ROC was used to evaluate its predictive performance. Subsequently, GO, KEGG, DO, and GSEA analyses were used to investigate the mechanism of their contribution to SLE. In addition, ssGSEA, an immune-related algorithm, was applied to assess the infiltration levels of different immune cell types. To verify the bioinformatic results, real-time quantitative PCR (qRT-PCR) was used to analyze the relative expression of the five optimal key genes in SLE and healthy control samples. Overall, we found five genes with potentially strong diagnostic effects in SLE patients, suggesting that they might be the new targets for studying SLE.

Download and processing of data

The flowchart is shown in Fig. 1. The raw gene expression data were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo/)[9]. Three datasets, GSE121239 (GPL13158, normal: 20, SLE: 292), GSE81622 (GPL10558, normal: 25, SLE: 30) and GSE11907 (GPL96, normal: 12, SLE: 103), were incorporated in the subsequent bioinformatics analysis. Two datasets, GSE65391 (GPL10558, normal: 72, SLE: 924) and GSE49454 (GPL10558, normal: 20, SLE: 157), were used as independent validation sets. The information on these datasets is presented in Supplementary Table 1. The microarray data were converted into log2 values, and the combat algorithm implemented in the R package "sva" was used for integration to remove the batch effect and form a combined dataset[10].

Differential expression analysis

DEGs were identified by differential expression analysis between SLE and control samples through the R package “limma” (criteria: | logFC| > 0.75, P value < 0.05). Volcano plot and heatmap showed the significantly upregulated and downregulated genes.

Weighted gene coexpression network analysis

The potential functional modules of SLE samples were identified through the R package “WGCNA”[10, 11]. According to the weighted correlation adjacency matrices and cluster analyses, genes with similar expression patterns were allocated to coexpression modules. The topological overlap matrix (TOM) was derived from the adjacency matrix, and genes were assigned into different modules on the basis of the difference between them in the TOM. Notably, the cut height was set to 0.25, the minimal module size was set to 50, and the soft-thresholding power was set as 25 (scale-free R² = 0.9). Furthermore, both gene importance (GS) and module membership (MM) need to be analyzed. The Spearman correlation coefficients and the corresponding P values were also analyzed. Finally, the corresponding genes extracted from the hub module were obtained for further analysis.

GSEA and correlation analysis

GSEA was used to identify the biological significance of optimal key genes[12, 13]. To obtain a normalized enrichment score, each analysis required the gene set permutations with 1000 times. And FDR < 0.05 was considered as a significant enrichment. Additionally, Pearson correlation analysis was utilized to analyze the correlations between optimal key gene expression levels.

Functional enrichment analysis

Through the above analysis, the overlapping genes between DEGs and module genes were further obtained. A Venn diagram was constructed to show the overlapping genes. The “clusterProfiler” and “DOSE” R packages were used to perform Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Disease Ontology (DO) enrichment analyses to explore the function and pathways of these genes[14]. Additionally, a protein-protein interaction (PPI) network was created to study the interaction of overlapping candidate genes through “STRING” (https://string-db.org/). The coexpression network was also obtained with the R package “igraph” to study the correlation among these candidate genes.

Further screening of optimal key genes

A total of three machine learning algorithms including LASSO, SVM-REF and RF, were applied to explore the disease status. The LASSO algorithm was used to select variables and adjust complexity[15]. The SVM-RFE algorithm was used to analyze the most appropriate key genes[16, 17]. For the RF algorithm, the principle is to find the most reliable results from a large number of underlying tree models[18, 19]. Since the importance of genes greater than 2 in the RF algorithm is a commonly used screening criterion, the top 10 key genes were selected as new gene markers to predict the prognosis of SLE. Finally, the common genes obtained from the intersection of these several algorithms were the optimal key genes.

The expression and predictive value of optimal key genes

The Wilcoxon rank-sum test was used to analyze the expression levels of optimal key genes. The ROC curve was used to verify the predictive value of the optimal key genes[20].

Analysis of hallmark gene sets and immune cell infiltration

The CIBERSORT algorithm is a tool that uses a deconvolution algorithm to predict cell proportions, which helps to calculate the cell composition of complex immune tissues based on standardized gene expression data and quantify the abundance of specific cell types[21]. In this study, the CIBERSORT algorithm was used to analyze the infiltration of 22 immune cells in SLE and normal samples. The difference in immune cell proportions was analyzed by the Wilcoxon rank-sum test, and P < 0.05 was considered statistically significant. In addition, the ssGSEA algorithm was used to quantify the relative levels of 50 marker gene sets[22, 23]. In addition, Spearman’s correlation between 50 hallmark gene sets and the optimal key genes was also obtained.

Isolation of human PBMCs

A total of 13 samples were collected from the First Affiliated Hospital of Soochow University, including 6 SLE samples and 7 healthy controls. This study was approved by the Ethics Committee of the First Affiliated Hospital of Soochow University, and informed consent forms were signed by the patients. Human peripheral blood mononuclear cells (PBMCs) were isolated by FicollPaque density gradient centrifugation with EDTA anticoagulant blood.

Real-time quantitative PCR

Total RNA from the PBMCs of SLE patients and healthy controls was extracted with TRIzol reagent, and cDNA was synthesized by reverse transcription with the PrimeScript™ RT Reagent Kit (TaKaRa, China). β-actin was used as an internal reference. Real-time quantitative polymerase chain reaction (qPCR) was performed with the SYBR Green PCR Kit (TaKaRa, China) based on the manufacturer’s protocol. Furthermore, the relative expression was calculated by the 2^−△△CT method. All reactions were repeated in triplicate. The primers are displayed in Supplementary Table 2.

Statistical analysis

R software (version 4.1.1) and GraphPad Prism (version 7.0) were used for data processing, statistical analysis and mapping. The Wilcoxon rank-sum test or Student's t test identified the differences between two groups. The correlations were analyzed with Pearson's or Spearman's correlation tests. P < 0.05 was considered statistically significant.

DEGs between SLE and control samples

From the GEO database, three microarray datasets, including the GSE121239, GSE81622 and GSE11907 datasets, were combined to obtain a total of 57 control and 425 SLE samples. After the batch effect was removed (Fig. 2A), 242 DEGs (Supplementary Table 3) were identified, including 87 downregulated genes and 155 upregulated genes (Fig. 2B). As shown in Fig. 2C, some genes, such as KIR2DL3, EIF1AY, CD247, ABCB1, DSC1, RPS4Y1, GJC1, ZNF674, SHCBP1, SPCS2, and RPLP2, were significantly upregulated, while some genes were significantly downregulated, such as MX2, SIGLEC1, IFIT3, RSAD2, IFI27, and PI3. Next, GSEA of KEGG was performed to select the enriched signaling pathways (Supplementary Table 4). The Ridgeline map displayed changes in gene metabolism in SLE, such as pyruvate metabolism, DNA replication and RNA degradation (Fig. 2D). Notably, mucin type O-glycan biosynthesis, microRNAs in cancer, and tight junctions were significantly enriched in the SLE group (Fig. 2E). However, DNA replication, aminoacyl-tRNA biosynthesis and mismatch repair were significantly enriched in the control group (Fig. 2F).

Application of WGCNA

The coexpression network was established by WGCNA. A total of 11256 genes were selected for clustering, and obviously abnormal samples were excluded by setting thresholds (Fig. 3A). Then, when R² = 0.9 and the average connectivity was high, the soft power threshold was set to 25 (Fig. 3B). After the associated modules were integrated, five modules were obtained. The initialized and integrated modules were finally shown under the clustering tree (Fig. 3C). The correlation between the modules revealed no significant correlation between the modules (Fig. 3D). In addition, the transcriptional correlation analysis revealed that there was also no relationship between them (Fig. 3E). The frontal correlation was applied to examine the correlation between clinical features and ME values, and the results showed that yellow and blue modules were significantly correlated with SLE (R = 0.46, P < 3e-67; R = 0.23, P < 2.7e-38) (Fig. 3F-H). In the yellow and turquoise modules, a total of 4348 candidate genes were obtained in the subsequent analysis (Supplementary Table 5).

Functional enrichment analysis of overlapping genes

A total of 214 candidate key genes were selected from the DEGs, yellow and turquoise modules (Supplementary Table 6 and Fig. 4A). To explore the potential biological function and enrichment pathways, GO, KEGG, and DO analyses were performed on these candidate key genes. In the GO analysis, the candidate key genes were significantly enriched in the mRNA catabolic process, RNA catabolic process and nuclear-transcribed mRNA catabolic process for the biological process (BP) category. For the cellular component (CC) category, the genes were significantly enriched in cytosolic ribosome, secretory granule membrane and ribosome. The molecular function (MF) category was significantly enriched in poly-pyrimidine tract binding, endopeptidase inhibitor activity and enzyme inhibitor activity (Fig. 4B). In addition, the KEGG enrichment analysis showed that these overlapping genes were particularly related to ribosomes, ubiquitin-mediated proteolysis, leishmaniasis, malaria and cytokine-cytokine receptor interactions (Fig. 4C-E). The DO analysis showed that they were mainly enriched in hepatitis, lupus erythematosus and malaria (Fig. 4F).

The results of these functional enrichment analyses indicated that SLE patients may share common immune processes with other diseases. To further reveal protein-protein interactions in SLE, PPI networks of candidate key genes were analyzed (Fig. 4G, H).

Further identification of optimal key genes through machine-learning algorithms

To further explore the key genes, three machine-learning algorithms were performed. According to the 214 candidate key genes obtained above, 34 key genes were further selected as diagnostic markers of SLE by the LASSO algorithm (Fig. 5A and Supplementary Table 7). In addition, 58 key genes were screened after fivefold cross-validation of 214 genes using the SVM-REF algorithm (Fig. 5B and Supplementary Table 8). For the RF algorithm, the top 10 key genes with importance > 2 were identified, consisting of MX2, SQRDL, RFTN1, KIR2DL3, ABCB1, RPS14, CD247, CHST11, CD6 and DSC1 (Fig. 5C, D). Finally, all the key genes obtained by these machine-learning algorithms overlapped again, and the 5 optimal key genes were screened out, consisting of ABCB1, CD247, DSC1, KIR2DL3 and MX2, which were regarded as potential targets of SLE (Fig. 5E).

Evaluation of the expression and diagnostic value of optimal key genes

The expression of five optimal key genes was further analyzed in 425 SLE samples and 57 normal samples. In Fig. 6A-E, the expression of ABCB1, CD247, DSC1 and KIR2DL3 was significantly downregulated in SLE samples, while the expression of MX2 was significantly upregulated, indicating that they might have critical roles in the development of SLE (all P < 0.01). In addition, ROC curve analysis was performed to assess the diagnostic value of the optimal key genes (Fig. 6F). The AUC values of the ROC curves were 0.848 for ABCB1 (Fig. 6G), 0.859 for CD247 (Fig. 6H), 0.843 for DSC1 (Fig. 6I), 0.807 for KIR2DL3 (Fig. 6J), and 0.907 for MX2 (Fig. 6K), indicating that the five key genes have high diagnostic value in evaluating the progression of SLE.

To obtain more reliable results, the optimal key genes were further verified in an external validation dataset consisting of 1081 SLE samples and 92 control samples. Before analysis, the GSE65391 and GSE49454 datasets were normalized (Supplementary Fig. 1). Fortunately, the expression of 5 optimal key genes in SLE samples was consistent with the above results (Fig. 7A-E, all P < 0.05). The diagnostic value of these genes was assessed by ROC curve analysis (Fig. 7F). The AUC values of the external validation datasets were also high: ABCB1 (AUC: 0.825), CD247 (AUC: 0.862), DSC1 (AUC: 0.780), KIR2DL3 (AUC: 0.694), MX2 (AUC: 0.898) (Fig. 7G–K). Therefore, these results again prove that all the optimal key genes are related to SLE.

Identification of the biological function of five key genes

To clarify the biological functions of the five optimal key genes, GSEA was employed. According to the median expression levels of these genes, SLE samples were segmented into two groups. In addition, immune-related pathways such as allograft rejection, mismatch repair, and protein export were significantly enriched in the high ABCB1 subgroup (Fig. 8A), but pathways such as osteoclast differentiation and type II diabetes mellitus were enriched in the low ABCB1 subgroup (Supplementary Fig. 2A). DNA replication, fatty acid elongation, mismatch repair, and protein export were significantly enriched in the high CD247 subgroup (Fig. 8B), but arachidonic acid metabolism and type II diabetes mellitus were significantly enriched in the low CD247 subgroup (Supplementary Fig. 2B). Allograft rejection, DNA replication, mismatch repair, protein export and ribosome were significantly enriched in the high DSC1 subgroup (Fig. 8C), but glycerophospholipid metabolism, notch signaling pathway, and viral protein interaction with cytokine and cytokine receptor were significantly enriched in the low DSC1 subgroup (Supplementary Fig. 2C). Allograft rejection, fatty acid elongation, and mismatch repair were significantly enriched in the high KIR2DL3 subgroup (Fig. 8D), but measles, type II diabetes mellitus, and viral protein interaction with cytokine and cytokine receptor were significantly enriched in the low KIR2DL3 subgroup (Supplementary Fig. 2D). The NOD-like receptor signaling pathway, notch signaling pathway and osteoclast differentiation were significantly enriched in the high MX2 subgroup (Fig. 8E), but fatty acid elongation, mismatch repair and protein export were enriched in the low MX2 subgroup (Supplementary Fig. 2E).

Immune cell infiltration analysis

The CIBERSORT algorithm was used to evaluate the differences in immune cell infiltration and hallmark gene sets between the SLE and control groups. As shown in Fig. 9A, B, the proportions of naive B cells, regulatory T cells (Treg cells), monocytes, M0/M1 macrophages, eosinophils and neutrophils in SLE samples were significantly increased compared with those in control samples, while the percentages of CD8⁺ T cells, NK cells, dendritic cells and resting mast cells were significantly decreased.

Furthermore, correlation analysis revealed that four optimal key genes (ABCB1, CD247, DSC1, and KIR2DL3) were significantly positively related to the infiltration of monocytes, M0/M1 macrophages, Treg cells, CD8⁺ T cells, resting dendritic cells and resting mast cells but negatively related to the infiltration of activated memory CD4⁺ T cells, activated dendritic cells, eosinophils and neutrophils (Fig. 9C-F). However, the MX2 gene is roughly the opposite of the above four genes (Fig. 9G). For instance, the ABCB1 gene was positively related to M1 macrophages (R = 0.58, P < 2.2e-16) but negatively related to neutrophils (R = − 0.74, P < 2.2e-10) (Supplementary Fig. 3). In Fig. 9H, I, the gene correlations were also performed. These five optimal key genes displayed significant correlations. For example, the correlation coefficient between ABCB1 and CD247 was 0.88, but the correlation coefficient between ABCB1 and MX2 was − 0.74, indicating that four optimal key genes (ABCB1, CD247, DSC1 and KIR2DL3) had a significant functional similarity, while the function of MX2 gene was opposite to the remaining four genes.

To explore whether there were differences in the hallmark gene sets between the SLE and control groups, the ssGSEA algorithm was used to identify the significance of differences in 50 hallmark gene sets between the two groups according to the enrichment score. The distribution between the SLE and control groups is shown in Fig. 10A.

The gene sets showed significant differences, such as allograft rejection, coagulation, heme metabolism, angiogenesis, P53 pathway, inflammatory response and hypoxia. Hence, compared with the control group, these hallmark gene sets might be overactivated in the SLE group. In addition, four optimal key genes except for MX2 were generally consistent. For example, the four optimal key genes (ABCB1, CD247, DSC1, KIR2DL3) were positively related to the oxidative phosphorylation hallmark gene set, while MX2 was negatively related to it (Fig. 10B). Therefore, the role of the above genes should be further explored in SLE.

Validation of the five optimal key genes

The relative expression of five optimal key genes was verified with qRT-PCR in SLE patients and healthy controls. Compared with the control group, the expression of ABCB1 (Fig. 11A), CD247 (Fig. 11B), and KIR2DL3 (Fig. 11D) was significantly downregulated, while MX2 (Fig. 11E) was significantly upregulated in SLE patients (all P < 0.05), which was consistent with the results of this bioinformatic analysis. However, the expression of DSC1 (Fig. 11C) was also downregulated in SLE samples, although there was no statistical significance (P>0.05).

SLE is an autoimmune-mediated diffuse connective tissue disease characterized by immune inflammation that involves multiple systems and organs of the whole body. The main clinical characteristics are immune disorder and systemic inflammation, leading to progressive and irreversible multiorgan damage[24]. The clinical manifestations are extensive, including kidney, dermatological, neuropsychiatric, and cardiovascular symptoms. In addition, SLE patients are always recurrent and remitted, and there are a large number of autoantibodies in the body. If not treated in time, it is easy to cause irreversible damage to affected organs and even death in patients. For many years, the pathogenesis of SLE has been the focus of international research. The current medical consensus is that the pathogenesis of SLE is the result of the interaction of genetic, environmental and random factors. Autoantibodies, including anti-dsDNA antibodies and anti-SM antibodies, are of great significance for the diagnosis of SLE, but the sensitivity is low. The existing SLE immune serological indices cannot reach high sensitivity and specificity at the same time, resulting in difficulties in the diagnosis of SLE. Therefore, it is urgent to explore new biological markers to improve the diagnostic rate of SLE.

In recent years, bioinformatic technology has been increasingly developed in the fields of medical research, such as the search for pathogenic genes and the screening of effective drug targets. In this study, candidate genes associated with SLE were obtained after overlapping the genes analyzed by DEG and WGCNA. Then, machine learning algorithms such as LASSO, RF and SVM-REF were used to select five optimal key genes (ABCB1, CD247, DSC1, KIR2DL3 and MX2). The ROC curve analysis confirmed that these genes were highly expressed in SLE samples in the above datasets. Furthermore, GO, KEGG, DO and GSEA analyses were used to study the biological functions of these genes. ssGSEA was carried out to assess differences in immune cell infiltration and hallmark gene sets between SLE and control samples.

At present, several studies have discussed the relationship between these genes and the pathogenesis of SLE[25–30]. However, studies on DSC1 and SLE have not been conducted. ABCB1 is a multidrug resistant protein mainly expressed in the brain, liver, skin and other organs[31]. Recently, ABCB1 has been reported to be expressed in a variety of immune cells, such as monocytes, antigen-presenting dendritic cells, T cells, and B cells, and is involved in the expulsion of inflammatory molecules[32, 33]. The expression level of ABCB1 was also reported to be significantly lower in SLE patients[27]. CD247, also known as the T-cell surface glycoprotein CD3 zeta chain, is one part of the T-cell antigen receptor (TCR) complex and plays a critical role in receptor expression and signaling. Studies have shown that low CD247 expression due to chronic inflammation is related to the decreased T-cell activity[34]. Due to its involvement in SLE, the expression of CD247 in SLE patients was also found to be significantly lower than that in healthy controls[28]. The protein encoded by DSC1 is a calcium-dependent glycoprotein, a member of the desmocolloidal subfamily of the calcium-adhesion protein superfamily, which is mainly expressed in the upper layer of the epidermis and is the autoantigen of IgA pemphigus[35]. Killer cell immunoglobulin-like receptor (KIR) is expressed on the surface of human natural killer cells (NK cells) and T cells and is associated with viral infection and tumor and autoimmune diseases. KIR2DL3 plays an important role in some immune-related diseases[36, 37]. Meanwhile, the KIR2DL3 gene was also significantly lower in SLE patients than in healthy subjects[29]. MX2 is an interferon inducible gene that is mostly known for its antiviral activity[38]. In addition, MX2 exerts immune effects by mediating multiple immune cells, including neutrophils, macrophages and T cells[39]. In addition, significantly high expression of MX2 was also found in SLE samples[30]. To make this analysis more credible, qRT-PCR results showed that the relative expression of ABCB1, CD247, and KIR2DL3 was significantly downregulated, while MX2 was significantly upregulated in SLE patients. Notably, DSC1 was also lower in SLE samples, but without statistical significance. In general, the validation results were in line with the bioinformatic analysis and previous studies.

In the analysis of immune cell infiltration, different types of immune cells, including naive B cells, regulatory T cells (Treg cells), monocytes, plasma cells and NK cells, were significantly regulated in SLE samples. Previous studies have also revealed that the above immune cells are considered as critical players in SLE. For instance, the distribution of B-cell subsets in the peripheral blood of SLE patients is significantly different from that of normal people. Patients with SLE have a low frequency of memory B cells, a high frequency of double-negative B cells, and a high frequency of naive B cells[40]. In addition, the onset of SLE involves the entire immune system, and T cells have a pivotal role in this process. Both numerical and functional disturbances have been reported in the main CD4⁺ T helper cell subsets including Th1, Th2, Th17, regulatory, and follicular helper cells[41]. In the GSEA, some signaling pathways were overactivated in SLE samples, such as the mTORC1 signaling pathway and IL-6-JAK-STAT3 signaling pathway. Studies have reported that IL-6 downregulates the expression of TGFβRI through the JAK-STAT3 pathway in SLE patients[42]. Furthermore, the activation of mTORC1 was also found to be associated with the pathogenesis and development of SLE[43, 44].

This study relies on bioinformatics analysis, and there are some deviations in the results. Due to the limited conditions of our team, appropriate experiments were lacking to verify the results.

In this study, a total of five optimal key genes were screened, providing new targets for the pathogenesis of SLE. Additionally, more machine learning algorithms were applied to improve the accuracy of genetic screening. Moreover, this study not only used external dataset verification but also carried out real-time qPCR experiment verification, making the screening results more reliable. In conclusion, the purpose of this study is to provide a new direction for the clinical diagnosis and precise treatment of SLE by machine-learning analysis.

biological process

cellular component

Disease Ontology

GEO

Gene Expression Omnibus

Gene Ontology

gene importance

GSEA

Gene set enrichment analysis

KEGG

Kyoto Encyclopedia of Genes and Genomes

KIR

Killer cell immunoglobulin-like receptor

LASSO

least absolute shrinkage and selection operation (LASSO)

molecular function

module membership

NK cells

natural killer cells

PPI

protein-protein interaction

random forest

ROC

receiver operating characteristic curves

SLE

systemic lupus erythematosus

SVM-REF

support vector machine-recursive feature elimination

TCR

T cell antigen receptor

TOM

Topological overlap matrix

WGCNA

weighted gene coexpression network.

Ethics approval and consent to participate

This project was approved by the Ethics Committee of the First Affiliated Hospital, Zhejiang University School of Medicine (2023-675), and all study protocols complied with the Declaration of Helsinki. All involved participants provided written informed consent.

Consent for publication

Not applicable.

Availability of data and materials

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

Competing interests

The authors declare that the study was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

This work was supported by the National Natural Science Foundation of China (Grant No. 82172335, 81971994, 81871709, 91846103), Zhejiang Provincial Key Research and Development Program (Grant No. 2020C03032).

Authors’ contributions

XZ, LD, DC, and JX designed the study. LD analyzed the data, participated in the data collection, and XZ prepared the manuscript. In addition, XZ prepared figures 1-11, and LD prepared supplementary files. DC was responsible for the qRT-PCR experiment. JX and DC helped the analysis with constructive discussions. All authors critically revised the manuscript.

Acknowledgments

We would like to thank the contributions of public databases such as the GEO to human medicine. Our manuscript had been edited for proper English language by the highly qualified native English speaking editors at AJE, which was verified on the AJE website using the verification code C136-F7D0-F522-3D85-ACA6.

Sawada T, Fujimori D, Yamamoto Y: Systemic lupus erythematosus and immunodeficiency. Immunol Med 2019, 42(1):1-9.
Rekvig OP: Autoimmunity and SLE: Factual and Semantic Evidence-Based Critical Analyses of Definitions, Etiology, and Pathogenesis. Front Immunol 2020, 11:569234.
Weinstein A, Alexander RV, Zack DJ: A Review of Complement Activation in SLE. Curr Rheumatol Rep 2021, 23(3):16.
Tenbrock K, Rauen T: T cell dysregulation in SLE. Clin Immunol 2022, 239:109031.
Basta F, Fasola F, Triantafyllias K, Schwarting A: Systemic Lupus Erythematosus (SLE) Therapy: The Old and the New. Rheumatology and Therapy 2020, 7(3):433-446.
Lee JM, Chen MH, Chou KY, Chao Y, Chen MH, Tsai CY: Novel immunoprofiling method for diagnosing SLE and evaluating therapeutic response. Lupus Sci Med 2022, 9(1).
Zhao X, Zhang L, Wang J, Zhang M, Song Z, Ni B, You Y: Identification of key biomarkers and immune infiltration in systemic lupus erythematosus by integrated bioinformatics analysis. J Transl Med 2021, 19(1):35.
Perez RK, Gordon MG, Subramaniam M, Kim MC, Hartoularos GC, Targ S, Sun Y, Ogorodnikov A, Bueno R, Lu A et al: Single-cell RNA-seq reveals cell type-specific molecular and genetic associations to lupus. Science 2022, 376(6589):eabf1970.
Barrett T, Wilhite SE, Ledoux P, Evangelista C, Kim IF, Tomashevsky M, Marshall KA, Phillippy KH, Sherman PM, Holko M et al: NCBI GEO: archive for functional genomics data sets--update. Nucleic Acids Res 2013, 41(Database issue):D991-995.
Langfelder P, Horvath S: WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics 2008, 9:559.
Yao M, Zhang C, Gao C, Wang Q, Dai M, Yue R, Sun W, Liang W, Zheng Z: Exploration of the Shared Gene Signatures and Molecular Mechanisms Between Systemic Lupus Erythematosus and Pulmonary Arterial Hypertension: Evidence From Transcriptome Data. Front Immunol 2021, 12:658341.
Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES et al: Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proceedings of the National Academy of Sciences 2005, 102(43):15545-15550.
Powers RK, Goodspeed A, Pielke-Lombardo H, Tan AC, Costello JC: GSEA-InContext: identifying novel and common patterns in expression experiments. Bioinformatics 2018, 34(13):i555-i564.
Sepulveda JL: Using R and Bioconductor in Clinical Genomics and Transcriptomics. J Mol Diagn 2020, 22(1):3-20.
Huang H: Controlling the false discoveries in LASSO. Biometrics 2017, 73(4):1102-1110.
Sanz H, Valim C, Vegas E, Oller JM, Reverter F: SVM-RFE: selection and visualization of the most relevant features through non-linear kernels. BMC Bioinformatics 2018, 19(1):432.
Lin X, Li C, Zhang Y, Su B, Fan M, Wei H: Selecting Feature Subsets Based on SVM-RFE and the Overlapping Ratio with Applications in Bioinformatics. Molecules 2017, 23(1).
Uddin S, Khan A, Hossain ME, Moni MA: Comparing different supervised machine learning algorithms for disease prediction. BMC Med Inform Decis Mak 2019, 19(1):281.
Goldstein BA, Polley EC, Briggs FB: Random forests for genetic association studies. Stat Appl Genet Mol Biol 2011, 10(1):32.
Cao R, Lopez-de-Ullibarri I: ROC Curves for the Statistical Analysis of Microarray Data. Methods Mol Biol 2019, 1986:245-253.
Chen B, Khodadoust MS, Liu CL, Newman AM, Alizadeh AA: Profiling Tumor Infiltrating Immune Cells with CIBERSORT. Methods Mol Biol 2018, 1711:243-259.
Leek JT, Johnson WE, Parker HS, Jaffe AE, Storey JD: The sva package for removing batch effects and other unwanted variation in high-throughput experiments. Bioinformatics 2012, 28(6):882-883.
Leek JT: svaseq: removing batch effects and other unwanted noise from sequencing data. Nucleic Acids Res 2014, 42(21):e161.
Scharer CD, Blalock EL, Mi T, Barwick BG, Jenks SA, Deguchi T, Cashman KS, Neary BE, Patterson DG, Hicks SL et al: Epigenetic programming underpins B cell dysfunction in human SLE. Nat Immunol 2019, 20(8):1071-1082.
Zhong Y, Zhang W, Hong X, Zeng Z, Chen Y, Liao S, Cai W, Xu Y, Wang G, Liu D et al: Screening Biomarkers for Systemic Lupus Erythematosus Based on Machine Learning and Exploring Their Expression Correlations With the Ratios of Various Immune Cells. Front Immunol 2022, 13:873787.
Martins M, Williams AH, Comeau M, Marion M, Ziegler JT, Freedman BI, Merrill JT, Glenn SB, Kelly JA, Sivils KM et al: Genetic association of CD247 (CD3zeta) with SLE in a large-scale multiethnic study. Genes Immun 2015, 16(2):142-150.
Joy MS, Roberts BV, Wang J, Hu Y, Hogan SL, Falk RJ: A pilot study of leukocyte expression patterns for drug metabolizing enzyme and transporter transcripts in autoimmune glomerulonephritis. Int J Clin Pharmacol Ther 2014, 52(4):303-313.
Gorman CL, Russell AI, Zhang Z, Cunninghame Graham D, Cope AP, Vyse TJ: Polymorphisms in the CD3Z gene influence TCRzeta expression in systemic lupus erythematosus patients and healthy controls. J Immunol 2008, 180(2):1060-1070.
Tozkir JD, Tozkir H, Gurkan H, Donmez S, Eker D, Pamuk GE, Pamuk ON: The investigation of killer cell immunoglobulin-like receptor genotyping in patients with systemic lupus erytematosus and systemic sclerosis. Clin Rheumatol 2016, 35(4):919-925.
Meng XW, Cheng ZL, Lu ZY, Tan YN, Jia XY, Zhang M: MX2: Identification and systematic mechanistic analysis of a novel immune-related biomarker for systemic lupus erythematosus. Front Immunol 2022, 13:978851.
Magarbeh L, Hassel C, Choi M, Islam F, Marshe VS, Zai CC, Zuberi R, Gammal RS, Men X, Scherf-Clavel M et al: ABCB1 Gene Variants and Antidepressant Treatment Outcomes: A Systematic Review and Meta-Analysis Including Results from the CAN-BIND-1 Study. Clin Pharmacol Ther 2023.
Seelig A, Li-Blatter X: P-glycoprotein (ABCB1) - weak dipolar interactions provide the key to understanding allocrite recognition, binding, and transport. Cancer Drug Resist 2023, 6(1):1-29.
Liu X: ABC Family Transporters. Adv Exp Med Biol 2019, 1141:13-100.
Bronstein-Sitton N, Cohen-Daniel L, Vaknin I, Ezernitchi AV, Leshem B, Halabi A, Houri-Hadad Y, Greenbaum E, Zakay-Rones Z, Shapira L et al: Sustained exposure to bacterial antigen induces interferon-gamma-dependent T cell receptor zeta down-regulation and impaired T cell function. Nat Immunol 2003, 4(10):957-964.
Ishii N: Significance of anti-desmocollin autoantibodies in pemphigus. J Dermatol 2023, 50(2):132-139.
Vilches C, Parham P: KIR: diverse, rapidly evolving receptors of innate and adaptive immunity. Annu Rev Immunol 2002, 20:217-251.
Deborska-Materkowska D, Perkowska-Ptasinska A, Sadowska-Jakubowicz A, Pazik J, Serwanska-Swietek M, Mikolajczyk N, Swider R, Nowak J, Durlik M: Antiviral prophylaxis, male sex, and killer immunoglobulin-like receptor KIR2DL3 as markers for stratifying the risk of BK polyomavirus-associated nephropathy in kidney transplant recipients. Pol Arch Intern Med 2023, 133(1).
Juraleviciute M, Nsengimana J, Newton-Bishop J, Hendriks GJ, Slipicevic A: MX2 mediates establishment of interferon response profile, regulates XAF1, and can sensitize melanoma cells to targeted therapy. Cancer Med 2021, 10(8):2840-2854.
Zhao L, Fu X, Han X, Yu Y, Ye Y, Gao J: Tumor mutation burden in connection with immune-related survival in uterine corpus endometrial carcinoma. Cancer Cell Int 2021, 21(1):80.
Gao M, Liu S, Chatham WW, Mountz JD, Hsu HC: IL-4-Induced Quiescence of Resting Naive B Cells Is Disrupted in Systemic Lupus Erythematosus. J Immunol 2022, 209(8):1513-1522.
Li H, Boulougoura A, Endo Y, Tsokos GC: Abnormalities of T cells in systemic lupus erythematosus: new insights in pathogenesis and therapeutic strategies. J Autoimmun 2022, 132:102870.
Yan Q, Chen W, Song H, Long X, Zhang Z, Tang X, Chen H, Lin H, Sun L: Tofacitinib Ameliorates Lupus Through Suppression of T Cell Activation Mediated by TGF-Beta Type I Receptor. Front Immunol 2021, 12:675542.
Zhao X, Wang S, Wang S, Xie J, Cui D: mTOR signaling: A pivotal player in Treg cell dysfunction in systemic lupus erythematosus. Clin Immunol 2022, 245:109153.
Oaks Z, Winans T, Huang N, Banki K, Perl A: Activation of the Mechanistic Target of Rapamycin in SLE: Explosion of Evidence in the Last Five Years. Curr Rheumatol Rep 2016, 18(12):73.

No competing interests reported.

Download PDF

Journal Publication

published 10 Nov, 2023

Read the published version in BMC Immunology →

Editorial decision: Major revision
02 Aug, 2023
Reviews received at journal
28 Jul, 2023
Reviewers agreed at journal
25 Jul, 2023
Reviewers invited by journal
25 Jul, 2023
Editor invited by journal
19 Jul, 2023
Editor assigned by journal
19 Jul, 2023
Submission checks completed at journal
18 Jul, 2023
First submitted to journal
07 Jul, 2023

You are reading this latest preprint version

Exploration of biomarkers for systemic lupus erythematosus by machine-learning analysis

Status:

Journal Publication

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

Background

Materials and methods

Download and processing of data

Differential expression analysis

Weighted gene coexpression network analysis

GSEA and correlation analysis

Functional enrichment analysis

Further screening of optimal key genes

The expression and predictive value of optimal key genes

Analysis of hallmark gene sets and immune cell infiltration

Isolation of human PBMCs

Real-time quantitative PCR

Statistical analysis

Results

DEGs between SLE and control samples

Application of WGCNA

Functional enrichment analysis of overlapping genes

Further identification of optimal key genes through machine-learning algorithms

Evaluation of the expression and diagnostic value of optimal key genes

Identification of the biological function of five key genes

Immune cell infiltration analysis

Validation of the five optimal key genes

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1