Bioinformatic Analysis of the Potential Common Pathogenic Mechanisms for Psoriasis and Metabolic Syndrome

The pathogeneses of psoriasis and metabolic syndrome are closely related; however, the underlying biological mechanisms are yet to be clarified. A psoriasis training set was downloaded from the Gene Expression Omnibus database and analyzed to identify the differentially expressed genes (|logFC|> 1 and adjust P < 0.05). Differentially expressed genes for metabolic syndrome were obtained from the GeneCards, Online Mendelian Inheritance in Man, and DisGeNET databases, and crosstalk genes were obtained for multiple enrichment analysis after identifying the disease intersection. Characteristic crosstalk genes were screened using the least absolute shrinkage and selection operator regression model and random forest tree model, and the genes with area under the receiver operating characteristic curve > 0.7 were selected for validation by the two validation sets. Differential analyses of immune cell infiltration were performed on psoriasis lesion and control samples using the CIBERSORT and ImmuCellAI methods, and correlation analyses were performed between the screened signature crosstalk genes and immune cell infiltration. Significant crosstalk genes were analyzed based on the psoriasis area and severity index and on the responses to biological agents. We found five signature genes (NLRX1, KYNU, ABCC1, BTC, and SERPINB4) were screened based on two machine learning algorithms, and NLRX1 was validated. The infiltration of multiple immune cells in psoriatic lesions and non-lesions was associated with NLRX1 expression. NLRX1 was found to be associated with psoriasis severity and response rate after the use of biologics. NLRX1 could be a significant crosstalk gene for psoriasis and metabolic syndrome.


INTRODUCTION
Psoriasis is a chronic immune-mediated inflammatory skin disease with a global prevalence of 2-3% [1]. As a systemic disease, psoriasis is often associated with arthritis and metabolic diseases, which seriously affect the quality of life and life expectancy of patients [2]. Metabolic syndrome is the most common comorbidity in psoriasis and has a prevalence of 20-50% among patients with psoriasis [2]. Metabolic syndrome includes a group of clinical syndromes, characterized by the combined appearance of various metabolic diseases, such as obesity, dyslipidemia, hypertension, and insulin resistance; it is a risk factor for cardiovascular disease [3] and a major cause of death in patients with psoriasis [4]. For such patients with metabolic syndrome, the systemic use of drugs, such as retinoic acid and cyclosporine, is limited owing to their potential of developing metabolic disorders [5]. Therefore, the mechanism underlying comorbidity between psoriasis and metabolic syndrome must be investigated carefully to improve clinical treatments and management of psoriasis. However, the pathological mechanisms underlying this comorbidity remain to be elucidated first.
Previous studies had shown that M1-type macrophages and TNFα are closely related to the pathogenesis of psoriasis [6] and metabolic syndrome; however, systematic, genetic, pathway-based, and cytological studies on the comorbidity mechanisms have not yet been performed.
The current study aimed to screen the gene sets and characteristic crosstalk genes of the two comorbidities using a machine learning approach and investigate the clinical significance of the genes to identify new targets for the treatment of patients with psoriasis as well as metabolic syndrome.

Data Acquisition and Study Design
Microarray expression datasets GSE117239, GSE117468, GSE136757, GSE30999, and GSE98895 were downloaded from the Gene Expression Omnibus (GEO, https:// www. ncbi. nlm. nih. gov/ geo/) database. The GSE117239, GSE117468, and GSE136757 datasets were merged using the "sva" package in R software (R.4.2.0, http:// www.r-proje ct. org/) to eliminate batch effects, selecting the samples before the drug was administered, with complete clinical baseline and follow-up data for the psoriasis training set. GSE30999 and GSE98895 were used as validation sets for psoriasis and metabolic syndrome to validate the accuracy of the models. The chip information and workflow chart are presented in Table 1 and Fig. 1, respectively. The workflow chart was created using BioRender.

Acquisition of Differential Genes
Differential expression analysis was performed on the psoriasis training set using the "limma" package for R software, with |logFC|> 1 and adjusted P value < 0.05 being the screening criteria to identify the differentially expressed genes for psoriasis. The GeneCards database (https:// www. genec ards. org/), Online Mendelian Inheritance in Man (OMIM) database (https:// www. omim. org/), and DisGeNET database (http:// www. disge net. org. home/) were searched using the key word "Metabolic Syndrome" to identify the disease genes of metabolic syndrome. Intersection of the two diseases was used to obtain the crosstalk genes between psoriasis and metabolic syndrome.

Enrichment Analysis of Crosstalk Genes
The R software packages "org.Hs.eg.db" and "clusterProfiler" were used to enrich the obtained crosstalk genes of psoriasis and metabolic syndrome in the Gene Ontology (GO) categories of molecular function (MF), biological process (BP), and cellular component (CC). Further, Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed on the obtained crosstalk genes in psoriasis and metabolic syndrome. Adjust P value < 0.05 was used as a filtering condition. The KEGG enrichment results were summarized according to the classification information available on the KEGG pathway website (https:// www. kegg. jp/ kegg/ pathw ay. html) to obtain the secondary pathway classifications.

Machine Learning-based Screening the Significant Crosstalk Genes
We used two machine learning algorithms, namely the least absolute shrinkage and selection operator (LASSO) regression and random forest tree models, to screen for key crosstalk genes in psoriasis and metabolic syndrome. LASSO regression can be performed by adding a penalty term to the loss function and adjusting the penalty coefficients such that the coefficients of the less influential features decay to zero and only the important features are retained. We used the R software package "glmnet" to select the genes, corresponding to the points with smallest cross-validation values, as the genes to be screened based on the differential expression of crosstalk genes. The random forest tree algorithm randomly selects samples from the sample set with a random return, generates a decision tree to partition the sample set based on the features of the selected samples, and repeats the process several times to determine the best partitioned features. We used the R software package "randomForest"  to perform the analysis, set the number of decision trees to 500, and select feature genes using a gene importance score greater than 2 as the filtering criterion. The genes obtained by the two machine learning algorithms were intersected, and the receiver operating characteristic (ROC) curve was plotted for the intersected genes. Genes with an area under the curve (AUC) greater than 0.7 were selected. Differential expression of the candidate genes in the psoriasis lesion and non-lesion groups, the metabolic syndrome disease group, and the normal group, and the area under ROC curve of the diagnostic model, established with the candidate genes, were verified using the test set, and the final crosstalk characteristic genes were identified based on P < 0.05 and AUC > 0.7 as the screening conditions.

Immune Cell Infiltration Analysis
We chose two immune infiltration analysis methods, namely CIBERSORT and ImmuCellAI, to analyze the psoriasis training dataset. CIBERSORT is a deconvolution method, based on the linear support vector principle, for the expression matrix of seven human immune cell subtypes [7], and it can transform the gene expression matrix into an immune cell infiltration matrix. The gene expression data of 22 immune cells and the CIBERSORT.R package were downloaded from the official website of CIBER-SORT (https:// ciber sort. stanf ord. edu/). Immune cell infiltration was analyzed using R software, and the immune cell expression matrix of plausible samples was obtained with a screening condition of P < 0.05. ImmuCellAI can calculate the relative abundance of samples in each cell type by identifying biomarkers of the immune cell gene set [8]. Gene expression matrix of the psoriasis training dataset was uploaded by logging into the ImmuCellAI website (http:// bioin fo. life. hust. edu. cn/ web/ ImmuC ellAI/), and infiltration results and enrichment scores of the 24 immune cell types were obtained. The immune infiltration results obtained by both methods were analyzed separately to determine the difference in infiltration between the psoriasis lesion and non-lesion groups. In addition, a correlation analysis of crosstalk signature gene expression and immune cell content was performed to investigate the effect of crosstalk signature genes on immune cell infiltration, with P < 0.05 used as the screening condition.

Clinical Significance of Significant Crosstalk Genes
EmpowerStats software (version 5.0) was used to perform smooth curve fitting and threshold saturation effect analyses of crosstalk gene expression, based on the psoriasis area and severity index scores in psoriasis training dataset, to explore the relationship between gene expression and disease severity. The predictive ability of characteristic crosstalk genes for response after biological treatment was observed by plotting the ROC curves and column line plots. The rates of change in PASI before and after treatment with different biological regimens were compared using ANOVA, and PASI75, PASI90 response rates with different biological regimens were compared using Fisher's exact probability method and a continuously corrected chi-square test.

Differential Genes and Crosstalk Genes
Our combined psoriasis training dataset contained 164 patients in GSE117239 (n = 66), GSE117468 (n = 86), and GSE136757 (n = 12), and differential analysis of the lesion and non-lesion groups yielded a total of 1076 differentially expressed genes, of which 617 genes were upregulated and 459 genes were downregulated in the psoriasis lesion group, as shown in Fig. 2a, b. A total of 15,952, 139, and 1125 metabolic syndrome disease genes were obtained from the GeneCards, OMIM, and DisGeNET databases, respectively, and 16,017 metabolic syndrome-related disease genes were obtained by combining and de-duplicating the genes, as shown in Fig. 2c. Considering the intersection of the two diseases, 802 intersecting genes were obtained, of which 476 were upregulated and 326 were downregulated in the psoriasis lesion group, as shown in Fig. 2d.

Enrichment Analysis
GO enrichment results showed that the set of genes associated with cell division terms, such as mitotic nuclear division, organelle fission, spindle microtubule, condensed chromosome, centromeric region, positive regulation of cytokine production, cellular response to molecules of bacterial origin, receptor ligand activity, cytokine activity, chemokine receptor binding, and interleukin-1 receptor binding, and the set of genes related to immune inflammation were significantly enriched among the crosstalk genes of psoriasis and metabolic syndrome, as shown in Fig. 3a. The KEGG enrichment results showed that inflammation-related pathways, such as the NOD-like receptor signaling pathway and JAK-STAT signaling pathway; metabolism-related pathways, such as the glutathione metabolism; and human diseaserelated pathways, such as influenza A, hepatitis C, and bladder cancer, were significantly enriched among the crosstalk genes of psoriasis and metabolic syndrome, as shown in Fig. 3b.

Screening for Signature Crosstalk Genes
We applied the LASSO regression model to screen 15 genes and the random forest tree method to screen 35 genes, from which five signature genes were obtained, namely, NLRX1, KYNU, ABCC1, BTC, and SERPINB4, as shown in Fig. 4. NLRX1 was differentially expressed in both disease and control groups in the two test sets, had good diagnostic discriminatory ability, and could be used as a comorbid crosstalk signature gene for psoriasis and metabolic syndrome, as shown in Fig. 5.

Immune Cell Infiltration Analysis
Immune cell infiltration analysis was performed using the CIBERSORT method; the immune cell infiltration matrix was filtered, and 183 plausible samples were obtained, including 100 cases in the psoriasis lesion group and 83 cases in the normal control group. The results showed increased infiltration of immune cells in psoriatic lesions, such as T cells CD4 memory activated, T cells follicular helper, macrophages M1, dendritic cells activated, and neutrophils, and decreased infiltration of T cells CD4 naïve, NK cells activated, macrophages M2, plasma cells, mast cells resting, and other immune cells (Fig. 6a). The expression of NLRX1 was positively correlated with immune cells, such as Dendritic cells activated, T cells CD4 memory activated, T cells follicular helper, and macrophages M1, and negatively correlated with mast cells resting, NK cells activated, plasma cells, macrophages M2, and T cells CD4 naïve (Fig. 7a). Immune cell infiltration analysis was performed using the ImmuCellAI method. Immune cells, such as neutrophils, CD4 + T, Th1, Tfh, and central memory cells, showed increased infiltration in psoriatic lesions, whereas NKT, Th2, Th17, and CD8 naïve cells showed decreased infiltration in psoriatic lesions (Fig. 6b). The expression of NLRX1 was positively correlated with Th1, Tr1, neutrophils, and Tfh cells and negatively correlated with Th17, naïve CD8, and Th2 cells (Fig. 7b).

Clinical Significance of the Characteristic Crosstalk Genes
To investigate the relationship between NLRX1 expression and psoriasis clinical symptoms, we performed a correlation analysis between NLRX1 gene expression and the PASI score. However, the results showed no linear relationship between them. Therefore, we performed threshold effect and saturation effect analyses and smoothing curve fitting. Results showed that when NLRX1 expression was less than 8.7, the PASI score increased, and when NLRX1 expression was higher than 8.7, the PASI score decreased (Fig. 8 and Table 2). We further investigated the relationship across NLRX1 expression, the PASI score, and biological response rate. When PASI75 was used as the criterion, the PASI values in the responder group were significantly higher than those in the non-responder group. However, when PASI90 was used as the criterion, no significant difference between the PASI values in the responder and non-responder groups was observed, although the expression of NLRX1 increased significantly in the responder group, as shown in Table 3. The above results confirmed that PASI scores are not good predictors of response to biologics. Therefore, we added NLRX1 gene expression to the model, due to which the new model could better predict the response of patients with psoriasis to biologics, either by PASI75 or PASI90. The patients included in the study received eight different biological regimens, it may have an impact on the prediction model. However, the accuracy of the biological prediction model may be reduced after stratified analysis of the biological agents due to the small sample size. Therefore, we compared the rates of change of PASI, PASI75, and PASI90 response rates, using different biological treatment regimens as grouping conditions, and the results showed no significant differences in the rates of change of PASI and PASI75 response rates before and after treatment in different groups, while the response rates of PASI90 showed group differences, as shown in Tables 4, 5 and 6.

DISCUSSION
Circulating proinflammatory mediators and a broad inflammatory response are central to the mechanisms underlying psoriasis and metabolic syndrome comorbidity [4]. The inflammatory cascade in psoriasis starts at the skin [4]. When the skin is infected or injured, damaged keratinocytes release antimicrobial peptides and combine with their own genetic material to activate dendritic cells to turn on adaptive immunity. In presence of helper T cells, such as Th1, inflammatory factors, such as IL-17 and TNFα, are released to activate the JAK-STAT pathway and promote keratinocyte proliferation. Over-proliferating keratinocytes, in turn, continue to secrete cytokines and chemokines into the skin, thus forming a positive feedback mechanism [1,9]. Th1 cytokines promote the polarization of pro-inflammatory M1 macrophages, and Th2 cytokines promote the polarization of anti-inflammatory M2 macrophages [10]. The dynamic balance of M1/M2 is important for maintaining the balance between proinflammatory Th1 and anti-inflammatory Th2 [11]. Under inflammatory conditions, the balance is disrupted by the increased secretion of Th1-related cytokines and polarization of M1 macrophages, which in turn can be involved in the production of pro-inflammatory factors and depletion of Treg cells, thereby forming a positive feedback loop that promotes the development of psoriasis [12,13]. Neutrophils can also infiltrate the dermis and epidermis to form Munro's microabscesses induced by chemokines [14], which is one of the important pathological manifestations of psoriasis. NK cells can kill overactivated macrophages and T cells; when NK cell function is impaired, the survival of activated immune cells is prolonged, thus leading to chronic inflammation [15].
The reduced infiltration of Th17 cells in psoriasis seems inconsistent with previous research. However, there is no consensus in the literature on the level of expression of the major cytokines of Th17 cells in psoriasis. Arican et al. [16] found no significant difference in serum IL-17 levels between patients with psoriasis and controls, and similar results were observed by Roh et al. [17] for serum IL-17A and IL-22 levels. In majority of the studies, the cytokine levels of Th17 cells were elevated in psoriasis [18][19][20]. The differences may be related to the severity of the disease or the clinical subtype [21]. We were unable to analyze the source of the differences due to the lack of more available information in the microarray, and such an analysis would require a larger sample of clinical cohorts. Although the IL-17 signaling pathway was significantly enriched in psoriatic lesions in our study, the innate immune system has also been shown to produce IL-17, independent of IL-23-mediated induction of Th17 cells [22]. This suggested that compared to that in healthy skin, Th17 cell infiltration in psoriatic lesions may be unchanged or reduced despite elevated IL-17 levels.
In metabolic syndrome, the inflammatory cascade starts in adipose tissue [23]. Excessive accumulation of adipose tissue in obesity produces saturated fatty acids that promote the secretion of inflammatory factors TNF-α and IL-6 by activating Toll-like receptors and increasing the polarization of pro-inflammatory M1 macrophages. When adipocytes are exposed to inflammatory factors, the JAK-STAT pathway is activated, which inhibits insulin signaling and leads to insulin resistance [23,24]. Insulin resistance can lead to increased IL-17 secretion, and patients with metabolic syndrome may have significantly higher serum IL-17 levels than the healthy population [25]. Thus, signaling pathways associated with cell proliferation and immune inflammation were significantly enriched in psoriasis and metabolic syndrome comorbidity genes owing to the massive replication of cytokines and proinflammatory cells.
The NOD-like receptor (NLR) signaling pathway involves a subgroup of the family of pathogen recognition receptors that can regulate innate immunity by enhancing or attenuating pro-inflammatory signaling cascades [26]. NLRs can be classified into five subgroups (NLRA, NLRB, NLRC, NLRP, and NLRX1) based on their structural and functional characteristics [27]. NLRs play important roles in many inflammatory diseases and metabolic disorders. Activation of NLRP3 inflammatory vesicles in keratinocytes can promote the skin T-cell response [28]. Tervaniemi et al. [29] found that NLR signaling pathway genes are upregulated in psoriatic epidermis, and Tang et al. [30] revealed that the NLR signaling pathway is a key pathway in psoriasis among Han Chinese population. NOD2 can promote the development of type 1 diabetes by participating in the induction of pancreatic Th1 cells and Th17 cell production [31]. Ginsenoside RG1 was reported to prevent the development of T1D by attenuating NLRP3 function and inhibiting IL-1β and IL-18 secretion in the liver and pancreas of mice [32].
NLRX1 is the only NLR that targets mitochondria; however, data on its function are often contradictory in the literature due to its dual inflammatory and metabolic  as the model; f predictive model of the response or non-response to psoriasis biologics, with PASI90 as the response criterion and NLRX1 as the model; g predictive line graph of the response or non-response to psoriasis biologics, with PASI90 as the response criterion and PASI90 and NLRX1 as the model. regulatory role and high individual cellular specificity [33]. In our study, NLRX1 was upregulated in both psoriasis and metabolic syndrome, probably due to the facilitative role of NLRX1 in cellular aerobic glycolysis. NLRX1 can attenuate the activity of respiratory chain complexes to maintain high glycolytic efficiency [34], whereas increased aerobic glycolysis can promote cell proliferation and inflammatory responses, causing psoriasis and metabolic diseases [35,36]. Compared to WT mice, NLRX1 mice showed increased cell proliferation and then apoptosis [37]. In addition, overexpression of NLRX1 activated ROS production, which amplified the JNK pathway [38]. In turn, the JNK pathway contributed to the development of inflammatory skin diseases such as psoriasis by increasing the expression of inflammatory chemokines such as CCL20 and inflammatory cytokines such as TNF-α and IL-23 [39,40]. In contrast, NLRX1 maintained cellular homeostasis after injury and suppressed excessive inflammatory responses by inhibiting NF-κB signaling pathway transactivation, promoting autophagy, and participating in mitochondrial innate immune regulation [33,41]. This explained why the expression of NLRX1 did not positively correlate with the PASI values but showed a threshold saturation effect.
In this study, a bioinformatics approach was used to screen for comorbid genes in psoriasis and metabolic syndrome, a gene set enrichment analysis was performed, and two machine learning algorithms were used to screen for comorbid signature crosstalk genes and perform external validation and clinical effect analyses. Further, we used CIBERSORT and ImmuCellAI for immune cell infiltration analysis to analyze the mechanism underlying psoriasis and metabolic syndrome comorbidity from genetic, pathway, cellular, and clinical aspects to suggest new targets for clinical treatment.
This study has some limitations. First, we did not validate the screened signature genes in clinical trials and ex-vivo experiments; however, we selected  Fisher's exact test 0.1781 microarray data from brand new psoriasis and metabolic syndrome datasets in the GEO database for testing, and the results showed that NLRX1 passed the validation test and had a level of clinical significance. Second, to extract PASI values and validate the predictive ability of signature genes for biological response rates, we selected microarray data from patients who used biologics. Patients using biologics are generally seriously ill and may not be representative of all patients with psoriasis. However, we selected a common psoriasis lesion microarray for the validation set, and the results showed that NLRX1 indeed has good diagnostic ability.