Landscape painting of cuproptosis related genes in ALS
10 cuproptosis related genes were analyzed based on package “limma”, as it was showed in a boxplot (Figure 1A) and a heatmap (Figure 1B), we found 4 ALS-positively related genes LIPT1, DLD, DLAT and PDHB were overexpressed in ALS group than non-ALS group which implies that cuproptosis related genes cause ALS mainly by activating the cuproptosis process and inducing the copper overload. Another gene circle diagram (Figure 1C) elucidates specific chromosomal loci where each cuproptosis relate genes is located.
Erection and superiority comparison of SVM and RF model
SVM and RF model were established as the premise to evaluate occurrence of ALS. We evaluated the applicability of the two models by residual analysis, boxplots of residual (Figure 2A) and reverse cumulative distribution of residual (Figure 2B) were presented as the visualization of consequences. Both diagrams suggest RF model fits better than SVM model when evaluating the occurrence of ALS due to a smaller residual (red dot stands for root mean square of residuals). Another ROC curve also elucidated the accuracy of this consequence (RF model suits better than SVM model) by showing the distribution area under the curve (RF = 1, SVM = 0.742) (Figure 2C). After the erection of RF model, we assessed four differentials expressed cuproptosis genes (LIPT1, DLD, DLAT, PDHB) by ranking them and verifying importance of each gene in ALS occurrence based on RF model. Result (Figure 2D) showed LIPT1 scored close to 100 whereas another three genes ranged between 60 to 80 which indicates LIPT1 may plays an important role in the pathogenesis of ALS and could be considered as a potential treatment target for ALS.
Speculating susceptibility of ALS by constructing Nomogram model
Nomograms are a way to express the interrelationships between variables in a forecasting model by integrating multiple predictors based on multi-factor regression analysis and then using scaled line segments that are drawn on the same plane at a certain scale and has been widely applied to predict the development of diseases[20-22]. In this research, we introduced Nomogram model and visualized it (Figure 3A) to foresee the prevalence and susceptibility of ALS patients. To validate reliability of our Nomogram, we furthermore constructed a calibration curve (Figure 3B), a decision curve (Figure 3C) and a clinical impact curve (Figure 3D) The calibration curve contains three lines tagged with “Apparent”,“Bias-corrected”and “Ideal”. Lines of “Bias-correct” and “Apparent” tend to coincide and converges to the “Ideal” line which signified credibility of our model. Based on decision curve analysis, we constructed decision curve. Here, decision curve was marked with red and was above the grey line and the black line when threshold probability increased from 0 to 1. The clinical impact curve also showed that our model exhibited ideal predicted power.
Cognition of two instinct cuproptosis pattern in ALS
By introducing consensus clustering method, we found cuproptosis function patterns. According to the fit of sample clustering, we found that the optimal sample clustering results when k = 2 (Figure 4A-C), thus, two instinct cuproptosis patterns were identified and they were labelled as coppercluster A (113 samples) and coppercluster B (120 samples). Subsequently, four cuproptosis candidates’ expression were compared between these two clusters. Heatmap (Figure 4D) and boxplot (Figure 4E) were plotted to elucidate those candidates were highly expressed in cluster B and depressed in cluster A. By using PCA (Figure 4F), we found that the four candidates do distinguish two different patterns. (P < 0.05 was seen as significant evidence, * : p < 0.05, ** : p < 0.01, *** : p < 0.001).
Analysis and prediction of biological functions of intersecting genes
After screening out the intersecting genes between coppercluster A and cluster B, we analyzed their potential biofunctions via KEGG and GO analysis. Results of KEGG showed that intersecting genes functioned as involving in Fc gamma R - mediated phagocytosis and Platelet activation which both diagrams were visualized as a boxplot (Figure 5A) and a bubble plot (Figure 5B). Consequences of GO analysis was demonstrated with a boxplot (Figure 5C) and a bubble plot (Figure 5D) showed that in cellar component (CC) section, genes mainly participated in focal adhesion and cell – substrate junction. In molecular function (MF), intersecting genes mainly interacts with structural constituent of cytoskeleton, protein N – terminus binding. In biological process parts (BP), genes mainly take part in cognitive and memory functions of the brain, response to IFNγ,receptor – mediated endocytosis, positive regulation of protein – containing complex assembly.
Recognition of differential expressed genes between two cuproptosis patterns
Since we identified two distinct cuproptosis functioning patterns and acquired intersecting genes, we subsequently sought the DEGs between two clusters and based on consensus clustering method and the distribution of differential genes in the samples, we re-clustered ALS samples and identified two types of gene clusters delineated by differential genes. We found that best sample clustering fit when k is equal to 2 (Figure 6A-C). This can be interpreted as the gene clusters distinguished by differential genes can be divided into two clusters (in this case, gene cluster A and gene cluster B). By constructing heatmap, we found that DEGs were mainly high expressed in gene cluster B whereas depressed in gene cluster A (Figure 6D) which suggests that cluster A may be associated with the inhibition of cuproptosis, while cluster B may be associated with the activation of cuproptosis. To identify this conjecture, we explored the relative expression of the four cuproptosis candidates between gene cluster A and B, interestingly, as we suspected, four cuproptosis candidates were indeed down-regulated in cluster A and up-regulated in cluster B (Figure 6E).
Constructing copperscore as a quantification criterion
By using a principal component analysis method based on data dimensionality reduction, we established “copperscore” scoring criteria to quantify copperclusters and gene clusters. Comparison of coppercluster A and B (Figure 7A) showed that cluster B had higher score than A (p < 2.22e-16) whereas comparison of gene cluster A and B (Figure 7B) showed that cluster A had a higher score than B (p < 2.22e-16). Results revealed that in terms of ALS, cuproptosis related samples were correlated a higher copperscore. In addition, in order to show the correlation between copperclusters, gene clusters and copperscore more visually, we constructed a Sankey diagram (Figure 7C). The Sankey diagram shows the subordination relationships among samples and copperscore.
Immune cells infiltration characteristic recognition
As been reported[23] before, neuroinflammation takes part in process of ALS. Naturally, with the development of ALS, microglia are exposed to oxidative stress and release pro-inflammatory substances, resulting in the death of motor neuron[24, 25]. Thus, in this research we investigated potential mechanism between cuproptosis and immune cells infiltration. We first investigated the immune infiltration properties of four cuproptosis-associated genes in ALS by plotting heatmap (Figure 8A), interestingly, we found that T cell family, B cell family and dendritic family were activated in ALS whereas mononuclear phagocyte system and nature killer cell family were depressed in ALS. Subsequently, according to gene importance, we further explored correlation between LIPT1 and immune cells. Boxplot (Figure 8B) demonstrates that the high expression status LIPT1 can recruit more immune cells than lower expressed LIPT1. Next, we explored correlation between immune cell infiltration and copperclusters and gene clusters. Boxplot (Figure 8C) suggested that coppercluster B obtained more intense immune cell infiltration that coppercluster A. Another boxplot (Figure 8D) delineated that immune cell infiltration is more severe in A than in B.