Mediaboss Marketing

Identify shared differential genes

When conducting PCA analysis on the expression matrices of GSE33570 (Fig.2a) and GSE29315 (Fig.2d), we observed a clear two-sided distribution of samples in both the disease group and the control group. In the analysis of the GSE35570 dataset, a total of 1572 distinct genes were detected as being differentially expressed. These DEGs were categorized into 824 up-regulated genes and 748 down-regulated genes (Fig.2b). Similarly, we observed 423 DEGs in the GSE29315 dataset, including 271 up-regulated DEGs and 152 down-regulated DEGs (Fig.2e). Next, the GEGs of the two datasets are displayed heatmaps for both datasets (Fig.2c,f). Furthermore, we employed a Venn diagram to identify the overlapping genes with the same directional trend, resulting in 64 genes being up-regulated (Fig.2g) and 37 genes being down-regulated (Fig.2h).

Differential expression gene analysis, function enrichment analysis and pathway enrichment analysis. (a) The PCA plot of GSE35570. (b, c) The Volcano plot and heatmap of DEGs in GSE33570. (d) The PCA plot of GSE29315. (e, f) The Volcano plot and heatmap of DEGs in GSE29315. (g) Venn plot of the up-regulated DEGs. (h) Venn plot of the down-regulated DEGs. (i) The KEGG enrichment analyses of DEGs. (j) The GO enrichment analyses of DEGs.

In order to enhance our comprehension of the fundamental biological functions linked to the 101 DEGs, an assessment of GO and KEGG enrichment was conducted using the clusterProfiler software package in R. An analysis of GO highlighted that these shared genes were mainly enriched in leukocyte mediated immunity, myeloid leukocyte activation, and antigen processing and presentation (Fig.2j). Additionally, the DEGs exhibited significant enrichment across the top five KEGG pathways, including Tuberculosis, Phagosome, Viral myocarditis, Inflammatory bowel disease, and Th1 and Th2 cell differentiation (Fig.2i). Apparently, the functions of differentially expressed genes are closely associated with the immune function of the body. The core genes primarily serve the purpose of activating immune cells.

To carry out the PPI analysis, we utilized the STRING online tool and visualized the outcomes using the Cytoscape software (Supplementary Fig. S1a). The PPI network showed 68 nodes and 498 edges. The DC value of each node was calculated, with a median value of 11. Based on this, we identified 17 hub genes of PPI network: TYROBP, ITGB2, STAT1, HLA-DRA, C1QB, MMP9, FCER1G, IL10RA, LCP2, LY86, CD53, CD14, CD163, HCK, MNDA, HLA-DPA1, and ALOX5AP. Subsequently, we employed the MCODE plug-in to identify six modules (Supplementary Fig. S1b,c), which included a total of 29 common DEGs. These DEGs were LCP2, TYROBP, CD53, LY86, ITGB2, FCER1G, MNDA, C1QB, HCK, IL10RA, HLA-DRA, ALOX5AP, MT1G, MT1F, MT1E, MT1X, ISG15, IFIT3, PSMB9, GBP2, CD14, CD163, VSIG4, CAV1, TIMP1, S100A4, SDC2, FGFR2, and STAT1. The most important module comprises 12 genes (LCP2, TYROBP, CD53, LY86, ITGB2, FCER1G, MNDA, C1QB, HCK, IL10RA, HLA-DRA, ALOX5AP), which were further analyzed using the ClueGO plug-in in Cytoscape software. The investigation revealed that these genes primarily function in activating neutrophils to participate in the immune response and activating innate immunity (Supplementary Fig. S1d).

In this study, we analyzed a total of 26 genes from six modules extracted from MCODE. To determine the importance of each gene, we employed the RF algorithm in two datasets, namely GSE35570 (Fig.3a) and GSE29315 (Fig.3b). By comparing the rankings of gene importance in both datasets, we identified the top eight genes that were consistently ranked highly. To visualize this overlap, we created a Venn diagram (Fig.3c), which revealed three genes (CD53, FCER1G and TYROBP) that were shared between the two datasets. Remarkably, these three genes overlap with the hub genes identified through the PPI analysis based on DC values, as well as the genes found in the most significant module. These three genes showed promising diagnostic potential for HT and PTC. To evaluate the diagnostic value of the common hub genes, we computed the Cutoff Value, sensitivity, specificity, AUC and 95% CI for each gene in the four datasets (Table 1). In the GSE35570 dataset (Fig.3d), the AUC values were as follows: CD53 (AUC 0.71, 95% CI 0.610.82), FCER1G (AUC 0.81, 95% CI 0.730.89), and TYROBP (AUC 0.79, 95% CI 0.710.88). In the GSE29315 dataset (Fig.3e), the AUC values were as follows: CD53 (AUC 1.00, 95% CI 1.001.00), FCER1G (AUC 1.00, 95% CI 1.001.00) and TYROBP (AUC 1.00, 95% CI 1.001.00). In the TCGA dataset (Fig.3f), we validated the diagnostic value of the common hub genes for PTC. The AUC values were as follows: CD53 (AUC 0.71 95% CI 0.610.82), FCER1G (AUC 0.74, 95% CI 0.640.89) and TYROBP (AUC 0.80, 95% CI 0.700.89). To further evaluate the diagnostic value of the common hub genes for PTC in HT, we computed the AUC and 95% CI for each gene using GSE1398198. In the GSE138198 dataset (Fig.3g), the AUC values were as follows: CD53 (AUC 0.83, 95%CI 0.571.00), FCER1G (AUC 0.92, 95% CI 0.721.00) and TYROBP (AUC 1.00, 95% CI 1.001.00). We also analyzed the difference box plots between the two groups in the four datasets (Supplementary Fig. S2). Our analysis using box plots revealed a noteworthy disparity in gene expression between the HT group and the control group in GSE29315. This disparity serves as an explanation for the AUC values of the three hub genes in GSE29315, all of which were observed to be 1.

Screening of hub genes and the diagnostic value of hub genes. (a) The rankings of gene importance in GSE35570. (b) The rankings of gene importance in GSE29315. (c) Venn plot of the top eight genes in GSE35570 and GSE29315. (d) Diagnostic value of hub genes in the GSE35570. (e) Diagnostic value of hub genes in the GSE29315, (f) Diagnostic value of hub genes in the TCGA. (g) Diagnostic value of hub genes in the GSE138198.

By using the GSE35570 dataset, we developed three diagnostic model specifically for PTC, incorporating these pivotal genes that were identified through our analysis. The ANN model (Fig.4a) had 4 hidden units, a penalty of 0.0108, and was trained for 537 epochs. The ANN model achieved an AUC of 0.94 (95% CI 0.910.98) in the training set, while in the test set, the AUC was 0.94 (95% CI 0.831.00) (Fig.4b). The XGBoost model had 8 mtry, 6 min_n, 3 max_depth, 0.001 learn_rate, and 0.07 loss_reduction and 0.97 sample_size. The XGBoost model achieved an AUC of 0.84 (95% CI 0.750.93) in the training set, while in the test set, the AUC was 0.62 (95% CI 0.420.83) (Supplementary Fig. S3a). The DT model had 0.0003 cost_complexity, 5 tree_depth and 6 min_n. The DT model achieved an AUC of 0.93 (95% CI 0.900.97) in the training set, while in the test set, the AUC was 0.83 (95% CI 0.651.00) (Supplementary Fig. S3b). Supplementary Table S1 displays the predictive performance of three machine learning models. The results indicate that the ANN model outperformed the other models, leading us to choose the ANN model for further analysis. TCGA dataset as external validation dataset was utilized to assess the diagnostic performance of the ANN model for PTC, yielding an AUC value of 0.77 (95% CI 0.660.87) (Fig.4c). The GSE138198 dataset was used to evaluate the ANN models diagnostic efficacy for PTC in HT. In the GSE138198 dataset (Fig.4d), the ANN model demonstrated a perfect AUC of 1.00 (95% CI 1.001.00). To provide clinicians with a better understanding of variable contributions, we utilized the SHAP algorithm to interpret the ANN prediction results. Figure4e, f, g illustrated how the attributed importance of features changed as their values varied. Our findings reveal that CD53 had the most significant impact on the output of the ANN model. Initially, it was positively associated with the risk of PTC and then became negatively correlated after a turning point of approximately 6. TYROBP and FCER1G showed a positive correlation with the occurrence of PTC.

ANN model construction and feature importance analysis. (a) The ANN was constructed based on the shared hub genes. (b) Diagnostic value of the ANN model in the GSE35570. (c) Diagnostic value of the ANN model in the TCGA. (d) Diagnostic value of the ANN model in the GSE138198. (e) A score calculated by SHAP was used for each input feature. (f, g) Distribution of the impact of each feature on the full model output estimated using the SHAP values.

We analyzed the protein expression of the hub genes based on the HPA database (Supplementary Fig. S4). CD53 was highly expressed in both tumor and normal tissues, while FCER1G and TYROBP showed higher expression in tumors compared to normal tissues. Furthermore, IF staining was performed to measure the expressions of CD53, FCER1G, and TYROBP in our clinical samples, including 10 HT-related PTC tissues and 6 NAT. By performing IF analysis (Fig.5), we obtained semi-quantitative results indicating significantly elevated fluorescence signal intensities for CD53, FCER1G, and TYROBP in the HT-related PTC group, as compared to the NAT group (P<0.05).

Microscopy scan of IF staining showed the distribution of CD53(green), FCER1G(green), and TYROBP(green), in HT-related PTC tissues and normal tissues adjacent to the tumour (NAT); as well as diagnostic value of CD53, FCER1G and TYROBP. MFI: Mean Fluorescence Intensity.

Considering the important roles of immune and inflammatory responses in the development of HT and PTC, we analyzed the differences in immune cell infiltration patterns between PTC, HT and normal samples using the CIBERSORT algorithm. By utilizing the GSE35570 dataset, we identified 12 immune subgroups that exhibited significant variations between PTC and normal samples (Supplementary Fig. S5a). Additionally, the analysis of the GSE29315 dataset revealed 5 immune subgroups that were significantly different between HT and normal samples (Supplementary Fig. S5b). Among these, 4 common immune subpopulations were found to be significantly higher in both PTC and HT samples compared to normal samples. These subpopulations included T cells CD8, T cells CD4 memory resting, macrophages M1 and mast cells resting. Additionally, we conducted spearman correlation analysis between hub genes and immune cells (Supplementary Fig. S5c,d). The results suggested that immune responses could potentially contribute to the involvement of hub genes in PTC and HT progression. IF staining was utilized to identify immune cell infiltration in 5 cases of PTC in HT tissues and 5 cases of NAT (Fig.6). The expression levels of CD4+T-cell marker Cd4, CD8+T-cell marker Cd8, and macrophage marker Cd86 were found to be significantly higher in the PTC in HT group compared to the NAT group. The IF staining results provided some extent of verification for the accuracy of the immune infiltration analysis results.

Microscopy scan of IF staining showed the distribution of Cd4(green), Cd8(green), and Cd86(green), in HT-related PTC tissues and normal tissues adjacent to the tumour (NAT). MFI: Mean Fluorescence Intensity.

Based on the three core genes screened in the RF algorithm, we conducted a search in the DGIdb database for relevant potential drugs. The results showed that only FCER1G had relevant drugs, while no relevant drugs were found for CD53 and TYROBP. FCER1G was predicted to have two potential drugs: benzylpenicilloyl polylysine and aspirin. Among these, benzylpenicilloyl polylysine had the highest score of 29.49, while aspirin had a score of only 1.26. We hypothesise that benzylpenicilloyl polylysine and aspirin may be effective in the treatment of HT and PTC and may prevent HT carcinogenesis.

See original here:
Identification and validation of potential common biomarkers for papillary thyroid carcinoma and Hashimoto's thyroiditis ... - Nature.com