Identification of stage-related and severity-related biomarkers and exploration of immune landscape for Dengue by comprehensive analyses

Gene expression datasets of Dengue patients were selected from a public database named the Gene Expression Omnibus databases (GEO, http://​www.​ncbi.​nlm.​nih.​gov/​geo). We followed these reference points: 1. datasets analyzed whole blood samples of different stages from Dengue patients including DF and DHF, and normal samples; 2. Datasets should include at least 20 dengue patients. Based on these criteria, we obtained GSE43777 dataset analyzing Two chip types (affymetrix HG-U133 plus 2 in GLP570 platform and HG-Focus in GLP201 platform) [41], GSE28405 dataset [42] and GSE51808 dataset [43].

In GSE43777, more than 200 samples (one sample for each stage) were collected from 51 DF and 13 DHF subjects in Maracay, Venezuela, and demographics and clinical, immunological and hematological characteristics of participants were preciously summarized [41]. In GSE28405, 31 clinically undifferentiated DENV RT-PCR positive patient samples were selected in Singapore within 72 h (Early Acute stage, EA), 4–7 days (Late Acute stage, LA) and 3–4 weeks (Convalescence stage, C) after self-reported fever onset, and other clinical information had been recorded as preciously described [42]. In GSE51808, whole blood samples were obtained from 47 Dengue patients (DF n = 31, DHF = 16) hospitalized at the Siriraj Hospital in Bangkok within days 2 and at 4 weeks or later after discharge (the Convalescence stage) and 9 normal subjects were analyzed， and detail clinical information were also showed [44].

Regarding gene expression levels in C or EA stages as baselines, these patients were divided into 3 groups, C vs EA, C vs LA and EA vs LA, as precious studies [41]. The Principal Component Analysis (PCA) analysis [45] was used to explore whether different stages can be distinguished clearly.

A website (http://​sangerbox.​com/​Tool) based on a R package “Limma” was applied to analyze DEGs among three phases (we showed the analysis regardless of serotypes and also performed separate analysis for each serotype) and between DHF and DF for Dengue patients. A R package “clusterProfiler” in R version 4.1.0 and a gene set enrichment analysis (GSEA) software (4.1.0) [46] were used to performed Gene Ontology (GO) enrichment analysis, Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis and GSEA to explore potential bio-function and enrichment pathways for DEGs.

We obtained a total of 4843 genes in the C vs EA group, 4472 genes in the C vs the LA group and 4463 genes in the EA vs LA group by a “WGCNA” package. We converted the adjacency matrix into the topological overlap matrix (TOM) when the power of β were equal to 5, 4 and 12. According to a height cutoff of 0.25, we merged similar modules and the module displaying the highest conformity with the development disease of Dengue was used to intersect with DEGs to identify more stable DEGs.

we estimated fractions of immune cells in whole blood samples by the “CIBERSORT” website (https://​cibersortx.​stanford.​edu/​index.​php) which can be used to input gene expression data and then obtain a fraction estimate of 22 immune cell types in whole blood samples [48].

ROC [49] was applied to identify, valid and test value of biomarkers and immune cells in distinguish stages and severity based on Area Under The Curve (AUC).

The Principal Component Analysis (PCA) shows three stage samples can be distinguished (Additional file 1: Figure S1 A–C). After normalizing the expression data from GSE43777 dataset on GLP201 platform, we identify 147 DEGs (121 up-regulated genes and 26 down-regulated genes) between the C and the EA stages (Fig. 2A, F and Additional file 7: Table S1), 124 DEGs (102 up-regulated genes and 22 down-regulated genes) between the C and the LA stages (Fig. 2B, G and Additional file 8: Table S2) and 195 DEGs (74 up-regulated genes and 121 down-regulated genes) between the EA and LA stages (Fig. 2C, H and Additional file 9: Table S3) by a “Limma” package based on |Log2FC|≥ 1, and adjusted P value < 0.05. Likewise, in the EA stage we obtained 30 DEGs (16 up-regulated genes and 14 down-regulated genes) (Fig. 2D, I and Additional file 10: Table S4) between DF (n = 15) and DHF (n = 11) samples from GSE43777 dataset on GLP570 platform, and 58 DEGs (48 up-regulated genes and 10 down-regulated genes) (Fig. 2E, J and Additional file 11: Table S5) between DF (n = 25) and DHF (n = 26) samples in the LA stage, based on |Log2FC|≥ 1 and P value < 0.05.

GO enrichment analysis results show that 147 DEGs (between the C and the EA stages) are mainly enriched in defense response to viruses and type I interferon (Fig. 2K); DNA replication is a common bio-activity for 124 DEGs (between the C and the LA stages) (Fig. 2L); the Fig. 2M shows that neutrophil activation and defense response to virus are dominating enrichment biofunctions for 195 DEGs between the EA and LA stages; bio-function involved in negative regulation of viral genome replication are more active for 30 DEGs between DF and DHF samples in Fig. 2N; neutrophil and humoral immune response are activated for DHF samples in the LA stage (Fig. 2O).

GO enrichment analysis results (down-regulated genes in DHF samples enriched in negative regulation of viral replication) in Fig. 2I and N caught our attention, because it suggested it was difficult to control viral replication in DHF patients in the EA stage. Relative studies [21, 24, 50] demonstrated DENV inhibited cell apoptosis through autophagy, thereby promoting its replication. Therefore, we suspecte related-autophagy genes may be activated in EA stage and LA stage. As the Additional file 1: Figure S1 J shown, in the LA stage, we identify definitely 1 different expression autophagy-related gene (CCL2) by intersecting DEGs between DHF and DH samples from GSE43777 analyzed on GLP570 platform with 222 autophagy-related genes downloaded from Human Autophagy Database (HADb) which provides a complete and an up-to-date list of human genes and proteins involved directly or indirectly in autophagy. The boxplot diagram (Additional file 1: Figure S1 K) shows that the expression levels of CCL2 increases significantly in DHF samples in the LA stage, suggesting that DENV replicates more strongly in DHF than DF which may be one reason resulting in DHF in the LA stage. While we do not identify different expression of autophagy-related genes between DF and DHF in the EA stage.

KEGG analysis results reveal that pathogen infection and NOD − like receptor signaling pathway are obvious enrichment pathways for 147 DEGs (between the C and the EA stages) (Fig. 2P); Cell replication, Legionellosis and Malaria, are major enrichment pathways for 124 DEGs (between the C and the LA stages) (Fig. 2Q); 195 DEGs (between the EA and LA stages) are enriched in pathogen infection and Cell replication (Fig. 2R); NOD − like receptor signaling pathway and pathogen infection are more active for 30 DEGs between DF (n = 15) and DHF (n = 11) samples in Fig. 2S; inflammatory pathways are activated for DHF samples in LA stage in Fig. 2T.

Using the GSE43777 dataset analyzed on GLP201 platform, we performed a “WGCNA” package to obtain key modules associated with procession features of Dengue. We select 5, 4 and 12 (Fig. 3A–C) as the soft-threshold power respectively and 0.25 (Additional file 1: Figure S1 D-F) as the cutoff height in the C vs EA group, in the C vs LA group and in the EA vs LA group. Modules consist of genes with similar expression properties (Additional file 1: Figure S1 G–I). In the cluster tree, we use each branch to represent a gene, and one color to show a co-expression module (Fig. 3D–F).

The association of module eigengene (ME) values and clinical stages is applied to assess the module-stage feature correlation. WGCNA results show 14, 12 and 8 modules, respectively (Fig. 3G–I). Compared the C with the EA group, the turquoise module is the most matched strongly with the EA group (Pearson co-efficient = − 0.89, P = 1e−33) and the C group (Pearson co-efficient = 0.89, P = 1e−33) in Fig. 3G. Compared the C with the LA group, the most significant correlation can be observed between the LA group (Pearson co-efficient = 0.71, P = 5e−20) and the blue module, and between the C group (Pearson co-efficient = − 0.71, P = 5e−20) and the blue module in Fig. 3H. While comparing the LA group with the EA group, the black module and the green module show the same highest correlation (Pearson co-efficient = 0.72) in Fig. 3I. We regard these modules mentioned above as key modules.

To filter reliable and strong hub genes (shared genes) accurately, Venn diagrams were used to visualize hub genes between DEGs and key modules by a website (http://​www.​ehbio.​com/​test/​venn/​#/​) [51]. The results illustrate 143, 99 and 187 hub genes by interacting the turquoise module with 147 DEGs (between the C and the EA stages) (Fig. 4A), interacting the blue module with 124 DEGs (between the C and the LA stages) (Fig. 4B), interacting the black and green modules with 195 DEGs (between the EA and LA stages) (Fig. 4C), respectively. CD38 and CDKN1C are shared by the three phases (Fig. 4D). When considering serotypes, CD38 and CDKN1C still express differentially in three comparison groups (C vs EA, C vs LA, EA vs LA) for each serotype and log|FC|> 1 or log|FC| is extremely closed to 1 (Additional file 12: Table S6). The Fig. 4G shows the expression levels of CD38 firstly increase and then decrease from day 0 to the C stage (***P < 0.001). While the expression of CDKN1C is firstly down-regulated and subsequently up-regulated from the day 0 to the C stage (***P < 0.001) (Fig. 4G). The GSEA shows CD38 is enriched in cellar proliferation and metabolic pathways (Fig. 4E) and CDKN1C is enriched in lipid metabolism and inflammation pathways (Fig. 4F). Correlation analyses show in Fig. 4H expression levels of CD38 are negatively correlated with CDKN1C (spearman correlation = − 0.5). We select CD38 and CDKN1C for follow-up analyses.

Because CD38 and CDKN1C express differently in three stages (Fig. 5A–B), we speculate that they can be used as biomarkers with staging characteristic to distinguish clinical stages for Dengue patients. To identify, verify and test the value of CD38 and CDKN1C, we regard GSE43777 datasets analyzed on the GPL201 platform (C stage, n = 48; EA stage, n = 47; LA stage, n = 73), GSE43777 datasets analyzed on the GPL570 platform (C stage, n = 24; EA stage, n = 26; LA stage, n = 51) and GSE28405 (C stage, n = 31; EA stage, n = 57; LA stage, n = 31) as a training dataset, a verifying dataset and a testing dataset, respectively. When analyzing single serotype, we regard GSE43777 datasets analyzed on the GPL201 platform (Serotype I: C stage, n = 26; EA stage, n = 26; LA stage, n = 44. Serotype II: C stage, n = 6; EA stage, n = 7; LA stage, n = 5. serotype III: C stage, n = 9; EA stage, n = 9; LA stage, n = 15. Serotype IV: C stage, n = 6; EA stage, n = 4; LA stage, n = 8.), GSE43777 datasets analyzed on the GPL570 platform (Serotype I: C stage, n = 7; EA stage, n = 3; LA stage, n = 14. Serotype II: C stage, n = 11; EA stage, n = 16; LA stage, n = 23. serotype III: C stage, n = 3; EA stage, n = 3; LA stage, n = 8. Serotype IV: C stage, n = 3; EA stage, n = 3; LA stage, n = 6.) as a training dataset and verifying dataset, respectively.

In the training group, CD38 has 0.960, 0.976, and 0.829 of AUC in three comparing groups (C vs EA, C vs LA, EA vs LA) respectively and CDKN1C has 0.852, 0.868 and 0.955 of AUC in three comparing groups (C vs EA, C vs LA, EA vs LA) respectively (Fig. 5D–F). In the verifying group, the staging diagnostic value of CD38 and CDKN1C in staging for Dengue is as follows (Fig. 5G–I): CD38 (AUC: 0.897, C vs EA; AUC: 0.957, C vs LA; AUC: 0.827, EA vs LA), CDKN1C (AUC: 0.575, C vs EA; AUC: 0.929 C vs LA; AUC: 0.876, EA vs LA). Figure 5J–L show high diagnostic values in the testing dataset: CD38 (AUC: 0.822, C vs EA; AUC: 0.996 C vs LA; AUC: 0.926, EA vs LA), CDKN1C (AUC: 0.744, C vs EA; AUC: 0.724 C vs LA; AUC: 0.842, EA vs LA). The analytical results above show that CD38 can distinguish admirably different phases for Dengue patients and has a higher distinguishable value than CDKN1C. Although we consider serotypes (serotype I–IV), results are similar among three comparison groups (C vs EA, C vs LA, EA vs LA) (Additional file 2: Figure S2, Additional file 3: Figure S3).

In addition, as Fig. 5C shown, CD38 distinguishes perfectly Dengue samples from normal samples with an AUC of 100% in GSE51808 dataset (9 normal samples and 28 Dengue samples). Therefore, CD38 shows the high value in distinguishing stages for Dengue patients. We selected CD38 as a biomarker for staging diagnosis of Dengue.

There is no significant difference in expression levels of CD38 and CDKN1C between DH and DHF samples (Additional file 1: Figure S1 L), and different serotypes of Dengue also do not show obvious difference in expression levels of CD38 and CDKN1C (Additional file 1: Figure S1 M).

After intersecting 30 DEGs between DF (n = 15) and DHF (n = 11) samples from GSE43777 dataset on GLP570 platform in the EA stage with 58 DEGs between DF (n = 25) and DHF (n = 26) samples in the LA stage, we identify 6 shared DEGs (LOC101928288, TCN1, DEFA4, FRG1B, LOC286087 and ZNF595) (Fig. 6A). We speculate that LOC101928288, TCN1, DEFA4, FRG1B, LOC286087 and ZNF595 can be used as biomarkers to discriminate DHF patients from DF patients due to different expression of them between DHF and DF. The GSE43777 dataset analyze on the GPL570 platform in the EA stage (DF = 15; DHF = 11) (Fig. 6B) and in the LA stage (DF = 25; DHF = 26) (Fig. 6C) are regarded as a training dataset; GSE43777 dataset analyzed on the GPL201 platform in Acute stage (DF = 40; DHF = 37) is regarded as a verifying dataset; GSE51808 (DF = 18; DHF = 10) and GSE18090 (DF = 8; DHF = 10) datasets are regarded as testing datasets. The ROC results (Fig. 6B–F) show ZNF595 always has acceptable diagnostic value in distinguishing DHF patients from DF patients. Therefore, ZNF595 is selected as a biomarker to predict DHF patients in Acute stages.

GO enrichment analyses for 58 DEGs between DHF samples and DF samples show neutrophil and humoral immune response are activated and KEGG pathway enrichment analyses show rich inflammatory pathways. Therefore, we analyze fraction changes for 22 types of immune cells during DENV infection by CIBERSORT. The GSE43777 dataset analyze on the GLP201 platform is used to explore fractions of 22 types of immune cells in whole blood samples from Dengue patients in three phases and the GSE43777 dataset analyzed on the GLP570 platform is applied to explore differently infiltrating-immune characteristic between DF and DHF samples.

As Fig. 7A–C and 7F shown, in the EA phase, fractions of activated dendritic cells, Neutrophils, Monocytes, and M1 Macrophages (P < 0.05) increase significantly compared to the LA and C phases. Compared the LA phase with the EA phase and the C phase, fractions of Plasma cells and activated memory CD4+ T cells (P < 0.05) increase clearly (Fig. 7A–C and F). While Compared with the EA phase and the LA phase, the increments in fractions of memory B cells, resting memory CD4+ T cells, resting dendritic cells, Eosinophils, and naïve B cells (P < 0.05) are more obvious in the C phase (Fig. 7A–C and F). Analytical results above mean that immune cells with antigen presentation, phagocytosis and chemotaxis, immune cells involved in humoral immunity and memory cells are more obviously active in the EA, LA and C phase, respectively. When analyzing in a single serotype, we can still get above results (Additional file 4: Figure S4).

In the LA stage, fractions of activated NK cells (P < 0.05) increase in DHF samples (Fig. 7E and H), but fractions of CD8+ T cells, gamma delta T cells, resting NK cells, M2 Macrophages and resting mast cells decrease (Fig. 7E and H) (P < 0.05), which imply immune response are damaged in DHF samples. In the EA stage, the immune response of DF patients is similar to that of DHF patients (Fig. 7D and G). The immune response of DHF patients in the LA stage is significantly impaired which explain rapid deterioration of DHF patients in the LA stage.

We analyze correlations between CD38 and immune cells in GSE43777 dataset (Fig. 7I). The infiltration level of Plasma cells, activated memory CD4+ T cells, gamma delta T cells and CD8+ T cells are positively related to CD38 (Fig. 7I); the infiltration level of naïve B cells, Monocytes, resting dendritic cells, resting memory CD4+ T cells, naïve CD4+ T cells, memory B cells, Eosinophils and activated mast cells are negatively related to CD38 (Fig. 7I). Therefore, the analytical results above show there is strong co-expression relationship between CD38 and plasma cells and between CD38 and activated memory CD4+ T cells (Pearson’s correlation > 0.5). In individual serotype, we can still draw this conclusion (Additional file 5: Figure S5). Correlation between immune cells and ZNF595 is not significant (Fig. 7J).

Because of high co-expression correlation between CD38 and plasma cells, we speculated that infiltrating-immune plasma cells could also show distinct differences in three stages. To explore this characteristic, we regarded GSE43777 analyzed by GPL201 as a training set and GSE43777 analyzed by GPL570 as a test set. As the Fig. 8A–C shown in the training group, the distinguishing value of Plasma cells, activated memory CD4+ T cells, and Monocytes in staging were good and Plasma cells have the highest distinguishing value (AUC:0.827, C vs EA; AUC:0.964, C vs LA; AUC: 0.832, EA vs LA). In the test group, the distinguishing value of three types of immune cells in stage for Dengue was still valuable and Plasma cells have the highest distinguishing value in Fig. 8D–F (AUC: 0.905, C vs EA; AUC: 0.949, C vs LA; AUC: 0.824, EA vs LA. Plasma cells (AUC = 0.968), activated memory CD4 + T cells (AUC = 0.845) and Monocytes (AUC = 0.869) can excellently distinguish Dengue samples from normal samples in GSE51808 (Fig. 8G). When considering different serotypes, above results were still obtained (Additional file 6: Figure S6). Therefore, we can discriminate three stages based on the fraction of Plasma cells, activated memory CD4 + T cells and Monocytes in Dengue patients.