An integrative genomic analysis of transcriptional profiles identifies characteristic genes and patterns in HIV-infected long-term non-progressors and elite controllers

Expression profiling studies including LTNPs or ECs were identified through Gene Expression Omnibus (GEO, http://​www.​ncbi.​nlm.​nih.​gov/​geo) using search term (“long term no progressor” OR “elite controller”). To ensure the relevant studies were not missed, search in Arrayexpress (http://​www.​ebi.​ac.​uk/​arrayexpress) was also performed. Nine microarray gene expression datasets, reporting expression data of LTNPs or ECs and normal patients were retrieved from public repositories. Three datasets (GSE24081, GSE6740 and GSE57730) profiling on the same version of microarray chips [Human Genome U133 Plus 2.0 Array (HG-U133_Plus_2) or Human Genome U133A ver2.0 (U133A)], met the inclusion criteria and were included in the integrated analysis to build a MADNT set. The characteristics of these datasets were listed in Table 1. Detailed clinical parameters of each patient involved were listed in Additional file 1: Table S1.

Microarray meta-analysis were carried out according to the guidelines described in [40]. Each datasets were log2 transformated and normalized by Agilent GeneSpring software (Version 11.5, Agilent, USA). Then, gene matching was done for all probes. When multiple probes matched the same gene symbol, the probe presented the greatest inner-quartile range (IQR) was selected to represent the target gene symbol. After matching all the probes to a common gene symbol, “MetaDE” R package [41] was exploited to merge the common gene symbols across multiple studies by p value combination using Fisher methods. Differentially expressed genes were selected with adjusted p value < 0.05, based on false discovery rate (FDR) by the Benjamini–Hochberg procedure and moderated t test.

Enrichment analysis for KEGG pathway and Gene Ontology terms were carried out by David online tool (https://​david.​ncifcrf.​gov). Gene set enrichment analysis (GSEA) [42] was carried out using GSEA version 3.0, downloaded from the Broad Institute (http://​www.​broadinstitute.​org/​gsea/​downloads.​jsp). Expression data sets and phenotype labels were created according to GSEA specifications. Gene set permutations were set to be done 1000 times for each analysis using the weighted enrichment statistic and signal to noise metric. Gene sets with FDR lower than 0.05 were considered significant.

Weighted gene coexpression network analysis (WGCNA) is a gene coexpression network-based approach [43, 44]. A gene co-expression network is defined as undirected, weighted gene network, in which the nodes represent expression profiles while edges represent pairwise correlation between gene expressions. Briefly, correlation coefficient Smn between characteristic gene m and gene n is calculated by their expression values between different samples using the formulation: Smn = |cor(mn). The correlation matrix was then transformed into an undirected network by raising the absolute value of each entry to a power β using 6 as correlation coefficient threshold. Genes were clustered into different modules using dynamic tree cutting method.

Protein–protein interactions (PPI) networks in each module were visualized by Cytoscape 3.6.0. The Network Analyzer examined the network for topological parameters, including degree, connectivity, betweenness and closeness.

To analyze the antiviral activity of several up-regulated genes, HIV infection assay was performed as previously described [45, 46]. Briefly, 200 ng pNL4-3-R-E-Luc vector and 150 ng pCMV-VSV-G packaging vector were transfected into the 293T cells together with 200 ng empty pCMV6-Entry vector, control gene (SLFN11, GADD45G) or pCMV6-Entry-CMPK1, ZNF395, METTL9, GADD45A, PKKAR2B, OAT, CHPT1, SPOCK2 and CBX7 in 12-well plate using Lipofectamine 2000 (Thermo Scientific) according to the manufacturer’s instructions. The old culture medium was replaced with fresh culture medium at 6 h post-transfection. Subsequently, the supernatants were collected at 48 h post-transfection and cleared by filtration with the 0.45 μm filter. Cell lysis were harvested for SDS-PAGE and western blotting with α-myc, p24 (Gag) and β-actin antibody, respectively. The amount of (infectious) virus particles in the supernatants was determined via one-cycle infection assay. For infection assays, 200 μl of 1 × 10⁵ SupT1 or Jurkat cells per well in 96 well plate were infected with supernatants collected from 293T cells. At 48 h post-infection, the SupT1 cells were lysed and luciferase activity was determined using a firefly Luciferase Assay System (Promega).

293T cells were maintained at 37 °C in high glucose Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% fatal bovine serum (Gibco). SupT1 and Jurkat cells were maintained at RPMI 1640 Medium with 10% fetal bovine serum.

We searched two public repositories for mRNA transcriptional profile related to viremic control using the search term” “elite controller” [OR] “long-term nonprogressors”. The inclusion criteria were set as (1) the microarray chips used in the study was from Human Genome U133A or Human Genome U133 plus 2 Array (Affemetrix Company); (2) sample source was CD4+, CD8+ T cells or whole blood. Three studies met the inclusion-criteria were selected for the meta-analysis study. Details about these datasets were outlined as in Table 1. The three series included 35 PPs, 34 ECs (except for two patients with extremely low viremia) and 5 VNPs. The following information was extracted from each data series including GEO number, author name, sample source, platform and the number of progressors and controllers.

To compile expression data for meta-analysis, Cel raw data was firstly preprocessed by GeneSpring 11.5 software (Agilent Technologies, Santa Clara, CA). The background subtraction, normalization, and log base 2 transformation of gene signals were carried out using Robust Multi-Array Analysis (RMA) summarization algorithm. Quality of each dataset was assessed in Genespring and expression data was retrieved from GeneSpring. MetaDE packages were exploited to combine the three datasets and differentially expressed genes (DEG) in NPs compared to normal chronic progressors. DEG p value in each study was calculated by moderated-t statistical analysis. Fisher combined probability test was chosen for meta-analysis statistical method. For statistics, Benjamini and Hochberg method was used for multiple testing correction. Genes with FDR less than 0.05 was accepted as DEGs. We found 30 upregulated DEG and 83 down-regulated DEGs in NPs compared with PPs (Fig. 1). Up-regulated genes included CMPK1, CD9, METTL9, EIF4A3, ZNF395 and so on, while down-regulated genes were overrepresented by CD38, LAG3, some Interferon-stimulated genes (ISGs) and inflammatory genes. Upregulated and down-regulated DEGs in the comparison of NPs and PPs were shown in Additional file 2: Table S2.

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis were performed for the functional evaluation of both up-regulated and down-regulated DEGs using DAVID (Database for Annotation, Visualization and Integrated Discovery) online tool. In regard to GO analysis, “type 1 interferon signaling pathway”, “defense response to virus”, “interferon-gamma mediated-signaling pathway”, “innate immune response” and “B cell signaling pathway” were enriched in down-regulated genes (Fig. 2a). In the pathway analysis, several anti-viral response pathways was enriched, including “Measles”, “Influenza A”, “Hepatitis B” and “Hepatitis C” (Fig. 2b). The analysis failed to enrich GO terms or KEGG pathways with FDR less than 0.05 in up-regulated genes. Unlike down-regulated genes, the biological meanings with up-regulated genes were sporadic and their relevance to AIDS disease progression was not immediately clear. Furthermore, night genes (CMPK1, METTL9, CHPT1, OAT, GADD45A, SPOCK2, CBX7, ZNF395 and PPKAR2B) were selected to validate the anti-HIV activity. 293T cells were transfected with NL4-3lucR-E- and VSV-G along with these genes, empty vector or control genes (GADD45G, SLFN11). 48 h post-transfection, the supernatant was harvested and used for infecting SupT1 and Jurkat cells. After another 48 h, luciferase activity in SupT1 or Jurkat cells was measured to assess the anti-HIV activity. CMPK1, GADD45A, PPKAR2B and CHPT1 suppressed HIV-1 production by more than twofold. METTL9, ZNF395 SPOCK2 and OAT exerted a moderate inhibitory effect on HIV-1 production. Notably, GADD45A inhibited HIV-1 production by more than eightfold in Jurkat cells (Fig. 2c). GADD45G, the expression of which varied little between NPs and PPs was used here as a negative control gene and showed no effect on HIV replication. SLFN11 which was reported to inhibit HIV-1 production in the previous studies was used here as a positive control [47]. To further investigate whether these up-regulated DEGs influences the expression of HIV gene, Gag expression in 293T cells was assessed by Western blotting. CHPT1, CMPK1, GADD45A, OAT, METTL9, PPKAR2B and ZNF395 significantly decreased the expression of HIV-1 Gag (Fig. 2d). The functional assay confirmed the reliability of our analysis results.

A pilot study reported that serum from LTNPs and ECs have distinct neutralizing capabilities and antibody-dependent cellular cytotoxicity (ADCC) activity [48]. To uncover the divergent transcription pattern between aviremic controllers and VNPs from different cell types, study subjects were grouped into three subsets aforementioned. Data of CD8+ T cells from two datasets were merged into a combined transcriptome by MetaDE package. 115 DEGs were identified in CD8+ T cells (Additional file 3: Table S3), which highly overlap with those identified in MADNT. Only 4 DEGs (OAS3, IFI44, IFI44L and EIF2AK2) were identified in CD4+ T cells of ECs (Fig. 2e), and no DEGs met the criteria (FDR < 0.05) in VNPs.

Taken together, statistical combination of the relevant datasets has created a meta-analysis derived nonprogressor transcriptome (MADNT) and different subgroups, which can be subject to downstream GSEA and WGCNA analysis.

Given the limited power to detect the transcriptional pattern in up-regulated genes in LTNPs and ECs, GSEA was carried out to investigate the intrinsic common features associated with virological or immunologic control. Rather than setting a cutoff value for single DE genes, GSEA evaluates the whole dataset at biological pathway level by performing unbiased global search for predefined gene sets. MADNT and different subgroups mentioned above were interrogated for pathway enrichment using canonical pathway from MSigDb 2 GO-BP collection and also C2 (KEGG and Reactome) collection. Fourteen pathways were enriched with FDR less than 0.05 in NPs which included “KEGG_RIBOSOME”, “3′UTR mediated translational regulation”, “Nuclear transcribed mRNA catabolic process nonsense-mediated decay“, ”SRP-dependent cotranslational protein targeting to membrane” and “translational initiation” (Fig. 3a). Besides, we noticed that KEGG pathway “Graft verse host disease” was also enriched at moderately significant level (FDR around 0.19 and nominal p < 0.03), which included HLA-DQB1, HLA-DQA1, HLA-DOB, HLA-DMA and HLA-DOA. More than 30 pathways were enriched in progressors, which included “responses to type I interferon”, “mitotic cell cycle”, “defense response to virus”, “innate immune response” and “regulation of interleukin_1 beta production”. Furthermore, to validate the reliability of these datasets from Affemetrix microarray, we selected another dataset (GSE87620) published recently, using illumina Human HT-12 V4 microarrays. Using GSE87620 as a validating set, we selected the up-regulated genes with FDR < 0.05 using online tool GEO2R and performed Gene Ontology annotation (Fig. 3b). The enriched pathways were highly consistent with the results from MADNT. “translational initiation”, “SRP-dependent cotranslational protein targeting to membrane” and “nuclear-transcribed-mRNA catabolic process, nonsense-mRNA decay” was enriched in NPs.

In the aspect of stratification subgroup, 6 pathways were enriched with FDR less than 0.25 in CD4+ T cells of ECs which included “multi organism metabolic process”, “translational initiation”, “Nuclear transcribed mRNA catabolic process nonsense-mediated decay”, “protein targeting to membrane”; Similarly, 32 pathways were enriched with FDR less than 0.01 in CD8+ T cells of ECs, most of which were associated with pathways related to translation and protein-targeting to membrane (Additional file 4: Table S4).

WGCNA provides insights to disease pathogenesis by studying co-expression genes between clinical samples based on gene expression correlation coefficients. The reliability of this algorithm depends on a large sample size, which we are able to apply the MADNT and two subgroups (CD8+ T cells (EC) and VNP) expression set into WGCNA analysis.

Using WGCNA, 15 distinct modules was found (Fig. 4a). The expression data from different genes within each calculated module was used to determine the module epigengenes (ME, the first component of the respectively module), which was correlated to the clinical parameter (AIDS progression). Several modules showed significant positive correlation with lose of viremic control, including the tan modules (p = 4e−07, correlation coefficient = 0.55) and purple modules (p = 0.001, correlation coefficient = 0.37) and grey60 module (p = 0.003, correlation coefficient = 0.34) (Fig. 4b). However, no modules were obviously correlated to viremic control. For the three modules related to disease progression, 39 characteristic genes were subject to functional annotation and pathway enrichment. They were totally enriched in 19 GO terms. The genes were largely related to type I interferon signaling pathway (IFIT1, IFIT3, IFI6, OAS1, OAS2, MX1), complement activation (IGHA1, IGHD, IGHG1, IGHM, IGKV1-17, IGKV1-37, IGKV1-39) and positive regulation of B cell activation (IGHA, IGHD, IGKC, IGLC1). Additionally, PPI networks of the three modules were visualized by Cytoscape 3.6 (Fig. 4c). RRM2, IGLC2 and RSAD2 were the hub genes of grey60, purple and tan modules, respectively.

As to the stratification subgroup, WGCNA approach was applied to CD8+ T cells of ECs and VNPs, respectively. We identified 24 distinct coexpression modules in CD8+ T cells containing 28 to 1174 genes per module. Salmon module showed a strong positive correlation with AIDS progression including CD38 and ISGs which is consistent with DEG and GSEA analysis. Magenta and purple module (Additional file 5: Table S5) showed a positive correlation with viral control in ECs including gene involved in regulation of cellular response to stress (DNAJA1, HSPA5, PIK3R1 and PMAIP1), leukocyte migration (SCL7A5, YES1, B4GALT1, NRAS, PDE4B, PIK3R1 and PODXL) and response to cytokines (REL, SKIL, MAPKAPK2 and NFKB2). Interestingly, two modules were significantly associated with immune control in VNPs (darkturquoise, ME correlation = 0.87, p-value = 2e−4; red, ME correlation = 0.77, p-value = 0.004) (Fig. 5a). ZNF395, PDCD4-AS1 and EIF4A3 that are among the 30 up-regulated genes were involved in the darkturquoise module, Notably, KIR2DS3 and KIR3DL3 which were reported to affect immune control of HIV by interacting with HLA-B molecules in a specific manner also appeared in the darkturquoise module. We failed to enrich GO or KEGG terms of these modules, probably indicating unknown important pathways or interactions between genes in these two modules accounting for the immune control in VNPs. PPI networks of the two modules were visualized by Cytoscape 3.6 (Fig. 5b, c).