Genetic and transcriptomic signatures of host control in HIV-1 infection

Abstract

Background

The interaction between HIV-1 and the host immune system plays a crucial role in the natural control and progression of the infection. Previous studies have identified APOBEC3G, Tetherin, SAMHD1, and SERINC5 as HIV-1 host restriction factors, which are counteracted by the viral proteins Vif, Vpu, and Nef, respectively. The blood expression levels of some of these host proteins are correlated with HIV load, suggesting that interindividual differences in the spontaneous control of HIV infection might lead to the identification of novel HIV restriction factors. Our study enrolled 150 participants from the Swiss HIV Cohort Study with human genome-wide genotyping data, pre-antiretroviral treatment peripheral blood mononuclear cells (PBMC) aliquots and HIV load measurements.

Results

Using BrB-seq, we quantified mRNA expression of all protein-coding genes and found significant associations between 792 genes and HIV load. Pathway analysis revealed that higher viral load associated with the upregulation of innate immune response, proteasome complex, mitochondrial and cell-cycle related activity, and with the downregulation of ribosomal transcripts and genes involved in cytokine-cytokine receptor interaction, including IL4R, IL7R, and TCF7. Mendelian Randomization confirmed the viral restriction activity of TRABD2A and identified new candidates as potential restriction factors.

Conclusions

These findings provide novel insights into the host–virus interplay and suggest additional genes such as TNFSF10, TTC3 and GRAP that may contribute to the natural control of HIV-1 infection.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12977-026-00679-4.

Background

Pathogen sensing is an essential first step in the host response to HIV-1 infection. This process provokes a cascade of immune reactions leading to the production of interferon and the consequent activation of interferon-responsive genes including restriction factors such as APOBEC3G, SAMHD1 and Tetherin, which function to limit HIV replication and dissemination.

Different molecular mechanisms are involved in viral restriction that target various steps of the viral replication cycle [1, 2]. Among the well-known restriction factors, APOBEC3G is a single-strand cytidine deaminase that interferes with reverse transcription by inducing numerous deoxycytidine to deoxyuridine mutations in the negative strand of the HIV DNA primarily expressed as cDNA [3]. SAMHD1 [4], a cellular enzyme, hydrolyses nucleotide triphosphates (NTPs) to triphosphate and nucleosides, thereby reducing the amount of NTPs necessary for viral cDNA synthesis. Tetherin (encoded by the BST2 gene) is an integral membrane protein that inhibits retrovirus infection by preventing the release of virus particles upon budding from infected cells [5]. These restriction factors are counteracted by HIV accessory proteins: Vifmaps target APOBEC3G for proteasomal degradation by recruiting a Cul1-E3 ligase. Vpu promotes Tetherin degradation by recruiting SCFs ligase complexes [6].

More recently, SERINC5 and SERINC3 have emerged as constitutively expressed restriction factors that do not require activation by interferon [7, 8]. Predominantly expressed in lymphocytes and monocytes, these proteins localize to the plasma membrane and act as a primary barrier against retroviral infection. We and others demonstrated that SERINC3 and SERINC5 are antagonized by the HIV-1 protein Nef. When expressed, Nef promotes the relocalization of SERINC3 and SERINC5 from the membrane to endosomal compartments with the involvement of clathrin-mediated trafficking. In absence of Nef, SERINC3 and SERINC5 are incorporated into virions and substantially inhibit viral infectivity [9, 10]. While the exact restriction mechanism is not fully understood, the two proteins seem to act as lipid-flipping membrane transporters that disrupt viral membrane asymmetry, a function abrogated by Nef through interaction with intracellular loop 4 [10]. Importantly, despite the presence of Nef, it was shown that overexpressing SERINC5 consistently impairs wild-type HIV-1 [8] and the infectivity of other viruses [9], suggesting that inter-individual variations of expression of restriction factors (e.g., due to eQTLs) may confer a protective effect proportional to the respective expression levels.

Genome-wide association studies (GWAS) [11] have shown that host genetic variations have significant statistical associations with HIV-1 viral load (VL). Variants in the HLA region, especially HLA-B27 and HLA-B57, as well as CCR5Δ32, have been associated with better spontaneous control of HIV infection. A portion of the remaining variance may be a result of genetic variation in regulatory regions affecting immune gene expression, including that of restriction factors. Integration of transcriptomic data and GWAS findings provides a mean to uncover novel genes with antiviral capabilities and understand a link between variations in regulatory regions and HIV VL.

Building on these findings, we investigate here the transcriptomic landscape of individuals from the Swiss HIV Cohort Study (SHCS), stratified by set point VL in the absence of antiretroviral therapy. Our study is guided by the following hypotheses: (i) viral control, as estimated by set point VL, correlates with the expression of known restriction factors such as BST2, APOBEC genes, SERINC3/5 and others; (ii) additional, previously uncharacterized restriction factors may exhibit similar correlations with viral load; (iii) genetic variants (eQTLs) may upregulate restriction factors expression, thereby enhancing host defence; and (iv) eQTLs may similarly modulate the expression of other unknown restriction factors with potential impact on HIV-1 control.

Methods

Study participants, samples and data

The SHCS (www.shcs.ch) is an ongoing, nationally representative cohort study of PWH. Clinical, behavioral, and laboratory data are collected at study registration and biannually thereafter. All centers’ local ethical committees approved the cohort study, and all participants provided written informed consent, including for human genetic testing.

For this study, we selected 150 SHCS participants of European ancestry, who had genome-wide genotyping data from previous studies and a PBMC aliquot available before antiretroviral treatment. PWH were selected based on their VL at set point (75 individuals among the top quartile; 75 individuals among the bottom quartile of VL distribution of all SHCS participants). The following clinical and laboratory parameters were extracted from the SHCS database and included in the analysis as stratifying factors and/or covariates: sex at birth, age at time of PBMC sampling, presence/absence of HLA-B*27:05, HLA-B*57:01 and the CCR5 delta 32 deletion (rs333), setpoint VL, number of CD4 + T cells and VL at time of PBMC sampling.

DNA genotyping

Genotyping was performed in the context of previous SHCS projects using genotyping arrays from Illumina and DNA extracted from PBMCs. For each genotyping batch, samples and single nucleotide polymorphisms (SNPs) were removed if the missingness exceeded 10% or the minor allele frequency deviated more than 20% from the 1000 Genomes Project reference. Missing genotypes were phased with EAGLE2 [12] and imputed using positional Burrows-Wheeler transformation (PBWT) [12], at the Sanger Imputation Service, using the 1000 Genomes Project Phase 3 panel as reference. Only high-quality SNPs with an imputation information score (INFO > 0.8) were retained following the imputation, after which the genotyping batches were combined. Principal components and, separately, population structure together with the 1000 Genomes reference panel was calculated using PLINK2 (v2.00a3LM). KING (v2.1.3) was used to ensure that no duplicate or cryptically related samples were included in the final dataset.

RNA extraction and sequencing

PBMC samples (~ 10⁶ cells) were stored at -80 °C. They were thawed in a water bath at 37 °C under manual agitation and rapidly resuspended and washed in 10 ml warm PBS (37 °C). Tubes were centrifuged 5 min at 1100 rpm, PBS removed and a second washing step with warm PBS was performed. Pellets were resuspended immediately in 350 µl RLT plus with β-mercaptoethanol (1:100) for cells lysis and processed for RNA extraction according to manufacturer’s instructions (RNeasy Plus Mini Kit, Qiagen Cat.No./ID :74134). Elution volumes were 18 µl and final concentrations were mostly between 15 and 50 ng/µl. From this step, all manipulations were done on ice. RNA quality (RIN: RNA Integrity Number) was analyzed using a Tapestation instrument and the RNA ScreenTape system kit (Agilent, ref. 5067–5576 and 5067–5577). RNA quantification was performed using the Qubit™ system and the RNA Quantification high sensitivity kit (ref.: Q32852).

To perform RNA sequencing, we applied the BrB-seq protocol (Alithea Genomics), which enables the simultaneous generation of RNA-seq libraries through high sample multiplexing and quantification of gene expression from 3’cDNA libraries [13]. The libraries were sequenced with an average coverage of 5Mio reads at the Gene Expression core facility of the EPFL on Illumina sequencers.

Gene expression and eQTL analyses

The calculation of gene expression matrices was obtained through the Brb pipeline provided by Alithea Genomics (https://alitheagenomics.com/blog/brb-seq-pipeline-downstream-analysis-best-practices). Preprocessing of raw gene counts was done using Scanpy [14] toolkit and in-house developed python scripts, which perform normalization of counts, log-transform and batch correction. We used DAVID Functional Classification Tool [15] for gene ontology analysis, including clustering and annotation charts. We used STRING [16] with standard settings for gene interaction network analysis.

Expression quantitative trait loci (eQTL) summary statistics were obtained from the eQTLGen phase I consortium database and eQTL genetic variants used as genetic instruments for gene expression. Transcriptome-wide Mendelian randomization (TWMR) analysis was then performed on genes showing evidence of association between genetically predicted expression and VL, requiring at least one valid instrumental variable (one eQTL variant) per gene.

Mendelian randomization

Mendelian randomization (MR) was used to investigate potential causal relationships between gene expression and HIV-1 viral load (VL), leveraging human genetic variants as instrumental variables to proxy genetically determined differences in expression levels. This framework enables the assessment of host-driven effects on gene expression that are independent of transcriptional changes induced by infection itself. We applied several complementary MR approaches, including inverse-variance weighted (IVW) analysis together with robust and penalized variants, as well as Transcriptome-Wide Mendelian Randomization (TWMR). In all analyses, gene expression was modeled as the exposure, HIV-1 VL as the outcome, and single nucleotide polymorphisms (SNPs) as instrumental variables. To assess the robustness of these findings and account for potential violations of Mendelian randomization assumptions, including horizontal pleiotropy and unobserved confounding, complementary sensitivity analyses were conducted using MRmix, MR-Egger, MR-PRESSO, MR-RAPS, and inverse-variance weighted (IVW) methods, together with robust and penalized IVW variants. Results across methods were compared to evaluate consistency of causal inference.

Genetic associations with the outcome (VL) were derived from a single GWAS summary statistics dataset, whereas genetic associations with the exposure (gene expression) were obtained from the eQTLGen consortium dataset, using whole-blood eQTL estimates as the primary source.

For all MR methods except TWMR, instrumental variables consisted of independent eQTLs (linkage disequilibrium, LD, R² < 0.1) showing nominal evidence of association with both gene expression and VL (P < 0.05). For TWMR, the instrument selection procedure additionally accounted for shared regulation and potential pleiotropy across nearby genes. For each focal gene, we first selected independent eQTLs (R² < 0.1) associated with that gene, then identified additional genes regulated by the same variants irrespective of their association with the outcome. The instrument set was subsequently expanded by including all independent eQTLs affecting both the focal gene and these co-regulated genes. To minimize pleiotropic bias, only independent QTLs with lowest standard error from each LD block (R² > 0.1) were retained.

Results

Transcriptomic profiling

From the Swiss HIV Cohort Study (SHCS), we enrolled 150 untreated people of European ancestry with HIV (PWH) based on set-point viral load (VL), specifically selecting 75 individuals each from the highest and lowest quartiles of the VL distribution of all participants of SHCS (35 females, 115 males, mean age 39yo ranging from 24 to 68 – S1 Fig). Bulk RNA sequencing of peripheral blood mononuclear cells (PBMCs) was performed using the BrB-seq protocol [13] in two separate batches. Following initial quality control—retaining only samples with more than 6,000 expressed genes—a total of 144 samples were included in the mRNA expression analysis displaying between 1 and 15 million reads/sample and allowing the detection of 10’000 to 18’000 total genes/sample. (S2 Fig). Genes with no detectable expression across samples were excluded, resulting in 19,231 retained genes. Of these, 11,753 genes expressed in fewer than 10% of the samples were further filtered out, leaving 7,478 genes for downstream analyses. Raw gene counts were normalized to counts per million (CPM), log-transformed, and scaled.

Associations between gene expression and HIV-1 viral load

For each gene, we explored the univariate interaction effect between normalized expression level and HIV-1 VL using a Log-Normal Generalized Linear Model. We also added the following covariates—sex, presence/absence of the HLA-B*27:05 and HLA-B*57:01 alleles, presence/absence of the CCR5 delta 32 deletion (rs333) and the 10 first principal components of the genotype matrix according to the formula:

1

We calculated the effect sizes and their related p-values corrected with the Benjamini-Hochberg procedure for false discovery rate. The only covariate that was significantly associated with VL was sex (adjusted p-value = 0.02), as already observed [17], whereas a marginal contribution was provided by HLA-B*57:01 (0.058).

This analysis resulted in a list of 792 genes with significant associations between mRNA expression and HIV-1 VL. The top 20 positive and negative correlations are shown in Table 1 (full list in S1 Table). As an additional non-parametric measure of association, we calculated the Spearman correlation between gene expression and VL, identifying 586 of 792 genes with statistically significant correlations (FDR < 0.05).

Table 1.

Top 20 genes with the strongest association between mRNA expression and HIV-1 viral load

Gene	Beta	Padj	CPM		Gene	Beta	Padj	CPM
			For lower quartile of VL	For upper quartile of VL				For lower quartile of VL	For upper quartile of VL
IGHGP	2.38	0.0001	Mean=15.1	Mean=49.11	EEF1A1P1	− 4	0.00093	Mean=8.08	Mean=6.65
			Std=18.3	Std=65.79				Std=3.19	Std=3.6
TUBA1B	2.11	3.20E-11	Mean=1225.7	Mean=2606.63	EEF1A1P38	− 2.97	5.20E-05	Mean=3.66	Mean=2.57
			Std=760.1	Std=1421.73				Std=2.04	Std=1.73
COX8A	1.7	0.00032	Mean=531.33	Mean=751.11	EEF1B2	− 2.36	0.0001	Mean=1028.47	Mean=855.77
			Std=228.66	Std=294.2				Std=310.1	Std=389.6
TMSB4X	1.63	0.00332	Mean=9922.03	Mean=12840.77	EEF1A1P13	− 2.01	0.00207	Mean=25.22	Mean=22.6
			Std=4331.85	Std=5024.92				Std=14.64	Std=13.6
TAF10	1.61	0.00242	Mean=266.94	Mean=381.71	TOMM7	− 1.97	2.30E-05	Mean=1450.06	Mean=1301.03
			Std=153.75	Std=177.04				Std=626.74	Std=487.05
IFI6	1.59	1.60E-08	Mean=259.68	Mean=732.74	EEF1A1P11	− 1.76	0.00615	Mean=10.14	Mean=8
			Std=240.63	Std=460.46				Std=4.6	Std=5.47
LY6E	1.58	3.20E-11	Mean=2025.97	Mean=4517.89	COMMD6	− 1.66	0.00015	Mean=3106.44	Mean=2858.57
			Std=1166.45	Std=2297.03				Std=1149.37	Std=1038.78
IFITM3	1.53	5.50E-11	Mean=1603.56	Mean=4027.37	EEF1A1P14	− 1.45	0.04776	Mean=1.77	Mean=1.14
			Std=1467.64	Std=2617.44				Std=1.8	Std=1.25
PPP4C	1.47	7.50E-05	Mean=817.44	Mean=1102.83	EIF4B	− 1.45	2.30E-05	Mean=921.28	Mean=804.09
			Std=391.53	Std=462.09				Std=375.76	Std=392.81
ARPC3	1.46	0.00433	Mean=2789.67	Mean=3554.86	SNHG6	− 1.39	0.00175	Mean=2586.56	Mean=2599.06
			Std=1340.13	Std=1710.04				Std=949.45	Std=984.21
APOBEC3A	1.46	1.10E-08	Mean=786.61	Mean=2255.80	SNHG29	− 1.37	1.40E-07	Mean=2242.03	Mean=1972.00
			Std=690.8	Std=1728.52				Std=758.92	Std=839.15
ISG15	1.44	6.90E-10	Mean=972.66	Mean=2801.49	NACA	− 1.35	0.00024	Mean=9499.78	Mean=9411.31
			Std=841.12	Std=2091.96				Std=3796.36	Std=4296.91
RBX1	1.42	0.00022	Mean=413.5	Mean=667.57	TPT1P9	− 1.33	0.00321	Mean=18.16	Mean=15.89
			Std=197.15	Std=283.10				Std=8.17	Std=8.79
C2	1.41	6.90E-10	Mean=44.25	Mean=112.06	EEF1B2P3	− 1.28	0.00247	Mean=14.03	Mean=11.34
			Std=49.86	Std=88.73				Std=7.7	Std=6.90
IFI44	1.39	2.60E-09	Mean=194.14	Mean=429.97	RACK1	− 1.21	3.50E-06	Mean=10267.75	Mean=9988.66
			Std=215.09	Std=260.16				Std=3877.37	Std=4046.77
BIRC5	1.39	2.60E-09	Mean=19.67	Mean=94	NOP53	− 1.21	2.10E-06	Mean=3331.3	Mean=2973.09
			Std=20.69	Std=82.45				Std=1247.13	Std=1141.17
TYMS	1.39	1.90E-09	Mean=51.67	Mean=219.49	TCF7	− 1.19	9.70E-06	Mean=1817.14	Mean=1122.8
			Std=601.7	Std=193.64				Std=1013.74	Std=557.1
IFI27	1.37	4.90E-09	Mean=51.5	Mean=1019.34	EEF1GP5	− 1.19	0.01633	Mean=2.44	Mean=1.31
			Std=58.52	Std=1273.42				Std=2.79	Std=1.17
COX6A1	1.34	0.00049	Mean=240.44	Mean=348.43	UBA52	− 1.19	0.0102	Mean=34.53	Mean=3017.54
			Std=105.36	Std=124.17				Std=45.81	Std=1081.1
PCLAF	1.31	6.10E-08	Mean=34.53	Mean=163.09	PABPC1	− 1.17	7.80E-06	Mean=12388.58	Mean=11281.09
			Std=45.81	Std=105.38				Std=5878.49	Std=5286.66

We excluded ribosomal proteins from the list of negatively interacting genes

Interestingly, the expression profiles of well-characterized HIV-1 restriction factors showed, overall, a positive correlation with viral load (VL) across individuals (Fig. 1) such as APOBEC3A and APOBEC3B, members of the APOBEC3 family of cytidine deaminases [18].

Fig. 1.

Correlations between mRNA expression levels of known HIV-1 restriction factors and HIV-1 viral load. Scatter plots show the relationship between normalized expression values (x-axis) of selected host restriction factors (SERINC5, APOBEC3G, BST2, SAMHD1, SERINC3, APOBEC3A, APOBEC3F, and MX2) and log-transformed plasma HIV-1 viral load (y-axis) in patient samples. Each blue dot represents an individual subject. Linear regression fits are shown in red, with 95% confidence intervals indicated by shaded areas. The slope (β) and significance level (p-value) for each correlation are reported in the panel titles. Significant positive correlations (e.g., APOBEC3G, BST2, APOBEC3A, and MX2) indicate that higher expression of these restriction factors is associated with higher viral load

Another key factor positively correlated with VL is BST2 (Tetherin), whose expression is induced by interferon. In general, out of the 792 genes, we detected VL association with 32 genes previously reported as restriction factors. Important interferon-induced genes include IFITM2 and IFITM3, which restrict entry of various enveloped viruses [19] by interfering with membrane fusion via binding to the HIV-1 Env glycoprotein [20]. Another antiviral factor targeting Env is LGALS3BP, which is active both intra- and extracellularly [21]. Similarly, GBP5 restricts HIV-1 by blocking furin-mediated cleavage of Env, leading to reduced incorporation of functional glycoproteins [22]. OAS1, OAS2 and OAS3 restrict HIV-1 by activating the RNase L pathway, leading to degradation of intracellular RNA [23]. RSAD2 (Viperin) has been reported to act against numerous viruses by inducing premature termination of RNA synthesis [24], but also to support sustained HIV-1 infection in monocyte-derived macrophages [25]. Other interferon-induced factors include ISG20 [26], PLSCR1 [27], SERPING1 [28] and others. Statistically significant associations of these genes appear to reflect an interferon-driven host immune response to elevated viral burden rather than intrinsic antiviral control. In contrast, SERINC3 and SERINC5, constitutively expressed restriction factors not strongly induced by interferon, did not show significant correlation with VL. This lack of association in PBMCs may be due to their specific expression in macrophages, a cell type poorly represented in peripheral blood [7, 8, 29] or Nef-mediated viral counteraction mechanisms [30]. Many genes previously shown to promote infection and participate in HIV-1 progression were positively correlated with VL. These included CCR5 [31], CFL1 [32], LGALS1 [33], LGALS3BP (Galectin-3) [34], and PPIB [35]. Heat shock proteins HSP90B1 [36] and HSPA4 [37] were also positively associated. Additional genes included members of the kinesin family (KIF15, KIF18B, KIF2C, KIF5C) [38], ELOB [39], and STAT1 [40]. Genes promoting viral transcription and reactivation were also enriched. These included LGALS9 [41], EZH2 [42], BIRC5 [43] and TYMS [44].). Conversely, several host factors showed a negative association with HIV VL, including RPLP1 [45], GAS5 [46], PRKRA/PACT [47], COMMD6 [48], RCAN3 [49], NCR3 [50], and TCF7 [51]. Pathway analysis indicated enrichment in high VL patients for antiviral defense (p = 3.4E-18) and response to interferon-beta (p = 1.6E-4), cell division ( p = 1.1E-7), proteasome (p = 2.4E-8), mitochondrion (p = 1.6E-4) (Fig. 2).

Fig. 2.

STRING database visualization of genes positively correlated with HIV viral load, and main related pathways. Protein–protein interaction (PPI) network generated using the STRING database of genes whose mRNA expression levels are positively correlated with HIV-1 viral load. Nodes represent individual proteins, and edges denote known or predicted functional associations. Colored circles highlight major functional clusters enriched among these genes, including pathways related to cell cycle and DNA replication, the proteasome, the innate immune response, and mitochondrial activity. The high connectivity within and between clusters suggests that HIV replication is associated with coordinated regulation of multiple host cellular processes

Negatively associated pathways included cytoplasmic translation (p = 1E-127) and related ribosomal transcript production (4.4E-69) and cytokine–cytokine receptor interactions (3E-3) (Fig. 3). The top 20 associated pathways are reported in Fig. 4.

Fig. 3.

STRING database visualization of genes negatively correlated with HIV viral load and main related pathways. Protein–protein interaction (PPI) network of host genes whose mRNA expression levels are negatively correlated with HIV-1 viral load, constructed using the STRING database. Functional clustering highlights two major enriched pathways: ribosome biogenesis (dark blue cluster), representing a dense network of ribosomal proteins and assembly factors, and cytokine–cytokine receptor interaction (yellow cluster), consisting of immune signaling molecules and receptors. The observed negative correlations suggest that higher expression of ribosomal biogenesis and cytokine signaling genes is associated with lower HIV-1 viral load, implicating these pathways as potential host protective mechanisms against viral replication

Fig. 4.

Differential pathway enrichment analysis of gene expression data using DAVID. Top left: Significantly upregulated pathways, including extracellular exosome, cytosol, defence response to virus, protein binding, and negative regulation of viral genome replication, are shown with corresponding -log10(p-value) and gene counts. Top right: Significantly downregulated pathways, predominantly related to cytoplasmic translation, ribosomal subunits, and RNA binding, are displayed with their -log10(p-value) and gene counts. Bottom: Most significant GO biological process pathways identified by FGSEA gene set enrichment analysis – predominantly cytoplasmic transmission, metabolic processes, are shown with corresponding -log₁₀(p-value) and gene counts. Color intensity represents the number of genes contributing to each pathway

Deconvolution of cellular composition

To interpret the results from the association analysis, we performed cell type deconvolution analysis of the gene expression of the PBMCs using EPIC [52]. From the analysis of the whole transcriptome dataset, we observe that the majority of cells were CD8 T cells (up to 42%), B cells and NK cells whereas CD4 + T cells represented 10% of the PBMC population (Fig. 5). In addition, we compared the deconvolution of 25 samples with the lowest VL versus 25 samples with the highest VL (Fig. 5B). As expected, low VL samples showed a significantly higher abundance of B cells and CD4 T cells compared to high VL samples, in which these populations were 70% depleted (log₂ fold change=-1.76, p = 0.029). Interestingly, the deconvolution of high VL samples showed a significant increase of neutrophils, supported by the positive association with VL of granulocyte-specific genes such as CTSG, FPR1/2, SRGN, and GSDMD. Indeed, even though granulocytes tend to be excluded from PBMC filtration, a subset of low-density neutrophils (LDN) and platelets might be retrieved in PBMC [53].

Fig. 5.

Deconvolution of white blood cell populations. (A) Analysis of the full dataset reveals a depletion of CD4⁺ T cells and neutrophils. (B) When comparing blood cell populations between PWH with low and high VL, B cells and CD4⁺ T cells are more abundant in the low VL group, whereas neutrophils are significantly more elevated in the high VL group (* p-value < 0.05)

Mendelian randomization

We aimed to estimate the causal effect of the correlated genes on HIV-1 VL. A standard approach for assessing causality is Mendelian Randomization (MR), which leverages naturally occurring genetic variation as instrumental variables to infer the direction and magnitude of causal relationships between gene expression and disease phenotypes. To enable MR, we integrated expression quantitative trait loci (eQTL) data—linking genetic variants to gene expression levels—with GWAS summary statistics for HIV-1 VL. This integrative approach allowed us to investigate whether the observed correlations reflect true causal effects and to identify potential HIV restriction factor genes.

Using the Transcriptome Wide Mendelian Randomization (TWMR) [54, 55] approach applied to GWAS VL summary statistics and eQTLGen phase I data [56] for genes with observed associations between gene expression and VL, 146 genes were identified as causal, each supported by at least one eQTL. Given that for some of these genes there might be unobserved confounding, which violates key assumptions of MR, we additionally show the significance obtained with MRmix [57], MR-Egger [58], MR-presso [59], Mr-RAPS [60] and IVW [61] along with their robust [62]and penalized variations [63](Fig. 6 and S2 Table).

Fig. 6.

Identification of genes strongly correlated with VL. “Arkanoid”- style chart which represents genes with the strongest evidence of concordance across different types of IVW Mendelian Randomization methods, common GWAS summary on outcome (VL) and exposure (gene expression) dataset. Genes previously mentioned in conjunction with HIV-1 in PubMed abstracts marked in bold. Visual representation of all investigated genes in S3 Fig

Our analysis revealed that part of the genetic regulation of ribosomal protein genes (e.g., RPL12 (p = 7.1E-05, supporting MR methods: 9/23), RPL17 (p < 1 × 10⁻²², 4/13), RPL18A (p < 1 × 10⁻²², 5/13), RPL23A (p < 1 × 10⁻²², 3/11), and others; S2 Table) is causally and negatively associated with VL, supporting the hypothesis that HIV-infected cells downregulate ribosomal pathways to mitigate viral replication [64]. In line with this, several genes involved in peptide chain elongation (e.g., EEF1B2 (p < 1 × 10⁻²², 1/12), ELOB (p < 1 × 10⁻²², 6/12) also showed causal associations, suggesting that host genetic variation may influence interactions between HIV-1 and the reverse transcription complex [65].

Among the negatively correlated genes with a significant causal effect on VL, TRABD2A (p = 0.003, 11/20) stands out as a recently identified HIV-1 restriction factor. Another strong signal was PNRC1 (p < 1 × 10⁻²², 3/13), a nuclear receptor coactivator involved in mRNA decay and transcriptional regulation. We also identified IL7R (p < 1 × 10⁻²², 6/12), encoding the interleukin-7 receptor. Other notable genes with previously observed interaction with HIV-1 are EEIG1 (p = 0.0375, 2/9), the early estrogen-induced gene 1; MAL (p = 0.0037, 6/9), a myelin and lymphocyte protein associated with lipid raft signaling in immune cells; the long non-coding RNA DANCR (p < 1 × 10⁻²², 4/12) expressed during the early stages of HIV-1 infection. Surprisingly, CD44 (p < 1 × 10⁻²², 4/13) is part of this group of genes despite being tagged as a host factor incorporated in virions to promote hyaluronan-mediated infection of Fibroblastic reticular cells [66].

Finally, we detected associations with genes not previously known as HIV-1 interactors, such as TTC3 (p = 0.0003, 8/20), ZNF33B (p = 2.6 × 10⁻⁷, 4/13), KLF9 (p = 3e-7, 6/13), GRAP (p < 4e-6, 3/13). These are intriguing candidates for further functional validation and may represent novel host factors influencing HIV-1 replication or latency.

Among the positively correlated genes, several apoptosis-related factors emerged, including FASLG (p = 1.4 × 10⁻⁶, 3/12), TNFSF10 (p < 1 × 10⁻²², 6/12), CASP7 (p < 1 × 10⁻²², 6/23), CTSC (p < 1 × 10⁻²², 7/19), LMNB1 (p = 3.6 × 10⁻⁴, 3/13), and DRAM1 (p = 2.6 × 10⁻⁶, 7/23).

We also identified a set of innate immunity–related genes—many of which are known interferon-stimulated genes (ISGs)—with strong causal links to HIV-1 VL (GO enrichment, adjusted p = 2 × 10⁻³). These include well-established HIV restriction or response factors such as APOBEC3A (p = 1.1 × 10⁻¹², 8/24), ISG20 (p = 5.7 × 10⁻⁵, 7/20), RSAD2/viperin (p = 0.03, 8/12), PML (p < 1 × 10⁻²², 7/13), JAK2 (p = 4.7 × 10⁻⁴, 7/13), MYD88 (p = 0.0498, 6/9), OAS1 (p = 1.2 × 10⁻⁸, 6/20), FCER1G (p = 0.01, 5/23), NLRP3 (p < 1 × 10⁻²², 6/23), and IFI27L1 (p < 1 × 10⁻²², 6/23).

Discussion

We explored the interaction between PBMC gene expression and HIV-1 set point VL as a proxy for spontaneous viral control in 144 PWH. Using a Log-Normal Generalized Linear Model, we identified 792 genes whose expression levels were significantly positively or negatively associated with VL. Together, these results highlight a complex interplay between host restriction factors, interferon-driven immune responses, and viral countermeasures in shaping HIV-1 dynamics. Interferon-induced antiviral genes (e.g., APOBEC3 family, BST2, IFITMs, OAS family) correlated positively with VL, likely reflecting virally induced immune activation rather than direct control of replication. Additionally, our cell type deconvolution revealed a depletion of B and CD4⁺ T cells and an enrichment of neutrophils in high VL samples. We hypothesize that the loss of B and CD4⁺ T cells is partly compensated by a shift toward innate, but non-specific, neutrophil-driven responses, where low-density neutrophils suppress adaptive immunity against HIV-1 through PD-L1/PD-1 and expand under high viral burden [67].

The enrichment of host dependency factors (CCR5, CFL1, LGALS family, PPIB, kinesins) among positively correlated genes reinforces their central role in HIV-1 replication and progression. Associations with transcriptional regulators (EZH2, STAT1, LGALS9) and heat shock proteins suggest mechanisms that facilitate viral reactivation and latency maintenance. Pathway-level enrichments in translation, NF-κB/NFAT signaling, and proteostasis further point to regulatory nodes that could be targeted therapeutically.

While no significant association was observed for SERINC3 and SERINC5, we found several negatively associated genes (RPLP1, GAS5, PRKRA, COMMD6, RCAN3, TCF7) recently characterized as important regulators of T-cell metabolism and viral control. For instance, TCF7 mRNA expression in PBMCs has been reported to be lower in rapid progressors compared to long-term non-progressors and healthy controls [68], and TCF7 knockdown disrupts mitochondrial function and reduces CD8 + T-cell proliferative capacity [69], highlighting the central role of metabolic regulation in TCF7-mediated antiviral control.

We also observed a negative association between VL and the expression of eukaryotic translation initiation and elongation factors (EEF1B2, EEF1D, EEF2, EIF3H, EIF4B and pseudogenes EEF1A1P1, EEF1A1P13), of which EIF3H, EEF1D and EEF2 have already been reported as exploited by HIV-1 for viral protein synthesis [70–72]. Consistent with prior findings, several ribosomal transcripts were depleted at high VL, suggesting translational repression as a host restriction mechanism. Additional negatively associated causal genes included TRABD2A, which inhibits virion assembly by degrading the Gag polyprotein, PNRC1, a nuclear receptor coactivator involved in mRNA decay, IL7R, MAL, DANCR, and GRAP. Positive correlations were observed for genes involved in viral entry, replication, latency, and apoptosis, including CCR5, CFL1, LGALS1/3BP, PPIB, heat shock proteins (HSP90B1, HSPA4), LGALS9, EZH2, BIRC5, TYMS, and DRAM1. Notably, DRAM1 is upregulated by HIV via p53 and may contribute to the elimination of infected cells [73], highlighting the interplay between host apoptotic pathways and viral replication.

Gene ontology and pathway analyses revealed a downregulation of ribosomal biogenesis and cytokine-cytokine receptor interactions at high VL, whereas interferon-stimulated genes, proteasome activity, cell-cycle/DNA replication, and mitochondrial pathways were upregulated. These results suggest that HIV-1 induces host transcriptional programs to enhance replication while simultaneously activating antiviral defense mechanisms. Positive correlation between cell cycle–related pathways and VL has been reported in several previous studies and is consistent with the “tap-and-drain” hypothesis [74]. In this model, ongoing HIV-1 infection continuously stimulates the immune system by driving persistent activation and compensatory proliferation of CD4 + T cells to replace cells lost during infection (the “tap”). At the same time, these newly activated cells are preferential targets for HIV-1 infection and are rapidly destroyed or functionally exhausted, thereby “draining” the CD4 + T-cell pool faster than it can be replenished. This self-sustaining cycle of activation, proliferation, infection, and depletion provides a mechanistic explanation for the paradoxical coexistence of increased cell cycle signaling with progressive CD4 + T-cell loss. This pattern was also supported by our deconvolution analysis comparing individuals with high versus low VL (Fig. 5).

In this context, our study identified the TNFSF10 (TRAIL) signaling pathway as a potential mediator of HIV-1–induced CD4⁺ T cell depletion. Previous work in humanized (NRG-hu HSC) mice demonstrated that blockade of TRAIL prevented HIV-1–induced CD4⁺ T cell loss in vivo [75]. Here, we observed transcriptomic signatures consistent with TRAIL+ innate immune cells, supporting the role of TRAIL-mediated apoptosis in shaping CD4⁺ T cell dynamics during infection.

The identification of TTC3, ZNF33B, KLF9, and GRAP as genes associated with HIV-1 VL is particularly intriguing, as none of these factors have previously been characterized as HIV-1 interactors. TTC3 encodes a tetratricopeptide repeat protein with E3 ubiquitin ligase activity, involved in protein ubiquitination and degradation of phosphorylated Akt [76]. Since it has already been shown that PI3K/Akt survival pathway is triggered by HIV to prevent apoptosis of HIV-infected cells [77], TTC3 may counteract this effect by promoting Akt degradation, thereby influencing the balance between survival and death of infected cells and potentially impacting reservoir persistence. ZNF33B, an uncharacterized zinc finger transcription factor, could influence HIV-1 by modulating host chromatin or viral promoter activity, as other zinc finger proteins (e.g., ZNF restriction factors) have been implicated in HIV-1 viral replication [78]. KLF9, a ubiquitously expressed member of the Sp1 C2H2-type zinc finger family of transcription factors, has roles in T-cell differentiation [79] and stress responses [80]. It was previously observed as a possible activator of HIV-1 L promoter via GC-box binding [81] therefore its negative association with VL is unexpected. However, as other member of the KLF family, KLF9 might also act a transcriptional repressor by competing with SP1 [82, 83]. Finally, GRAP encodes a lymphoid-specific adaptor protein in the Ras/MAPK pathway, a signaling cascade activated by T-cell receptor stimulation [84] and known to facilitate HIV-1 replication and reactivation from latency [85].

Conclusions

These findings provide novel insights into host–virus interactions and suggest additional genes that may contribute to the natural control of HIV-1 infection. Through a combination of analyses of gene-expression variation, its relationship to viral load, and Mendelian randomization, we confirm previously reported associations involving members of the APOBEC and IFITM families, as well as BST2 and OAS genes. We also identify associations that implicate transcription-elongation factors, several of which have been highlighted previously as contributors to viral replication. Our results underscore the importance of interferon-dependent restriction factors, translational repression, apoptotic pathways, and T-cell metabolic regulators, and further nominate new candidate genes and pathways, including TRAIL (regulated by TNFSF10), that may influence viral control. Moreover, we highlight a number of genes not previously described as HIV-1 interactors; for example, TTC3 and GRAP participate in pathways already implicated in the HIV-1 life cycle. Collectively, these findings advance our understanding of virus–host interactions and identify putative targets for validation in future functional studies and potential therapeutic interventions.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

S.O. performed all computational and statistical analyses and contributed to manuscript writing. C.W.T. assisted with data analyses and contributed to writing. D.A., J.M., and M.M. carried out the experimental work. A. Ciuffi contributed to manuscript writing and provided clinical samples. A.Calmy, L.E., J.N., A.M., L.B., P.S., and M.S. provided clinical samples within the framework of the Swiss HIV Cohort Study. J.F. contributed to the analyses and manuscript writing and co-supervised the study. F.S. coordinated computational and statistical analyses, wrote the manuscript, conceived and supervised the study. All authors read and approved the manuscript.

Funding

This study has been financed within the framework of the Swiss HIV Cohort Study, supported by the Swiss National Science Foundation (grant #CRSII5_213514 to J.F.), by SHCS project #812 and by the SHCS research foundation. The data are gathered by the five Swiss University Hospitals, two Cantonal Hospitals, affiliated hospitals and private physicians.

Data availability

The datasets generated and/or analysed during the current study are available in the Zenodo repository, with DOI 10.5281/zenodo.18161013.

Declarations

Ethics approval and consent to participate

The SHCS was approved by the ethics committees of the participating institutions, and written informed consent was obtained from all participants.

Consent for publication

Consents for publication are obtained from enrolled participants as part of consent to participate.

Competing interests

The authors declare no competing interests.