CircRNA profiles of extracellular vesicle-enriched fractions from ART suppressed pregnant women living with HIV identifies interactome networks key to inflammation and viral latency

Abstract

Background

Despite virologic suppression with ART, pregnant people living with HIV (PLWH) experience a disparate risk of chronic non-AIDS-related morbidities (NAMs) during pregnancy compared to people without HIV infection including preterm labor and small for gestational age. Many studies postulate that NAMs reflect immunological dysfunction from underlying continuous inflammation. Extracellular vesicles (EV) are fundamental to both HIV immune pathogenesis and maternal-placental-fetal systemic intercellular networking. Circular RNAs (CircRNA) are enriched within EVs and are readily transmissible with disease-specific functional outcomes in recipient cell populations. However, there is limited research addressing the intersection of EV circRNA and HIV during pregnancy.

Methods

Small RNA (smRNA) cargo of EV-enriched fractions derived from the plasma of pregnant PLWH (n = 8) was compared to that of pregnant people living without HIV (PLWoH, n = 10) using low-input RNA sequencing (Illumina PE150). Differential circRNA profiles were used to construct circRNA-miRNA-gene regulatory maps and circRNA-protein interactions (circInteractome and miRTarbase). Gene set enrichment analysis (GSEA) using Reactome on the predicted genes from the interactome networks, provided insights into the regulatory roles of the differentially expressed circRNAs of EV-enriched fractions.

Results

Differential expression analysis identified 27 significantly upregulated circRNAs of EV-enriched fractions for pregnant PLWH as compared to pregnant PLWoH. GSEA revealed physiological pathways influential to HIV reservoir establishment, latency, replication, immune activation, immunosenescence, as well as maternal-placental-fetal immune cross talk. FUS, AGO2, and EIF4A3 were the top circRNA binding proteins and are involved in scaffolding for circRNA biogenesis, EV sorting, and miRNA sponging. High-density miRNA-binding circRNAs of EV-enriched fractions highlighted circRNA-mediated suppression of HIV inhibitory mechanisms including extensive binding of hsa-miR-326 and hsa-miR-548c-3p.

Conclusions

While many studies have compared plasma RNA profiles for PLWH compared to PLWoH, few have conducted this analysis for EVs, and we were unable to find a study specific to the unique combination of exploring RNA cargo of EV-enriched fractions in PLWH during pregnancy. This study aims to address the knowledge gap for NAMs in pregnancy by investigating EV-based intercellular communication. The results imply that novel EV circRNA-mediated regulatory mechanisms may contribute to HIV persistence, inflammation, and immune modulation in pregnancy.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-08348-y.

Introduction

According to the World Health Organization (WHO), at the end of the year 2022 there were an estimated 39 million people living with HIV (PLWH) of which 76% received ART [1]. In the ART era, the prognosis for PLWH has radically shifted from a lethal to a chronic infectious disease [2]. However, PLWH receiving ART continue to face a disproportionate risk for non-AIDS-related morbidities (NAMs) that impact multiple physiological systems associated with poor obstetrical outcomes [2, 3]. More than half of the global population are female, and every year, ~ 1.3 million become pregnant [4]. Women are underrepresented in HIV research and frequently excluded due to pregnancy status or the possibility of childbearing [5].

Sex-specific differences reflected in HIV pathogenesis, residual viremia, ART response, and NAMs have been observed in women living with HIV from numerous studies [6, 7]. Pregnant PLWH also have a higher risk for NAMs with approximately 2–3 times an increased risk for prenatal-, delivery-, and postpartum-related complications [8]. These risks extend beyond the mother. HIV-exposed uninfected (HEU) infants exhibit increased infectious disease vulnerability and adverse neurodevelopmental outcomes, compared to non-HIV-exposed infants [9]. Given the combination of NAMs being reported at higher rates in pregnant PLWH, burden of obstetrical complications, and poor clinical outcomes in HEU infants, there is an urgent need for HIV research efforts specific to pregnant PLWH receiving suppressive ART. Moreover, with public health success in enhancing ART access globally, to reduce perinatal and postnatal vertical transmission, this understudied population is projected to continue to rapidly expand [10].

The etiology of NAMs is not well understood and considered multifactorial [3]. Several studies report on aberrant immunological responses from the direct and indirect effects of HIV residual viremia [11]. Other contributory mechanisms include ART toxicities, co-infections, and gastrointestinal barrier breakdown with microbiome disruption [11]. HIV plasma viremia is often undetectable with effective ART, although a wide range of tissues contain HIV-infected cells precluding eradication and resulting in rapid viral rebound following treatment interruption [12]. In turn, the HIV vault of replication-competent infected cells is a critical source of continuous residual inflammation that exhausts and dysregulates the immune system [13]. During pregnancy, HIV-associated persistent inflammation and immune exhaustion can disrupt the delicate immunological balance required for healthy gestation, leading to placental inflammation, adverse pregnancy outcomes, and altered fetal immune programming [14]. These immunological outcomes underscore the need to uncover the underlying mechanisms of NAMs in PLWH and the contributions of pregnancy to these, to design targeted interventions that support maternal and infant health.

Extracellular vesicles (EVs) are critical to HIV pathogenesis and maternal-placental-fetal cellular networking, but their role in pregnancy is not well understood [15, 16]. EVs are nanosized membrane-bound structures released from cells that act as local and systemic messengers of a plethora of bioactive molecules (e.g. RNA species, proteins, etc.). EV is a general term inclusive of exosomes, apoptotic bodies, and microvesicles. As EV transmission is in bulk, recipient cells experience complete genetic reprogramming events with many alterations in signaling pathways and phenotype [17]. During pregnancy, the placenta releases a high magnitude of EVs with pregnant women having an increased EV plasma concentration ~ 20-fold greater than that of their non-pregnant counterparts [18]. There is clear evidence that EVs engage in bidirectional maternal-placental-fetal cross talk [19]. In the murine models, fluorescently labeled EVs have been tracked crossing the placenta with localization to distinct sites in maternal and fetal tissue [20]. EVs in maternal and fetal circulation are essential to physiological adaptations to pregnancy and fetal development, including immunity, metabolic homeostasis, and vascular function [21]. However, disordered maternal-placental-fetal EV communication is attributed with various adverse pregnancy outcomes [22].

Infections are considered to alter the in-utero environment with associated changes in EV quantity, subtype, and composition [23]. In PLWH, plasma-derived EVs are comprised of a wide range of proinflammatory cargo that contribute to HIV-associated chronic inflammation, impaired immune recovery, and NAMs [16, 24]. In pregnancy, HIV infection increases production and circulating levels for total and trophoblastic EVs that were shown to modify dendritic cell secretion of monocyte chemoattractant protein-1 (MCP-1) [25]. Interestingly, trophoblastic EVs conveyed an antiviral regimen of chromosome 19 microRNA cluster (C19MC miRNAs) that prompted viral degradation in autolysosomes facilitating protection against DNA and RNA viruses [26]. Overall, the host’s cellular microenvironments appear to be conditioned by EVs to promote disease progression with viral reservoir reinforcement [27].

During pregnancy, EVs mediate the transfer of smRNAs (ie. circRNA, miRNA) to recipient cells, resulting in targeted gene expression changes with immunoregulatory actions [28]. CircRNAs are covalently closed continuous loop structures comprised of single-stranded RNA (lack 5’ and 3’ poly(A) ends) and generated from back splicing events of exons and/or introns [29]. Although circRNAs were originally considered artifactual, progression in sequencing and computational technologies has uncovered considerable gene regulatory function within physiological and pathological processes integral to reproduction and antiviral immunity [30, 31]. CircRNAs have tissue-specific function with diverse regulatory roles interacting with RNA, proteins, and DNA [31]. EVs selectively enrich for circRNAs due to a significantly elevated back-splicing ratio and RBP- mediated sorting [32]. Large sets of stable disease-specific circRNAs are transferred by EV communication networks and can be readily detected through liquid biopsies [32]. EV circRNA profiles are emerging as a promising source for diagnostic/prognostic biomarkers and a fundamental regulatory axis for complex cellular responses that can be therapeutically targeted [33].

Our long-term goal is to address the knowledge gap for NAM disparities that may be related to pregnancy through exploring EV-based communication with an emphasis on cargo profiling. While ART usage to preserve maternal health and prevent vertical transmission of HIV is an indisputable priority, it is necessary to determine the putative mechanisms responsible for observed health disparities and to improve pregnancy and fetal/infant outcomes. In this article, we aim to identify EV circRNA regulatory interactome networks involved in HIV pathogenesis during pregnancy to aid in determining molecular mechanisms that underlie the negative inflammatory and immunological consequences observed in pregnant PLWH and their infants.

Methods

Biorepository sample and data acquisition

Plasma samples from peripheral maternal blood were acquired through the Atlanta, GA clinical site for the Study of Treatment and Reproductive Outcomes (STAR) and Study of Pregnancy Outcomes in women with Respiratory illness due to suspected or confirmed coronavirus infection (SPORE) biorepositories. Specifically, samples from PLWoH (n = 10) and PLWH (n = 6) were obtained from STAR as well as PLWH samples (n = 2) from SPORE. Both biorepositories target the biological and psychosocial effects of infectious disease, particularly HIV, in reproductive-age women and during pregnancy. The eligibility criteria were as follows: inclusion criteria (1) pregnant cisgender women, (2) ≥ 18 years of age, (3) singleton pregnancy, (4) PLWH or PLWoH, (5) lack the presence of other infections (e.g. gonorrhea, chlamydia, trichomonas, syphilis, SARS-CoV-2, HBV, TB, etc.) and exclusion criteria (1) < 18 years of age, (2) multiple fetus pregnancy, (3) self-report for HBV or TB infection during pregnancy, and (4) positive for a STI at the time of sampling. Specimen-linked data was used to obtain demographic, pregnancy-specific, and HIV-related clinical characteristics. Demographic and clinical characteristics for the participants are outlined in Table 1.

Table 1.

Pregnant participant demographic & clinical characteristics

Standard operating procedures were followed for processing maternal blood collected into cell preparation tubes (CPTs) with sodium citrate. Centrifugation of CPTs was at 1,500-1,800 x g for 20–30 min at room temperature (RT). If the CPTs were processed immediately, plasma was aspirated without disturbing the cell layer, pooled per donor, and spun at 1,000 x g for 10 min. If the CPTs were processed the next day, they were similarly aspirated and pooled, however an additional centrifugation was applied at 400 x g for 10 min followed by 1,000 x g for 10 min. Platelet-free plasma was distributed into 1 mL aliquots and stored at -80 °C.

RNA isolation from EV-enriched fractions and RNAseq

RNA isolation from EV-enriched fractions, library preparation, and sequencing were performed at CD Genomics (USA). The Qiagen exoRNeasy Midi Kit (Germany, catalog number 77144) was used to isolate total EV RNA from 1 mL of filtered (0.22m) blood plasma according to the manufacturer’s protocol. In brief, the plasma sample was mixed with Buffer XBP, EVs were bound to the column, washed with Buffer XWP, lysed, and eluted with QIAzol. For total RNA isolation, chloroform followed by ethanol was added, EV RNA was bound, washed, and eluted. RNA quantity and quality were assessed through the Qubit RNA HS assay (ThermoFisher, catalog number Q32852) and Bioanalyzer 2100 Eukaryote Total RNA Nano (Agilent Technologies), respectively. Given the limited quantity of sample (~ 1 mL of plasma), purification and additional EV characterization experiments were not implemented. Sequencing libraries were constructed by a strand-specific cDNA synthesis with 3’ end adenylation and adaptor ligation. Prior to sequencing, quality control and normalization processes were applied. Low-input paired-end small RNA sequencing was conducted at 20 M reads/sample with the Illumina HiSeq X Ten PE150 (USA). Raw sequencing data was filtered for clean reads and mapped to the reference genome by hierarchical indexing for spliced alignment of transcripts 2 (HISAT2). All samples from both groups (pregnant PLWH and pregnant PLWoH) were processed under identical conditions to ensure internal consistency and minimize technical variability.

smRNA detection & annotation

The COMPSRA (“COMprehensive Platform for Small RNA-Seq data Analysis”) algorithm was used for annotation against multiple small RNA reference databases (miRbase, gtRNAdb, gencode, circBase, piRNABank, piRBase and piRNACluster), enabling reference-guided assignment of reads to small RNA classes, including circRNA-associated loci. The mapped reads’ positional information on the smRNA permitted abundance determination with the htseq-count software. The density distribution for each sample and the correlation between samples reinforced technical data consistency and reliability. To remove low-abundance features, a filter was applied to the raw count data that required a minimum count threshold of three reads in at least 30% of the samples. Density distribution and inter-sample correlation analyses were conducted. These approaches reduced noise from very low-expressed smRNAs and focused the downstream analyses (e.g., differential expression, heatmaps, and PLS-DA) on more reliably detected transcripts. Origin and integrity of the RNA signal was validated through read length distribution analysis. To ensure that the detected RNA signal reflected host-derived extracellular RNA rather than potential viral contamination, we implemented a multi-layer validation strategy combining annotation-based filtering, structural analysis, and targeted viral alignment.

Differential expression analysis

After the alignment, filtering, and annotation, DESeq2 software was used, to detect differential smRNA expression for pairwise comparisons, with a |log2Fold Change| ≥ 1.00 and adjusted p-value (FDR) < 0.05 set as the screening criteria. To visualize these expression changes, a volcano plot and heatmap were generated. For supervised multivariate analysis, we performed partial least squares discriminant analysis (PLS-DA).

Network construction & pathway enrichment analysis

Circular RNA Interactome was used to predict target miRNAs, of the differentially expressed circRNA of EV-enriched fractions from pregnant PLWH, with a context percentile score ≥ 90. Next, miRTarbase identified predicted genes targeted by the miRNAs and were screened to have at least 1 strong validation method (western blot, qPCR, or reporter assay confirmation). The circRNAs) candidates (annotated via circBase) were scanned using TargetScan algorithm feature of the CircInteractome database to designate “super-sponge” status (≥ 20 binding sites for a single miRNA). Interactome networks were generated with Gephi. Gene enrichment analysis (GSEA) of the predicted genes from the circRNA-miRNA-gene networks was evaluated through the Reactome database.

RBP analysis

To uncover which RBPs interacted with the differentially expressed circRNAs, in silico motif predictions were combined with experimental enhanced crosslinking and immunoprecipitation (eCLIP) data. FIMO-scanned binding‐site counts for each RBP were extracted from the circInteractome database and a “motif matrix” was built across our 27 significantly upregulated circRNAs of EV-enriched fractions to flag RBP–circRNA pairs with ≥ 20 predicted sites as candidates for high‑density interaction. The aggregated human eCLIP peaks (all cell types) were overlaid onto 100 nt–extended circRNA coordinates to compute, for each RBP–circRNA pair, the number of CLIP overlaps (hit count).

Assessment of potential viral RNA contribution

To evaluate whether the RNA signal detected in extracellular vesicle (EV)-enriched samples could reflect co-isolated viral particles, we performed a targeted alignment-based screening against HIV reference genomes. Sequencing reads were first aligned to the human reference genome (GRCh38 primary assembly), and non-human reads were subsequently extracted and aligned to a combined HIV-1 reference composed of multiple representative genomes (NC_001802.1, K03455.1, AF033819.3). Alignment files were processed using samtools to generate sorted and indexed BAM files. For each sample, we quantified the percentage of reads aligning to the HIV genome, high-confidence HIV-aligned reads using a mapping quality threshold of mapping quality score (MAPQ) ≥ 30, and genome-wide coverage profiles. To assess the structural consistency of viral signal, coverage was summarized in binned genomic intervals across the HIV genome. In addition, read length distributions were examined to evaluate library integrity and to determine whether mapped reads were consistent with expected small RNA sequencing profiles.

Results

Participant population characteristics

Group characteristics were compared by unpaired T-tests and distribution by contingency analysis (Fischer’s exact test). Most participants were Black/African American, non-Hispanic with 87.5% and 60% for pregnant PLWH and PLWoH, respectively. The age (p = 0.105), cumulative # of pregnancies (p = 0.113), cumulative # of live births (p = 0.218), as well as the distribution of parity status (p > 0.999) and trimester of sample collection (p = 0.261) did not statistically differ between groups. There were similar levels of tobacco/drug use and a low prevalence of pregnancy-related complications (e.g. gestational diabetes). While 37.5% had ≥ 95% adherence, all pregnant PLWH were on ART with an undetectable viral load (< 20 copies/mL) except for one participant who had been recently diagnosed with HIV infection.

Seven pregnant PLWH had a CD4 + T cell count > 500 cells/µL consistent with a functional immunological status. One pregnant PLWH had a CD4 + T cell count < 500, but > 200 cells cells/µL, the threshold for an AIDS-defining condition. Pregnant PLWH had a wide spread of 1362 cells/µL for the nadir CD4 + T cell count, ranging from 6 to 1368 cells/µL. The nadir CD4 + T cell count is the lowest value within the course of HIV, typically prior to ART initiation, and is a negative prognostic predictor en with sustained restoration of CD4 + T cell levels > 500 cells/µL. The duration of HIV infection, defined as the time elapsed since diagnosis, spanned 21.58 years with a median of 4.15 and mean of 6.24 years. The modes of transmission were heterosexual and perinatal/vertical.

Biotype distribution for circulating smRNA of EV-enriched fractions

A total of 89,109 unique smRNAs were mapped from plasma EVs. These smRNAs were categorized into six biotypes: circRNA, piRNA, snRNA, tRNA, miRNA, snoRNA. The distribution of these biotypes was evaluated between PLWH and PLWoH groups (Fig. 1A). CircRNA was the predominant biotype in both groups, comprising 98.26% (71,946 reads) of the total smRNAs in samples from PLWH and 98.47% (69,600 reads) in samples from PLWoH. The remaining biotypes were present in much smaller proportions: piRNA accounted for 0.54% (396 reads) and 0.44% (309 reads), snRNA for 0.46% (336 reads) and 0.40% (284 reads), tRNA for 0.35% (258 reads) and 0.36% (254 reads), miRNA for 0.23% (165 reads) and 0.21% (145 reads), and snoRNA for 0.16% (117 reads) and 0.12% (87 reads), in samples from PLWH and PLWoH, respectively. Statistical analysis using Fisher’s exact test revealed no significant differences in smRNA biotype composition between the two groups (p = 0.888). These results demonstrate a consistent distribution of smRNA species regardless of HIV status, with circRNA dominating the RNA cargo composition of EV-enriched fractions in both groups. We next asked whether any individual circRNA was uniquely detected in one group or the other. Applying a presence threshold (≥ 3 reads in ≥ 30% of samples) yielded 4,407 “present” circRNAs in total, of which 2,489 (70.6%) were shared between PLWH and PLWoH, 618 (17.5%) were exclusive to PLWH EVs and 419 (11.9%) were exclusive to PLWoH EVs (Fig. 1B).

Fig. 1.

Species diversity mapping and explorative profiling comparison for smRNA-seq of EV-enriched fractions derived from plasma based on HIV status in ART suppressed pregnant women. Note: PLWH (orange) and PLWoH (green). (A) Bar graph illustrates the total number and distribution of mapped unique smRNA references. (B) Exclusive vs. shared total circRNA detection. circRNA presence was defined at a threshold of ≥ 3 reads in ≥ 30% of samples. (C) Partial least squares discriminant analysis (PLS-DA) is a supervised learning technique for dimensionality reduction. This analysis approach accounts for variable interaction and prediction classification relevance. The distinct grouping separation indicates that the HIV status likely alters smRNA EV cargo. Approximately 30% of the variance (Component 1: 19% and Component 2: 11%) is explained by the selected predictors. Eclipses represent the 95% confidence interval to aid in visualization of intra-group variability. (D) Heat map and hierarchical clustering analysis for all the samples and their top 20 smRNA (circRNAs). Individual samples and circRNAs are represented by the columns and rows, respectively. Clustering trees are shown at the left (for circRNAs) and at the top (for samples). The annotation bar displays HIV status, and the color-coded scale indicates increased (red) and decreased (blue) circRNA levels relative to the mean for normalized expression values

SmRNA profiling of EV-enriched fractions reveals distinct grouping based on HIV status

From the initial dataset of 89,109 smRNAs mapped from plasma EVs, a filtering step was applied whereby smRNAs with fewer than three counts in at least 30% of the samples were excluded. This resulted in a refined dataset of 3,760 smRNAs for downstream analysis. To explore potential differences in smRNA profiles between HIV status, a partial least squares discriminant analysis (PLS-DA) was performed on the filtered dataset. This supervised dimensionality reduction approach identifies the most relevant components for distinguishing groups in high-dimensional data samples. The PLS-DA results demonstrated distinct separation between pregnant PLWH and PLWoH (Fig. 1C). Correspondingly, hierarchical clustering of the top 27 smRNAs using complete linkage and Euclidean distance revealed clear separation by HIV status, providing additional support for the discriminative potential of smRNAs of EV-enriched fractions (Fig. 1D).

Two exceptions, in which the participant clustered to the opposite group (e.g. pregnant PLWH participant that clustered with the pregnant PLWoH group), were B7 (PLWH) and B9 (PLWoH). Interestingly, participant B7 (pregnant PLWH) acquired HIV perinatally. Previous studies have suggested that perinatally-acquired PLWH experience immune/inflammatory changes that differ from other modes of HIV acquisition [34]. There are very few studies that explore this phenomenon of perinatally vs. non-perinatally acquired PLWH within the context of pregnancy and the in utero environment [34]. Participant B9 (pregnant PLWoH) verified the use of tobacco/cigarettes within the past year. Cigarette smoke exposure is known to modulate smRNA profiles that elicit inflammation and subsequent organ dysfunction [35]. The clustering of participants B2 (pregnant PLWH) and B6 (pregnant PLWH) independently from the other pregnant PLWH, was possibly due to the recent diagnosis with HIV and inflammatory-related conditions (e.g. gestational diabetes, hypertension), respectively. Both HIV-associated inflammation and ART-induced dysregulation of placental hormones exacerbate the pathophysiological mechanisms of gestational diabetes (GDM) and hypertensive disorders of pregnancy (HDP) [36]. The heightened levels of the top 27 smRNA as compared to PLWoH in participant B6 could have been amplified by GDM and HDP or could possibly have been already present and predisposed these individuals to the disorders. These findings underscore that smRNAs of EV-enriched fractions, particularly circRNAs, hold promise as potential biomarkers for distinguishing by HIV prognostic status, warranting further validation in larger cohorts.

Differential expression analysis identifies significant alterations of circRNA levels of EV-enriched fractions for pregnant PLWH as compared to pregnant PLWoH

Differential expression analysis identified 27 significant upregulated circRNAs in EV-enriched fractions from pregnant PLWH compared to pregnant PLWoH using thresholds of |log2FC| ≥ 1 and FDR < 0.05) (Supplementary Table 1). Among these, hsa-circ-0080427 (logFC = 3.9), hsa-circ-0063063 (logFC = 3.7), and hsa-circ-0063062 had the highest expression levels, as shown in Supplementary Table 1 and Fig. 2A. To further characterize the differentially expressed circRNAs, we used circBase to determine their genomic origins (Fig. 2). Sixteen upregulated circRNAs arise from SNORD116-10, one from SNORD116-6, and three from SNHG14 (Small Nucleolar RNA Host Gene 14). SNHG14 is a long non-coding RNA (lncRNA) that hosts multiple small nucleolar RNAs (snoRNAs), including the SNORD116 cluster, which is involved in RNA processing and epigenetic regulation. This observation is of particular interest given that deletions affecting SNHG14 and SNORD116 are linked to Prader-Willi syndrome, a neurodevelopmental disorder associated with dysregulated RNA metabolism, suggesting that circRNA dysregulation at these loci may have broader functional implications [37]. Additionally, three circRNAs mapped to WBSCR17 (GALNT17), located within the Williams-Beuren syndrome (WBS) region at 7q11.23 and implicated in membrane trafficking. Two circRNAs mapped to RBFOX2, encoding a master splicing regulator critical for tissue-specific alternative splicing placental development, with associations to preeclampsia [38]. Finally, two circRNAs mapped to TCONS_00023883, an uncharacterized transcript suggesting novel regulatory RNAs yet to be studied (Fig. 2B).

Fig. 2.

Differential expression exhibited by smRNA of EV-enriched fractions derived from plasma of ART suppressed pregnant PLWH as compared to PLWoH and circRNA-associated gene evaluation. (A) Volcano plot depicts the differentially expressed smRNA (labeled). Note: FDR = false discovery rate. Points colored in red with a right shift are upregulated smRNA with an |log2FC| ≥ 1 and FDR < 0.05 threshold. While there are no points colored in blue with a left shift, these would indicate downregulated smRNA. Points colored in grey are smRNA that lack statistical significance. (B) Chord diagram of the upregulated circRNAs and their host genes. circRNAs, shown on the left, are connected to their corresponding host genes, shown on the right, and are represented by a single color e.g. SNORD116-10 in yellow. (C) Ideogram shows the genomic organization of SNHG14 circRNA isoforms on human chromosome 15q11.2. 22 of the 27 upregulated circRNAs originate from the SNHG14 locus. The top panel highlights the 15q11.2 band in red. The middle panel is the SNHG14 lncRNA host gene colored in light blue and the green boxes denote the positions of the SNORD116 intronic snoRNA genes (SNORD116--30). The bottom panel has the differentially enriched circRNA isoforms from the SNHG14 locus arranged by their genomic start–end coordinates (gray bars labeled by circRNA ID). A common 2,849 bp region (chr15:25,312,558–25,315,406) shared by every isoform is colored in orange

Pathway enrichment of circRNA-miRNA-gene mapping displays associations with inflammation and viral latency

To uncover circRNA-driven regulatory mechanisms in ART-suppressed pregnant PLWH, we generated a circRNA–miRNA–gene network in two steps. First, we used CircInteractome (https://circinteractome.nia.nih.gov) to predict candidate miRNAs for each upregulated circRNA. This tool applies TargetScan algorithms to detect 7-mer or 8-mer seed-region complementarity and reporting a context^+ score that gauges the likelihood of functional binding. We kept only those miRNAs with scores ≥ 90. Second, we cross-referenced these miRNAs against miRTarBase 10.0 (https://mirtarbase.cuhk.edu.cn), retaining interactions experimentally validated by at least one of the following: luciferase reporter assays, Western blots, or qPCR. This two-tiered filtering approach removed low-confidence predictions and yielded high-confidence circRNA–miRNA–gene triads, which likely contribute to chronic inflammation and viral latency seen in this population.

The resulting network, visualized in Gephi, comprises 27 circRNA (pink nodes), 82 miRNA (orange nodes), and 388 genes (green nodes) (Fig. 3). Nodes size reflects degree centrality, with larger nodes indicating high connectivity. In our tripartite network, each miRNA node has an in-degree (the number of circRNAs that feed into it) and an out-degree (the number of genes that miRNA targets). Notably, many circRNA targeted the same miRNA, for example, hsa-miR-326 have relatively high in-degree (15 circRNAs) and a considerable out-degree (22 genes), giving a total degree of 37, illustrating potential substantial regulatory overlap. Subnetworks based on Reactome identified genes pertinent to HIV pathogenesis are depicted in Fig. 4.

Fig. 3.

Integrated circRNA-miRNA-gene network developed from differentially expressed circRNA of EV-enriched fractions derived from plasma ofART suppressed pregnant PLWH. The central nodes (pink) are circRNAs. The middle nodes are miRNAs (orange) that are bound by the circRNAs as determined by circInteractome with a context percentile score ≥ 90. The peripheral nodes (green) are the genes targeted by the miRNAs as determined by the miRTarbase database with support from at least 1 strong validation method such as western blot, qPCR, or reporter assay confirmation. Each connection is a specific interaction between the circRNAs, miRNAs, and target genes. Node size reflects the interaction quantity within the network (e.g. highly connected nodes appear larger vs. less connected nodes appear smaller)

Fig. 4.

Subset circRNA-miRNA-gene networks based on Reactome identified genes pertinent to HIV pathogenesis. Following Reactome gene set enrichment analysis on predicted genes, select signaling pathways essential to HIV pathogenesis are depicted as subsets from the complete network described in Fig. 3 that was developed from differentially expressed circRNA of plasma-derived EVs in pregnant PLWH. The central nodes (pink) are circRNAs. The middle nodes are miRNAs (orange) that are bound by the circRNAs. The peripheral nodes (green) are the genes targeted by the miRNAs

A “super-sponge” network of the circRNA-miRNA high density binding interactions (> 20 binding sites) with AGO2 sites (ranging from 47 to 79) identified seven miRNA as heavily bound up by circRNAs to prevent target mRNA degradation and translational repression (Fig. 5). Hsa-miR-326, and has-miR-548c3p have been previously linked to HIV through directly decreasing HIV replication and inhibiting TGF-β and potentially supporting antiviral immune function, respectively [39, 40]. By suppressing HIV inhibitory mechanisms related to blocking HIV replication or enhancing host immune response, the EV circRNA could be mediating processes of viral protection.

Fig. 5.

CircRNA-miRNA “super-sponge” network. The TargetScan algorithm feature of the CircInteractome database was used to scan every annotated circRNA from the complete network described in Fig. 3 for 7--mer seed matches to all human miRNAs. A threshold of ≥ 20 sites for a single miRNA was used to designate a circRNA as a “super-sponge.” Correspondingly, the majority of high-density miRNA-binding circRNAs had numerous AGO2 cross-link peaks. The central nodes (pink) are circRNAs and the peripheral nodes (green) are miRNAs (orange) that are bound by the circRNAs

The Reactome biological pathways database was used to perform functional gene enrichment analysis of the miRNA-target genes. The top 20 statistically significant pathways (p-adjusted of 1e-07–3e-07) are shown in Fig. 6.

Fig. 6.

Gene set enrichment analysis (GSEA) of predicted genes from the circRNA-miRNA-gene network. Gene targets of miRNAs bound by the top differentially expressed circRNAs of EV-enriched fractions derived from plasma of ART suppressed pregnant PLWH was assessed for physiological function through the Reactome biological pathways database. The 20 most enriched biological pathways are displayed. GSEA identifies whether a gene list is significantly overrepresented within a particular biological pathway. The gene ratio expresses the gene list representation within the complete gene set of the pathway. P-values were calculated from the enrichment scores and portray the likelihood for biological relevance of the gene list to the pathway

Mapping of RNA-binding proteins (RBP)

To investigate the mechanisms driving circRNA formation in plasma EVs, we used CircInteractome to scan the 100-nt regions immediately upstream and downstream of each differentially expressed circRNA junction. Remarkably, binding motifs for FUS, AGO2, and EIF4A3 were predicted at nearly every junction, consistent with their established roles in looping distant splice sites together and catalyzing back-splicing (Fig. 7). In contrast, factors such as TDP-43, LIN28A, IGF2BP1/2/3, TIAL1, DGCR8, HuR, FMRP, AGO1 and ZC3H7B appeared more selectively, each on a subset of junctions—suggesting that while EIF4A3/FUS/AGO2 form the core “matchmakers,” additional RBPs fine-tune circRNA biogenesis in an RNA-specific manner.

Fig. 7.

RNA binding protein (RBP) motif-scanning and Crosslinking and Immunoprecipitation (CLIP) integration of differentially expressed circRNA of EV-enriched fractions derived from plasma of ART suppressed pregnant PLWH. Find Individual Motif Occurrences (FIMO)‐predicted RBP binding sites was extracted from CircInteractome database to report potential matches for RBP’s position weight matrix (PWM) across every annotated circRNA sequence. A motif matrix of circRNA x RBP counts was generated, with the subset of the 27 identified differentially expressed circRNA, to determine candidate high-density interactions (≥ 20 sites). To corroborate these in silico predictions with experimental binding data, aggregated human eCLIP peak sets (all cell lines) was overlapped via Granges with the RBP motif scanning findings. For each circRNA–RBP pair, the CLIP hit count (number of overlapping peaks), and Mean CLIP score (average crosslink score across those peaks) were computed This combined motif + CLIP approach highlighted FUS, AGO2 and EIF4A3 as the top RBPs likely to be engaged with the circRNAs of interest in our study of plasma-derived EVs from pregnant PLWH

Of all RBPs surveyed, EIF4A3 motifs showed the highest and most consistent enrichment, underscoring its central role in circRNA production. Intriguingly, EIF4A3 also antagonizes IRF3-driven type I interferon induction and thereby promotes replication of multiple RNA viruses [41]. Our data therefore raise the possibility that EV-associated circRNAs—by virtue of recruiting EIF4A3 at their junctions—could couple enhanced back-splicing efficiency with localized suppression of innate immune signaling in recipient cells. Moreover, the co-occurrence of multiple RBP motifs on individual circRNAs reinforces a multilayered control model, in which combinatorial RBP binding not only specifies which circles are formed and exported, but may also dictate their stability, subcellular routing, and downstream regulatory functions once delivered to placental or fetal targets.

Evaluation of potential HIV RNA contribution to EV-associated signal

To evaluate whether the detected RNA signal could be influenced by viral contamination, we examined read alignment patterns to HIV reference genomes. A small fraction of reads aligned to the HIV genome across samples, without consistent association with clinical groups (Supplementary Figure 1A). After applying stringent mapping quality filtering (MAPQ ≥ 30), high-confidence HIV-aligned reads were nearly absent across all samples, with most samples containing zero reads (Supplementary Figure 1B). Genome coverage analysis revealed a sparse, non-uniform signal with recurrent hotspots shared across samples, inconsistent with authentic viral genome representation (Supplementary Figure 1C). Consistently, HIV-aligned reads showed markedly lower mapping confidence compared to human-derived reads (Supplementary Figure 1D). Read length distributions exhibited a dominant peak consistent with library insert size and a secondary peak characteristic of small RNA species (Supplementary Figure 1E), further supporting the absence of structured viral RNA. In contrast, host-derived RNA signals showed clear biological structure. circRNA-associated loci mapped to coherent genomic regions, including the SNHG14/SNORD116 locus (Fig. 2C), supporting structured host-derived transcription. Together, these results indicate that HIV-derived RNA contribution is negligible, whereas host-derived RNA signals are structured and biologically consistent, supporting their origin from EV-associated RNA cargo rather than viral contamination.

Discussion

In this study, circRNA cargo of EV-enriched fractions derived from the plasma of pregnant PLWH and pregnant PLWoH was characterized through smRNA sequencing. Subsequent circRNA-miRNA-gene mapping with biological pathway enrichment analysis and circRNA-protein interaction identification was performed to determine potential circRNA of EV-enriched fractions regulatory processes pertinent to pregnant PLWH. While many studies have compared plasma RNA profiles for PLWH vs. PLWoH, few have conducted this analysis for EVs, and we were unable to find a study specific to the unique combination of circRNA cargo of EV-enriched fractions for PLWH during pregnancy. With the advancing fields of circRNAs and EVs, their vital role in both pregnancy and viral infections necessitates novel multidisciplinary research reflective of the inherent physiological complexities.

For explorative analysis, the smRNA biotype composition and PLS-DA/hierarchical clustering by HIV status were evaluated. CircRNA was the most abundant smRNA present in EVs with low levels of the other species for both pregnant PLWH and pregnant PLWoH. The finding of similar composition profiles for smRNA of EV-enriched fractions for both groups is consistent with other studies on EV smRNA cargo of PLWH (non-pregnant), however the type of smRNA that was most to least abundant varied [42]. EV cargo is recognized to display an extensive degree of composition diversity predominately reflective of the EV subpopulation, parental cell origin, and individual physiological status [43]. The PLS-DA and hierarchical clustering data demonstrated clear separation by group with markedly different profiles for smRNA cargo of EV-enriched fractions for pregnant PLWH vs. pregnant PLWoH indicating circRNAs of EV-enriched fractions may potentially be beneficial biomarkers for HIV effects during pregnancy. Twenty-seven circRNAs of EV-enriched fractions were demonstrated to be significantly differentially expressed between groups. Similarly, plasma smRNAs might distinguish elite controllers, chronic HIV progressors, and PLWoH from one another [44].

CircRNAs are most known for their ability to act as a ‘sponge’ for RBPs and microRNA, thereby exerting transcriptional control through circRNA-miRNA-mRNA/gene regulatory axes [45]. There is growing evidence that circRNA dysregulation supports several viral infections, most recently SARS-CoV2 [46]. Only a limited number of studies have explored circRNA-miRNA-mRNA/gene regulation with retroviral infections such as HIV [47]. We were unable to find any articles that explored this phenomenon within the context of EV cargo during pregnancy [47]. We systematically generated a circRNA-miRNA-gene map based on the significantly upregulated circRNA of EV-enriched fractions of the pregnant PLWH using predictor algorithms, circular RNA interactome and miRTarbase. Next, functional gene enrichment analysis via the Reactome biological pathways database was used to identify biological pathways interconnected to circRNA regulation within pregnant PLWH. From the top 20 significantly enriched pathways, we highlight those most relevant to HIV pathogenesis and consistent with the current theories of inflammation and viral latency in NAM development (Fig. 8). These include the following: IL-4/IL-13 cytokine signaling, Transforming growth factor-β (TGF-β) signaling, Estrogen receptor (ESR) -mediated signaling, PIP3 activates protein kinase B (AKT) signaling, Toll-like receptor 4 cascade, and Mitogen-activated protein kinase (MAPK) signaling cascades.

Fig. 8.

CircRNA of EV-enriched fractions differentially expressed in pregnant PLWH exhibit regulatory control on pathways fundamental to immune function, HIV persistence, and HIV-induced chronic inflammation. (1) Immune Evasion – IL-4/IL-13 signaling. Th2 dominance promotion results in a deficient cytotoxic T cell response thus protecting the virus against host immune defenses. (2) Reservoir Reinforcement – TGF-β and Estrogen signaling. TGF-β directs latency initiation through increased infection with viral transcription inhibition, lymphoid homing, and central memory differentiation. Estrogen is a potent repressor of viral reactivation. (3 and 4) Enhanced Viral Reactivation and Dysfunctional T Cell Activity – TLR cascade, MAPK, and PI3K/Akt. A cycle occurs whereby viral persistence drives continuous low-grade inflammation that activates T-cells harboring HIV provirus to proliferate, thus maintaining the latent reservoir and further exacerbating inflammation. TLR and MAPK signaling are involved in an inflammatory cascade supportive of HIV propagation and reactivation. PI3K signaling reinforces T cell survival and proliferation by reducing apoptosis induction and inhibiting autophagy. While these effects may seem paradoxical when considered simultaneously, within a cycle of an inflammatory exacerbation that reactivates the virus to expand the latent reservoir followed by reactivation inhibition and reservoir reinforcement enables for continued maintenance of the viral reservoirs with host immune response disruptions. Created with BioRender

Given the opposing cross-regulation of Th1 and Th2 polarization, immune responses will be dominated by either, but not both cell types. Different diseases have been noted to markedly interfere with this Th1/Th2 balance [48]. In the context of a viral infection, IL-4/IL-13 are recognized as quintessential Th2 cytokines and considered to be an evasion tactic that exploits the reliance of cytotoxic T cell (CTL) induction on Th1 cells [48]. Aberrant cytokine production induced from progression to chronic HIV infection supports a Th2 dominant state. As the Th1 support of CTLs declines, antiviral efforts are impeded and variants with rapid replicative capabilities, such as the syncytium-inducing (SI) virus, are promoted [49]. CTLs have intricate polyfunctional antiviral properties that extend past the terminology of “killer T cells” [50]. The noncytotoxic antiviral response (CNAR)/CD8 + T cell antiviral factor (CAF) are non-TcR-mediated/-HLA-restricted mechanisms that suppress HIV as well as other retroviral replication [50]. Cell-to-cell contact is not a necessity; in fact EV secretion potently reduces HIV transcription through a protein moiety for both acute and chronic infection models [51]. The quality of CTLs is critical to protecting against HIV disease progression; elite controllers have CTLs with high avidity and reinforced polyfunctionality [52]. IL-4R antagonist or IL-13Rα2 adjuvanted vaccine strategies that inhibit IL-4/IL-13 signaling have demonstrated improvements in CTL antiviral efficacy in macaques [53]. In utero exposure to immune dysregulation, HIV, and ART of pregnant PLWH influences immune ontogeny in HEU infants [9, 54]. The deficit in Th1 immunity within pregnant PLWH appears to be reflected within HEU infants [55]. During pregnancy, there is a dominant bias towards a type-2 cytokine environment, however this skew is more pronounced for cord blood mononuclear cells of HEU infants and may further impair Th1 function contributing to impaired responsiveness to intracellular pathogens and conventional immunizations [55].

TGF-β directs effects for pro- and anti-inflammatory peripheral immune responses to HIV and remains at elevated levels within PLWH [56]. The persistence of TGF-β has been connected to immunosuppression in AIDS development and in NAM progression (e.g. cardiovascular, hepatic, and pulmonary) [57]. TGF-β persistence may be a physiological response to the HIV-related chronic inflammatory state that exhausts the immune system [56]. Unexpectedly, this adaptive measure appears to result in an even more pronounced escalation towards immunosenescence [56]. Emerging studies underscore TGF-β as central to the establishment and expansion of the latent HIV reservoir [58]. The primary proposed TGF-β-mediated HIV latency mechanisms are enhanced HIV infection followed by latency initiation, central memory T cell differentiation, and lymphoid homing. TGF-β also increases the expression of CCR5, thus assisting in the R5-tropic CD4 + T cell susceptibility to HIV infection [58]. Simultaneously, TGF-β upregulates B lymphocyte-induced maturation protein-1 (BLIMP-1) via modifying the microRNAome with consequent suppression of viral transcription [59]. Overall, the viral load is increased with latency expansion [59]. Pro-inflammatory exacerbations in disease pathogenesis reactivate HIV and promulgate this cycle [59]. T cell migration to secondary lymphoid organs is directed by CCR7 and is actively upregulated by TGF-β signaling resulting in HIV dissemination and protection within pockets of immunological privilege [58]. In addition, TGF-β triggers effector cells to undergo differentiation to central memory CD4 + T cells that are essential to the longevity of the HIV latent reservoir [60]. Consistently, following blockade of TGF-β with galunisertib in macaques, SIV was successfully reactivated, lymphoid tissue had reduced provirus, and T cells exhibited a shift towards an effector phenotype [61]. TGF-β is vital to maternal-placental-fetal immunological cross talk and other essential pregnancy-related processes such as embryonic growth and uterine spiral artery remodeling; thus the clinical ramifications of TGF-β blockade in pregnant PLWH should be carefully considered [62]. TGF-β has been associated with an array of pregnancy complications; increases have been linked with induction of pre-eclampsia whereas decreases can result in recurrent spontaneous abortion from disruption in maternal immune tolerance [63]. Perturbations in maternal cytokine patterning provoked by viral infections have been demonstrated to adversely affect pregnancy outcomes and fetal development pertaining to immune function. For perinatally-HIV-infected and HEU infants, TGF-β + Treg cells were speculated to be expanded to alleviate mucosal inflammation with bystander immune suppressive effects [64].

Continuing in the vein of the latent HIV reservoir, ESR-mediated signaling has been shown to strongly repress HIV reactivation [65]. There appears to be sex specific differences as a result of estrogen responsivity that change reservoir size [65]. There are also evident sex-based differences in HIV pathogenesis that likely stem from sex hormone regulation of HIV-susceptible immune cells [65, 66]. A molecular inhibitory mechanism of HIV transcription through 17β-estradiol was identified, Erα, and β-catenin complexed and settled on the HIV LTR to productively block promoter activity [66]. The estrogen receptor is considered a strong repressor of HIV reactivation and, while frequently overlooked, is important to the regulation of latency [65]. Coherently, the inducible viral reservoir is smaller in women and may impede latency-reversing agents [65]. Fluctuations in estrogen levels due to menstruation, reproductive stage, or pregnancy need to be addressed when considering targeting of the HIV latent reservoir [65]. Estrogen hormone therapy for transgender women living with HIV is proposed to increase risk for NAM development due to activating toll-like receptor 4 and worsening of HIV-associated chronic inflammation [67].

The phosphoinositide 3-kinase (PI3K)/ Akt (aka protein kinase B) pathway is exploited by many viruses [68]. HIV hijacks the PI3K/Akt signaling to regulate replication and cellular survival following viral entry [68]. HIV activation of Akt improves cellular survival by inhibiting autophagy, lowering the stimulation of caspases, and blocking pro-apoptotic factors [69]. In contrast to Akt signaling in resting T cells, hyperactive states with Akt activation promote HIV reactivation [70, 71]. The concurrent AKT activation of NF-kB followed by a proinflammatory cytokine cascade is also considered targetable with implications for reducing NAM risk [71]. EVs have previously been associated with modulating PI3K/Akt signaling to reactivate HIV through the transmission of c-Src cargo [72]. Inhibition of the PI3K/Akt pathway has garnered attention in the block-and-lock HIV strategy [73]. However, PI3K signaling appears to be critical to pregnancy, with interruption leading to compromised maternal metabolic adaptations and fetal growth within murine models [74].

Systemic immune activation followed by immune exhaustion with diminished T cell function in PLWH is associated with the degree of toll-like receptor (TLR) activity [75]. Myeloid differentiation primary response protein 88 (MyD88) dependent signaling, initiated by all TLRs apart from TLR3, induces pro-inflammatory NF-kB and mitogen-activated kinases (MAPK) with secretion of IL-12 and interferons [76]. Chronic HIV stimulation of TLR-4 signaling increases programmed death ligand-1 (PD-L1), a primary marker of exhaustion, expression that drives immunological unresponsiveness and correlates with disease progression [75]. Sex-based differences in immune response to viral infections are partially explained by variability in TLR response [77]. More robust TLR immunity appears to be pervasive in women with correlations to chromosome X inactivation avoidance [78]. In pregnant women with viral infection, TLR responses incite an inflammatory maternal-placental-fetal interface with immune tolerance disruptions and pregnancy complications [79]. More than half of HEU infants were found to have TLR4 anergy with insufficient pathogen sensing corresponding to the clinical window of vulnerability to gram-negative bacteria [9, 80]. Targeting TLRs to provoke a controlled inflammatory environment conducive to HIV reactivation is being explored for therapeutic latency reversal [81]. Conflicting accounts have been reported for the effects of the TLR function of HIV infection, replication, and reactivation [77]. TLR9 agonist treatment, MGN1703, in PLWH directed HIV reactivation with an improved CTL response [82]. MAPK inhibition has been shown to reverse viral-mediated blockade of apoptosis and repress HIV LTR promoter activity [83]. Phosphatidylethanolamine-binding protein 1 (PEBP1) inactivation of MAPK/NF-kB signaling through cytoplasmic sequestration advanced HIV latency [84].

The MAPK family signaling cascades are important to numerous viral infections by supporting replication/infectivity, altering immune response, and apoptosis potential [85]. Consequently, MAPK signaling has been proposed as a target for the development of broad-spectrum antiviral agents [85]. Elevated MAPK signaling has been observed to be maintained by secreted viral proteins and inflammation (canonical pathway) in the absence of virus [85]. For HIV, stress-related MAPK p38 has been attributed with cytokine-mediated inflammation, macrophage activation, and induction of apoptosis by HIV proteins gp120 and tat [86]. MAPK p38 has been intrinsically linked with inflammation and small-molecule inhibitors are actively being pursued for chronic inflammatory conditions [87]. Intriguingly, SIV-infected macaques treated with a p38 MAPK inhibitor and ART had significantly reduced activation markers and plasma levels of IFNα, IFNγ, TNFα, IL-6, IP-10, sCD163 and C-reactive protein [86]. In the context of chronic SIV infection with ART, the p38 MAPK inhibitor again reduced immune activation and improved T cell recovery [86].

Our discovery that maternal plasma EVs are selectively enriched in circRNAs harboring high-affinity FUS, AGO, and EIF4A3 binding motifs suggests a previously unrecognized mechanism by which pregnant PLWH might shape their offspring’s innate defenses (Fig. 9). EIF4A3, a core exon-junction complex factor, not only drives back-splicing but has been shown to antagonize IRF3-mediated type I interferon induction, thereby promoting RNA virus replication [41]. By packaging circRNAs studded with multiple EIF4A3 recognition sites into EVs, the maternal compartment could inadvertently deliver “splicing–immune” signals to the placenta and fetal tissues. In HEU infants, this EV-mediated transfer may blunt early interferon responses, compromising antiviral defenses. Furthermore, persistent trafficking of EIF4A3-loaded circRNAs to classical HIV reservoir niches could reinforce the low-level viral persistence that ART cannot fully clear. Going forward, it will be critical to track the uptake of these maternal EV circRNAs by neonatal immune cells, measure subsequent IFN pathway activity in HEU cohorts, and correlate specific circRNA signatures with infection outcomes. Such studies may ultimately reveal both a predictive biomarker and a novel target for therapeutic intervention in the prevention of pregnant PLWH and HEU morbidity.

Fig. 9.

Proposed molecular mechanisms of the differentially expressed circRNA of EV-enriched fractions in pregnant PLWH in targeting RNA binding proteins (FUS and AGO) and miRNA sponging (e.g. miR-326). Given the large size of the EV circRNA, the high-density of miRNA binding, and AGO2 overlap; scaffolding and sponging/decoying to compete with endogenous smRNA and proteins are likely their main functions of the numerous regulatory capabilities of circRNA. For protein scaffolding, the EV circRNA could be involved in facilitating FUS elicited back splicing with subsequent circRNA biogenesis as well as in targeting smRNA species to selectively be sorted into EVs. For miRNA sponging, miR-326 was a prime example in the ‘super-sponge’ network of an EV circRNA-mediated suppression of HIV inhibitory processes. Created with BioRender

There are several limitations within our study due to EV isolation approach, small plasma volumes, EV purity, possible clinical confounders, and bioinformatic methodology restrictions in circRNA identification. As highlighted in MISEV 2023, affinity-based EV isolation methods are more specific for EVs than standard precipitation techniques (e.g., PEG-based methods); however, they do not distinguish EVs from other extracellular particles or specific subtypes [88, 89]. Therefore, in accordance with MISEV guidelines, the resulting fractions were termed EV-enriched. Limited volumes of plasma (~ 1 mL) were able to be obtained, thus additional EV characterization experiments were not feasible without compromising RNA yield and downstream analysis. These constraints are common in studies involving valuable clinical samples, particularly in pregnancy cohorts where sample availability is inherently limited. Notably, similar studies utilizing limited-volume clinical samples (from people living with HIV or pregnant participants) have relied on affinity-based or low-input approaches coupled with downstream molecular or computational validation when extensive physical characterization is not feasible [42, 90, 91].

The complexity of plasma also introduces potential co-isolation of viral particles. Given the overlap between EVs and HIV virions in size and biophysical properties, there are challenges to effective separation. Despite this limitation, multiple lines of evidence support that the observed RNA profiles predominately reflect of host derived EV-associated RNA rather than virion-associated material. All the pregnant PLWH were ART-supressed with undetectable viral loads (except one recent diagnosis), suggesting minimal viral contribution. To further assess potential viral contribution, sequencing reads were aligned to HIV reference genomes. Although a small fraction of reads aligned to HIV, high-confidence alignments were nearly absent across all samples, and genome coverage was sparse, fragmented, and non-uniform, with recurrent hotspot regions observed across both PLWH and PLWoH samples. These patterns are inconsistent with authentic viral RNA. In contrast, human-aligned reads showed robust high-confidence mapping, and host-derived RNA features mapped to coherent genomic loci, including the SNHG14/SNORD116 region. Read length distributions were also consistent with expected small RNA sequencing profiles, further supporting the biological integrity of the dataset. Collectively, these findings indicate that the observed RNA profiles are not driven by viral contamination and instead reflect structured host-derived EV-associated RNA cargo.

There were no statistically significant differences for the clinical parameters considered between the groups and hierarchical clustering did not unveil overlapping patterns for the 27 clinical comparisons. This would imply successful matching of baseline characteristics between groups, or the low number of participants may not have provided sufficient power to fully detect clinical influences. Additionally, residual clinical confounders, those that were not assessed such as other pregnancy complications or inflammation-related conditions, could have impacted our findings.

It is important to note that circRNA annotation in this study was primarily based on reference-guided assignment of reads to circRNA-associated genomic loci (e.g., circBase), rather than comprehensive de novo detection of back-splice junctions using junction-aware algorithms. While this approach enables efficient identification of circRNA-enriched regions, it may conflate true circular RNAs with linear host transcripts or lncRNA fragments transcribed from the same loci (for example, SNHG14). Therefore, the reported circRNAs signals should be interpreted as circRNA-associated loci Future studies should incorporate junction-aware circRNA callers (e.g. CIRCexplorer2, find_circ) and experimental validation approaches, such as RNase R treatment and back splice junction–specific RT-PCR, to confirm circular RNA structure [94].

Conclusions

While women are more than half of the global HIV populous and experience severe NAMs that extend into pregnancy, they are substantially underrepresented in HIV research. This study addresses the knowledge gap for NAMs through a novel multidisciplinary approach specific to EV cargo transport during pregnancy exposed to HIV and ART. We demonstrated differential profiles in circRNAs of plasma-derived EV-enriched fractions, generated multiple interactome regulatory networks (circRNA-miRNA-gene, circRNA-protein, circRNA-miRNA super-sponge), and conducted pathway enrichment analysis that reinforced the theory of chronic inflammation underlying NAM development within pregnant PLWH. We also postulated potential molecular mechanisms of HIV progression based on EV circRNA scaffolding and miRNA/RBP sponging. Within the landscape of precision medicine, studying immune response and EV cargo could aid in the identification of potential biomarkers and therapeutic targets.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

HIV

Human Immunodeficiency Virus

ART

Antiretroviral therapy

PLWH

People living with HIV

NAMs

Non-AIDS-related morbidities

EV

Extracellular vesicles

CircRNA

Circular RNA

smRNA

Small RNA

PLWoH

People living without HIV

GSEA

Gene set enrichment analysis

WHO

World Health Organization

HEU

HIV-exposed uninfected

MCP-1

Monocyte chemoattractant protein-1

C19MC miRNAs

Chromosome 19 microRNA cluster microRNAs

miRNA

MicroRNA

STAR

Study of Treatment and Reproductive Outcomes

SPORE

Study of Pregnancy Outcomes in women with Respiratory illness due to suspected or confirmed coronavirus infection

SARS-CoV-2

Severe Acute Respiratory Syndrome Coronavirus 2

HBV

Hepatitis B Virus

TB

Tuberculosis

STI

Sexually transmitted infection

AA

African American

PI

Protease inhibitor

NRTI

Nucleoside reverse transcriptase inhibitor

INSTI

Integrase strand transfer inhibitor

CPTs

Cell preparation tubes

RT

Room temperature

HISAT2

Hierarchical indexing for spliced alignment of transcripts 2

COMPSRA

COMprehensive Platform for Small RNA-Seq data Analysis

PLS-DA

Partial least squares discriminant analysis

qPCR

Quantitative polymerase chain reaction

eCLIP

Enhanced crosslinking and immunoprecipitation

GDM

Gestational diabetes mellitus

HDP

Hypertensive disorders of pregnancy

SNORD116

Small nucleolar RNA, C/D Box 116

SNHG14

Small Nucleolar RNA Host Gene 14

lncRNA

Long non-coding RNA

snoRNAs

Small nucleolar RNAs (snoRNAs)

WBSCR17

Williams-Beuren Syndrome chromosomal region 17

GALNT17

Polypeptide N-acetylgalactosaminyltransferase 17

WBS

Williams-Beuren Syndrome

RBFOX2

RNA binding fox-1 homolog 2

RBP

RNA binding proteins

FIMO

Find Individual Motif Occurrences

IL-4

Interleukin-4

IL-13

Interleukin-13

TGF-β

Transforming growth factor-β

ESR

Estrogen receptor

PIP3

Phosphatidylinositol (3,4,5)-trisphosphate

PI3K

Phosphoinositide 3-Kinase

AKT

Protein kinase B

MAPK

Mitogen-activated protein kinase

TLR

Toll-like receptor

CTL

Cytotoxic T cell

SI

Syncytium-inducing

CNAR

CD8 + noncytotoxic antiviral response

CAF

CD8 + T cell antiviral factor

BLIMP-1

B lymphocyte induced maturation protein-1

SIV

Simian Immunodeficiency Virus

MyD88

Myeloid differentiation primary response protein 88

NF-kB

Nuclear Factor kappa-light-chain-enhancer of activated B cells

PD-L1

Programmed death ligand-1

PEBP1

Phosphatidylethanolamine-binding protein 1

MAPQ

Mapping Quality Score

Author contributions

Conceptualization, D.B., V.S., V.B., and E.J.; Sample acquisition, A.S., T.T., M.B., R.F.; Data curation, C.M., R.F., D.B., & A.C., Analysis & interpretation – D.B., V.S. E.J., Original draft preparation, D.B., V.S., E.J.; Funding, R.C., E.J. All authors participated in the manuscript’s review and editing process.

Funding

The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This study was by supported by the National Institute of Child Health and Human Development (NICHD) R01HD97843 (EJ and RC). This work was also supported in part by the Research Centers in Minority Institutions (RCMI) Grant Number U54MD007602 from the National Institute of Minority Health and Health Disparities (NIMHD). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NICHD, NIMHD, or the NIH.

Data availability

Data access is openly available on the Gene Expression Omnibus (GEO) under the series GSE309395.

Declarations

Ethics and consent to participate

This study was approved by the Morehouse School of Medicine IRB (#2038242-3). Written informed consents were obtained from all participants.

Ethic declaration

This study was conducted in accordance with the Morehouse School of Medicine Human Research Protection Program (accredited March 2015) and adhered to the ethical principles outlined in the Belmont Report, the Declaration of Helsinki, and the Nuremberg Code, as well as applicable provisions of the Code of Federal Regulations (Titles 21 and 45). Informed consent for biospecimen collection and access to clinical data was obtained through the Study of Treatment and Reproductive Outcomes (STAR) Atlanta biorepository.

Competing interests

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.