Viral load dependent cellular heterogeneity in Trichomonas vaginalis at the single cell level

Summary

Trichomonasvirus (TVV) is a double-stranded RNA virus that persistently infects Trichomonas vaginalis, a protozoan parasite responsible for the most common non-viral sexually transmitted infection. While TVV has been implicated in modulating parasite biology and drug sensitivity, the extent and significance of viral heterogeneity remain unclear. Using ultra-deep single-cell RNA sequencing (scRNA-seq) on three TVV-positive T. vaginalis isolates, we reveal that viral distribution is highly heterogeneous and dynamic, with marked variability in infection status and viral load across individual cells. Time-resolved scRNA-seq datasets confirm that this variability is not static but fluctuates across replicates. By integrating bootstrap-based pseudobulk analysis with viral load-stratified trend analysis, we uncover infection-associated transcriptional changes involving stress response, adhesion, and chromatin regulation. Notably, MLF-like and histone genes are selectively upregulated in cells with high viral loads. These findings redefine current models of TVV-host interactions and lay a foundation for future mechanistic studies of virus-mediated pathogenesis in protozoan parasites.

Graphical abstract

Highlights

• •scRNA-seq reveals mosaic Trichomonasvirus distribution in T. vaginalis
• •Viral load heterogeneity dynamically fluctuates within clonal T. vaginalis populations
• •High viral burden drives transcriptional changes linked to drug susceptibility and adhesion
• •MLF-like and histone genes show selectively upregulated under high infection states

Parasitology; Virology; Omics

Introduction

Trichomonas vaginalis is an anaerobic, flagellated protozoan parasite and the causative agent of trichomoniasis, the most prevalent non-viral sexually transmitted infection worldwide. While often asymptomatic, infections in women can cause vaginal discharge, vulvovaginal inflammation, and various reproductive complications,^1,2 including increased susceptibility to HIV-1 acquisition, cervical neoplasia, infertility, and adverse pregnancy outcomes.³ Metronidazole, a nitroimidazole compound, remains the cornerstone of trichomoniasis treatment, typically achieving cure rates of 84–98%.^4,5,6 However, emerging reports of reduced drug efficacy and resistance among clinical isolates underscore an urgent need to elucidate the molecular mechanisms driving parasite adaptation and drug responsiveness.^7,8,9

A unique feature of T. vaginalis biology is its frequent co-infection with Trichomonasvirus (TVV), a genus of double-stranded RNA viruses within the family Pseudototiviridae. TVVs possess a non-segmented genome of approximately 4.5–5.5 kilobases, containing two overlapping open reading frames that encode a capsid protein and an RNA-dependent RNA polymerase.^10,11 Phylogenetic analyses have identified five distinct species, designated as TVV1 through TVV5, that can co-infect the same T. vaginalis isolate. Despite sharing conserved structural and replicative features, these viruses are genetically diverse and do not appear to have originated from a common ancestor.^12,13

Although TVVs are non-cytolytic, they can amplify inflammatory responses in the human host, particularly through the activation of Toll-like receptor 3 (TLR3) by double-stranded RNA.¹⁴ While the host-directed effects of TVV have been relatively well characterized, much less is known about the direct impact of TVV infection on the parasite itself. Accumulating evidence suggests that TVV may modulate parasite biology by altering gene expression and enhancing cytopathogenicity. Yet, transcriptomic studies investigating these parasite-intrinsic effects have yielded inconsistent and sometimes contradictory results. As summarized in Tables S1–S3, previous research examining the impact of TVV on T. vaginalis protein expression, metronidazole susceptibility, and clinical symptoms has been substantially constrained by confounding factors, including inter-isolate genetic variability, heterogeneous culture conditions, and bulk population averaging, making it difficult to disentangle and attribute observed effects specifically to TVV presence or load. Critically, no previous study has addressed the within-population heterogeneity of viral infection at single-cell resolution, leaving a significant knowledge gap in our understanding of how TVV shapes parasite biology and phenotypic diversity.

To address these challenges, we combined bulk RNA sequencing (RNA-seq) and polymerase chain reaction (PCR) to systematically determine isolate-level TVV infection status, followed by single-cell RNA sequencing (scRNA-seq) and immunofluorescence assays to analyze viral heterogeneity and distribution at the single-cell level. We focused on three key isolates: ATCC 50148 (harboring TVV1, TVV2, and TVV3), ATCC PRA-98 (TVV2 and TVV3, used to investigate dynamic temporal changes), and ATCC 30236 (TVV1-only, enabling differential gene expression analysis under metronidazole-treated and untreated conditions). Together, these approaches provide the first high-resolution characterization of intra-population viral heterogeneity in T. vaginalis and directly address the central question: Is TVV distribution heterogeneous at the single-cell level, and how does viral load shape parasite phenotypes and transcriptional programs?

Results

Characterization of Trichomonasvirus infection in T. vaginalis isolates

To confirm the presence of Trichomonasvirus (TVV) in various T. vaginalis isolates, RNA sequencing (RNA-seq) data were mapped to reference sequences of Trichomonasvirus. The infection status of 24 isolates was systematically analyzed and summarized in Table 1. Among these, the isolate ATCC 50143 was confirmed to be virus-free, whereas ATCC 30236 was infected by TVV1 only. Notably, double infections were detected in isolate ATCC PRA-98 (harboring TVV2 and TVV3), while a triple infection was observed in isolate ATCC 50148 (harboring TVV1, TVV2, and TVV3) (Figures 1A and 1B). These four isolates were selected for further investigation. Initial validation via RT-PCR (Figure 1C) confirmed findings consistent with the in-silico RNA-seq analysis, reinforcing the reliability of the sequencing approach in viral identification.

Table 1.

Trichomonasvirus species infecting T. vaginalis isolates

Isolate	Number of Species	TVV1	TVV2	TVV3	TVV4	TVV5
NYCB20	0
ATCC 50143 (CDC085)	0
ATCC 50167 (B7RC2)	0
T1	1	●
B7268	1	●
ATCC 30236 (JH 31A #4)	1	●
T016	1		●
ATCC 30001 (C-1:NIH)	1		●
ATCC 30238 (JH 32A #4)	1			●
GOR69	2	●	●
NYCF20	2	●		●
ATCC 50142 (RU393)	2	●		●
LSU180	2	●			●
CDC1103	2	●			●
CDC1123	2	●			●
CDC1132	2	●			●
NYCG31	2	●				●
PRA-98 (G3)	2		●	●
ATCC 50148 (NYH286)	3	●	●	●
NYCA04	3	●		●		●
NYCE32_8	3	●		●		●
SD2	3	●		●		●
NYCC37	4	●	●	●		●
NYCD15	5	●	●	●	●	●

Figure 1.

Trichomonasvirus species infection in T. vaginalis isolates

The presence of TVV in the T. vaginalis isolates (ATCC 30236, ATCC 50143, ATCC 50148, ATCC PRA-98) was determined by mapping RNA-seq reads to Trichomonasvirus reference sequences obtained from the NCBI Virus Database using QIAGEN CLC Genomics Workbench (v20.0.3).

(A) Summary of consensus coverage, average coverage, and total read counts.

(B) Schematic representation of the consensus contig alignment for each Trichomonasvirus species.

(C) Detection of Trichomonasvirus isolates using RT-PCR with virus species-specific primers, with PCR products separated on a 1% agarose gel.

Single-cell RNA-seq reveals mosaic Trichomonasvirus distribution within clonal parasite populations

To investigate the distribution of Trichomonasvirus within T. vaginalis isolates, single-cell RNA sequencing (scRNA-seq) was employed. Sequencing data were processed using the Cell Ranger pipeline (10x Genomics, Pleasanton, CA, USA), followed by downstream analysis with Seurat v5.1.0 and Monocle v1.3.7 in R. These tools enabled the classification of individual cells based on viral transcript abundance. Visualization of viral distribution revealed significant heterogeneity in the presence of viral infection across single cells within the same isolate (Figure 2A). This observation was further confirmed by immunofluorescence assays, which demonstrated specific cytoplasmic staining of viral double-stranded RNA (dsRNA) using a dsRNA-specific antibody (Figure 2B). Collectively, these findings confirm that Trichomonasvirus displays an uneven, mosaic-like distribution at the single-cell level, even within genetically clonal parasite populations.

Figure 2.

Heterogeneous distribution of Trichomonasvirus in different T. vaginalis isolates

(A) viral populations among the T. vaginalis isolates ATCC 50148. The plots show the presence of Trichomonasvirus species and their co-occurrence patterns within isolates.

(B) Detection of viral dsRNA using the dsRNA-specific antibody 9D5 in different T. vaginalis isolates. Nuclei were stained with DAPI (blue), while viral dsRNA was detected using the 9D5 antibody conjugated with Alexa Fluor 594 (red). Images indicate the heterogeneity of viral dsRNA within the parasite. Scale bars = 10 μm.

Trichomonasvirus subtype proportions fluctuate over time within the same isolate

To investigate the temporal dynamics of Trichomonasvirus (TVV) infection, we compared the distribution of TVV subtypes across three independently collected single-cell datasets (Figure 3), representing distinct experimental time points (Exp. 1, Exp. 2, and Exp. 3). Each cell was categorized into one of the four groups based on infection status: TVV2, TVV3, co-infection with TVV2 and TVV3, or TVV-negative (TVV-). Notably, the relative abundance of each subtype exhibited substantial variation across time. Exp. 1 was dominated by TVV2 and uninfected cells, while Exp. 2 showed a dramatic increase in TVV3-infected cells. In Exp. 3, co-infection (TVV2+TVV3) became the predominant subtype, accompanied by a reduction in both TVV3-only and TVV-negative populations. This temporal shift in viral composition suggests a dynamic process influencing subtype prevalence over time.

Figure 3.

Temporal dynamics of Trichomonasvirus (TVV) species composition

Bar plots show the distribution of Trichomonasvirus (TVV) infection types across three independent biological replicates (EXP. 1, EXP. 2, and EXP. 3). Cells were classified into four groups based on viral content: TVV2-only, TVV3-only, co-infection with TVV2 and TVV3, or virus-negative (TVV-). Although percentages are shown for clarity, statistical comparison was performed using absolute cell counts. A chi-square test revealed significant differences in subtype composition across experiments (∗∗∗∗p < 0.0001), highlighting that the virus species distribution is dynamic rather than fixed within clonal populations.

A chi-square test based on cell counts confirmed that these compositional differences were statistically significant (∗∗∗∗p < 0.0001), supporting the hypothesis that the distribution of TVV subtypes varies non-randomly across experimental time points.

Having established that TVV burden varies temporally and spatially within isolates, we next sought to evaluate how this heterogeneity influences host gene expression under drug pressure.

Sub-lethal metronidazole treatment enables single-cell resolution of early stress responses

To assess how viral heterogeneity modulates drug-induced transcriptional responses at single-cell resolution, we first optimized metronidazole exposure conditions that induce cellular stress without triggering extensive cell death. A time-course dose-response analysis was performed using the TVV1-positive isolate ATCC 30236, with metronidazole concentrations ranging from 4 to 8 μM. Among these, 6 μM was identified as the optimal sublethal concentration (Figure 4). This concentration entirely suppressed parasite proliferation by 6 h post-treatment while preserving more than 90% cell viability during the 6- to 9-h window. These parameters were selected to ensure that cells remained transcriptionally active during the early stress response phase without progressing to late-stage cytotoxicity. The 6 μM condition was therefore used for the downstream single-cell RNA sequencing experiment.

Figure 4.

Establishment of sub-lethal metronidazole exposure conditions for single-cell RNA sequencing in T. vaginalis

Cell concentration curves of ATCC 30236 following exposure to metronidazole at 4, 6, and 8 μM, measured over a 24-h time course. All three concentrations led to complete growth arrest by ∼6 h post-treatment, though the degree of suppression varied slightly among them. We selected 6 μM (red line) as a representative sub-lethal condition, as it reflects an intermediate response and preserves >90% cell viability between 6 and 9 h. This concentration was therefore chosen to best capture the early transcriptional response window while maintaining cellular integrity. Data represent mean ± SEM from three independent biological replicates (n = 3). Note: the y axis was truncated and compressed to allow both the high cell concentrations of the NC group and the suppressed growth of drug-treated groups to be visualized in the same panel.

Viral burden varies widely across cells from the same culture

To investigate the distribution of Trichomonasvirus infection at single-cell resolution, scRNA-seq was performed on two samples derived from the same T. vaginalis isolate (ATCC 30236), obtained by splitting a single culture. One sample was briefly exposed to a sub-lethal concentration of metronidazole (MTZ) for 1 h, while the other remained untreated. For downstream analysis, cells were categorized into three infection states based on normalized viral transcript abundance: High_infection (top 25% of virus-positive cells), Infected (remaining virus-positive cells), and Uninfected. As shown in Figure 5A, viral transcript levels varied substantially across individual cells, with cells infected at a high level exhibiting the highest abundance of viral transcripts. The distribution of infection states also differed between the two samples (Figure 5B): sample 30236 consisted primarily of uninfected cells, whereas 30236_M displayed a higher proportion of virus-positive cells. Given that both samples originated from the same isolate and were processed in parallel, the observed differences likely reflect inherent variability in the distribution of the virus.

Figure 5.

Viral load distribution and infection group composition in single-cell RNA-Seq Data

This figure illustrates the distribution of viral RNA and the relative abundance of infection groups in two T. vaginalis single-cell RNA-seq samples (30236 and 30236_M).

(A) Violin plots display normalized viral transcript levels (per 10⁴ UMIs) across three infection states: High_infection, Infected, and Uninfected.

(B) Stacked bar plots show the proportion of cells assigned to each infection group within each sample.

Bootstrap pseudobulk strategy robustly identifies reproducible differentially expressed genes across sampling parameters

Due to the inherent limitation of single-cell data sparsity, we employed a bootstrap-based pseudobulk strategy to improve the robustness of differential expression analysis. In our implementation, each bootstrap was performed with five resampling iterations, and we systematically varied. the number of bootstrap iterations (500 vs. 1000) and the number of cells sampled per replicate (150 vs. 200). These settings were applied to two biological samples (30236 and 30236_M) across both the infected vs. uninfected and high infection vs. uninfected comparisons. The DEG sets identified under different configurations exhibited substantial overlaps, as illustrated in the four-way Venn diagrams (Figure 6). In the infected vs. uninfected comparison, 82.5% (12,908 genes) and 79.8% (8,151 genes) of DEGs were consistently detected across all sampling schemes in samples 30236 and 30236_M, respectively. For the high infection vs. uninfected condition, these overlaps further increased to 85.6% (13,474 genes) and 82.3% (8,440 genes), respectively, with the high infection group consistently yielding slightly more DEGs compared to the general infected group. These findings suggest that cells with elevated viral burden may exhibit broader transcriptional alterations and confirm that our pseudobulk bootstrap framework yields highly consistent DEG results across varying sampling parameters. All result was summarized in Table S4.

Figure 6.

Consistent detection of differentially expressed genes across bootstrap pseudobulk analyses

Four-way Venn diagrams illustrate the overlap of DEGs identified under different bootstrap sampling configurations (500 or 1000 iterations; 150 or 200 cells per replicate).

(A) Sample 30236 was compared between infected and uninfected conditions.

(B) Sample 30236_M for the same contrast.

(C) Sample 30236 in the high-infection versus uninfected comparison.

(D) Sample 30236_M under the high infection vs. uninfected comparison. Across all conditions, a large proportion of DEGs are shared between sampling settings, demonstrating high technical reproducibility. Notably, 85.6% of DEGs (13,474 genes) in panel (C) and 82.3% (8,440 genes) in panel (D) are consistent across all bootstrap replicates.

Elevated viral load induces broad transcriptional remodeling

To elucidate the impact of infection severity on host gene expression dynamics, we conducted a stratified analysis focusing specifically on cells exhibiting high levels of viral burden. Traditional differential expression analyses often compare all infected cells as a single group against uninfected counterparts; however, this approach may mask transcriptional responses uniquely associated with severe infections, particularly when substantial heterogeneity exists within the infected population. To address this limitation, we defined a “high infection” subgroup based on viral transcript abundance and performed parallel differential expression (DE) analyses comparing these cells to uninfected controls.

Using bootstrap-based pseudobulk DE testing across two experimental groups (30236 and 30236_M), we calculated absolute log₂ fold change (|log₂FC|) values under four sampling configurations (500 or 1000 replicates × 150 or 200 cells per group). In all scenarios, the high infection group consistently exhibited significantly greater |log₂FC| values than the general infected group (Wilcoxon test, ∗∗∗p < 0.001), indicating that higher infection burden is associated with more pronounced transcriptional remodeling (Figure 7).

Figure 7.

Comparison of absolute log₂ fold changes (|log₂FC|) across bootstrap pseudobulk DEG analyses under varying infection severity

Bar plots show the mean absolute log₂ fold change (|log₂FC|) ± standard deviation (SD) for differentially expressed genes (DEGs) identified in two samples (30236 and 30236_M) under two infection conditions: infected vs. uninfected (blue) and high infection vs. uninfected (red). Each bootstrap analysis was performed using combinations of 500 or 1000 replicates and 150 or 200 cells per replicate. For both samples, the high infection condition consistently yielded significantly higher |log₂FC| values across all parameter settings (Wilcoxon test, ∗∗∗p < 0.001).

Building on these findings, we next sought to capture subtle, continuous transcriptional changes across the full spectrum of viral load. While the bootstrap pseudobulk framework provided robust group-level contrasts, it was unable to resolve the continuous expression gradients inherent to single-cell data. To address this, we implemented a trend-based differential expression analysis, dividing infected cells into quartiles (Q1-Q4) based on their viral load. This approach enabled us to identify genes whose expression levels changed monotonically with increasing infection severity, providing a more in-depth understanding of how the host transcriptional program adapts to different levels of intracellular pathogen burden.

Viral load positively correlates with genes involved in metronidazole susceptibility and adhesion

In our single-cell RNA sequencing analysis, we identified several virulence- and adhesion-associated genes whose expression positively correlated with intracellular viral load. While the functional roles of these genes particularly in drug metabolism, adhesion, and stress response are well-established in Trichomonas vaginalis, previous studies have reported conflicting findings regarding whether viral infection directly modulates their expression. Using our trend-based single-cell approach, we confirmed that genes implicated in metronidazole (MTZ) activation, including flavodoxin-like fold proteins (TVAGG3_0053980, TVAGG3_0213430, TVAGG3_0585360, TVAGG3_0940660),¹⁵ thioredoxin-like proteins (TVAGG3_0038110, TVAGG3_0191900),¹⁶ disulfide oxidoreductase (TVAGG3_0154220),¹⁷ and nitroreductase family proteins (TVAGG3_0195320, TVAGG3_0427810, TVAGG3_0695620).¹⁸ Interestingly, no differential expression was observed for pyruvate:ferredoxin oxidoreductase (PFO), a well-established factor in MTZ resistance.^19,20 Two annotated PFO precursor genes (TVAGG3_0282970, TVAGG3_0890230) were markedly upregulated in the high viral load group within the control sample. (Figure 8A).

Figure 8.

Viral load-associated expression of functionally characterized gene groups in Trichomonas vaginalis

Trend-based differential expression analysis was performed by stratifying virus-positive single cells into viral load quartiles (Q1-Q4) and comparing each bin to viral load-negative cells.

(A) Drug susceptibility-associated genes, including flavodoxin-like fold proteins (TVAGG3_0053980, TVAGG3_0213430, TVAGG3_0585360, TVAGG3_0940660), thioredoxin-like proteins (TVAGG3_0038110, TVAGG3_0191900), disulfide oxidoreductase (TVAGG3_0154220), nitroreductase family proteins (TVAGG3_0195320, TVAGG3_0427810, TVAGG3_0695620), and pyruvate:ferredoxin oxidoreductase proprotein genes (TVAGG3_0282970, TVAGG3_0890230).

(B) Ubiquitin-related genes (TVAGG3_0129220, TVAGG3_0221350, TVAGG3_0281680, TVAGG3_0586770, TVAGG3_0981600) are potentially involved in protein turnover and cellular stress responses.

(C) Adhesion-associated genes, including TVAGG3_0540230, TVAGG3_0979910, TVAGG3_0163930, TVAGG3_0221780, and the heteropolysaccharide binding protein HPB2 (TVAGG3_0232100), were previously identified as adherence factors. Panels display per-bin expression distributions across viral load quartiles (Q1-Q4), overlaid with median trend lines.

Additionally, ubiquitination-related genes (TVAGG3_0129220, TVAGG3_0221350, TVAGG3_0281680, TVAGG3_0586770, TVAGG3_0981600) exhibited enhanced expression in high-viral load cells (Figure 8B), suggesting that an elevated viral burden may amplify cellular stress response and proteostasis pathways. This finding is consistent with the observation that higher MTZ susceptibility-associated gene expression (Figure 8A) may facilitate a more rapid or pronounced drug-induced cytotoxic response.

For adhesion-associated factors, we observed that elevated viral loads were accompanied by the increased expression of several key adhesion genes, including TVAGG3_0540230, TVAGG3_0979910, TVAGG3_0163930, and TVAGG3_0221780. Notably, the upregulation of TVAGG3_0232100, which encodes the heteropolysaccharide binding protein HPB2, has been recently identified as an important adherence factor²¹ (Figure 8C). These findings provide high-resolution evidence supporting the notion that viral infection modulates key virulence programs in T. vaginalis.

Novel stress- and chromatin-associated genes are upregulated in high viral load conditions

In addition to previously reported genes involved in drug susceptibility and adhesion, our analysis revealed a set of novel genes associated with viral load that may play roles in stress response and chromatin regulation in T. vaginalis Notably, although T. vaginalis lacks a canonical cyst stage, under adverse conditions it can adopt a pseudocyst-like form,²² we observed the upregulation of a myeloid leukemia factor (MLF) homolog (Figure 9A), which has been previously described as an encystation-induced protein in other protozoan parasites.²³ Several MLF-related genes, including TVAGG3_0389720, TVAGG3_0515090, TVAGG3_0703230, TVAGG3_0912280, and TVAGG3_0969640, showed strong positive correlations with viral load, especially in cells treated with metronidazole.

Figure 9.

Novel viral load-associated genes linked to stress response and chromatin regulation in T. vaginalis

(A and B) Trend-based differential expression analysis revealed genes with positive correlations to viral load, including (A) myeloid leukemia factor (MLF)-related genes (TVAGG3_0389720, TVAGG3_0515090, TVAGG3_0703230, TVAGG3_0912280, TVAGG3_0969640) and (B) multiple histone-related genes (e.g., TVAGG3_0221660, TVAGG3_0309740, TVAGG3_0408090, TVAGG3_0473430). Boxplots show gene expression distributions across viral load quartiles (Q1-Q4), with median trend lines overlaid.

In parallel, we detected a marked upregulation of histone-related genes (Figure 9B; e.g., TVAGG3_0221660, TVAGG3_0309740, TVAGG3_0408090, TVAGG3_0473430, TVAGG3_0504870, TVAGG3_0571710, TVAGG3_0504880, TVAGG3_0804230, TVAGG3_0804220, TVAGG3_0829970, TVAGG3_0874660, TVAGG3_0963230, TVAGG3_0963220, TVAGG3_0976870, and TVAGG3_1000540) in high viral load cells. These genes predominantly encode core histone proteins (H2A, H2B, H3, and H4), suggesting potential alterations in chromatin structure and nucleosome dynamics. The observed expression pattern may reflect a stress-induced chromatin remodeling response or transcriptional reprogramming triggered by viral burden. These findings not only broaden our understanding of T. vaginalis-virus interactions but also identify candidate genes involved in epigenetic regulation under viral and drug-related stress.

Discussion

Our study provides compelling evidence that the distribution of Trichomonasvirus (TVV) within T. vaginalis is not only highly heterogeneous but also dynamically regulated at the single-cell level. Initial single-cell RNA sequencing (scRNA-seq) of the TVV-positive isolate ATCC 50148 revealed striking heterogeneity in viral distribution, with a substantial proportion of cells lacking detectable viral transcripts despite originating from the same clonal population. To validate this observation, we performed immunofluorescence staining for double-stranded RNA (dsRNA), which further confirmed the presence of distinct virus-positive and virus-negative subpopulations. These results fundamentally challenge the long-standing assumption that TVV is uniformly present in all cells of a positive isolate, a notion that has underpinned much of the prior transcriptomic and functional studies on TVV. within TVV-positive isolates, highlighting the need to assess infection status at the single-cell level. Notably, in the virus-negative isolate ATCC 50143, a weak dsRNA signal was detected, likely representing non-viral endogenous small RNAs (e.g., tRNA fragments²⁴ or siRNAs²⁵), underscoring the necessity for stringent validation to distinguish true viral infection from background.

To determine whether the observed heterogeneity was a stable feature or reflected dynamic shifts over time, we examined three independently cultured biological replicates of the TVV-positive isolate ATCC PRA-98. scRNA-seq profiling of these replicates revealed striking temporal variation in both the proportion of virus-positive cells and the dominant TVV subtypes. These findings demonstrate that the distribution of Trichomonasvirus within T. vaginalis is inherently dynamic, likely shaped by intercellular viral exchange, cell division, or intrinsic regulatory mechanisms.

This temporal plasticity raised an important question: can such variation in viral burden within a genetically uniform population influence host biology in a meaningful way? To address this, we performed ultra-deep scRNA-seq on the TVV-positive isolate ATCC 30236, which allowed high-resolution quantification of viral transcripts per cell. This analysis revealed a gradient of viral load among infected cells, providing an ideal system to dissect the dose-dependent effects of TVV on host gene expression.

The inherent heterogeneity of viral distribution has significant implications for interpreting previous transcriptomic and proteomic studies, which often assumed uniform infection status within isolates. Unrecognized variability in viral load or the proportion of infected cells may have contributed to previously conflicting conclusions regarding TVV’s influence on drug response and gene regulation. By using a single-isolate model under controlled culture conditions, our study avoided inter-isolate confounders and enabled direct interrogation of within-population viral effects.

Within this refined framework, we observed that TVV presence was associated with increased metronidazole susceptibility and modest upregulation of adhesion-related genes. In contrast, canonical virulence factors such as cysteine proteases were not consistently induced at the transcript level.²³ Instead, our approach uncovered two previously uncharacterized transcriptional modules responsive to viral burden: the myeloid leukemia factor (MLF)-like gene family and histone-related genes. MLF homologs, previously implicated in stress responses and encystation-like states in Giardia lamblia,^22,26 were specifically upregulated in high viral load cells and in metronidazole-treated conditions, suggesting a role in pseudocyst-like adaptations. Likewise, the coordinated induction of core histones (H2A, H2B, H3, H4) in high viral-load cells implies that chromatin remodeling may be a downstream consequence of infection, potentially altering cell cycle progression, transcriptional plasticity, or stress adaptation in T. vaginalis.^27,28 These findings echo histone-mediated regulatory mechanisms observed in Plasmodium falciparum²⁹ and Toxoplasma gondii,³⁰ pointing to conserved epigenetic responses to intracellular infection across protozoan parasites. Together, these findings broaden our understanding of T. vaginalis-virus interactions and highlight candidate genes involved in epigenetic regulation under conditions of viral and drug-related stress.

Although TVV-positive and TVV-negative cells exhibited significant differences in viral transcript abundance and, in the expression of selected host genes (e.g., HPB2, MLF-like genes), their global transcriptional profiles did not form discrete clusters based on infection status in unsupervised UMAP or t-SNE projections. This likely reflects the continuous and subtle nature of TVV-driven transcriptional remodeling, which does not dominate overall cellular variance. To address this, we implemented a dual-pronged analytical framework to more sensitively detect infection-associated transcriptional changes. First, a bootstrap-based pseudobulk strategy aggregated randomly sampled single cells to mitigate sparsity and enhance statistical robustness. Second, we stratified virus-positive cells into viral load-based bins and performed trend-based differential expression analysis to uncover gradual, monotonic gene expression changes across the infection gradient. This integrative approach allowed us to resolve both condition-specific shifts and viral load-dependent programs that would otherwise be masked in global clustering analyses.^31,32 To detect these nuanced transcriptional responses, we employed a dual-pronged analysis pipeline. First, a bootstrap-based pseudobulk framework mitigated sparsity and technical variability by aggregating randomly sampled single cells into stable expression profiles. Second, we stratified virus-positive cells into viral load-based bins and applied trend-based differential expression analysis to identify gradual, monotonic changes across the infection gradient. This integrative strategy enabled us to capture both condition-specific shifts and load-dependent transcriptional programs with enhanced interpretability and robustness.

dsRNA viruses have also been described in other protozoan parasites associated with human disease, including Leishmania,^{33,34,35,36,37} Giardia,^38,39,40 Cryptosporidium,^41,42,43 and Plasmodium vivax,⁴⁴ suggesting a potentially conserved role for persistent viral infections across diverse parasitic species. Building on our findings, such endogenous protozoan viruses may represent previously overlooked modulators of parasite biology and virulence-and could even be leveraged as therapeutic targets or delivery platforms in future antiparasitic strategies.

Finally, our findings have broader implications for both experimental design and therapeutic exploration. The observed heterogeneity in viral presence and load within clonal parasite populations highlights the need to account for intra-isolate variation in future studies. At the same time, a major current limitation is the lack of an established system to experimentally clear TVV from T. vaginalis and subsequently reintroduce the virus in a controlled manner. To address this gap, we are developing a reinfection platform that includes virus-cleared T. vaginalis lines and extracellular vesicle-mediated TVV delivery.⁴⁵ This system will allow controlled manipulation of viral load, decoupled from isolate-specific genetic backgrounds, thereby facilitating rigorous functional assays to dissect the causal roles of TVV in host biology, immune modulation, and drug susceptibility.

In conclusion, our study demonstrates that TVV distribution in T. vaginalis is neither static nor homogeneous. Through ultra-deep single-cell transcriptomics and robust computational analyses, we uncovered previously underappreciated complexity in TVV-host interactions. These data not only revealed novel host pathways linked to stress adaptation and immune evasion, but also established a foundational framework for future mechanistic investigations. Ultimately, our work underscores the importance of considering persistent viral infections as dynamic modulators of protozoan pathobiology, with potential implications for understanding disease progression and identifying new therapeutic targets.

Limitations of the study

While our single-cell transcriptomic approach offers unprecedented resolution in delineating Trichomonasvirus (TVV) distribution and corresponding host transcriptional responses, several limitations should be acknowledged. First, this study relies on Trichomonas vaginalis isolates naturally infected with TVV; the absence of a system that enables complete viral clearance and controlled reinfection precludes direct causal validation of TVV-mediated effects. Second, despite extensive bootstrap-based statistical analyses, the intrinsic sparsity and stochastic dropout in single-cell RNA sequencing may have masked low-abundance transcripts or transient virus-host interactions. Third, all observations were derived from in vitro cultures, which may not fully replicate in vivo infection dynamics or the influence of host immune factors. Future development of experimental systems that permit precise TVV manipulation, coupled with integrative multi-omics analyses, will be critical for elucidating the mechanistic underpinnings of viral load-dependent phenotypic heterogeneity.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Petrus Tang (petang@mail.cgu.edu.tw).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Data: All RNA sequencing data used in this study have been deposited in the NCBI SRA database. Publicly available bulk RNA-seq datasets include project PRJNA280779 (NYCB20A, B7268, ATCC 30236, GOR69, NYCF20, NYCG31, NYCA04, SD2, NYCC37, NYCD15), SRR585698 (TO16), and SRX1686500 (NYCE32_8). In addition, sequencing data generated in this study have been submitted to the SRA under the following accessions: scRNA-seq raw data PRJNA1238186 and bulk RNA-seq raw data PRJNA1241774.
• •Code: code used in generating the results and figures in this article are publicly available at https://doi.org/10.5281/zenodo.16271640.
• •Additional information: Any additional information required to reanalyze the data reported herein is available from the lead contact upon request.

Acknowledgments

We would like to thank the Microscopy Center at Chang Gung University for its technical assistance in laser confocal microscopy and Prof. Min-Chi Chen from the Department of Public Health and Biostatistics Consulting Center, School of Medicine, Chang Gung University, for assistance with biostatistical advice. This research was supported by the Chang Gung Memorial Hospital. (CMRPD1J0311∼1J0CMRPD1J0311∼1J0313313, CMRPD1M0571∼1M0572) and National Science and Technology Council of Taiwan (NSTC 113-2320-B-182-017-MY3). H.W.L. is a recipient of Doctoral Scholarship from the Ministry of Education, Taiwan.

Author contributions

H.W.L. and S.C.O., designed and performed experiments. H.W.L. and J.W.S., analyzed the data. C.Y.T., P.J.H., C.C.L., Y.M.Y., and R.L. provided resources and methodologies. H.W.L. and P.T. wrote the article. P.T. acquired funding.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-dsRNA [9D5]	Absolute Antibody	https://www.antibodyregistry.org/AB_2920603
Goat anti-Rabbit IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 594	Thermo Fisher Scientific	RRID: AB_2534079
Chemicals, peptides, and recombinant proteins
Trypan Blue Solution	Thermo Fisher Scientific	Cat#15250061
(+)-Sodium L-ascorbate	Sigma-Aldrich	Cat#A4034
L-Cysteine	Sigma-Aldrich	Cat#C7352
D(+)-Glucose anhydrous	PanReac AppliChem	Cat#141341
Yeast Extract	Thermo Fisher Scientific	Cat#211929
Sodium chloride	Honeywell Research Chemicals	Cat#S9888 Fluka
Potassium phosphate monobasic	Sigma-Aldrich	Cat# P5655
Potassium phosphate dibasic	J.T.Baker	Cat#15568104
SuperScript™ III Reverse Transcriptase	Thermo Fisher Scientific	Cat#18080085
FluoroVue™ Nucleic Acid Safe stain	SMOBIO Technology	Cat#NS1000
Poly-L-lysine	Sigma-Aldrich	Cat#P4707
Formaldehyde	63 Pure Chemicals	Cat#0007
Phosphate buffered saline	Sigma-Aldrich	Cat#P3813
Tween 20	Thermo Fisher Scientific	Cat#85113
Triton X-100	Thermo Fisher Scientific	Cat#85111
Bovine Serum Albumin	Sigma-Aldrich	Cat#A5611-10G
Critical commercial assays
GENEzol™ TriRNA Pure Kit	Geneaid Biotech	Cat#GZXD100
Universal Plus™ mRNA-Seq with NuQuant® Kit	Tecan Life Sciences	Cat#0520-A01
Chromium Next GEM Single Cell 3ʹ LT Reagent Kits v3.1 (Dual Index)	10X Genomics	Cat#PN-1000325
Chromium Next GEM Single Cell 3ʹ Reagent Kits v3.1 (Dual Index)	10X Genomics	Cat#PN-1000269
2X Super Hi-Fi Taq PCR MasterMix with loading dye	Biotools	Cat#KTT-BB05
FluoroQuest™ Antifade Mounting Medium	AAT Bioquest	Cat#20007
Deposited data
T. vaginalis reference genome	NCBI	https://ftp.ncbi.nlm.nih.gov/genomes/all/GCF/026/262/505/GCF_026262505.1_NYU_TvagG3_2/
Trichomonas vaginalis virus 1 reference genomes	NCBI virus	Taxid:674953
Trichomonas vaginalis virus 2 reference genomes	NCBI virus	Taxid:674954
Trichomonas vaginalis virus 3 reference genomes	NCBI virus	Taxid:170965
Trichomonas vaginalis virus 4 reference genomes	NCBI virus	Taxid:1008292
Trichomonas vaginalis virus 5 reference genomes	NCBI virus	Taxid:3047136
ScRNA-seq Raw Data	NCBI SRA	PRJNA1238186
Bulk RNA-seq Raw Data	NCBI SRA	PRJNA1241774
Bulk RNA-seq Raw Data (NYCB20A, B7268, ATCC 30236, GOR69, NYCF20, NYCG31, NYCA04, SD2, NYCC37, NYCD15)	NCBI SRA	PRJNA280779
Bulk RNA-seq Raw Data (TO16)	NCBI SRA	SRR585698
Bulk RNA-seq Raw Data (NYCE32_8)	NCBI SRA	SRX1686500
Experimental models: Organisms/strains
Trichomonas vaginalis ATCC PRA-98	ATCC	G3
Trichomonas vaginalis ATCC 30236	ATCC	JH 31A #4
Trichomonas vaginalis ATCC 50148	ATCC	NYH 286
Oligonucleotides
Primer: TVV1F2875 ATTAGCGGTGTTTGTGATGCA	Goodman et al.46	N/A
Primer: TVV1R3443 CTATCTTGCCATCCTGACTC	Goodman et al.46	N/A
Primer: TVV2F2461 GCTTGAGCACTGCTCGCG	Goodman et al.46	N/A
Primer: TVV2R3068 TCTCTTTTGGCATCGCTT	Goodman et al.46	N/A
Primer: TVV3F61 AAATTAATCAACACCCTCC	Goodman et al.46	N/A
Primer: TVV3R482 CAGATCACTTTGTGTGTC	Goodman et al.46	N/A
Primer: TVV4F1338 ATGCCAGTTGCTTTCCG	Goodman et al.46	N/A
Primer: TVV4R1834 TTCCCCAATAGTTATCAG	Goodman et al.46	N/A
Software and algorithms
QIAGEN CLC Genomics Workbench (v20.0.3)	QIAGEN Digital Insights	N/A
Cell Ranger v9.0.0	10x Genomics	https://github.com/10XGenomics/cellranger
Loupe Browser v8.0	10x Genomics	https://www.10xgenomics.com/support/software/loupe-browser/latest
Seurat v5.1.0	https://doi.org/10.1038/s41587-023-01767-y	https://satijalab.org/seurat/
R v4.4.1	R	https://www.r-project.org/
R studio v4.4.1	R studio	http://www.rstudio.com/.
BiocManager v1.30.26	https://doi.org/10.1186/gb-2004-5-10-r80	https://www.bioconductor.org/
edgeR v4.3.0	https://doi.org/10.1093/bioinformatics/btp616	https://bioconductor.org/packages/edgeR
ggVenn v0.1.10	https://doi.org/10.32614/CRAN.package.ggvenn	https://github.com/yanlinlin82/ggvenn
tidyverse v2.0.0	https://doi.org/10.21105/joss.01686.	https://dplyr.tidyverse.org/
Matrix v1.7-0	https://doi.org/10.32614/CRAN.package.Matrix	https://cran.r-project.org/web/packages/Matrix
Magick v2.8.7	https://doi.org/10.32614/CRAN.package.magick	https://cran.r-project.org/web/packages/magick/index.html
Patchwork v1.3.1	https://doi.org/10.32614/CRAN.package.patchwork	https://cran.r-project.org/web/packages/patchwork/index.html
Graph pad PRISM v10	GraphPad	https://www.graphpad.com/; RRID: SCR_002798
Custom scripts	This paper	https://doi.org/10.5281/zenodo.16271640
Other
NanoDrop™ 2000	Thermo Fisher Scientific	Cat#ND-2000
96-Well Thermal Cycler	Applied Biosystems	Cat#12333653
ZEISS LSM510 META NLO Microscope	ZEISS	N/A
Illumina NovaSeq 6000	Illumina	N/A
Illumina NovaSeq X Plus	Illumina	N/A
Bioanalyzer 2100	Agilent	N/A

Method details

Trichomonas vaginalis culture

The T. vaginalis isolates (ATCC 30236, ATCC50143, ATCC 50148, ATCC PRA-98) used in this study were obtained from the American Type Culture Collection (ATCC, Manassas, Virginia, USA). The cells were initially cultured at a density of approximately 5 × 10⁵ cells/mL and maintained axenically in yeast extract, iron-serum (YI-S) medium⁴⁷ at 37°C. For RNA-sequencing experiments, cells were harvested during the mid-exponential growth phase (1.8-2.3 × 10⁶ cells/ml). The viability of the cells was assessed using the trypan blue exclusion assay.

RNA extraction and quality assessment

Total RNA was extracted from mid-log phase cultures of each T. vaginalis isolate using the GENEzol TriRNA Pure Kit (Geneaid Biotech Ltd., New Taipei City, Taiwan) according to the manufacturer’s instructions. Briefly, cells were harvested from logarithmic phase cultures, the supernatant was discarded, and the cell pellet was washed with PBS. The pellet was then incubated with GENEzol reagent for 5 min. An equal volume of absolute ethanol was subsequently added and gently mixed. The mixture was transferred to RB Columns and centrifuged at 14,000 × g for 1 min. The RB Columns were then sequentially rinsed with a pre-wash buffer followed by a wash buffer. Finally, purified RNA was eluted from the RB Columns using 20 μL of RNase-free water. RNA concentration was measured using a NanoDrop spectrophotometer (Thermo Fisher Scientific, Massachusetts, USA). The 230/280 and 260/280 absorbance ratios, calculated by the NanoDrop software, were used to assess RNA quality and potential protein contamination.

Reverse transcription-polymerase chain reaction (RT-PCR)

The total RNA was reverse-transcribed using the SuperScript III Reverse Transcriptase (Thermo Fisher Scientific, Massachusetts, USA). The target genes of Trichomonasvirus were amplified using species-specific primers: TVV1F2875-TVV1R3443, TVV2F2461-TVV2R3068, TVV3F61-TVV3R482, and TVV4F1338-TVV4R1834, as designed by Goodman et al.⁴⁶ The amplification was performed with the 2× Super Hi-Fi Taq PCR MasterMix with loading dye (BIOTOOLS Co., Ltd., New Taipei City, Taiwan). Each reaction mixture (25 μL) contained 10 μM of each primer, 12.5 mL of the 2 × Super Hi-Fi Taq PCR MasterMix with loading dye, and 1 μg of the template derived from reverse transcription, following the manufacturer’s protocol. The thermal cycling conditions were as follows: inactivate reverse transcriptase at 94°C for 3 min, followed by 30 cycles of 94°C for 30 s, 55°C for 30 s, and 72°C for 1 min, concluding with a final extension step at 72°C for 5 min. The resulting amplified cDNA fragments were visualized after separation in a 1% agarose gel prepared with Tris-Acetate-EDTA (TAE) buffer with 0.01% FluoroVue™ Nucleic Acid Safe stain (SMOBIO Technology Inc., Hsinchu City, Taiwan).

Bulk RNA sequencing

The extracted RNA was used for library construction following a directional library preparation protocol. cDNA libraries were manually prepared using the Universal Plus mRNA-Seq with NuQuant Kit (Tecan, Männedorf, Zürich, Switzerland) and sequenced on the Illumina system (Illumina, Inc., San Diego, CA, USA).

Detection of Trichomonasvirus in T. vaginalis RNA-seq data

RNA-seq reads were aligned to the reference genomes of TVV species (TVV1-5) downloaded from the National Center for Biotechnology Information (NCBI) virus database. The analysis was performed using the RNA-Seq Analysis module in the CLC Genomics Workbench. Alignment parameters were set as follows: masking mode = no masking, match score = 1, mismatch cost = 3, linear gap cost model with insertion and deletion costs = 3, length fraction = 0.9, and similarity fraction = 0.8. The results section details the TVV species successfully aligned to their corresponding reference sequences among the T. vaginalis isolates.

Indirect immunofluorescent assay

A total of 3 × 10^∧6 trichomonads were inoculated onto a 0.1% Poly-L-lysine (Merck, Darmstadt, Germany) coated slide and incubated for 15 min at 37°C in an anaerobic chamber. After incubation, the medium was carefully removed, and the cells were fixed with 4% formaldehyde (FA) for 10 min at room temperature. The cells were then gently washed twice with 0.1% PBS-Tween20 (PBST) and permeabilized with 0.1% Triton X-100 in PBS, followed by blocking with 3% bovine serum albumin (BSA). Next, the cells were incubated for 1 h at room temperature with the anti-dsRNA [9D5] primary monoclonal antibody (Absolute Antibody Ltd, Cleveland, UK) at a 1:1200 dilution in blocking solution. Following two washes with PBST, a goat anti-rabbit secondary antibody conjugated with Alexa Fluor 594 (Thermo Fisher Scientific Inc., Massachusetts, USA) was applied at a 1:1000 dilution, along with DAPI (4′,6-diamidino-2-phenylindole). The sample was incubated for an additional hour at room temperature. After two more washes with PBST, the slides were air-dried, and 4 μL of FluoroQuest™ Antifade Mounting Medium (AAT Bioquest, Inc., California, USA) was applied to mount the slides. Stained slides were visualized using a ZEISS LSM510META NLO Microscope (ZEISS, Stuttgart, Germany).

Optimization of sub-lethal metronidazole concentrations

To identify a metronidazole concentration suitable for single-cell transcriptomic profiling while preserving parasite viability, we performed a series of dose-response assays using the Trichomonas vaginalis ATCC 30236 isolate, which harbors endogenous TVV1. Exponentially growing trophozoites were seeded at a density of 1 × 10⁶ cells/mL and exposed to metronidazole (Sigma-Aldrich, Missouri, USA) at final concentrations of 2, 4, 6, and 8 μM. Cell density and viability were assessed every 3 h over a 24-h period using trypan blue exclusion and manual counting with a hemocytometer. Growth inhibition was evaluated by comparing treated samples to untreated controls at each time point.

Single-cell RNA sequencing

Single-cell RNA sequencing (scRNA-seq) was conducted using three Trichomonas vaginalis isolates: ATCC 30236, ATCC PRA-98, and ATCC 50148. For ATCC 30236, two experimental conditions were prepared: an untreated control and cells treated with 6 μM metronidazole for 1 h, as determined through prior dose optimization. All samples were harvested during the mid-logarithmic growth phase, and cell viability was assessed using trypan blue exclusion to ensure high-quality input for downstream processing.

Single-cell encapsulation and library construction were performed using the 10x Genomics Chromium Single Cell 3′ platform following the manufacturer’s protocol. Cells were loaded onto a Chromium Next GEM Chip and processed on a Chromium Controller (10x Genomics, Pleasanton, CA, USA). For ATCC 30236, the Next GEM Single Cell 3' Reagent Kits v3.1 (Dual Index) were used; for ATCC PRA-98 and ATCC 50148, the v3.1_LT (Dual Index) version of the reagent kits was employed. Individual cells were encapsulated alongside Single Cell 3′ Gel Beads containing barcoded oligonucleotides with unique molecular identifiers (UMIs). Library construction involved cell lysis, barcoded reverse transcription, cDNA amplification, fragmentation (∼200 bp), adapter ligation, and dual indexing.

Sequencing was performed on the Illumina NovaSeq X Plus platform for ATCC 30236 samples and on the NovaSeq 6000 platform for ATCC PRA-98 and ATCC 50148 (Illumina, San Diego, CA, USA). To enable detailed transcriptomic profiling, ultra-deep coverage of approximately 100 gigabases per sample was obtained for ATCC 30236, while ∼35 gigabases per sample were generated for ATCC PRA-98 and ATCC 50148.

To assess temporal variation in Trichomonasvirus distribution within clonal populations, three independent biological replicates of ATCC PRA-98 were cultured, processed, and sequenced at distinct time points. These time-resolved scRNA-seq datasets allowed for downstream analysis of dynamic changes in viral subtype prevalence and infection state heterogeneity.

Single-cell RNA-seq reference construction and preprocessing

Preliminary data processing was performed using the Cell Ranger pipeline (version 7.1.0, 10x Genomics) within a Linux-based computational environment. A custom reference genome was constructed by integrating genome assemblies and corresponding annotation files for T. vaginalis and its associated Trichomonasvirus (TVV) species, retrieved from the NCBI FTP server and the NCBI Virus database. Manual curation was applied to refine gene annotations and ensure consistency across all chromosomes and contigs. For viral references, full-length sequences of each TVV species were incorporated to enable species-specific transcript detection. Sequencing reads were aligned to the combined host-virus reference to quantify both parasite and viral transcripts. The viral infection status at the single-cell level was determined based on the presence of virus-specific transcripts, allowing individual cells to be stratified accordingly. An initial assessment of viral transcript distribution was performed using the Loupe Browser (version 8.0, 10x Genomics) to verify species-specific expression and evaluate the overall quality of the viral signal.

Subpopulation classification in single-cell RNA-seq data

To define infection status at the single-cell level, we analyzed raw viral transcript counts in two virus-exposed sample groups derived from the T. vaginalis ATCC 30236 isolate, designated as 30236 and 30236_M. Gene expression matrices were first filtered to retain high-quality cells based on RNA content (nFeature_RNA >200 and <5000, nCount_RNA <20,000), and downsampled to equalize cell numbers. Data normalization was performed using the NormalizeData function in the Seurat package (version 5.1.0) with default parameters to enable downstream comparative analyses.

The viral load for each cell was calculated by summing the raw UMI counts of viral genes, dividing by the total cellular UMI count, and then multiplying by 10,000 to yield a normalized viral transcript index. Cells with a viral load of zero were categorized as uninfected. Among the remaining cells exhibiting non-zero viral load, the top 25% with the highest values were defined as the high-infection group, while the rest were assigned to the infected group. This stratification enabled downstream comparisons across three infection states: uninfected, infected, and high-infection.

Bootstrap-based pseudobulk differential expression analysis

To overcome data sparsity and enhance the reproducibility of DEG detection from scRNA-seq data, we implemented a bootstrap-based pseudobulk framework using edgeR (version 4.3.0) to evaluate differential gene expression (DEG) from single-cell RNA-seq data, to improve robustness to cell-to-cell variability. For each biological condition, individual cells were randomly sampled and aggregated to construct pseudobulk replicates. Four combinations of sampling parameters were tested: 500 or 1000 bootstrap iterations, and 150 or 200 cells per pseudobulk replicate. In each iteration, pseudobulk expression matrices were generated for both experimental and control groups. Differential expression analysis was then conducted using the quasi-likelihood negative binomial framework implemented in edgeR (glmQLFit and glmQLFTest),⁴⁸ and DEGs were defined as genes with an adjusted false discovery rate (FDR) < 0.05.

To assess the consistency of DEG detection across different bootstrap settings, we generated four-way Venn diagrams using the ggvenn (version 0.1.10) R package. DEG sets from each bootstrap configuration were compared within each sample (30236 or 30236_M) and each contrast (infected vs. uninfected or high infection vs. uninfected). Gene intersections and the corresponding percentages relative to the total union were visualized to quantify reproducibility.

Comparison of absolute log₂ fold changes across infection severities

To investigate whether infection severity influenced the magnitude of differential gene expression by analyzing the absolute values of log₂ fold changes (|log₂FC|) for genes identified as DEGs in both comparisons. For each gene, |log₂FC| values were summarized across all bootstrap replicates. The mean ± standard deviation of |log₂FC| was computed for each group (infected vs. uninfected and high infection vs. uninfected) under each bootstrap configuration. Differences in expression magnitude between groups were statistically evaluated using the Wilcoxon signed-rank test.

Trend-based differential expression analysis

Building upon the bootstrap-based DEG framework, we extended our analysis to explore viral load-associated transcriptional dynamics using edgeR (v4.3.0). Specifically, we stratified virus-positive cells into equal-sized bins to represent increasing viral burden. In this analysis, cells were divided into four equally sized groups based on ascending viral load, rather than conventional statistical quartiles. For each bin, we performed differential expression analysis by comparing virus-positive cells within the bin to virus-negative cells using edgeR’s quasi-likelihood framework (glmQLFit and glmQLFTest). Genes expressed in at least five cells were retained for analysis.

Bin-wise differential expression results were aggregated into a per-gene trend summary table capturing multiple metrics, including the average log fold change across bins (avg_logFC), the minimum observed FDR (min_FDR), and the Spearman correlation coefficient between bin rank and logFC. For each gene, we also quantified the number of bins exhibiting significant upregulation or downregulation, defined as FDR <0.1 with a positive or negative logFC, respectively. Genes were classified as upregulated if they showed an average logFC greater than 0.25, a minimum FDR less than 0.1, and significant upregulation in at least two bins. Conversely, genes with an average logFC less than −0.25, a minimum FDR below 0.1, and significant downregulation in two or more bins were classified as downregulated.

This edgeR-based approach enabled robust detection of differentially expressed genes in response to viral load gradients, capturing not only binary condition-specific effects but also continuous transcriptional shifts across bins. While this strategy yielded a broad set of candidate genes, downstream interpretation focused on a targeted subset previously implicated in drug resistance, virulence, or stress response in T. vaginalis, thereby enhancing the biological relevance of the results beyond purely computational output.

Quantification and statistical analysis

All statistical analyses were performed using R (version 4.4.1; https://www.r-project.org/) and GraphPad Prism 10 (GraphPad Software, San Diego, CA, USA). Temporal variation in viral subtype composition across ATCC PRA-98 samples was assessed using the Chi-square test in GraphPad Prism. Wilcoxon signed-rank tests were applied to compare absolute log₂ fold changes between infection groups. Differential expression analysis was conducted using the quasi-likelihood framework implemented in the edgeR package (glmQLFit and glmQLFTest functions). Spearman’s correlation coefficients were used to assess monotonic trends between viral load and gene expression. The use of technical and biological replicates is detailed in the figure legends.

Published: November 27, 2025