Study participants and background dataWe analyzed plasma cytokine levels, and microbiome data of the same study participants (N = 241) from our previous investigation21, considering the participants’ reported sexual activity. The cohort was exclusively MSM, and the average age (±standard deviation [SD]) of study participants was 41 ± 16 years (Table 1). The majority of these 1984–1985 MACS participants were white (95%) and achieved undergraduate or post-graduate college degrees (63%). The group included a higher number of non-smokers, with 57% either having never smoked or were former smokers, and current smokers comprising 42% of the total. A higher percentage of participants (56%) were categorized as heavy drinkers (consumption of three or more drinks per day at least once a month), as opposed to low drinkers (42%, defined as consumption of less than two drinks per day no more than once a month). Prevalent oral bacterial antibiotic use was reported by 53% of the cohort. Approximately 16% of participants reported having had at least one sexually transmitted infection (STI), including syphilis, non-specific/nongonococcal urethritis, chlamydia, herpes, and gonorrhea. A notable proportion (81%) reported past use of one or more substances including marijuana, poppers, cocaine, uppers, ecstasy, heroin, speedball, phencyclidine, downers, ethyl chloride, gamma-hydroxybutyrate, or other unspecified drugs. About 40% of the participants were hepatitis B virus (HBV) negative (anti-hepatitis B virus core (anti-HBc) negative and hepatitis B surface antigen (HBsAg) negative), 55% had resolved HBV infection (anti-HBc positive and HBsAg negative), and 3% were HBV positive at the time of the survey (HBsAg positive). About 95% of participants were hepatitis C virus antibody negative, 2.5% were positive, and the remaining 2.5% had missing test results. Participants were categorized into four ordered groups (G1 to G4) based on the number of partners with whom they engaged in receptive anal intercourse (see Methods for detailed definitions).Table 1 Summary of demographic, clinical, and behavior featuresExploratory analysesAll the data of this study were collected in the early years of the AIDS epidemic before the availability of effective anti-HIV-1 therapy. We found that the HIV-1 infection rates monotonically increased with the number of sexual partners with whom a participant had receptive anal intercourse during the previous 2 years (p < 0.001 Fig. 1). This finding suggests that the risk of HIV-1 infection increases with the number of partners with whom a man had receptive anal intercourse.Fig. 1: HIV-1 infection monotonically increases with the number of receptive anal intercourse partners.The X-axis categorizes participants (N = 241) into exposure groups, ranging from Group 1 (G1) through Group 4 (G4). The Y-axis represents the percentage distribution of the participants with subsequent HIV-1 infection (Case, depicted in red) and negative controls (Ctrl, depicted in blue). Combined, these percentages total 100%. As participants move from G1 to G4, indicative of an increase in the number of partners, there is a marked rise in the percentage of HIV infection. This increase is statistically significant supported by the likelihood ratio test (LRT) statistics of 0.0021 and a p-value (one-sided) of less than 0.001, as determined by the constrained linear mixed effects (CLME) test, suggesting a monotonic increasing trend.There were no statistically significant differences observed among the four sexual exposure groups, or between the HIV-1 infection status, with respect to mean ages, alcohol consumption status, education level, and smoking status (Fig. 2). However, recreational substance use was positively associated with the number of receptive anal intercourse partners (p < 0.01, Fig. 2a) and the risk of HIV-1 infection (p = 0.02, Fig. 2b). These findings suggest that substance use acts as a potential confounder between the sexual exposure groups and outcome variables. Antibiotic usage was positively associated with sexual exposure groups (p = 0.03, Fig. 2a) and with HIV-1 infection (p = 0.01, Fig. 2b). We hypothesized that antibiotic usage did not directly impact HIV-1 infection but instead, influenced levels of the biomarkers (cytokines and gut microbiome). Consequently, instead of treating antibiotic usage as a confounder, we treated it as a covariate in the subsequent analyses between the sexual exposure groups and biomarkers. Both HBV infection and history of STI positively correlated with HIV-1 infection but were not correlated with the number of partners with whom a participant had receptive anal intercourse (Fig. 2). Hence, rather than being confounders in this study, they could serve as exposures in a different causal pathway leading to HIV-1 infection.Fig. 2: Forest plot depicting the associations of demographic, clinical, and behavior features with exposures and the outcome.This forest plot visualizes the associations of participants’ features (N = 241) with a sexual exposure groups, analyzed using ordinal logistic regression models, and b the HIV-1 infection status outcome, analyzed using logistic regression models. The X-axis denotes the odds ratio (OR), while the Y-axis lists the various demographic and clinical characteristics. Each feature’s effect size (OR) is symbolized by a diamond. An accompanying horizontal line represents the 95% confidence interval (CI), indicating the range in which the actual effect size is likely to reside. A vertical red line at the OR of 1.0 serves as a reference for no effect. For each feature, the exact OR, 95% CI, and the p-value (two-sided)—derived from logistic regression models—are presented in an adjacent table.Association analysesTo evaluate whether changes in biomarker levels, namely, inflammatory cytokines, and the gut microbiome mediate the effects of sexual behavior on HIV-1 infection, we first examined the associations between levels of these biomarkers with both exposure and outcome. Notably, our previously published work21 detailed associations between each of these biomarkers and HIV-1 infection status. In this current study, we (a) investigated the association between the biomarkers, namely, cytokines, and microbiome, and the sexual exposure groups, and (b) reanalyzed the association between the microbiome and HIV-1 infection status using ANCOM-BC224, a recently developed methodology that has better performance characteristicsSexual exposure groups and plasma inflammatory cytokine levelsFor each participant, we assessed plasma for sCD14, soluble scavenger receptor CD163 (sCD163), interleukin 6 (IL-6), and lipopolysaccharide-binding protein (LBP) to detect monotonic increasing trends with an increase in the number of partners with whom a participant had receptive anal intercourse. As shown in Fig. 3a–c, there was a significant increasing trend in sCD14 levels with sexual exposure groups (p = 0.014), and marginally significant increasing trends in C-reactive protein (CRP) (p = 0.083) and sCD163 (p = 0.068) with sexual activity groups.Fig. 3: Bar plots demonstrating monotonically increasing trends between exposure groups and cytokines levels.Effect sizes of a CRP, b sCD14, and c sCD163 levels from participants’ plasma samples (N = 241) were shown here. The X-axis details the exposure groups of participants (N = 241), from G1 to G4. The Y-axis reflects the cytokine’s effect size determined by the CLME model. It is important to note that these are not raw concentrations but fitted values under a monotonic trend. Each bar’s error bars denote the 95% CI. Pairwise p-values (one-sided), used for contrasting the exposure groups, are displayed above the brackets that span the respective bars. The p-value for a monotonically increasing trend is provided within the plot.Sexual exposure groups and gut microbiomeA total of 13,073,544 sequence reads were generated for the 243 stool samples, with a median of 51,125 (range 67–126,903) reads per stool sample. The analyses did not demonstrate significant associations between alpha diversity metrics, including richness and the Shannon diversity index, and the ordered sexual exposure groups. This held true across both an increasing trend (Supplementary Fig. 2a, b) and a decreasing trend (Supplementary Fig. 2c, d). Furthermore, using the Bray–Curtis dissimilarity, no discernible differences were observed in Beta diversity across sexual activity groups at the species level (Supplementary Fig. 3).The ANCOM-BC2 pattern analysis was employed to evaluate monotonic increasing or decreasing trends in abundances of microbial species with sexual exposure groups (Fig. 4a). As the number of partners with whom a participant has receptive anal intercourse increased from Group 1 to Group 4, we discovered a significant decreasing trend (p < 0.05) in the abundance of some of the well-known commensal bacteria and those involved in the production of short-chain fatty acids such as A. muciniphila (p = 0.001), B. uniformis (p < 0.001), Bacteroides spp. (p < 0.001), Bifidobacterium spp. (p < 0.001), A. onderdonkii (p = 0.007), Anaerovibrio spp. (p = 0.009), B. adolescentis (p = 0.021), B. caccae (p = 0.03), B. fragilis (p = 0.026), Butyricimonas spp. (p = 0.034), Lachnobacterium spp. (p = 0.029), Lachnospira spp. (p = 0.006), Megasphaera spp. (p = 0.006), Odoribacter ssp. (p = 0.015), Paraprevotella spp. (p = 0.02), and Succinivibrio spp. (p = 0.024). The decreasing trends in B. uniformis, Bacteroides spp., and Bifidobacterium spp. were significant even after performing multiple testing p-value corrections, and the significance of Bifidobacterium spp. is robust to the presence of zero counts based on the sensitivity analysis. On the other hand, there was a significant increasing trend (p < 0.05) in the abundance of C. celatum (p = 0.018), Dehalobacterium spp. (p = 0.013), Methanobrevibacter spp. (p = 0.016), and RFN20 spp. (p = 0.021) from Group 1 to Group 4 (Fig. 4a).Fig. 4: Heatmaps depict the ANCOM-BC2 pattern analysis.Monotonic increasing and decreasing trends in microbial species abundances from participants’ stool samples (N = 241) were assessed in relation to a exposure groups (G1 is the reference) and b outcome of HIV-1 infection status. The X-axis delineates contrasts between exposure groups or outcomes, while the Y-axis lists species identified as significant via ANCOM-BC2 pattern analysis. Species that remained significant post-adjustment for multiple comparisons using the Benjamini–Yekutieli (BY) are highlighted in green. Fold-changes (natural log base) are superimposed within each cell. The color spectrum, from blue to red with a neutral white midpoint, visualizes the fold changes. Specifically, blue cells indicate reduced abundance relative to the reference group, and red cells signal increased abundance compared to the reference group. A white cell represents no effect, where the fold change equals 1. Species that are significantly associated with both sexual exposure groups, as well as HIV-1 infection status are highlighted in the red boxes.We implemented an ANCOM-BC2 two-group comparison to discern species exhibiting differential abundance (DA) in relation to HIV-1 infection status. Notably, among the species identified as significantly differentially abundant with regards to HIV-1 infection status (Fig. 4b), the following species also displayed monotonic trends with sexual exposure group: A. muciniphila (p = 0.043), B. caccae (p = 0.043), B. fragilis (p < 0.001), B. uniformis (p = 0.006), Bacteroides spp. (p < 0.001), Butyricimonas spp. (p = 0.001), Dehalobacterium spp. (p < 0.001), Methanobrevibacter spp. (p = 0.006), and Odoribacter spp. (p = 0.013). The effect sizes in B. caccae and Dehalobacterium spp. remained significant after performing multiple testing p-value corrections. Moreover, the significance of B. caccae proved to be robust to zero counts, as verified by our sensitivity analysis. Interestingly, both Dehalobacterium spp. and Methanobrevibacter spp. potentially increased in abundance in future infection relative to uninfected participants, while the remaining species decreased in abundance among HIV-1 infected relative to uninfected participants.Interactions between microbial abundances and plasma cytokine levels that were associated with sexual exposure groupsThe differences in levels of inflammatory markers CRP, sCD14, and sCD163, previously established as significantly associated with the exposure variable (number of receptive anal intercourse partners), were not found to be significantly associated with the DA species among the sexual exposure groups, with two exceptions noted within the genera Lachnospira and RFN20 (Supplementary Table 1).Mediation analysisThe primary aim of this study was to assess whether inflammatory cytokines and gut microbiota mediate the relationship between an increase in the number of partners for receptive anal intercourse and subsequent HIV-1 infection (refer to Supplementary Fig. 4a). We focused on specific biomarkers including cytokines sCD14 and sCD163, and DA species A. muciniphila, B. caccae, B. fragilis, B. uniformis, Bacteroides spp., Butyricimonas spp., Dehalobacterium spp., Methanobrevibacter spp., and Odoribacter spp. These biomarkers were selected since they were identified as having significant associations with both sexual exposure groups, as well as the outcome variable, HIV-1 infection status.Using the natural effect model consisting of sexual exposure groups as the exposure variable, biomarkers (sCD14 and sCD163) as the mediators, and HIV-1 infection status as the outcome variable, we discovered a significant natural direct effect (NDE) of the exposure on the outcome while controlling for biomarker levels. NDE quantifies the effect of the exposure on the outcome not through the mediator. Specifically, conditional on substance use, increasing the exposure from Group 1 to another group (while maintaining sCD14 and sCD163 at the same level) significantly increased the odds of HIV-1 infection. The odds ratios for Groups 2, 3, and 4 are exp(1.92) = 6.82, exp(2.56) = 12.94, and exp(3.55) = 34.81, respectively (Table 2). This increasing trend is highly significant with p-value < 0.001 (Table 2)25. In addition to NDE, we also investigated the natural indirect effect (NIE) of the exposure on the outcome that acts through the mediator. Unlike NDE, NIE represents the effect of the intervention on the outcome solely due to its effect on the mediator. The natural effect models indicated a significant NIE of sCD14 and sCD163 levels on HIV-1 infection while controlling for the exposure among participants who had the largest number of receptive anal intercourse partners (p = 0.02, Table 2). A trend analysis further strengthened this result with a significant increase (p = 0.007, Table 2) in NIE corresponding with biomarker levels for participants that were exposed to a higher number of partners for receptive anal intercourse. Specifically, conditional on substance use, shifting levels of sCD14 and sCD163 from those observed in Group 1 to those seen in Group 4, while holding the exposure constant at any given group, increases the odds of HIV-1 infection with an odds ratio of exp(0.33) = 1.39.Table 2 Results of the natural effect models consisting of sexual exposure groups as the exposure variable, biomarkers (cytokines, microbial species, or their combined effects) as the mediators, and HIV-1 infection status as the outcome variableA similar natural effect model consisting of sexual exposure groups as the exposure variable, microbiota A. muciniphila, B. caccae, B. fragilis, B. uniformis, Bacteroides spp., Butyricimonas spp., Dehalobacterium spp., Methanobrevibacter spp., and Odoribacter spp. as mediators, and HIV-1 infection status as the outcome variable, we discovered a significant NDE of the exposure on the outcome while controlling the microbial abundances. Conditional on substance use, changing the exposure from Group 1 to any other group, while maintaining microbial abundance, increased the odds of HIV-1 infection (odds ratios for Groups 2, 3, and 4 are exp(2.08) = 8.00, exp(2.61) = 13.60, and exp(3.58) = 35.87, respectively; Table 2). A trend analysis further strengthened this result with a significant increasing trend (p < 0.001, Table 2). Moreover, this model highlighted a significant NIE of microbial abundances in HIV-1 infection, particularly among participants with the largest number of receptive anal intercourse partners (p = 0.04, Table 2). Trend analysis further validated a notable increase in NIE associated with microbial abundance for participants that were exposed to a greater number of receptive anal intercourse partners, with a p-value of 0.033 (Table 2). Specifically, transitioning microbial species abundances from those of Group 1 to those in Group 4, while holding exposure constant, increased HIV-1 infection odds by exp(0.35) = 1.42.When the cytokines and microbial species were taken together as mediators into the natural effect models, which is biologically reasonable to consider, the results of NDE for this combined model were similar to NDE results when separate natural models were considered for cytokines and microbial species (Table 2). However, there was a substantial increase in the NIE estimates from the combined model consisting of cytokines and microbiomes as mediators (Table 2) when compared to the two sets of mediators, cytokines, and microbiomes, modeled separately. For example, the NIE estimates of cytokines and microbiomes in Group 4 relative to Group 1, was about 1.5 times the NIE estimates of either cytokines or microbiomes in Group 4 relative to Group 1, when modeled separately. Thus, the microbiome and the cytokines synergistically mediated the effects of sexual exposure on HIV-1 infection. Interestingly, sexual exposure groups had significant NDE on the outcome through many microbial species and cytokines when taken individually (i.e., separate models for each of these variables) (Supplementary Table 2). However, individually, almost none of these variables demonstrated significant NIE mediating the effect of sexual activity on HIV-1 infection (Supplementary Table 2). We also investigated if substance use could be a potential mediator of sexual exposure, as defined in this paper, on HIV-1 infection and found it not significant (data not presented). Taken together, the above results suggest that collectively rather than individually the biomarkers sCD14 and sCD163, and the microbial species A. muciniphila, B. caccae, B. fragilis, B. uniformis, Bacteroides spp., Butyricimonas spp., Dehalobacterium spp., Methanobrevibacter spp., and Odoribacter spp. mediate the effects of sexual behavior on HIV-1 infection.
