Starting to have sexual intercourse is associated with increases in cervicovaginal immune mediators in young women: a prospective study and meta-analysis

  1. Sean M Hughes
  2. Claire N Levy
  3. Fernanda L Calienes
  4. Katie A Martinez
  5. Stacy Selke
  6. Kenneth Tapia
  7. Bhavna H Chohan
  8. Lynda Oluoch
  9. Catherine Kiptinness
  10. Anna Wald
  11. Mimi Ghosh
  12. Liselotte Hardy
  13. Kenneth Ngure
  14. Nelly R Mugo
  15. Florian Hladik  Is a corresponding author
  16. Alison C Roxby  Is a corresponding author
  1. Department of Obstetrics and Gynecology, University of Washington, United States
  2. Vaccine and Infectious Disease Division, Fred Hutchinson Cancer Research Center, United States
  3. Department of Laboratory Medicine & Pathology, University of Washington, United States
  4. Department of Global Health, University of Washington, United States
  5. Centre for Virus Research, Kenya Medical Research Institute, Kenya
  6. Centre for Clinical Research, Kenya Medical Research Institute, Kenya
  7. Department of Medicine, University of Washington, United States
  8. Department of Epidemiology, University of Washington, United States
  9. Department of Epidemiology, Milken Institute School of Public Health, The George Washington University, United States
  10. Department of Clinical Sciences, Unit of Tropical Bacteriology, Institute of Tropical Medicine, Belgium
  11. Department of Community Health, Jomo Kenyatta University of Agriculture and Technology, Kenya

Abstract

Background:

Adolescent girls and young women (AGYW) are at high risk of sexually transmitted infections (STIs). It is unknown whether beginning to have sexual intercourse results in changes to immune mediators in the cervicovaginal tract that contribute to this risk.

Methods:

We collected cervicovaginal lavages from Kenyan AGYW in the months before and after first penile-vaginal sexual intercourse and measured the concentrations of 20 immune mediators. We compared concentrations pre- and post-first sex using mixed effect models. We additionally performed a systematic review to identify similar studies and combined them with our results by meta-analysis of individual participant data.

Results:

We included 180 samples from 95 AGYW, with 44% providing only pre-first sex samples, 35% matched pre and post, and 21% only post. We consistently detected 19/20 immune mediators, all of which increased post-first sex (p<0.05 for 13/19; Holm-Bonferroni-adjusted p<0.05 for IL-1β, IL-2, and CXCL8). Effects remained similar after excluding samples with STIs and high Nugent scores. Concentrations increased cumulatively over time after date of first sex, with an estimated doubling time of about 5 months.

Our systematic review identified two eligible studies, one of 93 Belgian participants, and the other of 18 American participants. Nine immune mediators were measured in at least two-thirds of studies. Meta-analysis confirmed higher levels post-first sex for 8/9 immune mediators (p<0.05 for six mediators, most prominently IL-1α, IL-1β, and CXCL8).

Conclusions:

Cervicovaginal immune mediator concentrations were higher in women who reported that they started sexual activity. Results were consistent across three studies conducted on three different continents.

Funding:

This research was funded by R01 HD091996-01 (ACR), by P01 AI 030731-25 (Project 1) (AW), R01 AI116292 (FH), R03 AI154366 (FH) and by the Center for AIDS Research (CFAR) of the University of Washington/Fred Hutchinson Cancer Research Center AI027757.

Editor's evaluation

This study finds that the levels of many immune markers are higher in vaginal samples in women taken after initiation of vaginal sex than before initiation of vaginal sex. This result may indicate that initiation of vaginal sex potentially influences vaginal immune responses in adolescents and young adults. This study will be of the highest interest to those interested in how immune markers can change within individuals over time.

https://doi.org/10.7554/eLife.78565.sa0

Introduction

Adolescent girls and young women (AGYW) aged 15–24 are at high risk for sexually transmitted infections (STIs) and are disproportionately affected by human immunodeficiency virus (HIV), accounting for as many as 80% of new HIV infections in some countries (UNAIDS, 2019). In Kenya, in particular, a quarter of new HIV infections occur in AGYW (National AIDS Control Council, 2018). Because of the disproportionate impact of HIV on AGYW, understanding the behavioral and physiological components of heightened STI risk in this population presents an opportunity to reduce HIV infections.

The period following first penile-vaginal sexual intercourse (“first sex”) marks the start of vulnerability to STIs and is associated with higher acquisition of bacterial vaginosis (BV) and STIs than later in life. The reasons for this increased susceptibility remain unclear, because (1) studies of mucosal immunity in AGYW are challenging, and (2) it can be difficult to distinguish immune changes that are a consequence of STI acquisition from changes that occur independently of STIs. Understanding changes in cervicovaginal tract (CVT) immune mediators following first sex may help identify interventions to decrease the risk of STI and HIV acquisition in AGYW.

In this study, we measured immune mediators in cervicovaginal lavage (CVL) specimens collected from a unique longitudinal cohort of Kenyan AGYW (Casmir et al., 2021; Yuh et al., 2020), an especially vulnerable population. By comparing specimens collected in the months before and after first sex, our goal was to measure changes in CVT immune mediators following start of sexual intercourse. Extensive clinical information was available about participants, including acquisition of STIs.

To generalize our findings to broader populations, we also conducted a systematic review of published literature to identify other studies of cervicovaginal immune mediators in AGYW before and after first sexual intercourse. We sought individual participant data from study authors and performed meta-analyses of individual immune mediators, assessing changes in immune mediator concentration before and after first sex.

Materials and methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Biological sample (Homo sapiens)Cervicovaginal lavage samplesThis paperThe dataset is included in Source code 1
Commercial assay or kitQuantifiler Duo DNA Quantification KitThermo FisherCatalog# 4387746
Commercial assay or kitPierce BCA Protein Assay KitThermo FisherCatalog# 23225
Commercial assay or kitMIG R-Plex assayMeso Scale DiscoveryCatalog# F210I
Commercial assay or kitRANTES R-Plex assayMeso Scale DiscoveryCatalog# F21ZN
Commercial assay or kitIL-1RA U-Plex assayMeso Scale DiscoveryCatalog# K151XPK
Commercial assay or kitIFN-γ U-Plex assayMeso Scale DiscoveryCatalog# K151TTK
Commercial assay or kitIL-10 U-Plex assayMeso Scale DiscoveryCatalog# K151TZK
Commercial assay or kitIL-12p70 U-Plex assayMeso Scale DiscoveryCatalog# K151UAK
Commercial assay or kitIL-1β U-Plex assayMeso Scale DiscoveryCatalog# K151TUK
Commercial assay or kitIL-2 U-Plex assayMeso Scale DiscoveryCatalog# K151TVK
Commercial assay or kitIL-6 U-Plex assayMeso Scale DiscoveryCatalog# K151TXK
Commercial assay or kitIL-8 U-Plex assayMeso Scale DiscoveryCatalog# K151TYK
Commercial assay or kitTNF-α U-Plex assayMeso Scale DiscoveryCatalog# K151UCK
Commercial assay or kitIFN-a2a U-Plex assayMeso Scale DiscoveryCatalog# K151VHK
Commercial assay or kitIL-17a U-Plex assayMeso Scale DiscoveryCatalog# K151UTK
Commercial assay or kitIL-18 U-Plex assayMeso Scale DiscoveryCatalog# K151VJK
Commercial assay or kitIL-1α U-Plex assayMeso Scale DiscoveryCatalog# K151UNK
Commercial assay or kitIL-7 U-Plex assayMeso Scale DiscoveryCatalog# K151UPK
Commercial assay or kitIP-10 U-Plex assayMeso Scale DiscoveryCatalog# K151UFK
Commercial assay or kitMIP-1a U-Plex assayMeso Scale DiscoveryCatalog# K151UJK
Commercial assay or kitMIP-1β U-Plex assayMeso Scale DiscoveryCatalog# K151UKK
Commercial assay or kitMIP-3α U-Plex assayMeso Scale DiscoveryCatalog# K151UZK
Commercial assay or kitHuman Kallikrein 3/PSA DuoSet ELISAR&D SystemsCatalog# DY1344
Software, algorithmR Project for Statistical ComputerR Core TeamRRID:SCR_001905
Software, algorithmR package: lmerTestComprehensive R Archive NetworkRRID:SCR_015656
Software, algorithmR package: metaComprehensive R Archive NetworkRRID:SCR_019055
Software, algorithmR package: tidyverseComprehensive R Archive NetworkRRID:SCR_019186
OtherGhosh 2018 datasetGhosh et al., 2018The dataset is included in Source code 1
OtherJespers 2016 datasetJespers et al., 2016The dataset is included in Source code 1

Clinical cohort and study procedures

Request a detailed protocol

This study used specimens collected from the Kenya Girls Study, a previously described longitudinal cohort study of AGYW (Casmir et al., 2021; Yuh et al., 2020). Briefly, AGYW aged 16–20 were recruited in Thika, Kenya. Enrolled participants returned quarterly, where they were interviewed about their sexual behavior and provided vaginal swabs, CVLs, and blood. For the sub-study described in this paper, we selected a group of 195 samples from the larger Kenya girls study, including samples from participants who provided only pre-first sex samples, participants who provided only post-first sex samples, and participants who provided both pre- and post-first sex samples. The sample size was chosen based on selection of all available samples from all participants who reported sexual intercourse and a matching number of participants who did not.

Human subjects approval was obtained from the Kenya Medical Research Institute Scientific Ethics Review Unit (protocol 2760) and the University of Washington Institutional Review Board (number 00000946). Participants under age 18 provided written informed assent, and written informed consent was obtained from a parent/guardian. Participants assented privately from parents/guardians, after asking questions and deciding whether they wanted to participate free of parental influence. Participants age 18 or older provided written informed consent.

CVLs were collected using flexible tubing in the vagina to avoid a speculum examination. 5 cc of sterile saline was instilled into the vagina via tubing by the study clinician, left for 15 s, then aspirated into the syringe. Lavage fluid was spun for 10 min at 800×g; supernatant was removed, re-spun for 10 min, and stored in 2 mL aliquots at –80°C.

Participants were tested for STIs and BV. Vaginal swabs were tested for NG, CT, and TV using the Gen-Probe Aptima test (Hologic, Marlborough, MA) and for HSV-1 and HSV-2 by in-house PCR. Nugent scoring was performed on smears from vaginal swabs for BV; scores ≥7 were considered BV, scores 4–6 were considered intermediate, and scores ≤3 were considered no dysbiosis (Nugent et al., 1991). Blood was tested for HIV using the Vironostika HIV Uni-Form II Ag-Ab (Biomerieux, Marcy-l'Etoile, France) and for HSV-1 and HSV-2 antibodies using in-house Western blot.

STI tests were performed annually in the parent study. Therefore, STI testing had not been performed at some timepoints selected for this sub-study. For samples with no concurrent STI result, we determined STI status as follows: If the annual tests both before and after the untested visit were negative, we inferred a negative result for that visit. If either annual test was positive, STI testing was performed for that timepoint. Negative CT and NG tests were inferred for 28 samples as well as negative TV tests for 45 samples.

Participants were tested for pregnancy via rapid urinary pregnancy test if they reported missed menses. Serum samples were tested for progesterone by automated immunoassay (Cobas E411, Roche, Basel, Switzerland). Participants were defined to be in the follicular phase if serum progesterone was <3 ng/mL, in the luteal phase if serum progesterone was ≥3 ng/mL, and ‘other’ if more than 35 days since the start of their last menstrual period had passed (indicating irregular menstruation, use of hormonal contraception, or pregnancy). Participants were assumed not to be using contraception until they reported sexual intercourse.

PSA was measured in CVL using the Human Kallikrein 3/PSA DuoSet ELISA (R&D Systems, Minneapolis, MN). Samples were considered positive if the PSA concentration was ≥10 ng/mL. Total protein concentrations were measured using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, Waltham, MA). PSA and total protein concentrations were calculated using four-parameter logistic curves fit to the standard curves. Vaginal swabs were tested for Y-chromosome DNA by Quantifiler Duo DNA Quantification Kit (Life Technologies).

The immune mediator data reported in this article was previously included in a meta-analysis of changes in immune mediator concentrations during the menstrual cycle (Hughes et al., 2022), where it is referred to as “Hughes-unpublished”.

Immune mediator quantification

Request a detailed protocol

We selected 20 immune mediators to measure in CVLs. These immune mediators were selected based on prior human and/or non-human primate studies as being consistently up- or down-regulated prior to HIV or STI transmission events, and therefore playing a key role in genital tract immunity to STIs. Concentrations of immune mediators were measured in CVL using Meso Scale Discovery (MSD) R-Plex/U-Plex kits according to the manufacturer’s instructions and read on the MESO QuickPlex SQ 120. Pre- and post-first sex samples were present on every plate, and the scientists were blinded to sample identity. CVLs were diluted 100-fold for measurement of IL-1RA and 10-fold for CXCL9 and CCL5. Concentrations were determined using four parameter logistic fits in MSD Discovery Workbench software.

Definition of first sexual intercourse

Request a detailed protocol

Samples were categorized as pre- or post-first sex based on participant report of ever engaging in penile-vaginal penetrative sex. In addition, visits were categorized as post-first sex if at that visit or a previous visit, the participant was pregnant, Y-chromosome DNA was detected in a vaginal swab sample, PSA was detected in a CVL, or a vaginal swab tested positive for NG, CT, or TV.

Statistical analysis

Request a detailed protocol

Data analysis was conducted using R (R. Core Team, 2018) version 4.0.0 with the packages plater (M Hughes, 2016) and tidyverse (Wickham et al., 2019) in addition to the statistical packages described below.

For MSD, concentrations of each analyte were averaged across replicate wells and log2-transformed. Replicates with concentrations below the reported lower limits of detection were assigned the value of half the lower limit of detection. Samples were defined as detectable if the concentration was above the lower limit of detection for at least one of the two replicate wells. Some samples were excluded due to rare STIs or missing data (fully described in Figure 1), these exclusion criteria were not pre-established.

We performed primary and secondary analyses. For the primary analysis, we used univariate models to estimate the difference between cytokine concentrations in pre- and post-first sex samples (assessed by effect size and p-value). Pre- and post-first sex were modeled as a binary fixed effect and participant as a random effect to account for multiple samples per participant using mixed-effect models with the R packages lme4 (Bates et al., 2015) and lmerTest (Kuznetsova et al., 2017). Log2-transformed immune mediator concentrations were used as the outcome for immune mediators if more than 50% of samples were above the limit of detection. For the remaining immune mediators, logistic models were used with the outcome being above/below the lower limit of detection. Logistic models were used for immune mediators with low levels of detection because the distributions become increasingly non-normal as more samples fall below the lower limit of detection. We adjusted for multiple comparisons by the Holm-Bonferroni method (Holm, 1979).

We performed several secondary analyses. We tested whether pregnancy, STIs, BV, and contraception explained our results by repeating our primary, univariate analysis on the subset of samples negative for all of these variables. In addition, we repeated our univariate models using immune mediator concentrations normalized to total protein. Furthermore, we performed univariate models on the subset of samples from participants who provided both pre- and post-first sex samples. Finally, we fit multivariate models with additional fixed effects: age, education at enrollment, rural/urban residence at enrollment, and roof material at enrollment (an indicator of poverty). In all cases, we assessed whether these factors had a major influence on results by comparing the effect sizes from the primary analysis to the effect sizes from the secondary analysis.

We additionally wished to assess whether immune mediator concentrations changed over time with respect to the date of first sex. To assess this question, we fit mixed-effect models with linear splines, using days since first sex as a fixed effect and participant as a random effect, with the outcome being the concentration of immune mediator (log2 pg/mL). For the splines, we used a knot at day 0, which was the day of first sex.

All code and raw data are available in Source code 1 and on GitHub at https://github.com/seaaan/ThikaPrePostSexManuscript/ (copy archived at swh:1:rev:ef3bd4bee186be79944ebd821fd5e1bb2e89a26b; Hughes, 2022). All results are available in Supplementary file 1.

Systematic review and meta-analysis

We designed our systematic review and meta-analysis in compliance with the preferred reporting items for a systematic review and meta-analysis of individual participant data (PRISMA-IPD) guidelines (Stewart et al., 2015). The PRISMA-IPD checklist is included as Reporting standard 1.

Study eligibility criteria

Request a detailed protocol

Studies were eligible if they reported original immunoassay data on immune mediator concentrations in CVT fluid samples from AGYW before and after first vaginal sexual intercourse. Immunoassay data included any antibody-based methods (such as ELISA, bead-based assays, and electrochemiluminescence assays). CVT fluid samples could be collected by CVL, menstrual cup, or swab. Each study was required to include samples from before and after first penile-vaginal sexual intercourse, but we did not require that there be pre- and post-first sex samples from each participant. Samples were required to be categorized as pre- or post-first sex based on participant report of ever engaging in penile-vaginal penetrative sex. In addition, visits were categorized as post-first sex based on pregnancy, detection of sperm antigen or DNA, or a positive STI test.

Study identification and inclusion

Request a detailed protocol

We searched PubMed on August 28, 2021 for articles published in English using the search strategy in Supplementary file 2. Abstracts from search results were reviewed independently by two reviewers (SMH and CNL) using abstrackr (Wallace et al., 2012). If either reviewer judged the study to be potentially eligible based on the abstract, then full-text articles were obtained and reviewed by both reviewers. Study inclusion was determined by consensus between the two reviewers.

Data collection and verification

Request a detailed protocol

We sought individual participant data via email with the authors of eligible studies. For each sample, we collected the participant identifier, the pre/post-first sex status of the sample, and the immune mediator concentrations in pg/mL. For each study, we collected the type of immunoassay used, the type of sample collected, and the country of sample origin. Individual participant data were verified by replicating analyses published in the prior manuscripts. Pre/post-first sex status was standardized across studies: in particular, one study included samples from participants reporting genital touching without penile-vaginal sexual intercourse. We categorized these samples as pre-first sex in accordance with our definition above. No personally identifiable information was obtained from the studies included in the meta-analysis. Participants in both studies provided informed consent and consent to publish as described in the original manuscripts (Jespers et al., 2016; Ghosh et al., 2018).

Data analysis

Request a detailed protocol

The outcome of interest was the difference in immune mediator concentrations between pre- and post-first sex samples. We performed meta-analyses on all immune mediators present in at least two included studies. We used a two-stage approach for meta-analysis. First, we reanalyzed the individual participant data from the eligible studies. Studies with multiple samples per participant were analyzed using mixed-effect models as above; studies with single samples per participant were analyzed using simple linear models. For all studies, immune mediator concentrations (pg/mL) were log2-transformed. In the second stage, we performed random-effects meta-analysis using the inverse-variance method in the R package meta (Balduzzi et al., 2019). We quantified heterogeneity using I2. We did not attempt to explore variation in effects by study-level characteristics due to the low number of studies available.

Risk of bias

Request a detailed protocol

We assessed risk of bias within each study using a modified Newcastle-Ottawa Quality Assessment scale (Supplementary file 3), ranging from 0 (high risk of bias) to 7 (low risk of bias).

Results

Characteristics of cohort

We selected 195 CVL samples for this study. We excluded 15 samples due to unavailable covariate information (such as total protein or Prostate-specific antigen [PSA]), extensive blood contamination, and presence of rare infections (n=1–2 samples per infection: Trichomonas vaginalis (TV), Neisseria gonorrhoeae (NG), genital HSV-1 or HSV-2 DNA; Figure 1A). The final sample set included 180 samples from 95 participants.

Cervicovaginal lavage sample selection.

(A) Cervicovaginal lavage sample selection, collected from adolescent girls and young women in a longitudinal cohort study. (B) Timepoints for collection of cervicovaginal lavage specimens pre- and post-first sexual intercourse.

Of these, 111 samples from 75 participants were classified as pre-first sex, and 57 samples from 45 participants were obtained after reported first sex. In addition, 12 samples from 8 participants were classified as post-sex due to presence of Y-chromosome DNA, PSA, or because of pregnancy or a positive Chlamydia trachomatis (CT) test at that visit or a prior visit. Thus, there were a total of 69 post-sex specimens from 53 participants. Pre-first sex samples were collected a median of 344 days before first sex (interquartile range [IQR] 72–687), while post-first sex samples were collected a median of 66 days (27-105) after. We captured matched pre- and post-first sex specimens from 33 participants (66 specimens); the remaining participants provided either only pre-first sex specimens (42 participants; 78 specimens) or only post-first sex specimens (20 participants; 36 specimens; Figure 1B).

Table 1 summarizes demographic and clinical variables of the samples from the final sample set. The median age of participants in the final sample set was 19.1 years (IQR 18.1–19.6). All participants were Black Africans assigned female at birth. Post-first sex samples came from participants who were, on average, older and more likely to have BV (Table 1). By definition, only the post-first sex group included samples from pregnant participants, participants using contraception, and samples where Y-chromosome, PSA, CT, or HSV-2 was detected.

Table 1
Comparison of social, demographic, and biological characteristics of adolescent girls and young women at timepoints of cervicovaginal lavage specimens selected for the study.

Each participant can appear in this table more than once. Pregnant indicates pregnant at the time of sample collection. All participants listed here were negative for HSV-1 and HSV-2 genital DNA (indicative of active infection). Mixed effect regression models included a random intercept for participant.

Pre-first sex timepointsN=111 specimens*Post-first sex timepointsN=69 specimens*p-Value
Age (years)18.80 (17.80, 19.45)19.30 (18.60, 19.80)<0.001
Education at enrollment0.934§
 Completed high school18 (16%)12 (17%)
 In high school93 (84%)57 (83%)
Income at enrollment0.876§
 No regular income57 (51%)43 (62%)
 Regular income54 (49%)26 (38%)
Residence at enrollment0.575§
 Rural72 (65%)28 (41%)
 Urban39 (35%)41 (59%)
Roof material at enrollment0.814§
 Not tile or metal10 (9%)1 (1.4%)
 Tile or metal101 (91%)68 (99%)
Menstrual phase
 Luteal39 (35%)26 (38%)0.73§
 Follicular64 (58%)35 (51%)0.36§
 Other8 (7.2%)8 (12%)0.21§
Pregnant-
 Negative111 (100%)64 (93%)
 Positive0 (0%)5 (7.2%)
Contraception-
 None111 (100%)54 (78%)
 Emergency pills0 (0%)4 (5.8%)
 Condoms0 (0%)11 (16%)
Y-chromosome-
 Negative111 (100%)55 (80%)
 Positive0 (0%)14 (20%)
Prostate-specific antigen-
 Negative111 (100%)67 (97%)
 Positive0 (0%)2 (2.9%)
Bacterial vaginosis
 Negative (Nugent score 0–3)102 (92%)54 (78%)0.08§
 Intermediate (4–6)6 (5.4%)6 (8.7%)0.82§
 Positive (7–10)3 (2.7%)9 (13%)0.02§
Chlamydia-
 Negative111 (100%)59 (86%)
 Positive0 (0%)10 (14%)
HSV-1>0.9§
 Seronegative4 (3.6%)2 (2.9%)
 Seropositive107 (96%)67 (97%)
HSV-2-
 Seronegative111 (100%)66 (96%)
 Seropositive0 (0%)3 (4.3%)
  1. *

    Statistics presented: median (IQR) or n (%).

  2. Linear mixed effects regression.

  3. Time-invariant variables, with data only available from the enrollment visit.

  4. §

    Logistic mixed effects regression with binary combinations, comparing each category to all other samples (e.g. luteal compared to pooled follicular and other).

  5. Not tested for difference between groups because this variable can only be positive at post-first sex time points by definition.

Immune mediator measurements pre- and post-first sex

Concentrations of 20 immune mediators were measured (Figure 2). IL-1RA was detected at the highest concentrations. Sixteen immune mediators were detectable in most specimens. Four were detected in fewer than half of the samples: IFN-γ (27%), IL-10 (23%), IL-12p70 (20%), and IFN-α2A (7%); due to lack of detection, IFN-α2A was excluded from further analysis.

Immune factor concentrations in pg/mL from cervicovaginal lavage specimens from adolescent girls and young women, comparing specimens from before and after first sexual intercourse.

Blue indicates specimens prior to first sex, and green indicates samples after first sex. Specimens below the limit of detection are indicated as lighter colors and were set to half the limit of detection. Percentages indicate the percent of samples within the detectable range of the assays. The boxes range from the 25th to 75th percentiles of the data, with the middle vertical line indicating the median. The whiskers stretch to the values no farther from the edge of the box than 1.5 times the interquartile range. The sample size is 180 specimens from 95 participants.

As shown in Figure 2, immune mediator concentrations were higher after first sex. We quantified the differences between pre- and post-first sex samples (Figure 2). In all cases, the concentration (Figure 3A, Table 2) or proportion detectable (Figure 3B, Table 3) was higher. The differences were significant at p<0.05 for most (13/19) immune mediators and remained significant for IL-1β, IL-2, and CXCL8 after adjustment for multiple comparisons (adjustment for 19 immune mediators).

Table 2
Comparison of immune mediators in cervicovaginal lavage samples pre- and post-first sexual intercourse.

Results of univariate mixed-effect models with first sex as fixed effect and participant as random effects. Mean log2 pg/mL difference values above 0 indicate higher concentrations post-first sex. Ratio indicates the multiplicative ratio of post over pre-sex (on a linear scale), so a ratio of 1 indicates no difference, ratios >1 indicate higher post, and ratios <1 indicate higher pre. The ratio is calculated as 2^mean log2 difference. Adjusted p-values are adjusted by Holm-Bonferroni for 19 immune mediators.

Immune mediatorMean log2 differenceStandard errorRatio (95% CI)p-ValueAdjusted p-value
IL-20.900.251.86 (1.32–2.62)4.7E-40.009
IL-1β1.330.432.51 (1.4–4.51)0.0020.044
CXCL81.010.332.02 (1.29–3.17)0.0030.045
IL-1α0.670.241.59 (1.15–2.2)0.0060.088
CCL30.620.221.53 (1.13–2.07)0.0060.091
IL-17A0.640.251.56 (1.11–2.18)0.0110.154
TNF-α0.690.291.61 (1.08–2.39)0.0200.236
CXCL100.740.321.67 (1.08–2.57)0.0210.236
CXCL90.600.261.51 (1.07–2.14)0.0210.236
IL-1RA0.480.221.39 (1.03–1.88)0.0320.258
IL-60.810.391.75 (1.03–2.98)0.0410.284
CCL40.420.271.33 (0.93–1.92)0.1240.621
CCL200.530.361.44 (0.88–2.34)0.1460.621
CCL50.390.281.31 (0.9–1.92)0.1660.621
IL-180.440.361.35 (0.83–2.22)0.2320.621
IL-70.020.181.01 (0.79–1.3)0.9210.921
Table 3
Comparison of immune mediators in cervicovaginal lavage samples pre- and post-first sexual intercourse.

Results of univariate mixed-effect models with first sex as fixed effect and participant as random effects. Log odds ratio values above 0 indicate that the immune mediator was more often detected post-first sex. Odds ratio indicates the odds ratio of detection for post over pre-sex (on a linear scale), so a ratio of 1 indicates no difference, ratios >1 indicate higher detection post, and ratios <1 indicate higher detection pre. Adjusted p-values are adjusted by Holm-Bonferroni for 19 immune mediators.

Immune mediatorLog-oddsStandard errorOdds ratio (95% CI)p-ValueAdjusted p-value
IL-101.150.463.15 (1.29–7.72)0.0120.154
IFN-γ1.040.462.82 (1.14–6.98)0.0250.236
IL-12p702.021.037.52 (1.00–56.59)0.0500.300
Comparison of immune mediators in cervicovaginal lavage samples pre- and post-first sexual intercourse.

Univariate mixed-effect models with first sex as fixed effect and participant as random effects. Symbols indicate the mean and horizontal lines indicate the 95% confidence intervals. Filled symbols indicate p<0.05, while open symbols indicate p≥0.05. Vertical lines at 1 indicate no difference between pre- and post-first sex. (A) Multiplicative ratio of post-first sex concentrations (pg/mL) to pre-first sex concentrations on a log scale. A value of 1 indicates no difference, and numbers higher than 1 indicate greater quantities of immune mediators post-first sex. Ratios and 95% confidence intervals are shown at right. The ratio is calculated as 2^mean log2 difference shown in Table 2. ( B) Odds ratio (on a log scale) comparing the odds of the immune mediator being detected above the lower limit of detection (for immune mediators detected in fewer than half of the samples). A value of 1 indicates no difference, and numbers higher than 1 indicate greater detection of immune mediators post-first sex. Ratios and 95% confidence intervals are shown at right. The sample size is 180 specimens from 95 participants.

Immune mediator concentrations over time

We next assessed whether immune mediator concentrations increased all at once at first sex or cumulatively over time. We therefore evaluated immune mediator concentrations relative to date of first sex. Dates were available for 80 pre-first sex samples from 59 participants and 60 post-first sex samples from 49 participants.

As shown in Figure 4 and Table 4, immune mediator concentrations were generally stable for 3 years prior to first sex and then increased sharply in the year following first sex. The change in slope following first sex was statistically significant at p<0.05 for 17/19 immune mediators. This pattern is consistent with cumulative increases in immune mediator concentrations following first sex.

Concentration of immune mediators in cervicovaginal lavage samples relative to day of first sex.

Each symbol depicts the concentration in a single sample. Black lines show the slopes from mixed-effect models with days since first sex as fixed effect and participant as random effects with linear splines using a knot at day 0. Concentrations are shown in pg/mL on a log10 scale. The sample size is 80 pre-first sex samples from 59 participants and 60 post-first sex samples from 49 participants.

Table 4
Change in concentration of immune mediators with time.

Results of mixed-effect models with 100 days since first sex as a fixed effect and participant as random effects, using a spline with a knot at 0 (the day of first sexual intercourse). The slope column indicates the change in log2 immune mediator concentration (pg/mL) per 100 days. A value of 1 would indicate an increase of 1 log2 pg/mL in 100 days, while a value of –1 would indicate a decrease of the same. The p-values test whether the post-first sex slope differs from the pre-first sex slope.

Slope pre-first sexSlope post-first sexp-Value for difference between slopes
Slope95% CISlope95% CI
IL-20.0003(–0.1, 0.1)0.99(0.59, 1.39)0.00003
IL-12p70–0.02(–0.07, 0.03)0.49(0.28, 0.69)0.00003
IL-1β–0.04(–0.21, 0.13)1.60(0.92, 2.29)0.0001
IL-100.01(–0.05, 0.07)0.57(0.32, 0.81)0.0002
CXCL8–0.03(–0.16, 0.09)1.09(0.57, 1.61)0.0002
TNF-α–0.01(–0.12, 0.1)0.98(0.52, 1.44)0.0002
IL-17A0.04(–0.06, 0.14)0.93(0.51, 1.34)0.0003
CCL4–0.03(–0.13, 0.07)0.75(0.33, 1.17)0.001
IFN-γ0.03(–0.05, 0.11)0.62(0.28, 0.97)0.003
IL-1α0.02(–0.09, 0.13)0.71(0.29, 1.13)0.005
IL-6–0.08(–0.24, 0.08)0.88(0.24, 1.53)0.01
CCL30.05(–0.03, 0.13)0.53(0.2, 0.87)0.01
CCL20–0.10(–0.25, 0.06)0.69(0.07, 1.3)0.03
CXCL90.01(–0.1, 0.12)0.54(0.1, 0.97)0.03
CXCL100.03(–0.11, 0.18)0.70(0.14, 1.26)0.04
IL-1RA0.07(–0.03, 0.18)0.54(0.14, 0.94)0.04
IL-7–0.02(–0.09, 0.05)0.31(0.02, 0.59)0.04
CCL50.09(–0.03, 0.22)0.56(0.08, 1.04)0.09
IL-180.08(–0.06, 0.22)0.46(–0.12, 1.04)0.25

Assessment of other contributing factors

We next sought to determine whether the differences between pre- and post-first sex could be explained by covariates that differed between the groups or that are known to affect immune mediator concentrations (Table 1). In particular, pregnancy, CT infection, HSV-2 seropositivity, use of contraception, and elevated Nugent score are all known to affect cervicovaginal immune mediator concentrations. Because of this, we wished to know whether one or more of these factors was responsible for the differences we observed between pre- and post-first sex samples. To address this possibility, we repeated our analysis on a restricted sample set, excluding all samples where at the visit the participant was pregnant, positive for CT or HSV-2, reported using contraception, or had a Nugent score of 4 or greater. The analysis of these 137 samples (102 pre- and 35 post-first sex) is shown in Figure 5AB. The effects remained positive for all immune mediators in this subset analysis, indicating that the differences we saw between pre- and post-first sex samples were not solely a result of acquisition of STIs or pregnancy.

Alternative analysis strategies for the association of first sexual intercourse and quantities of cervicovaginal immune mediators.

Filled symbols indicate p-value <0.05, and open symbols indicate p-values ≥0.05. Vertical lines at 1 indicate no difference between pre- and post-first sex. Black symbols show the same primary analysis as Figure 3. Green symbols show the primary, univariate models applied to a subset of samples negative for pregnancy, contraception, bacterial vaginosis (BV), chlamydia, and HSV-2. Orange symbols show the primary univariate analysis performed on concentrations normalized to total protein concentrations (pg/mg protein). Purple symbols show the primary analysis repeated on only those samples from participants who provided both pre- and post-first sex samples. Pink symbols show a multivariate analysis adjusted for age, education, income, urban/rural residence, and roof type. (A) Multiplicative ratio of post-first sex concentrations to pre-first sex concentrations (pg/mL) on a log scale. A value of 1 indicates no difference, and numbers greater than 1 indicate higher values post-sex. (B) Odds ratio (on a log scale) for the immune mediator being detected above the lower limit of detection. A value of 1 indicates no difference, and numbers greater than 1 indicate higher odds of detection post-sex. Symbols indicate the mean, and horizontal lines indicate the 95% confidence intervals. The error bar for the fully paired analysis of IL-12 extends off-scale. (C) Total protein concentrations in cervicovaginal lavage specimens. The boxes range from the 25th to 75th percentiles of the data, with the middle horizontal line indicating the median. The whiskers stretch to the values no farther from the edge of the box than 1.5 times the interquartile range. The sample size is 180 specimens from 95 participants.

In addition, we sought to assess whether the differences between pre- and post-first sex would be explained by social determinants of health. There was limited diversity in our cohort in these variables, in particular in terms of race and ethnicity. We adjusted the models from our primary analysis for age, education, whether the participants reported a regular source of income, whether the participants lived in an urban or rural environment, and whether the participants lived in a home with a metal or tile roof (lacking such a roof being an indicator of poverty). As shown in Figure 5AB, the effect sizes from the multivariate models were similar to those from the univariate models, suggesting that these variables do not explain the results we observed.

To determine whether the pre- and post-first sex differences were driven by differences in protein recovery, we normalized the immune mediator concentrations to total protein concentrations. This analysis modestly reduced the effect size estimates between pre- and post-first sex (Figure 5AB). This result is consistent with the slightly higher total protein concentrations observed in post-first sex CVLs (Figure 5C, 0.25 log2 fold change mg/mL, p=0.12). In all cases except for IL-7, the immune mediator concentrations remained higher post-first sex after normalization to total protein.

Fully paired analysis

As described above, only a minority of participants provided both pre- and post-first sex samples; about two-thirds of participants provided only pre-first sex or only post-first sex samples. This is a consequence of the difficulty of assessing participants during dynamic life events and is consistent with prior studies (Ghosh et al., 2018; Jespers et al., 2016).

We repeated our univariate analysis on the 66 samples from the 33 participants who provided paired pre- and post-first sex samples. As shown in Figure 5AB, immune mediator concentrations were generally higher post-first sex in this subset of samples, but the effects were smaller and all but one (IL-12p70) failed to reach statistical significance.

The specimens from the fully paired subset were generally comparable to the specimens from the participants who only provided post-specimens, with one important difference: the post-specimens in the fully paired subset were obtained sooner after reported date of first sex (p=4.7E-8). In the fully paired subset, the post-first sex samples came only about a month after first sex (median 31.5 days). In contrast, the post-first sex samples from the other participants came about 4 months after first sex (median 157 days). The age of the participants at the time of first sex, however, was similar (18.9 years in each subset, p=0.90).

Systematic review and meta-analysis of immune mediator changes post-first sex

We performed a systematic review to identify comparable studies. Of 147 abstracts retrieved through our search, 7 were assessed as potentially eligible, and 2 studies were determined to be eligible after review of the full texts. We obtained individual participant data from both studies. The first was a cross-sectional study by ELISA of 11–19-year olds in Washington DC with 18 CVL samples from 10 participants post-first sex and 8 participants pre-first sex. This study found a statistically significant decrease in TNF-α, while most other mediators were higher (but not statistically significant) post-first sex (Ghosh et al., 2018). The other study longitudinally followed 14–19-year-old AGYW in Belgium, with 269 swab samples from 93 participants, 9 of whom provided both pre- and post-first sex samples, 43 of whom provided only pre-samples, and 41 of whom provided only post-samples. Using Luminex, this study found increases in IL-1α, IL-1β, and CXCL8 post-first sex (Jespers et al., 2016). No important issues were identified in checking the integrity of the individual participant data received for these two studies. Risk of bias was low (scores of 7/7 for Jespers et al., 2016 and the study reported here) to moderate (score of 4/7 for Ghosh et al., 2018). Individual participant data (IPD) was obtained from all eligible studies, so there is no additional risk of bias from missing data.

Including our study reported here, nine immune mediators were measured in at least two studies and were eligible for meta-analysis (Figure 6; Table 5). Meta-analyses identified six immune mediators as having higher concentrations post-first sex (IL-1α, IL-1β, IL-6, CXCL8, CCL4, and CCL5, all p<0.05; all p<0.05 after adjustment for multiple comparisons except for IL-6 and CCL4 [adjusted p=0.066]). The remaining three immune mediators all had meta-analysis p>0.05, with two higher post-first sex (CXCL10 and CCL20) and TNF-α showing the opposite. Supplementary file 4 contains detailed plots for each immune mediator included in the meta-analyses: the concentration data from every sample for each study and a forest plot showing each study’s effect size and weighting for each immune mediator.

Meta-analysis comparing three studies measuring cervicovaginal immune mediators in adolescent girls and young women before and after first sexual intercourse.

Multiplicative ratio of post-first sex concentrations to pre-first sex concentrations (pg/mL) on a log scale. A value of 1 indicates no difference, and numbers greater than 1 indicate higher values post-first sex. Symbols indicate the mean, and horizontal lines indicate the 95% confidence intervals. Filled symbols indicate p-value <0.05 while open symbols indicate p-values ≥0.05. Vertical lines at 1 indicate no difference between pre- and post-first sex. Colors indicate the source of the data (blue, primary study; gold, Ghosh 2018; pink, Jespers 2016; black, meta-analysis).

Table 5
Meta-analyses of immune mediator concentrations in cervicovaginal samples comparing pre- and post-first sexual intercourse.

Results of random effects meta-analyses using inverse variance pooling. Mean log2 difference values above 0 indicate higher concentrations post-first sex. Ratio indicates the multiplicative ratio of post over pre-sex (on a linear scale), so a ratio of 1 indicates no difference, ratios >1 indicate higher post, and ratios <1 indicate higher pre. Adjusted p-values are adjusted by Holm-Bonferroni for measuring nine immune mediators. I2 is a measure of heterogeneity between studies, with a proportion of 0 indicating that all between-study variation can be explained by random sampling and higher I2 proportions values indicating clinical, biological, or methodological diversity between studies.

Immune mediatorMean log2 differenceStandard errorRatio (95% CI)p-ValueAdjusted p-valueNumber of studiesI2
IL-1α0.970.221.96 (1.45–2.64)1.0E-59.1E-530.28
CXCL81.080.252.12 (1.51–2.98)1.5E-51.2E-430
CCL50.630.161.55 (1.25–1.91)5.2E-53.6E-420.05
IL-1β1.060.272.09 (1.44–3.03)9.3E-55.6E-430
IL-60.500.201.41 (1.07–1.86)0.0130.06630
CCL40.420.171.34 (1.06–1.7)0.0140.06620
CXCL100.450.321.36 (0.88–2.11)0.1640.49220.42
CCL200.150.401.11 (0.64–1.92)0.712120.54
TNF-α–0.170.920.89 (0.26–3.08)0.851120.89

We used statistical heterogeneity (I2) to assess the comparability of the three studies included in the meta-analysis, with an I2 of 0 indicating that all between-study variation can be explained by random sampling and higher I2 proportions indicating clinical, biological, or methodological diversity between studies (Higgins et al., 2021). In our meta-analysis, statistical heterogeneity was generally low (I2 of 0 for IL-1β, IL-6, CXCL8, and CCL4) to moderate (I2 of 0.05–0.55 for IL-1α, CCL5, CXCL10, and CCL20; Table 5). TNF-α had high statistical heterogeneity between studies (I2=0.885).

Discussion

In this study of a large cohort of Kenyan AGYW with specimens collected pre- and post-first sex, we observed an association of increased levels of CVT immune mediators with the start of sexual activity. The strongest evidence was for IL-1β, IL-2, and CXCL8, but all measured immune mediators followed a similar pattern. In addition, we observed strong linear associations between time since first sex and immune mediator concentration, compared to flat levels over time pre-first sex. These results were tested with several analytical approaches: limiting to only paired samples, limiting to samples without known causes of inflammation (BV, STIs, or pregnancy), adjusting for socioeconomic cofactors, and adjusting for multiple comparisons, but with all of these methods, the increased levels of immune mediators post-sex remained consistent.

We further performed a systematic review and meta-analysis of individual participant data, identifying two previous studies conducted in Belgium and the United States (Jespers et al., 2016; Ghosh et al., 2018). Meta-analysis combining all three studies found that the start of sexual activity was associated with increased concentrations for 8/9 immune mediators, with particularly strong evidence for IL-1α, IL-1β, IL-6, and CXCL8. The impressive agreement across these three studies strengthens the conclusion that the start of sexual activity is associated with increases in a wide range of CVT immune mediators.

While we observed a clear association between sexual activity and higher immune mediator concentrations, there is room for caution about whether the association is causal. For example, there could be differences in social determinants of health between the participants who do and do not engage in sexual activity. These differences were largely unmeasured in all three studies and may contribute to the changes we observed in immune mediator levels pre- and post-first sex. In our Kenyan study, we had information from the baseline visit about participant age, education, income, and housing. The effects of first sex on immune mediators remained the same after adjusting for these variables, suggesting that age, education, income, and housing at baseline do not explain the differences we observed. However, we did not assess these variables for changes over time, only at baseline.

The fully paired subset of our cohort (where participants provided pre- and post-samples) had smaller effect sizes than the full cohort. Most effects remained positive, but they were smaller in magnitude than in the primary analysis, and all were non-significant. A possible explanation for the smaller effects is that immune mediator concentrations increase over time post-first sex (Figure 4). The post-first sex samples in the fully paired subset were collected sooner after first sex (median 32 days) than those from the participants who provided only post-first sex samples (median 157 days). Thus, the effects may have been weaker in the fully paired subset, in part, because the samples were obtained such a short time after first sex. Another possible explanation for the smaller effects is unmeasured confounding in the full cohort. Because participants in this subset provided both pre- and post-first sex samples, the subset largely controls for unmeasured confounding factors. These unmeasured factors could have inflated the effect sizes in the full cohort. A future study with a larger number of participants followed from pre- to post-first sex and with additional specimens collected for a longer period of time post-sex would be necessary to resolve this uncertainty.

While it has long been suspected that sexual activity induces CVT changes, our study, with post-first sex samples collected a median of only 66 days after first sex, clarifies that the association of inflammatory changes with sexual activity occurs very early after the start of sexual activity. In addition to an early start of the inflammatory changes post-first sex, we observed an accumulation over time. The rates of increase over time varied between immune mediators, but on average indicated a doubling in concentration within about 5 months. For example, the immune mediator with the median rate of increase was CCL20 and it increased by 0.69 log2 pg/mL per 100 days (Figure 4). Furthermore, the association appears even without any incipient BV or STI, and might need to be considered a normal component of sexual activity.

The observed increases in inflammation may be important clinically in modifying risk of STI acquisition and in promoting fertility. There has been speculation that AGYW are at increased risk of STI due to physiological differences in the CVT compared to older women, in addition to behavioral risk factors (Hwang et al., 2011; Masson et al., 2015; McKinnon and Karim, 2016; Venkatesh and Cu-Uvin, 2013). Our research shows that there is an association between starting to have sex and increases in immune mediator concentration. Whether the increase in immune mediators brings protection from or vulnerability to STIs is unclear. Increased inflammation may prepare the CVT to prevent infection upon exposure. At the same time, increased expression of inflammatory mediators may recruit CD4 T cells, increasing the abundance of HIV target cells and potentially the risk of HIV infection. In addition to preventing infection, immune mediators play important roles in implantation and pregnancy. Therefore, the associations we observed with increases in immune mediators and start of sexual activity may be relevant for fertility. For example, IL-6 induces sperm capacitation, increasing its fertilizing ability (Laflamme et al., 2005). In addition, trophoblast cells secrete immune mediators to attract and regulate immune cells within the placenta, which is a necessary process for successful pregnancy (Mor et al., 2011).

Mechanism of immune mediator increase

Several mechanisms may play a role in the association of sexual activity with increased CVT immune mediator concentrations: BV, semen exposure, vaginal practices including washing, and physical microtrauma.

Numerous studies show that BV is rare prior to first sex, increases after first sex (Tabrizi et al., 2006; Fethers et al., 2012; Fethers et al., 2011; Fethers et al., 2009; Francis et al., 2020; Francis et al., 2019; Mitchell et al., 2012; Thoma et al., 2011a), is associated with recent sexual activity, and is associated with exposure to semen (Gajer et al., 2012; Ma et al., 2013; Jewanraj et al., 2021; Mngomezulu et al., 2021; Thoma et al., 2011b). Therefore, vaginal bacteria may play a role in these increased immune mediator levels. Jespers et al. reported that the increased CVT cytokines they observed post-sex were partly mediated by increases in detection and concentration of Gardnerella vaginalis and Atopobium vaginae (Jespers et al., 2016). Our finding of higher IL-1α and IL-1β post-first sex is consistent with literature linking BV to increased CVT IL-1α and IL-1β. However, BV is also associated with reduced CXCL10 (Masson et al., 2019), which conflicts with our finding of increased CXCL10 after first sex. Our data demonstrate conclusively that even when limited to only women with no BV, immune mediators were increased post-first sex. However, a leading hypothesis for increased CVT concentrations post-first sex may be the establishment of non-Lactobacillus bacterial populations.

Exposure to semen has been associated with increases in cervical leukocytes (Pandya and Cohen, 1985; Thompson et al., 1992; Sharkey et al., 2012) and inflammatory cytokines, especially IL-6 and CXCL10, in most (Jewanraj et al., 2021; Mngomezulu et al., 2021; Sharkey et al., 2012; Francis et al., 2016; Jespers et al., 2017; Kyongo et al., 2012) but not all studies (Agnew et al., 2008; Nakra et al., 2016). In vitro exposure to seminal plasma induces cytokines including IL-1α, IL-6, and CXCL8 in cervical and vaginal epithelial cells (Sharkey et al., 2007) and cervical explants (Adefuye et al., 2014; Denison et al., 1999; Introini et al., 2017). Others have suggested that increases in immune mediators following exposure to semen may be a result of semen directly transferring cytokines into the CVT (Deese et al., 2021). However, detection of immune mediators directly transferred by semen is unlikely to explain our results, because semen was rarely detected in our study (by Y-chromosome DNA and PSA).

Vaginal washing changes the microbiome and increases BV (Lokken et al., 2019; Low et al., 2011). In this cohort, vaginal washing was reported by about 56% of participants, but data was not available for vaginal washing for the particular specimens analyzed here, and data was not available for how vaginal washing practices may have changed after first sex.

Penile-vaginal sexual intercourse can cause microtrauma in the vaginal walls (Norvell et al., 1984), and this process may be inflammatory. However, exposure to semen may be necessary for the inflammatory response to intercourse, because condoms have been shown to block that response (Sharkey et al., 2012). Our study was unable to evaluate the role of microtrauma in the inflammatory response observed.

Strengths and limitations

The strengths of our study include a large, well-characterized cohort with multiple specimens per participant. Our sampling technique allowed collection of CVL without a speculum exam, which improved the feasibility of this research in a population with limited sexual experience. In addition, our systematic review and meta-analysis place our results in the context of what is known in the literature and synthesize all three studies. Despite the diversity of these studies and populations, all three studies showed consistent results.

An important limitation of our study is possible misclassification of specimens from pre-sexual activity. It is important to note that all three cohorts presented here were specifically designed to assess the sexual activity of adolescents and used best practices in ascertaining this challenging key variable. Of the 123 samples where participants reported no sexual activity, we classified 12 (9.8%) as post-first sex based on STI, pregnancy, Y-chromosome, or PSA results. Misclassification of specimens as pre/post-first sex is a challenge in this field, which we did our best to mitigate, but residual misclassification is possible.

In a secondary analysis, we removed all samples where the participant used contraception, was pregnant, had a Nugent score above 3, or was positive for HSV-2 or CT. The results of this analysis were similar to our primary, univariate analysis of the full cohort. These factors have complex relationships with sexual activity: Each can be caused by sexual activity and several of these variables could also be influenced (in part) by changes in immune mediator level. Thus, they may mediate the effect of sex on immune mediator levels, or they may be a consequence of it. Our goal with this secondary analysis was not to precisely estimate the direct effect of beginning sexual activity in the absence of these factors, but rather to determine whether these factors solely explained our results. Because the effects remained after removing these samples, we conclude that these factors by themselves do not solely explain the results we observed.

We observed similar results across three different studies in both low- and high-income settings. The largest limitation is that the total sample size across all three studies was less than 500 AGYW. Our study did include a number of cases of chlamydia and BV, increasing its generalizability, given the commonness of these conditions. However, our study notably included few cases of HSV-2, a very common STI, which may reduce its generalizability if HSV-2 modifies the effect of sexual activity on immune mediators. Similarly, we excluded a small number of samples due to trichomonas or gonorrhea infection or active local HSV-1/HSV-2 production because there were too few samples to reliably analyze. Moreover, our Kenyan cohort included little socioeconomic diversity, with participants being similar in income, housing, education, and ethnicity but still revealed similar results to AGYW on other continents.

We observed higher total protein concentrations in the post-first sex samples, translating to smaller differences between pre- and post-first sex groups when immune mediator concentrations were normalized to total protein. It is possible that the increases in total protein concentration post-first sex are associated with the general increases in immune mediators observed post-first sex. However, immune mediators remained higher post-first sex even after normalizing to total protein, so global protein upregulation does not completely explain this phenomenon.

The three studies included in the meta-analysis differed in important ways which could affect comparability. These differences included country of origin, presence of BV in the population, sample collection method (Dezzutti et al., 2011), and immunoassay platform (Chowdhury et al., 2009). However, statistical heterogeneity was low to moderate for most immune mediators, indicating that the between-study variability was largely what would be expected due to random sampling. The high statistical heterogeneity for TNF-α indicates a need for further study of that mediator. Because we obtained the raw individual participant data from each study, we were able to analyze all three studies using the same methods, which improves their comparability.

Conclusions

We identified consistent increases in cervicovaginal immune mediators among participants after first penile-vaginal sexual intercourse. The dynamic nature of the inflammatory milieu after sexual activity may be a catalyst for changes that could promote acquisition of STIs. Because these specimens are from AGYW with little pre-existing BV or STI, our research is suggestive that this inflammatory milieu may provide risk independent of STIs or vaginal dysbiosis. Further research should focus on the exact causes of inflammation associated with first sex and assess whether this inflammation is potentially harmful or whether it offers any benefits. We envision that this information will contribute to expanding our toolset in the fight against the high STI susceptibility observed in AGYW.

Data availability

All raw data and analysis code are available in the supplementary files.

References

  1. Book
    1. Higgins J
    2. Thomas J
    3. Chandler J
    4. Cumpston M
    5. Li T
    6. Page M
    7. Welch V
    (2021)
    Cochrane Handbook for Systematic Reviews of Interventions (6.2 Edn)
    Cochrane.
  2. Report
    1. National AIDS Control Council
    (2018)
    Kenya AIDS Response Progress Report
    Ministry of Health.
    1. Norvell MK
    2. Benrubi GI
    3. Thompson RJ
    (1984)
    Investigation of microtrauma after sexual intercourse
    The Journal of Reproductive Medicine 29:269–271.
  3. Book
    1. R. Core Team
    (2018)
    R: A Language and Environment for Statistical Computing
    Vienna, Austria.

Decision letter

  1. Talía Malagón
    Reviewing Editor; McGill University, Canada
  2. Ricardo Azziz
    Senior Editor; University at Albany, SUNY, United States

Our editorial process produces two outputs: i) public reviews designed to be posted alongside the preprint for the benefit of readers; ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Cervicovaginal immune mediators increase when young women begin to have sexual intercourse: a prospective study and meta-analysis" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen Ricardo Azziz as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1. Please consider and discuss the potential for unmeasured confounding and mediation in the analysis, with potentially changes to the regression models in response to reviewer comments.

2. Please temper causal claims in the abstract and Discussion and provide further discussion as to alternative potential explanations for the findings.

3. Provide further description in the methods of the spline analysis.

4. Discuss the consistency in effect sizes between the paired and unpaired analyses.

5. Provide further discussion on whether there was difficulty in obtaining pre-sex samples from women, and whether there is a potential for selection bias due to unrepresentative sampling prior to sex.

Reviewer #1 (Recommendations for the authors):

– The authors use causal language throughout the paper, implying that sex increases the level of immune markers. While they are welcome to hypothesize this in the discussion, I would suggest being more cautious when describing the results in the main text, abstract, and title and use more descriptive language instead (Cervicovaginal immune mediators are higher in vaginal samples of women who report having initiated sex, rather than "Cervicovaginal immune mediators increase when young women begin to have sexual intercourse" – increase is a verb that implies causality), given the observational nature of the study and the potential for bias. I would suggest discussing more of the possibility of the biases mentioned below.

– In Table 1 I would have liked to see more demographic and socioeconomic variables measured in the study. As mentioned, studies of sexual behavior are very vulnerable to confounding due to social determinants of health. While it is possible that these variables are not different in pre- and post- sex samples in this study, it is necessary to show this for the reader to assess whether there is a potential for confounding.

– The variables included in the multivariate model are not appropriate if the objective is to adjust for confounders, as many of the variables are not confounders and are rather mediators or potentially downstream effects of immune markers. Adjusting for these would lead to a biased estimate of the causal effect of vaginal sex. The authors should take more time to reflect on the potential causal relationships between these variables and how they influence their analysis. Some reflections:

– Age: is the most clear-cut confounder in this analysis and should certainly be included in the model, as age influences sexual behavior and is likely to be correlated with levels of immune markers.

– Menstrual phase: is appropriate to include in the model despite not necessarily meeting the traditional definition of a confounder. The menstrual phase is unlikely to influence the initiation of sex. However, there may be chance differences in menstrual phase between samples pre- and post-sex that may warrant adjustment. As the menstrual phase is not a mediator or a downstream effect of sex, adjustment for this variable will not bias results.

– Pregnancy and contraception: these variables are potential mediators on the causal pathway between initiation of sex and levels of immune markers, and should not be included in the model. If the objective of the authors is to measure only the direct (non-mediated) effect of sex, then a more complex mediation analysis would be needed to disentangle the effects of sex, pregnancy, and contraception, and the model would need to include potential confounders which are associated with pregnancy and contraception use (ex. socioeconomic variables).

– Nugent score, Chlamydia infection, and HSV-2 seropositivity: these are complex variables that could be both mediators of the effects of sex but also downstream effects of changes in levels of immune markers. It is therefore problematic to adjust for these as they are potentially colliders on the causal pathway, and adjustment would potentially lead to selection (collider) bias. See the article by Hernán (Hernán A Structural Approach to Selection Bias, Epidemiology: 2004 – Volume 15 – Issue 5 – p 615-625 https://doi.org/10.1097/01.ede.0000135174.63482.43), especially figures 3-4.

– Unmeasured confounders: there is potentially some residual unmeasured confounding from variables which influence sexual initiation and which also potentially influence immune response which are not included in the model. The most likely are variables related to social determinants of health (socioeconomic status, ethnicity, etc.). If the authors have access to these variables from the original study, they should consider their inclusion in the model, especially if they are strongly correlated with sexual initiation.

– The strongest evidence that there is likely to be potential selection bias and unmeasured confounding in the main (unpaired) analysis comes from the paired analysis, where the point estimates for the effect sizes are much smaller than in the main analysis. I think this analysis is the strongest analysis in the paper and am surprised in fact that this was not considered as the main analysis. The authors should discuss the results from this analysis more, as the paired analysis controls for many of the unmeasured confounders that would bias the results in the main analysis, so the results from this analysis are more likely to be valid estimates of causal effect than the results from the unpaired analysis.

– While the analysis for effect of time in Figure 4 is interesting, I would have liked to see p-values for the spline effect in the model. It is easy to fit different curves pre- and post-sex, but it does not necessarily follow that the addition of a spline leads to a significant improvement in the model performance. If the p-values for the spline are not significant, it would suggest the data does not support that there is an effect of time since initiating sex. Furthermore, this analysis should be described in the methods, I did not see this analysis described anywhere.

– The analysis of the correlation between estimates from different models (Figure 5D) has no analytical utility and should be removed. It is expected the results from different models would be correlated as they are based on the same data. High correlation between estimates from different analyses does not mean that the results or the model structure is valid; the validity of an analysis is based on subject matter knowledge and consideration of whether the analysis is appropriate for the question at hand. This analysis is therefore not informative.

– The logistic and linear regressions provide different information about the data; the logistic regression informs us on whether a particular marker is more likely to be detected in post-sex samples, whereas the linear regression informs us on whether the level of a marker is higher in post-sex samples. I would have therefore been interested in seeing the results from both regression models for all of the markers, instead of showing the linear regression results for some and the logistic regression results for others.

– I think Figure 3 is a good way to present results (multiplicative change in marker levels). I would suggest also adding the point estimates of the ratios and their confidence intervals on the right-hand side of each estimate (forest plot style), as the ratio estimates are more intuitive than the regression coefficients in Table 2. Also, the figures should distinguish between the scale of the axes (log 2) vs the units of the axes. Usually, the parentheses in the labels are used to indicate the axis units. The (log2) in the label may therefore confuse readers, as the axes do not show the effect of an increase in a log2 unit, but the ratios pre- vs. post-sex on a log2 scale. This could be better explained in the legend.

– Figure 5C and 5D appear to be mislabeled.

Reviewer #2 (Recommendations for the authors):

– How easy technically to get CVL from AGYW pre-first sex? How is it possible to discern the impact of the sampling on the mediators?

– Why exclude STIs if these can be controlled for in the MV models? Also unclear why differentiating sexual activity and STIs is so critical if the effects of both often co-occur after sexual activity is initiated. Perhaps this makes sense since STIs were rare, but it seems like it could affect generalizability.

– The sensitivity analysis of participants sampled only pre- and post- sex is a strength, as this comparison will likely reduce heterogeneity between individuals. Here the focus should be less on P-values and more on consistency between effect sizes. A correlation is presented, but this could be heterogeneous because of estimates being stronger or weaker in the paired analysis. Therefore it may make more sense to compare these on a case-by-case basis, depending on the consistency in these trends.

– Abstract- doesn't make sense to say a "median" 54% increase, as the % increase will mean something different for different markers, depending on their individual distributions. Markers with wider distributions will naturally have great % increases.

– Re the 57 samples reported after first sex, and the 12 samples deemed as such due to pregnancy, presence of semen biomarkers, STIs etc- is the assumption that the 12 samples did not overlap with the 57, e.g. represent under-reporting of first sex?

– In figure 5, even though most covariates didn't affect the estimates, would some of these be considered possible "mediators" as opposed to "confounders". E.g. BV has been associated with lower CXCL10 levels, so changes in BV post sex (which has also been shown) could lead to decreases in some of these key chemokines. This is worth considering, in particular since participants were sampled for the most part a long time before/after sex. It does appear in the Discussion. Is there clear enough evidence to conclude that the effects are unlikely to be due to sex alone? The Discussion of this on page 16 re: CXCL10 does not seem to match how the data are presented.

– Adjusting for protein concentration- certainly worth reporting this increases post-sex, but could this be due to general increases in inflammation following sex.

– Bottom of page 12, paragraph starting with "The specimens from the fully-paired…" – it is unclear what the point of this paragraph is.

https://doi.org/10.7554/eLife.78565.sa1

Author response

Essential revisions:

1. Please consider and discuss the potential for unmeasured confounding and mediation in the analysis, with potentially changes to the regression models in response to reviewer comments.

We have substantially changed the multivariate modeling approach and provided a fuller discussion of the potential for unmeasured confounding, as detailed in the response to reviewer 1 below.

2. Please temper causal claims in the abstract and Discussion and provide further discussion as to alternative potential explanations for the findings.

We have replaced language implying causality with language specifying associations throughout the Abstract and Discussion as well as in the title. We have also added two paragraphs near the beginning of the Discussion where we discuss alternative possible explanations for our findings. These changes are detailed in the response to reviewer 1 below.

3. Provide further description in the methods of the spline analysis.

To address this point, we added a new paragraph to the Methods and a new table to the Results section (new Table 4), which are further described in the response to reviewer 1 below.

4. Discuss the consistency in effect sizes between the paired and unpaired analyses.

We expanded our presentation of the differences between these two analyses and made it more nuanced. We moved it from the Strengths and Limitations section towards the end of the Discussion to feature more prominently close to the beginning. This change is detailed in the response to reviewer 1 below.

5. Provide further discussion on whether there was difficulty in obtaining pre-sex samples from women, and whether there is a potential for selection bias due to unrepresentative sampling prior to sex.

Our study attempted to recruit from as broad of a population as possible by using a community recruitment strategy (as opposed to recruiting from adolescents seeking clinical care). We collected CVL specimens without the use of a speculum exam, which made it more feasible to collect pre-sex specimens. In addition, we used the meta-analysis as a further way to verify results from outside of our own cohort to try to assess whether these results were applicable in other geographic populations and with different laboratories. We feel satisfied that our cohort results are similar to other cohorts in our meta-analysis. This issue is further discussed in the response to reviewer 2 below.

Reviewer #1 (Recommendations for the authors):

– The authors use causal language throughout the paper, implying that sex increases the level of immune markers. While they are welcome to hypothesize this in the discussion, I would suggest being more cautious when describing the results in the main text, abstract, and title and use more descriptive language instead (Cervicovaginal immune mediators are higher in vaginal samples of women who report having initiated sex, rather than "Cervicovaginal immune mediators increase when young women begin to have sexual intercourse" – increase is a verb that implies causality), given the observational nature of the study and the potential for bias. I would suggest discussing more of the possibility of the biases mentioned below.

We have changed the language throughout describing the results as an association rather than suggesting causality. These changes are tracked in the revised manuscript; many were very small modifications of sentences, so we have not reproduced them here.

We added two paragraphs to the Discussion (pages 18-19) where we give a more nuanced assessment of the possibility of causality as well as alternative explanations for the results we observed. These paragraphs read as follows:

“While we observed a clear association between sexual activity and higher immune mediator concentrations, there is room for caution about whether the association is causal. For example, there could be differences in social determinants of health between the participants who do and do not engage in sexual activity. These differences were largely unmeasured in all three studies and may contribute to the changes we observed in immune mediator levels pre- and post-first sex. In our Kenyan study, we had information from the baseline visit about participant age, education, income, and housing. The effects of first sex on immune mediators remained the same after adjusting for these variables, suggesting that age, education, income, and housing at baseline do not explain the differences we observed. However, we did not assess these variables for changes over time, only at baseline.

“The fully paired subset of our cohort (where participants provided pre and post samples) had smaller effect sizes than the full cohort. Most effects remained positive, but they were smaller in magnitude than in the primary analysis and all were non-significant. A possible explanation for the smaller effects is that immune mediator concentrations increase over time post-first sex (Figure 4). The post-first sex samples in the fully paired subset were collected sooner after first sex (median 32 days) than those from the participants who provided only post-first sex samples (median 157 days). Thus, the effects may have been weaker in the fully paired subset, in part, because the samples were obtained such a short time after first sex. Another possible explanation for the smaller effects is unmeasured confounding in the full cohort. Because participants in this subset provided both pre- and post-first sex samples, the subset largely controls for unmeasured confounding factors. These unmeasured factors could have inflated the effect sizes in the full cohort. A future study with a larger number of participants followed from pre- to post-first sex and with additional specimens collected for a longer period of time post-sex would be necessary to resolve this uncertainty.”

We changed the title to “Starting to have sexual intercourse is associated with increases in cervicovaginal immune mediators in young women: a prospective study and meta-analysis”

– In Table 1 I would have liked to see more demographic and socioeconomic variables measured in the study. As mentioned, studies of sexual behavior are very vulnerable to confounding due to social determinants of health. While it is possible that these variables are not different in pre- and post- sex samples in this study, it is necessary to show this for the reader to assess whether there is a potential for confounding.

Thank you for the helpful comments about social determinants of health both here and throughout this review. We have added information about education, income, rural/urban residence, and roof material (an indicator of poverty) to Table 1 (page 7). These questions were only asked at the enrollment visit. In addition, we now state that all participants were Black Africans assigned female at birth (page 6). Kenyan ethnicity was not collected from participants.

– The variables included in the multivariate model are not appropriate if the objective is to adjust for confounders, as many of the variables are not confounders and are rather mediators or potentially downstream effects of immune markers. Adjusting for these would lead to a biased estimate of the causal effect of vaginal sex. The authors should take more time to reflect on the potential causal relationships between these variables and how they influence their analysis. Some reflections:

– Age: is the most clear-cut confounder in this analysis and should certainly be included in the model, as age influences sexual behavior and is likely to be correlated with levels of immune markers.

– Menstrual phase: is appropriate to include in the model despite not necessarily meeting the traditional definition of a confounder. The menstrual phase is unlikely to influence the initiation of sex. However, there may be chance differences in menstrual phase between samples pre- and post-sex that may warrant adjustment. As the menstrual phase is not a mediator or a downstream effect of sex, adjustment for this variable will not bias results.

– Pregnancy and contraception: these variables are potential mediators on the causal pathway between initiation of sex and levels of immune markers, and should not be included in the model. If the objective of the authors is to measure only the direct (non-mediated) effect of sex, then a more complex mediation analysis would be needed to disentangle the effects of sex, pregnancy, and contraception, and the model would need to include potential confounders which are associated with pregnancy and contraception use (ex. socioeconomic variables).

– Nugent score, Chlamydia infection, and HSV-2 seropositivity: these are complex variables that could be both mediators of the effects of sex but also downstream effects of changes in levels of immune markers. It is therefore problematic to adjust for these as they are potentially colliders on the causal pathway, and adjustment would potentially lead to selection (collider) bias. See the article by Hernán (Hernán A Structural Approach to Selection Bias, Epidemiology: 2004 – Volume 15 – Issue 5 – p 615-625 https://doi.org/10.1097/01.ede.0000135174.63482.43), especially figures 3-4.

– Unmeasured confounders: there is potentially some residual unmeasured confounding from variables which influence sexual initiation and which also potentially influence immune response which are not included in the model. The most likely are variables related to social determinants of health (socioeconomic status, ethnicity, etc.). If the authors have access to these variables from the original study, they should consider their inclusion in the model, especially if they are strongly correlated with sexual initiation.

Thank you for these insightful comments. In response, we made the following major changes:

1) Removed the multivariable model

2) Added a new multivariable model with a new model that adjusts for age, education, income, rural/urban residence, and roof material (proxy for household wealth). The effect sizes from this new model are very similar to the effect sizes from the univariate model.

a. Results page 13: “In addition, we sought to assess whether the differences between pre- and post-first sex would be explained by social determinants of health. There was limited diversity in our cohort in these variables, in particular in terms of race and ethnicity. We adjusted the models from our primary analysis for age, education, whether the participants reported a regular source of income, whether the participants lived in an urban or rural environment, and whether the participants lived in a home with a metal or tile roof (lacking such a roof being an indicator of poverty). As shown in Figure 5AB, the effect sizes from the multivariate models were similar to those from the univariate models, suggesting that these variables do not explain the results we observed.”

b. Discussion pages 18-19: Quoted two comments above.

3) Conducted a new univariate analysis on a subset of samples that were negative for pregnancy, contraception, BV, HSV-2, and CT. Our goal with this analysis was not to precisely estimate the direct effect of beginning sexual activity in the absence of these factors, but rather to determine whether these factors solely explained our results. Essentially we felt that if the differences between pre- and post-first sex completely disappeared when we removed these samples, then our results weren’t very interesting. That finding would indicate that it was just one of those factors causing the difference between pre- and post. Because the differences don’t disappear, we conclude that those factors are not the sole cause of the differences we observe – the differences must be caused by either beginning sexual intercourse or some other unmeasured variable(s).

a. Results page 12: “We next sought to determine whether the differences between pre- and post-first sex could be explained by co-variates that differed between the groups or that are known to affect immune mediator concentrations (Table 1). In particular, pregnancy, CT infection, HSV-2 seropositivity, use of contraception, and elevated Nugent score are all known to affect cervicovaginal immune mediator concentrations. Because of this, we wished to know whether one or more of these factors was responsible for the differences we observed between pre- and post-first sex samples. To address this possibility, we repeated our analysis on a restricted sample set, excluding all samples where at the visit the participant was pregnant, positive for CT or HSV-2, reported using contraception, or had a Nugent score of 4 or greater. The analysis of these 137 samples (102 pre- and 35 post-first sex) are shown in Figure 5AB. The effects remained positive for all immune mediators in this subset analysis, indicating that the differences we saw between pre- and post-first sex samples were not solely a result of acquisition of STIs or pregnancy.”

b. Discussion page 21: “In a secondary analysis, we removed all samples where the participant used contraception, was pregnant, had a Nugent score above 3, or was positive for HSV-2 or CT. The results of this analysis was similar to our primary, univariate analysis of the full cohort. These factors have complex relationships with sexual activity: Each can be caused by sexual activity and several of these variables could also be influenced (in part) by changes in immune mediator level. Thus, they may mediate the effect of sex on immune mediator levels or they may be a consequence of it. Our goal with this secondary analysis was not to precisely estimate the direct effect of beginning sexual activity in the absence of these factors, but rather to determine whether these factors solely explained our results. Because the effects remained after removing these samples, we conclude that these factors by themselves do not solely explain the results we observed.”

4) Removed claims that we have eliminated all important confounding factors and wrote a more nuanced paragraph about causality in the Discussion (quoted two comments above).

– The strongest evidence that there is likely to be potential selection bias and unmeasured confounding in the main (unpaired) analysis comes from the paired analysis, where the point estimates for the effect sizes are much smaller than in the main analysis. I think this analysis is the strongest analysis in the paper and am surprised in fact that this was not considered as the main analysis. The authors should discuss the results from this analysis more, as the paired analysis controls for many of the unmeasured confounders that would bias the results in the main analysis, so the results from this analysis are more likely to be valid estimates of causal effect than the results from the unpaired analysis.

We have added text near the beginning of the Discussion addressing the fully-paired analysis in much more detail and in a much more balanced fashion. We describe how this analysis suggests that the analysis of the full cohort may be over-estimating the effect. In addition, we describe how the shorter time since first sex in the fully-paired subset may also contribute to the smaller effect sizes. We conclude by recommending a larger paired study in the future with longer follow-up. The paragraph reads as follows (Discussion pages 18-19):

“The fully paired subset of our cohort (where participants provided pre and post samples) had smaller effect sizes than the full cohort. Most effects remained positive, but they were smaller in magnitude than in the primary analysis and all were non-significant. A possible explanation for the smaller effects is that immune mediator concentrations increase over time post-first sex (Figure 4). The post-first sex samples in the fully paired subset were collected sooner after first sex (median 32 days) than those from the participants who provided only post-first sex samples (median 157 days). Thus, the effects may have been weaker in the fully paired subset, in part, because the samples were obtained such a short time after first sex. Another possible explanation for the smaller effects is unmeasured confounding in the full cohort. Because participants in this subset provided both pre- and post-first sex samples, the subset largely controls for unmeasured confounding factors. These unmeasured factors could have inflated the effect sizes in the full cohort. A future study with a larger number of participants followed from pre- to post-first sex and with additional specimens collected for a longer period of time post-sex would be necessary to resolve this uncertainty.”

– While the analysis for effect of time in Figure 4 is interesting, I would have liked to see p-values for the spline effect in the model. It is easy to fit different curves pre- and post-sex, but it does not necessarily follow that the addition of a spline leads to a significant improvement in the model performance. If the p-values for the spline are not significant, it would suggest the data does not support that there is an effect of time since initiating sex. Furthermore, this analysis should be described in the methods, I did not see this analysis described anywhere.

Thank you for your comments about the spline analysis in Figure 4. In the first submission, we fit separate models pre- and post-first sex. In response to this comment, we changed to single models using linear splines, with a knot at day 0 (the day of first sex). This new analysis is shown in Figure 4. In addition, we added a new table (new Table 4; page 12), including the p-values for the time-dependent spline effects (of which 17/19 are significant at p<0.05).

Results page 10-11: “As shown in Figure 4 and Table 4, immune mediator concentrations were generally stable for three years prior to first sex and then increased sharply in the year following first sex. The change in slope following first sex was statistically significant at p<0.05 for 17/19 immune mediators. This pattern is consistent with cumulative increases in immune mediator concentrations following first sex.”

In addition, we have more thoroughly described this analysis in the Methods section. Thank you for pointing out that we had neglected to describe this analysis in the Methods.

Methods page 28: “We additionally wished to assess whether immune mediator concentrations changed over time with respect to the date of first sex. To assess this question, we fit mixed-effects models with linear splines, using days since first sex as a fixed effect and participant as a random effect, with the outcome being the concentration of immune mediator (log2 pg/mL). For the splines, we used a knot at day 0, which was the day of first sex.”

– The analysis of the correlation between estimates from different models (Figure 5D) has no analytical utility and should be removed. It is expected the results from different models would be correlated as they are based on the same data. High correlation between estimates from different analyses does not mean that the results or the model structure is valid; the validity of an analysis is based on subject matter knowledge and consideration of whether the analysis is appropriate for the question at hand. This analysis is therefore not informative.

We have removed the correlations and appreciate the correction.

– The logistic and linear regressions provide different information about the data; the logistic regression informs us on whether a particular marker is more likely to be detected in post-sex samples, whereas the linear regression informs us on whether the level of a marker is higher in post-sex samples. I would have therefore been interested in seeing the results from both regression models for all of the markers, instead of showing the linear regression results for some and the logistic regression results for others.

The reason we used logistic regressions for low-detected immune mediators is that the distributions of the concentrations deviates strongly from the normal distribution when most samples are below the limit of detection. We didn’t feel that linear regression was appropriate in those cases and while logistic regression certainly provides different information, we felt that it was the best alternative. Similarly, we could not have performed logistic regression on every immune mediator, as several immune mediators were detected in every sample.

We have added a sentence to the methods explaining our rationale for this choice (page 28: “Logistic models were used for immune mediators with low levels of detection because the distributions become increasingly non-normal as more samples fall below the lower limit of detection”).

For completeness, the Author response table 1 shows the results of linear and logistic models fit on every immune mediator. The Coef columns are the log2 difference for the linear models and the log-odds for the logistic model. The direction and significance of the changes overall are similar by both models. The coefficients have the same sign whether linear or logistic for every immune mediator except CCL5. The biggest p-value differences between models appear for immune mediators which were detected in most samples, which often had low p-values by linear analysis but had higher p-values by logistic analyses (IL6, CCL3, IL1b, CXCL10, CXCL9), presumably because such a small number of samples were undetectable for those immune mediators.

Author response table 1
LinearLogistic
AssayCoefStd. ErrorP-valueCoefStd. ErrorP-valuePercent detectable
IL-12p700.240.140.082.021.030.0520.0
IL-100.320.160.051.150.460.0122.8
IFNγ0.490.200.021.040.460.0226.7
IL-17A0.640.250.010.800.340.0250.6
IL-70.020.180.920.370.340.2750.6
TNF0.690.290.020.900.440.0468.3
IL-20.900.250.00051.280.520.0177.8
CCL40.420.270.120.740.560.1986.1
IL-60.810.390.041.942.240.3989.4
CCL30.620.220.0060.240.570.6891.7
CCL50.390.280.17-7.404.970.1493.9
CCL200.530.360.150.650.830.4395.6
IL-1b1.330.430.0020.460.850.5996.1
CXCL100.740.320.0232.92a8078958199.4
CXCL90.600.260.02176.10a8078958199.4
CXCL81.020.330.003b100
IL-1a0.670.240.006b100
IL-1RA0.480.220.03b100
IL-180.440.360.23b100

a One single pre-sex sample was undetectable and no post-sex samples were, explaining highly imprecise values.

b No samples were undetectable, so logistic models could not be fit.

– I think Figure 3 is a good way to present results (multiplicative change in marker levels). I would suggest also adding the point estimates of the ratios and their confidence intervals on the right-hand side of each estimate (forest plot style), as the ratio estimates are more intuitive than the regression coefficients in Table 2. Also, the figures should distinguish between the scale of the axes (log 2) vs the units of the axes. Usually, the parentheses in the labels are used to indicate the axis units. The (log2) in the label may therefore confuse readers, as the axes do not show the effect of an increase in a log2 unit, but the ratios pre- vs. post-sex on a log2 scale. This could be better explained in the legend.

Thank you for these helpful suggestions. We made the following changes:

– Relabeled the axes and improved the legends of Figures 3, 5, and 6

– Added ratios and CIs to the side of each estimate in Figure 3

– We did not add these values to Figures 5 or 6 because they would have made the figures too complicated. The values are available in Additional File 2 for both figures as well as in the forest plots in Additional File 6 for Figure 6.

– We also added the ratios and CIs to Tables 2, 3, and 5. This way, when looking at the tables, the reader can interpret either log2 differences or ratios, whichever they find more intuitive.

– Figure 5C and 5D appear to be mislabeled.

Thank you for pointing out the error. We have corrected this mistake.

Reviewer #2 (Recommendations for the authors):

– How easy technically to get CVL from AGYW pre-first sex? How is it possible to discern the impact of the sampling on the mediators?

We used an innovative sampling technique that collected CVL without use of a speculum, which was thought to be unpleasant and unnecessary in AGYW pre-first sex. We placed a small amount of IV tubing in the fornix and instilled saline in that way, this was well tolerated. We have added a sentence to the Strengths and limitations section of the Discussion noting that this method made it technically possible to perform the study (page 20: “Our sampling technique allowed collection of CVL without a speculum exam, which improved the feasibility of this research in a population with limited sexual experience”).

In addition, we sought to recruit from a broad population. Our study used a community recruitment strategy and did not recruit from adolescents seeking clinical care, therefore we might have recruited youth who were uncomfortable with clinical procedures. We used the meta-analysis as a further way to verify results from outside of our own cohort to try to assess whether these results were applicable in other geographic populations and with different laboratories. We feel satisfied that our cohort results are similar to other cohorts in our meta-analysis.

– Why exclude STIs if these can be controlled for in the MV models?

We excluded 5 samples due to rare STIs because we felt we could not accurately control for them in the multivariate models given the small sample sizes (n=1-2 per STI). We clarified this point in the Results section. However, we included samples with more common (in our cohort) STIs.

Excluded: Trichomonas (1 sample), Gonorrhea (1 sample), active genital HSV-1 infection (1 sample), active genital HSV-2 infection (2 samples), where “active” means DNA was detected in a genital sample.

Included: Chlamydia, HSV-1 seropositive, HSV-2 seropositive, (samples with BV were also included)

Also unclear why differentiating sexual activity and STIs is so critical if the effects of both often co-occur after sexual activity is initiated. Perhaps this makes sense since STIs were rare, but it seems like it could affect generalizability.

Our goal was to determine if sexual activity itself affects the genital immune milieu. For that reason, we felt it was important to determine whether any changes were solely a result of STIs. If we had observed changes in the univariate model that disappeared after controlling for STIs, we would have concluded that cytokine changes post-first sex were only a result of STIs. Instead, we observed that the changes persisted, suggesting that they are associated with sexual activity itself (or some other unknown factor), but not specifically the STIs present in the cohort.

– The sensitivity analysis of participants sampled only pre- and post- sex is a strength, as this comparison will likely reduce heterogeneity between individuals. Here the focus should be less on P-values and more on consistency between effect sizes. A correlation is presented, but this could be heterogeneous because of estimates being stronger or weaker in the paired analysis. Therefore it may make more sense to compare these on a case-by-case basis, depending on the consistency in these trends.

Thank you for these comments. As also discussed in the response to Reviewer 1, we have removed the correlations. We have substantially expanded our discussion of this analysis and focused on comparing effect sizes in our interpretation. We have additionally tempered our claims of causality based on this analysis.

– Abstract- doesn't make sense to say a "median" 54% increase, as the % increase will mean something different for different markers, depending on their individual distributions. Markers with wider distributions will naturally have great % increases.

We have removed this passage from the abstract.

– Re the 57 samples reported after first sex, and the 12 samples deemed as such due to pregnancy, presence of semen biomarkers, STIs etc- is the assumption that the 12 samples did not overlap with the 57, e.g. represent under-reporting of first sex?

Thank you for pointing out that this issue was confusing. We added a sentence to the Results clarifying that there were a total of 69 post-first sex samples (57 reported and 12 inferred based on test results). The relevant section (page 5) now reads:

“Of these, 111 samples from 75 participants were classified as pre-first sex and 57 samples from 45 participants were obtained after reported first sex. In addition, 12 samples from 8 participants were classified as post-sex due to presence of Y-chromosome DNA, PSA or because of pregnancy or a positive Chlamydia trachomatis (CT) test at that visit or a prior visit. Thus, there were a total of 69 post-sex specimens from 53 participants.”

The (important!) issue of underreporting is discussed in the Strengths and limitations section of the Discussion:

Page 20-21: “An important limitation of our study is possible misclassification of specimens from pre-sexual activity. It is important to note that all three cohorts presented here were specifically designed to assess the sexual activity of adolescents and used best practices in ascertaining this challenging key variable. Of the 123 samples where participants reported no sexual activity, we classified 12 (9.8%) as post-first sex based on STI, pregnancy, Y-chromosome, or PSA results. Misclassification of specimens as pre/post-first sex is a challenge in this field, which we did our best to mitigate, but residual misclassification is possible.”.

– In figure 5, even though most covariates didn't affect the estimates, would some of these be considered possible "mediators" as opposed to "confounders". E.g. BV has been associated with lower CXCL10 levels, so changes in BV post sex (which has also been shown) could lead to decreases in some of these key chemokines. This is worth considering, in particular since participants were sampled for the most part a long time before/after sex. It does appear in the Discussion. Is there clear enough evidence to conclude that the effects are unlikely to be due to sex alone? The Discussion of this on page 16 re: CXCL10 does not seem to match how the data are presented.

Thank you for your comments about certain co-variates being mediators rather than confounders. This issue was also raised by Reviewer 1. We have substantially changed our multivariate model and reworked the presentation and discussion of results pertaining to the covariates. We additionally added a new analysis of a subset of the full cohort containing only the subset of samples that were negative for pregnancy, contraception, BV, HSV-2, and CT. Because the effects remain positive in that subset, we conclude that these factors are not solely causing the effects we see. Rather, the effects likely derive from sex itself as well as possible unmeasured variables.

We removed the Discussion of CXCL10 because it derived from the earlier multivariable model, which is no longer in the paper. We agree with you about CXCL10: BV decreases CXCL10 and BV increases post-first sex. Though we still see an increase in CXCL10 post-first sex, BV is probably reducing its magnitude compared to the increase that would be seen without BV. Thoroughly disentangling these interactions is beyond the scope of this paper and set of samples.

– Adjusting for protein concentration- certainly worth reporting this increases post-sex, but could this be due to general increases in inflammation following sex.

Thank you for this suggestion. We have added it to the Discussion on page 21: “It is possible that the increases in total protein concentration post-first sex are associated with the general increases in immune mediators observed post-first sex.”

– Bottom of page 12, paragraph starting with "The specimens from the fully-paired…" – it is unclear what the point of this paragraph is.

We agree that the purpose of this paragraph was not clear as written. The purpose was to discuss the comparability of specimens from the fully-paired subset to the rest of the specimens and we have rephrased the paragraph to make that clear (now on page 15).

“The specimens from the fully-paired subset were generally comparable to the specimens from the participants who only provided post specimens, with one important difference: the post specimens in the fully-paired subset were obtained sooner after reported date of first sex (p = 4.7E-8). In the fully-paired subset, the post-first sex samples came only about a month after first sex (median 31.5 days). In contrast, the post-first sex samples from the other participants came about four months after first sex (median 157 days). The age of the participants at the time of first sex, however, was similar (18.9 years in each subset, p = 0.90).”

https://doi.org/10.7554/eLife.78565.sa2

Article and author information

Author details

  1. Sean M Hughes

    Department of Obstetrics and Gynecology, University of Washington, Seattle, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Visualization, Methodology, Writing – original draft
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-9409-9405
  2. Claire N Levy

    Department of Obstetrics and Gynecology, University of Washington, Seattle, United States
    Contribution
    Data curation, Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3204-211X
  3. Fernanda L Calienes

    Vaccine and Infectious Disease Division, Fred Hutchinson Cancer Research Center, Seattle, United States
    Contribution
    Data curation, Investigation
    Competing interests
    No competing interests declared
  4. Katie A Martinez

    Department of Obstetrics and Gynecology, University of Washington, Seattle, United States
    Contribution
    Data curation, Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3411-065X
  5. Stacy Selke

    Department of Laboratory Medicine & Pathology, University of Washington, Seattle, United States
    Contribution
    Data curation, Investigation, Methodology, Project administration, Writing – review and editing
    Competing interests
    No competing interests declared
  6. Kenneth Tapia

    Department of Global Health, University of Washington, Seattle, United States
    Contribution
    Formal analysis, Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  7. Bhavna H Chohan

    1. Department of Global Health, University of Washington, Seattle, United States
    2. Centre for Virus Research, Kenya Medical Research Institute, Nairobi, Kenya
    Contribution
    Resources, Project administration, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2125-6799
  8. Lynda Oluoch

    Centre for Clinical Research, Kenya Medical Research Institute, Nairobi, Kenya
    Contribution
    Resources, Project administration, Writing – review and editing
    Competing interests
    No competing interests declared
  9. Catherine Kiptinness

    Centre for Clinical Research, Kenya Medical Research Institute, Nairobi, Kenya
    Contribution
    Resources, Project administration, Writing – review and editing
    Competing interests
    No competing interests declared
  10. Anna Wald

    1. Vaccine and Infectious Disease Division, Fred Hutchinson Cancer Research Center, Seattle, United States
    2. Department of Laboratory Medicine & Pathology, University of Washington, Seattle, United States
    3. Department of Medicine, University of Washington, Seattle, United States
    4. Department of Epidemiology, University of Washington, Seattle, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3486-6438
  11. Mimi Ghosh

    Department of Epidemiology, Milken Institute School of Public Health, The George Washington University, Washington, United States
    Contribution
    Resources, Writing – review and editing
    Competing interests
    No competing interests declared
  12. Liselotte Hardy

    Department of Clinical Sciences, Unit of Tropical Bacteriology, Institute of Tropical Medicine, Antwerp, Belgium
    Contribution
    Resources, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8785-5081
  13. Kenneth Ngure

    1. Department of Global Health, University of Washington, Seattle, United States
    2. Department of Community Health, Jomo Kenyatta University of Agriculture and Technology, Nairobi, Kenya
    Contribution
    Conceptualization, Resources, Funding acquisition, Project administration, Writing – review and editing
    Competing interests
    Kenneth Ngure was supported by the International AIDS Society to attend AIDS 2022. The author has no other competing interests to declare
  14. Nelly R Mugo

    1. Department of Global Health, University of Washington, Seattle, United States
    2. Centre for Clinical Research, Kenya Medical Research Institute, Nairobi, Kenya
    Contribution
    Conceptualization, Resources, Funding acquisition, Project administration, Writing – review and editing
    Competing interests
    Nelly R Mugo received honoraria from MERCK Ltd in support of a presentation on HPV vaccination uptake in LMIC. The author has no other competing interests to declare
  15. Florian Hladik

    1. Department of Obstetrics and Gynecology, University of Washington, Seattle, United States
    2. Vaccine and Infectious Disease Division, Fred Hutchinson Cancer Research Center, Seattle, United States
    3. Department of Medicine, University of Washington, Seattle, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Methodology, Writing – original draft, Project administration
    Contributed equally with
    Alison C Roxby
    For correspondence
    florian@uw.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0375-2764
  16. Alison C Roxby

    1. Vaccine and Infectious Disease Division, Fred Hutchinson Cancer Research Center, Seattle, United States
    2. Department of Global Health, University of Washington, Seattle, United States
    3. Department of Medicine, University of Washington, Seattle, United States
    4. Department of Epidemiology, University of Washington, Seattle, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Methodology, Writing – original draft, Project administration
    Contributed equally with
    Florian Hladik
    For correspondence
    aroxby@uw.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9449-8062

Funding

Eunice Kennedy Shriver National Institute of Child Health and Human Development (HD091996)

  • Alison C Roxby

National Institute of Allergy and Infectious Diseases (AI030731)

  • Anna Wald

National Institute of Allergy and Infectious Diseases (AI116292)

  • Florian Hladik

National Institute of Allergy and Infectious Diseases (AI154366)

  • Florian Hladik

National Institute of Allergy and Infectious Diseases (AI027757)

  • Florian Hladik

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The authors would like to thank the following for their contributions: study staff and participants in Thika, Kenya; the Endocrine Technologies Core (NIH P51OD011092) at the Oregon National Primate Research Center for measuring progesterone concentrations; Dr Scott McClelland and the University of Washington/University of Nairobi East Africa STI Laboratory in Mombasa, Kenya for STI testing; Diana Louden from the University of Washington Health Sciences Library for devising the systematic review search strategy; Dr. Veronica Gomez-Lobo for performing the clinical components of the Washington DC study. Funding: This research was funded by R01 HD091996-01 (ACR), by P01 AI 030731–25 (Project 1) (AW), R01 AI116292 (FH), R03 AI154366 (FH), and by the Center for AIDS Research (CFAR) of the University of Washington/Fred Hutchinson Cancer Research Center AI027757. Laboratory work was done at the CFAR Core Immunology Laboratory, supported by AI027757. The funders had no role in study design, data collection and analysis, or preparation of the manuscript. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Ethics

Human subjects approval was obtained from the Kenya Medical Research Institute Scientific Ethics Review Unit (protocol 2760) and the University of Washington Institutional Review Board (number 00000946). Participants under age 18 provided written informed assent, and written informed consent was obtained from a parent/guardian. Participants assented privately from parents/guardians, after asking questions and deciding whether they wanted to participate free of parental influence. Participants age 18 or older provided written informed consent. No personally identifiable information was obtained from the studies included in the meta-analysis. Participants in both studies provided informed consent and consent to publish as described in the original manuscripts.

Senior Editor

  1. Ricardo Azziz, University at Albany, SUNY, United States

Reviewing Editor

  1. Talía Malagón, McGill University, Canada

Publication history

  1. Received: March 11, 2022
  2. Preprint posted: April 1, 2022 (view preprint)
  3. Accepted: September 13, 2022
  4. Version of Record published: October 25, 2022 (version 1)

Copyright

© 2022, Hughes et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 726
    Page views
  • 46
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Sean M Hughes
  2. Claire N Levy
  3. Fernanda L Calienes
  4. Katie A Martinez
  5. Stacy Selke
  6. Kenneth Tapia
  7. Bhavna H Chohan
  8. Lynda Oluoch
  9. Catherine Kiptinness
  10. Anna Wald
  11. Mimi Ghosh
  12. Liselotte Hardy
  13. Kenneth Ngure
  14. Nelly R Mugo
  15. Florian Hladik
  16. Alison C Roxby
(2022)
Starting to have sexual intercourse is associated with increases in cervicovaginal immune mediators in young women: a prospective study and meta-analysis
eLife 11:e78565.
https://doi.org/10.7554/eLife.78565
  1. Further reading

Further reading

    1. Immunology and Inflammation
    Lyra O Randzavola, Paige M Mortimer ... David C Thomas
    Research Article

    EROS (Essential for Reactive Oxygen Species) protein is indispensable for expression of gp91phox, the catalytic core of the phagocyte NADPH oxidase. EROS deficiency in humans is a novel cause of the severe immunodeficiency, chronic granulomatous disease (CGD), but its mechanism of action was unknown until now. We elucidate the role of EROS, showing it acts at the earliest stages of gp91phox maturation. It binds the immature 58kDa gp91phox directly, preventing gp91phox degradation and allowing glycosylation via the oligosaccharyltransferase (OST) machinery and the incorporation of the heme prosthetic groups essential for catalysis. EROS also regulates the purine receptors P2X7 and P2X1 through direct interactions and P2X7 is almost absent in EROS deficient mouse and human primary cells. Accordingly, lack of murine EROS results in markedly abnormal P2X7 signalling, inflammasome activation and T cell responses. The loss of both ROS and P2X7 signalling leads to resistance to influenza infection in mice. Our work identifies EROS as a highly selective chaperone for key proteins in innate and adaptive immunity and a rheostat for immunity to infection. It has profound implications for our understanding of immune physiology, ROS dysregulation and possibly gene therapy.

    1. Computational and Systems Biology
    2. Immunology and Inflammation
    Mingyao Pan, Bo Li
    Short Report Updated

    T cells are potent at eliminating pathogens and playing a crucial role in the adaptive immune response. T cell receptor (TCR) convergence describes T cells that share identical TCRs with the same amino acid sequences but have different DNA sequences due to codon degeneracy. We conducted a systematic investigation of TCR convergence using single-cell immune profiling and bulk TCRβ-sequence (TCR-seq) data obtained from both mouse and human samples and uncovered a strong link between antigen-specificity and convergence. This association was stronger than T cell expansion, a putative indicator of antigen-specific T cells. By using flow-sorted tetramer+ single T cell data, we discovered that convergent T cells were enriched for a neoantigen-specific CD8+ effector phenotype in the tumor microenvironment. Moreover, TCR convergence demonstrated better prediction accuracy for immunotherapy response than the existing TCR repertoire indexes. In conclusion, convergent T cells are likely to be antigen-specific and might be a novel prognostic biomarker for anti-cancer immunotherapy.