Daily Vaginal Swabs and Mobile Phone Sex Report for Assessing HIV Virion Exposure Prospectively Among a Cohort of Young Sexually Active Women in South Africa (HVTN 915)

Abstract

Background:

Measurements of HIV exposure could help identify subpopulations at highest risk of acquisition, and improve the design of HIV prevention efficacy trials and public health interventions. The HVTN915 study evaluated the feasibility of self-administered vaginal swabs for detection of HIV virions to assess exposure.

Methods:

Fifty 18-to-25-year-old sexually active HIV-seronegative women using contraception were enrolled in Soweto, South Africa. Participants self-administered daily vaginal swabs and answered sexual behavior questions via mobile phone for 90 days. Clinician-administered vaginal swabs, behavioral questionnaires, HIV diagnostic testing, and counselling were performed at eight clinic visits. Glycogen concentrations assessed adherence to swabbing. Y-chromosome DNA (Yc-DNA) assessed the accuracy of reported condom use. HIV exposure was measured by virion PCR in swabs from 41 women who reported unprotected vaginal sex during follow-up.

Results:

Glycogen was detected in 315/336 (93.8%) participant-collected and in all clinician-collected swabs. 20/39 daily swabs (51.3%) linked to mobile reports of unprotected sex tested positive for Yc-DNA, whereas 10/187 swabs collected after three days of abstinence or protected sex (5.3%) had detectable Yc-DNA. No participant became HIV infected during the study, yet exposure to HIV was detected by nucleic acids in two vaginal swabs from one participant, collected less than one hour post-coitus.

Conclusion:

There was high adherence to daily vaginal swabbing. Daily mobile surveys had accurate reporting of unprotected sex. Detection of HIV in self-collected vaginal swabs from an uninfected participant demonstrated it was possible to measure HIV exposure, but the detection rate was lower than expected.

INTRODUCTION

Globally, HIV is a leading cause of death in women of reproductive age.¹ More than 44% of new infections in women are among 15–24-year-olds¹, so studying and preventing infections in this population is of utmost importance.

Current research and interventions to prevent HIV and other genital tract infections (GTIs) rely mainly on face-to-face interviewing of participants to collect sexual behaviors. This approach is susceptible to social desirability bias, recall bias, and misclassification.^2,3 Because of the absence of an interviewer, Computer Assisted Self-Interviewing (CASI) could provide more accurate sex reports, with less social desirability bias. However, CASI has shown only small improvements in protected sex reporting over face-to-face interviewing when evaluated using biomarkers of semen in vaginal secretions.^4–6

Recently, mobile phone services have been used for health promotion^7–10 and for sexual behavior surveys.^8,11 However, it is unclear whether this survey method is any less susceptible to biases, as it has not been compared to biomarkers of unprotected sexual activity. Biomarker evaluations of unprotected sex, such as GTIs,^12,13 prostate-specific antigen (PSA),^14,15 or Y chromosome DNA (Yc-DNA)¹⁶ are more costly and invasive than interviewing methods. However, their independence from self-reporting may increase the validity of research and more accurately inform public health policy.¹⁷

Although seminal biomarkers and GTIs associate well with the practice of unprotected sex and the risk of HIV infection, there is currently no method to determine vaginal HIV exposure. A daily, direct indicator of HIV exposure could 1) improve accuracy in identifying transmission events, and 2) contribute to the screening of individuals at risk of HIV for recruitment into interventions.

Self-administered vaginal swabs have successfully been used for diagnosis of several GTIs, and have shown applicability in public interventions addressing sexual health.^18–21 Diagnosis of chlamydia, gonorrhea, and human papillomavirus (HPV) via PCR in self-administered vaginal swabs correlates well with clinician collections, with high specificity and sensitivity.^21–23 Self-collection was preferred to clinical examination among clinicians and young women.^20,24,25 Thus, this method may support greater privacy, adherence and frequent sampling among young women at risk of HIV.

HVTN915 was a prospective observational study investigating the use of daily mobile phone sexual activity reporting and self-collected vaginal swabs for detection of HIV exposure among sexually active 18–25-year-old women in Soweto, South Africa.

METHODS

Study Design:

HVTN915 enrolled 50 healthy, HIV-seronegative, sexually active women, 18–25 years of age, reporting at least three acts of vaginal sex with men in the 30 days prior to screening. Women were recruited through community face-to-face contact with study recruiters in hotspots around Soweto, South Africa. (e.g. taverns, brothels). In addition to age and recruitment location, sexual behavior eligibility criteria used epidemiological data on HIV prevalence in South Africa²⁶: i) women who had vaginal intercourse 4 or more times per week in the 30 days preceding screening, ii) women who engaged in transactional sex, iii) unmarried women living together with a partner, iv) women who engaged in unprotected vaginal or anal sex (sex without a condom), and v) women with any history of genital ulcer disease.

Eligible participants underwent HIV diagnostic testing and counselling, provided informed consent, and were capable of using the mobile phone provided as part of the study. They had negative pregnancy tests (QuickVue One-Step [Quidel Corporation, San Diego, CA]) at enrolment, were not breastfeeding, and agreed to use effective contraception throughout the study.

At enrolment, all participants underwent a gynecological exam with Papanicolaou smear (if no examination in the past 12 months), GTI testing for gonorrhea, chlamydia, and if symptomatic, for bacterial vaginosis and Trichomonas vaginalis. After enrolment, GTI symptoms were solicited at all seven follow-up visits and all symptomatic GTIs were managed syndromically per South African standards.²⁷

For 90 days, participants were asked to perform self-administered vaginal swabbing every morning and respond to a brief behavioral questionnaire via an application installed in study-provided phones, which participants were able to keep after study completion. They attended clinic visits at weeks 0, 1, 2, 3, 4, 6, 8, and 12, and received reimbursement of 150 ZAR (~USD15) per visit. Volunteers underwent risk reduction, adherence, and contraceptive counselling; social impact assessments; face-to-face interviews; returned self-collected swabs; and received new swabs. At weeks 0, 1, 2, 6, and 8, participants underwent a clinician-administered vaginal swab. HIV diagnostic testing in blood was conducted at weeks 6 and 12.

Ethics Approvals:

The study was approved by the University of the Witwatersrand Human Research Ethics Committee. All study participants were informed of study procedures and signed consent documentation prior to enrolment.

HIV Blood Tests:

HIV blood testing was performed using two concurrent rapid tests (Determine HIV1/2 [Abbott, Umhlanga, South Africa]; and UniGold Recombigen HIV [Trinity Biotech, Wicklow, Ireland]). In case of discordancy among test results, a confirmatory chemiluminescent microparticle immunoassay (Abbott HIV-1/2 Ag/Ab Combo 4^th generation) completed diagnosis.

Vaginal Swab Collection and Elution:

Vaginal swabs were collected using the Gen-Probe Aptima Swab Kit (Hologic, Marlborough, MA), preventing degradation of HIV nucleic acids at room temperature for up to four weeks (data not shown). Clinicians and participants were asked to insert swabs 7cm into the vagina for 10–30 seconds, then place the swab into a tube containing 2.9mL of Aptima media. Swabs were matched to mobile phone behavioral data using swab number and date. The use of Aptima swabs for detection of seminal Y chromosome or HIV was not documented in the product label.

Self-collected samples were stored at room temperature until returned to the clinic; eluates from clinician swabs were processed on the day of collection. Processing consisted of squeezing the cotton bud, mixing eluate by pipetting, aliquoting, and storing at −80°C.

Glycogen Quantitation:

Vaginal glycogen was detected using EnzyChrom Glycogen Assay (BioAssay Systems, Hayward, CA), using fluorescence at 530/585nm read in a Spectramax iD3 multimode plate reader (Molecular Devices, San Jose, CA). For maximum sensitivity, adherence assessments for all samples were run in undiluted eluates in duplicate. If above the standard curve, dilutions were carried out. For analysis of adherence, the glycogen data was analyzed as presence (at or above lower limit of quantitation [LLQ]) or absence (below LLQ). For comparisons between clinic and participant collected swabs, absolute concentrations in swab eluates were compared.

Yc-DNA PCR:

DNA was isolated from swab eluates using the QIAamp DNA Isolation Kit (Qiagen, Hilden, Germany), and one hundredth of the total Aptima eluate was used to run Quantifiler Duo PCR reactions in the Quantstudio Flex System (both AB Biosciences, Concord, MA). The assay measures simultaneously the amplification of a synthetic sequence (for detection of inhibitors), the amplification of Human Ribonuclease P RNA Component H1 gene (RPPH1; present in all human DNA regardless of sex), and the amplification of primers in the SRY region (unique to the male Yc-DNA). PCR reactions containing inhibitors or near LLQ were re-run to confirm results. For adherence testing, presence (above LLQ + standard deviation [SD]) or absence (below LLQ + SD) of RPPH1 was compared. For unprotected sex testing, semen detection was defined by the presence (above LLQ + SD) of Yc-DNA.

Real-time PCR for HIV exposure:

All eluates collected within 4h of coitus were run on COBAS AmpliPrep/COBAS TaqMan HIV-1 Test, v2.0 (Roche Molecular Systems, Basel, Switzerland) platform, which detects both the HIV-1 gag and LTR regions, and has an LLQ of 20 copies/mL. Samples with Yc-DNA collected 4–24h post-coitus were run in Abbott m2000sp and m2000rt instrumentation (Abbott), which detects HIV pol and has an LLQ of 40 copies/mL. Eluates were vortexed three times for 2–3 seconds, then centrifuged at 2000×g for 5 minutes before loading. HIV exposure in eluates was defined by the presence (above LLQ) of HIV nucleic acids.

Behavioral Data Collection:

Each participant was provided a Samsung Galaxy Star S5280 smartphone, with preloaded data, and a survey application programmed using SurveyCTO (http://www.surveycto.com). At enrolment, participants learnt how to use the phone and application, and submitted a practice survey to demonstrate proficiency. Thereafter, participants received daily short messaging service (SMS) reminders, and airtime minutes equivalent to ZAR 5 (~USD0.50), upon completion of each survey. The questions assessed how many vaginal sex acts had taken place from 7am the previous day to 7am that day; whether a condom was used every time; if they had swabbed and time thereof; swab code number; as well as how much time had passed between their last sex event and swab collection.

During their eighth clinic visit, each participant was interviewed face-to-face regarding the past seven days of sexual behavior. Sexual behavior questions included sex with main, casual, and/or new partners; number, age, and concurrent relationships of partners; transactional sex; number of sex acts; condom use during vaginal, oral, and anal sex in the previous sex act; and the frequency and amount of alcohol or drugs used during the past seven days. The Community Advisory Board provided guidance for questionnaire development, in terms of culturally responsive implementation and language.

For behavioral comparisons between week 1 and 12 of trial participation, the age of sex partners, number of sex partners, and frequency of sex per week as continuous variables were compared. The frequency of condom use (always vs. never/sometimes), condom usage in the previous sex act (yes/no), and participation in transactional sex (yes/no) were compared as dichotomous variables.

RESULTS

Study Conduct

Fifty Black African females, median age 22 (IQR 21–24), were enrolled and followed for a total of 4219 person days from 08/2015 to 04/2016. At screening (30 day recall), participants reported a median of three partners (IQR 2–3), a median of 23.5 sex acts (IQR 16–34), with inconsistent condom use 98% of the time (Table 1). 42% reported engaging in transactional sex (n=21); 66% reported one of their partners having concurrent sexual partners (n=33). As detailed in Figure 1, participants returned swabs to the clinic routinely for 12 weeks; four women left the study before completion.

Table 1.

Indicators of HIV risk among HVTN915 study participants prior to enrolment.

Variable	Variable Description	N	Group	Count (%) or Median (IQR)
Age of sexual partners	Age of sexual partners (main, casual, and new)			26 (24.5–29)
Number of sexual partners in the last 30 days	In the last 30 days, how many main, casual, and new partners have you had sex with?			3 (2–3)
Frequency of sex	In the last 30 days, how many times have you had vaginal sex with your main, casual, and new partner?			23.5 (16–34)
Condom usage in last 30 days	In the last 30 days, how often was a condom used during vaginal sex?	50	never	11 (22%)
			sometimes	38 (76%)
			always	1 (2%)
Condom usage in previous sex encounter	Was a condom used the last time you had sex?	50	Yes	5 (10%)
			No	45 (90%)
Transactional sex	In the last 30 days, did your main, casual, new partner give you money, goods, shelter or services for sex?	50	Yes	21 (42%)
			No	29 (58%)
Concurrent sexual relationships	In the last 30 days, did your main, casual or new partner also have sex with other women and/or men?	50	Yes	33 (66%)
			No/Don’t know	17 (34%)
Genital tract infections or dysbiosis at enrolment	N. gonorrhea (Abbot NAAT)	49	positive	4 (8.2%)
	C. trachomatis (Abbot NAAT)	49	positive	12 (24.5%)
	T. vaginalis (OSOM)	10	positive	1 (10%)
	Bacterial Vaginalis (Clue cells in wet mount)	11	positive	0 (0%)

Figure 1. Study conduct.

Shown are HVTN915 screening, study conduct and protocol terminations. In-clinic visits are depicted in the diagram, including the week of the study, the number of swabs (with matched mobile phone entries) returned at each visit, the number of matched swabs tested by adherence (by glycogen or human RPPH DNA) and the number of participants. Reasons for not enrolling and terminating the study are also shown.

Changes in Sexual Behavior during Study Participation

As shown in Table 2, in the comparison of in-clinic questionnaires (seven day recall) collected at Week 1 and 12, participants indicated a reduction in the number of sex partners (p=0.001), number of weekly sex events (p=0.004), and engagement in transactional sex (p=0.008). Participants who dropped out did not have discernable differences in HIV risk at enrolment from participants who concluded the study (data not shown). Both the mobile phone (data not shown) and in-clinic questionnaires (Table 2) suggested that sex with condoms was inconsistent throughout the study, despite risk reduction counselling and availability of condoms. The results indicated that participants reduced some, but not all, of their HIV risk during study participation.

Table 2.

Questionnaire Results at first and last clinic visits.

Question	Description	Answers	Week 0		Week 12		p valuea
			N	Count (%) or Median (IQR)	N	Count (%) or Median (IQR)
Age of sexual partners	Age of sexual partners (main, casual, and new)		40	25.75 (23.92–28.5)	40	26 (25–29)	0.589
Number of sexual partners in the last 7 days	In the last 7 days, how many main, casual, and new partners have you had sex with?		50	2 (1–2)	46	1 (1–2)	<0.001
Frequency of sex	In the last 7 days, how many times have you had vaginal sex with your main, casual, and, new partner?		50	3 (2–6)	46	3 (1–4)	0.004
Condom usage in last 7 days	In the last 7 days, how often was a condom used during vaginal sex?	Always	50	6 (12%)	37	10 (27%)	0.058
		Never/Sometimes		44 (88%)		27 (73%)
Condom usage in previous sex encounter	Was a condom used the last time you had sex?	Yes	50	11 (22%)	46	15 (32.6%)	0.285
		No		39 (78%)		31 (67.4%)
Transactional sex	In the last 7 days, did your main, casual, new partner give you money, goods, shelter or services for sex?	Yes	50	9 (18%)	46	2 (4.3%)	0.008

^aPaired tests were performed including the Wilcoxon signed-rank test for continuous responses, McNemar’s test for paired binary responses, and Bowker’s test of symmetry for categorical responses with more than two levels. Paired tests only account for participants responding at both visits. All p-values are two-sided and no adjustments for multiple comparisons.

Adherence to Self-collected Vaginal Swabbing

Of the 4219 days of follow-up, 3025 (71.7%) mobile phone entries were matched to individual vaginal swabs (Figure 1). Participants returned a median of 66 swabs (IQR 47–77) with matching mobile phone information during follow-up (73.3% of expected swabs). No adverse events associated with swabbing were reported.”

Two assays were performed to assess adherence with swab collections. Vaginal secretions contain elevated glycogen, unlike empty, buccal, seminal, or cutaneous swabs²⁸. In preliminary testing, we also demonstrated elevated glycogen was not observed in rectal swabs (data not shown), so glycogen could identify accuracy in sampling of the vaginal compartment.

To demonstrate vaginal insertion, we assessed 336 swab eluates representing three to seven self-administered samples from each participant (Figure 2A). We stratified self-collected swab selection by study visit, in order to assess adherence throughout the study (Figure 1). Glycogen concentrations were monitored on swab eluates from participants reporting no unprotected sex in the past 4h, because, in preliminary testing, vaginal glycogen was unaffected by menstruation, but provided inconclusive results when sampled immediately after coitus (data not shown).

Figure 2. High Adherence to vaginal swabbing as assessed by glycogen and human RPPH DNA.

A) Adherence of vaginal swabbing using glycogen. For each study participant, three to seven self-collected samples, stratified by study visit, were selected to measure adherence. As a positive control, one clinic collected swab for each participant was also assayed. Glycogen concentrations were measured by EnzyChrom Glycogen Assay and quantification of the enzymatic reaction was performed by fluorescence. Samples under the LLQ of 3.15μg/mL were considered to not have swabbed vaginal compartments. B) Glycogen concentrations in Aptima swab eluates from self-collected and clinic-collected swabs. Dashed line is the LLQ, and all swabs below the LLQ were plotted at this location. Bars depict median glycogen concentration and interquartile ranges (IQR). Differences among clinic and self-collected swabs were examined in Mann-Whitney test (p=0.797). C) Detection of Human and Yc-DNA in self-collected vaginal swabs stratified by sexual behavior. Samples of abstinence, condom/abstinence for 7 consecutive days and unprotected sex (4–24h post coitus) were selected at random from all collected swabs and used to assess adherence (n=80). Samples of condom/abstinence for 3 days were selected at random after exclusion of one participant who had no glycogen in the previously tested swabs. All were assayed for human and Yc-DNA using Quantifiler Duo PCR. The presence of human RPPH DNA was used as a measure of adherence, whereas Yc-DNA indicated male DNA contamination indicative of unprotected sex. Positive PCRs were defined to be above the LLQ of 23pg (7 DNA copies) per 10ul of swab eluate tested.

To compare with swabs collected at the clinic, fifty clinician-collected swabs were used as positive controls; and all contained glycogen (Figure 2A). In comparison, 315 out of 336 self-collected swabs had glycogen (93.8%), indicating high adherence to collecting the requested samples and high accuracy in swabbing the vaginal vault. Glycogen levels in self-collected swabs (median 432μg/mL, IQR 142.1–944.4) were comparable to clinician-administered swabs (median 357.4μg/mL, IQR 185.9–887.9), indicating comparable sample collection (Figure 2B; Mann-Whitney p=0.797). The 21 (6.2%) non-adherent swabs were not randomly distributed; one participant contributed 5 (23.8%) empty swabs, and was excluded from further Yc-DNA quantitation and HIV exposure assays.

To confirm the adherence rate, a second subset of 80 swab eluates was used to assess human RPPH DNA (Figure 2C). One swab contained inhibitors of the PCR reaction, and 73 (92.4%) contained measurable levels of RPPH DNA above the LLQ of 2.3ng/mL of swab eluate, similar to adherence levels estimated by glycogen measurements (Figure 2A).

Evaluating the Accuracy of Mobile Phone Surveys for Sexual Behavior

To assess the validity of sexual behavior reporting in the mobile phone questionnaire, we examined the presence of Yc-DNA, with a half-life of 18h-4 days after inoculation/unprotected sex.^29–32 Using the behavior reported in the mobile phone questionnaire, we compared 40 random samples reporting unprotected sex and swabbing 4–24 hours post-coitus, versus swab eluates collected after 1) seven consecutive days of reporting abstinence (n=20), 2) protected sex and abstinence for seven consecutive days (n=20), or 3) protected sex and abstinence for 3 consecutive days of (n=160; Figure 2C). Although we detected RPPH DNA in 187 of the abstinent and protected sex samples, Yc-DNA was present in only 10 swabs (5.3%). In contrast, 20 swabs linked to unprotected sex reports (51.3%) contained detectable Yc-DNA out of 39 samples with detectable RPPH DNA. The mobile phone application supported accurate reporting of protected/abstinent behaviors (negative predictive value of 90%) and unprotected sex (positive predictive value of 67%).

To evaluate the mobile phone data and compare it to clinic behavioral data, we examined Yc-DNA detection in 77 swabs where the clinic report did not match mobile phone data (Figure 3A; 2.5% of collected swabs with paired mobile phone app data). Three of these swabs contained no detectable RPPH DNA and were excluded, as they were likely empty of vaginal secretions. Because there are different reporting intervals covered by in-clinic (seven days) and mobile phone (1 day) questionnaires, we aggregated the mobile phone data and summarized it in person-weeks. A comparison indicated that mobile phone reporting more accurately captured unprotected sex events as indicated by the detection of Yc-DNA in the vaginal compartment (p=0.003).

Figure 3. Comparison of Yc-DNA detection among the mobile phone and In-Clinic Behavioral Surveys.

A) Unprotected sex reporting via mobile phone survey is more accurate than in-clinic behavioral reports. Yc-DNA was measured using Quantifiler Duo PCR in 77 swabs where the clinic report did not match the mobile data. Thirty-six samples corresponded to 16 participants from 21 clinic visits reporting condom usage/abstinence in the past seven days at the clinic visit, but indicating at least one unprotected sex event in the mobile phone survey during the same interval. Another 38 samples corresponded to 13 participants from 16 clinic visits reporting condom usage/abstinence in the mobile phone questionnaire, but indicating at least one unprotected sex event in the past seven days at the clinic visit. Three of these swabs contained no detectable human RPPH DNA and were excluded from the table. Swab numbers in italics were aggregated in person-weeks (regular font) to facilitate statistical comparisons among similar 7-day study periods using a two-sided Fisher’s exact test (p=0.0031). B) Errors in reporting of in-clinic sex behaviors are associated with longer recall periods. DNA from swabs collected from participants who reported abstinence/condom use on the in-clinic survey, but unprotected sex on the mobile phone survey were graphed with respect to the time between swab collection (and mobile phone reporting) and the in-clinic survey date. Yc-DNA positive (black bars) and Yc-DNA negative (white bars) swabs were measured using Quantifiler Duo PCR, and graphed to indicate that unprotected sex events confirmed by male DNA occur later in the recall period. Differences assessed using a one-sided Wilcoxon Test p=0.0384.

As an additional exploratory analysis, we examined whether recall issues could be causing less accurate responses to the in-clinic survey, as seven days recall of sexual activity were needed (Figure 3B). Within the samples that reported abstinence/condom use on the in-clinic survey and unprotected sex on the mobile survey, Yc-DNA was more frequently detected in situations when the reported unprotected sex event had taken place earlier during the seven day recall period (p=0.038).

Real-Time PCR for HIV Exposure by in Vaginal Swabs

No HVTN915 participants seroconverted during the study period, so HIV RNA detection in vaginal samples only indicates HIV exposure.

As the mobile phone survey demonstrated some accuracy in reporting condom usage (Figure 2C and 3), all vaginal swab eluates from participants reporting swab collections within 0–4h of unprotected sex in the mobile phone survey were tested for HIV virions via real-time-PCR. Only one eluate out of 69 collected 0–1h post-coitus had 104 HIV copies/mL (see Figure, Supplemental Digital Content 1). None of the 247 samples collected 1–4h post-coitus without a condom had detectable HIV.

Because HIV nucleic acid stability in the vagina is unknown, we tested swabs collected 24h-72h after the positive result in the HIV exposure test (Supplemental Digital Content 1). Eluate from the swab collected 24h later had 64 HIV copies/mL and represented a collection that took place 1 hour after unprotected sex, but whose swab number did not perfectly match the mobile phone survey, and thus, had been excluded from previous analysis. The participant reported abstinence 48h post-HIV exposure, and unprotected sex with a swab collected 8–12h post-coitus 72h post HIV exposure; both samples had undetectable HIV.

To improve the odds of detecting HIV in samples collected 4–24h post-coitus without a condom, we examined Yc-DNA to select samples with evidence of seminal exposure (Supplemental Digital Content 1). Of 451 samples tested, 440 (97.5%) contained human RPPH DNA; 185 (41.0%; n=28 participants) also contained Yc-DNA. All samples with Yc-DNA had undetectable HIV, despite the presence of semen.

DISCUSSION

This study evaluated the feasibility of self-administered vaginal swabs for detection of HIV-1 transferred through a penetrative vaginal sex act. The detection of HIV in two swabs from an uninfected participant demonstrates it is possible to detect an HIV exposure using self-collected vaginal swabs.

HIV incidence among young women in Johannesburg ranges from 2.5% to 7.4%^26,33,34 and the probabilities of HIV-1 transmission per coital act range from 1/100 to 1/2000^35,36. Thus, in a cohort of 50 women followed for 3 months, we would have expected 0.3–0.9 seroconversions, and 31 to 1,850 potential exposures to HIV. As reported by the mobile phone survey, 39.5% would be prevented by condom use. In a study of HIV seropositive men from Kenya not taking antiretrovirals, 45.5% of semen ejaculates contained detectable HIV.³⁷ Thus, 9 to 509 exposures with detectable HIV were expected in our cohort, and 70% would have swabs with mobile phone data for analysis. Therefore, we expected to detect 6 to 356 exposures, and our finding of only two HIV exposures is lower than expected.

One possibility to explain the detection of only two HIV exposures in our cohort is that the HIV incidence estimates and per-sex-act transmission probabilities used in our calculations are overestimating the 2016 HIV exposure rates of women in Soweto. The widespread introduction of HIV testing and improved anti-retroviral therapy (ART) coverage could further decrease new transmission events in this population. Indeed, several models have predicted decreases in incidence ranging from 40%-54% over the 2002–2012 decade,^38,39 consistent with the 2012 survey estimates for incidence in South African women ages 15–24²⁶.

Another explanation for the lower detection of HIV exposures in our cohort is the reduction in HIV risk behavior during the study. In the in-clinic questionnaire, participants reported three risk behaviors that reduced during study participation: number of sex partners, number of weekly sex acts, and engagement in transactional sex. It is possible that risk-reduction counselling, daily swabbing, and sex reporting via mobile phone survey contributed to increased awareness of risk during sexual behavior. In fact, the HIV-positive swabs were identified at days eight and nine post-enrolment, when participants had yet to report altered behavior.

It is also possible that some HIV exposures were missed in the swabs collected 4–24h after unprotected sex, as only those with Yc-DNA evidence were tested for HIV by real-time PCR. Only 41% of these swabs tested positive for semen, despite the unprotected sex report. Other studies have used DNA from the entire vaginal lavage to precisely detect semen days after coitus^29–32. In our study, only 1/100^th of the entire swab eluate was examined for Yc-DNA, to save sufficient sample for simultaneous HIV exposure testing. Nevertheless, we expected HIV exposure testing would be most sensitive in the swabs with higher Yc-DNA, yet none of the samples with evidence of semen had detectable virus 4–24h after unprotected sex.

Lastly, the number of hours post-coitus may also define the ability to detect HIV. Seminal plasma is diluted post-coitus in cervicovaginal fluid, and decreases 90% within 4h of collection as estimated by PSA.^17,30 Ejaculate fluid is further diluted by cervicovaginal secretions, and reaches undetectable levels by 24h post-coitus.^17,30 Thus, the identification of HIV in two samples collected less than 1 hour after coitus, but none 1–24h post-coitus, supports the idea that cell-free HIV nucleic acids can become diluted early after sex, below the LLQ of the HIV RT-PCR assays. HIV is also susceptible to nucleases in the vaginal compartment,^40,41 potentially limiting the applicability of this assay to strict temporal requirements. The temporal restrictions for HIV detection in vaginal secretions may limit the current assay use in screening for women at risk of HIV, or for identifying transmission events retroactively, such as testing for vaginal HIV in quarterly swabs collected prior to seroconversion.⁴²

The average semen ejaculate volume is 3.2mL,⁴³ and HIV-seropositive semen in the absence of antiretrovirals contains a median of 371–3400 HIV copies/mL.^37,44 If vaginal swabbing collects 125μL of secretions and dilutes the semen in 0.4mL of secretions⁴⁵, the average HIV-positive sample would contain approximately 41–377 viral copies, if collected soon after coitus. As the secretions in this study are diluted in Aptima collection media, 14–127 copies/mL would be the typical assay result. We detected 64 and 104 viral copies/mL in our samples, confirming that detection of HIV-1 virus in vaginal secretions is above the LLQ of the assay (20–40 copies/mL). However, it is possible that some exposures are below the LLQ, especially if collected several hours post-coitus, or after vaginal washing, when further degradation/dilution takes place. Future studies of HIV exposure may benefit from obtaining information about vaginal hygiene practices, partner HIV status, and semen viral load, to define the sensitivity of the assay.

Importantly, our data suggests that the HIV-1 exposure rate detected does not appear to be the product of consistent condom use, inaccurate protected sex reporting in mobile phone surveys, or women not sampling their vaginal compartments. Inconsistent condom use was reported throughout the trial and was confirmed by Yc-DNA.

Mobile phone self-reporting proved more accurate (with only 6% of human RPPH positive samples having Yc-DNA and reporting abstinence/condom use) than in studies correlating biomarkers to face-to-face interviews (11% to 39% false positives)^15,46,47 and CASI questionnaires (12% to 34%).^4–6 Our analysis of inconsistent clinic versus mobile phone reports also suggests unprotected sex reporting via mobile phone is more accurate due to a shortened recall interval.

Study participants also accurately conducted vaginal self-swabbing, as 93.8% of swabs contained glycogen and 92.4% contained human RPPH DNA. Of the mobile phone data collected, 3025 (71.7%) entries matched a vaginal swab.

Together, these results suggest that self-collected sampling of the vaginal compartment is feasible and, in association with mobile phone surveys, may improve our understanding of sexual health in women at risk of HIV infection. To improve the detection of HIV exposures, further work is needed to increase the sensitivity of HIV PCR assays, and to understand the turnover of viral nucleic acids in the vaginal compartment.