In southern and eastern Africa, the incidence of HIV in young women (aged <25 years) remains high, despite extensive prevention campaigns and scale-up of antiretroviral therapy (ART).¹ The Joint UN Programme on HIV/AIDS (UNAIDS) has identified young women in these regions as the highest priority group for HIV prevention to help achieve the global goal of ending AIDS as a public health threat by 2030.¹

A substantial imbalance in HIV prevalence exists between men and women throughout southern and eastern Africa, and women acquire HIV at a much earlier age than their male peers (appendix p 4). Although the reasons for these differences are not fully understood, the sexual network structure and dynamics might have prevalence² suggest that older men are probably the main source of infection for young women, this suggestion has been disputed in two recent studies.^3,4 However, findings from these studies were based on reported ages of male partners presumed to be the source of, or at least related to, HIV infections in women. The lack of clarity and definitive data on whether age-disparate sexual partnering is important for HIV transmission has hampered HIV prevention in southern and eastern Africa, which hosts more than 50% of the global HIV burden.¹ Current prevention programmes in South Africa and several other countries have steered clear of including this sociobehavioural factor in their prevention messages and education programmes.

Current HIV prevention programmes have had limited impact on reducing HIV incidence in young women. Prevention strategies targeting the key underlying drivers of HIV transmission could provide a more focused approach to HIV prevention in this high-risk group. Phylogenetic analysis can be used to identify drivers of HIV transmission, as seen in the ATHENA cohort study of men who have sex with men in the Netherlands.⁵ Phylodynamic models can assess how transmission dynamics change over time, as used in a study in men who have sex with men in Detroit, USA.⁶ However, longitudinal cohorts with detailed clinical and behavioural data are rare in Africa.⁷

The purpose of our large-scale community-wide phylogenetic study was to examine the underlying HIV transmission dynamics and the source and consequences of high rates of HIV infection in young women in a district with a high burden of HIV in KwaZulu-Natal, South Africa. In this community, the reported incidence of HIV in women younger than 25 years is 10.2 per 100 person-years (36 seroconversions during 353 person-years of follow-up in the placebo group of a microbicide trial).⁸

Enrolment for the study began on June 11, 2014, and ended on June 22, 2015. 14 618 households were eligible; of these, 2144 (14.7%) heads of households refused to take part in the study, 753 (5.2%) households had no one at home after three visits, and 432 (3.0%) households had no one home after the first visit, but were never revisited because the protocol-specified sample size was met before they needed to be. 11 289 (77.2%) households were enrolled, 398 of whom did not have an eligible person living in the household. 10 891 potentially eligible individuals were approached for participation in the study; 577 (5.3%) refused to take part (246 men and 331 women), 488 (4.5%) were still in the process of completing enrolment when the study ended, and 14 (0.1%) had no HIV results available. 9812 individuals were enrolled and tested for HIV, of whom 3969 (40.5%) were HIV positive (weighted 36.3%, 95% CI 34.8–37.8; table 1). 2955 (47.2%) of 6265 women (44.1%, 42.3–45.9) had HIV compared with 1014 (28.6%) of 3547 men (28.0%, 25.9–30.1; p<0.0001; table 2). HIV prevalence was 62.0% (1758 of 2835; weighted 59.8%) in women aged 25–40 years and was highest in those aged 35–39 years (517 [68.0%] of 760; weighted 66.4%; figure 1). In individuals younger than 25 years, HIV prevalence was about three times higher in women than in men (p<0.0001; table 2).

1592 (40.1%; weighted 42.3%) of the 3969 HIV-positive individuals reported that they were on ART (table 1), coverage that was similar to national treatment coverage in ART-eligible individuals at the time of the study.¹ 1412 (88.8%; weighted 88.3%) of 1590 participants who reported using ART and whose viral loads were tested had HIV RNA viral loads of less than 1000 copies per mL, whereas 697 (29.5%; weighted 27.1%) of 2366 individuals who reported that they were not on ART had viral loads less than 1000 copies per mL. 13 HIV-positive individuals had samples insuffcient for viral load testing. 2292 (57.9%; weighted 58.0%) of the 3956 HIV-positive individuals whose viral loads were tested had detectable viral loads of at least 20 copies per mL; 1847 (46.5%) had more than 1000 copies per mL and were selected for sequencing (table 1).

Blood samples were available from 1688 individuals whose viruses were selected for sequencing. The viruses of 95 of these individuals could not be sequenced and four sequences did not pass the quality control step, resulting in 1589 pol sequences of 1200 bp for 86.0% of the individuals with viral loads of more than 1000 copies per mL. Applying the criteria of more than 90% SH-aLRT clade support and genetic distance of less than or equal to 4.5%, 469 (29.5%) sequences grouped into 202 transmission clusters (see appendix p 3 for cluster size distribution). 151 (32.2%) of the 469 sequences were from men and 318 (67.8%) were from women, reflecting the sex distribution in the whole sequenced dataset (1102 women and 487 men).

90 of the 202 clusters had at least one woman and one man (figure 2). 123 women were linked to 103 men in the 90 clusters. Within these clusters, 43 (35.0%) women were younger than 25 years, 56 (45.5%) were aged 25–40 years, and 24 (19.5%) were older than 40 years.

49 women (39.8%) were aware of their serostatus compared with 27 men (26.2%; p=0.035; table 1). Because only individuals with viral loads of more than 1000 copies per mL are included in this analysis, the small numbers of people in transmission clusters who reported being on ART were expected (table 1). Median viral load was significantly higher in men than in women (31 000 copies per mL in men vs 16 629 copies per mL in women; p=0.002); 41 (33.3%) women and 37 (35.9%) men had viral loads of more than 50 000 copies per mL.

In the 90 clusters, there were 188 possible pairings between the 123 women and 103 men. A comparison of all possible pairings across age and sex categories (figure 3) identified two more frequent sets of linkages: between women aged 25–40 years and men aged 25–40 years (58 [30.9%] pairings) and between women younger than 25 years and men aged 25–40 years (37 [19.7%] pairings]), which constituted just over 50% of the linkages. In the 60 possible pairings between the 43 women younger than 25 years who were linked to 41 probable male partners, the mean age difference was 8.7 years (95% CI 6.8–10.6; p<0.0001). In these 60 possible pairings, 18 (30.0%) of the probable male partners were younger than 25 years, 37 (61.7%, 95% CI 49.7–74.3) were aged 25–40 years, and five (8.3%) were aged 41–49 years. Of the 41 probable male partners linked to a woman younger than 25 years, 16 (39.0%) were also linked to a 25–40-year-old woman (figure 4); the mean age of these male partners was 29.6 years (SD 7.1) and they were linked to both a younger woman (mean age 22.1 years) and an older woman (mean age 32.6 years). In the 92 possible pairings between the 56 women aged 25–40 years who were linked to 36 probable male partners, the mean age difference was 1.1 years (95% CI–0.6 to 2.8; p=0.111). Of 79 men (mean age 31.5 years) linked to women younger than 40 years, 62 (78.5%) were unaware of their HIV-positive status, 76 (96.2%) were not on ART, and 29 (36.7%) had viral loads of more than 50 000 copies per mL.

The mean age of male partners of HIV-positive women aged 20 years or younger was 30.5 years; the mean age difference was 11.5 years (SD 7.8; p<0.0001). The age gap decreased to a mean difference of 7.0 years (SD 6.7) in HIV-positive women aged 21–25 years and to a mean difference of 1.5 years (SD 9.1) in women aged 26–30 years (table 3).

The age gap between HIV-negative women and their self-reported male partners (not linked through clustering) was smaller (plus or minus 3.8 years) and more consistent across the age groups. Educational levels, age of sexual debut, number of lifetime and current sexual partners, sexual frequency, and condom use were similar in the various subpopulations, irrespective of HIV status and linkage in clusters (table 1). Overall, the sample included 4% of the census-defined population. Sequencing coverage of HIV-positive individuals in the sample was higher than 25% in all age categories, being highest in men (402 [52.4%; weighted 51.3%] of 767) and women (926 [39.8%; weighted 36.5%] of 2325) aged 40 years and younger (table 2). Since sequencing coverage was not skewed and there is no evidence of super-spreaders driving HIV transmission in the opposite direction to the prevalence gradient, the dominant directionality of transmission was determined to occur from the highest prevalence of 59.8% (women aged 25–40 years) to 40.3% (probable male partners aged 25–40 years) to 22.3% (women aged <25 years; table 2, figures 3 and 4).

Phylogenetic linkage, together with prevalence gradients, suggest that men aged 25–40 years were the primary source of the high rates of HIV acquisition in adolescent girls and young women (15–25 years). Many of these men had acquired HIV infection from women aged 25–40 years, which is the group with the highest prevalence of HIV. Over time, when the current group of adolescent girls and young women reach their 30s, we expect that they will then constitute the next group of women with high HIV prevalence, thereby perpetuating the cycle of HIV transmission to men in their 30s who will infect the next cohort of adolescent girls and young women (figure 4). The risks of HIV transmission within this cycle are enhanced at a community level by two key factors. First, two-fifths of HIV-positive individuals in this community were not aware of their HIV-positive status, and second, only two-thirds of individuals who knew they had HIV had initiated ART, resulting in suboptimal viral suppression. In particular, many of the men linked to women in the dual age groups (ie, <25 years and 25–40 years) were likely to have recently acquired HIV, were unaware of their HIV status, and had viral loads associated with high rates of transmission.¹⁹

Our results differ from those reported elsewhere; age disparity was not a risk factor for HIV acquisition in either a cohort study in which women reported the age of their most recent sexual partner on a yearly basis³ or in a clinical trial in which the age of a single primary partner was reported by women at baseline for infections occurring up to a year later.⁴ The major shortcoming in both studies is that the men whose ages were analysed might not have been the source of the women’s HIV exposure or infection. Investigators of both studies were unable to determine accurately which men transmitted HIV to the women enrolled.

Rapidly advancing bioinformatics methods such as phylogenetics²⁰ and phylodynamics^12,21 overcome this limitation and provide information about clearly linked individuals within a transmission cluster. A strength of our approach is that the phylogenetic clusters minimise the assumptions required to identify the chains of transmission. However, an important assumption remains; that viral transmission occurred among linked individuals within a cluster, acknowledging that not all partners in a chain of transmission may have been included in a cluster.

Our study also has several weaknesses. First, although this was a large cross-sectional study, our sample of 9812 individuals consisted of only 4% of the population, limiting the extent to which we were able to uncover linkages. Since the 4% sample was representative, we have extrapolated its findings to the community, in recognition of the limitations of this sample size.

Second, we did not have detailed longitudinal, clinical, and behavioural data to assist in establishing directionality. We therefore inferred directionality on the basis of the prevalence gradients, which is a crude approach and might not apply uniformly across the 188 possible pairings in the 90 male–female clusters. Although it is a reasonable assumption that most transmission will probably occur from high to low prevalence, we were not able to quantify the exact proportion of transmissions occurring against the gradient. However, there was no evidence of super-spreaders to drive transmission against the prevalence gradient.

Third, our study sampled an urban and a neighbouring rural community from one province in one country and is therefore not representative of all southern and eastern African countries. However, phylogenetic studies require a large sample size from a single community to maximise the opportunity to identify transmission clusters. Although several similar large-scale studies will be required in other settings to assess the generalisability of our results, the differences in HIV prevalence in young men and women in South Africa are also present in many southern and eastern African countries (appendix p 4). Findings from a phylogenetic study in Rakai, Uganda, showed that a substantial proportion of HIV infections were transmitted within married couples.²² A bioinformatics study in Mochudi, Botswana, identified acute subepidemics introduced from outside the region between 1997 and 2007.²³ However, both studies did not specifically address the issue of the source of infection in young women.

Fourth, our analysis excluded individuals on successful therapy and is therefore skewed to more recent infections because only individuals with viral loads greater than 1000 copies per mL could be included. Alternatively, this limitation can be viewed as a strength, because the data describe the current and recent epidemic with limited contamination from long- established infections and people on treatment who pose minimal risk of spreading HIV. The high rate of sequencing failure with current sequencing technology in individuals with low viral loads will make these kinds of studies much more difficult to do as treatment coverage increases. In this community with 42% ART coverage, our large study of almost 10 000 individuals and more than 1500 sequences identified only 90 male–female clusters. Although this number is fewer than expected, it reflects what is possible from large studies at this stage of the HIV epidemic when ART is being rapidly rolled out and clusters becoming increasingly diffcult to identify. Finally, a potential bias could be the non-response rate. In this study, non-response at household and individual levels were within the range factored into the design.

Age disparity in sexual relationships has long been postulated as an important contributor to HIV transmission in Africa.²⁴ Empirically, age disparity has been difficult to quantify but an analysis of a 1998–2000 community survey in Zimbabwe reported a small but significant increase in HIV risk in men caused by age-disparate relationships.²⁵ Our phylogenetic analysis, which was made possible by the rapidly decreasing cost of gene sequencing, provides an empiric demonstration of the key role of the high HIV incidence in adolescent girls and young women transmitted by men (on average 8 years older) as a key component of HIV transmission in South Africa.

In several countries, including South Africa, men have been more challenging than women to reach with HIV testing and ART because men are often highly mobile while seeking job opportunities.^26,27 However, other options to reduce HIV transmission are available. Prevention efforts need to include clear messages about age-disparate relationships between young women and men in their 30s, as part of an overall longer-term effort to change community norms in this regard. Over and above existing prevention and treatment programmes, three targeted interventions could help to break the cycle of HIV transmission in this setting: rapid scale-up of universal testing and treatment of men and women aged 25–40 years to reduce the risk of these high HIV prevalence groups transmitting HIV; scale-up of circumcision in men younger than 25 years to decrease their risk of acquiring HIV in their highest risk years; and provision of pre-exposure prophylaxis for young women to empower them to reduce the high risk of acquiring HIV. This combination prevention approach targeting three points in the transmission cycle could help to reduce HIV incidence in young women, an essential initial step to set Africa on the path to the UNAIDS 2030 goal of ending AIDS.