Circumcision status at HIV infection is not associated with plasma viral load in men: analysis of specimens from a randomized controlled trial

Male circumcision (MC) has been shown in randomized controlled trials (RCTs) to confer approximately 60% protection to men from acquiring HIV infection via heterosexual sex [1–3]. It has become a key component of HIV prevention efforts in sub-Saharan Africa, with nearly 15 million voluntary medical male circumcisions (VMMC) performed since 2008 [4]. Widely-accepted possible mechanisms for the protective effect include: the removal of the inner foreskin’s high concentration of Langerhans cells, through which HIV can access the immune system; the foreskin’s vulnerability to microabrasions; the elimination of the anaerobic sub-preputial environment that supports a pro-inflammatory microbiome; and the greater incidence of genital ulcer disease (GUD) among uncircumcised men [5].

Nevertheless, many circumcised men will ultimately become HIV-infected. Determinants of plasma viral load (VL) set point among infected individuals are not entirely understood, but in addition to well-established associations like host immunogenetics [6] and viral genotype [7], it is associated with VL in “source” partners [8–10]. Although this relationship is sometimes attributed to viral genotype also, another suggested mechanism is that high source partner VL results in a higher infecting inoculum, which then results in higher VL in the newly-infected partner [11, 12]. (The majority of HIV infections resulting from heterosexual sex are believed to result from a single virion [9, 13], so the effect of inoculum size on VL is not hypothesized to result directly from a higher initial virion population. Instead, a larger inoculum may lead to increased host T-cell activation, which in turn enhances replication of the ultimately ‘successful’ virion [10]). Supporting evidence for inoculum size as an independent determinant of VL includes the association between source partner genital tract factors which would affect magnitude of viral shedding, such as bacterial vaginosis and genital herpes suppression, and VLs in seroconverting partners [12].

If source inoculum affects VL, the same mechanisms that confer partial protection against infection could operate in circumcised men who do become infected to decrease the proportion of virions breaching their skin, lowering inoculum and thus VL. Fig. 1 illustrates this proposed causal chain. If present, such an effect could confer substantial population-level benefit, as VL is a major determinant of both progression to clinical disease and infectiousness [14].

Evidence for the plausibility of this result comes from one study that found MC in genotype-confirmed male source partners was associated with lower HIV VLs among the female partners they infected [10, 15] (adjusted mean difference = −.63 log₁₀ copies/mL, p = .03). The authors hypothesized the above mechanism in reverse: lower innocula transferred to female partners due to the removal of the male’s infectious foreskin Langerhans cells via MC. This study did not find evidence of lower VL set points associated with MC among men infected by female source partners, but this outcome was limited by small sample size (N = 55) and the fact that participants were not randomized with respect to circumcision status.

We used stored plasma samples from the original RCT of MC for HIV prevention conducted in Kisumu, Kenya [2], to investigate whether MC status of male participants who became HIV-infected was associated with their VL.

A strong linear correlation was found between VL and sample age (Fig. 2; r = −.48, p < .0001, R² = .23); a quadratic term was tested for significance to assess for a nonlinear association, but was not significant and was therefore dropped. No significant correlation was found between VL and duration of infection at time of blood draw (Fig. 3; r = .07, p = 0.40).

No significant difference in log VL was found between circumcised and uncircumcised men in the as-treated (means 4.00 and 4.03 log copies/mL respectively, p = .88) or per-protocol (means 4.06 and 3.99 log copies/mL, p = .76) analyses. This remained true after adjusting for STI diagnosis within 3.5 months of seroconversion and age of sample, for both the as-treated (means 3.96 and 4.09 log copies/mL, p = .47) and the per-protocol (means 3.99 and 4.04 log copies/mL, p = .80) analysis.

Our results do not support an effect of circumcision status larger than 1.26 log₁₀ copies/mL at the time of seroconversion on VL after 6 months. This does not rule out a clinically significant effect: a 1 log difference in VL was found in Uganda to be associated with an adjusted hazard ratio of 3.09 for death and 2.75 for progression to AIDS [17] and the 1 log difference between a VL of 10,000 and 100,000 copies/mL has been modeled to be associated with approximately a 5-year difference in asymptomatic survival [18]. However, this finding is consistent with the observational results reported by Lingappa et al. in participants from southern and eastern African countries, and represents gold-standard data due to randomization which is unlikely to become available from future studies. It also helps establish the assumptions needed for sample size planning for any observational studies further examining the issue. Given the mechanistic biological nature of the proposed effect, results are also likely to be valid across populations.

If the lack of observed effect reflects biological fact, the explanation may lie in the set of portals of entry available to HIV virions on the male genitalia. In addition to the foreskin, the glans, urethral mucosa and possibly fossa navicularis are all portals across which translocation of cell-associated HIV-1 has been demonstrated in vivo [19]. Given the low per-act probability of female-to-male transmission through any of these portals (4 per 10,000 infectious exposures in a recent systematic review) [20], it seems to follow that the probability of transmission through any single portal is even lower, and thus that in any one act leading to infection, the most likely scenario is that only one portal is involved. Thus circumcision would remove one portal, decreasing infection risk, but would not affect the size of an inoculum passing through a different portal, thus not affecting VL if infectious becomes established. Others have found higher per-act transmission probabilities such as 0.38% in developing countries in the absence of commercial sex work or genital ulceration, [21] but also at this higher magnitude, a single portal seems likely to be involved in the majority of transmissions.

Conversely, the previously-observed lower VLs in women with circumcised male source partners are more likely to be related to total male viral shedding, which would be expected to come from multiple “portals of exit” during sex, including semen and penile mucosa. Shedding has been shown via coronal lavage to be reduced in HIV-infected men after healing from MC, compared to pre-MC levels [22].

It is also possible that infecting inoculum does not actually impact VL, though this would not explain Lingappa et al.’s [12] observed association between male infecting partner circumcision and lower female partner VL in their baseline data.

Finally, it is possible that there is a population-level effect of circumcision status at time of infection on men’s VL, but that it is mediated by the known protective associations of MC with risk for genital ulcer disease (GUD) [23], thus decreasing VL indirectly through the same mechanisms postulated for a direct effect. Lingappa et al. [12] did not find an association between HSV-2 serology and VL set point, but thoroughly testing this possibility would require a much larger sample of circumcised and uncircumcised seroconverters with sufficient prevalence of GUD to estimate a true effect.

The major potential confounder is the effect of sample age on VL. Although controlling for this did not impact results, the effect of storage time on VL can have an unpredictable [24] component, which could have caused residual confounding. However, if substantial, this component should have eroded all observed associations; the observed correlations between VL and sample age, and between VLs from the same participant, should not have been seen.

Another potential confounder is ART initiation, which was not captured in the study data. However, between 2005 and 2011, Kenyan National HIV Treatment guidelines set a CD4 threshold for ART of <=200/mm³ in the absence of clinical indications, and the maximum elapsed time of 5.1 years between calculated infection date and blood draw in our sample makes it unlikely that participants would have progressed to meet this criterion and start treatment. If any did, elimination of samples with undetectable viral load would also have been likely to exclude them. In other respects, although circumcised participants were somewhat more likely to report risky behavior, the difference was not significant, and VLs would not necessarily be higher in infections contracted through high-risk sex.

Finally, for participants with only one sample, stability of VL over time cannot be demonstrated; therefore, VL cannot necessarily be considered equivalent to VL set point.

Though circumcision in HIV-positive men has an observed association with HIV VL in female partners infected by them, this study provides no evidence that circumcision status at HIV seroconversion affects VL after 6 months in men.