Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Molecular Epidemiology of HIV-1 Subtype B Reveals Heterogeneous Transmission Risk: Implications for Intervention and Control

Molecular Epidemiology of HIV-1 Subtype B Reveals Heterogeneous Transmission Risk: Implications... Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 Molecular Epidemiology of HIV-1 Subtype B Reveals Heterogeneous Transmission Risk: Implications for Intervention and Control Erik M Volz, Stephane Le Vu, Oliver Ratmann, Anna Tostevin, David Dunn, Chloe Orkin, Siobhan O’Shea, Valerie Delpech, Alison Brown, Noel Gill, Christophe Fraser, UK HIV Drug Resistance Database Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 The Journal of Infectious Diseases MAJOR ARTICLE Molecular Epidemiology of HIV-1 Subtype B Reveals Heterogeneous Transmission Risk: Implications for Intervention and Control 1 1 1 2 2 3 4 5 5 Erik M. Volz, Stephane Le Vu, Oliver Ratmann, Anna Tostevin, David Dunn, Chloe Orkin, Siobhan O’Shea, Valerie Delpech, Alison Brown, 5 6 Noel Gill, and Christophe Fraser ; on behalf of the UK HIV Drug Resistance Database Department of Infectious Disease Epidemiology and the National Institute for Health Research Health Protection Research Unit on Modeling Methodology, Imperial College 2 3 4 London; Institute for Global Health, University College London; Barts Health NHS Trust, London; Infection Sciences, Viapath Analytics, Guy’s and St Thomas’ NHS Foundation 5 6 Trust, London; Public Health England, London; and Li Ka Shing Centre for Health Information and Discovery, Oxford University, United Kingdom (See the Editorial commentary by Baeten, on pages 1509–11.) Background. e Th impact of HIV pre-exposure prophylaxis (PrEP) depends on infections averted by protecting vulnerable indi - May viduals as well as infections averted by preventing transmission by those who would have been infected if not receiving PrEP. Analysis of HIV phylogenies reveals risk factors for transmission, which we examine as potential criteria for allocating PrEP. Methods. We analyzed 6912 HIV-1 partial pol sequences from men who have sex with men (MSM) in the United Kingdom combined with global reference sequences and patient-level metadata. Population genetic models were developed that adjust for stage of infection, global migration of HIV lineages, and changing incidence of infection through time. Models were extended to simulate the effects of providing susceptible MSM with PrEP. Results. We found that young age <25 years confers higher risk of HIV transmission (relative risk = 2.52 [95% confidence inter - val, 2.32–2.73]) and that young MSM are more likely to transmit to one another than expected by chance. Simulated interventions indicate that 4-fold more infections can be averted over 5 years by focusing PrEP on young MSM. Conclusions. Concentrating PrEP doses on young individuals can avert more infections than random allocation. Keywords. HIV; men who have sex with men; phylodynamics; pre-exposure prophylaxis. The effectiveness of public health interventions (PHIs) to enhancing transmission risk and increasing epidemic spread com bat human immunodeficiency virus (HIV), such as pre- [7–9]. exposure prophylaxis (PrEP) with antiretroviral medications While the factors that shape transmission risk are under- (ARVs) depends on unknown variability of transmission risk in stood qualitatively, it is challenging to obtain robust quanti- the infected population. The impact of PHIs can be enhanced if tative estimates of transmission probabilities or transmission OA-CC-BY the intervention can be focused on patients with higher trans- risk. HIV genetic sequence data from routine drug resistance mission risk, due to, for example, different risk behaviors or testing is one of the few sources of widely available observa- epidemiological settings [1]. HIV transmission risk is highly tional data that are directly informative about HIV transmis- variable over time, over the course of infection, between risk sion patterns and transmission risk [10, 11]. Donor-recipient groups, and geographically [2]. Biological, behavioral, and envi- transmission pairs harbor virus that is genetically closely ronmental factors shape individual HIV transmission risk in related compared to the population as a whole [12, 13]. At complex ways. Transmission probabilities per coital act depend longer evolutionary time scales, populations or risk groups on viral load [3], sexual positioning [4], male circumcision [5], with higher transmission rates will tend to have a paraphyletic and comorbidities [6]. Viral load is in turn mediated by the relationship with populations that are primarily recipients of natural history of HIV infection, and many previous investiga- infection. And over long periods of time, HIV genetic diver- tions have elucidated the role of early/acute HIV infection in sity is informative about the effective population size of the virus and epidemic growth rates [14, 15]. Genetic clustering of Received 15 August 2017; editorial decision 1 December 2017; accepted 22 January 2018; potential transmission pairs in large HIV sequence databases published online February 26, 2018. is a simple and scalable approach to characterizing transmis- Presented in part: HIV Dynamics and Evolution conference, Skye, United Kingdom, 25 May 2017. Correspondence: E. M.  Volz, PhD, Department of Infectious Disease Epidemiology, Imperial sion patterns [16], but genetic clustering is a highly unreliable College London, Norfolk Place, London W2 1PG, UK (e.volz@imperial.ac.uk). proxy for transmission risk and inferences based on clusters The Journal of Infectious Diseases 2018;217:1522–9 are known to be biased by correlations with stage of infection © The Author(s) 2018. Published by Oxford University Press for the Infectious Diseases Society of America. This is an Open Access article distributed under the terms of the Creative Commons at time of sampling [17–20]. In this study, we computed genetic Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted clusters for a large sample of HIV-1 subtype B sequences in reuse, distribution, and reproduction in any medium, provided the original work is properly cited. the United Kingdom and used these results to heuristically DOI: 10.1093/infdis/jiy044 1522 • JID 2018:217 (15 May) • Volz et al Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 identify factors that may enhance individual-level transmis- e w Th ork was conducted as part of the National Institute sion risk, with a particular focus on the role of young age on for Health Research Health Protection Research Unit (NIHR mediating HIV transmission in men who have sex with men HPRU) at Imperial College London (Modelling Methodology), (MSM). For selected variables with significant clustering asso - a partnership with PHE. ciations, we performed a more robust phylodynamic analysis Genetic Clustering using formal phylogenetic and population genetic modeling Heuristic genetic clustering analyses were carried out using [21]. Well-designed population genetic models can account for threshold evolutionary distance, as described in [13]. Clusters observed genetic diversity resulting from differential sampling were computed using thresholds of 0.5% and 1.5% distance effort over time [ 22] and between risk groups [23], and can using a TN93 substitution model, and sequence ambiguities account for nonlinear epidemic dynamics through time [24]. were averaged when computing evolutionary distances between This analysis provided estimates of transmission risk ratios for sequences. To identify variables that may be related to trans- selected biological and demographic covariates, which in turn mission risk, univariate logistic regression models were used informed a mathematical model of a PrEP intervention. to quantify the relative risk of clustering at different genetic distance thresholds. Multivariate logistic regression mod- METHODS els including an indicator for early HIV infection (EHI) were Data used to adjust for upwards skew in frequency of clustering of The UK HIV Drug Resistance Database contains more than recent HIV infections [20]. Young men who have sex with men 100 000 sequences from more than 60 000 patients at the end (YMSM) were defined as patients with sequences sampled while of 2015 (http://www.hivrdb.org.uk/). We extracted 6912 par- the patient had an age less than 25 years, which corresponds to tial pol HIV-1 sequences and associated metadata (patient- the bottom 10.7% of the age distribution. Both age at time of level variables) that met the following criteria: (1) the sequence sampling and absolute age (corresponding to year of birth) may was subtype B determined using REGA [25]; (2) the sequence be important determinants of evolutionary history of an HIV length was >1200 nucleotides; (3) the patient reported being lineage; however, we defined YMSM based on age at sampling MSM; and (4) the patient was treatment naive. We excluded because we anticipate this variable to have stronger association all but the first sequence per patient if multiple sequences are with recent transmission history of that lineage. The age thresh - available. We further restricted our analysis to samples that had old for defining YMSM is shared by recent studies [ 28] and was a CD4 and/or recent infection testing algorithm (RITA) [26] motivated by observed increasing odds of clustering with young result within 1  year of the sequence sample date in order to age and the objective of identifying a relatively small risk group adjust for the effect of recency of infection on clustering and that would benefit from PrEP prioritization. phylodynamic analyses. Sequences were collected between 1991 and the end of 2014 with 50% of samples collected Phylogenetic Analysis aer 2009. ft Phylogenetic trees were estimated by maximum likelihood To account for importation of HIV lineages, we added 1006 using ExaML and the R package big.phylo with a general time subtype B global reference sequences corresponding to unique reversible model of nucleotide substitution and gamma distri- sequences with highest similarity (using bitscore) aer a B ft LAST bution for rate heterogeneity among sites [29]. One hundred search for each of the UK sequences. The BLAST database trees were reconstructed from bootstrap alignment. Three sub - comprised 18 544 global reference sequences (excluding UK type G reference strains from Los Alamos database were used as sequences) obtained from Los Alamos HIV sequence data- outgroup for rooting the subtype-specific trees. base (https://www.hiv.lanl.gov, accessed October 2016). Drug resistance mutation sites as listed in the 2015 update from the Molecular Clock International Antiviral Society-USA [27] were stripped from We calculated root-to-tip distance from the phylogenetic tree the alignment using the R package big.phylo (https://github. and regressed distance by time of sampling. By iterations of com/olli0601/big.phylo). Grubb’s algorithm [30] (https://CRAN.R-project.org/pack- Multiple demographic and clinical covariates were available age=outliers), we identified and excluded 0.3% sequences as for each patient from Public Health England’s (PHE’s) HIV and outliers in terms of divergence time and evolutionary rate. We AIDS Reporting System (HARS), which included persons diag- applied least-square dating (LSD) algorithm [31] on rooted nosed with HIV and seen for care. These data were linked to the trees and sampling times to estimate the substitution rate and UK Resistance database and included: (1) region of diagnosis dates of ancestral nodes. To ensure accuracy of time-scaled phy- corresponding to 12 reporting regions of PHE in the United logenies, the fast LSD method was compared to slower state of Kingdom, (2) year of birth, (3) ethnicity, (4) CD4 counts, and the art Bayesian methods (BEAST) [32] for a single clade using (5) viral loads. lineage-through-time statistics. Estimated lineages through Heterogeneous Transmission Risk of HIV • JID 2018:217 (15 May) • 1523 Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 time using LSD and BEAST are compared in the Supplementary YMSM transmitting to another YMSM and the probability of information. an older MSM (OMSM) transmitting to a YMSM. To facilitate computation with very large phylogenies, we Coalescent analyses were implemented using the phydynR R divided the maximum likelihood tree into 21 disjoint clades package (https://github.com/emvolz-phylodynamics/phydynR) defined by threshold time to most recent common ancestor and model parameters were estimated using maximum likeli- (TMRCA). The threshold TMRCA was chosen such that the hood. Confidence intervals (CI) for transmission risk ratios of maximum number of sampled lineages in any clade was fewer EHI and YMSM, age assortativity parameters, and exogenous lin- than 1000 and the minimum clade size was 300. All clades had eage importation rates were computed using likelihood proles. fi a TMRCA before 1980 and thus larger clades included both A complete specification of the model equations, code, and closely and distantly related sequences. Clades should not be estimation methodology is available in the Supplementary confused with genetic distance clusters. Phylodynamic analyses information. were run in parallel on each clade. Predicting PrEP Intervention Effectiveness Phylodynamic Analysis We simulated a PrEP intervention based on provision of ARVs A structured coalescent model [33] was developed to estimate to approximately 15 000 susceptible individuals who are vul- transmission risk ratios from the time-stamped HIV phylogeny nerable to HIV infection. This strategy is a modest scale-up of while adjusting for stage of infection at time of sampling and current plans to provide PrEP to 10 000 eligible individuals over differential sampling effort among young MSM and the remain - 3 years [38]. ing population. To adjust for stage of infection, we assigned Two scenarios were considered in order to evaluate the ben- each lineage to a CD4 stage described by Cori et  al [34]. We efit of prioritizing YMSM with higher risk of both infection defined the EHI stage to include both recent/acute infection and transmission. In the first scenario, PrEP was randomly and patients with high CD4 >500. Thus the EHI period is likely allocated to all MSM irrespective of age, and in the second to encompass more than a year of the initial infectious period scenario, all PrEP was allocated to YMSM. Note that PrEP for most patients. Stage assignments were based on the CD4 will not be allocated completely at random, and the first sce - result collected nearest in time to sequence sampling (maxi- nario is used as a benchmark rather than to model a likely mum 1 year) as well as the RITA test result if available. outcome or standard of care. All simulations assumed 90% To model the dynamics of the number of infections and effectiveness in preventing infection. The population-level transmission rates within and between each deme, we devel- impact of PrEP depends on the proportion of susceptible oped a compartmental infectious disease model consisting of individuals treated, and the number of susceptible MSM 7 ordinary differential equations, which described the number was extrapolated from recent HIV prevalence estimates and of infections in each of 3 stages of infection and 2 transmission number diagnosed in the United Kingdom [39]. Given an risk levels corresponding to age group. The transmission rate estimated 45 000 MSM diagnosed and undiagnosed living was modeled as a product of 3 factors: (1) risk level according with HIV at the end of 2014 and HIV prevalence among to a binary covariate such as being a young MSM; (2) stage of MSM aged 15–44 between 4.1% and 5.8%, we infer there to infection (EHI, chronic, or AIDS); and (3) secular trends in be between 731 000 and 1.05 million susceptible MSM. We transmission rate. We modeled incidence of infection in each therefore examined a range of proportions receiving PrEP clade using a susceptible-infected-removed (SIR) model and of 1.4%–2.1% for all MSM or alternatively 13.4%–19.2% of estimated susceptible population size and transmission rate YMSM. This simulation exercise did not account for self- separately in each clade. medication with PrEP or potential differences in self-prophy - Importation of lineages into the United Kingdom was mod- laxis between age groups. The number of new HIV infections eled using a single deme to represent the global HIV reservoir. was simulated under both scenarios over a 5-year horizon by e r Th eservoir deme was designed to have exponentially growing modifying the mathematical model fitted to HIV phylogenies effective population size with 2 free parameters which were esti - and reducing transmission rates in proportion to the number mated independently using a skyspline model [35]. Importation of susceptibles receiving PrEP. from the reservoir is modeled as a source-sink relationship RESULTS with a constant rate per lineage. Once a lineage migrates from the reservoir, we assume that it may not circulate back to the Relative to older age groups, YMSM were more likely to be sam- reservoir. pled with recent infection corresponding to higher CD4s and Whereas transmission patterns between age groups may be more frequent RITA-positive test results (Table 1). YMSM had highly nonrandom [36, 37] and transmissions are more likely significantly higher rates of EHI defined as CD4>500 or RITA between people in similar age groups, 2 additional parameters positive test result (41% versus 23%, 2-sample binomial test). were estimated that describe the conditional probability of a YMSM of subtype B were also more likely to reside outside of 1524 • JID 2018:217 (15 May) • Volz et al Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 Table 1. Demographic and Clinical Characteristics of YMSM and OMSM Finally, we estimated the average time that a HIV lineage Included in the Analysis and all MSM in the Database has circulated in the United Kingdom prior to sampling. This was 27.6 years (95% CI, 26.6–28.75), suggesting that most sub- YMSM/B OMSM/B All MSM type B infections in MSM are derived from introductions that (n = 706) (n = 6206) (n = 30 711) occurred in the late 1980s [14]. Year of birth (IQR) 1987(1984–1990) 1971(1964–1977) 1971(1964–1979) e fi Th tted population genetic model provided estimates of CD4 (IQR) 469(337–620) 415(259–583) 420(260–590) RITA+ 34% 25% 28% the number of effective infections through time ( Figure  1 and London 39% 55% 59% Supplementary Figure S1) which approximately corresponds White 84% 88% 85% to the number of MSM living with HIV within the 21 clades Immigrant 19% 26% 30% included in this analysis. Note that these estimates were based Clustered (0.5%) 31% 20% on a subset of all HIV-1 genetic diversity and the absolute num- Clustered (1.5%) 77% 62% ber of infections does not correspond to the total number of YMSM and OMSM count only patients that meet  all inclusion criteria. Statistical tests compare YMSM/B or OMSM/B and all MSM, except for the clustered outcome for which infections in the population. Nevertheless, the estimated rates YMSM and OMSM are compared. Significance levels were determined using a 2-sample t of growth of MSM living with HIV are similar to estimates test for continuous variables and Fishers exact test for binary variables. Entries in bold have a P value < .001. obtained by Public Health England based on surveillance data Abbreviations: IQR, interquartile range; OMSM, older men who have sex with men; (Figure 1B and Supplementary Figure S2). We estimated that the YMSM, young men who have sex with men. 21 clades included samples from 60% of people living with HIV descended from the clades, the remainder not meeting inclusion criteria, not having sequences, or not being diagnosed in the the London metropolitan area and less likely to be born outside United Kingdom. Note that the sample proportion is influenced of the United Kingdom. Relative to OMSM, YMSM were more by the fact that approximately 80% are diagnosed, not all diag- likely to be genetically clustered with at least 1 other patient nosed have a sequence in the database, and approximately 50% (31% versus 20%). More than 60% of both YMSM and OMSM of lineages were excluded due to lack of adequate biomarkers clustered with at least 1 other patient using 1.5% threshold or because they were collected from ART-experienced patients. evolutionary distance and at 0.5% threshold distance about a er Th e was substantial variation in the proportion of each clade quarter of patients clustered. Small but statistically significant that are YMSM, and the proportion of transmissions attribut- associations were found between the odds of clustering at 0.5% able to YMSM in each clade (Figure  1). The proportion of the and 1.5% evolutionary distance and EHI, age, and location of clade that was YMSM was not significantly associated with sampling. Patients sampled with EHI clustered slightly more in growth rates of effective infections in each clade (F test P = .42). multivariate analyses (odds ratio [OR] = 1.14, 0.5% threshold) Simulated PrEP interventions based on the fitted popula - as do YMSM (OR = 1.09, 0.5% threshold). tion genetic model showed large gains from focusing PrEP on Coalescent-based phylodynamic analysis showed strong evi- YMSM in comparison to random allocation to all MSM. Note dence of higher transmission risk for both EHI and YMSM. that, in reality, PrEP would not be allocated randomly, and ran- e t Th ransmission risk ratio of YMSM relative to all other MSM dom allocation should be interpreted as a benchmark rather (OMSM) is 2.52 (95% CI, 2.32–2.73). The transmission risk than a likely outcome or standard of care. Figure  2 shows the ratio of EHI (CD4  >500 and/or RITA positive) relative to all predicted cumulative infections averted by PrEP over 5  years other stages of infection is 3.70 (95% CI, 3.36–4.09). These rep - if 15 000 susceptible individuals were provided PrEP in 2015, resent independent effects, and YMSM with EHI were predicted which was the end point for sequence data included in this to have the highest transmission risk. study. Simulations reflect not only direct impacts from prevent - e p Th hylodynamic analysis also revealed highly nonrandom ing infections in treated individuals, but also indirect effects transmission patterns by age. The probability that the recipient from preventing transmission by individuals who would have is YMSM given an infected donor in the OMSM risk group was become infected without PrEP. We predicted that 749 (636–857) 20.0% (95% CI, 17.7%–22.7%), which is roughly twice the pro- infections would be averted over 5  years if PrEP was focused portion of the population that is YMSM (approximately 10% on YMSM and that 179 (150–207) infections would be averted of MSM by definition). However, the probability that the recip - with random allocation. PrEP for YMSM averted 4.2 times as ient is YMSM given a YMSM donor was very much higher: many infections over 5 years as random allocation. 83.3% (95% CI, 78.4%–87.2%). Consequently, most YMSM were infected by other YMSM and not by older age groups. 75% DISCUSSION of infections in YMSM were attributable to other YMSM, and 87% of infections in OMSM were attributable to other OMSM. e co Th mbination of high levels of transmission and high assor - Age assortativity and transmission patterns are summarized in tativity amplifies incidence in YMSM and PrEP effectiveness in Figure 1. YMSM. PrEP in YMSM averts transmissions that would occur Heterogeneous Transmission Risk of HIV • JID 2018:217 (15 May) • 1525 Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 A C 9.1 0.5 9.0 Clade size / 100 8.9 0.4 6000 4 8.8 8.7 0.3 0 8.6 9.8 10.0 10.2 10.4 1980 1990 2000 2010 05 10 15 20 log MSM living with HIV Clade index Pair probabilities Population attributable fraction of transmissions 17% 7% 83% YMSM 80%35% YMSM 46% Other MSM Other MSM 20% 12% Figure 1. Effective number of infections through time, cumulative sequences sampled through time, and proportion of transmissions attributable to young men who have sex with men (YMSM) for 21 subtype B clades in UK MSM. A, Effective infections and cumulative samples combined for 21 clades. B, Number of MSM living with human immunodeficiency virus (HIV) estimated from surveillance data between 2004 and 2012 [ 39] versus phylodynamic estimates (log-transformed). A  regression with slope constrained to 1 is shown in red. C, The estimated proportion of transmissions attributable to YMSM (age <25) for 21 clades and the number of samples in each clade. D, Estimated transmission patterns between YMSM and other MSM (OMSM). Left: The probability that a recipient is in each age group given that a donor is either YMSM or OMSM. Right: The proportion of all transmissions that are attributable to each pair of age groups. Abbreviation: PAF, population attributable fraction. if YMSM were infected, and because YMSM are more likely to HIV genetic diversity in the United Kingdom showed a clear infect one another, subsequent generations of the epidemic pro- pattern consistent with higher risk of infection in YMSM and cess are further reduced. These transmission patterns yield higher higher risk of transmission EHI. The effect of young age on incidence in YMSM, and it remains to corroborate observations transmission risk remained aer co ft ntrolling for higher rates based on phylodynamic analysis by estimating age-specific inci - of EHI among patients diagnosed at young age. YMSM have dence in UK MSM from nongenetic surveillance data. the hallmarks of a small high-risk core group [40], having both higher intrinsic transmission rates and preferential attachment within the risk group. YMSM show very high levels of assort- ative mixing. Most infections (75%) in YMSM arose via inter- action with other YMSM despite comprising around 10% of the infected MSM population. These findings are consistent with previous reports of higher rates of HIV genetic clustering 500 in YMSM [41, 42]. We find evidence for a modest net flow of Random transmissions from OMSM to YMSM which is in line with the Targetted findings of age-discordant clustering among young MSM found by Wolf et al [28]. These transmission patterns also differ from studies of genetic clustering in heterosexual populations, which have shown much greater age-discordancy within clusters that is hypothesized to arise from net flows of transmission from older males to younger females [43]. 2015 2016 2017 2018 2019 2020 In reality, PrEP will not be allocated randomly, and age may provide one of many criteria for PrEP. Further studies Figure 2. The number of infections averted over a 5-year horizon using a random are warranted to examine how age can be used in combina- allocation of pre-exposure prophylaxis (PrEP) or a targeted PrEP strategy focused on tion with other transmission and infection risk factors. Recent young men who have sex with men. 1526 • JID 2018:217 (15 May) • Volz et al E ective number of infections (line) Cumulative samples (dash) Infections averted log e ective MSM living with HIV PAF YMSM Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 randomized clinical trials have examined the direct protec- Potential coni fl cts of interest. All authors: No reported con- tive effects of PrEP in MSM but have not accounted for indi - flicts of interest. All authors have submitted the ICMJE Form rect transmission effects included in our simulations [ 44, 45]. for Disclosure of Potential Conflicts of Interest. Conflicts that Clinical trials, including studies conducted with UK MSM [46], the editors consider relevant to the content of the manuscript have focused on individuals at higher risk of infection than have been disclosed. MSM as a whole (eg, those with recent sexually transmitted References infection [STI] testing history and condomless sex), so that esti- mates of infections averted are not directly comparable with our 1. Abu-Raddad LJ. Role of acute HIV infection in driving HIV simulation results, which were constrained by the available data transmission: implications for HIV treatment as preven- to focus on random allocation within age groups. tion. PLoS Med 2015; 12:e1001803. This study only examined subtype B sequences, which com - 2. Chemaitelly H, Awad SF, Abu-Raddad LJ. The risk of HIV prise the large majority of sequences among MSM in the United transmission within HIV-1 sero-discordant couples appears Kingdom. Less-prevalent subtypes have different demographic to vary across sub-Saharan Africa. Epidemics 2014; 6:1–9. and clinical characteristics, notably CRF02AG which has higher 3. Quinn TC, Wawer MJ, Sewankambo N, et al. Viral load and proportions of recent African migrants (results not shown), and heterosexual transmission of human immunodeficiency these results may not generalize to those clades. Compared to all virus type 1.  Rakai Project Study Group. N Engl J Med sequences from MSM, the HIV lineages included in this analy- 2000; 342:921–9. sis come from patients that are less likely to reside in London or 4. Boily MC, Baggaley RF, Wang L, et  al. Heterosexual risk be foreign born (Table 1). of HIV-1 infection per sexual act: systematic review and Young age in MSM is a simple and easily identifiable proxy meta-analysis of observational studies. Lancet Infect Dis for transmission risk that may form one of many inputs into 2009; 9:118–29. algorithms for prioritizing or promoting PrEP. Parameters 5. Wawer MJ, Makumbi F, Kigozi G, et  al. Circumcision in estimated in this study may provide useful inputs into more HIV-infected men and its effect on HIV transmission to detailed simulations for designing realistic PrEP interventions female partners in Rakai, Uganda: a randomised controlled [47]. Access to PrEP and knowledge of PrEP may be lower in trial. Lancet 2009; 374:229–37. young MSM [48], providing further impetus to provide benefits 6. Baeten JM, Overbaugh J. Measuring the infectiousness of for that group. YMSM can not be prioritized to the exclusion persons with HIV-1: opportunities for preventing sexual of other age groups, nor should age be the sole criterion used. HIV-1 transmission. Curr HIV Res 2003; 1:69–86. Many risk groups were not considered in this phylogenetic 7. Hollingsworth TD, Pilcher CD, Hecht FM, Deeks SG, Fraser analysis and it is important to have diverse as well as focused C. High transmissibility during early HIV infection among allocation of PrEP [49]. Other variables, such as recent STI test- men who have sex with men-San Francisco, California. ing history or self-reported risk exposures, are likely to also cor- J Infect Dis 2015; 211:1757–60. relate highly with transmission risk, and it remains to examine 8. Volz EM, Ionides E, Romero-Severson EO, Brandt MG, such covariates within a phylodynamics framework. Mokotoff E, Koopman JS. HIV-1 transmission during early infection in men who have sex with men: a phylody- Supplementary Data namic analysis. PLoS Med 2013; 10:e1001568; discussion Supplementary materials are available at e Th Journal of Infectious e1001568. Diseases online. Consisting of data provided by the authors to 9. Marzel A, Shilaih M, Yang WL, et  al.; Swiss HIV Cohort benefit the reader, the posted materials are not copyedited and Study. HIV-1 Transmission during recent infection and are the sole responsibility of the authors, so questions or com- during treatment interruptions as major drivers of new ments should be addressed to the corresponding author. infections in the Swiss HIV Cohort Study. Clin Infect Dis 2016; 62:115–22. Notes 10. Dennis AM, Herbeck JT, Brown AL, et  al. Phylogenetic Acknowledgments. This research was made possible by the studies of transmission dynamics in generalized HIV epi- Steering Committee and Coordinating Centre of the UK Drug demics: an essential tool where the burden is greatest? Resistance Database the Centres contributing data to the UK J Acquir Immune Defic Syndr 2014; 67:181–95. Drug Resistance Database. A full list of contributors is provided 11. Pillay D, Herbeck J, Cohen MS, et  al.; PANGEA-HIV in the online Supporting material. Consortium. PANGEA-HIV: phylogenetics for generalised Funding. This work was supported by the UK Health epidemics in Africa. Lancet Infect Dis 2015; 15:259–61. Protection Research Units in Modeling Methodology and 12. Romero-Severson E, Skar H, Bulla I, Albert J, Leitner T. Timing Sexually Transmitted Infections and by the National Institutes and order of transmission events is not directly reflected in a of Health (grant number R01AI087520). pathogen phylogeny. Mol Biol Evol 2014; 31:2472–82. Heterogeneous Transmission Risk of HIV • JID 2018:217 (15 May) • 1527 Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 13. Wertheim JO, Leigh Brown AJ, Hepler NL, et  al. e Th 28. Wolf E, Herbeck JT, Van Rompaey S, et  al. Short commu- global transmission network of HIV-1. J Infect Dis 2014; nication: phylogenetic evidence of HIV-1 transmission 209:304–13. between adult and adolescent men who have sex with men. 14. Hué S, Pillay D, Clewley JP, Pybus OG. Genetic analy- AIDS Res Hum Retroviruses 2017; 33:318–22. sis reveals the complex structure of HIV-1 transmission 29. Kozlov AM, Aberer AJ, Stamatakis A. ExaML version within defined risk groups. Proc Natl Acad Sci U S A 2005; 3: a tool for phylogenomic analyses on supercomputers. 102:4425–9. Bioinformatics 2015; 31:2577–9. 15. Faria NR, Rambaut A, Suchard MA, et  al. HIV epidemi- 30. Grubbs FE. Sample criteria for testing outlying observa- ology. e e Th arly spread and epidemic ignition of HIV-1 in tions. Ann Math Stat 1950; 21:27–58. human populations. Science 2014; 346:56–61. 31. To TH, Jung M, Lycett S, Gascuel O. Fast dating using 16. Poon AF, Gustafson R, Daly P, et al. Near real-time monitoring least-squares criteria and algorithms. Syst Biol 2016; of HIV transmission hotspots from routine HIV genotyping: 65:82–97. an implementation case study. Lancet HIV 2016; 3:e231–8. 32. Drummond AJ, Suchard MA, Xie D, Rambaut A. Bayesian 17. Le Vu S, Ratmann O, Delpech V, et al. Comparison of clus- phylogenetics with BEAUti and the BEAST 1.7. Mol Biol ter-based and source-attribution methods for estimating Evol 2012; 29:1969–73. transmission risk using large HIV sequence databases [pub- 33. Volz EM. Complex population dynamics and the coalescent lished online ahead of print 20 October,  2017]. Epidemics under neutrality. Genetics 2012; 190:187–201. doi: 10.1016/j.epidem.2017.10.001. 34. Cori A, Pickles M, van Sighem A, et al. CD4+ cell dynam- 18. Dearlove BL, Xiang F, Frost SDW. Biased phylodynamic ics in untreated HIV-1 infection: overall rates, and effects inferences from analysing clusters of viral sequences. Virus of age, viral load, sex and calendar time. AIDS 2015; Evol 2017; 3:vex020. 29:2435–46. 19. Poon AF. Impacts and shortcomings of genetic clustering 35. Volz EM, Romero-Severson E, Leitner T. Phylodynamic methods for infectious disease outbreaks. Virus Evol 2016; inference across epidemic scales. Mol Biol Evol 2017; 2:vew031. 34:1276–88. 20. Volz EM, Koopman JS, Ward MJ, Brown AL, Frost SD. 36. Prah P, Copas AJ, Mercer CH, Nardone A, Johnson AM. Simple epidemiological dynamics explain phylogenetic Patterns of sexual mixing with respect to social, health clustering of HIV from patients with recent infection. PLoS and sexual characteristics among heterosexual couples Comput Biol 2012; 8:e1002552. in England: analyses of probability sample survey data. 21. Volz EM, Koelle K, Bedford T. Viral phylodynamics. PLoS Epidemiol Infect 2015; 143:1500–10. Comput Biol 2013; 9:e1002947. 37. Phillips AN, Cambiano V, Nakagawa F, et al. Increased HIV 22. Volz EM, Frost SD. Sampling through time and phylody- incidence in men who have sex with men despite high levels namic inference with coalescent and birth-death models. of ART-induced viral suppression: analysis of an extensively J R Soc Interface 2014; 11:20140945. documented epidemic. PLoS One 2013; 8:e55312. 23. De Maio N, Wu CH, O’Reilly KM, Wilson D. New routes to 38. NHS England. NHS England announces world’s largest single phylogeography: a bayesian structured coalescent approxi- PrEP implementation trial to prevent HIV infection, 2017. mation. PLoS Genet 2015; 11:e1005421. https://www.england.nhs.uk/2017/08/nhs-england-an- 24. Ratmann O, Hodcroft EB, Pickles M, et al.; PANGEA-HIV nounces-worlds-largest-single-prep-implementation-tri- Consortium. Phylogenetic tools for generalized HIV-1 epi- al-to-prevent-hiv-infection/. Accessed 10 February 2018. demics: findings from the PANGEA-HIV methods compar - 39. Skingsley A, Yin Z, Kirwan P, et al. HIV in the UK–Situation ison. Mol Biol Evol 2017; 34:185–203. Report 2015: data to end 2014. London: Public Health 25. Pineda-Peña AC, Faria NR, Imbrechts S, et  al. Automated England, 2015. subtyping of HIV-1 genetic sequences for clinical and sur- 40. Caldarelli G, Catanzaro M. Networks: A  very short intro- veillance purposes: performance evaluation of the new duction. Oxford: Oxford University Press, 2012. REGA version 3 and seven other tools. Infect Genet Evol 41. Pao D, Fisher M, Hué S, et al. Transmission of HIV-1 during 2013; 19:337–48. primary infection: relationship to sexual risk and sexually 26. Aghaizu A, Murphy G, Tosswill J, et  al. Recent infection transmitted infections. AIDS 2005; 19:85–90. testing algorithm (RITA) applied to new HIV diagnoses in 42. Lubelchek RJ, Hoehnen SC, Hotton AL, Kincaid SL, Barker England, Wales and Northern Ireland, 2009 to 2011. Euro DE, French AL. Transmission clustering among newly Surveill 2014; 19:20673. diagnosed HIV patients in Chicago, 2008 to 2011: using 27. Wensing AM, Calvez V, Günthard HF, et al. 2015 update of phylogenetics to expand knowledge of regional HIV trans- the drug resistance mutations in HIV-1. Top Antivir Med mission patterns. J Acquir Immune Defic Syndr 2015; 2015; 23:132–41. 68:46–54. 1528 • JID 2018:217 (15 May) • Volz et al Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 43. de Oliveira T, Kharsany AB, Gräf T, et al. Transmission net- (PROUD): effectiveness results from the pilot phase of works and risk of HIV infection in KwaZulu-Natal, South a pragmatic open-label randomised trial. Lancet 2016; Africa: a community-wide phylogenetic study. Lancet HIV 387:53–60. 2017; 4:e41–50. 47. Cambiano V, Miners A, Dunn D, et al. Cost-effectiveness 44. Grant RM, Lama JR, Anderson PL, et al.; iPrEx Study Team. of pre-exposure prophylaxis for HIV prevention in men Preexposure chemoprophylaxis for HIV prevention in men who have sex with men in the UK: a modelling study who have sex with men. N Engl J Med 2010; 363:2587–99. and health economic evaluation. Lancet Infect Dis 2018; 45. Buchbinder SP, Glidden DV, Liu AY, et  al. HIV pre-ex- 18:85–94. posure prophylaxis in men who have sex with men and 48. Bauermeister JA, Meanley S, Pingel E, Soler JH, Harper transgender women: a secondary analysis of a phase 3 ran- GW. PrEP awareness and perceived barriers among single domised controlled efficacy trial. Lancet Infect Dis 2014; young men who have sex with men. Curr HIV Res 2013; 14:468–75. 11:520–7. 46. McCormack S, Dunn DT, Desai M, et  al. Pre-exposure 49. e L Th ancet HIV. Better late than never: PrEP in England. prophylaxis to prevent the acquisition of HIV-1 infection Lancet HIV 2017; 4:e1. Heterogeneous Transmission Risk of HIV • JID 2018:217 (15 May) • 1529 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The Journal of Infectious Diseases Oxford University Press

Molecular Epidemiology of HIV-1 Subtype B Reveals Heterogeneous Transmission Risk: Implications for Intervention and Control

Loading next page...
 
/lp/oxford-university-press/molecular-epidemiology-of-hiv-1-subtype-b-reveals-heterogeneous-IHJe9UOHjX

References (53)

Publisher
Oxford University Press
Copyright
Copyright © 2022 Infectious Diseases Society of America
ISSN
0022-1899
eISSN
1537-6613
DOI
10.1093/infdis/jiy044
Publisher site
See Article on Publisher Site

Abstract

Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 Molecular Epidemiology of HIV-1 Subtype B Reveals Heterogeneous Transmission Risk: Implications for Intervention and Control Erik M Volz, Stephane Le Vu, Oliver Ratmann, Anna Tostevin, David Dunn, Chloe Orkin, Siobhan O’Shea, Valerie Delpech, Alison Brown, Noel Gill, Christophe Fraser, UK HIV Drug Resistance Database Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 The Journal of Infectious Diseases MAJOR ARTICLE Molecular Epidemiology of HIV-1 Subtype B Reveals Heterogeneous Transmission Risk: Implications for Intervention and Control 1 1 1 2 2 3 4 5 5 Erik M. Volz, Stephane Le Vu, Oliver Ratmann, Anna Tostevin, David Dunn, Chloe Orkin, Siobhan O’Shea, Valerie Delpech, Alison Brown, 5 6 Noel Gill, and Christophe Fraser ; on behalf of the UK HIV Drug Resistance Database Department of Infectious Disease Epidemiology and the National Institute for Health Research Health Protection Research Unit on Modeling Methodology, Imperial College 2 3 4 London; Institute for Global Health, University College London; Barts Health NHS Trust, London; Infection Sciences, Viapath Analytics, Guy’s and St Thomas’ NHS Foundation 5 6 Trust, London; Public Health England, London; and Li Ka Shing Centre for Health Information and Discovery, Oxford University, United Kingdom (See the Editorial commentary by Baeten, on pages 1509–11.) Background. e Th impact of HIV pre-exposure prophylaxis (PrEP) depends on infections averted by protecting vulnerable indi - May viduals as well as infections averted by preventing transmission by those who would have been infected if not receiving PrEP. Analysis of HIV phylogenies reveals risk factors for transmission, which we examine as potential criteria for allocating PrEP. Methods. We analyzed 6912 HIV-1 partial pol sequences from men who have sex with men (MSM) in the United Kingdom combined with global reference sequences and patient-level metadata. Population genetic models were developed that adjust for stage of infection, global migration of HIV lineages, and changing incidence of infection through time. Models were extended to simulate the effects of providing susceptible MSM with PrEP. Results. We found that young age <25 years confers higher risk of HIV transmission (relative risk = 2.52 [95% confidence inter - val, 2.32–2.73]) and that young MSM are more likely to transmit to one another than expected by chance. Simulated interventions indicate that 4-fold more infections can be averted over 5 years by focusing PrEP on young MSM. Conclusions. Concentrating PrEP doses on young individuals can avert more infections than random allocation. Keywords. HIV; men who have sex with men; phylodynamics; pre-exposure prophylaxis. The effectiveness of public health interventions (PHIs) to enhancing transmission risk and increasing epidemic spread com bat human immunodeficiency virus (HIV), such as pre- [7–9]. exposure prophylaxis (PrEP) with antiretroviral medications While the factors that shape transmission risk are under- (ARVs) depends on unknown variability of transmission risk in stood qualitatively, it is challenging to obtain robust quanti- the infected population. The impact of PHIs can be enhanced if tative estimates of transmission probabilities or transmission OA-CC-BY the intervention can be focused on patients with higher trans- risk. HIV genetic sequence data from routine drug resistance mission risk, due to, for example, different risk behaviors or testing is one of the few sources of widely available observa- epidemiological settings [1]. HIV transmission risk is highly tional data that are directly informative about HIV transmis- variable over time, over the course of infection, between risk sion patterns and transmission risk [10, 11]. Donor-recipient groups, and geographically [2]. Biological, behavioral, and envi- transmission pairs harbor virus that is genetically closely ronmental factors shape individual HIV transmission risk in related compared to the population as a whole [12, 13]. At complex ways. Transmission probabilities per coital act depend longer evolutionary time scales, populations or risk groups on viral load [3], sexual positioning [4], male circumcision [5], with higher transmission rates will tend to have a paraphyletic and comorbidities [6]. Viral load is in turn mediated by the relationship with populations that are primarily recipients of natural history of HIV infection, and many previous investiga- infection. And over long periods of time, HIV genetic diver- tions have elucidated the role of early/acute HIV infection in sity is informative about the effective population size of the virus and epidemic growth rates [14, 15]. Genetic clustering of Received 15 August 2017; editorial decision 1 December 2017; accepted 22 January 2018; potential transmission pairs in large HIV sequence databases published online February 26, 2018. is a simple and scalable approach to characterizing transmis- Presented in part: HIV Dynamics and Evolution conference, Skye, United Kingdom, 25 May 2017. Correspondence: E. M.  Volz, PhD, Department of Infectious Disease Epidemiology, Imperial sion patterns [16], but genetic clustering is a highly unreliable College London, Norfolk Place, London W2 1PG, UK (e.volz@imperial.ac.uk). proxy for transmission risk and inferences based on clusters The Journal of Infectious Diseases 2018;217:1522–9 are known to be biased by correlations with stage of infection © The Author(s) 2018. Published by Oxford University Press for the Infectious Diseases Society of America. This is an Open Access article distributed under the terms of the Creative Commons at time of sampling [17–20]. In this study, we computed genetic Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted clusters for a large sample of HIV-1 subtype B sequences in reuse, distribution, and reproduction in any medium, provided the original work is properly cited. the United Kingdom and used these results to heuristically DOI: 10.1093/infdis/jiy044 1522 • JID 2018:217 (15 May) • Volz et al Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 identify factors that may enhance individual-level transmis- e w Th ork was conducted as part of the National Institute sion risk, with a particular focus on the role of young age on for Health Research Health Protection Research Unit (NIHR mediating HIV transmission in men who have sex with men HPRU) at Imperial College London (Modelling Methodology), (MSM). For selected variables with significant clustering asso - a partnership with PHE. ciations, we performed a more robust phylodynamic analysis Genetic Clustering using formal phylogenetic and population genetic modeling Heuristic genetic clustering analyses were carried out using [21]. Well-designed population genetic models can account for threshold evolutionary distance, as described in [13]. Clusters observed genetic diversity resulting from differential sampling were computed using thresholds of 0.5% and 1.5% distance effort over time [ 22] and between risk groups [23], and can using a TN93 substitution model, and sequence ambiguities account for nonlinear epidemic dynamics through time [24]. were averaged when computing evolutionary distances between This analysis provided estimates of transmission risk ratios for sequences. To identify variables that may be related to trans- selected biological and demographic covariates, which in turn mission risk, univariate logistic regression models were used informed a mathematical model of a PrEP intervention. to quantify the relative risk of clustering at different genetic distance thresholds. Multivariate logistic regression mod- METHODS els including an indicator for early HIV infection (EHI) were Data used to adjust for upwards skew in frequency of clustering of The UK HIV Drug Resistance Database contains more than recent HIV infections [20]. Young men who have sex with men 100 000 sequences from more than 60 000 patients at the end (YMSM) were defined as patients with sequences sampled while of 2015 (http://www.hivrdb.org.uk/). We extracted 6912 par- the patient had an age less than 25 years, which corresponds to tial pol HIV-1 sequences and associated metadata (patient- the bottom 10.7% of the age distribution. Both age at time of level variables) that met the following criteria: (1) the sequence sampling and absolute age (corresponding to year of birth) may was subtype B determined using REGA [25]; (2) the sequence be important determinants of evolutionary history of an HIV length was >1200 nucleotides; (3) the patient reported being lineage; however, we defined YMSM based on age at sampling MSM; and (4) the patient was treatment naive. We excluded because we anticipate this variable to have stronger association all but the first sequence per patient if multiple sequences are with recent transmission history of that lineage. The age thresh - available. We further restricted our analysis to samples that had old for defining YMSM is shared by recent studies [ 28] and was a CD4 and/or recent infection testing algorithm (RITA) [26] motivated by observed increasing odds of clustering with young result within 1  year of the sequence sample date in order to age and the objective of identifying a relatively small risk group adjust for the effect of recency of infection on clustering and that would benefit from PrEP prioritization. phylodynamic analyses. Sequences were collected between 1991 and the end of 2014 with 50% of samples collected Phylogenetic Analysis aer 2009. ft Phylogenetic trees were estimated by maximum likelihood To account for importation of HIV lineages, we added 1006 using ExaML and the R package big.phylo with a general time subtype B global reference sequences corresponding to unique reversible model of nucleotide substitution and gamma distri- sequences with highest similarity (using bitscore) aer a B ft LAST bution for rate heterogeneity among sites [29]. One hundred search for each of the UK sequences. The BLAST database trees were reconstructed from bootstrap alignment. Three sub - comprised 18 544 global reference sequences (excluding UK type G reference strains from Los Alamos database were used as sequences) obtained from Los Alamos HIV sequence data- outgroup for rooting the subtype-specific trees. base (https://www.hiv.lanl.gov, accessed October 2016). Drug resistance mutation sites as listed in the 2015 update from the Molecular Clock International Antiviral Society-USA [27] were stripped from We calculated root-to-tip distance from the phylogenetic tree the alignment using the R package big.phylo (https://github. and regressed distance by time of sampling. By iterations of com/olli0601/big.phylo). Grubb’s algorithm [30] (https://CRAN.R-project.org/pack- Multiple demographic and clinical covariates were available age=outliers), we identified and excluded 0.3% sequences as for each patient from Public Health England’s (PHE’s) HIV and outliers in terms of divergence time and evolutionary rate. We AIDS Reporting System (HARS), which included persons diag- applied least-square dating (LSD) algorithm [31] on rooted nosed with HIV and seen for care. These data were linked to the trees and sampling times to estimate the substitution rate and UK Resistance database and included: (1) region of diagnosis dates of ancestral nodes. To ensure accuracy of time-scaled phy- corresponding to 12 reporting regions of PHE in the United logenies, the fast LSD method was compared to slower state of Kingdom, (2) year of birth, (3) ethnicity, (4) CD4 counts, and the art Bayesian methods (BEAST) [32] for a single clade using (5) viral loads. lineage-through-time statistics. Estimated lineages through Heterogeneous Transmission Risk of HIV • JID 2018:217 (15 May) • 1523 Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 time using LSD and BEAST are compared in the Supplementary YMSM transmitting to another YMSM and the probability of information. an older MSM (OMSM) transmitting to a YMSM. To facilitate computation with very large phylogenies, we Coalescent analyses were implemented using the phydynR R divided the maximum likelihood tree into 21 disjoint clades package (https://github.com/emvolz-phylodynamics/phydynR) defined by threshold time to most recent common ancestor and model parameters were estimated using maximum likeli- (TMRCA). The threshold TMRCA was chosen such that the hood. Confidence intervals (CI) for transmission risk ratios of maximum number of sampled lineages in any clade was fewer EHI and YMSM, age assortativity parameters, and exogenous lin- than 1000 and the minimum clade size was 300. All clades had eage importation rates were computed using likelihood proles. fi a TMRCA before 1980 and thus larger clades included both A complete specification of the model equations, code, and closely and distantly related sequences. Clades should not be estimation methodology is available in the Supplementary confused with genetic distance clusters. Phylodynamic analyses information. were run in parallel on each clade. Predicting PrEP Intervention Effectiveness Phylodynamic Analysis We simulated a PrEP intervention based on provision of ARVs A structured coalescent model [33] was developed to estimate to approximately 15 000 susceptible individuals who are vul- transmission risk ratios from the time-stamped HIV phylogeny nerable to HIV infection. This strategy is a modest scale-up of while adjusting for stage of infection at time of sampling and current plans to provide PrEP to 10 000 eligible individuals over differential sampling effort among young MSM and the remain - 3 years [38]. ing population. To adjust for stage of infection, we assigned Two scenarios were considered in order to evaluate the ben- each lineage to a CD4 stage described by Cori et  al [34]. We efit of prioritizing YMSM with higher risk of both infection defined the EHI stage to include both recent/acute infection and transmission. In the first scenario, PrEP was randomly and patients with high CD4 >500. Thus the EHI period is likely allocated to all MSM irrespective of age, and in the second to encompass more than a year of the initial infectious period scenario, all PrEP was allocated to YMSM. Note that PrEP for most patients. Stage assignments were based on the CD4 will not be allocated completely at random, and the first sce - result collected nearest in time to sequence sampling (maxi- nario is used as a benchmark rather than to model a likely mum 1 year) as well as the RITA test result if available. outcome or standard of care. All simulations assumed 90% To model the dynamics of the number of infections and effectiveness in preventing infection. The population-level transmission rates within and between each deme, we devel- impact of PrEP depends on the proportion of susceptible oped a compartmental infectious disease model consisting of individuals treated, and the number of susceptible MSM 7 ordinary differential equations, which described the number was extrapolated from recent HIV prevalence estimates and of infections in each of 3 stages of infection and 2 transmission number diagnosed in the United Kingdom [39]. Given an risk levels corresponding to age group. The transmission rate estimated 45 000 MSM diagnosed and undiagnosed living was modeled as a product of 3 factors: (1) risk level according with HIV at the end of 2014 and HIV prevalence among to a binary covariate such as being a young MSM; (2) stage of MSM aged 15–44 between 4.1% and 5.8%, we infer there to infection (EHI, chronic, or AIDS); and (3) secular trends in be between 731 000 and 1.05 million susceptible MSM. We transmission rate. We modeled incidence of infection in each therefore examined a range of proportions receiving PrEP clade using a susceptible-infected-removed (SIR) model and of 1.4%–2.1% for all MSM or alternatively 13.4%–19.2% of estimated susceptible population size and transmission rate YMSM. This simulation exercise did not account for self- separately in each clade. medication with PrEP or potential differences in self-prophy - Importation of lineages into the United Kingdom was mod- laxis between age groups. The number of new HIV infections eled using a single deme to represent the global HIV reservoir. was simulated under both scenarios over a 5-year horizon by e r Th eservoir deme was designed to have exponentially growing modifying the mathematical model fitted to HIV phylogenies effective population size with 2 free parameters which were esti - and reducing transmission rates in proportion to the number mated independently using a skyspline model [35]. Importation of susceptibles receiving PrEP. from the reservoir is modeled as a source-sink relationship RESULTS with a constant rate per lineage. Once a lineage migrates from the reservoir, we assume that it may not circulate back to the Relative to older age groups, YMSM were more likely to be sam- reservoir. pled with recent infection corresponding to higher CD4s and Whereas transmission patterns between age groups may be more frequent RITA-positive test results (Table 1). YMSM had highly nonrandom [36, 37] and transmissions are more likely significantly higher rates of EHI defined as CD4>500 or RITA between people in similar age groups, 2 additional parameters positive test result (41% versus 23%, 2-sample binomial test). were estimated that describe the conditional probability of a YMSM of subtype B were also more likely to reside outside of 1524 • JID 2018:217 (15 May) • Volz et al Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 Table 1. Demographic and Clinical Characteristics of YMSM and OMSM Finally, we estimated the average time that a HIV lineage Included in the Analysis and all MSM in the Database has circulated in the United Kingdom prior to sampling. This was 27.6 years (95% CI, 26.6–28.75), suggesting that most sub- YMSM/B OMSM/B All MSM type B infections in MSM are derived from introductions that (n = 706) (n = 6206) (n = 30 711) occurred in the late 1980s [14]. Year of birth (IQR) 1987(1984–1990) 1971(1964–1977) 1971(1964–1979) e fi Th tted population genetic model provided estimates of CD4 (IQR) 469(337–620) 415(259–583) 420(260–590) RITA+ 34% 25% 28% the number of effective infections through time ( Figure  1 and London 39% 55% 59% Supplementary Figure S1) which approximately corresponds White 84% 88% 85% to the number of MSM living with HIV within the 21 clades Immigrant 19% 26% 30% included in this analysis. Note that these estimates were based Clustered (0.5%) 31% 20% on a subset of all HIV-1 genetic diversity and the absolute num- Clustered (1.5%) 77% 62% ber of infections does not correspond to the total number of YMSM and OMSM count only patients that meet  all inclusion criteria. Statistical tests compare YMSM/B or OMSM/B and all MSM, except for the clustered outcome for which infections in the population. Nevertheless, the estimated rates YMSM and OMSM are compared. Significance levels were determined using a 2-sample t of growth of MSM living with HIV are similar to estimates test for continuous variables and Fishers exact test for binary variables. Entries in bold have a P value < .001. obtained by Public Health England based on surveillance data Abbreviations: IQR, interquartile range; OMSM, older men who have sex with men; (Figure 1B and Supplementary Figure S2). We estimated that the YMSM, young men who have sex with men. 21 clades included samples from 60% of people living with HIV descended from the clades, the remainder not meeting inclusion criteria, not having sequences, or not being diagnosed in the the London metropolitan area and less likely to be born outside United Kingdom. Note that the sample proportion is influenced of the United Kingdom. Relative to OMSM, YMSM were more by the fact that approximately 80% are diagnosed, not all diag- likely to be genetically clustered with at least 1 other patient nosed have a sequence in the database, and approximately 50% (31% versus 20%). More than 60% of both YMSM and OMSM of lineages were excluded due to lack of adequate biomarkers clustered with at least 1 other patient using 1.5% threshold or because they were collected from ART-experienced patients. evolutionary distance and at 0.5% threshold distance about a er Th e was substantial variation in the proportion of each clade quarter of patients clustered. Small but statistically significant that are YMSM, and the proportion of transmissions attribut- associations were found between the odds of clustering at 0.5% able to YMSM in each clade (Figure  1). The proportion of the and 1.5% evolutionary distance and EHI, age, and location of clade that was YMSM was not significantly associated with sampling. Patients sampled with EHI clustered slightly more in growth rates of effective infections in each clade (F test P = .42). multivariate analyses (odds ratio [OR] = 1.14, 0.5% threshold) Simulated PrEP interventions based on the fitted popula - as do YMSM (OR = 1.09, 0.5% threshold). tion genetic model showed large gains from focusing PrEP on Coalescent-based phylodynamic analysis showed strong evi- YMSM in comparison to random allocation to all MSM. Note dence of higher transmission risk for both EHI and YMSM. that, in reality, PrEP would not be allocated randomly, and ran- e t Th ransmission risk ratio of YMSM relative to all other MSM dom allocation should be interpreted as a benchmark rather (OMSM) is 2.52 (95% CI, 2.32–2.73). The transmission risk than a likely outcome or standard of care. Figure  2 shows the ratio of EHI (CD4  >500 and/or RITA positive) relative to all predicted cumulative infections averted by PrEP over 5  years other stages of infection is 3.70 (95% CI, 3.36–4.09). These rep - if 15 000 susceptible individuals were provided PrEP in 2015, resent independent effects, and YMSM with EHI were predicted which was the end point for sequence data included in this to have the highest transmission risk. study. Simulations reflect not only direct impacts from prevent - e p Th hylodynamic analysis also revealed highly nonrandom ing infections in treated individuals, but also indirect effects transmission patterns by age. The probability that the recipient from preventing transmission by individuals who would have is YMSM given an infected donor in the OMSM risk group was become infected without PrEP. We predicted that 749 (636–857) 20.0% (95% CI, 17.7%–22.7%), which is roughly twice the pro- infections would be averted over 5  years if PrEP was focused portion of the population that is YMSM (approximately 10% on YMSM and that 179 (150–207) infections would be averted of MSM by definition). However, the probability that the recip - with random allocation. PrEP for YMSM averted 4.2 times as ient is YMSM given a YMSM donor was very much higher: many infections over 5 years as random allocation. 83.3% (95% CI, 78.4%–87.2%). Consequently, most YMSM were infected by other YMSM and not by older age groups. 75% DISCUSSION of infections in YMSM were attributable to other YMSM, and 87% of infections in OMSM were attributable to other OMSM. e co Th mbination of high levels of transmission and high assor - Age assortativity and transmission patterns are summarized in tativity amplifies incidence in YMSM and PrEP effectiveness in Figure 1. YMSM. PrEP in YMSM averts transmissions that would occur Heterogeneous Transmission Risk of HIV • JID 2018:217 (15 May) • 1525 Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 A C 9.1 0.5 9.0 Clade size / 100 8.9 0.4 6000 4 8.8 8.7 0.3 0 8.6 9.8 10.0 10.2 10.4 1980 1990 2000 2010 05 10 15 20 log MSM living with HIV Clade index Pair probabilities Population attributable fraction of transmissions 17% 7% 83% YMSM 80%35% YMSM 46% Other MSM Other MSM 20% 12% Figure 1. Effective number of infections through time, cumulative sequences sampled through time, and proportion of transmissions attributable to young men who have sex with men (YMSM) for 21 subtype B clades in UK MSM. A, Effective infections and cumulative samples combined for 21 clades. B, Number of MSM living with human immunodeficiency virus (HIV) estimated from surveillance data between 2004 and 2012 [ 39] versus phylodynamic estimates (log-transformed). A  regression with slope constrained to 1 is shown in red. C, The estimated proportion of transmissions attributable to YMSM (age <25) for 21 clades and the number of samples in each clade. D, Estimated transmission patterns between YMSM and other MSM (OMSM). Left: The probability that a recipient is in each age group given that a donor is either YMSM or OMSM. Right: The proportion of all transmissions that are attributable to each pair of age groups. Abbreviation: PAF, population attributable fraction. if YMSM were infected, and because YMSM are more likely to HIV genetic diversity in the United Kingdom showed a clear infect one another, subsequent generations of the epidemic pro- pattern consistent with higher risk of infection in YMSM and cess are further reduced. These transmission patterns yield higher higher risk of transmission EHI. The effect of young age on incidence in YMSM, and it remains to corroborate observations transmission risk remained aer co ft ntrolling for higher rates based on phylodynamic analysis by estimating age-specific inci - of EHI among patients diagnosed at young age. YMSM have dence in UK MSM from nongenetic surveillance data. the hallmarks of a small high-risk core group [40], having both higher intrinsic transmission rates and preferential attachment within the risk group. YMSM show very high levels of assort- ative mixing. Most infections (75%) in YMSM arose via inter- action with other YMSM despite comprising around 10% of the infected MSM population. These findings are consistent with previous reports of higher rates of HIV genetic clustering 500 in YMSM [41, 42]. We find evidence for a modest net flow of Random transmissions from OMSM to YMSM which is in line with the Targetted findings of age-discordant clustering among young MSM found by Wolf et al [28]. These transmission patterns also differ from studies of genetic clustering in heterosexual populations, which have shown much greater age-discordancy within clusters that is hypothesized to arise from net flows of transmission from older males to younger females [43]. 2015 2016 2017 2018 2019 2020 In reality, PrEP will not be allocated randomly, and age may provide one of many criteria for PrEP. Further studies Figure 2. The number of infections averted over a 5-year horizon using a random are warranted to examine how age can be used in combina- allocation of pre-exposure prophylaxis (PrEP) or a targeted PrEP strategy focused on tion with other transmission and infection risk factors. Recent young men who have sex with men. 1526 • JID 2018:217 (15 May) • Volz et al E ective number of infections (line) Cumulative samples (dash) Infections averted log e ective MSM living with HIV PAF YMSM Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 randomized clinical trials have examined the direct protec- Potential coni fl cts of interest. All authors: No reported con- tive effects of PrEP in MSM but have not accounted for indi - flicts of interest. All authors have submitted the ICMJE Form rect transmission effects included in our simulations [ 44, 45]. for Disclosure of Potential Conflicts of Interest. Conflicts that Clinical trials, including studies conducted with UK MSM [46], the editors consider relevant to the content of the manuscript have focused on individuals at higher risk of infection than have been disclosed. MSM as a whole (eg, those with recent sexually transmitted References infection [STI] testing history and condomless sex), so that esti- mates of infections averted are not directly comparable with our 1. Abu-Raddad LJ. Role of acute HIV infection in driving HIV simulation results, which were constrained by the available data transmission: implications for HIV treatment as preven- to focus on random allocation within age groups. tion. PLoS Med 2015; 12:e1001803. This study only examined subtype B sequences, which com - 2. Chemaitelly H, Awad SF, Abu-Raddad LJ. The risk of HIV prise the large majority of sequences among MSM in the United transmission within HIV-1 sero-discordant couples appears Kingdom. Less-prevalent subtypes have different demographic to vary across sub-Saharan Africa. Epidemics 2014; 6:1–9. and clinical characteristics, notably CRF02AG which has higher 3. Quinn TC, Wawer MJ, Sewankambo N, et al. Viral load and proportions of recent African migrants (results not shown), and heterosexual transmission of human immunodeficiency these results may not generalize to those clades. Compared to all virus type 1.  Rakai Project Study Group. N Engl J Med sequences from MSM, the HIV lineages included in this analy- 2000; 342:921–9. sis come from patients that are less likely to reside in London or 4. Boily MC, Baggaley RF, Wang L, et  al. Heterosexual risk be foreign born (Table 1). of HIV-1 infection per sexual act: systematic review and Young age in MSM is a simple and easily identifiable proxy meta-analysis of observational studies. Lancet Infect Dis for transmission risk that may form one of many inputs into 2009; 9:118–29. algorithms for prioritizing or promoting PrEP. Parameters 5. Wawer MJ, Makumbi F, Kigozi G, et  al. Circumcision in estimated in this study may provide useful inputs into more HIV-infected men and its effect on HIV transmission to detailed simulations for designing realistic PrEP interventions female partners in Rakai, Uganda: a randomised controlled [47]. Access to PrEP and knowledge of PrEP may be lower in trial. Lancet 2009; 374:229–37. young MSM [48], providing further impetus to provide benefits 6. Baeten JM, Overbaugh J. Measuring the infectiousness of for that group. YMSM can not be prioritized to the exclusion persons with HIV-1: opportunities for preventing sexual of other age groups, nor should age be the sole criterion used. HIV-1 transmission. Curr HIV Res 2003; 1:69–86. Many risk groups were not considered in this phylogenetic 7. Hollingsworth TD, Pilcher CD, Hecht FM, Deeks SG, Fraser analysis and it is important to have diverse as well as focused C. High transmissibility during early HIV infection among allocation of PrEP [49]. Other variables, such as recent STI test- men who have sex with men-San Francisco, California. ing history or self-reported risk exposures, are likely to also cor- J Infect Dis 2015; 211:1757–60. relate highly with transmission risk, and it remains to examine 8. Volz EM, Ionides E, Romero-Severson EO, Brandt MG, such covariates within a phylodynamics framework. Mokotoff E, Koopman JS. HIV-1 transmission during early infection in men who have sex with men: a phylody- Supplementary Data namic analysis. PLoS Med 2013; 10:e1001568; discussion Supplementary materials are available at e Th Journal of Infectious e1001568. Diseases online. Consisting of data provided by the authors to 9. Marzel A, Shilaih M, Yang WL, et  al.; Swiss HIV Cohort benefit the reader, the posted materials are not copyedited and Study. HIV-1 Transmission during recent infection and are the sole responsibility of the authors, so questions or com- during treatment interruptions as major drivers of new ments should be addressed to the corresponding author. infections in the Swiss HIV Cohort Study. Clin Infect Dis 2016; 62:115–22. Notes 10. Dennis AM, Herbeck JT, Brown AL, et  al. Phylogenetic Acknowledgments. This research was made possible by the studies of transmission dynamics in generalized HIV epi- Steering Committee and Coordinating Centre of the UK Drug demics: an essential tool where the burden is greatest? Resistance Database the Centres contributing data to the UK J Acquir Immune Defic Syndr 2014; 67:181–95. Drug Resistance Database. A full list of contributors is provided 11. Pillay D, Herbeck J, Cohen MS, et  al.; PANGEA-HIV in the online Supporting material. Consortium. PANGEA-HIV: phylogenetics for generalised Funding. This work was supported by the UK Health epidemics in Africa. Lancet Infect Dis 2015; 15:259–61. Protection Research Units in Modeling Methodology and 12. Romero-Severson E, Skar H, Bulla I, Albert J, Leitner T. Timing Sexually Transmitted Infections and by the National Institutes and order of transmission events is not directly reflected in a of Health (grant number R01AI087520). pathogen phylogeny. Mol Biol Evol 2014; 31:2472–82. Heterogeneous Transmission Risk of HIV • JID 2018:217 (15 May) • 1527 Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 13. Wertheim JO, Leigh Brown AJ, Hepler NL, et  al. e Th 28. Wolf E, Herbeck JT, Van Rompaey S, et  al. Short commu- global transmission network of HIV-1. J Infect Dis 2014; nication: phylogenetic evidence of HIV-1 transmission 209:304–13. between adult and adolescent men who have sex with men. 14. Hué S, Pillay D, Clewley JP, Pybus OG. Genetic analy- AIDS Res Hum Retroviruses 2017; 33:318–22. sis reveals the complex structure of HIV-1 transmission 29. Kozlov AM, Aberer AJ, Stamatakis A. ExaML version within defined risk groups. Proc Natl Acad Sci U S A 2005; 3: a tool for phylogenomic analyses on supercomputers. 102:4425–9. Bioinformatics 2015; 31:2577–9. 15. Faria NR, Rambaut A, Suchard MA, et  al. HIV epidemi- 30. Grubbs FE. Sample criteria for testing outlying observa- ology. e e Th arly spread and epidemic ignition of HIV-1 in tions. Ann Math Stat 1950; 21:27–58. human populations. Science 2014; 346:56–61. 31. To TH, Jung M, Lycett S, Gascuel O. Fast dating using 16. Poon AF, Gustafson R, Daly P, et al. Near real-time monitoring least-squares criteria and algorithms. Syst Biol 2016; of HIV transmission hotspots from routine HIV genotyping: 65:82–97. an implementation case study. Lancet HIV 2016; 3:e231–8. 32. Drummond AJ, Suchard MA, Xie D, Rambaut A. Bayesian 17. Le Vu S, Ratmann O, Delpech V, et al. Comparison of clus- phylogenetics with BEAUti and the BEAST 1.7. Mol Biol ter-based and source-attribution methods for estimating Evol 2012; 29:1969–73. transmission risk using large HIV sequence databases [pub- 33. Volz EM. Complex population dynamics and the coalescent lished online ahead of print 20 October,  2017]. Epidemics under neutrality. Genetics 2012; 190:187–201. doi: 10.1016/j.epidem.2017.10.001. 34. Cori A, Pickles M, van Sighem A, et al. CD4+ cell dynam- 18. Dearlove BL, Xiang F, Frost SDW. Biased phylodynamic ics in untreated HIV-1 infection: overall rates, and effects inferences from analysing clusters of viral sequences. Virus of age, viral load, sex and calendar time. AIDS 2015; Evol 2017; 3:vex020. 29:2435–46. 19. Poon AF. Impacts and shortcomings of genetic clustering 35. Volz EM, Romero-Severson E, Leitner T. Phylodynamic methods for infectious disease outbreaks. Virus Evol 2016; inference across epidemic scales. Mol Biol Evol 2017; 2:vew031. 34:1276–88. 20. Volz EM, Koopman JS, Ward MJ, Brown AL, Frost SD. 36. Prah P, Copas AJ, Mercer CH, Nardone A, Johnson AM. Simple epidemiological dynamics explain phylogenetic Patterns of sexual mixing with respect to social, health clustering of HIV from patients with recent infection. PLoS and sexual characteristics among heterosexual couples Comput Biol 2012; 8:e1002552. in England: analyses of probability sample survey data. 21. Volz EM, Koelle K, Bedford T. Viral phylodynamics. PLoS Epidemiol Infect 2015; 143:1500–10. Comput Biol 2013; 9:e1002947. 37. Phillips AN, Cambiano V, Nakagawa F, et al. Increased HIV 22. Volz EM, Frost SD. Sampling through time and phylody- incidence in men who have sex with men despite high levels namic inference with coalescent and birth-death models. of ART-induced viral suppression: analysis of an extensively J R Soc Interface 2014; 11:20140945. documented epidemic. PLoS One 2013; 8:e55312. 23. De Maio N, Wu CH, O’Reilly KM, Wilson D. New routes to 38. NHS England. NHS England announces world’s largest single phylogeography: a bayesian structured coalescent approxi- PrEP implementation trial to prevent HIV infection, 2017. mation. PLoS Genet 2015; 11:e1005421. https://www.england.nhs.uk/2017/08/nhs-england-an- 24. Ratmann O, Hodcroft EB, Pickles M, et al.; PANGEA-HIV nounces-worlds-largest-single-prep-implementation-tri- Consortium. Phylogenetic tools for generalized HIV-1 epi- al-to-prevent-hiv-infection/. Accessed 10 February 2018. demics: findings from the PANGEA-HIV methods compar - 39. Skingsley A, Yin Z, Kirwan P, et al. HIV in the UK–Situation ison. Mol Biol Evol 2017; 34:185–203. Report 2015: data to end 2014. London: Public Health 25. Pineda-Peña AC, Faria NR, Imbrechts S, et  al. Automated England, 2015. subtyping of HIV-1 genetic sequences for clinical and sur- 40. Caldarelli G, Catanzaro M. Networks: A  very short intro- veillance purposes: performance evaluation of the new duction. Oxford: Oxford University Press, 2012. REGA version 3 and seven other tools. Infect Genet Evol 41. Pao D, Fisher M, Hué S, et al. Transmission of HIV-1 during 2013; 19:337–48. primary infection: relationship to sexual risk and sexually 26. Aghaizu A, Murphy G, Tosswill J, et  al. Recent infection transmitted infections. AIDS 2005; 19:85–90. testing algorithm (RITA) applied to new HIV diagnoses in 42. Lubelchek RJ, Hoehnen SC, Hotton AL, Kincaid SL, Barker England, Wales and Northern Ireland, 2009 to 2011. Euro DE, French AL. Transmission clustering among newly Surveill 2014; 19:20673. diagnosed HIV patients in Chicago, 2008 to 2011: using 27. Wensing AM, Calvez V, Günthard HF, et al. 2015 update of phylogenetics to expand knowledge of regional HIV trans- the drug resistance mutations in HIV-1. Top Antivir Med mission patterns. J Acquir Immune Defic Syndr 2015; 2015; 23:132–41. 68:46–54. 1528 • JID 2018:217 (15 May) • Volz et al Downloaded from https://academic.oup.com/jid/article/217/10/1522/4915980 by DeepDyve user on 19 July 2022 43. de Oliveira T, Kharsany AB, Gräf T, et al. Transmission net- (PROUD): effectiveness results from the pilot phase of works and risk of HIV infection in KwaZulu-Natal, South a pragmatic open-label randomised trial. Lancet 2016; Africa: a community-wide phylogenetic study. Lancet HIV 387:53–60. 2017; 4:e41–50. 47. Cambiano V, Miners A, Dunn D, et al. Cost-effectiveness 44. Grant RM, Lama JR, Anderson PL, et al.; iPrEx Study Team. of pre-exposure prophylaxis for HIV prevention in men Preexposure chemoprophylaxis for HIV prevention in men who have sex with men in the UK: a modelling study who have sex with men. N Engl J Med 2010; 363:2587–99. and health economic evaluation. Lancet Infect Dis 2018; 45. Buchbinder SP, Glidden DV, Liu AY, et  al. HIV pre-ex- 18:85–94. posure prophylaxis in men who have sex with men and 48. Bauermeister JA, Meanley S, Pingel E, Soler JH, Harper transgender women: a secondary analysis of a phase 3 ran- GW. PrEP awareness and perceived barriers among single domised controlled efficacy trial. Lancet Infect Dis 2014; young men who have sex with men. Curr HIV Res 2013; 14:468–75. 11:520–7. 46. McCormack S, Dunn DT, Desai M, et  al. Pre-exposure 49. e L Th ancet HIV. Better late than never: PrEP in England. prophylaxis to prevent the acquisition of HIV-1 infection Lancet HIV 2017; 4:e1. Heterogeneous Transmission Risk of HIV • JID 2018:217 (15 May) • 1529

Journal

The Journal of Infectious DiseasesOxford University Press

Published: Apr 23, 2018

Keywords: hiv; hiv-1; infections; men who have sex with men; pre-exposure prophylaxis; phylogeny

There are no references for this article.