Phylogenetic prioritization of HIV-1 transmission clusters with viral lineage-level diversification rates

Abstract

Background and objectives Public health officials faced with a large number of transmission clusters require a rapid, scalable, and unbiased way to prioritize distribution of limited resources to maximize benefits. We hypothesize that transmission cluster prioritization based on phylogenetically-derived lineage-level diversification rates will perform as well as or better than commonly used growth-based prioritization measures, without need for historical data or subjective interpretation. Methodology 9822 HIV pol sequences collected during routine drug resistance genotyping were used alongside simulated sequence data to infer sets of phylogenetic transmission clusters via patristic distance threshold. Prioritized clusters inferred from empirical data were compared to those prioritized by the current public health protocols. Prioritization of simulated clusters was evaluated based on correlation of a given prioritization measure with future cluster growth, as well as the number of direct downstream transmissions from cluster members. Results Empirical data suggest diversification rate-based measures perform comparably to growth-based measures in recreating public heath prioritization choices. However, unbiased simulated data reveals phylogenetic diversification rate-based measures perform better in predicting future cluster growth relative to growth-based measures, particularly long-term growth. Diversification rate-based measures also display advantages over growth-based measures in highlighting groups with greater future transmission events compared to random groups of the same size. Furthermore, diversification rate measures were notably more robust to effects of decreased sampling proportion. Conclusions and implications Our findings indicate diversification rate-based measures frequently outperform growth-based measures in predicting future cluster growth and offer several additional advantages beneficial to optimizing the public health prioritization process. Lay Summary Public health officials faced with more transmitting groups than can be immediately addressed require an efficient way to prioritize limited resources. We demonstrate that phylogenetic approximations of transmission derived from viral sequence data offer advantages over other common methods of prioritizing clusters, while meeting and frequently exceeding existing performance standards.

Optimized phylogenetic clustering of HIV-1 sequence data for public health applications

“…To determine whether clusters that were prioritized under a ΔAIC-optimized method reflected accurately the actual distribution of incident cases, we simulated an epidemic on a network using FAVITES [ 68 ] under an HIV-1 transmission model [ 72 ] that was calibrated to the heterogeneous, localized HIV-1 epidemic in British Columbia, Canada [ 69 ]. This configuration simulated the spread of HIV-1 over ten years in a population of 26,746 individuals who were connected by a preferential attachment contact network [ 70 ], starting from 3,110 infections at time 0.…”

“…Next, an epidemic was seeded with 3,110 infections in a population of 26,746 individuals, and propagated for 10 simulation-years through the network using the program GEMFsim [ 71 ] under the HIV-1 transmission model described by Granich and colleagues [ 72 ]. This model was parameterized so that HIV-1 transmission occurred predominantly from untreated individuals at an early stage of infection [ 69 ]. Infected individuals are sampled upon initiating ART.…”

“…To simulate a virus phylogeny from the resulting transmission tree, virus population dynamics within each host individual was simulated under a logistic coalescent model. Finally, HIV-1 sequence evolution was simulated along the virus phylogeny under a generalized time-reversible model of nucleotide substitution that was previously parameterized to actual HIV-1 pol sequences [ 69 ]. These sequences were labelled by sampling times in simulation time units, which we converted to standard dates.…”

“…To evaluate whether subtree clusters extracted under our optimized criteria actually represent epidemiologically-related infections, i.e., whether the incident sequences are being correctly grafted onto known clusters, we used FAVITES (version 1.2.10) [68] to simulate an epidemic in a network and to sample a transmission tree and simulate sequence evolution. We used the same configuration as a previous study of using phylogenetic clustering to prioritize HIV prevention measure [69]. In brief, FAVITES was configured to generate a random network under the Baraba ´si-Albert [70] preferential attachment model, in which highly connected nodes tend to accumulate more connections.…”