Inferring human immunodeficiency virus 1 proviral integration dates with Bayesian Inference

Human immunodeficiency virus 1 (HIV) proviruses archived in the persistent reservoir currently pose the greatest obstacle to HIV cure due to their evasion of combined antiretroviral therapy and ability to reseed HIV infection. Understanding the dynamics of the HIV persistent reservoir is imperative for discovering a durable HIV cure. Here, we explore Bayesian methods using the software BEAST2 to estimate HIV proviral integration dates. We started with within-host longitudinal HIV sequences collected prior to therapy, along with sequences collected from the persistent reservoir during suppressive therapy. We built a BEAST2 model to estimate integration dates of proviral sequences collected during suppressive therapy, implementing a tip date random walker to adjust the sequence tip dates and a latency-specific prior to inform the dates. To validate our method, we implemented it on both simulated and empirical data sets. Consistent with previous studies, we found that proviral integration dates were spread throughout active infection. Path sampling to select an alternative prior for date estimation in place of the latency-specific prior produced unrealistic results in one empirical data set, while on another data set the latency-specific prior was selected as best-fitting. Our Bayesian method outperforms current date estimation techniques with a root mean squared error of 0.89 years on simulated data relative to 1.23–1.89 years with previously developed methods. Bayesian methods offer an adaptable framework for inferring proviral integration dates.Funding provided by: Canadian Institutes of Health ResearchCrossref Funder Registry ID: http://dx.doi.org/10.13039/501100000024Award Number: PJT-159625Funding provided by: National Institutes of HealthCrossref Funder Registry ID: http://dx.doi.org/10.13039/100000002Award Number: NIHR21A12702