Timing HIV infection with a simple and accurate population viral dynamics model.

Daniel B Reeves,E Fabian Cardozo-Ojeda,Joshua T Schiffer,Morgane Rolland,Bryan T Mayer,Peter Gilbert,Yifan Li,Bethany L Dearlove,Merlin L Robb

doi:10.1098/rsif.2021.0314

Abstract

Clinical trials for HIV prevention can require knowledge of infection times to subsequently determine protective drug levels. Yet, infection timing is difficult when study visits are sparse. Using population nonlinear mixed-effects (pNLME) statistical inference and viral loads from 46 RV217 study participants, we developed a relatively simple HIV primary infection model that achieved an excellent fit to all data. We also discovered that Aptima assay values from the study strongly correlated with viral loads, enabling imputation of very early viral loads for 28/46 participants. Estimated times between infecting exposures and first positives were generally longer than prior estimates (average of two weeks) and were robust to missing viral upslope data. On simulated data, we found that tighter sampling before diagnosis improved estimation more than tighter sampling after diagnosis. Sampling weekly before and monthly after diagnosis was a pragmatic design for good timing accuracy. Our pNLME timing approach is widely applicable to other infections with existing mathematical models. The present model could be used to simulate future HIV trials and may help estimate protective thresholds from the recently completed antibody-mediated prevention trials.

Highlights

A key challenge for HIV prevention trials is dating the exposure that led to breakthrough infection
Infection is difficult to study in practice; even if prospective sampling were available, HIV RNA is not detectable in blood during early HIV infection and not all participants can accurately point to potential recent exposure events
We assumed that HIV infection begins with an infecting exposure, which is the target time of estimation

Summary

Introduction

A key challenge for HIV prevention trials is dating the exposure that led to breakthrough infection. The estimation of infection time subsequently allows the inference of the concentration of the protective agent at exposure, which is critical to understanding why HIV acquisition was not prevented. To estimate the time of infection—or the eclipse phase, the period between HIV acquisition and first detectable viral load—a model or inference technique is required. Several use viral sequence data and evolutionary models to trace time back to the founder sequence [1,2,3,4]. Mathematical models of viral load have been used for timing HIV infection [5] (and SARS-CoV-2 [5]). Still other approaches apply diagnostic windows leveraging Fiebig staging [6] and prior knowledge of eclipse-phase duration [7,8]. Combinations of some of these approaches have been integrated into a comprehensive statistical framework [9]

Methods

Results

Conclusion