Machine Learning Bolsters Evidence That D1, Nef, and Tat Influence HIV Reservoir Dynamics

Area of Science:

• Virology
• Immunology
• Genetics

Background:

• The establishment of a persistent HIV reservoir early in infection is the primary obstacle to achieving a cure.
• Understanding the evolutionary dynamics of proviral sequences within the reservoir is crucial for developing novel treatment strategies.

Purpose of the Study:

• To characterize the decay dynamics of the intact HIV reservoir using longitudinal sequence data.
• To identify specific HIV genomic elements associated with proviral decay and persistence using machine learning.

Main Methods:

• Analysis of longitudinal near full-length HIV-1 sequences from individuals on antiretroviral therapy.
• Application of Bayesian mixed-effects models to assess single-phase and multiphasic decay models.
• Development of a machine-learning logistic regression model to identify genomic correlates of proviral decay and persistence.

Main Results:

• Biphasic decay models provided a significantly better fit for intact HIV reservoir dynamics than single-phase models.
• Estimated half-lives for the intact reservoir were 18.2 months (fast phase) and 433 months (slow phase).
• Machine learning identified the splice donor site D1 as a key element associated with reservoir contraction, while HIV nef and tat were linked to T-cell protection and clonal expansion, respectively.

Conclusions:

• The long-lived HIV reservoir comprises at least two distinct compartments with differing decay rates.
• Specific HIV genomic elements exert opposing forces, influencing reservoir contraction (e.g., D1) and persistence/expansion (e.g., nef, tat).
• These findings provide critical insights into the complex forces shaping the HIV reservoir over time and suggest potential therapeutic targets.