Despite the use of combination antiretroviral therapy, many HIV associated conditions, such as HIV associated neurocognitive disorders and frailty still exist in people living with HIV (PLWH). A potential biomarker reflective of these conditions is resting state functional connectivity (rsFC). Changes in rsFC strength have been hypothesized to reflect a compensatory reaction due to damage caused by persistent inflammation and chronic immune activation. Within a large cohort of PLWH and HIV- controls we identified networks most affected over the life span of HIV infection using machine learning methods.
A total of 538 rsFC scans from 318 PLWH (mean age 47.2y, 77% male, 31% Caucasian, mean duration of infection 12.8y +/-9.4, 84% viral load <200) collected from studies at Washington University School of Medicine (WUSM) and 2791 scans from 2133 HIV- controls (mean age 44.4y, 42% male, 69% Caucasian) collected from studies at WUSM and other sources were analyzed. Ages ranged from 20 to 70 years old (figure 1b). Ten rsFC networks were evaluated, and preprocessing was performed using in house methods. Correlation matrices were generated for all participants, and an average correlation matrix was computed for each year of age for both groups. A Relief feature selection algorithm was used to identify the strongest predictive networks of HIV status. We then evaluated which networks showed significantly different trajectories with respect to aging among PLWH and controls.
The Relief algorithm identified the strongest predictors of HIV status as multiple connections between the somatomotor, cingulo-opercular, and dorsal attention networks. The strongest difference in average connectivity strength as a function of aging between PLWH and controls were between the dorsal attention and both the fronto-parietal (figure 1a) and cingulo-opercular networks (R2 = -.82, -.52, P<.01), as well as the ventral attention and somatomotor networks (R2 = -.38, P<.01).
Our data suggest changes in rsFC occurs as a result of HIV infection, and are primarily associated with motor, control, and attention. Further, changes exhibit specific temporal patterns between networks which is independent of reorganization that occurs due to normal aging, and these changes begin around midlife. These results will allow for a more tailored treatment approach for PLWH, and should be considered when conducting clinical trials.