Mapping Brain Changes in HIV/AIDS

This 3-year project will provide enormous insight into HIV/AIDS. Knowledge of how HIV affects the brain is sorely lacking. Working with members of the UCLA Center for Computational Biology (CCB), we will develop a multidimensional, computational atlas of HIV/AIDS, including the most powerful tools to track HIV disease progression in the brain. We will investigate how specific factors affect the rate of brain degeneration, such as anti-retroviral therapy, duration of illness, age and gender, and viral load. We will relate brain changes over time to immune system measures, such as CD4+ T-cell counts. This effort will create the computational tools to study how deficits in HIV patients emerge and spread in the living brain. We will test novel tools for large-scale analysis and mapping of brain change (Aims 1 and 2), novel tools to correlate these brain changes with immune deterioration and imminent clinical decline (Aim 3), and methods to gauge and optimize the power of these tools for large-scale drug trials (Aims 4 and 5). Our new computational tools will map 3D dynamic changes in serial brain MRI scans. We will link detected brain changes to serum measures (T-cell decline or recovery) and to treatment parameters (e.g., whether the patient is using HAART therapy or not). We will relate changes to clinical data from detailed neuropsychological exams. We will examine 100 AIDS patients with a broad range of disease severity, and 100 healthy controls matched for HIV-associated risk factors (see Experimental Design). The project will be very efficient as these scans have already been collected – they are stored in the LONI/CCB database, and urgently await analysis.

Specific Aims

  • Aim 1: Map Disease Progression in the Cortex. We will map the 3D profile of cortical thickness in 100 HIV/AIDS patients and 100 matched controls, scanned twice with high-resolution MRI at a 2-year interval. We will create average maps of gray matter thickness for each subject group at each time point, and analyze changes statistically to localize regions with significant changes. We will answer the following questions: How do cortical brain deficits progress in HIV/AIDS? Are some brain systems protected, and to what degree? Is gray matter in the motor cortices selectively damaged? Hypothesis: In HIV/AIDS, cortical deficits will progress from ensorimotor and adjacent association cortices, to pervade the entire cortex late in the illness.
  • Aim 2: Map Rates of Structural Brain Change Over Time. Using 3D MRI data acquired longitudinally from 100 HIV/AIDS patients and 100 matched controls over a 2-year interval, we will determine a 3D map of the mean rate of atrophy in each subject group. We will create 3D statistical maps to answer the following questions: What is the pattern of brain change (volume loss) over time? Which brain regions change the most? Can tensor-based morphometry (TBM), based on fluid image registration techniques developed at the CCB, identify systematic changes in a group of patients? How do changes in deep gray matter structures – the caudate, putamen, and hippocampus – correlate with the cortical changes mapped in Aim 1? Hypothesis: Deep nuclei will show greatest deficits early in the illness, as they have the highest viral load. We expect significant longitudinal changes in the supplementary motor cortex, corpus callosum and ventricles, based on our cross-sectional studies. Over time, white matter atrophy will spread radially outwards to the cortex as CD4+ T-cell counts and neuropsychological function decline.
  • Aim 3: Links to Declining Cognition, Immune Function, and Viral Load. We will correlate the brain changes mapped in Aim 2 and 3 with clinical data measured at the time of the baseline and follow-up scans on neuropsychiatric test performance, CD4+ T-cell counts, viral load, duration of illness, and medication status. We will answer the following questions: Are brain changes mapped with MRI linked with declining levels of CD4+ T-cells, the commonest biomarker of HIV disease progression? Does brain change depend on whether the patient is on HAART therapy? Are brain changes correlated with viral load, measured over a 2- year interval in patients? In which specific brain regions are these processes linked? Do deficits, detected in the initial MRI scan, predict future cognitive decline, ascertained by tests of psychomotor speed, executive function, and memory performance? Hypothesis: Dynamic rates of tissue atrophy are expected to link with T-cell counts, but not with viral load, which can fluctuate dramatically over relatively short intervals.
  • Aim 4: Compute Power Estimates for Drug Trials
    Using the neuroimaging measures from Aims 1 and 2, we will compute how many subjects would be needed in a drug trial to detect a 20%, 10% or 5% slowing of disease progression, over a 2-year follow-up interval. We will also determine how much better the automated technique is (in statistical power) than conventional morphometry, where regions are hand-traced on brain images and their volumes assessed? Hypothesis: We expect the cortical maps in Aim 1 to afford greater power per subject than the fluid maps of brain change in Aim 2, but the fluid maps should provide higher throughput and greater automation, i.e. greater power per unit processing time in a drug trial.
  • Aim 5: Algorithm Optimization. We will assess whether two alternative choices of algorithm parameters can further increase our power to detect the HIV effects on the brain. Using False Discovery Rate (FDR) plots to compare detection power, we will assess the influence of the image registration “cost function” (Jensen-Renyi divergence versus Mutual Information) for recovering brain changes mapped in Aims 2 and 3. We expect that the Jensen-Rényi divergence will outperform Mutual Information, which is the current standard in the field. Second, for detecting brain changes in HIV/AIDS, we will compare Lie group methods – which analyze the full tensor-valued information – versus standard univariate statistical methods. Hypothesis: We predict that the Lie group methods will outperform standard statistics; our preliminary data support this hypothesis and detect HIV effects with greater power; Chiang et al., 2006, Lepore et al., 2006, in Appendix). This final step will optimize the processing sequence for use in drug trials.