Tensor-valued diffusion MRI detects brain microstructure changes in HIV infected individuals with cognitive impairment

Abstract Despite advancements, the prevalence of HIV-associated neurocognitive impairment remains at approximately 40%, attributed to factors like pre-cART (combination antiretroviral therapy) irreversible brain injury. People with HIV (PWH) treated with cART do not show significant neurocognitive changes over relatively short follow-up periods. However, quantitative neuroimaging may be able to detect ongoing subtle microstructural changes. This study aimed to investigate the sensitivity of tensor-valued diffusion encoding in detecting such changes in brain microstructural integrity in cART-treated PWH. Additionally, it explored relationships between these metrics, neurocognitive scores, and plasma levels of neurofilament light (NFL) chain and glial fibrillary acidic protein (GFAP). Using MRI at 3T, 24 PWH and 31 healthy controls underwent cross-sectional examination. The results revealed significant variations in b-tensor encoding metrics across white matter regions, with associations observed between these metrics, cognitive performance, and blood markers of neuronal and glial injury (NFL and GFAP). Moreover, a significant interaction between HIV status and imaging metrics was observed, particularly impacting total cognitive scores in both gray and white matter. These findings suggest that b-tensor encoding metrics offer heightened sensitivity in detecting subtle changes associated with axonal injury in HIV infection, underscoring their potential clinical relevance in understanding neurocognitive impairment in PWH.


Introduction
Human immunode ciency virus (HIV) in ltrates immune system cells and crosses the blood-brain barrier (BBB) shortly after seroconversion and leads to brain injury (1).This in ltration triggers a cascade of effects, including axonal disruption, myelin loss, astrogliosis, and to a lesser extent, (2).Approximately 50% of people with HIV (PWH) may experience mild cognitive impairment.These de cits can affect cognitive domains such as executive function, attention, ne motor skills, and information processing speed (2)(3)(4).
Despite the adoption of combination antiretroviral therapy (cART), chronic mild neuroin ammation is believed to be the primary reason for HIVassociated cognitive impairment.The key contributors to neuroin ammation are activated microglia and perivascular macrophages, with some involvement from astrocytes.An additional contribution is the transmigration to the central nervous system (CNS) of activated monocytes, which, after differentiation, increase the pool of perivascular macrophages (1,(5)(6)(7)(8).As cART becomes more accessible, the understanding of brain abnormalities and cognitive de cits in HIV patients has become increasingly complex.Aging individuals receiving cART may develop comorbid medical conditions that independently lead to brain damage and cognitive changes.Furthermore, certain antiretroviral regimens have been associated with brain damage, complicating treatment strategies (9).Hence, there is an urgent need for advanced techniques to deepen our understanding of the pathogenesis of tissue changes in the brain due to HIV infection.Conventional approaches to evaluate neuronal injury involve serum or cerebrospinal uid markers.However, these methods entail invasive procedures and lack the exceptional spatial resolution offered by magnetic resonance imaging (MRI).As a result, there is a growing interest in sensitive, reliable, readily accessible, and reproducible noninvasive imaging approaches for evaluating the brain injury.Although MRI doesn't offer information at a cellular level, it enables the characterization of various biophysical tissue properties tightly coupled with neuroin ammatory processes.
Advanced MRI pulse sequences and post-processing methods provide novel quantitative measures re ective of brain injury (10)(11)(12)(13)(14). Utilizing micrometer-scale displacement of tissue water, diffusion MRI (dMRI) can noninvasively detect microstructural abnormalities in the brain (15)(16)(17).It provides excellent sensitivity to microstructural damage associated with HIV (18-21).However, conventional read-outs are signi cantly impacted by the dispersion of regional ber orientations, such as crossing bers, posing challenges in detecting regional pathology.This issue is not con ned to standard diffusion tensor imaging (DTI) but extends to more advanced metrics when the effects of ber orientation dispersion are not considered.For instance, the fractional anisotropy (FA) of white matter in DTI is closely linked to densely packed and myelinated axonal structures, as well as the presence of glial cells in disease (22,23).However, interpreting the FA is challenging due to the blending of mesoscopic tissue features (e.g., ber orientation dispersion and crossings) with microscopic features (e.g., axons, cells, and density).These complexities may result in FA changes misinterpreted as pathology (24).Given that 90% of white matter voxels involve crossing bers, the imperfect alignment of axonal bers makes it nearly impossible to separate tissue microstructural anisotropy from macrostructure using FA (25-28).For example, increased FA in Alzheimer's disease was attributed to changes in ber orientation dispersion rather than microscopic anisotropy alterations (29,30).
Tensor-valued diffusion encoding is a new technique which employs diffusion encoding in multiple directions before image readout.While encoding in a single direction -as is done for DTI -yields linear tensor encoding (LTE), encoding in all directions with equal sensitivity yields spherical tensor encoding (STE).By contrasting LTE and STE, additional information about the tissue microstructure can be obtained, such as the separation of microscopic anisotropy and orientation dispersion (31).While a similar objective has been de ned for many modelling methods using LTE, such methods are prone to bias due to modeling degeneracy (32).Unlike DTI, where the interpretation of FA relies on both microscopic features and the bulk tract orientation dispersion, b-tensor encoding separates these effects through diffusional variance decomposition (33)(34)(35).This approach enables the assessment of axonal integrity by measuring microscopic fractional anisotropy (µFA) as well as isotropic and anisotropic diffusional variance (MKi and MKa) at the sub-voxel level.Thus, b-tensor encoding measures may emerge as a sensitive biomarker for evaluating brain microstructure (both gray and white matter) in vivo (36, 37).To date, b-tensor encoding has been used to assess microstructural abnormalities in several diseases (34,(38)(39)(40).
In addition to advanced MRI, plasma levels of neuro lament light (NFL) chain and glial brillary acid protein (GFAP) serve as uid biomarkers of brain injury.Although higher concentrations of NFL and GFAP are measured in cerebrospinal uid, recent advances have increased the sensitivity of the assays making their assessment possible in peripheral blood samples.
The main goals of this work were to investigate whether b-tensor encoding metrics show better sensitivity to HIV infection compared to metrics from conventional DTI, and if they correlate with cognitive performance and blood markers.

Study Subjects
Twenty-four PWH (age = 55±10 years, male/female = 17/7) and 31 matched healthy controls (HC) (age = 55±15 years, male/female = 24/7) were enrolled from Rochester NY, and vicinity area.The Institutional Research Subjects Review Board (RSRB) at the University of Rochester thoroughly reviewed and approved the study.All participants provided written informed consent prior to enrollment and underwent clinical, laboratory, neurocognitive, and brain MRI examinations.All experiments were conducted in accordance with relevant guidelines and regulations.Detailed baseline demographics are presented in Table 1.Our previous report (41), provides detailed descriptions of the inclusion and exclusion criteria as well as all study procedures.To brie y summarize, PWH meeting inclusion criteria had stable cART for a minimum of 3 months before screening and were aged ≥ 18. Exclusions encompassed individuals with symptomatic cardiovascular diseases (angina, myocardial infarction, stroke, or other peripheral atherosclerotic disease) and uncontrolled vascular risk factors such as diabetes mellitus and hypertension.Additionally, those with severe premorbid or comorbid psychiatric disorders (schizophrenia, bipolar disorder, active depression), brain infections other than HIV-1, space-occupying brain lesions, dementia from any cause, and metallic implants were excluded.The control population differed from PWH based on HIV status, level of education and race.

Neuropsychological assessments
Assessments of neurocognitive and functional performance were performed in all subjects.Before conducting analyses, Z-scores were computed for each cognitive domain, including a total Z-score for each participant.
All MR images underwent thorough inspection for artifacts, including motion, geometric distortion, and signal dropout.T1w images underwent structural segmentation using the anatomical processing script (fsl_anat) from FMRIB's Software Library (FSL) (49).The processing pipeline involved image reorientation and cropping, radio-frequency bias-eld correction, linear and nonlinear registration to MNI 2mm standard space through FLIRT and FNIRT, brain extraction via BET(50), tissue segmentation using FAST, and subcortical structure segmentation employing the FIRST algorithm.Lesion segmentation was carried out using volBrain, an automated online MRI brain volumetry system(51), based on T1w and FLAIR images.
Tensor-valued or b-tensor encoding: NIfTI images were prepared for compatibility with the multidimensional diffusion MRI framework (52) available at https://github.com/markus-nilsson/md-dmri.Brie y, the diffusion-weighted images from each participant underwent a three-step processing approach: 1) Correction for eddy current-induced distortion and inter-volume subject motion was achieved by registering the images to an extrapolated reference (53) using ElastiX (54).The use of extrapolation-based references is crucial for accurate registration of high b-value images.2) Smoothing of the images was carried out using a 3D Gaussian kernel with a standard deviation of 0.4 voxels.3) Voxel-by-voxel normalized anisotropic, isotropic, and total diffusional variance (MKa, MKi, MKt), as well as microscopic anisotropy (µFA), were obtained through linear least squares tting of the log signal while correcting for heteroscedasticity.
DTI: The DWI were corrected for eddy current-induced distortion, inter-volume subject motion, and susceptibility-induced distortion using "topup" and "eddy" tools in FSL (55,56).DTI metrics such as FA and MD were computed by using DTIFIT in FSL (49).
ROI analysis: For pre-speci ed regions of interest (ROIs), we calculated average ROI values for all MRI metrics (DTI -FA, MD; b-tensor encoding -µFA, MKi, MKa, MKt).We used Harvard-Oxford cortical and subcortical, and the Johns Hopkins University WM (JHU-WM) atlases available in FSL in standard MNI152-2 mm space for ROI extraction.Prior to this, all MRI metrics were registered to the high-resolution T1w images of the same individual using a 12-DOF linear registration (FLIRT tool in FSL).Then, individuals' T1w images were spatially normalized to the MNI-T1-152 standard template using nonlinear registration (using ANTs) (48, 57).The transformation matrix and the warping eld from these two steps were applied to DTI, and b-tensor encoding metrics.We then extracted mean values from the MRI metrics for the followings: global white matter (GWM), cortical gray matter (CGM), subcortical gray matter (SGM) using the corresponding masks as ROIs, and four white matter tracts encompassing coherent, crossing, and fanning bers: Genu of Corpus Callosum (GCC), Anterior Corona Radiata (ACR), Forceps Minor (FMin), Superior Fronto-Occipital Fasciculus (SFOF).

Statistical Analyses
Statistical analyses were performed in Python (version 3.7.4)and MATLAB (version R2022b).An unpaired t-test was used to compare the differences in ROIs between the two cohorts.Spearman correlation analyses were performed to nd the associations between imaging metrics and blood markers and cognitive scores after controlling for age.A p-value of < 0.05 was considered statistically signi cant for a single hypothesis testing problem.For inferential problems that involved multiple hypotheses, the Benjamini-Hochberg multiple testing procedure was used to control the false discovery rate (FDR) at the < 0.05 level (58).

Participant Characteristics
Detailed information about demographic, clinical, neurocognitive and MRI data of the study participants are presented in Table 1.The Welch's Two Sample t-test did not reveal any statistically signi cant age difference between the HC and PWH (p = 0.947).Nonetheless, we incorporated age as a covariate in all multivariate regression analyses to mitigate any lingering confounding effects.Furthermore, in comparison to PWH, those who were HC exhibited signi cantly higher education levels and were to a higher degree Caucasians (p < 0.001).
Group comparisons of MRI metrics: Fig. 1 represents mean voxel-by-voxel b-tensor encoding and DTI maps from a PWH subject.Our analysis revealed signi cant differences in b-tensor metrics in several white matter ROIs, while no signi cant ndings were observed for DTI metrics (Table 2).In Fig. 2, we illustrate the comparisons between PWH and HC cohorts across various white matter tracts, encompassing coherent, crossing, and fanning bers.
Notably, we observed a signi cant decrease in µFA (p = 0.042 for GCC, p = 0.002 for FMin, p = 0.007 for ACR, p = 0.042 for SFOF) and MKa (p = 0.042 for GCC, p = 0.006 for FMin, p = 0.007 for ACR, p = 0.049 for SFOF), along with a signi cant increase in MKi (p = 0.027 for SCC, p = 0.005 for FMin, p = 0.001 for ACR, p = 0.034 for SFOF) among PWH.Although FA exhibited a decrease in PWH, it did not reach statistical signi cance.We also found increased MD in PWH cohort but not signi cant (not shown).The trend of changes in DTI metrics is consistent with previous works (19,(59)(60)(61)(62)(63).For example, several previous studies reported that PWH had a decreased FA in several brain regions, including genu and splenium of corpus callosum (GCC, SCC), and SFOF (19,(59)(60)(61)(62)(63).Although PWH had a decreased µFA and MKa and increased MKi compared to healthy controls in global ROIs i.e., for GWM, CGM and SGM, none of the metrics exhibited signi cant difference.

Group comparisons of cognitive performance and blood markers
Welch's two group t-test showed the total cognitive score was lower in the PWH cohort compared to the HC cohort (t = 2.22, p = 0.030).However, while the average concentrations of NFL and GFAP were slightly elevated in the PWH cohort compared to the HC cohort, these differences did not reach statistical signi cance.

Relationship between cognitive scores and b-tensor metrics
We investigated the correlation between total cognitive z-scores and b-tensor encoding-based µFA, MKi, and MKa in global white matter (WM), subcortical gray matter (SGM), and cortical gray matter (CGM), among both PWH and HC individuals (see Fig. 3).Signi cant relationships were observed between total cognitive z-scores and b-tensor metrics in the PWH cohort, while no statistical signi cance was found for HC subjects except for MKa in SGM.Additionally, no signi cant associations were found for FA, except for CGM in PWH.Correlations between total cognitive scores and btensor encoding metrics in four white matter tracts, which involve coherent, crossing, and fanning bers (GCC, ACR, FMin, and SFOF), are presented in Supplementary Fig. 1.Similar trends of changes were identi ed within those ROIs.
Further, we conducted correlation analyses between cognitive domain scores (i.e., Attention/Working Memory, Speed of Information Processing, Executive Function, Language, Learning, Memory, and Motor Skills) and b-tensor metrics, as well as DTI metrics, speci cally µFA and FA for global ROIs (see Supplementary Table 1).Our ndings suggest that executive function, attention, and motor skills display increased sensitivity to microstructural tissue changes measured by b-tensor encoding compared to DTI metrics.The trend of correlations aligns with previous studies involving DTI-derived FA.
We also performed two-way ANOVA to measure the effects of HIV status, MRI metrics of ve ROIs (such as GWM, CGM and SGM, GCC and ACR) and their interactions on cognitive scores.Table 3 shows the representative results for total cognitive scores while Supplementary Tables 2-4 are for subdomains (such as executive function, attention, and motor functions).Relationship between blood markers and b-tensor metrics: Fig. 4 illustrates the associations between average neuro lament light chain (NFL) concentrations, as well as GFAP with b-tensor and DTI metrics.NFL concentrations showed a negative correlation with µFA and MKa while being positively correlated in WM, and signi cance is mostly found in PWH subjects (p < 0.05).GFAP also shows similar trends.We did not nd any signi cant interactions between HIV status and MRI metrics with blood markers.

Discussion
This is the rst study to apply diffusion MRI with b-tensor encoding in the context of HIV-associated neuropathology to better understand the underlying brain tissue microstructure and to investigate the association between b-tensor metrics and cognitive performance and blood markers of brain injury.Tensor-valued diffusion encoding proves valuable in unraveling orientation dispersion and sub-voxel anisotropy, surpassing the capabilities of conventional diffusion techniques like DTI, as it increases the amount of microstructure information encoded into the diffusion-weighted images (32).Our hypothesis posits that axonal injury would associate with elevated plasma levels of NFL and GFAP and lower cognitive performance in PWH.Our ndings reveal that a) b-tensor encoding metrics (µFA, MKa, MKi, MKt) demonstrate stronger sensitivity to microstructural changes in the brain attributed to HIV infection than DTI metrics (FA and MD); b) b-tensor encoding metrics are signi cantly associated with cognitive scores in PWH but not with FA and MD; and c) b-tensor encoding metrics in white matter are signi cantly associated with blood markers such as GFAP and NFL in PWH .
In alignment with observations in other neuroin ammatory and neurodegenerative disorders (24,(64)(65)(66), PWH exhibit a reduction in anisotropy-related metrics (FA, µFA, MKa) and an elevation in diffusivity-related metrics (MD, MKi) compared to their healthy counterparts.This indicates a widespread loss of tissue microstructural integrity and possible edema.The mean values for diffusion metrics (both DTI and b-tensor encoding) align with previous studies involving both healthy and diseased subjects (67).
Given that both FA and µFA serve as indices of diffusion anisotropy, their comparability is noteworthy.Mean FA values typically ranged from 0.39 to 0.48 in white matter and 0.22 to 0.28 in gray matter, consistently lower than mean µFA values, which fell in the range of 0.49 to 0.76 in white matter and 0.44 to 0.64 in gray matter, respectively.This discrepancy is likely attributed to the presence of crossing and fanning bers in brain tissues, which attenuate FA measurements without affecting µFA.The results suggest that µFA, along with other b-tensor encoding metrics, serves as a more sensitive and speci c measure for detecting microstructural changes compared to FA.
Our ndings indicate signi cant reductions by 10-15% in µFA values in various white matter regions, speci cally in coherent (genu of corpus callosum), crossing (SFOF), and fanning (ACR) bers, while FA showed up to 3% non-signi cant changes, in PWH compared to healthy controls.This implies that microstructural changes, as measured by µFA, are predominantly due to the loss of local anisotropy rather than disruption of white matter ber coherence in the HIV cohort.Since µFA is proposed as a measure of axonal integrity rather than myelin (23,68), this decrease in µFA suggests widespread axonal damage resulting from HIV infection.Notably, FA values were observed to be higher in some ROIs in individuals with HIV compared to healthy controls.For instance, FA increased by 2% in the SFOF, while µFA exhibited a 10% decrease compared to controls.One plausible explanation is that axonal degradation within a single bundle in a crossing ber region, such as the SFOF, may diminish the regional dispersion of ber orientation, consequently leading to an increase in regional FA.Furthermore, we observed a 17-24% decrease in MKa and up to a 4-10% decrease in MKt, along with a 10-21% increase in MKi, in white matter regions in PWH compared to controls.This is noteworthy, as conventional dMRI without b-tensor encoding cannot separate MKi and MKa, as it can only detect MKt, which is the sum of the two.By using b-tensor encoding to dissociate the two, larger differences between the groups were found.
The cognitive performance, as measured by the total cognitive Z-score, demonstrated a stronger correlation with b-tensor encoding metrics compared to DTI metrics in PWH, underscoring the sensitivity of b-tensor encoding metrics.Signi cantly, the interaction between b-tensor metrics and HIV status was observed for the b-tensor encoding-based anisotropy metrics (i.e., µFA and MKa in crossing and fanning ber regions).This suggests that the cognitive effects in the HIV cohort are primarily linked to the loss of local anisotropy, impacting cognitive performance.Moreover, cognitive domain scores, particularly in executive function, attention, and motor functions, exhibited robust associations with anisotropy metrics, speci cally µFA, compared to FA.However, there were no signi cant interactions between the imaging metrics and cognitive domain scores.The observed trend in correlations aligns with previous studies involving DTI-derived FA.Decreased FA has also been noted in various white matter regions, correlating with decreased memory and executive function in PWH exhibiting HIV-associated neurocognitive disorders, particularly in studies with larger sample size (63, 69, 70).
The presence of Neuro lament light chain (NfL) in plasma has emerged as blood marker indicative of neuroaxonal degradation (71).NfL is released into the brain's extracellular space (ECS) following axonal injury and subsequently detected in the cerebrospinal uid (CSF) and blood (71,72).Elevated NfL levels consistently manifest in various neurological and neurodegenerative disorders, including HIV infection (72)(73)(74)(75).Furthermore, activated glial cells are recognized for releasing microparticles expressing Glial Fibrillary Acidic Protein (GFAP) into circulation during brain injury (76-80), cognitive impairment (81), and viral infections such as HIV infection(82, 83).This study unveils a notably stronger association between b-tensor encoding metrics and both NFL and GFAP in white matter, compared to DTI metrics in individuals with HIV.This nding suggests enhanced sensitivity in detecting relevant correlations.
However, it is essential to acknowledge several limitations within this study.Firstly, despite the careful age matching between PWH and healthy controls, there exists an imbalance in the proportion of male and female participants.This discrepancy could introduce gender-related confounding factors.Despite concerted efforts to include female participants, the representation remains at a minimum of 25% in each cohort.Nevertheless, the proportion of males and females in both cohorts are not signi cantly different.However, the proportion of Caucasians and African Americans was signi cantly imbalanced.Moreover, due to a lack of SMS in the early implementation of the FWF sequence the collection of b-tensor encoding metrics encountered limitations in resolution compared to DTI measures (2x2x4 mm 3 vs.1.5x1.5x1.5 mm 3 ).This discrepancy was a result of time constraints during data acquisition.These limitations should be considered when interpreting the results and may warrant further investigation in future studies with larger and more diverse cohorts.

Conclusion
In this study, we investigated the effectiveness of tensor-valued diffusion encoding and associated analysis in delineating tissue microstructural degradation in people living with HIV.Our ndings indicate that metrics based on b-tensor encoding demonstrate greater sensitivity in quantifying subtle changes associated with HIV infection.Moreover, we demonstrated a signi cant correlation between b-tensor encoding metrics, cognitive scores, and plasma levels of NFL and GFAP in PWH.Therefore, the utilization of b-tensor encoding offers a more comprehensive and clinically relevant insight into abnormalities in brain tissue microstructure related to HIV infection compared to the conventional DTI approach.

Figure 1 Example
Figure 1

Figure 2 Comparison
Figure 2

Figure 3 Relationships
Figure 3

Table 1
Study coordinators trained and supervised by an experienced neuropsychologist, administered all neuropsychological tests.The test battery covered diverse cognitive domains, such as Attention/Working Memory (CalCAP CRT 4; CRT 14), Speed of Information Processing (Stroop Color Naming, Digit Symbol Modalities Test), Executive Function (Trail Making Test B, Stroop Interference Task), Language (letter and category uency), Learning (Rey Auditory Verbal Learning Test Trials 1-5, Rey Complex Figure Test ImmediateRecall), Memory (RAVLT Trial 7, RCFT Delayed Recall), and Motor Skill (Grooved Pegboard).Assessment of premorbid intellectual functioning and English language uency was limited to the baseline, utilizing the Wide Range Achievement Test (WRAT) 4-Reading subtest.was performed on a 3T whole-body scanner (MAGNETOM Prisma Fit, Siemens, Erlangen, Germany, software version VE11c) equipped with a 64channel head coil.The maximum gradient strength is 80 mT/m with a slew rate of 200 mT/m/s.Anatomical Imaging: The T1-weighted (T1w) images were acquired using a 3D magnetization prepared rapid acquisition gradient-echo (MPRAGE) MRI

Table 2
Diffusion MRI metrics for HIV and healthy controls in brain tissues Bold font indicates correlations that were statistically signi cant after FDR correction.

Table 3
Two-Way ANOVA to measure the effects of imaging metrics, HIV status, and their interactions on total cognitive scores.