Linking Neuroinammation and Extracellular Free Water in HIV Infection: A Longitudinal Study

Initiation of combination antiretroviral therapy (cART) reduces inammation in HIV-infected (HIV+) individuals. Recent studies demonstrated that diffusion MRI based extracellular free water (FW) modeling can be sensitive to neuroinammation. Here, we investigate the FW in HIV-infection, its temporal evolution, and its association with blood markers, and cognitive scores. Using 96 age-matched participants, we found that FW was signicantly elevated in grey and white matter in cART-naïve HIV+ compared to healthy controls (HIV-) at baseline. Similarly, at baseline, HIV+ participants had increased neurolament light chain (NfL) values that correlated with FW and CD4 counts. FW in grey and white matter, as well as NfL decreased in the HIV+ after 12 weeks of cART treatment. No signicant FW differences were noted between the HIV+ and HIV- cohorts at 1 and 2-year follow-up. Results suggest that FW elevation in cART-naïve HIV+ participants is likely due to neuroinammation. The correlation between FW and NfL and the improvement in both FW and NfL after 12 weeks of cART treatment further reinforces this conclusion. The longer follow-up at 1 and 2 years suggests that cART helped control neuroinammation as inferred by FW. Therefore, FW could be used as a biomarker to monitor HIV-associated neuroinammation.


Introduction
Despite the successful suppression of viral replication and improved immune function with combination antiretroviral therapy (cART), approximately 30-50% of HIV-infected individuals still develop HIVassociated neurological disorders, including cognitive impairment 1 . There is evidence that in ammation persists despite undetectable viral loads, and it is thought to be the primary mechanism behind HIV disease progression and chronicity, and comorbidities such as HIV-associated cognitive impairment 1,2 .
Microglia and perivascular macrophages are the primary actors of neuroin ammation 3 . These cell types are also the primary source of productive HIV infection 4,5 . Quantifying the presence of HIV brain infection, neuroin ammation and neuronal injury associated with the infection, are key factors in evaluating the HIV reservoir, the burden of neuroin ammation and consequences of such neuroin ammation. Evaluation of brain HIV infection in HIV-infected individuals is currently limited to indirect measurements such as those performed using cerebrospinal uid (CSF). However, CSF viral load is undetectable in >90% of well-treated individuals with undetectable plasma viral load, thus, CSF viral load may not adequately represent the extent of intra-parenchymal infection 6 .
To date, neuroin ammation due to HIV infection has been assessed using several neuroimaging biomarkers, including elevated glial markers via magnetic resonance spectroscopy (MRS) 7,8 , increased glucose metabolism via Positron Emission Tomography (PET) and increased microglial activation via translocator protein (TSPO) PET tracers [9][10][11] . While PET ligands offer a more direct quanti cation of microglia activation 12 , there are several drawbacks. PET imaging is expensive, it exposes individuals to signi cant ionizing radiation that could limit frequent scanning, ligands are not always widely available, and it is non-informative on how the signal captured is re ective of tissue damage. Speci cally, in cART treated individuals, the expected level of activated microglia is small and the signal captured does not provide information on the effect on neurons 13 . In contrast, the FW index is an emerging diffusion MRI metric that indirectly and noninvasively measures the leakage of extracellular uid into brain regions. FW index has been found to be associated with neuroin ammatory responses, atrophy, vascular risk factors and cognitive declines 7,[13][14][15] . While the exact interpretation of the FW index is still unclear, one of the hypotheses is that excessive extracellular FW stems from the leakage of the blood brain barrier (BBB) with leaking serum proteins that cause axonal damage and demyelination. It is well established that the BBB plays a crucial role in the pathogenesis of the HIV-infection 16 .
Free water (FW) imaging, a diffusion MRI post-processing method, has been used to differentiate the extracellular non-owing free water from the contribution of water that diffuses in proximity to the axons 17 . The FW index, a measure of the relative fraction of freely diffusing water in the extracellular space (ECS), can reveal neuroaxonal damage that affects the tissue diffusion characteristics. FW has been found to be correlated with neuroin ammation and neurodegeneration in several neurological disorders [18][19][20][21][22] . The relevance of FW as a marker of in ammation has been further evaluated by correlating FW with levels of microglia activation measured by TSPO via a PET study 23 . Furthermore, previous animal studies [24][25][26] have shown that neuroin ammation alters brain ECS.
The main goals of this work were to investigate whether increased FW index would be observed in cART naïve HIV-infected individuals and if it improves with cART treatment, thus providing a marker of responsiveness to treatment. We also posited that if FW index were indicative of in ammation, it would be correlated with other makers of brain injury such as neuro lament light chain (NfL) and markers of immune function such as CD4 count.

Methods And Materials
Participants Forty-four treatment-naïve HIV+ participants (4 females and 40 males; mean age ± standard error, SE =34.48 ± 1.95 years, range, 20 -63 years) and 52 age-matched HIV-healthy controls (26 females and 26 males; mean age ± SE =37.02 ± 1.66 years, range, 18-63 years) were enrolled and followed between 2013 to 2019 prior to and then following cART treatment. The study was reviewed and approved by the Institutional Research Subjects Review Board (RSRB) at the University of Rochester Medical Center. All participants provided written informed consent before enrollment, and then underwent clinical, laboratory, neurocognitive and brain MRI exams. All experiments were carried out in accordance with relevant guidelines and regulations. HIV+ participants were evaluated at baseline prior to initiation of cART and at three follow-up timepoints [12 weeks (n = 38), 1 year (n = 29), and 2 years (n = 19)] after cART treatment had been started, while the HIV-participants were evaluated at the baseline, 1-year (n = 46), and 2-year (n = 19) time points. The 12-week evaluation of the HIV-participants were not included in the study design as signi cant changes in MRI and clinical measures in healthy controls are not expected within such a short period. Detailed baseline demographics (including age, sex, clinical characteristics) are presented in
All MRI images were checked for any severe artifacts such as gross geometric distortion, bulk motion, and signal dropout. DWI images were corrected for eddy current-induced distortion, inter-volume participant motion, and susceptibility-induced distortion using the "topup" and "eddy" tools in FSL 30,31 . A bi-tensor model was used to create FW maps on a voxel-by-voxel basis from the DWI data using a previously described algorithm 21 and the processing was carried out using Next ow 32 pipeline with all software dependencies bundled in a Singularity Container 33 . The FW values are in the range of 0 to 1. Values approaching 0 correspond to negligible FW diffusion in the ECS while 1 corresponds to maximum FW diffusion (i.e., water in a voxel that diffuses completely freely). All the T1w images were co-registered to the MNI space (1mm) using ANTs 29 . The same transformations were applied to the FW maps. Since the data was collected either with Siemens TIM Trio or Siemens Prisma due to scanner upgrade, we used ComBat algorithm to harmonize data across scanner for the DWI derived FW index for each ROI 34 . An empirical Bayes framework is used in the ComBat algorithm to estimate additive and multiplicative scanner effects.
The FW values were calculated from whole Grey Matter (GM) and White Matter (WM) using corresponding masks. In addition, the Harvard-Oxford (subcortical) and JHU-ICBM (WM and tracts) atlases were used to calculate region averages in standard space (1mm) in 25 pre-de ned regions-ofinterest (ROI) relevant to HIV infection 35,36 (shown in supplementary tables). FW values were averaged over bilateral ROIs (except some WM tracts) for each participant.
Statistical analyses were performed in R 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria). Marginal comparisons between two independent groups were conducted by either two-group Welch's unequal variances t-test (for continuous variables) or Fisher exact test (for categorical variables). Paired ttests were used to compare the levels of continuous variables in the HIV+ and HIV-participants between baseline and follow-up visits. Pearson correlation tests were used to test the marginal associations between two continuous variables. 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 37 .
Linear mixed effects regression modeling: Due to the longitudinal nature of the data, most multivariate analyses performed in this study were based on linear mixed effects regressions (LMER), with perparticipant random intercepts to account for serial correlations between multiple time points. Empirical evidence shows that the cART treatment effect was most prominent in the rst 12 weeks; its long-term effect was subtler and, in some cases, different from its short-term effect. Therefore, we performed two separate LMER analyses to study the longitudinal associations between covariates (such as HIV status and cART treatment) and the response variables (such as FW index and blood markers).
The Short-Term Model (STM) was applied to data collected at baseline (both HIV+ and HIV-participants) and Week-12 (HIV+ participants only). Covariates included were HIV status (cohort), short-term cART treatment (visit), and age.
The Long-Term Model (LTM) was applied to data collected from the HIV-participants at baseline and the HIV+ participants at week-12 (used as the new baseline time point in LTM), year-1 and year-2 data collected from both cohorts. Covariates included were HIV status, visit, age, and the interaction between HIV status and visit to account for possibly different temporal patterns of a response variable between two cohorts.
For both models, the parameters were estimated by the restricted maximum likelihood (REML) criterion, and the statistical signi cance was assessed by the adjusted ANOVA F-test and regression t-test provided by R package lmerTest 38 .

Participant Characteristics
Demographic and clinical data of the participants are presented in Table 1. Welch's Two Sample t-test indicates that there is no statistically signi cant age difference between the HIV-and HIV+ participants (p = 0.324). However, we included age as a covariate in all multivariate regression analyses to control for their remaining confounding effects. Compared to HIV+, HIV-participants had a higher number of males than females (p<0.001) and higher education levels (p = 0.039) at baseline.
Baseline FW comparison for HIV+ vs. HIV-cohorts Figure 1A-B represents the comparisons of FW between the HIV+ and HIV-at baseline for GM and WM. Figure 1C. Marginal comparisons based on Welch's two-sample t-test showed that FW index was higher in GM (t=4.74, adjusted p-value p adj <0.001) and in WM (t=2.11, p adj =0.038) in the HIV+ cohort than in HIV-cohort ( Figure 1, Table 1).

Mean voxel-by-voxel FW map from HIV+ cohort is shown in
We also compared the FW in both cohorts categorized by age such as young adults (age £30, 22 HIV+, 18 HIV-participants) and older adults (age >30, 15 HIV+, 32 HIV-participants). FW was signi cantly higher in GM (t=10.51, p adj <0.001) and WM (t=4.072, p adj <0.001) in the HIV+ young adult than HIV-young adult. In contrast, FW was signi cantly higher only in GM (t=2.589, p adj =0.007) in the HIV+ older adult compared to HIV-older adults.
Among 25 ROIs (See Supplementary Table S1) that included subcortical GM structures and WM tracts, 3 ROIs (Thalamus, Amygdala and Hippocampus) had signi cantly higher FW in the HIV+ cohort than in HIV-cohort (p<0.05).

Baseline comparisons of blood marker for HIV+ vs. HIV-cohorts
Welch's two group t-test showed that average NfL concentration was marginally higher in HIV+ compared to the HIV-cohort (t=2.10, p=0.042) ( Table 1).

Short Term effects of cART on FW
In paired comparisons between baseline and week-12 for HIV+ participants (n=31 with measures at both baseline and week-12), we found that the FW index decreased signi cantly in GM (t=4.57, p adj <0.001) and WM (t=2.60, p adj =0.014) (Figure 2A-B). This was also re ected in ROI based analyses (Supplementary Table S2).
Based on the STM that modeled the cohort and treatment effect simultaneously and adjusted for the confounding effects of age, we found that GM and WM are associated with signi cant higher FW in the HIV+ cohort (b cohort , GM =0.044, p adj,GM <0.001; (b cohort , WM =0.005, p adj,WM =0.004). Even after accounting for the effects of HIV and treatment, age was signi cantly associated with the increase of FW for most of the ROIs. More information on individual ROIs can be found in SupplementaryTable S3.

Short Term effects of cART on blood markers
Paired t-test showed that the average NfL concentration decreased (t=1.61, p=0.115) after 12 weeks of cART treatment ( Figure 2C).

Long Term effects of cART on FW
Paired t-tests revealed that FW increased in WM and GM at year-1 compared to week-12 and baseline for the HIV+ and HIV-cohorts respectively (p<0.05). However, at year-1 and year-2 visits there were no signi cant differences in GM and WM between the HIV+ and HIV-( Figure 3A-B).
The LTM showed there were no signi cant cohort effect and the interaction of cohort and visit in GM and WM suggesting that levels of FW in the HIV+ cohort were stabilized after 12 weeks of cART treatment ( Figure 3A-B). The LTM on ROIs also revealed similar results as presented in Supplementary Table S4. Age is invariably associated with signi cant increase of FW.

Long Term effects of cART HIV+ on NfL
The average NfL concentrations were found to become stable after 12 weeks (new baseline) of cART treatment and follow-up visits in the HIV+ cohort. Similar to FW, after 12 weeks of cART treatment, no signi cant differences in NfL were found between the HIV+ and HIV-cohorts during the follow-up visits ( Figure 3C).
Baseline FW association with blood markers and cognitive performance Pearson correlation analysis for HIV+ at baseline showed that average NfL concentration was correlated with FW in GM and WM (b »0.6, p adj <0.001). In addition, we found signi cant positive correlations for FW vs. NfL in 19 of 25 ROIs. The CD4 cell counts were negatively associated with FW in GM and WM (b »-0.4, p adj =0.041), as well as in 14 out of 25 ROIs. However, no signi cant correlations were found for FW vs. VL for any ROIs. Details are provided in Supplementary Table S5. The total cognitive score was also lower in the HIV+ cohort compared to the HIV-cohort at baseline (p=0.028). Paired t-test showed that total cognitive score increased after 12 weeks of cART treatment in HIV+ cohort (p<0.001). After 12 weeks of cART treatment and follow-up visits in HIV+ cohort, total cognitive scores became stable. The total cognitive score was not signi cantly correlated with FW for WM and GM in both cohorts at baseline; however, there was a trend toward a negative correlation in the GM (r= -0.26, p=0.11). In the STM and LTM, FW in WM and GM were not signi cantly associated with the total cognitive scores. Visit and interaction between cohort and visit were found signi cant for FW in WM and GM in the LTM while visit was signi cant in the STM (p adj <0.05).

Discussion
In this study, we used the FW index to indirectly assess neuroin ammation associated with HIV infection. The rationale to use FW as a putative marker of in ammation is based on previous studies in other brain diseases [18][19][20][21]23,26 . We reasoned that HIV infected cART naïve participants would be at higher risk of in ammation (higher FW) and that treatment would reduce in ammation (lower FW). We also hypothesized that active neuroin ammation would be associated with higher plasma levels of NfL and lower cognitive performance. Four main ndings emerged from this work: 1) FW index and NfL were higher in cART naïve HIV+ compared to the HIV-participants at baseline; 2) FW index and NfL decreased dramatically in GM and WM after 12 weeks of cART treatment in the HIV+; 3) FW levels were comparable between the HIV+ and HIV-at 1 year and 2 years of follow-up and similar trends were observed in NfL; and 4) the baseline measures of FW index in GM and WM in the HIV+ were strongly associated with NfL concentration.
To the best of our knowledge, this is the rst study to investigate FW as a possible biomarker for neuroin ammation in the HIV+ cohort. Prior work demonstrates the involvement of subcortical GM and WM structures in neuronal damage in the HIV+ participants 7,35,36,39,40 . We found signi cantly higher FW index in GM and WM in cART-naïve HIV+ compared to the HIV-participants at baseline, consistent with previous work in other neurological disorders 15,[18][19][20][21]41 . Increased FW in GM and WM in the HIV+ cohort might be related to abnormal neuroimmune response 42,43 .
The plasma level of the Neuro lament light chain (NfL) is a promising blood marker for neuroaxonal degradation 44 . NfL is considered a direct measure of neuronal damage since it is released into the brain's ECS following axonal injury and consequently into the CSF and blood 44,45 . Elevated NfL levels are observed in several neurological and neurodegenerative disorders including HIV+ cohort 41,[45][46][47] . In this work, the average NfL concentration was signi cantly higher, by ~39%, in the HIV+ compared to HIVparticipants. The FW index was found to be positively correlated with NfL and negatively correlated with CD4 counts. Several previous studies suggested that low CD4 counts might be linked to brain atrophy including cortical thinning, volume reduction in GM and WM as well as ventricular enlargement [48][49][50] . As such, low CD4 is associated with elevated extracellular FW in brain tissue. These results provide additional evidence that FW may be related to neuroin ammatory processes. However, in the present study we did not nd an association between FW and VL.
A signi cant nding was that FW decreased drastically in whole GM and WM and several ROIs in the brain after 12 weeks of cART treatment. That is, FW was decreased signi cantly by 21% and 8% in the GM and WM respectively due to the cART treatment. After 12 weeks of cART treatment, FW values were close to those of the HIV-participants implying that short term cART treatment normalized FW in the HIV+ cohort. In contrast, the average NfL concentration was reduced by ~17% after 12 weeks of cART treatment in the HIV+ cohort. A previous study also reported that NfL levels decreased to normal level after 6 months of cART initiation in acute HIV-infection 51 . These ndings indicate that the change in neuroin ammation due to the short-term treatment effect can be estimated using FW index.
After the initial changes due to 12-weeks of cART exposure, no signi cant differences in FW were observed between the HIV + and HIV-participants during the two years of follow-up. However, we found that FW increased signi cantly at year-1 compared to that of baseline in the HIV-participants and compared to 12 weeks of cART treatment in the HIV+ participants. Comparison of young adults and older adults revealed that these differences in FW were driven more by the young adults (age<30 years) in both cohorts. It is not clear whether a yearly change in FW water should be expected as both of our cohorts showed no further changes at year-2 of follow-up although it is worth to notice that the sample size was much reduced at year-2 follow-up compared to year-1. Further investigation with large sample is required to con rm these ndings. Overall, FW and NfL exhibited similar temporal trends over 2 years follow-up visits, suggesting that after cART initiation there is likely minimal level of neuroin ammation.
Of notice, there was no signi cant correlation between cognitive performance and FW. It is possible that changes in cognitive performance require some neuronal and glia structural damage while FW dynamics can happen at any stage including those where cellular integrity is minimally altered. The trend observed in negative correlation between cognitive performance and FW may also suggest that the relatively small sample size may have contributed to the limited association.
This study has a few limitations worth considering. First, the proportion of male and female participants was not equal, however the FW index in GM and WM was not signi cantly different in males vs. females in our HIV-participants. Second, the number of participants was lower for follow-up visits, especially at year-2 and blood markers were not collected for all participants.

Conclusions
In summary, our ndings suggest that extracellular FW is elevated in the brains of cART naïve HIV+ participants at a time when neuroin ammation is expected to be high. Higher levels of NfL and its correlation with FW tend to support this possibility. Most importantly, short term cART treatment effectively reduces the levels of FW and stabilizes it over the 2 years of follow-up. Although the biological underpinnings of the elevated FW index are still unclear, FW is a potential in ammatory marker that could be used to monitor disease course and response to interventions in HIV-infected individuals.