Resting state network connectivity alterations in HIV: Parallels with aging

Abstract The increasing incidence of age‐related comorbidities in people with HIV (PWH) has led to accelerated aging theories. Functional neuroimaging research, including functional connectivity (FC) using resting‐state functional magnetic resonance imaging (rs‐fMRI), has identified neural aberrations related to HIV infection. Yet little is known about the relationship between aging and resting‐state FC in PWH. This study included 86 virally suppressed PWH and 99 demographically matched controls spanning 22–72 years old who underwent rs‐fMRI. The independent and interactive effects of HIV and aging on FC were investigated both within‐ and between‐network using a 7‐network atlas. The relationship between HIV‐related cognitive deficits and FC was also examined. We also conducted network‐based statistical analyses using a brain anatomical atlas (n = 512 regions) to ensure similar results across independent approaches. We found independent effects of age and HIV in between‐network FC. The age‐related increases in FC were widespread, while PWH displayed further increases above and beyond aging, particularly between‐network FC of the default‐mode and executive control networks. The results were overall similar using the regional approach. Since both HIV infection and aging are associated with independent increases in between‐network FC, HIV infection may be associated with a reorganization of the major brain networks and their functional interactions in a manner similar to aging.


| INTRODUCTION
In resource-rich countries, people with HIV (PWH) now have a life expectancy similar to that of uninfected individuals (Samji et al., 2013). HIV eradication, however, remains elusive and HIV infection continues to be one of the leading causes of global disease burden (Ortblad et al., 2013). Additionally, with aging and long-term chronic disease, age-related comorbidities have become a key concern in PWH ( Rasmussen et al., 2015;Wing, 2016) and studies have shown an increased incidence of such comorbidities in PWH compared with uninfected controls (Pathai et al., 2014). These comorbidities include cardiovascular disease (Martin-Iguacel et al., 2015), liver disease (Joshi et al., 2011), renal disease (Ando et al., 2012), and neurologic diseases such as HIV-associated neurocognitive disorder (HAND; Heaton et al., 2010). The increased prevalence of age-related disease in PWH has led to theories of accelerated aging in PWH, which have been supported by studies of DNA methylation (Gross et al., 2016;Lew et al., 2021). Additionally, emerging literature using machine learning methods and structural MRI has also tested the hypothesis that brain aging is accelerated in PWH, finding that PWH have greater brain predicted age difference/brain age gap (Cole et al., 2017;Kuhn et al., 2018).
Studies using functional magnetic resonance imaging (fMRI) have become highly promising in identifying the changes that occur in the brain of PWH. Task-based studies have shown patterns of hyperactivity in PWH compared with controls (Hakkers et al., 2017), including increases in prefrontal and cingulate activity during attention tasks (Chang et al., 2004(Chang et al., , 2008(Chang et al., , 2013 and during working memory tasks (Chang et al., 2001;Ernst et al., 2002). One meta-analysis of such task-based fMRI studies identified consistent HIV-related aberrations in the fronto-striatal circuitry (Du Plessis et al., 2014).
Similarly, a number of resting state fMRI studies have identified comparable changes in functional connectivity (FC) in PWH. That is, HIV-related changes in FC have principally identified alterations in the frontal cortices, specifically the executive control network (ECN;Ipser et al., 2015;Ortega et al., 2015;Thomas et al., 2013) and default mode network (DMN; Cole et al., 2018;Thomas et al., 2013). Studies have also identified altered cortico-striatal FC related to HIV infection (Ipser et al., 2015;Ortega et al., 2015). In all, these studies suggest HIV infection is related to aberrancies in frontal cortices and striatal circuits, however, the added impact of aging on these changes is not fully understood.
In uninfected individuals, normative aging appears to be related to reduced brain modularity. That is, with increasing age, withinnetwork FC appears to decrease, while between-network FC increases (Betzel et al., 2014;Damoiseaux, 2017;Spreng et al., 2016).
Overall, this global pattern of decreasing network segregation with aging provides a concise framework to understand how functional connectivity changes with normative aging.
Previous aging studies of HIV have been inconsistent, with some studies identifying patterns of accelerated aging in PWH as shown by HIV-by-age interactions (Arif et al., 2020;Chang et al., 2013;Egbert et al., 2018Egbert et al., , 2019Lew et al., 2020), while other studies not showing such interactions, instead finding independent and additive effects of age and HIV (Ances et al., 2010(Ances et al., , 2012Cole et al., 2018;Ipser et al., 2015;Juengst et al., 2007;Thomas et al., 2013). Similar inconsistencies have emerged in neuroimaging studies of HAND, with some studies finding correlations between neuropsychological performance and corresponding functional brain networks (Chaganti et al., 2017;Egbert et al., 2018;Wang et al., 2019), while others show little or no FC impairments in HAND (Cole et al., 2018;Guha et al., 2016;Janssen et al., 2017). In sum, a clear pattern of neural changes associated with HIV, aging, and cognitive function remains elusive, and few studies have examined PWH through the lens of an aging framework.
In this study, we aimed to examine the relationships between aging and HIV infection in resting state FC. Additionally, we aimed to understand the contribution of HAND toward this relationship. We studied a large sample of PWH and uninfected controls, sampled evenly from 22 to 72 years old and computed estimates of FC using resting state fMRI. To study these relationships through the lens of an aging phenotype, we assessed within-and between-network connectivity, and hypothesized that PWH would show evidence of advanced aging in the form of increased between-network connectivity, and decreased within-network connectivity, relative to the healthy control group. We further studied how these metrics differ in participants with HAND relative to cognitively unimpaired PWH, hypothesizing that those with HAND would show larger functional aberrations and age acceleration. Lastly, to ensure the validity of our findings regardless of the brain atlas used, we conducted further analyses using regional FC. nation of HAND, the study followed the same protocol as described in Lew et al. (2018). Briefly, Composite scores for multiple functional domains (executive functioning, attention, speed of processing, fine motor, verbal learning and memory, and language) were computed by calculating demographically normalized z-scores and taking an average of the z-scores for all tests within that domain. Along with an assessment of activities of daily living, these scores were used to diagnose HAND according to the Frascati guidelines (Antinori et al., 2007). A global composite score was created by averaging across domains.

| Resting-state fMRI preprocessing
The rs-fMRI data were preprocessed using SPM12 and the DPABI Toolbox (Yan et al., 2016). Preprocessing procedures included removal of the first 10 volumes, motion correction to the first volume with rigid-body alignment; co-registration between the functional scans and the anatomical T1-weighted scan; linear detrending; regression of motion parameters, and their derivatives (24-parameter model; Friston et al., 1996), as well as white matter (WM), cerebrospinal fluid (CSF) time series (using a component-based noise reduction method, five principal components; Behzadi et al., 2007); and spike censoring (volumes with a framewise displacement [FD] ≥0.5 were regressed out; Satterthwaite et al., 2013); spatial normalization of the functional images into Montreal Neurological Institute (MNI) stereotaxic standard space; and spatial smoothing with a 6-mm at full-width at halfmaximum Gaussian kernel. Finally, bandpass filtering was applied at (0.01-0.1) Hz (Cordes et al., 2001).
For quality control of the imaging data, participants with excess head motion, defined as >0.8 in mean FD (Power et al., 2012) and/or >2.5 mm of maximum motion, were excluded from the study. After removing these individuals, 86 PWH and 99 uninfected controls had both structural and functional MRI data that could be used for further processing (Table 1).

| Extraction of network functional connectivity measures
We used the Yeo 7-network atlas (Yeo et al., 2011), a previously established functional brain atlas based on 1000 resting state fMRI scans, to partition the functional connectome into seven replicated

| Statistics
We utilized linear regression models to examine the independent and interactive effects of HIV and aging. Each within-and between-T A B L E 1 Demographic information of the participants.

| Analysis of HIV-associated neurocognitive disorder
To determine the impact of HAND, we performed post hoc analyses repeating our statistical comparisons after splitting the PWH group by HAND. Specifically, we performed pairwise comparisons between uninfected controls, unimpaired PWH, and participants with HAND.
These comparisons were performed for all within-and betweennetwork functional connectivity estimates.

| Parallel analyses with network based statistics
To ensure that our findings were not dependent on the technique and brain atlas used, we further conducted analyses using a regional approach. To do so, we first extracted FC between each pair of regions, based on the 512-region automated anatomical labeling (AAL) atlas (Crossley et al., 2013;Tzourio-Mazoyer et al., 2002;Zalesky, Fornito, Harding, et al., 2010). We then, use the network-based statistic (NBS) toolbox v1.2

| Effects of aging and HIV in between-network functional connectivity
Significant main effects of age were identified in all seven BN-FC metrics (see Table 2 and Figure 2). All of these effects showed consistent increases in between-network functional connectivity with increasing age.
With regard to HIV status, BN-FC metrics showed a significant main effect of HIV status in the LIM, DMN, and ECN (Table 2, Figure 2), All of these significant effects showed an increase in BN-FC in PWH relative to controls. Adding in the HIV-by-age interaction to the models showed no interaction effects in any of the BN-FC metrics. To further validate and explore these effects, we also performed a leave-one-out cross validation on all within-and between-network functional connectivity models. These results supported the general findings and are reported in Figures S1 and S2.
Because of the significant impact of preprocessing on FC (Tozzi et al., 2020), we also examined these BN-FC data after applying Global Signal Regression during preprocessing, and interrogated the impact of such processing on long versus short-range connections.
Results showed overall consistent patterns and a weak relationship between motion correlation and connectivity distance, which are reported in Figures S3 and S4).
To further probe the effects of HIV and age on BN-FC, we per-  Figure S5. Additionally, matrices using the 512-regions atlas are shown in Figure S6.

| HAND impact on network FC
To examine whether our effects of HIV were driven by participants with HAND, we split the HIV group by HAND status and performed comparisons between controls, unimpaired PWH, and those with HAND. Ultimately, no significant differences were identified between unimpaired PWH and those with HAND. Similarly, significant differences between controls and unimpaired PWH were not identified in BN-FC. However, significant differences between controls and PWH with HAND were present in FC between the VAN, ECN, and DMN between-network FC metrics (all p < .05; Figure S7).
Given the significant group-by-age interaction in WN-FC of the VAN, we also performed follow-up pairwise comparisons of this interaction effect. The control versus HAND comparison of WN-FC showed a significant interaction of group by age ( p = .032), while in the control versus unimpaired PWH comparison there was not a significant group by age interaction ( p = .143; Figure S8). This suggests that cognitive impairment may be driving this differential aging trajectory.

| Parallel analyses with network-based statistics
We conducted NBS analyses to corroborate and expand upon the findings from the network-based FC analyses using a region-based approach. Since the size of a significant component or subnetwork is associated with the chosen primary threshold in the NBS test, we found more significant networks for the lower threshold (t = 4.5) and fewer for the higher threshold (t = 5.5), as expected. Nonetheless, the results with our network-based findings (Figure 4). The vast majority of the significant links were between nodes from different networks (gray links on Figure 4).
In regard to the similarity of the HIV effect on the networkrelated findings, PWH showed significant increases in FC in a total of 24 links, which were largely between the frontal regions of the DMN and ECN (light and dark blue), as well as between the nodes belonging to the ECN and VAN (green; Figure 4). In contrast to the network approach, the NBS also identified significantly reduced FC links in the PWH compared with controls. However, these links were relatively limited and largely localized within the DMN, and particularly the dorsal part of the network (dorsal medial prefrontal cortex vs. precuneus/ angular gyri; light blue, Figure 4).
Finally, no significant HIV-by-age interactions or differences between PWH and HAND were detected.

| DISCUSSION
Our study examined a large sample of PWH and uninfected controls spanning age 22-72 and identified independent effects of age and HIV in resting-state FC between the major brain networks. Agerelated increases in BN-FC were widespread, and PWH displayed further increases above and beyond aging, specifically between networks related to higher order functions, such as the default-mode and executive control networks. Employing a complementary regional-based F I G U R E 1 Within-network functional connectivity by age and HIV status. The average correlation within each network was calculated and z transformed. Regressions testing the independent effects of HIV and age showed no significant main effects of HIV, and a significant main effect of age only in the visual network. Scatter plots display z values of each within-network functional connectivity metric by age, with uninfected controls in blue and PWH in red. HAND status is differentiated by shape for display purposes. Linear fits for each group are displayed with 95% confidence intervals. Boxplots displaying group differences are added to the right of each plot, and visual representations of each network are inset. denotes significant ( p < .05) main effect of age. denotes significant ( p < .05) HIV by age interaction. Related statistics are reported in Table 2.
approach further validated these results and extended our findings by displaying the specific nodes that showed the largest effects. These findings broadly support the framework that HIV infection leads to alterations in brain functional connectivity that are highly similar to the aging phenotype. Below we discuss these findings in the context of previous literature.
Our data showed expected increases in BN-FC with increasing age, which is a well-established pattern of aging (Damoiseaux, 2017).
Using our framework, we additionally found that HIV infection was also related to widespread increases in BN-FC, independent and additive to the effects of aging. Our findings may be consistent with previous network level investigations of HIV, which found more positive BN-FC in PWH compared with controls (Thomas et al., 2013).
Importantly, our results expand upon these previous studies by studying this pattern across aging, establishing that this pattern occurs across adulthood and mimics the aging phenotype seen in our sample.
Overall, these findings suggest that PWH display a connectivity profile that resembles an advancement in age.
These network-level resting-state FC findings were confirmed and expanded upon by employing the NBS toolbox (Zalesky, Fornito, Harding, et al., 2010). NBS detected altered FC at the brain region level, offering a finer granularity of the FC differences. While we found an overall similar pattern to the network-based findings, we identified interesting results of both stronger and weaker FC in the PWH relative to the uninfected controls. Similar to the resting-state networks analysis, NBS revealed widespread BN-FC changes with F I G U R E 2 Between-network FC by age and HIV status. The correlation between each network was calculated, z transformed, and then averaged for each network. Regressions testing the independent effects of HIV and age showed significant main effects of age in every network. Significant main effects of HIV were identified in limbic, default mode, and executive control networks. Scatter plots display z values of each between-network functional connectivity metric by age, with uninfected controls in blue and PWH in red. HAND status is differentiated by shape for display purposes. Linear fits for the groups (controls and PWH) are displayed with 95% confidence intervals. Boxplots displaying group differences are added to the right of each plot, and visual representations of each network are inset. *p < .05 with motion covaried out. denotes significant (p < .05) main effect of age. Related statistics are reported in Table 2. increasing age, significant group effects where PWH had increased FC between the ECN and DMN compared with controls, no significant age-by-HIV interactions, and no significant main effects of HAND. In particular, there was significantly stronger FC in PWH compared with controls, especially between nodes in the frontal regions among the DMN and ECN, which has been reported in the literature (Ipser et al., 2015;Ortega et al., 2015;Thomas et al., 2013). Also, reduction in DMN FC has been reported in multiple resting-state studies of PWH (Cole et al., 2018;Thomas et al., 2013). Here we replicate and expand upon prior studies (Ances et al., 2010(Ances et al., , 2012Ipser et al., 2015;Juengst et al., 2007;Thomas et al., 2013) by showing that these changes are present at the regional level, independent of aging. In fact, at the network level we did not observe lower FC within the DMN, but this did emerge at the regional level, with lower FC between the dorsal medial prefrontal cortex and angular gyri/precu- Our analyses of HAND status showed no significant differences between unimpaired PWH and HAND, suggesting our effects were not driven by cognitive dysfunction. However, when comparing these subgroups to uninfected controls, BN-FC with the VAN, DMN, and ECN were significantly increased in those with HAND, but not unimpaired PWH. Additionally, other follow-up analyses also suggested that participants with HAND drove the interaction effect in WN-FC of the VAN. Caution is needed when interpreting these pairwise findings, as unimpaired PWH did not significantly differ from participants with HAND. However, taken together, this may indicate that the VAN is particularly sensitive to cognitive impairment. Supporting this, previous studies have also found HAND-related differences in the salience network (Chaganti et al., 2017) and therefore this network may be a promising indicator of HAND. Previous studies have also identified the presence of functional deficits related to HIV in the absence of HAND, with one study specifically noting no difference in restingstate FC by HAND status (Guha et al., 2016). This ultimately suggests that a reorganization of brain networks occurs irrespective of neuropsychological changes. At the same time, our sample of PWH with HAND may not represent a severe enough sample to detect further exacerbations of functional connectivity. That is, the majority of our participants with HAND were of the mildest category (asymptomatic neurocognitive impairment). This could have led our HAND group to be similar to the unimpaired group. Further study is needed to examine the more severe presentations of HAND and to reconcile these data with that from magnetoencephalographic (MEG) imaging, which F I G U R E 3 Pairwise functional connectivity statistics: effects of age and HIV. Matrix statistics performed on betweenand within-(diagonal) network connectivity matrices allowed pairwise between-network connectivity to be examined. has frequently reported neurophysiological differences among cognitively impaired (i.e., HAND) and unimpaired PWH (Arif et al., 2020;Groff et al., 2020;Lew et al., 2018;Spooner et al., 2020;Wiesman et al., 2018;Wilson et al., 2016).
Regarding HIV-by-age interactions, none of the BN-FC models showed any significant interactions. This is in agreement with multiple studies showing that the effects of age and HIV on brain function are independent (Ances et al., 2010;Cole et al., 2018;Juengst et al., 2007;Ortega et al., 2015). However, importantly, these independent main effects of HIV and aging still fit into an aging frame-  (Pathai et al., 2014).
One notable exception in our data is that we did identify a significant HIV-by-age interaction in the ventral attention network, such that PWH showed a greater decrease in FC compared with controls.
Indeed, other neuroimaging studies have identified HIV by age interactive effects (Lew et al., 2021;Petersen et al., 2022), which may suggest an accelerated aging profile. Therefore, an accelerated agerelated pattern of decreased within network connectivity may have been seen in our PWH, albeit specific to one network. Interestingly, this is the same network that showed HAND-specific changes in BN-FC. Ultimately, the ventral attention network may be altered differentially with HIV infection, and further studies might examine this network more specifically.
Interestingly, we also identified increased WN-FC of the VIS network with age. While it has been previously reported that this network is among the least influenced by aging (Doucet et al., 2021), other studies have also reported a positive association with aging (Seidler et al., 2015;Zhou et al., 2023;Zonneveld et al., 2019). We believe that such differences may be related to a variety of factors F I G U R E 4 Results from the network-based statistics toolbox. Significant functional connectivity (FC) with age and between groups with primary threshold t-score = 5. (top) FC positively correlated with age, with a large majority being between-networks throughout the brain (gray connections). (middle) Significantly weaker FC in PWH compared with uninfected controls, primarily between nodes in the frontal, parietal, and occipital regions within the DMN, within the ECN, and between the DMN and ECN. (bottom) Significantly stronger FC in PWH compared with controls, especially between nodes in the frontal regions between the DMN and ECN. Legend to the right presents the color of the nodes (and edges for within-network connectivity) representing functional networks based on the Yeo 7-network atlas. DAN, dorsal attention network (yellow); DMN, default mode network (periwinkle); ECN, executive control network (dark blue); HC, hippocampus (purple); LIM, limbic network (light blue); SMN, somato-motor network (orange); Th, thalamus (pink); VAN, ventral attention network (green); VIS, visual network (red).
such as different brain atlas parcellations (Doucet et al., 2019) and sample size and diversity (Marek et al., 2022), or different preprocessing steps (Power et al., 2012). Ultimately, this network did not show any group effects and future study is needed to validate this finding.
While our study has several strengths, such as a large sample size of PWH and the use of parallel analyses involving both network and region-based approaches, it also has a number of limitations that must be considered. Firstly, while our dataset of PWH throughout aging is relatively large given the population, much larger samples are needed to ensure replicability and address generalizability (Marek et al., 2022).
While we performed a cross validation with our data, further studies using an independent dataset are needed to confirm these results are robust. Similarly, we chose a relatively lenient threshold for head motion exclusion and did not use global signal regression (GSR) in our main analyses. Regarding head motion, there is currently no clear consensus on what exact threshold should be used and it is relatively accepted that it needs to be adapted to each study/clinical sample.
Accordingly, in line with previous literature (Ciric et al., 2017), we chose instead to use a censoring approach (censored volumes were regressed out) and we further used the mean head motion as a covariate in all group analyses. As per GSR, the use of GSR in preprocessing has also been a relatively controversial step and previous studies have suggested that GSR may distort group differences (Saad et al., 2012), as it might remove meaningful FC differences between clinical versus healthy groups. In this context, we focused on the main effect of age on FC and found relatively consistent results, even after GSR ( Figure S1). The main difference was detected in the limbic network, with a reverse pattern in PWH. However, we believe this may be partially due to the instability of the limbic network, as shown in studies suggesting this network has relatively low reliability and high intersubject variability, due to its location in the lower part of the brain where the signal-to-noise ratio is often lower than the other brain networks (Chen et al., 2015;Sbaihat et al., 2022). Additionally, we identified a weak relationship between distance and motion correlation, suggesting long versus short range connections may be differentially impacted by motion (Ciric et al., 2017). It will be important for future studies to adopt more strict criteria for motion exclusion, as well as test the impact of GSR on resting-state FC in an independent sample of PWH. Secondly, as mentioned previously, our sample of PWH had very well-managed HIV-infection in the form of effective cART and undetectable viral loads. Our sample also had no other substantial neurologic/psychiatric comorbidities, and therefore, these data may not generalize well to a broader population of PWH that may have a variety of other complications. Second, we did not examine common health comorbidities such as obesity, hypertension, or diabetes, nor did this study assess the effects of socioeconomic factors or health habits. Further study is needed to examine the impact of these other common health factors on our findings. Finally, future studies should examine older adults (beyond age 72) and PWH with more severe types of HAND.
In conclusion, HIV infection may be related to a reorganization of

ACKNOWLEDGMENTS
We thank the participants and all the study staff. We also acknowledge the significant contribution from Kevin R. Robertson, PhD, professor of neurology and director of the AIDS Neurological Center at the University of North Carolina at Chapel Hill. Dr. Robertson designed the neuropsychological testing battery and analyzed the data, and sadly died during the conduct of the study. We also acknowledge the enormous contribution of Dr. Susan Swindells who advised on the clinical aspects of the study and was central to our recruitment and study management. Dr. Swindells retired during the study. The data that support the findings of this study are available from the corresponding author upon reasonable request.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.