Structural network alterations induced by ART-naive and ART-treated subjects infected with HIV.

OBJECTIVE
To investigate how the structural connectivity altered in combined antiretroviral therapy-treated (cART+) HIV patients and cART-naive (cART-) HIV patients by conducting Network analysis of Diffusion Tensor Imaging (DTI) data.


METHODS
We enrolled 22 cART-, 23 cART+ and 28 normal controls (NC) in our current study. Firstly, the DTI imaging data pre-processing was conducted and the asymmetric 90 × 90 matrix for each participant from their DTI data was obtained with the use of PANDA. Then, we applied a graph-theoretical network analysis toolkit, GRETNA v2.0, to calculate metrics such as small-"worldness," characteristic path length, clustering coefficient, global efficiency, local efficiency, and nodal "betweenness". Finally, we took comparisons among the three groups to investigate topological alterations.


RESULTS
Results (1) the regional characteristics (nodal efficiency) were altered in cART- and cART+ patients predominantly in the frontal cortical regions; (2) changes in various network properties in cART+treat and cART-patients were associated with the performance of behavior functions; (3) Hubs redistributed in HIV subjects especially in cART+ patients.


CONCLUSION
The regional characteristics (nodal efficiency) were altered in cART- and cART+ patients predominantly in the frontal cortical region, and changes in various network properties in cART- and cART+ patients were associated with the performance of behavior functions. In addition, Hubs redistributed in HIV subjects especially in cART+ patients.


Conclusion
1) The regional characteristics (nodal e ciency) were altered in cART-naïve and cART-treated patients predominantly in the frontal cortical regions; (2) changes in various network properties in cART-naïve and cART-treated patients were associated with the performance of behavior functions; (3) reduced network segregation was associated with lower current CD4 count in cART-naïve participants, suggesting that brain network segregation may have been adversely affected by a history of greater immune suppression.
(4) Hubs redistributed in HIV subjects especially in cART-treated patients.

Background
The human immunode ciency virus (HIV) invades the central nervous system(CNS) Shortly after seroconversion (~ 8 days) following signi cant immune suppression and notable cognitive impairment. [1] Combination antiretroviral therapy (cART) has been reported to have extensively decreased the morbidity and mortality of HIV-infected individuals and extended their life expectancy, but HIV-associated neurocognitive disorders (HAND) are still prevalent. [2][3][4] Currently, it is not fully understood why cognitive impairment persists in HIV infected subjects who have been virologically suppressed for several years. [5][6][7][8] One possible interpretation may be that irreversible CNS impairment takes place quite early after HIV infection, which causes subtle cognitive alterations measurable by comprehensive neuropsychological tests. In such case, cART would effectively prevent the progression of CNS damage but not reverse occured injury. [9] However, there are also HIV patients with normal cognition even after long term HIV infection with cART. By studying their injury characteristic may help in our further understanding the pathomechanism of HIV infection.
Nowadays, noninvasive magnetic resonance imaging (MRI) provides idle measures to detect early alterations in HIV infected subjects. As one of them, Diffusion tensor imaging (DTI) is capable of detecting and measuring white matter integrity and ber connectivity in vivo. Previous studies have found extensive white matter impairments (including corticospinal tracts, the cerebral peduncle, the cerebellar peduncles, the internal capsule, the uncinated fasciculus, the corona radiate and the corpus callosum (CC) etc). [10][11][12][13][14][15] The conbination of graph theory and DTI analysis, has provided interesting insights into the organization of the complex and comprehensive brain network. Network analysis based on the white matter network has shown that HIV-positive participants presented with disrupted global and nodal connectivity in structural connectome, [16,17] preferential alterations in clustering coe cient, path length, and local e ciency, indicative of structural segregation and structural integration. However, little is known about how network metrics change in HIV-positive individuals with normal cognition after regular cART, and what the differences are between the so-called "cognition-negative" and the healthy controls.
In our study, we initially combined DTI-based tractography and graph theoretic methods to explore global and nodal alterations among three groups: HIV-positve subjects with regular cART treatment. HIV-positive naive to treatment, and HIV-negtive controls, aimed at investigating differences of topological organization among three groups. By doing so, we not only discovered effective biomarkers for earlier diagnosis, we also could monitor the effect of cART on HIV-positive individuals. We then examined the associations between network metrics and cognitive performance. We hypothesized that except for the path length, topological metrics would show positive correlations with the cognition. Moreover, the cART group may not have displayed a better result due to the existing brain damage.

Subjects and data acquisition
From March 2016 to November 2016, a total of 73 subjects: 22 cART-naïve, 23 cART-treated, 28 normal controls, were recruited at Capital Medical University Beijing You An Hospital in China. This study was approved by the Ethics Committee of Beijing You An Hospital. All the Participants had submitted a written, signed, and dated informed consent after acknowledging of this study. The exclusion criteria for HIV subjects included: history of alcohol or substance abuse, history of neurodegenerative disease except for HIV (including Alzheimer's disease), diagnosis of neurological comorbidities (including leukoencephalopathy or HIV encephalitis), obvious brain lesion (including tumors or trauma), The clinical markers, such as current CD4 + cell counts, CD4+/CD8 + ratios, and plasma viral loads were collected for all patients. The duration of HIV infection was con rmed by patients' self-reports. The NC subjects were recruited from the same urban regions as the patients. The exclusion criteria for NC subjects included: obvious brain lesions (such as trauma or tumors), any history of previous neurodegenerative diseases (such as diabetes, and Alzheimer's disease), or other systemic diseases.

Neuropsychological evaluation
The Neuropsychological test conducted for each participant had been validated for HIV positive populations. The following six competency areas were evaluated with reference to the Frascati guidelines: (1) speed of information processing using the tracking test part A; (2) use of the same pair of continuous performance tests (CPT-IP), Wechsler memory Table III (WMS-III), pacing auditory serial addition (PASAT) for attention/working memory; (3) motor function using grooved nail plate test; (4) using Wisconsin Card Sorting Test-64 (WCST-64 Abstract/execution function; (5) learning and recall using the revised Hopkins Language Learning Test (HVLT-R); short visual space memory test revision (BVMT-R); (6) using category Fluency and language/language for animal naming tests. To further assess cognitive functions, composite cognitive scores were created to evaluate the six domains tested. The raw scores for each test were normalized to T scores and adjusted for age, gender, and education. The T scores of each cognitive domain in one or more tests are averaged over the domain to calculate a T score for a particular domain and then further applied to the correlation analysis. Additionally, cognitive impairment is diagnosed as ANI in patients with at least two cognitive domains that exhibit at least 1 SD below the mean of neuropsychological testing.

Image acquisition
For each participant, both T1-weighted MRI and DTI were obtained on a 3.0T Siemens scanner (Allegra, Siemens Medical System, Erlangen, Germany) using a 32-channel phased array coil. A standard birdcage head coil and restraining foam pads were used to minimize head motion. A Structural T1-weighted MRI was acquired with a spoiled gradient recall sequence with the following parameters: Slices = 176, TR = 1900 ms, TE = 2.2 ms, inversion time (TI) = 900 ms, FA = 9°, eld of view (FOV) = 256 × 256 mm, acquisition matrix = 256 × 256, and thickness = 1 mm. For DTI, we used single shot echo planar imaging (EPI) sequences in contiguous axial planes covering the whole brain with the following parameters:in 32 independent, non-collinear directions of a b-value = 1000s/mm2 and one additional image with no diffusion weighting(b = 0), slices = 60, TR = 11000 ms, TE = 98 ms, FA = 90°, FOV = 256 mm × 256 mm, acquisition matrix = 128 × 128, and thickness = 2 mm.

Image processing
Imaging processing and network construction was performed on PAND, procedures include: Data converses from DICOM to NIfTI; the correction of Eddy current and head motion of FDT diffusion imaging data; the brain structure and tissue extraction; Realignment; the calculation of Fractional anisotropy; Diffusion tensor tractography. Tractography was operated to construct 3-D streamlines which was representative of ber tract connectivity. [18] Network metric construction Firstly, according to the anatomical automatic labeling atlas, the white matter brain network was constructed by calculating the paired Pearson relative coe cients between 90 regions of interest [19]. Further, whole-brain and regional network topological properties (including small-worldness(Sigma), global e ciency (Eg), clustering coe cient (Cp), nodal betweenness (Bnod), characteristic path length (Lp) and local e ciency (Eloc)) were calculated.

Statistical Analysis
We conducted statistical analysis with SPSS v20.0. All the demographic data and neuropsychological scores except for sex variables (with a Pearson chi-square test) were examined with the use of one-way ANOVA. Group effects of global and regional network metrics were performed among 3 groups using oneway ANOVA. Further, we explored the correlations between network measures and neuropsychological scores by partial correlation analysis with demographic variables, as covariates.
The nodes with signi cant group differences were performed in order to identify the correlation between neuropsychological test scores with speci c brain regions.

Results
Global topology of the white matter connectome All the three groups did not present signi cant small-world organization. ANCOVAs showed that there are no signi cant group effects on global e ciency(P = 0.380), characteristic path length(P = 0.279), and clustering coe cient(P = 0.433).
Regional characteristics in anatomical brain networks in CART-, cART-treated and control patients The regional alterations in cART-treated, cART-naïve and NC networks were found to have signi cantly altered nodal characteristics (i.e., betweenness centrality, local e ciency, path length, and clustering coe cient) in cortical regions, which were mainly located in the right hemisphere. We found that betweenness alterations predominately located in the prefrontal lobe (e.g., OLF and SFGmed) and temporal lobe (e.g., TPOsup) (Fig. 1). We also found that clustering coe cient widely described across occipital (e.g.,LING), parital (e.g., POCG), subcortical (e.g., CAU) and prefrontal (e.g., SFGmed)lobes (Fig. 2). In addition, signi cant local e ciency differences mainly concentrate on subcortical (e.g., CAU) and prefrontal (e.g., SFGmed) lobes (Fig. 3). Finally, path length changes were all discovered in pareital lobes (e.g., POCG, PCG, SPG, IPL) (Fig. 4) Differences in betweenness centrality, local e ciency, path length and clustering coe cient There were no signi cant changes among the three groups. ANOVAs presented signi cant group differences in betweenness connectivity, path length, clustering coe cient, and local e ciency. We further localized the nodes with changed betweenness centrality among the three groups. Regions with signi cant differences across the three groups were located in left TPOsup, right SFGmed and OLF. Post hoc tests showed an increase in SFGmed in cART-treated group and TPOsup in cART-naïve group versus the control group. In addition, a decrease in OLF in the CART-group than cART-treated ( Fig. 1) We also investigated the nodes with alternated clustering coe cient and excitedly we found signi cant changes in nodes located in left POCG, right LING, CAU, and SFGmed. Also, the following post hoc test indicated an increase in left POCG and right LING in cART-treated and cART-naïve than control. Increases in CAU were also found in CART-naïve than in cART-treated. On the other hand, decreases were found in SFGmed in cART-treated and CART-naïve group versus the control. (Fig. 2) As for the local e ciency, we interestingly discovered that nodes that lied in left ACG, right CAU and SFGmed displayed signi cant transformation. However, the post hoc test only prompted an increase in SFGmed and CAU in control group compared with cART-treated group. Furthermore, there was a decrease in CAU in cART-treated verus in the cART-naïve. (Fig. 3) Finally, the signi cant length path changes mainly concentrated on the right part of the brain, for instance, PCG, POCG, SPG and IPL. The post hoc test indicated an increase in PCG, POCG, SPG and IPL in the cART-treated group versus in the control group. Otherwise, there was a decrease in the cART-naïve group as compared with the cART-treated group. (Fig. 4)

Hub regions
In the control group, ve regions including three hetero modal or unimodal association cortex regions, two subcortical cortex regions were identi ed as the hubs because of large values in betweenness (Table 2). In the cART-naïve group, 10 regions including ve heteromodal or unimodal association cortex regions and four subcortical cortex regions and one temporal cortex were identi ed as the hubs (Table 3). In the cART-treated group, 15 regions including seven heteromodal or unimodal association cortex regions, four subcortical cortex regions, two paralimbic cortex regions and two primary cortex regions were identi ed as the hubs (Table 4) Correlation between topological metrics and clinical markers By examining the relationship between the signi cant topological metrics and the current CD4 T cell counts, we found that in the cART-naïve group, the left POCG in clustering coe cient was negatively correlated with the current CD4 T cell counts, while CAU in the clustering coe cient was positively correlated with the current CD4 T cell counts. However, in the cART-treated group, CAU in local e ciency was negatively correlated with the current CD4 T cell counts (Fig. 5)

Correlation between topological metrics and cognitive performances
We further examined the relationship between network metrics and cognitive performances, and dedicated that the cART-naïve changes of topological metrics were signi cantly associated with declined cognitive functions. We also found that cART-naïve subjects with decreased clustering coe cient (e.g., SFGmed.R, LING.R) in WM network had lower attention/working memory, speed of information processing and abstract/executive function, Also, cART-naïve subjects with decreased local e ciency of several cortical regions (e.g., SFGmed.R) in WM network had lower speed of information processing functions and abstract/executive function, while patients with longer shortest path length (e.g., PCG.R, POCG) had lower motor function. There was an extraordinary phenomenon discovered in which patients with lower local e ciency had higher memory (e.g., CAU.R) For cART-treated patients, we saw quite the opposite result. The longer shortest length path was positively correlated with the increased speed of information processing, abstract/executive function, and attention/working memory. Increased local e ciency was correlated with lower abstract/executive function. Clustering coe cient was negatively correlated with memory. There was also an extraordinary phenomenon discovered in which the Clustering coe cient was positively correlated with motor function (Fig. 6)

Discussion
Our current study combined diffusion tensor Image and graph theory to demonstrate changes in the organization and segregation of structural networks in cART-naïve and cART-treated subjects. Our main ndings were as follows: (1) the regional characteristics (nodal e ciency) were altered in cART-naïve and cART-treated subjects preferentially in the frontal cortical regions; (2) alterations in some topological metrics in cART-naïve and cART-treated patients correlated with cognitives performances; (3) reduced network segregation was associated with lower current CD4 T cell counts in cART-naïve patients, indicating that brain network segregation may have been adversely affected by a history of enhanced immunosuppression; (4) Hubs redistributed in HIV subjects especially in cART-treated patients. These ndings suggested that WM degeneration altered the structural connectivity pattern of WM network in HIV individuals, and cART failed to reverse the existing disruption of structural topology. However, this may, just slow the progression, so early diagnosis and treatment are imperative.
In our study there were no signi cant small-worldness among the three groups, which is quite different from extensive studies [20,21]. This may be due to our limited samples. Global differences in our study also failed to present results like previous studies. It is possible that the average time of infection was too short to present such outcomes.
The regional alterations in cART-treated, cART-naïve and NC networks were detected to have signi cantly altered nodal characteristics (i.e., betweeness centrality, local e ciency, path length and clustering coe cient) in cortical regions, which were mainly lied in the right hemisphere. We found that betweenness alterations preferentially lied in the prefrontal lobe (e.g., OLF and SFGmed) and temporal lobe (e.g., TPOsup). And previous studies have shown that these frontal regions exhibited HIV-related abnormalities in the WM integrity [22][23][24] and gray matter morphology. The temporal pole, which includes linguistic integration, emotion, and semantic memory, also veri ed that there was atrophy and neuronal loss [25][26][27][28][29] in HIV patients and SIV infected rhesus monkeys. We also found that clustering coe cient widely described across occipital, parital, subcortical and prefrontal lobes. In addition, signi cant local e ciency differences mainly concentrated on the subcortical and prefrontal lobes.
Finally, path length changes were all discovered in the pareital lobe.
According to most of the previous studies, a combination of shortest path lengths and high clustering coe cients gave rise to the most optimal network balance between segregation and integration [30].
However, HIV infection can be a disconnection element. Betweenness centrality measures the importance of nodes for information transmission [31]. Our ndings of betweeness centrality alterations among cARTtreated, cART-naïve and control groups indicated that cART-treated had signi cantly higher betweenness centrality than the cART-naïve group, which suggested that prescribed cART does improve structual connectivity to some extent, which is inconsistent with previous studies [32]. However, the right SFG and the right TPOsup showed increased betweenness centrality, which is quite opposite to previous research. We speculated that the increase of these nodes may be compensatory for the reduction of OLF centrality.
Clustering coe cient measures network segregation which re ects specialization of discrete brain regions or systems in conducting Speci c psychological operations [33]. In our study, the right SFGmed implied an increase in Clustering coe cient in control group compared with cART-naïve and cART-treated group, suggesting a stronger local specialization, which has been veri ed in several studies [32]. On the other hand, there was an obvious decrease in the Clustering coe cient in the control group than the other two groups. This may be due to a possible compensation for the decrease in SFG.
Local e ciency in the right ICU and the right SFGmed in the control group was signi cantly higher than that in the cART-treated group, indicating that the e ciency of information transformation was affected by the white matter disruption, which is quite consistent with the previous studies.
Short path lengths in the brain networks ensure e cient transmission or rapid transfer of information between or among remote areas considered to be the basis of cognitive processes [34]. The increase in HIV-related pathways may therefore re ect the disruption of neuronal integration across distant regions associated with impaired cognitive function [35]. These ndings indicated that CART may have differential effects on regional variability with regular treatment.
We further examined the relevance between network affairs and cognitive performances. Results suggested that the cART-naïve changes of network metrics were signi cantly associated with the decrease of cognitive functions, which is in accordance with the previous studies [36]. Meanwhile, for the cART-treated groups, we obtained quite the opposite results. The speculation for this phenomenon may be that the compensatory mechanism is reversed even during the ART era.
We further examined the relevance between network affairs and cognitive performances. Results suggested that the cART-naïve changes of network metrics were signi cantly associated with the decrease of cognitive functions, which is in accordance with the previous studies [36]. Meanwhile, for the cART-treated groups, we obtained quite the opposite results. The speculation for this phenomenon may be that the compensatory mechanism is reversed even during the ART era.
Among cART-naïve patients, lower current CD4 T cell counts were positively correlated with decreased clustering coe cient (eg, right CAU), indicating that historical immunesuppression plays a key role in brain network segregation. However, lower current CD4 T cell counts were negatively associated with the reduced clustering coe cient (e.g., left POCG). That may be explained by the compensatory mechanism of the neighborhood nodes. While among cART-treated patients current CD4 T cell counts were negatively correlated with right CAU local e ciency, indicative of thereserved compensatory mechanism with the application of ART.
We investigate alterations of hub pro le, which plays a crucial role in identifying the most relevant nodes to information tra c. To identify the hub regions, we examined nodal betweenness centrality of each cortical region across the three subject groups (see material and methods). We found that these identi ed hubs were preferentially located in regions of association cortex (PUT, STG, and MOG), indicating their paramount roles in the structural networks. When hubs among three groups were further analyzed, we interestingly found the phenomenon that hubs in the cART-naïve and cART-treated groups redistributed separately. According to the "hubs overload and failure" theory, brain disease can break down the optimal balance of network, diminishing the information handling of hub nodes. As a result, scenario rerouting and rewiring contributed to hubs overload followed by hub failure [37]. Furthermore, more hubs in cARTtreated patients were reproduced from prefrontal to occipital, apart from the cognitive injury resulting from pathological affairs, the cART patients received may also lead to hub redistributed in drugs vulnerable regions, which was suspected of higher metabolic activity from these regions [38] There are several limitations in our study. In the rst place, we constructed the structural network with deterministic tractography which has been widely used in previous studies. While it failed in accurately mapping out such bers as crossing bers and long-distance bers, the sample size was too small to draw an effective conclusion regarding our method, though we found signi cant regional alterations among the three groups. Moreover, The cross-sectional nature of our study made us unable to con rm progress of HIV and the effect of cART over time. Thereofore, a longitudinal study would be bene cial in future studies.

Conclusion
In conclusion, we employed diffusion tensor MRI tractography to construct WM networks of cART-naïve, cART-treated patients and NCs. However, there was no signi cant small-world organization, and global differences among the three groups were not obvious. Our study showed that HIV individuals on stable cART had decreasd betweeness centrality, local e ciency, clustering coe cient, and longer length path within certain nodes compared to cART-naïve. In one distance, the mixture of increased and decreased connectivity would result in alterations in network organization. Moreover, we found that the HIV subjects had decreased nodal e ciency of cortical regions preferentially lied in the frontal lobe. Furthermore, we also found that the changes of HIV subjects highly correlated with cognitive functions. Finally, hub regions were redistributed in HIV subjects, especially in cART-treated patients. Our ndings support the WM degeneration hypothesis of changed brain networks in HIV participants, and suggest that the changes in uence the cognitive performances of patients. This HIV infected human brain WM networks, based on topology-based analysis, provides a new approach to show patterns of structural disconnectivity in neuropsychiatric diseases. This current study also helps us better understand human brain neurodegenerative connectome in diseases. Authors' contributions Jiaojiao Liu and Hongjun Li designed the paper, Jiaojiao Liu Wrote the paper, Weiwang, Yuanyuan Wang, Mingming Liu, Ruili Li collect the data, Dan Liu participated in the data analysis, Hongjun Li and Quansheng Gao reviewed the paper. Table 1 Demographic and neuropsychological evaluation of CART-, CART+ patients and normal controls Table 2 Regions showing high betweenness in cortical networks of normal subjects Table 3 Regions showing high betweenness in cortical networks of CART-subjects