Alteration in topological properties of brain functional network after 2‐year high altitude exposure: A panel study

Abstract Introduction High altitude (HA) exposure leads to cognitive impairment while the underlying mechanism is still unclear. Brain functional network is crucial for advanced functions, and its alteration is implicated in cognitive decline in multiple diseases. The aim of current study was to investigate the topological changes in HA‐exposed brain functional network. Methods Based on Shaanxi‐Tibet immigrant cohort, neuropsychological tests and resting‐state functional MRI were applied to evaluate the participants' cognitive function and functional connection (FC) changes, respectively. GRETNA toolbox was used to construct the brain functional network. The gray matter was parcellated into 116 anatomically defined regions according to Automated Anatomical Labeling atlas. Subsequently, the mean time series for each of the 116 regions were extracted and computed for Pearson's correlation coefficients. The relation matrix was further processed and seen as brain functional network. Correlation between functional network changes and neuropsychological results was also examined. Results The cognitive performance was impaired by HA exposure as indicated by neuropsychological test. HA exposure led to alterations of degree centrality and nodal efficiency in multiple brain regions. Moreover, two subnetworks were extracted in which the FCs significantly decreased after exposure. In addition, the alterations in FCs within above two subnetworks were significantly correlated with changes of memory and reaction time. Conclusions Our results suggest that HA exposure modulates the topological property of functional network and FCs of some important regions, which may impair the attention, perception, memory, motion ignition, and modulation processes, finally decreasing cognitive performance in neuropsychological tests.


| INTRODUC TI ON
High altitude (HA) exposure is becoming more and more common as the increased human activities in high plateaus and mountains.
Most human, even those who lived in plateau for generations (Ebert-Santos, 2017;Hill et al., 2016), are not totally adapted to HA exposure, as indicated by headache, sickness, poor sleep quality (Sakamoto et al., 2017), cognitive impairment , polycythemia, pulmonary hypertension (Robinson et al., 2017), right ventricular hypertrophy, and fertility reduction (Parraguez et al., 2013;Simonson, 2015;Whayne, 2014). With the huge demand of heath care for people exposed to HA, various studies were carried to explore mechanisms underlying HA-induced health challenges.
Notably, cognitive impairment was common especially in those undergoing long-term HA exposure and had great influence on operation capability of people working under HA environment, while the pathophysiological mechanism underlying such HA-induced alteration remains largely unclear.
Neuroimaging studies were widely employed in this field aiming at clarifying such phenomenon. Previous MRI studies have revealed some structural and functional changes after HA exposure which may partially explain the cognitive impairment. A previous study investigated 16 young man lived in high altitude for two years, revealing that HA subjects have significant changes of amplitude of low frequency fluctuation (ALFF), an indicator of spontaneous brain activity in the functional MRI signal, at bilateral occipital cortices, right anterior insula and extending to the caudate, putamen, inferior frontal orbital cortex, temporal pole, and superior temporal gyrus when compared with matched sealevel natives . Our previous study followed 69 participants who immigrated to Tibet for 2 years, finding that chronic HA exposure decreased regional homogeneity in the left putamen, superior temporal gyrus, superior parietal lobule, anterior cingulate gyrus, and medial frontal gyrus, while increased regional homogeneity was found in the hippocampus. Moreover, the functional connection (FC) between left putamen and superior temporal gyrus, anterior/middle cingulate gyrus, and other brain regions weakened after exposure .
Notably, human brain is particularly complex and different brain regions interact with each other to form a complicated network for advanced functions (Cao, Huang, & He, 2017). Among various network analysis approaches, graph theory analysis has been increasingly emphasized (Du et al., 2018). The graph theory analysis exacts mathematical representation of complex networks to form a framework comprised of nodes and edges. Each node represents a relatively independent brain region, and each edge refers to the FC among two nodes (Wang et al., 2015). The human brain network has some important topological properties which have been found to be associated with human neurological disorders, for example, degree centrality, a robust index of focal connectivity in which the number of direct connections from one node to all nodes is counted, reflecting its relative importance within a network (Yang, Dong, & Chawla, 2014). When compared with healthy controls, patients with bipolar disorder showed significant decrease in degree centrality in cortical regions including the middle temporal pole, inferior temporal gyrus, and ventral prefrontal cortex, which may be the underlying neural mechanisms in bipolar disorder (Zhou et al., 2017). The nodal efficiency of the network refers to the averaged reciprocal shortest path length between the node and the other nodes, representing the ability of information transfer from itself to other nodes in the entire network (Achard & Bullmore, 2007). Similarly, the alteration in nodal efficiency is also observed in multiple psychiatric and neurological disorders. Based on previous studies, it is intriguing to investigate whether HA exposure leads to topological abnormalities in the functional brain networks and whether such abnormalities are correlated with cognitive decline.
Our previous study found that some connections between different brain regions were changed after HA exposure by using seed-based analysis . However, there was still no study about the changes in FCs from a perspective of whole brain networks. Notably, in current study, we constructed the functional brain networks and compared the topological properties before and after exposure. We hypothesized that chronic HA exposure may alter the topological properties in cognitive-related brain regions and FCs among these regions would be weakened. Moreover, the topological and FC changes may be correlated with HA-induced neuropsychological changes. To the best of our knowledge, this is the first study focusing on the chronic HA exposure induces brain functional network changes through graph-based theoretical approaches.

| Participants
The study was ratified by the Ethics Committee of the Medical Faculty of Air Force Medical University (No. KY20143344-1), and written informed consent was obtained from each participant. We always abided by the Declaration of Helsinki during the whole process of research.
All 69 participants in current study came from Shaanxi-Tibet immigrant cohort (STI cohort). The detailed information about this cohort has been described in our previous article .
The 69 young healthy high school graduates from Shaanxi (400 m) were admitted into Tibet University (3,650 m) for higher education and recruited to the STI cohort study which was launched in 2014.

| Neuropsychological test
We conducted four tests to monitor the memory ability and reaction of each participant, including (a) verbal memory test, which was design to measure working memory for words, including the exami-

| MRI data acquisition
Functional MRI data were acquired with General Electric Discovery We strictly employ the same MRI parameters in the whole process in order to ensure the comparability of the images. Standard T1-weighted 3D anatomical data were acquired with the 3D magnetization-prepared rapid gradient echo (3D MPRAGE) sequence The following echo-planar imaging (EPI) sequence which covers the entire brain was applied during resting-state functional MRI data acquirement: repetition time: 2,000 ms; echo time: 30 ms; flip angle: 90°; field of view: 220 mm × 220 mm; acquisition/reconstruction matrix: 128 × 128; slice thickness: 4 mm; section gap: 0.6 mm; number of slices: 30, voxel size = 1 × 1 × 1 mm 3 , scam time: 6 min; and total volumes: 180. We use earplugs and custom-built head coil cushions to dampen scanner noise and minimize head motion.
During data acquisition, subjects were asked to remain alert with eyes closed and to keep their head still.

| Data preprocessing
The fMRI data were preprocessed using GRETNA (v2.0) (GRETNA, RRID:SCR_009487) (Wang et al., 2015). In summary, the preprocessing includes (a) converting DICOM files to NIfTI format; (b) discarding the first 10 volumes of NIfTI format; (c) slice-timing correction; (d) the functional MRI data were realigned to correct for head motion; (e) the T1 images were further utilized for normalization and segmentation, and each subjects' brain image was segmented into gray matter, white matter, and cerebrospinal fluid; (f) functional images were normalized to standard MNI space and smoothed with a 6-mm full width at half maximum Gaussian kernel; (g) linear detrend and band-pass filtering (0.01-0.08 Hz) were conducted; and (h) regressing several nuisance signals including head motion, global mean, and signals from the cerebrospinal fluid and white matter from the data. For the before exposure group, six subjects were excluded for some reasons: one for the artifact of T1 images, four for excessive head motion, and another one for unsatisfied normalization. For the after exposure group, 10 subjects were also excluded for some reasons: one for the artifact of functional images, four for excessive head motion, and five for unsatisfied normalization.

| Brain network construction
GRETNA, a graph theoretical network analysis toolbox for imaging connectomics, was employed for construction of functional brain network (Wang et al., 2015). The whole cerebral and cerebellar gray matter was parcellated into 116 anatomically defined regions according to Automated Anatomical Labeling (AAL) atlas.
Subsequently, the mean time series for each of the 116 regions were extracted and computed for Pearson's correlation coefficients. Fisher's z transformation was performed to turn the data into a z-value which was close to a normal distribution. Moreover, the graphic model of the brain functional network was constructed by a binary connection matrix, which was converted by the z-values with a series of threshold (Sparsity: 5%-50%) of the relation matrix.

| Global and nodal network properties
The brain network is seen as a graph in math. The mathematic study of graph is named as graph theory. In this study, applying graph theory to brain network, each brain region is defined as a node and each connection between two nodes is defined as an edge in graph. Thereby, the network property can be quantified by some indexes in graph theory. Generally, these indexes can be categorized from global and nodal perspectives. Degree of a given node means the number of edges connecting it to the rest of the graph. It is easy to understand that there are many paths between two nodes in graph. The optimal path which has minimal edge numbers is named as shortest path length. For two brain regions, it is easier to communicate with each other when the shortest path length is smaller. The efficiency of a given node is inversely proportional to its shortest path length to other nodes in whole graph. Besides, the local efficiency for a given node measures how efficient the communication is among the first neighbors of this node when it is removed. The centrality captures the importance of a given node (Hagmann et al., 2008), which can be calculated from degree or betweenness. A node with higher degree also has higher degree centrality, which reflects its information communication ability in graph. Moreover, a node with high betweenness centrality lies on many of the shortest paths that link other nodes in the network to one another, so it has significant impact on information flow between other nodes. The clustering coefficient of a given node measures the likelihood its neighborhoods are connected to each other. Modularity refers to the existence of multiple densely connected communities of regions in a brain network. The participant coefficient reflects the ability of an index node in keeping communication between its own module and the other modules.

| Statistical analysis
To determine the significant before-after differences of all the network matrices (global and nodal properties and the AUC of each network metric), paired t test analysis was used in current study. FDR correction of p < .05 was applied to control for multiple comparisons in nodal properties. The internodal connections were also compared by paired t test. Network-based statistic (NBS) was used to control the error rate (Edge p = .0001, Component p = .01, Number of iterations = 1,000). Moreover, Pearson's correlation was performed to determine the relationship between brain functional network alterations and neuropsychological changes. The statistical analyses were performed with GRETNA and GraphPad (GraphPad Prism, RRID:SCR_002798). p < .05 was considered statistically significant.

| HA exposure induced cognitive function impairment
The current study enrolled young adults who immigrated to HA region for higher education. We followed them for 2 years aiming F I G U R E 1 The changes of regional nodal properties after high altitude (HA) exposure. Nodes with increased degree centrality or nodal efficiency are shown in red. Nodes with decreased degree centrality or nodal efficiency are shown in blue. The figure is made by Brain Net Viewer (http://www.nitrc.org/proje cts/bnv/). (a) Degree centrality and (b) node efficiency at evaluating HA-induced cognitive and brain functional network alteration. The result of neuropsychological tests revealed that HA exposure decreased accuracy in the verbal/visual memory test and prolonged responding time in the visual/auditory reaction time test.
The details about cognitive function impairment were described elsewhere .

| All global network properties remained intact after exposure
Global metrics we focused in this study included small-world parameters, local efficiency and global efficiency, modularity, rich club, assortativity, synchronization, and hierarchy. By using GRETNA, we calculated and compared these metrics before and after exposure.
As a result, there were no significant changes among all global network properties.

| Degree centrality and nodal efficiency changed in some regions after exposure
Alterations in degree centrality were revealed in 20 brain regions (|t| > 2.9878, p < .05 FDR corrected). Among them, the degree centrality of 12 regions decreased and eight regions elevated after exposure. See details in Figure 1 and Table 1.
Alterations in nodal efficiency were revealed in 23 brain regions (|t| > 2.8255, p < .05 FDR corrected). Among them, the nodal efficiency of 14 regions decreased and nine regions elevated after HA exposure. See details in Figure 1 and Table 2

| FCs within two subnetworks changed after exposure
Apart from above network metrics, some FCs were also changed in response to HA challenges. NBS correction revealed two subnetworks with statistically significant changes ( Figure 2 and Table 3).
Notably, all FCs were decreased in both subnetworks (Figure 2).

| Significant correlations between brain network alterations and cognitive changes
To understand the effects of HA exposure on brain network and function, we performed correlation analyses between brain network alterations and cognitive decline. Notably, some alterations in FCs within above two subnetworks were associated with changes of memory and reaction time (Figure 3a

| D ISCUSS I ON
High altitude-induced cognitive impairment is an important public health problem remaining to be solved in HA regions. Extensive efforts have been made to clarify the underlying mechanisms while limited progress has been made in past years. In this study, by investigating HA-induced changes in topological properties of brain functional networks, we revealed some topological and FC changes after two-year HA challenge, which may be the mechanism underlying HA-induced cognitive impairment.
First of all, there were no significant changes in global metrics which include small-world parameters, local efficiency, global efficiency, modularity, rich club, assortativity, synchronization, and hierarchy. Such phenomenon may suggest that two-year HA exposure was not enough to cause topological properties changes from global perspectives.
However, according to our results, some nodal metrics were significantly altered after two-year HA exposure (Tables 1 and 2). Twenty brain regions were observed to have changes with regard to degree centrality, and 23 brain regions changed in nodal efficiency.
Notably, nearly all brain regions with changed degree centrality also have nodal efficiency alteration. Moreover, the alteration in nodal degree and efficiency was consistent in each region. reported to not only influence spatial and nonspatial short-term memory but also involved in sound localization discrimination or auditory object discrimination (Lancelot et al., 2003). The cerebellum, which is crucial for motor coordination, is also influenced after HA exposure. Collectively, two-year HA exposure decreased the degree centrality and nodal efficiency of some crucial regions that regulate attention control, reaction, memory, and motor coordination, finally leading to worse performance in neuropsychological tests. Notably, most of abovementioned regions located in the left hemisphere, suggesting that left hemisphere is more vulnerable under hypoxic condition, which is consistent with a previous study (Yoo, Shin, Chang, & Caplan, 1998). The difference of vascular structure between left and right hemisphere may account for such left-preferring phenomenon. In pre-and postmortem studies, the mean vertebral artery diameter was found to be larger on the left side than the right side (Khan, Cloud, Kerry, & Markus, 2007).
So we speculate that left hemisphere may be accustomed to massive blood and oxygen supply, which may make it more vulnerable to hypoxic condition.
The degree centrality and nodal efficiency increased in several brain regions, including right orbital middle frontal gyrus, right middle occipital gyrus, right inferior occipital gyrus, left superior parietal gyrus, bilateral inferior parietal gyri, and bilateral angular gyri.
Most of these regions belong to association cortex which is an evolutionarily recent type of cortex for advanced information processing related to cognition (Khodagholy, Gelinas, & Buzsaki, 2017).

TA B L E 3
Subnetworks with changed functional connections (FCs) after HA exposure

Subnetwork Brain regions with FC Trends
Subnetwork 1 Putamen_L ↓

Vermis_8
Putamen_R ↓ Association regions integrate multiple functional systems, such as memory and attention systems, and are mainly involved in intelligent processing (Mesulam, 1998). We inferred that the improvement in association cortices could be a compensation to the decline of some nodal metrics in cognition-related regions.

Vermis_8
Apart from these, two changed subnetworks were extracted after exposure  (Zhao et al., 2017). Thereby, we propose that the decrease in FCs between left anterior cingulate gyrus and bilateral temporal lobes may impair the auditory perception and comprehension, finally influencing performance in neuropsychological tests. In addition, the FCs of motor and memory-associated regions also decreased, which influence the test score from other aspects. In subnetwork 2, the hubs turned to be some cerebellar regions, including parts of cerebellar hemisphere and vermis. The other sides of these connections are components of bilateral basal ganglia. As is well known, the cerebellum and basal ganglia function as two pivotal regions for motor coordination. The lesions occurred in cerebellum, and basal ganglia could lead to motor coordination impairment, such as cerebellar ataxia (Slapik et al., 2019) and Parkinson's disease (Rolinski et al., 2016). Moreover, recent studies revealed that cerebellum and basal ganglia interacted with each other during predictive motor timing which is crucial flexible movement adjustment (Caligiore et al., 2017;Husárová et al., 2013;Kunimatsu, Suzuki, Ohmae, & Tanaka, 2018). Thereby, the subnetwork 2 mainly connects two motor coordination regions, namely cerebellum and basal ganglia, bearing the responsibility of motion response in neuropsychological tests. Thereby, the HA-induced FCs decrease in subnetwork 2 may be associated with the slower response in simple reaction time task from a motor execution perspective. What's more, the alterations in FCs within above two subnetworks were significantly correlated with changes of memory and reaction time.
As shown in Part A of Figure Figure 3 were relatively difficult to understand. The FC changes between right cerebelum_3 and bilateral putamen positively correlated with ASRT change, which indicated that HA induced these FC decreases may shorten ASRT. However, actually, the ASRT significantly prolonged after HA exposure. We speculate that the prolonged ASRT may be caused by other alteration, like decreased regional homogeneity in right superior temporal gyrus . The alterations in FC between right cerebelum_3 and bilateral putamen may be inadequately compensatory responses to HA-induced ASRT prolongation. Moreover, the DVBM change negatively correlated with FC change between vermis_3 and right putamen. Similarly, this decreased FC may also be an inadequately compensatory response to HA-induced memory decline. However, such speculations still need to be examined in future studies.
Notably, there are some limitations regarding current study. First of all, this is a self-control study aiming at clarifying HA-induced brain functional network changes, while it lacks a sea-level control cohort to be more rigorous. We plan to establish a sea-level college student cohort in future studies. Second, this study only included young adults while other people who also endure HA challenges in HA regions, like children, adults, and the olders were not considered in current study. Thereby, further studies may focus on different population and reveal the alterations in their brain functional networks after HA exposure. What's more, the AAL template was employed in current study to define the nodes of functional network.
However, this template is not fine enough for its relative low resolution. The more complex templates, like probabilistic atlas, should be applied in future studies.

| CON CLUS ION
In our current study, we constructed the functional brain network and compared the topology properties before and after 2-year HA exposure. We found that HA-exposed young adults exhibited altered degree centrality and nodal efficiency in multiple brain regions which were involved in attention control, reaction, memory, and motor coordination. Moreover, two changed subnetworks were extracted after exposure. The subnetwork 1 connected brain regions controlling auditory perception, comprehension, motion initiation, and memory. The subnetwork 2 connected the two motor coordination system, which helps to regulate motion response in neuropsychological tests. In a word, through modulating the topological properties of functional network and FCs of some important regions, HA exposure may impair the attention, perception, memory, motion ignition, and modulation processes, finally decreasing cognitive performance in neuropsychological tests.

ACK N OWLED G M ENTS
This study was financially supported by the National Science Foundation of China (No. 8133045, 81730053, 81803194, 81502770) and the Military Logistics Research Project (No. AWS14L008, AWS16J022, AWS17J013).

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

AUTH O R S ' CO NTR I B UTI O N S
W-JL, J-YC, and W-BZ contributed equally to the conception and design of the study. Z-LX and X-MC were responsible for data processing and manuscript drafting. Z-LX, X-MC, QZ, J-YW, and Y-BX collected MRI and cognition data. QZ, J-YW, JL, B-JL, and X-RD contributed to the data analysis. All authors revised and approved the final draft of this article.

E TH I C A L A PPROVA L
The study was ratified by the Ethics Committee of the Medical Faculty of Air Force Medical University (No. KY20143344-1), and written informed consent was obtained from each participant. We always abided by the Declaration of Helsinki during the whole process of research.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.