Changes in functional brain networks and neurocognitive function in Chinese gynecological cancer patients after chemotherapy: a prospective longitudinal study

The details of the study subjects have been previously described [24] but briefly described as Chinese adult women with a primary diagnosis of cervical, ovarian, or uterine cancer were ready for chemotherapy treatment. Age-matched women without cancer history were recruited as healthy controls. All age-matched healthy controls were recruited from among staff members at this hospital. This study obtained ethical approval from the ethics committees at both The Hong Kong Polytechnic University and The Third Affiliated Hospital of Guangzhou Medical University. All research participants joined this study voluntarily and provided written informed consent.

The details of neurocognitive function assessment have been previously described [24] but briefly summarized as follow: this study took the recommendation of the International Cognition and Cancer Task Force (ICCTF): three core neurocognitive tests of the Hopkins Verbal Learning Test - revised (HVLT-R), the Trail Making Test (TMT), and the Controlled Oral Word Association Test (COWA) were used [25]. This study used the Chinese version of the Auditory Verbal Learning Test - revised version (AVLT-R) [26]; the Chinese version of TMT and the Chinese version of the COWA [27], respectively. As attention and working memory were the most common neurocognitive dysfunctions in Chinese gynecological cancer patients [28], this study also included the WAIS-III Digit Span test for measuring attention and working memory [29]. Cognitive performance testing in the patient group was conducted at post-surgery and post-chemotherapy, respectively. The duration of cognitive performance tests for patients were four months. As recruiting healthy controls needs to be age-matched with patients, it took two months to recruit eligible healthy controls, the duration of cognitive performance tests for healthy controls was two months. AVLT tests consisted of three successive learning trials and other procedures of cognitive tests were conducted consistently in both groups.

According to the ICCTF recommendations for neuroimaging studies in cancer and cognition, a minimal set of MRI sequences should include an rs-fMRI and a high-resolution T1-weighted anatomical MRI scan to assess functional brain networks [6]. Whole brain rs-fMRI data were collected on a Philips 3.0 T scanner (Achieva; Philips, Best, The Netherlands), using an 8-channel SENSE head coil at The Third Affiliated Hospital of Guangzhou Medical University, China. Throughout the rs-fMRI data acquisition, patients were instructed to close their eyes and relax, but to remain in a maximally alert state. A T2-weighted gradient-echo EPI sequence was used to obtain the rs-fMRI scan. A total of 240 whole brain EPI volumes were acquired using the following parameters: TR = 2000 ms, TE = 30 ms, flip angle = 90°, in-plane imaging resolution = 3 × 3 × 3 mm, in-plane field of view (FOV) = 256 × 256 mm, slice thickness = 4 mm, axial slices = 33. The rs-fMRI scan time was 8 min 6 s. T1-weighted imaging was achieved for morphometric (GM volume, cortical thickness and surface area) analysis using three-dimensional fast spoiled-gradient recalled acquisition in steady state (3D-FSPGR) in 164 coronal slices with the following parameters: acquisition matrix = 256 × 256; TE = 3.8 ms; TR = 8.2 ms; flip angle = 7°; FOV = 256 mm × 256 mm; slice thickness = 1 mm; voxel resolution =1 × 1 × 1 mm. The 3D-T1 scanning time was 5 min 58 s.

The rs-fMRI images were preprocessed using GRETNA: a graph theoretical network analysis toolbox for imaging connectomes [30]. During the preprocessing process, the first 10 volumes for signal were removed to reach a steady state, leaving 230 functional volumes for each subject. The remaining functional volumes were corrected for acquisition time delay between slices (slice timing) and head motion between volumes (realignment). Other steps in preprocessing these functional data consisted of spatial normalizing by DARTEL (warping individual functional images to the standard MNI space by applying the transformation matrix that can be derived from registering the final template file), spatially smoothing with a Gaussian kernel (full width at half-maximum of 4 mm), regressing out covariates (white matter, cerebral spinal fluid, global signals, and head-motion profiles are removed to avoid noise signals by multiple regression analysis), temporally linear detrending, temporal band-pass filtering (0.01–0.1 Hz), and scrubbing to reduce the effects of head motion on rs-fMRI data. The networks are constructed based on a voxel or region of interest approach. The Automated Anatomical Labeling (AAL) atlas was used to parcellate the brain into 90 regions (cerebellum excluded). Functional brain networks were constructed by thresholding the correlation matrices with a density of 5% [31]. All network analyses were performed using GRETNA [30]. The values were mapped onto the cortical surface using BrainNet Viewer [32]. Data preprocessing and network analyses are shown in Fig. 1.

There are various methods to identify functional hubs. Some research suggests that hub regions can be defined as degree, betweenness centrality, and/or clustering coefficient values exceeding 1 SD (Standard Deviation) above the mean network, thus indicating hub status [33]. Other research indicates that nodes with a high degree, exceeding 1.5 SD above the mean network, can be identified as functional hubs, meaning that they exhibit high connectivity to the rest of the brain [31]. This study defined functional hubs of research participants with node degree values exceeding 1.5 SD above the mean network.

Descriptive statistics, correlation and comparison analyses were conducted using SPSS for Windows (version 21; IBM SPSS Statistics, Armonk, NY, U.S.). Descriptive statistics are presented as mean, standard deviation (SD), and range. According to ICCTF recommendations, cancer patients were rated as experiencing cognitive dysfunction “if two or more neurocognitive tests (AVLT, TMT, COWA and Digit Span test) had a Z-score at or below -1.5, and/or one test had a Z-score at or below -2.0 of the healthy control group” [25]. Transformation of Z-scores was computed as subjects’ raw score minus the mean group score and divided by the standard deviation. Correlations of neurocognitive function with brain functional connectivity were made using Pearson correlation coefficients. Group differences were tested with independent or paired t-tests. All statistical tests performed were two-sided, and a P value of less than 0.05 was considered statistically significant.

To evaluate the robustness or reproducibility of the result, this study modified two key parameters of global signal regression and different connectivity density thresholds. For the validation of correlation between functional global metrics and associations with neurocognitive outcomes, this study used partial correlation analysis to adjust global signal as the covariate, as global signal of the whole brain is an important confounding factor for brain network analyses based on rs-fMRI [34]. Taken consideration of the connectivity density as an important parameter of the network topological structure, this study also validated study findings with network densities of 3 and 7% suggested by previous research [31]. A network null model for each group was designed with a density of 5% for comparison, as the weights of connections survived after thresholding with a density of 5% were applied.

Of 37 eligible patients, a total of 20 patients agreed to join this study and completed the baseline rs-fMRI and neurocognitive assessment. Four patients refused to attend the MRI scans and neurocognitive assessment post-chemotherapy. There were 20 healthy control subjects who were matched in terms of age, marital and menopausal status. The demographic and clinical characteristics of the research participants are summarized in Table 1.

As illustrated in Table 2, with the exception of information processing speed, there was no significant difference at T1 (pre-chemotherapy) in the neurocognitive test mean scores between patients and healthy controls (Ps > 0.05). There was a significant difference in neurocognitive test scores, (including Digit Span tests, AVLT immediate recall and delayed recall, and TMT-A) (all Ps < 0.05) at T2 (post-chemotherapy). Transformation of patient Z-scores was computed as patients’ raw score minus the mean of the control group score at T1 and divided by SD. Z-scores of cognitive tests at T1 and T2 adjusted for education and employment status. From Table 3, there was a significant difference in Z scores of neurocognitive tests (including Digit Span tests, AVLT and TMT-A) (all Ps < 0.05) between patients and healthy controls at T1 and T2. There were seven patients at T1 and nine patients at T2 who reported cognitive dysfunction, respectively (Table 4).

All participants in the patient and healthy control groups demonstrated a small-world organization as indicated by small-worldness greater than 1. There were significant differences in small-worldness at T1 and T2 between patients and healthy controls (P = 0.04, and P = 0.02, respectively) (Table 5). There was a significant increase in characteristic path length at T2 between patients and healthy controls (P = 0.01). Results from the longitudinal graph analysis revealed a reducing trend of local and global efficiency in the patient group (Table 5). Lower raw TMT-A scores were significantly associated with lower local efficiency (r = 0.28, P = 0.04), and lower verbal memory scores were statistically significant and associated with lower global efficiency (r = 0.41, P = 0.03) in the patient group, but not in the healthy control group.

Brain regions of research participants were evaluated for network hub status based on nodal degree values exceeding 1.5 SD above the mean network [31]. Hub characteristics of brain regions are shown in Figs. 2 and 3. As seen in Fig. 2, functional hub brain regions for cancer patients are mainly located in temporal regions, while parietal regions are the functional hubs in healthy controls. Within the patient group, left hippocampus, left parahippocampal gyrus, left and right insula; middle temporal gyrus, and superior temporal gyrus are functional hubs for patients with neurocognitive dysfunction (Fig. 3).