Changes in Default Mode Network Connectivity in Resting-State fMRI in People with Mild Dementia Receiving Cognitive Stimulation Therapy

Group cognitive stimulation therapy (CST) is a 7-week activity-based non-pharmacological intervention for people with mild to moderate dementia. Despite consistent evidence of clinical efficacy, the cognitive and brain mechanisms of CST remain unclear. Theoretically, group CST as a person-centred approach may work through promoting social interaction and personhood, executive function, and language use, especially in people with higher brain/cognitive reserve. To explore these putative mechanisms, structural MRI and resting-state functional MRI data were collected from 16 people with mild dementia before and after receiving CST, and in 13 dementia controls who received treatment as usual (TAU). Voxel-based morphometry (VBM) and resting-state functional connectivity (rs-FC) analyses were performed. Compared with TAU, the CST group maintained the total brain volume/total intracranial volume (TBV/TICV) ratio. Increased rs-FC in the default mode network (DMN) in the posterior cingulate cortex and bilateral parietal cortices nodes was observed in the CST over TAU groups between pre- and post-intervention timepoints. We provided preliminary evidence that CST maintains/enhances brain reserve both structurally and functionally. Considering the role of DMN in episodic memory retrieval and mental self-representation, preservation of personhood may be an important mechanism of CST for further investigation.


Introduction
Dementia, including Alzheimer's disease, vascular dementia, frontotemporal dementia, Lewy body disease, mixed dementia, and other subtypes, affects 50 million people worldwide [1]. Before a cure for dementia can be identified, interventions to maximise functioning and quality of life are necessary [2]. Apart from symptomatic pharmacological treatments, non-pharmacological interventions for dementia are being actively researched to contain this rapidly increasing global disease burden due to population ageing.
Group-based cognitive stimulation therapy (CST) is currently the only manualised non-pharmacological intervention recommended in clinical guidelines for people with mild to moderate dementia to promote cognition and quality of life [3], in view of its effectiveness in delaying cognitive decline and improving quality of life with evidence from meta-analyses, with an effect size for cognitive outcome comparable to anti-dementia medication treatment [4,5]. The standard protocol of group CST involve 14 sessions This study aims to explore changes in the DMN, CEN, LAN during resting state associated as putative brain networks implicated in CST, in the context of individual differences in baseline and change in cognitive reserve among people with mild dementia.

Materials and Methods
Data were drawn from an MRI substudy of a prospective non-randomised controlled trial comparing participants receiving CST versus treatment as usual (TAU) in Hong Kong Chinese people with mild dementia for cognitive and quality of life outcomes. The study was approved by the Human Research Ethics Committee of The University of Hong Kong (HREC Reference Number: EA1501019). Written informed consent was obtained from both the participants and their carers prior to their participation in the study.

Participants
Participants of the MRI substudy were recruited from a CST trial [4] where they received referrals from community dementia services in Hong Kong. Following inclusion criteria in a key CST trial [4], the inclusion criteria for this substudy were: (1) a clinical diagnosis of dementia; (2) a Cantonese Mini-Mental State Examination (MMSE) score of over 18 [29]; (3) absence of other psychiatric disorders; (4) able to communicate and understand communication; and (5) able to see and hear well enough to participate in a meaningful assessment. The following criteria were added for this MRI study: (6) right-handedness assessed using the Edinburgh Handedness Inventory [30]; (7) no contraindications to MRI, including the presence of cardiac pacemakers, incompatible metal implants, and claustrophobia; and (8) without a history of stroke, heart attack, or head trauma leading to loss of consciousness. A total of 20 participants from the CST group and 17 participants in the control group of the trial were enrolled into this MRI study. Among the 20 participants enrolled in the CST group, follow-up MRI data were available for 16 participants (80% retention rate), with four lost to follow-up or withdrawn. Among the 17 recruited as controls, 15 were eligible for and completed MRI at baseline, and 13 (87% retention rate) completed the follow-up assessments.

Assessments and Procedures
Participants referred from community dementia services were invited for a screening assessment to determine eligibility. After screening and baseline (T0) assessment, participants allocated to the CST group then received a standard group CST protocol with 14 sessions delivered twice a week, of approximately 45 min duration and [6] tailored for Hong Kong Chinese [31]. Participants in the control group received TAU. The same set of assessments were repeated at follow-up (T1), which is upon completion of all CST sessions for the intervention group.
Behavioural assessments included cognitive performance, mood, communication, as well as the participants' demographics and proxies for cognitive reserve (years of education and work). Cognitive performance was measured using the ADAS-Cog [32], which covers multiple cognitive domains and has three subscales: (1) memory and learning; (2) language; and (3) praxis. A higher score on ADAS-Cog (score range, 0-70) denotes worse performance. We defined a change in cognition by subtracting T1 scores from T0 score. The Cornell Scale of Depression in Dementia (CSDD) [33] was used to assess participants' mood based on informant and patient interviews. A higher score suggests worse depression (score range, 0-38). We used the Holden Communication Scale (HCS) [34] to assess participants' communication, and a higher score indicates more difficulties (score range, 0-48). MRI scans were arranged on average 31 days (SD = 25) after the behavioural assessments at T0, and on average within 42 days (SD = 37) at T1. The average interval between T0 and T1 scan was 145 days (SD = 72).

Magnetic Resonance Imaging (MRI) Data Acquisition and Preprocessing
Structural and functional brain imaging data were acquired using a 3.0-Tesla Philips Achieva whole-body MRI scanner (Philips Healthcare, Best, The Netherlands). Following the echo-planar imaging (EPI) sequence, high-resolution T1-weighted magnetisationprepared rapid gradient-echo (MP-RAGE) imaging images were acquired for structural morphometric analysis and for preprocessing of fMRI data. Using a 2D gradient EPI sequence, we acquired 180 T2-weighted functional image volumes, using the following parameters: repetition time (TR) = 2000 ms, echo time (TE) = 30 s, flip angle = 90 • , number of slices (Nslices) = 32, matrix = 64 × 64, slice thickness = 4 mm, field of view (FoV) = 240 × 240 mm 2 , ascending interleaved slice ordering. Head motion was restricted using firm padding that surrounded the participants' head throughout the whole scanning process. For resting-state functional imaging, participants were asked to rest inside the scanner with their eyes open, and fixate on a cross in the middle of a projected image inside the scanner, and avoid engaging in any specific thoughts.
T1-weighted MP-RAGE images were preprocessed following the CAT12 protocol (Structural Brain Mapping group, Jena University Hospital, Jena, Germany) implemented as a toolbox in Statistical Parametric Mapping (SPM) 12 package (Institute of Neurology, London, UK). Images were corrected for bias-field inhomogeneities, then spatially normalised and segmented into GM, WM, and cerebrospinal fluid (CSF) within the same generative model [35]. A quality check was performed after the preprocessing, and according to the automated quality insurance protocol of CAT12, all scans included in this study passed the check. All GM images were then smoothed with an 8 mm, full width-athalf-maximum, isotropic Gaussian kernel to minimise between subject cortical variation of the gyrus.
Functional images were preprocessed using the SPM 12 package. None of the participants were excluded based on the criteria of a displacement > 3 mm in x, y, or z or an angular rotation > 1.5 • in any direction. Imaging data were then slice-time corrected and realigned. Functional volumes were co-registered and resliced to a voxel size of 2 mm 3 , normalized to the Montreal Neurological Institute (MNI) template brain, and smoothed with an 8-mm, full width-at-half-maximum, isotropic Gaussian kernel. The denoising process was carried out with the Functional Connectivity toolbox (CONN, V16.a) [36] implemented in the MATLAB environment and included three steps: (1) Linear regression of potential confounding effects in the BOLD signal. To identify the factors, the anatomical component-based noise correction procedure (aCompCor) was used to identify noise components associated with segmented WM and CSF [37], estimated subject-motion parameters [38], and identified scrubbing parameters [39] without affecting intrinsic functional connectivity [40]. (2) Application of a 0.008-0.09 Hz bandpass filter to reduce the effects of low-frequency drift and high-frequency noise [41]. (3) Linear detrending.

Data Analyses
Behavioural data were analysed using SPSS (version 23.0). Independent sample t or χ 2 tests were used to explore differences in characteristics between CST and control groups. Repeated measure ANOVA was used to analyse the areas of significance change from T0 to T1 between the two groups in ADAS-Cog, CSDD, and HCS.
Structural images were analysed using the VBM technique using CAT12, which provides comprehensive information of brain morphometric features while avoiding biases due to structural differences [42]. Total intracranial volume (TICV) was calculated as the sum of the GM, WM, and CSF volumes; and the total brain volume (TBV) as the sum of GM and WM, and the TBV/TICV ratio was then calculated to allow for more direct comparisons between participants. To detect the TBV/TICV ratio changes in participants, paired sample t-tests were conducted on the processed images. Repeated-measures ANCOVA and mixeddesign ANCOVA with age and gender as nuisance covariates were used to explore withinsubject changes between baseline and follow-up conditions.
Resting-state functional connectivity (rs-FC) was performed by applying a seed-based approach using the CONN toolbox [36]. Previously defined regions of interest (ROIs) based on literature were used as seeds for seed-based rs-FC analyses [43]. All seed ROIs were 10 mm diameter spheres specified in MNI coordinates. The ROIs for DMN were the MPFC ( [20]. The ROIs for Language Network were AI (39,19,14), sTG (−55, −18, 8), iFG (46,36,18), and mTG (62, −37, −3) [24]. See Supplementary Figure S1 for a brain surface map with the different ROIs corresponding to the three networks of interest. Seed-to-voxel analysis was used, in the first-level (individual) analysis, correlations were generated in the CONN toolbox by extracting the mean resting-state BOLD time course from each seed ROI and calculating correlation coefficients with the BOLD timecourse of each voxel throughout the whole brain, and the resulting coefficients were Z transformed [15]. The rs-FC of each network was calculated by averaging the rs-FC of selected seeds within the network, paired sample t-tests were conducted to analyse individual level changes. In the second-level between group analysis, the differences of rs-FC of DMN, CEN, and LAN between CST and TAU groups pre-and post-intervention were measured by applying mixed ANCOVA with age, gender, and education as covariates. Pearson's correlation coefficient was performed to explore bivariate correlations between demographics, cognitive/brain reserve proxies and ADAS-Cog change. Table 1 shows the baseline and follow-up characteristics of the sample with complete baseline and follow-up data. Both groups were characterised by a female dominance with an average age of around 80 years. The CST group had significantly fewer people with any formal education, although both groups had a low level of education, which is common among the current cohort of older persons in Hong Kong due to historical reasons. The TAU group had worse communication as shown in the HCS score. See Supplementary  Table S1 for a correlation matrix of the key variables. Table 1. Baseline and follow-up demographic characteristics, cognitive reserve proxies, and cognition.

Within-Subject Changes
Paired-sample t-test and mixed-design ANCOVA were used to explore within-subject changes. Table 2 summarises the changes and results from paired sample t-tests in ADAS-Cog, CSDD, HCS, brain volumes and rs-FC of DMN, CEN, and LAN.  Two-sided contrast maps without Bonferroni corrected are shown in Figure 1. Increased correlation in the CST over TAU in the follow-up over baseline was shown in the lLP, rLP, and PCC (see Figure 1).  Table 1 shows the baseline and follow-up characteristics of the sample with complete baseline and follow-up data. Both groups were characterised by a female dominance with an average age of around 80 years. The CST group had significantly fewer people with any formal education, although both groups had a low level of education, which is common among the current cohort of older persons in Hong Kong due to historical reasons. The TAU group had worse communication as shown in the HCS score. See Supplementary Table 1 for a correlation matrix of the key variables.

Predictors of Cognitive Changes
We investigated the bivariate correlation between age, cognitive reserve proxies (education, work), baseline brain reserve proxy (TBV/TICV, rs-FC), and ADAS-Cogs (baseline, follow-up, and change). For the CST group, while age was related to ADAS-Cog score at baseline and follow-up (r = 0.56 and 0.47, respectively, p < 0.01), it was unrelated to the change in ADAS-Cog score. Years of education in this low-education sample was not related to ADAS-Cog score at baseline or follow-up, or its change. Gender was not related to any ADAS-Cog score either, but female gender was positively correlated with higher baseline TBV/TICV ratio (r = 0.56, p < 0.05). We observed a significant positive correlation between years of work and improvement in ADAS-Cog (r = 0.45, p < 0.05), which was not related to baseline or follow-up ADAS-Cog score. Baseline TBV/TICV was not related to baseline ADAS-Cog, but was correlated with follow-up ADAS-Cog (r = −0.50, p < 0.05), and also positively correlated with improvements in ADAS-Cog score in the CST group (r = 0.57, p < 0.01). Partial correlation controlling for age and gender revealed a strong correlation between years of work and improvement in ADAS-Cog (r = 0.75, p < 0.01; Table 3), and the partial correlation between baseline TBV/TICV and improvement in cognition was non-significant (p = 0.14). In the TAU group, a significant correlation was observed between baseline TBV/TICV and cognitive improvement (r = 0.63, p < 0.05), years of work was not correlated with cognitive change. Partial correlation controlling for age and gender still revealed significant correlation between baseline TBV/TICV and improvement in ADAS-Cog (r = 0.66, p < 0.05). Figure 2a illustrates the relationship between z-transformed years of work and ADAS-Cog change score after controlling for age and gender in both groups; and Figure 2b illustrates the relationship between z-transformed baseline TBV/TICV with standardised ADAS-Cog change score. Table 3. Partial correlation between baseline network connectivity and cognitive changes controlling for age and gender, and Fisher's Z transformation.    Baseline rs-FC in DMN, CEN, or LAN was not related to ADAS-Cog (baseline, followup, and change) in all participants in general, except that baseline CEN connectivity was positively correlated with improvements in ADAS-Cog in control group (r = 0.68, p < 0.05; Supplementary Materials Table S1). Partial correlations controlling for age and gender were performed between baseline network connectivity and cognitive improvement in both groups, and no significant results were found (Table 3). To further compare differences in correlations between two groups, Fisher's Z transformation was performed [44], and comparison of transformed Z revealed marginal difference in the correlation between years of work and improvement in cognition (z = 1.83, p = 0.06).

Discussion
This is the first study exploring the putative impact of CST on functional connectivity of three key brain networks in people with mild dementia, as compared with controls receiving TAU. The results suggested that CST may increase resting-state DMN connectivity, particularly in the PCC, LLP, and RLP nodes, which happened despite a reduction in grey matter volume regardless of the intervention or TAU. Baseline TBV/TICV ratio positively predicted cognitive change in both groups, and years of work is predictive of cognitive improvement only in the CST group. These initial findings provided several directions for future research.
The potential implication of CST on DMN, but not CEN or LAN, opens an avenue for further research into the mechanisms of action in evidence-based non-pharmacological interventions for dementia. To date, there have been very few studies examining the effects of non-pharmacological intervention for dementia on structural or functional brain changes of the clients. A study among healthy older adults found that a multi-domain combined training reorganized the functional connectivity of DMN by increased rs-FC in PCC regions [45]. A recent study used magnetoencephalography (MEG) to examine changes in brain activity of people with dementia (PWD) before and after receiving nonpharmacological treatment, results revealed a reduced alpha activity in the right temporal lobe and fusiform gyrus, as well as an increased low-gamma activity in the right angular gyrus [46]. Both studies confirm that with dementia or not, aging brains have the capacity for plasticity. The estimated number needed to treat (NNT) with the standard protocol for CST to deliver clinically important cognitive improvement ranged from 6 to 14, suggesting modest efficacy similar to pharmacological intervention [6,47]. Due to the complex nature of the intervention, to further improve the clinical efficacy and reduce the NNT, an understanding of the 'active ingredient' in CST as opposed to other services/interventions is needed. The finding of increased resting-state DMN connectivity after CST therefore provided direction for further investigation and refinement of intervention design, with reference to our current understanding of the function of DMN.
In a recent review consolidating previous hypotheses and evidence, DMN is identified as a network responsible for 'sense-making' in a social world, by integrating incoming external information with a person's idiosyncratic memory and knowledge [48]. Research has increasingly shown that, in contrast to the previous task-unrelated, 'mind wandering' hypothesis, DMN activities may support ongoing cognition for detailed experiences in a task [49], automated adaptive responses to environmental demands that govern daily lives [50], forming context-dependent models of the situation for establishing shared meaning. This updated understanding of the function of DMN is in line with the emphasis on person-centredness in CST [6] and biopsychosocial theoretical underpinning [7]. Personcentred care emphasises ways to enable a person in making the fullest use of one's abilities and to remain a social being [51], which is realised in CST principles such as 'maximising potential, 'building/strengthening relationships', 'respect', 'involvement', and 'inclusion'. These principles distinguish CST from other interventions that place more emphasis on training and rehabilitation of individual cognitive domains. Future study investigating DMN connectivity change in non-pharmacological interventions with varying level of person-centredness would provide further insight.
The increased connectivity in the medial and bilateral parietal cortices suggested a role of the representation of mental self. Earlier research showed that the medial parietal region may be a nodal structure in self-representation [52] and the lateral parietal cortex may support episodic memory [53]. The role of self in memory has recently been revisited in dementia [54], which points toward the impairment in a sense of self-continuity across temporal contexts linking to a reduced capacity for expressing episodic memory and future thinking. In CST, the principles of 'using reminiscence as an aid to the here-and-now' and 'continuity and consistency between sessions' may facilitate or reinstate this sense of selfcontinuity. In a recent study investigating the role of self-referential thinking and memory in people with amnestic mild cognitive impairment [55], both autobiographic recall and narrative conditions were shown to benefit memory, and the authors concluded that the association of information to the 'self' in people with cognitive impairment provides a useful schema, which relies on the autonoetic experience of inward reflection and subjective experience of relating to one's knowledge and experience. Research including behavioural measurements of the subjective experience of self-representation would allow investigation of this putative active ingredient in CST.
The increased DMN connectivity in the CST group happened against a context of a significant reduction in grey matter volume (observed in both groups), which suggests that CST does not alter disease progression in terms of structural neuropathology. In the large-scale prevention research Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability (FINGER) involving 2 years of intensive multimodal intervention for older persons, similar negative findings have been reported on MRI changes in brain volumes, cortical thickness, and white matter lesion volume, despite improved cognition with the intervention [56]. Although it is less likely that cognitively stimulating activities act through structural (versus functional) brain changes, our initial finding of reduced TBV/TICV ratio only in the control group may also suggest it as a potential outcome measure on top of other absolute structural brain changes.
The FINGER study, nevertheless, identified a higher baseline cortical thickness in Alzheimer's disease signature areas, and possibly a higher hippocampal volume, as predictors for cognitive benefits from the multimodal intervention [56]. On the other hand, the earlier report in healthy older persons using rsfNIRS [26] found that, among people with high mid-life occupation achievement (a proxy for cognitive reserve), those with high small-worldness (associated with higher IQs [15], also a cognitive reserve proxy) are less likely to benefit from intellectual engagement activities. These earlier findings suggested the relevance of baseline brain structural and functional characteristics as predictors of intervention response in healthy older persons. Our results provided support to the use of baseline work years as predictors of response to CST in people with mild dementia, supporting the argument that people with a higher cognitive reserve (i.e., those with excess disability given the same disease severity level) are more likely to benefit, although resting-state functional connectivity of the three brain networks of interest in this study did not predict cognitive gain from CST. On the other hand, baseline TBV/TICV predicted improved cognition in both groups, while baseline CEN connectivity is associated with cognitive improvement in the control group only. The mechanisms by which these brain reserves interact with exposure to everyday activities in people with dementia regardless of non-pharmacological intervention may be of interest for developing daily life strategies to enhance cognition.
This study has several limitations. First, the small sample size limited the possibility in analysis, including a limited statistical power to detect a significant difference in cognitive change between the two groups. Although the study was not designed to detect the effectiveness of CST given the available evidence, this limitation posed difficulties in group comparison and results interpretation, with a lack of support of a behavioural outcome difference between groups. The finding of a difference in DMN connectivity change only in the CST group nevertheless provided a basis for future resting-state fMRI study with sufficient power to detect cognitive outcome changes. Second, due to the exploratory nature of the study, a quasi-experimental design was used, with potential bias in group characteristics. At baseline, the control group had worse communication, while fewer people in the CST group had any formal education. These differences might have contributed to the outcomes and our findings should be interpreted with this consideration. Finally, due to logistics difficulties, the time intervals between behavioural assessment and MRI scan at both the baseline and follow-up were long (CST group: 182 days, SD = 73; control group: 94 days, SD = 26), which might have introduced noise and uncertainties in the data. Notwithstanding these limitations, this study provided evidence for the first time of a brain functional change in people with mild dementia after exposure to a non-pharmacological intervention recommended by clinical guidelines, and suggested directions for further research that may improve its efficacy, as well as a proposed mechanism of action that would be relevant to a wide range of person-centred interventions for dementia.

Conclusions
This is the first study to explore the neuropsychological mechanisms of CST using resting-state functional connectivity MRI. We provided preliminary evidence that CST maintains/enhances brain reserve both structurally and functionally. The role of DMN in episodic memory retrieval and mental self representation suggested further investigation into the preservation of personhood as an important mechanism of CST and other personcentred intervention strategies.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/brainsci11091137/s1, Table S1: Correlations between age, cognitive reserve proxies, brain reserve proxy, and ADAS-Cog scores in CST and TAU groups. Figure S1: Regions of interest used in the functional connectivity analysis, superimposed on a three-dimensional brain template.