Localised Grey Matter Atrophy in Multiple Sclerosis and Clinically Isolated Syndrome—A Coordinate-Based Meta-Analysis, Meta-Analysis of Networks, and Meta-Regression of Voxel-Based Morphometry Studies

Background: Atrophy of grey matter (GM) is observed in the earliest stages of multiple sclerosis (MS) and is associated with cognitive decline and physical disability. Localised GM atrophy in MS can be explored and better understood using magnetic resonance imaging and voxel-based morphometry (VBM). However, results are difficult to interpret due to methodological differences between studies. Methods: Coordinate-based analysis is a way to find the reliably observable results across multiple independent VBM studies. This work uses coordinate-based meta-analysis, meta-analysis of networks, and meta-regression to summarise the evidence from voxel-based morphometry of regional GM hanges in patients with MS and clinically isolated syndrome (CIS), and whether these measured changes are relatable to clinical features. Results: Thirty-four published articles reporting forty-four independent experiments using VBM for the assessment of GM atrophy between MS or CIS patients and healthy controls were identified. Analysis identified eight clusters of consistent cross-study reporting of localised GM atrophy involving both cortical and subcortical regions. Meta-network analysis identified a network-like pattern indicating that GM loss occurs with some symmetry between hemispheres. Meta-regression analysis indicates a relationship between disease duration or age and the magnitude of reported statistical effect in some deep GM structures. Conclusions: These results suggest consistency in MRI-detectible regional GM loss across multiple MS studies, and the estimated effect sizes and symmetries can help design prospective studies to test specific hypotheses.


Introduction
Areas of inflammation, axonal loss, demyelination and gliosis, occurring throughout the brain and spinal cord, are the distinctive features of multiple sclerosis (MS) [1]. Although MS has been considered a condition affecting the white matter (WM) and the hyperintense lesions on T2-weighted "clinically isolated syndrome" OR CIS) and TITLE-ABSTR-KEY("voxel based morphometry" OR VBM) and TITLE-ABSTR-KEY("grey matter" OR "gray matter" OR GM) and TITLE-ABSTR-KEY(atrophy).

Study Selection
Inclusion criteria are (a) involved participants with MS or CIS, (b) compared patients to healthy controls, (c) performed whole-brain VBM for assessing GM atrophy, and (d) reported coordinates for GM volume changes in either Talairach [22] or Montreal Neurological Institute (MNI) reference space. Exclusions were made due to unreported coordinates or unavailable full text. Two independent researchers assessed these criteria of the individual studies and the MNI or Talairach coordinates.

Study Properties
Information extracted for analysis: the censoring threshold, i.e., the smallest Z value the study considered as significant, the reported coordinates, and either the Z score, estimated degrees of freedom and t statistic, or uncorrected p-value; t statistics and uncorrected p-values are converted automatically to Z scores.
Details about the algorithms incorporated into CBRES and CBMAN are presented in [19,20]. In both algorithms a clustering algorithm [23] is used to determine where the coordinates reported by multiple independent studies are spatially concordant (clustered). Once clusters are formed, the reported Z scores are converted to standardised effect sizes by dividing by the square root of the number of subjects. In CBRES, a random effect meta-analysis of these effect sizes is performed in each cluster. In CBMAN, the test statistic is the correlation of standardised effect sizes performed pairwise between clusters. Where a study does not report a coordinate within a cluster, or where no effect sizes are reported by a study, the contribution to the cluster is estimated using the study censoring threshold.
The significant results of the CBRES meta-analysis are clusters of reported coordinates where the estimated effect size is statistically different to zero after controlling the false cluster discovery rate (FCDR), a type 1 error control method based on the false discovery rate (FDR) [24]. The expected proportion of clusters incorrectly declared significant is controlled at 5% by default. Clusters indicate both spatial and effect size concordance across studies, which is an unlikely chance event suggesting that atrophy at the location of the clusters is a general feature of MS.
Significant results reported by CBMAN are clusters where standardised reported statistical effects are correlated between clusters. This indicates a significant pattern of reported effect that is represented as a network of nodes (clusters) and edges (correlations). The FDR is used to control type 1 error rate of the effect size correlations. The clusters analysed by CBMAN and CBRES are identical, since the same clustering algorithm is employed, but the results may differ due to the different hypotheses tested. A feature of both CBRES and CBMAN is that the results declared significant are reported as a function of the FDR. Any that just miss the threshold for significance can therefore be explored.
Analysis can also be performed on subgroups of studies. This estimates a subgroup-specific effect size in each of the clusters found significant during the full analysis (using all studies); this is useful since clusters may not be significant if the subgroup is small, yet the effect size might be of interest. Furthermore, the use of standardised effect sizes makes meta-regression possible by looking for significant correlation between a specified covariate and the standardised effect size in each cluster.

Experimental Procedure
Multiple experiments reported on the same subjects were pooled into single independent experiments to prevent correlated results inducing apparent concordance that is not due to a generalisable MS process [25].
All planned analyses were performed controlling the FDR at 0.05. For each, the next most significant clusters were explored, and reported, to make sure that none had just been missed at this threshold.

Main Analysis
The main meta-analysis was performed using both CBRES and CBMAN and involved all studies meeting the inclusion criteria.

Subanalyses
Subanalyses for CIS, benign MS (BMS), Relapsing Remitting MS (RRMS), Primary Progressive MS (PPMS) and Secondary Progressive MS (SPMS) studies were performed, with subtypes as defined in the reporting studies. This analysis estimates effects of the respective subgroup within significant clusters discovered using all studies.

Meta-Regression
Regression analyses were performed for covariates that might influence the grey matter volume: mean age (years), MS disease duration (years; excluding CIS studies with no MS disease duration), MSFC, and EDSS (all studies and including RRMS studies only).

Included Studies and Sample Characteristics
The literature search yielded 237 potential studies, of which 34 met the inclusion criteria ( Figure 1). The 34 included research papers reported 45 whole-brain VBM experiments comparing MS subtypes and controls (see Appendix A for study details). cluster.

Experimental Procedure
Multiple experiments reported on the same subjects were pooled into single independent experiments to prevent correlated results inducing apparent concordance that is not due to a generalisable MS process [25].
All planned analyses were performed controlling the FDR at 0.05. For each, the next most significant clusters were explored, and reported, to make sure that none had just been missed at this threshold.

Main Analysis
The main meta-analysis was performed using both CBRES and CBMAN and involved all studies meeting the inclusion criteria.

Subanalyses
Subanalyses for CIS, benign MS (BMS), Relapsing Remitting MS (RRMS), Primary Progressive MS (PPMS) and Secondary Progressive MS (SPMS) studies were performed, with subtypes as defined in the reporting studies. This analysis estimates effects of the respective subgroup within significant clusters discovered using all studies.

Meta-Regression
Regression analyses were performed for covariates that might influence the grey matter volume: mean age (years), MS disease duration (years; excluding CIS studies with no MS disease duration), MSFC, and EDSS (all studies and including RRMS studies only).

Included Studies and Sample Characteristics
The literature search yielded 237 potential studies, of which 34 met the inclusion criteria ( Figure  1). The 34 included research papers reported 45 whole-brain VBM experiments comparing MS subtypes and controls (see Appendix A for study details).

Primary Meta Analysis
The analysis found eight significant clusters involving basal ganglia and cortical regions; effect sizes are given in Table 1 and the complete list of Talairach regions, automatically detected [26], covered by each cluster given in online materials [dataset] (DOI: 10.17639/nott.7049) [21]. Significant clusters and a depiction of the covariance of standardised effect sizes between clusters are shown in Figure 2. Forest plots for the most significant clusters according to CBRES are shown in Figure 3. In Figure 4, a scatter plot of standardised effect sizes reported in the left and right thalamic clusters shows clear correlation detected by CBMAN. The first non-significant cluster according to CBRES was at a FCDR of 0.18. The first non-significant edges discovered by CBMAN were at a FDR of 0.057, where a further nine significant edges and two extra clusters (right Caudate peaking at Talairach

Subanalyses
Results from the subanalysis are given in Table 1, which shows estimated effects sizes considering only the respective subgroup within each of the clusters from the primary meta-analysis. It is apparent Brain Sci. 2020, 10, 798 8 of 16 that the estimated effects are lowest in magnitude in the CIS and PPMS groups and highest in the SPMS and BMS groups, while RRMS generally falls in the middle.

Age
A single cluster was found with age as a significant covariate, as shown in Table 2. The first non-significant cluster was at a FCDR of 0.18.

MS Disease Duration
This analysis included 39 of the independent experiments reported MS disease duration due to exclusion of CIS studies where MS disease duration is zero. Three clusters were found with disease duration as a significant covariate, as shown in Table 2. The first non-significant cluster was found at a FCDR of 0.1.

MSFC
Regression analysis was not performed because only 9 out of 44 studies reported MSFC.

EDSS
No significant clusters with EDSS as a covariate. The first non-significant cluster was found at a FCDR of 0.1.

Discussion
The results of 34 voxel-based morphometry studies of MS and CIS are summarised using CBMA, showing that GM atrophy in MS not only occurs in some regions more than in others and that regions of predilection are not independent. Eight regions were identified using two algorithms testing different null hypotheses. Results indicate a consistent pattern, rather than independent clusters, of reported effects both spatially and in terms of effect size.
The pattern of localised GM atrophy involves both cortical and subcortical regions. Considering the correlation of reported statistical effect size suggests that these regions do not develop independently, but rather together and with some hemispheric symmetry as shown in Figure 2. Meta-regression analysis suggests that the standardised effect magnitude increases with disease duration by several % per year of disease on average. This is also reflected in the mean statistical effect size estimates within the disease type subgroups, where the estimates for the CIS group tend to be lower than the RRMS group, which are in turn lower than the SPMS group. The PPMS subgroup reported intriguingly low statistical effect sizes, while the BMS group almost as large as the SPMS group in some clusters, which might reflect that BMS is indistinct from MS with long enough follow up. These estimates should be considered with caution because of the small subgroup sizes, however they could be prospectively tested.
Brain Sci. 2020, 10, 798 9 of 16 GM tissue damage is an important pathological process in MS that underlies neurological disability [27]. It has been suggested that distribution of cortical GM atrophy is related to the effects of WM lesions on cortical regions that are network hubs, with trans-synaptic degeneration then extending from these hubs [28], or that the preferential accumulation of WM lesions in some regions would induce tract-mediated effects through secondary retro-or anterograde degeneration. The relationship between GM atrophy and WM abnormalities is weaker in people with PPMS or SPMS [29]. The loss of volume is the result of many dynamic processes, with a balance between destructive and reparative mechanisms with interaction among neurons, oligodendrocytes, axons, microglia, astrocytes, inflammatory cells, endothelial cells and water distribution [30].
In the thalamic clusters, the standardised effect sizes were found to correlate negatively with disease duration. Both imaging and pathology studies have demonstrated the involvement of thalamus in early RRMS [31], CIS [32] and paediatric MS [33]. Cifelli and colleagues [34] conducted a study of normalised thalamic volume measurements in SPMS patients. Volumes of manually outlined thalami were normalised by intracranial volumes and showed a mean decrease of 17%. Thalamic volume loss may be due in part to disconnection created by WM lesions [35,36].
Atrophy of the left and right putamen MS has also been detected. The putamen is part of the dorsal striatum and the basal ganglia and plays a role in the regulation of movement, coordination, motor function and cognition [37][38][39]. It is also involved in modulation of sensory and motor aspects of pain [40]. Thus, a pathology such as neurodegeneration might be expected to cause a broad spectrum of clinical manifestation from motor dysfunction to psychiatric disorder [41,42]. Previous studies have demonstrated progressive atrophy of the putamen in both RRMS and SPMS [43]. Kramer and colleagues [44] recently reported a significant relationship between putamen volume and disease duration in MS, which was also indicated by the present study.
Pre-and postcentral gyrus (bilateral) is consistently reported. The clusters have density peaks reported in the right precentral and the left postcentral gyrus, but the coordinates forming the clusters cover both pre-and postcentral gyrus on each side. Li et al. [45] used diffusion tensor imaging (DTI) and demonstrated neuroconnectivity changes in the left postcentral gyrus, and reduced communicability correlating with the 25 foot walk test results.
Clusters covering the left/right superior temporal gyrus and insula ware also detected. The superior temporal gyrus is associated with auditory and speech comprehension [46,47] and perception of emotions in facial stimuli [48,49]. In addition, it is an essential structure in the pathway containing prefrontal cortex and amygdala that are responsible for social cognition processes [48,50]. The study conducted by Achiron and colleagues [51] suggested correlation between reduced cortical thickness in superior temporal gyrus and global cognitive score, attention, information processing speed and motor skills. The insula is primarily a visceral-somatic region [52]. Studies have shown a relationship between functional connections of the basal ganglia and insula and fatigue severity in case of MS patients [53][54][55].
A similar coordinate-based meta-analysis in MS was performed by Chiang et al. at the same time as the present study [56]. That study used the popular ALE algorithm and produced similar results. The study also used functional meta-analytic connectivity modelling (fMACM) [57] to explore functional coactivation of clusters as a network, estimated using non-MS studies. By contrast, the present study investigates the network-like properties of GM atrophy and uses the included MS-specific studies.
The limitations of CBMA are that bias and methodological issues in the primary studies might be reflected in the results. Therefore, CBMA results should be considered hypothesis generating and used to inform robust prospective studies. To this end, the presented results provide a priori regions of interest for testing as well as statistical effect size estimation for sample size calculations.

Conclusions
This CBMA of VBM studies of MS and CIS has identified a pattern of related cortical and subcortical GM atrophy. Relationships are indicated by the covariance of reported statistical effects.
Disease duration was found to be a significant covariate of the standardised reported effect sizes in the thalamic clusters and a cluster covering the left claustrum/putamen/insula. The estimated statistical effect sizes may be important for powering prospective studies of GM atrophy in MS to test specific hypotheses.