Participants with MS were enrolled from a larger prospective longitudinal study of psychiatric comorbidity in immune-mediated inflammatory disorders (18). A subgroup of participants with MS were recruited for this sub-study (19) and underwent additional neuroimaging, cognitive, and psychiatric testing. Expanded Disability Status Scores (20), lesion load, and disease duration was recorded for participants in the MS group. Participants were excluded if they had comorbid neurodegenerative disease or brain tumors, or if they had any MRI contraindications. The sub-study also recruited healthy controls who did not have any chronic health conditions, history of head injury, or chronic medication use apart from oral contraceptives or hormonal replacement therapy. All participants provided written consent and were required to have sensorimotor function adequate to perform the cognitive tests, have sufficient English language knowledge to undergo the test protocol, and were aged 18 years or older.

Cognitive testing included the Weschler Test of Adult Reading (WTAR) (21) and the oral form of the Symbol Digit Modalities Test (SDMT) (22). The WTAR assessed premorbid intelligence (23). The SDMT is the most commonly used test of information processing speed in MS (24), and is a component of the Brief International Cognitive Assessment for Multiple Sclerosis (BICAMS) test battery (25). We used local regression-based age-, sex-, and education-adjusted SDMT scores (26) to compare the MS and HC samples. The cognitively preserved (MS-CP) group included individuals with MS whose SDMT z-score was ≥0 standard deviations (SD) of the HC reference group. The MS-CI group included individuals with SDMT z-scores <−1.5SD below the HC mean (27). The nearest age-matched participants from our HCs were selected to form control groups for the MS-CI and MS-CP groups (labeled HC_MS-CI and HC_MS-CP, respectively).

The acquisition parameters and the preprocessing, denoising, and initial data analyses are described in detail in our previously published protocol paper (19). Briefly, images were collected using a 3 T Siemens TIM Trio MRI system. Anatomical T₁-weighted images were acquired using the MPRAGE sequence. T₂-weighted FLAIR scans were acquired for subsequent lesion detection. Resting state fMRI was acquired using the Human Connectome Project “cmrr_mbep2d_bold” sequence with an acquisition time of 7:12 min. Lesion segmentation was completed with the Lesion Segmentation Tool in SPM12, and lesion masking and filling were accomplished in FSL. Lesion-filled MS structural raw data and HC structural raw data were entered into the processing pipeline. Functional data for MS and HC were processed identically. In brief, rs-fMRI data were spatially preprocessed using SPM12 which included removal of the first 10 brain volumes, realignment, distortion correction, coregistration to MNI space, and smoothing with a 4.0 mm full-width half-maximum 3D Gaussian kernel. The images were then imported to the CONN toolbox (version 18b) (28) for subsequent pre-preprocessing. The scan-to-scan change in average BOLD signal within the global-mean mask was scaled to standard deviations and changes of >5 standard deviations were flagged as outliers which were included as a first-level covariate for scrubbing. Using the Artifact Detection Tool (29) (ART) within CONN, time points with default values of volume-to-volume intensity changes of z > 3.0, values of rotational motion of >0.05 degrees in any plane and/or translational motion >0.50 mm in any direction were identified. The estimated subject-motion parameters were regressed from the data in the denoising step of the preprocessing pipeline. To minimize motion-related variability in the BOLD signal, the 3 rotation and 3 translation parameters, plus their associated first-order derivatives, were regressed from the BOLD signal timeseries. The time-course of eroded white matter and cerebrospinal fluid were similarly regressed from the BOLD signal timeseries to correct for physiological noise. A temporal band-pass filter (0.008–0.09 Hz) was used to isolate only low frequency fluctuations (30).

We summarized participant characteristics using descriptive statistics, and compared groups using chi-square tests, Student’s t-tests, and Mann–Whitney U tests as appropriate.

Functional connectivity analysis was performed using brain parcellations included with CONN software (28). The CONN atlas includes 132 ROIs which are an amalgamation of the Harvard-Oxford cortical and subcortical structural atlas (31) and the cerebellar portions of the AAL atlas (32). An ROI-to-ROI analysis was used to calculate the bivariate correlations between each pair of ROIs using the General Linear Model (GLM, correlation (bivariate) settings, no weighting applied). First, we performed the broader between-subjects contrast [MS (1) HC (−1)] controlling for age and sex. Second, we performed the contrast comparing cognitively impaired MS patients with HC, controlling for sex [MS-CI (1) HC_MS-CI (−1)]. Third, we performed the contrast comparing cognitively preserved MS patients and HC, controlling for sex [MS-CP (1) HC_MS-CP (−1)]. Finally, we contrasted the cognitively impaired MS participants with cognitively preserved MS participants [MS-CI (1) MS-CP (−1)]. Significant ROI-to-ROI connections were thresholded by intensity at p < 0.05 corrected for multiple comparisons using the false discovery rate (33).

A linear regression analysis was run for the MS group alone, identifying the association between the FC and the MS group’s SDMT scores (controlling for age). The ROI-to-ROI connection threshold was set to p < 0.05 corrected for multiple comparisons using the false discovery rate (33).

We included 107 participants with MS and 94 controls after excluding 4 participants with MS and 6 HC participants due to insufficient MRI data quality. The MS group was older and had a higher proportion of women relative to the HC group. As expected, the MS group had lower SDMT scores than the HC group, but the two groups did not differ with respect to premorbid IQ based on the WTAR (Table 1).

For the contrast of MS vs. HC, we found 37 pairwise differences (ROI-to-ROI pairings) that distinguished the FC of the MS group from HC (Table 2).

Twenty-six of these pairwise differences involved lower FC in the MS group relative to controls, shown in Figure 1 with blue squares and listed in Table 2 in the section MS < HC. Of those, 5 pairwise differences were between two ROIs associated with visual processing, and 7 pairwise differences involved at least one ROI in a brain region associated with visual processing. Two pairwise differences involved at least one ROI in a brain region associated with language processing. The remaining pairwise differences involved either a cerebellar or subcortical ROI.

Eleven of the pairwise differences involved higher FC in the MS group relative to controls, shown in Figure 1 with red squares and listed in Table 2 in the section MS > HC. Of those, 2 involved pairings between ROIs associated with visual processing, 5 involved a pairing between visual and language regions, and the remaining pairings involved other cerebellar or subcortical ROIs.

Overall, in terms of associated function, 20 of the ROI-to-ROI pairings included at least one ROI associated with visual processing, and 8 included at least one ROI associated with language. Eleven pairings included at least one ROI in the cerebellar vermis, a region associated with affective processing (34).

After categorization by SDMT z-score, we retained 43 people in the MS-CI group and 23 in the MS-CP group. Both MS-CI and MS-CP groups had a higher proportion of women than their nearest age-matched counterparts. The MS-CI group performed worse and the MS-CP group performed better on the SDMT than age-matched controls (Table 3).

We found 11 pairwise differences in FC between MS-CI and HC_MS-CI (Table 4), 10 of which involved lower FC for the MS group compared to the HC (Figure 2 blue squares), while one involved higher FC for the MS group (Figure 2 red square).

The pairwise differences indicating lower FC for the cognitively impaired MS subgroup as compared to HC included 6 pairings that included at least one ROI in a brain region associated with visual processing and 3 pairings including an ROI in a brain region associated with language. The remaining pairwise differences involved cerebellar or subcortical ROIs.

The single pairwise difference that showed higher FC for the MS-CI group compared to their matched control group involved an ROI associated with language processing.

Overall, in terms of associated function, 6 pairings involved an ROI associated with visual processing, 3 pairings involved an ROI associated with language processing, and 6 of the pairwise differences involved a cerebellar lobule which is associated with cognition (34).

We did not find any differences in FC between the MS-CP group and matched controls (HC_MS-CP).

We did not find any differences in FC between the MS-CI and the MS-CP groups.

One pairwise connection was significantly correlated with the SDMT scores in the MS group. FC between the left cerebellum 9 and the left inferior division of the lateral occipital cortex ROIs was observed (T = 4.72, pFDR = 0.039).

In the group of MS-CI individuals with impaired IPS, we showed altered FC in the cerebellum, thalamus, and basal ganglia, as well as areas associated with VIS and LAN networks. Current literature on associations between FC and IPS is contradictory, with conflicting findings on the correlation between SDMT scores and FC measures (1, 7, 10, 14, 15). Using a definition of the MS-CI group that is similar to our own (26, 27), Meijer et al. (17) examined the difference between CI and CP groups in whole-brain FC and structure. They reported that, alongside worsened structural integrity and reduced deep brain grey matter volume, their CI group also showed an increased FC metric compared to CP individuals with MS. Given the differences in study methodology and that the authors did not include the cerebellum in their analysis, only limited comparisons are possible. We found mainly lowered FC between cerebellar lobules and visual ROIs in our MS-CI group, as well as increased FC between right aMTG and the subcallosal cortex. More work that explores specific brain regions in an MS-CI group is needed.

There were FC changes in common at 2 ROI pairs between the analysis of all MS and HC, and the analysis of our MS-CI group compared to nearest age-matched controls. Both the MS group and the MS-CI group had lower FC between left occipital pole and right SCC and lower FC between left thalamus and cerebellar vermis 9. These results are consistent with previous systematic reviews which indicate altered FC involving thalamic and visual areas in MS (3, 6). They also reinforce the results of the first analysis we performed by indicating the same cerebellar, thalamic, and visual ROIs. In contrast, we also found altered FC between 9 ROI pairs in the MS-CI group that were not implicated, pairwise, by the analysis of the entire MS group. It is notable that 8 of the 9 ROI pairs involved lower FC of cerebellar ROIs with subcortical, visual, and language-related regions. Similarly, the connection between the left cerebellar 9 lobule and the inferior lateral occipital cortex being related to SDMT scores in the MS group complements the finding from the group-wise comparisons highlighting the relevance of FC between cerebellar and visual cortex to impaired cognition in MS.

We propose that regions of FC change in common between MS versus HC and MS-CI versus HC_MS-CI (i.e., thalamus, cerebellar vermis 7 and 9, OP, SCC) may represent the FC changes that are common to individuals with MS. A larger sample size of cognitively preserved individuals with MS may then detect changes in these same common regions. The additional regions present only in the MS-CI vs. HC_MS-CI contrast (i.e., cerebellum 6, 7b, 8) may represent a signature of additional IPS impairment. FC of cerebellar lobule 9 was implicated in impaired cognition in MS, based on the results of our linear regression of SDMT scores in the MS group. These posterior cerebellar lobules should be used as seeds in follow-up analyses that examine the relationship between FC and IPS in individuals with MS. While we observed ROI pairings that significantly differed between cognitively impaired participants with MS and age-matched healthy controls, these results were less robust than those observed in the overall MS vs. HC contrast. We speculate that this may be due to a combination of the fact that (1) the contrast between cognitively impaired participants with MS and healthy controls provides a narrower reflection of differences between MS and healthy state, relative to the FC underlying information processing speed impairment and (2) the contrast between cognitively impaired MS participants and their matched healthy controls included a smaller sample size for the contrast groups, compared to the overall MS vs. HC contrast. Thus, this contrast had less power to detect group differences than the overall MS vs. HC contrast. With a larger sample size of cognitively impaired participants with MS and controls, it may be possible to detect additional differences. In addition, larger sample sizes of cognitively impaired MS participants would support analyses focused on individual variability and associations between MS, FC, and cognitive impairment. Future work should attempt to replicate these findings in larger, more diverse samples, and should investigate the network properties of these regions. We expect that this will bring further insight into the FC changes that underlie MS-related declines in IPS.

The contrast between HC and participants with MS and CI cannot distinguish FC differences related exclusively to CI from those related to MS, given that the groups varied both in their cognitive status as well as by MS status. This was the rationale for including the contrast of the cognitively preserved MS participants and their matched healthy controls, and the contrast of the cognitively intact vs. cognitively preserved participants within the MS group. Our results did not isolate FC differences specific to MS without cognitive impairment in our comparison of cognitively preserved MS patients to HC. Neither did we isolate FC differences specific to cognition impairment in MS in our comparison of cognitively impaired vs. cognitively preserved individuals with MS. The small sample size of the MS-CP group, which was common to both contrasts, may account for this. Future work using a larger sample size of cognitively preserved participants with MS will determine if this is a false negative due to low power.

We made attempts to follow the standardization and study quality recommendations from Jandric and coauthors when designing this study, which include the study of phenotype-specific influences, controlling for age, using well-established measures, defining ROIs consistently with previous research, and conducting model-led research (7). We controlled for age as a covariate of no-interest in our analysis or by age-matching. We used an established and validated cognitive measure and a standard brain atlas for parcellation. We fell short of the recommendations in two ways. The number of MS participants with primary progressive MS and secondary progressive MS was small and we were therefore unable to consider potential differences by MS subtype. However, many studies to date have used either mixed samples or a sample consisting of solely individuals with relapsing-remitting MS (7). Our samples were appropriate to our aim of describing broad trends in FC that apply across disease subtypes. Finally, we did not design our study to assess a specific disease model; however, we expect that the results obtained here will inform hypotheses for future model-led study designs. For example, much work remains to determine the adaptive vs. maladaptive implications of increases and/or decreases of FC in MS compared to healthy state. Many studies have now identified difference in FC for those with MS compared to the healthy population (3), however, far fewer studies have examined how resting state FC relates to cognitive impairment in MS (9–16, 18). Without longitudinal research to allow assessment of changes over time, in association with changes in cognition, it is challenging to assert whether the differences in FC that we have observed in the present study reflect adaptive (compensatory) vs. maladaptive changes in those with MS. Continued research in this area is necessary to fully elucidate the implications for the difference in FC presently observed. Further, the objective of the present study was to identify differences between those with MS and healthy controls relative to the FC underlying cognitive impairment. Informed by the present findings, future research can and should aim to determine potential causal mechanisms underlying the differences in cognitive-impairment related-FC between those with MS and HC. This could include, for example, exploring how individual variability in measures like lesion load, which are central to the pathology of MS, relate to cognitive impairment and associated differences in FC.