Introduction

Disease trajectory in relapsing-remitting multiple sclerosis (RRMS) is highly variable, and difficult to predict early in the disease course. Improved objective prognostic markers, which are sensitive to early RRMS pathological features, are needed to better inform early treatment decisions from the point of diagnosis, and for future clinical trials.

Episodic immune-mediated demyelination is thought to result in axonal damage and neuronal dysfunction [1]. Identification of new or enlarging white matter lesions (WML) visible on T2-weighted or T2-FLAIR (fluid-attenuated inversion recovery) MRI is widely used as an indicator of interval inflammatory disease activity [2]. The number and volume of WML do not, however, adequately explain clinical disability in RRMS [3, 4]. Microscopic damages to grey matter (GM) and normal-appearing white matter (NAWM) are also pathological features of RRMS, [5] and heterogeneity of microstructural damage within WML that appears similar on T2-weighted MRI is observable using quantitative microstructure-sensitive MRI methods [6,7,8,9].

A subset of T2-weighted WML are visibly hypointense on T1-weighted MRI, reflecting T1 prolongation, [10] and have variously been described as ‘black hole’ or ‘T1w-hypointense’ lesions [11]. Such ‘T1w-hypointense’ lesions are thought to represent more severely damaged white matter [12] and are associated with worse clinical disability [13, 14]. Radiological evaluation of T1-weighted hypointensity is nevertheless subjective and qualitative, and will not capture intralesional T1 heterogeneity or subtle change in normal-appearing tissue.

The spin-lattice relaxation time (T1) of protons in brain parenchyma is inversely proportional to myelin density [11]. Previous imaging-histological correlation indicates that T1 quantification may provide a non-invasive marker of myelin density in multiple sclerosis (MS), [12] and is sensitive to subtle abnormalities in normal-appearing GM and NAWM in RRMS [5, 15,16,17,18,19,20]. Widespread T1 prolongation is seen in RRMS compared with healthy controls, and correlates with disability [14, 20, 21]. Quantitative mapping of T1 may also distinguish microstructural heterogeneity within WML; substantially prolonged T1 tissue volumes appear to be a better predictor of disability than visual assessment of T1-weighted hypointense WML [22]. The contribution of prolonged T1 components in WML and other brain tissues to clinical disability in early disease, and how these evolve with time remain to be explored.

The primary aims of this study were to assess whether changes in tissue microstructure integrity, reflected in T1 heterogeneity and prolongation, are quantifiable within WML and other brain regions over the first year following RRMS diagnosis, and to ascertain the relationship between T1 prolongation and clinical disability. We hypothesised that prolonged T1 components would progress with time and be associated with worsening clinical disability.

Our secondary exploratory aims were to examine intralesional T1 heterogeneity and its relationship with disability, investigate single-time point measures as predictors of clinical disability trajectory following RRMS diagnosis and establish the test-retest reliability of T1 measures.

Materials and methods

Ethics approval was obtained from the South East Scotland Research Ethics Committee 02 (REC 15/SS/0233). The study conformed to the Declaration of Helsinki 2000 (amendments in 2002 and 2004) and Good Clinical Practice ICH guidelines. All participants gave written informed consent and data were pseudo-anonymised.

Participants

Seventy-nine patients were recruited to a single-centre sub-study of FutureMS, a prospective multi-centre longitudinal inception cohort study of people with recently diagnosed RRMS across Scotland (Fig. 1) [23, 24]. Inclusion criteria were as follows: < 6 months from diagnosis; aged ≥ 18 years; and capacity to provide informed consent. Exclusion criteria were as follows: previous use of disease-modifying therapies (DMTs); clinical trial participation prior to baseline assessment; and contraindications for MRI. FutureMS sample size calculation has been previously described; [24] Edinburgh participants were pragmatically offered extended MRI once the sub-study recruitment opened in November 2017 until March 2019. Twelve age- and sex-matched healthy controls were recruited for test-retest purposes. Exclusion criteria were contraindication for MRI and incidental MRI findings.

Fig. 1
figure 1

Flowchart showing inclusion and exclusion criteria for FutureMS quantitative T1 sub-study

Full size image

MRI and Expanded Disability Status Scale (EDSS) score were acquired at baseline (between November 2017 and July 2019) and 1-year follow-up. Handling of missing data and loss to follow-up are detailed elsewhere; briefly, data missing at random were imputed using predictive mean matching [23].

MRI protocol

MRI was performed on a 3-Tesla Siemens MAGNETOM Prisma clinical system at the University of Edinburgh Imaging Facility. Whole-brain T1 mapping was performed using a variable flip angle (VFA) 3D multi-echo spoiled gradient recalled echo sequence (total acquisition time: 12 min 28 s) as part of a comprehensive MRI protocol, including a 2D T2-weighted FLAIR and 3D T1-weighted MPRAGE (Table S1) [24]. Control subjects underwent an identical MRI protocol, repeated within 2 weeks, to determine test-retest agreement.

Structural MRI data processing

WML segmentation was performed on baseline 2D T2 FLAIR images using an adaptation of a previously reported thresholding method, [25] with manual correction [24]. Tissues were segmented with FreeSurfer (v6.0, http://surfer.nmr.mgh.harvard.edu/) from the T1-weighted MPRAGE [24]. These included NAWM; cortical GM (cGM); medial temporal regions (hippocampi and amygdala); basal ganglia (caudate, pallidum and putamen); thalami; and global deep GM (DGM; combined thalami, basal ganglia and medial temporal regions; Fig. 2). WML load was calculated as the percentage of occupied intracranial volume [24].

Fig. 2
figure 2

An axial T1 map with superimposed brain tissue segmentations: normal-appearing white matter (yellow), white matter lesions (red), cortical grey matter (light blue), thalami (dark blue), basal ganglia (green) and medial temporal regions (pink)

Full size image

T 1 data processing

T1 parametric maps were obtained from VFA gradient echo images in MATLAB (R2018b) using previously described equations (Fig. 3; software available: https://doi.org/10.7488/ds/2965) [26, 27]. Tissue segmentations were registered to T1 maps using FSL (v6.0.1) FLIRT [28].

Fig. 3
figure 3

Axial T1 (colour) map brain slices for a person with recently diagnosed relapsing-remitting multiple sclerosis taking part in FutureMS. The colour bar represents T1 in seconds of the colour scale

Full size image

Median T1 within each tissue was calculated per patient (fslstats). To investigate T1 WML heterogeneity, the proportion of WML voxels within each 0.25-s interval between 0.5 and 2.00 s was calculated for each patient at baseline and follow-up (RStudio v4.0.2).

As T1 prolongation is associated with demyelination [12] and disability, [3] we also characterised WML components with relatively prolonged T1 using a thresholding approach [22]. The number of WML voxels with (1) T1 ≥ 2.00 s (here termed ‘prolonged T1’) and (2) supramedian T1 (calculated per patient relative to the baseline cohort mean of WML median T1) was thus calculated at each time point.

Statistical analysis

All statistical analyses were performed in RStudio (v4.0.2).

Test-retest healthy controls

To establish test-retest reference levels, Bland-Altman limits of agreement were calculated from healthy control data, following previously reported methodology [29]. To assess whether the difference between time points was statistically significant, sign tests were performed.

Longitudinal change in T 1 metrics

Paired two-sided t-tests were used to examine longitudinal changes in median T1 for each tissue and, as separate analyses, 1-year changes in the number of prolonged/supramedian T1 voxels and WML load. To account for WML voxels forming during the 1-year follow-up period, which could positively skew WML T1 data, a post hoc paired t-test was performed including only WML voxels already present at baseline. T1 changes in NAWM and cGM were plotted against test-retest limits for visual comparison. To control for type II error, False Discovery Rate (FDR) correction (q < 0.05) was applied for primary analyses [30].

Relationship with clinical disability

The relationship between T1 and EDSS score at baseline was investigated using ordinal logistic regression. In view of limited EDSS changes within our recently diagnosed cohort, 1-year change in EDSS score was dichotomised as either worsening EDSS, defined as ≥ 0.5 points, or stable/improving; the relationship between disability groups and both baseline and 1-year changes in T1 measures was investigated using binomial logistic regression, adjusted for covariates and interactions as appropriate (Table S2 for extended methods). FDR correction was applied separately for the three sets of analyses, which are assumed to be distinct families of hypotheses [30].

Regression analyses were repeated for prolonged and supramedian T1 WML measures. The relationship between WML prolonged T1 measures at 1-year follow-up and EDSS score at follow-up was also investigated using ordinal logistic regression.

Results

Demographics

Complete data were available from sixty-two RRMS patients (Fig. 1), and demographics (Table 1) were comparable to the wider FutureMS cohort [23]. Six participants (n = 3 worsening versus n = 3 stable/improving disability) reported a relapse in the 6 weeks prior to baseline MRI (range: 11–34 days), one of whom received oral steroid treatment (27 days before MRI). No relapses or steroid treatments were reported in the 6 weeks prior to follow-up. At follow-up, 61% (n = 38) of participants had begun treatment with DMTs, mainly dimethyl fumarate (n = 23; Table S3). There was no significant difference in DMT status between disability groups (Χ2 = 0.79, p = 0.37).

Table 1 Cohort demographics of the relapsing-remitting multiple sclerosis (RRMS) participants. Differences between disability groups were assessed with Welch’s t-test for age, baseline/follow-up Expanded Disability Status Scale (EDSS) score, white matter lesion load and interval between baseline and follow-up MRI, and Pearson’s chi-squared (Χ2) test of independence with Yates’ continuity correction for sex and disease-modifying therapy (DMT) status
Full size table

One control participant was excluded due to an unexpected incidental finding; test–retest data was therefore available from eleven controls (age 44 ± 11, 7 female).

Test-retest reliability

Test-retest agreement determined from control data is shown in Fig. S1. The difference in median T1 was not significant for all brain regions (Table S4).

Characterising T 1 in white matter lesions

Qualitative assessment showed varied proportions of WML voxels within each 0.25-s T1 range between patients, with a greater proportion in higher T1 ranges at follow-up (Fig. 4b) relative to baseline (Fig. 4a). Visual inspection showed discrete regions of relative T1 prolongation were present across WML (Fig. 5a–c), generally coinciding with T1-weighted ‘black holes’ (Fig. 5d–f).

Fig. 4
figure 4

White matter lesion T1 heterogeneity in our cohort of people with recently diagnosed relapsing-remitting multiple sclerosis (RRMS) (n = 62). Each bar represents one participant with RRMS. The colours of the stacked plot relate to the proportion of a given participant’s white matter lesion voxels within a given T1 range at (a) baseline and (b) 1-year follow-up

Full size image
Fig. 5
figure 5

Spatial distribution (a–c) of prolonged T1 (i.e. ≥ 2.00 s) white matter lesion (WML) voxels (red) superimposed on all WML voxels (yellow) and whole-brain T1 maps (greyscale) in three representative relapsing-remitting multiple sclerosis study participants, with corresponding T1-weighted MPRAGE structural images (d–f)

Full size image

Longitudinal change in T 1 metrics

WML median T1 decreased by a mean average of 31 ms longitudinally (t(61) = 2.98, FDR-corrected p < 0.05, 95% CI [− 52, − 10]); no other tissues showed significant longitudinal T1 changes (Table 2). Longitudinal change in NAWM and WML T1 for the majority of patient data points fell within test-retest Bland-Altman limits established in healthy controls (Fig. S2).

Table 2 Brain tissue median T1 summary statistics and results for paired t-tests
Full size table

A longitudinal increase in WML load (mean average 0.235%, t(61) = 11.3, p < 0.001, 95% CI [0.193, 0.277]) prompted a post hoc analysis investigating only WML voxels present at baseline. This showed no significant change in median WML T1 (t(61) = 0.50, p = 0.618; Table 2).

There was a groupwise mean increase of 176 prolonged T1 WML voxels (t(61) = 5.11, p < 0.001, 95% CI [107, 246]) and an increase of 463 supramedian WML voxels (t(61) = 6.22, p < 0.001, 95% CI [314, 612]). When WML voxels which had formed during the follow-up period were excluded, changes in both measures remained significant (p < 0.001).

Relationship with clinical disability

Baseline median T1 was not associated with baseline EDSS score (all p > 0.05; Table S5) nor with EDSS worsening (all p > 0.05; Table S6) in any tissue studied. One-year changes in T1 (Table S7) within NAWM, WML, cGM, DGM, the basal ganglia and thalami (but not the medial temporal regions) were however positively associated with a greater risk of EDSS worsening (all p < 0.05; Fig. 6, and Table 3). The relationship between change in WML T1 and EDSS worsening was mediated by an interaction with age, whereby older age strengthened the association.

Fig. 6
figure 6

One-year change in T1 in our recently diagnosed relapsing-remitting multiple sclerosis cohort (n = 62). Boxplots are shown for normal-appearing white matter (NAWM), cortical grey matter (cGM), global deep grey matter (DGM), thalami, medial temporal regions, basal ganglia and white matter lesions (WML, b) including only WML voxel present at baseline. Boxplots are grouped according to dichotomised change in Expanded Disability Status Scale (EDSS) score over 1 year of  < / ≥ 0.5 points. Asterisk (*) denotes significant (p < .05 with False Detection Rate correction) association between difference over 1 year in T1 and disability group after adjusting for age, baseline T1, 1-year change in lesion load, disease-modifying therapy status and any significant interaction effects (see text for details); n.s., not significant

Full size image
Table 3 Relationship between 1-year change in T1 and change in disability in newly diagnosed relapsing-remitting multiple sclerosis (n = 62). Results are shown for binomial logistic regression models investigating the relationship between the change in median T1 in each tissue and dichotomised change in Expanded Disability Status Scale (EDSS) score (stable/improving vs worsening EDSS [< / ≥ 0.5 points]) over 1 year. Covariates were age, median baseline T1, 1-year change in white matter lesion load and disease-modifying therapy status. Continuous variables were centred and scaled, and adjusted odds ratios are for standardised data
Full size table

At baseline, greater numbers of prolonged T1 and supramedian WML voxels were independently associated with higher baseline EDSS (both p < 0.05; Table 4); neither measure was predictive of EDSS worsening (Table 5). One-year changes in prolonged T1 and supramedian WML voxel counts were positively associated with EDSS worsening (p < 0.05 and p < 0.01 respectively; Table 5). Prolonged and supramedian WML T1 measures at follow-up were also positively related with follow-up EDSS score (p < 0.05 and p < 0.01 respectively; Table 4).

Table 4 Cross-sectional relationship between prolonged/supramedian T1 white matter lesion (WML) voxel count and disability in newly diagnosed relapsing-remitting multiple sclerosis (n = 62). Results are shown for ordinal logistic regression models investigating the relationship at (a) baseline and (b) 1-year follow-up between the number of prolonged/supramedian T1 WML voxels and Expanded Disability Status Score (EDSS). WML voxels were classed (non-exclusively) as prolonged T1, where T1 ≥ 2.00 s, and supramedian T1, where T1 ≥ cohort mean of baseline median WML T1. Covariates were age and WML load (at the respective time point)
Full size table
Table 5 Relationship between prolonged/supramedian T1 white matter lesion (WML) voxel counts and 1-year change in disability in newly diagnosed relapsing-remitting multiple sclerosis (n = 62). Results are shown for binomial logistic regression models investigating the relationship between (a) baseline and (b) 1-year change in prolonged/supramedian T1 voxel counts and 1-year change in dichotomised Expanded Disability Status Scale (EDSS) score (stable/improving vs worsening EDSS [< / >  = 0.5 point increase]). WML voxels were classed as prolonged T1 if T1 ≥ 2.00 s and classed (non-exclusively) as supramedian if T1 ≥ the mean average of the baseline cohort median WML T1. Model covariates were §age and lesion load; §§age, baseline voxel count, 1-year change in WML load and disease-modifying treatment status at follow-up (untreated versus treated)
Full size table

Discussion

Brain T1, and its relationship with clinical worsening, was quantified longitudinally in a newly diagnosed RRMS cohort. Longitudinal cohort-wide changes in median T1 were not observed for GM, NAWM or WML. There was significant T1 heterogeneity in WML; WML voxels with more marked T1 prolongation increased in number over time, and were associated with disability at baseline and 1-year follow-up. Greater T1 increases in normal-appearing brain and prolonged T1 WML components also contributed significantly to evolving disability in the year following RRMS diagnosis.

T 1 variation

Measured T1 was higher than literature values, [17, 22] which is attributable to differences in acquisition methods. Average WML T1 values were higher than NAWM, consistent with demyelination [31]. As with previous findings in RRMS, [22] wide variance in WML T1 and substantial overlap with cGM T1 values suggest heterogeneous myelin loss; [12] in our cohort, average WML T1 was lower than cGM, which may reflect the earlier disease stage and less severe WML damage. Medial temporal T1 was significantly higher than other tissues; measurements in this region are likely confounded by tissue interface effects, and therefore unreliable.

Demyelination appears to be the major determinant of T1 prolongation in white matter, [31] except within fully demyelinated WML where there is a ‘myelin floor effect’, and axonal content dominates variation in T1 [12, 32]. Increased tissue water content, [11] due to neuroinflammation or oedema, [33, 34] may also increase T1. Subtle T1 prolongation in NAWM is therefore likely to represent demyelination, whereas more markedly prolonged T1 components within WML may result from both myelin and axonal loss. Myelin and axonal content are nonetheless strongly associated, [31] and we therefore use the term microstructural damage to reflect the difficulty in establishing their individual contributions to disability in WML and normal-appearing brain.

Our multiple threshold approach in WML is similar to methods applied previously, [22, 35] and analogous to histogram-based analyses [35, 36]. Our findings indicate varying degrees of microstructural damage across individuals at diagnosis, independent of WML load, which accords with lesion variation reported in more advanced disease across MS subtypes, [35, 37,38,39] and from other microstructure-sensitive methods [6,7,8].

Longitudinal change in T 1 and prolonged WML T 1 components

We found no significant longitudinal median T1 change in grey or white matter regions over the year following diagnosis, in line with previous serial T1 measurements in early RRMS [5]. A groupwise reduction in WML T1 over time was attributable to new or enlarging WML; when only WML voxels that were present at baseline were included, no significant T1 change was found. New WML voxels with less tissue damage developing over the follow-up period therefore appear to drive the cohort-wide WML T1 decrease, rather than remyelination. WML T1 reduction over time in early RRMS has been reported previously, [19] and this potential confound should be considered for future study design.

Marked intersubject variability in the proportions of higher WML T1 components guided our focus on WML voxels with supramedian T1 and values over 2.00 s; both aim to identify more disrupted microstructure [12]. We found that the number of thresholded T1 WML voxels increased significantly over 1 year following RRMS diagnosis, indicating increasing microstructural damage within existing WML. To our knowledge, this study is the first to examine the longitudinal evolution of prolonged T1 components, although thresholding approaches using T1 [12, 22] and normalised T1-weighted images [35] have been applied cross-sectionally in MS.

T 1 and clinical disability

Although positive associations between clinical disability and T1 within NAWM [14, 21] and WML [14] have been described previously, we did not find an association between cross-sectional median T1 measures and clinical disability. Previous studies were in mixed MS subtypes at varying disease stages however, and findings varied; [40] the earlier disease stage of our more homogeneous RRMS cohort may account for these differences.

Conversely, cross-sectional prolonged T1 and supramedian WML voxel counts were associated with EDSS score at the time of diagnosis, a relationship that persisted at 1-year follow-up. The significant increase seen in these WML components over time was also associated with EDSS worsening. This is in line with the literature, where thresholded T1 has been shown to provide disability-relevant information additional to visual assessment of ‘T1-w hypointense’ WML in more advanced disease [22]. Our data suggest that prolonged T1 WML components contribute disproportionately to clinical disability in early RRMS, which may reflect both demyelination and axonal loss [12]. Thresholded WML T1 measures may therefore provide useful disease stratification in early RRMS.

Longitudinal increases in median T1 in WML tissue present at baseline in patients with worsening disability, and decreases in those with stable or improving disability, are consistent with disability-relevant microstructural damage and remyelination, respectively. An earlier study did not find any association between a derived WML load-weighted T1 measure and 1-year change in EDSS score in RRMS, [10] which may reflect differences in cohort size, disease duration and analytical approach.

In our cohort, greater 1-year increase in median T1 in NAWM, cGM and thalamus was also associated with EDSS worsening, whereas T1 was relatively stable or decreasing in participants with stable or improving EDSS, resulting in an absence of net cohort-wide T1 change in GM or NAWM. T1 therefore captures disability-relevant demyelination that is not visible on conventional MRI, and the direction of T1 change over 1 year may indicate disability trajectory in early MS.

None of our baseline measures were predictive of EDSS worsening over the following year. High individual variance in baseline brain microstructural integrity, despite a relatively homogeneous diagnosis point, and limited 1-year clinical worsening following RRMS diagnosis may be limiting factors.

Limitations

T1 quantification may be useful as a group-level patient stratifier for targeted clinical trials; the small magnitude of changes and reproducibility of measures may, however, limit application in individual patients. VFA acquisition for T1 approximation is rapid and readily implemented, but does not correct for B1 inhomogeneities, which may reduce sensitivity to biological change and introduce spatial bias, although unlikely to substantially effect longitudinal measures. WML were segmented on 3-mm 2D FLAIR slices rather than 3D FLAIR due to better contrast resolution, which may introduce partial volume effects. Although all participants were recruited within 6 months of diagnosis, the relationship with true disease onset is variable and difficult to determine reliably; [41] diagnosis was therefore chosen as a pragmatic and measurable time point. Effects of intercurrent treatment are likely to be minimal: MRI was performed a minimum of 3 weeks post-steroid treatment; DMT status was not related to disability outcome; however, treatment variation in this cohort precludes meaningful sensitivity analysis. Innate EDSS limitations are well-recognised [42]. Moreover, 1-year change in EDSS score showed significant intersubject variability (Fig. S3); overall disability worsening was minor, and confounded by improvement following the clinical episode that initiated RRMS diagnosis. Five-year follow-up of the FutureMS cohort [23] will allow more reliable evaluation of T1 measures as predictors of disease trajectory.

Conclusion

Quantitative T1 mapping is sensitive to disability-relevant microstructural changes in the year following RRMS diagnosis. T1 thresholding approaches demonstrate heterogeneity of microstructural damage in WML; WML components with prolonged T1 increase with time, and are significantly associated with clinical disability. Widespread subtle demyelination can also be detected in NAWM and GM in patients with worsening disability. Further studies in well-characterised RRMS cohorts over longer time periods will help clarify the utility of T1 mapping as a disease stratifier, and for potential application in future therapeutic trials.