Demographics

The cohort of 119 progressive MS cases comprised 102 cervical, 116 thoracic, and 107 lumbar spinal cord levels. Cases with varied demographic profiles were included to maximise generalisability of the results (Table 1). 83 cases were female (69.7%) with a cohort mean age of 61.7 ± 12.5 years, disease duration of 28.4 ± 11.2 years, time from onset to wheelchair of 17.3 ± 10.3 years, and post-mortem interval of 16.7 ± 6.6 h.

Prevalence of spinal cord demyelination and inflammatory activity

At least one spinal cord lesion was observed in 91/119 (76.5%) of cases with 80/91 (87.9%) harbouring an inflammatory lesion (Fig. 2a). Lesions were more likely to be present in the cervical compared to the lumbar region in a categorical variable analysis (OR = 10.01, SE = 3.96, p = 0.0056) with a significant linear trend observed in an ordinal variable analysis (Cervical > Thoracic > Lumbar, p = 0.0018) (Fig. 2b). Inflammatory sections were more common than purely inactive sections at all spinal cord levels (Cervical: Probability = 0.84, SE = 0.065, p = 0.0001; Thoracic: Probability = 0.70, SE = 0.067, p = 0.0031; Lumbar: Probability = 0.73, SE = 0.074, p = 0.0022) though all spinal cord levels were equally likely to be inflammatory (Fig. 2c).

Fig. 2

Case-Level Prevalence of Demyelination in the MS Spinal Cord. a Pie charts highlighting the observed proportion of cases with spinal cord lesions and those that harboured at least 1 inflammatory (active or mixed/active inactive) lesion. b–c Stacked bar charts depicting the proportion of MS cases that harboured lesions at each level of the spinal cord irrespective of stage (b) and classified by the presence of inflammation (c). Proportions represent observed values, and the asterisks indicate significant post-hoc pairwise comparisons following logistic mixed modelling and multivariate adjustment for multiple comparisons (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

In total, a total of 460 lesions were identified in the spinal cord of which 177 (38.5%) were cervical, 152 (33.0%) were thoracic, and 131 (28.5%) were lumbar. The proportion of lesions was higher in the cervical cord when compared to the lumbar region (Ratio = 1.36, SE = 0.156, p = 0.0216) with a significant linear trend (Cervical > Thoracic > Lumbar p = 0.0080) (Fig. 3a). Across all cord levels, 268 lesions (58.3%) were inflammatory with 78 (17.0%) being active and 190 (41.3%) being mixed active/inactive. 192 lesions (41.7%) were inactive (Fig. 3b,d). Overall, inflammatory lesions were more common than inactive lesions (Ratio = 0.72, SE = 0.068, p = 0.0004) (Fig. 3b), a finding observed in cervical (Ratio = 0.65, SE = 0.10, p = 0.0058) and lumbar (Ratio = 0.66, SE = 0.12, p = 0.019) levels but not the thoracic level (Ratio = 0.85, SE = 0.14, p = 0.33) (Fig. 3c). Mixed active/inactive and inactive lesions were more common than active lesions (Mixed active/inactive: Ratio = 2.46, SE = 0.331, p < 0.0001; Inactive: Ratio = 2.44, SE = 0.328, p < 0.0001) (Fig. 3d), a finding that was observed at every cord level (Cervical: Active-Inactive Ratio = 2.50, SE = 0.56, p = 0.0002, Active-Mixed Active/Inactive Ratio = 2.82, SE = 0.62, p < 0.0001; Thoracic: Active-Inactive Ratio = 2.69, SE = 0.62, p < 0.0001, Active-Mixed Active/Inactive Ratio = 2.15, SE = 0.51, p = 0.0033; Lumbar: Active-Inactive Ratio = 2.17, SE = 0.54, p = 0.0050, Active-Mixed Active/Inactive Ratio = 2.29, SE = 0.56, p = 0.0019) (Fig. 3e). Lesion proportions of all types were similar across levels (Fig. 3c,e).

Fig. 3

Lesion-Level Prevalence of Demyelination in the MS Spinal Cord. a Pie charts highlighting the proportion of lesions found within the cervical, thoracic, and lumbar spinal cord. b, d Pie charts depicting the proportion of lesions classified by the presence of inflammation across all levels. c, e Stacked bar chaerts illustrating the proportion of lesions at each cord level classified by inflammatory activity. Proportions represent observed values, and the asterisks indicate significant post-hoc pairwise comparisons following logistic mixed modelling and multivariate adjustment for multiple comparisons (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

Concordance of spinal cord lesion presence across levels

We next explored whether the presence of a cervical spinal cord lesion impacted the likelihood of a lesion to be present in the thoracic and/or lumbar spinal cord. The sensitivity of a cervical spinal cord lesion to predict the presence of thoracic lesions was high (84.2%) though specificity was poor (59.5%), a finding also seen in the lumbar spinal cord (sensitivity = 86.0%; specificity = 51.0%).

Topography of spinal cord demyelination and inflammatory activity

Representative histological patterns and heatmap summaries of lesion predilection sites demonstrated that the dorsal columns, lateral columns, and grey matter as a whole were consistently affected. Interestingly, the subpial surface was relatively spared at all spinal cord levels (Fig. 4a–c). A complementary analysis assessing the prevalence of anterior, posterior, and lateral funiculi involvement was undertaken (Fig. 4d). The probability of the lateral and posterior funiculi being impacted was consistently higher than the anterior funiculus (Fig. 4e).

Fig. 4

Pathological Patterns and Topography of Demyelination. a Representative histological images of lesion distribution patterns. The dorsal columns (green box), lateral columns (red box), and the whole grey matter (orange box) were consistently affected throughout the cohort (scale bars: 0.5 mm). Interestingly, the subpial surface was commonly spared (red, green, and orange boxes). When present, subpial demyelination only affected a limited circumference of the cord (blue box). b Lesion frequency heatmaps of total, inflammatory, and inactive lesions in the cervical, thoracic, and lumbar spinal cord. c The subtraction images highlight areas more commonly affected by inflammatory lesions (red), inactive lesions (blue), and those equally affected by both types (white). d Anatomical landmarks of the anterior (red), lateral (blue) and posterior (green) funiculi. e Bar charts depicting the proportion of MS cases that exhibited demyelination of the anterior, posterior, and lateral funiculi at each level of the spinal cord and globally. Proportion values represent estimated marginal means and standard errors derived from mixed logistic regression models. Proportions add up to over 100% as cases can have involvement of more than one funiculus. The asterisks indicate significant post-hoc pairwise comparisons after the multivariate adjustment for multiple comparisons (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

Prevalence and extent of demyelination at cerebrospinal fluid interfaces

Given the relative subpial sparing, demyelination at CSF interfaces was analysed. 76/91 (83.5%) cases with lesions exhibited demyelination of the white matter abutting any part of the subpial surface of which 61/76 (80.3%) exhibited inflammatory demyelinating activity (Fig. 5a). Contrasts derived from logistic regression mixed models indicated that subpial involvement was not significantly more likely to be at any spinal cord level (Fig. 5b). Sections with inflammatory subpial involvement were more common than those found to be inactive at all levels with this relationship being significant at cervical and lumbar levels (Cervical: Probability = 0.72, SE = 0.0714, p = 0.0083; Thoracic: Probability = 0.64, SE = 0.071, p = 0.055; Lumbar: Probability = 0.69, SE = 0.074, p = 0.020) (Fig. 5c). When comparing across levels, sections from the cervical, thoracic, and lumbar spinal cord were equally likely to have inflammatory subpial disease (Fig. 5c).

Fig. 5

Prevalence of Subpial White Matter Demyelination in the MS Spinal Cord. a Pie charts highlighting the observed proportion of cases with subpial white matter spinal cord lesions and those that harboured at least 1 inflammatory (active or mixed/active inactive) subpial white matter lesion. b–c Stacked bar charts depicting the proportion of MS cases that harboured subpial white matter lesions at each level of the spinal cord irrespective of stage (b) and classified by the presence of inflammation (c). Proportions represent observed values, and the asterisks indicate significant post-hoc pairwise comparisons following logistic mixed modelling and multivariate adjustment for multiple comparisons (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

While presence of subpial demyelination was common across cases, the proportion of the subpial surface affected by lesions (i.e. spinal cord circumference affected by subpial lesions relative to total cord circumference) was lower than random chance (Cervical, Thoracic, and Lumbar: p < 0.001) and did not differ significantly across levels (Cervical: 19.2%, Thoracic: 23.2%, Lumbar: 19.7%) (Fig. 6a). When only considering cases with subpial lesions, the proportion of the subpial surface affected by lesions remained relatively low (Cervical: 29.3%, Thoracic: 32.5%, Lumbar: 26.4%) (Fig. 6b). When stratified by lesion stage, inflammatory lesions impacted a larger subpial circumference than inactive lesions in the cervical (Odds Ratio = 0.448, SE = 0.18, p = 0.049) and lumbar (Odds Ratio = 0.363, SE = 0.16, p = 0.025) spinal cord but not in the thoracic spinal cord (Odds Ratio = 1.20, SE = 0.47, p = 0.64) (Fig. 6c). The proportional circumference impacted by active subpial lesions was significantly less than mixed active/inactive (Cervical: Odds Ratio = 24.83, p < 0.0001; Thoracic: Odds Ratio = 9.63, p = 0.0004; Lumbar: Odds Ratio = 6.70, p = 0.0012) and inactive lesions (Cervical: Odds Ratio = 13.49, p = 0.0007; Thoracic: Odds Ratio = 15.15, p < 0.0001; Lumbar: Odds Ratio = 4.10, p = 0.025) at all levels (Fig. 6d).

Fig. 6

Extent of Subpial Demyelination in the MS Spinal Cord. a–b Stacked bar charts depicting the proportion of the subpial white matter surface affected by lesions at each level of the spinal cord irrespective of stage including cases with lesions of all types (a) and those with only subpial lesions (b). c–d Stacked bar charts depicting the proportion of the subpial white matter surface affected by lesions stratified by inflammatory activity. Proportions represent observed values, and the asterisks indicate significant post-hoc pairwise comparisons following linear mixed modelling and multivariate adjustment for multiple comparisons (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

We explored the relationship between subpial demyelination in the spinal cord and the well-documented pattern of subpial demyelination in the cerebral cortex [6, 21, 25]. 38/68 (55.9%) cases exhibited subpial demyelination in the motor cortex. Cases with evidence of cortical subpial demyelination did not demonstrate more extensive subpial demyelination at any spinal cord level compared to cases with no subpial cortical demyelination (Cervical: Odds Ratio = 1.20, SE = 0.61, p = 0.73, Thoracic: Odds Ratio = 0.55, SE = 0.29, p = 0.079, Lumbar: Odds Ratio = 1.29, SE = 0.75, p = 0.86) (Supplementary Fig. 1, online resource).

We also investigated the central canal, as it represents another CSF interface. 67/91 (73.6%) cases had lesions affecting the central canal with 46/67 (68.7%) categorised as inflammatory (Supplementary Fig. 2a, online resource). Contrasts derived from logistic regression mixed models indicated that central canal lesions were equally distributed at all spinal cord levels (Supplementary Fig. 2b, online resource). Sections with inflammatory central canal involvement were not more common than those that were found to be inactive at all levels. All spinal cord levels were also equally likely to have inflammatory disease of the central canal (Supplementary Fig. 2c, online resource).

Given that recent MRI work suggested a gradient of spinal cord lesions extending from fluid interfaces inwards (i.e. subpial or central canal to the parenchyma), we recapitulated the analysis by Ouellette et al. 2020. One-way ANOVA analysis demonstrated that normalised lesion volumes differed amongst bins at all levels (Cervical: F = 27.59, p < 0.0001; Thoracic: F = 29.75, p < 0.0001; Lumbar: F = 19.85; p < 0.0001). Post-hoc analyses comparing all bins to the innermost bin (i.e. furthest from CSF boundaries) demonstrated a consistent decrease of normalised lesion volumes in the bins positioned closest to the CSF (bin 1 and bin 10) (Fig. 7a–c).

Fig. 7

Distance-Based Lesion Segmentation Analysis of Spinal Cord Demyelination. Bar charts displaying normalised lesion fractions within concentric distance zones extending from the subpial surface (bin 1) to the central canal (bin 10) in the cervical (a), thoracic (b), and lumbar (c) spinal cord. Differences were assessed within each level individually with a one-way repeated measures ANOVA. Post-hoc comparisons with Dunnett’s correction were made compared to bin 5 (the region mid-way between the central canal and subpial surface). The asterisks indicate significant post-hoc pairwise comparisons (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

Clinicopathological relationships

Finally, we evaluated the relationship between spinal cord demyelination and clinical outcome. Time from disease onset to wheelchair and disease onset to death (disease duration) were used as surrogate markers of disease severity and response variables for multivariate cox proportional hazard analyses (Fig. 8). Sex and age of onset were included as covariates in all analyses. Cases with lesions anywhere in the spinal cord were more likely to be wheelchair-dependent (HR 2.45, 95% CI 1.43–4.20) or die (HR 2.65, 95% CI 1.61–4.34) at younger ages (Fig. 8a–b). Subpial location had no impact on clinical milestones (time to wheelchair: HR 1.32, 95% CI 0.71–2.44; time to death: (HR 1.71, 95% CI 0.93–3.13) (Fig. 8c–d). We also assessed the impact of inflammatory activity on clinical disease milestones. In a case-level analysis, cases with evidence of inflammatory activity did not exhibit a faster clinical decline (time to wheelchair: HR 1.01, 95% CI 0.53–1.93; time to death: HR 1.78, 95% CI 0.93–3.38) (Fig. 8e–f). However, a more granular analysis using lesion proportions demonstrated that a higher proportion of inflammatory lesions was observed in cases that died at younger ages (HR 2.76, 95% CI 1.45–5.29) though no relationship was observed with wheelchair trajectory (HR 1.01, 95% CI 0.53–1.94) (Fig. 8g–h). Interestingly, the relationship with time to death was driven by the proportion of active lesions (HR 4.04, 95% CI 1.63–10.03) and not mixed/active inactive lesions (HR 0.89, 95% CI 1.88–3.95). Conversely, an increasing proportion of inactive lesions was associated with a slower time to death (HR 0.19, 95% CI 0.36–0.69) without impacting time to wheelchair (HR 0.99, 95% CI 0.52–1.90) (Fig. 8i–j).

Fig. 8

Clinicopathological Relationships Between Demyelination and Clinical Milestones. Forest plots displaying the hazard ratio and 95% confidence interval are derived from multivariate cox proportional hazards models controlling for sex and age of onset. Both case-level (a–f) and lesion-level (g–j) analyses were performed. The impact of spinal cord lesion presence on the time from MS onset to wheelchair use (a) and MS onset to death (b) was tested. In cases with lesions, the impact of subpial lesion presence was tested on the time from MS onset to wheelchair use (c) and MS onset to death (d). The impact of inflammatory status was also tested on the time from MS onset to wheelchair use (e) and MS onset to death (f). Inflammatory and inactive lesion proportions were tested on the time from MS onset to wheelchair use (g, i) and MS onset to death (h, j) Coefficients and p-values are highlighted in red for significant predictors (p < 0.05) in the models