Brainstem–cortex disconnection in amyotrophic lateral sclerosis: bulbar impairment, genotype associations, asymptomatic changes and biomarker opportunities

Imaging data from 198 patients with ALS and 108 healthy controls (HC) were included in this study (Table 1). All participants gave informed consent in accordance with the Ethics Approval of this research project (Beaumont Hospital, Dublin, Ireland – IRB REC08/90). A prospective study design was implemented with the recruitment of incidence ALS cases. The mean symptom duration of patients from symptom onset to first scan was 15.5 months (median 16) with a range of 5–27 months. Participating ALS patients were diagnosed according to the revised El Escorial criteria. Exclusion criteria included prior neurosurgery, prior cerebrovascular events, traumatic brain injury, comorbid neoplastic, or neuroinflammatory diagnoses. Patients with comorbid psychiatric disease, patients who were unable to tolerate MR scanning, patients with incomplete MR acquisition and patients without genetic information were excluded. Inter-scan interval for longitudinal follow-up was four months. In three main analysis streams, patients were either assessed as a single group (analysis stream 1), stratified by C9orf72 status (analysis stream 2) or by the presence bulbar symptoms (analysis stream 3). Basic demographic and clinical variables (age, sex, handedness, years of education, medications, body region of symptom onset, family history of ALS or FTD), were recorded on the day of MRI scanning and all patients had their total ALSFRS-r and ALSFRS-r sub-scores documented at the time of their scan. All patients were screened for intronic GGGGCC repeat expansion in C9orf72 by repeat-primed PCR. Capillary electrophoresis outcomes were visualised using GeneMapper version 4.0 and patients exhibiting 30 or more repeats were considered C9orf72-positive. Additionally, all participating patients were screened for a panel of protein-altering, exonic or splice-site variants present in 32 genes linked to ALS in the ALS online database (ALSod). [6]

MR data were acquired on a 3 Tesla Philips Achieva platform and the protocol included structural T1-weighted (T1w), resting-state functional MR (rsfMRI) and diffusion-weighted (DWI) pulse-sequences (Fig. 1). The imaging protocol has been previously described [7]. Briefly, T1-weighted (T1w) images were acquired with a 3D Inversion Recovery prepared Spoiled Gradient Recalled echo (IR-SPGR) sequence with the following parameters; field-of-view (FOV) of 256 × 256 × 160 mm, flip angle = 8°, spatial resolution of 1 mm3, SENSE factor = 1.5, TR/TE = 8.5/3.9 ms, TI = 1060 ms. Diffusion tensor images (DTI) were acquired with a spin-echo echo planar imaging (SE-EPI) pulse sequence using a 32-direction Stejskal-Tanner diffusion encoding scheme, FOV = 245 × 245 × 150 mm, 60 slices with no interslice gap, spatial resolution = 2.5 mm3, TR/TE = 7639/59 ms, SENSE factor = 2.5, b-values = 0, 1100 s/mm2, dynamic stabilisation and spectral presaturation with inversion recovery (SPIR) fat suppression. Echo-planar imaging (EPI) was used to investigate fluctuations in the blood oxygen level-dependent (BOLD) signal for resting state functional imaging with eyes closed using the following imaging parameters: 30 axial slices, repetition time (TR) / echo time (TE) = 2000 ms/35 ms, flip angle (FA) = 90°, pixel bandwidth = 1780, Hz/Px. Spatial resolution: 2.875 × 2.875 × 4 mm.

T1-wieghted structural data were pre-processed in for cortical thickness (CT) calculations, medullary volume estimation, brainstem shape profiling and for downstream image registration. CT and medullary volumes were calculated using FreeSurfer version 7.1.0 [8], including automated image segmentation, surface reconstruction and individual CT map output. We defined the bulbar segment of the motor cortex based on the Brainnetome atlas [9], which provides functional cortical parcellation based on multimodal imaging data. We refer to labels “A4tl_L” and “A4tl_R” (“tongue/larynx”) of the Brainnetome atlas as the “bulbar cortex”. For brainstem segmentation and subsequent medullary volume estimation, we relied on FreeSurfer’s segmentBS pipeline [10], which uses a Bayesian algorithm to delineate probabilistic boundaries. To investigate brainstem outline deformations, the first pipeline [11] of the FMRIB´s Software Library was implemented [12]; images were skull-stripped and bias-corrected using FSL’s fsl_anat pipeline. Subsequently, subcortical structures were segmented using a Bayesian approach and parametrized those labels as surface meshes to run statistical comparisons.

Functional data were pre-processed to estimate functional connectivity (FC). FSL’s feat [13] was utilized for brain extraction, slice-time correction, and motion correction. Additionally, FSL’s AROMA algorithm [14] was implemented to correct for head-motion-related artifacts. Each patient’s pre-processed functional image was linearly co-registered to the native high-resolution structural scan using 6 degrees of freedom (DOFs), and for higher-level group comparisons, non-linearly warped to the MNI152 2 mm standard space (12 DOFs). FC was defined as Fisher z-transformed correlation between the mean time courses of the brainstem and the bulbar cortex (separately for the two hemispheres). FC was calculated within Matlab R2021b using tools from the CoSMoMVPA toolbox [15].

To determine structural connectivity (SC), diffusion-weighted (DW) data were used. Tools from MRtrix3 (version 3.0.3) [16] were used for pre-processing including noise removal [17], removal of Gibb’s Ringing artifacts [18], motion / eddy current [19] and bias-field corrections [20]. Pre-processed DW images were then aligned to high-resolution T1-weighted data. As fractional anisotropy (FA) is a composite metric of the three eigenvalues (λ₁, λ₂, λ₃), it is histologically non-specific to the underlying white matter pathology and merely offers an overall proxy of white matter integrity. Accordingly, in our study a diffusion-tensor model was fitted to the data to estimate cortico-medullary axial diffusivity (AD) (λ₁) and radial diffusivity (RD) ((λ₂ + λ₃)/2) separately [21]. Tractograms were calculated between the medulla and the right/left bulbar cortices separately in each participant in native space. A probabilistic algorithm was implemented to produce a fixed number of streamlines (n = 5000) between the pairwise ROIs. We then binarized these tractograms to extract the mean AD/RD values per tract in each subject using the previously calculated tensor maps. Multimodal MR data of patients and controls were interrogated both for baseline differences (cross-sectional modelling) as well as differences in progression (longitudinal modelling). Not only disease-associated signatures were explored (analysis stream 1) by contrasting all ALS patients to controls, but two specific ALS cohorts were further evaluated. The imaging profile of hexanucleotide repeat expansion carriers in C9orf72 (“C9 + ”) (analysis stream 2) and bulbar asymptomatic (BA) patients (analysis stream 3) i.e. C9-, spinal onset patients with a maximum ALSFRS-r bulbar score, were evaluated in dedicated analyses.

The age and education profile of patients and controls were contrasted using Welch two-sample t-tests and the ratios of gender and handedness distributions examined by Chi-square testing. All statistical analyses were performed within RStudio (R version 4.2.1) [22]. Multimodal MR data of patients and controls were interrogated both for baseline differences (cross-sectional modelling) as well as differences in progression (longitudinal modelling). Not only disease-associated signatures were explored (analysis stream 1) by contrasting all ALS patients to controls, but two specific ALS cohorts were further evaluated. The imaging profile of hexanucleotide repeat expansion carriers in C9orf72 (“C9 + ”) (analysis stream 2) and sporadic bulbar asymptomatic (BA) patients (analysis stream 3) i.e. C9-, spinal onset patients with a maximum ALSFRS-r bulbar score, were evaluated in dedicated analyses. Each of these streams were explored cross-sectionally and longitudinal, whereby the cross-sectional model comprised a simple linear model for analysis stream 1 and a one-way analysis of variance (ANOVA) for analysis streams 2 and 3 (given that these contrasted three groups). Age, sex and handedness were included as covariates and volumetric analyses were also adjusted for total intracranial volume (TIV). Additionally, post-hoc Tukey’s Honest Significance Difference (HSD) testing was implemented to explore pairwise differences between groups. To evaluate brainstem outline alterations at baseline, non-parametric statistical comparisons were implemented between all ALS patients and HC (analysis stream 1), C9 + vs. C9- ALS patients (analysis stream 2) and BA vs. bulbar symptomatic (BS) C9-patients (analysis stream 3). For each comparison, we applied FSL’s randomise [23] algorithm with 5000 permutations and 2D-optimized threshold-free cluster enhancement (TFCE), using a two-sample t-test design covarying for age, gender and handedness. Since two-sided testing was performed, significance threshold was set to p ≤ 0.025. Longitudinal changes were evaluated in linear mixed effects models, where Time (i.e. imaging timepoint) was considered as a random effect and the subjects as fixed effects. In “analysis stream 1”, the interaction between Time and Group, in “analysis stream 2” progression differences between C9 + vs. C9–/HC and in “analysis stream 3”, differences between C9- BA and BS patients were evaluated. Similar to the cross-sectional analyses at baseline, age, sex and handedness were included as covariates. To test the discriminatory potential of specific imaging metrics in our panel of radiological indices, we ran Receiver Operator Characteristics (ROC) analyses between all ALS patients and controls and tested the likelihood that a given area under the curve (AUC) value differed from 0.5. Given the relentless clinical progression in ALS, we hypothesized that AUCs may increase over time, therefore we included only patients who had a baseline and a 12-month follow-up scan (N = 67 vs. 147 HC) and computed ROC and corresponding AUC values for each imaging metrics at the two time-points 12-months apart. To test the hypothesis of increasing AUCs, we ran a one-sided paired t-test between the AUC of all 9 analysed imaging metrics at baseline and follow-up. The ROC/AUC analyses were carried out within RStudio using the pROC package [24]. Clinico-radiological associations were explored for bulbar motor (ALSFRS-r) scores. Cross-sectional associations with bulbar subscores at baseline were tested in a linear model incorporating the relevant covariates. Longitudinal associations were investigated based on the change in ALSFRS-r bulbar subscores as the independent, and the difference between the given MRI metric as dependent variables, correcting for age, gender and handedness.

Group-level outputs and additional information on data processing can be requested from the corresponding author. Individual-patient clinical, genetic and imaging data cannot be transferred due to institutional and departmental policies.

Demographic a clinical data are presented in Table 1. Welch two-sample t-test revealed no significant age differences between ALS patients and controls (t(241.6) =  – 0.58, p = 0.57). Chi-square test captured sex differences (X² (1, N = 306) = 6.06, p = 0.01), but no differences in handedness (X² (1, N = 306) = 0.08, p = 0.78). All patients tested negative for a panel of protein-altering, exonic or splice-site variants in 32 genes linked to ALS.

The main statistical outputs of neuroimaging analyses are presented in Table 2 and illustrated in Figs. 2, 3, 4, 5. Considering all ALS patients (Fig. 2), more rapid cortico-medullary disconnection was identified in patients compared to controls in the right hemisphere. The interaction effects (Group x Time) in the longitudinal models revealed RD increase [t(237) = 2.030, p = 0.044], AD increase [t(237) = 2.210, p = 0.028] and FC decline over time [t(237) =  – 2.187, p = 0.030]). Interestingly, the CT of the bilateral bulbar cortex – while not exhibiting progressive change – was significantly thinner in patients at baseline (both RH/LH: p < 0.001). No medullary volume (MV) or brainstem shape (BrS) differences were detected between the study groups.

Our ROC analyses and the derived AUC values confirmed the discriminatory potential of nearly all analysed neuroimaging metrics. At baseline, medullary volume showed the highest discriminatory power for distinguishing patients from controls (AUC = 0.611, p = 0.006, Fig. 3A), followed by CT (LH: AUC = 0.619, RH: AUC = 0.567) and RD (LH: AUC = 0.606, RH: AUC = 0.606), all exhibiting high AUCs. Notably, the AUCs of all analysed metrics increased over time (t(8) = 5.431, p < 0.001; Fig. 3B).

Following patient stratification by C9orf72 status (Fig. 4), CT was identified as an important moderator of progression: bulbar cortex CT loss in C9 + patients was more rapid than in C9- or HC group [RH: t(236) =  – 2.040, p = 0.042; LH: t(236) =   – 3.505, p < 0.001; based on the Group x Time interaction effect with C9 + as a reference group]. Moreover, ANOVA confirmed significant bilateral CT differences already at baseline [RH: F(2298) = 14.11, p < 0.001; LH: F(2298) =  – 2.040, p = 0.052], whereas post-hoc Tukey HSD testing revealed that C9 + patients were the main drivers of this effect [RH: thinner CT of C9 + vs. C9–: p < 0.001; LH: thinner CT of C9 + vs. C9–: p < 0.001]. The comparison of AD also revealed a genotype-effect [RH: F(2298) = 14.11, p = 0.0391], but pairwise group differences could not be confirmed by post-hoc Tukey HSD testing. Data visualization (Fig. 4C) reveals a tendency for increased AD in both ALS groups vs. HC. Brainstem volumes and brainstem shape profiles were not modulated by genetic status.

Spinal onset C9- patients without bulbar manifestations (“bulbar asymptomatic”- BA) exhibit marked radiological changes at baseline (Fig. 5) based on four neuroimaging metrics: (1) medullary volume [F(2276) = 1.03e-6, p = 0.023]; (2) bilateral CT [RH: F(2, 276) = 0.300, p = 0.010; LH: F(2, 276) = 0.433, p < 0.001], (3) bilateral cortico-medullary RD [RH: F(2276) = 1.65e-7, p = 0.012; LH: F(2276) = 1.18e-7, p = 0.049]] and (4) bilateral cortico-medullary AD [RH: F(2276) = 1.03e-6, p = 0.003; LH: F(2276) = 8.01e-7, p = 0.012]. Post-hoc Tukey HSD tests confirmed left-hemispheric CT reduction even in bulbar asymptomatic patients compared to controls (p = 0.017). From a cortico-medullary connectivity perspective, BA patients did not differ from BS patients in right-hemispheric RD (p = 0.061) and with regards to left-hemispheric RD, BA patients exhibited higher values compared to BS (p = 0.044). No post-hoc differences were detected in medullary volumes or FC in pairwise comparisons. Longitudinally, right-hemispheric structural connectivity – both RD and AD – deteriorated in BA similarly to BS with reference to controls [RD: t(219) =  – 2.183, p = 0.030; AD: t(219) =  – 2.304, p = 0.022. Moreover, right-hemispheric FC tended to decrease in BA similarly to BS vs. controls [FC: t(219) = 1.961, p = 0.051]. These observations suggest that considerable disease burden can be ascertained in anatomical regions associated with bulbar function before bulbar disability develops, both at a cortical level and also from a cortico-medullary connectivity perspective. No presymptomatic shape alterations were detected in the BA group.

No direct correlations were identified between bulbar ALSFRS-r subscores and any of the radiological integrity metrics (CT, AD, RD, FC, MV, BrS), neither cross-sectionally nor longitudinally. Output statistics are summarized for all neuroimaging metrics in Table 2 to demonstrate the dissociation between motor disability and cerebral imaging measures.