Relationship between brain white matter damage and grey matter atrophy in hereditary spastic paraplegia types 4 and 5

Abstract

Background and purpose

White matter (WM) damage is the main target of hereditary spastic paraplegia (HSP), but mounting evidence indicates that genotype‐specific grey matter (GM) damage is not uncommon. Our aim was to identify and compare brain GM and WM damage patterns in HSP subtypes and investigate how gene expression contributes to these patterns, and explore the relationship between GM and WM damage.

Methods

In this prospective single‐centre cohort study from 2019 to 2022, HSP patients and controls underwent magnetic resonance imaging evaluations. The alterations of GM and WM patterns were compared between groups by applying a source‐based morphometry approach. Spearman rank correlation was used to explore the associations between gene expression and GM atrophy patterns in HSP subtypes. Mediation analysis was conducted to investigate the interplay between GM and WM damage.

Results

Twenty‐one spastic paraplegia type 4 (SPG4) patients (mean age 50.7 years ± 12.0 SD, 15 men), 21 spastic paraplegia type 5 (SPG5) patients (mean age 29.1 years ± 12.8 SD, 14 men) and 42 controls (sex‐ and age‐matched) were evaluated. Compared to controls, SPG4 and SPG5 showed similar WM damage but different GM atrophy patterns. GM atrophy patterns in SPG4 and SPG5 were correlated with corresponding gene expression (ρ = 0.30, p = 0.008, ρ = 0.40, p < 0.001, respectively). Mediation analysis indicated that GM atrophy patterns were mediated by WM damage in HSP.

Conclusions

Grey matter atrophy patterns were distinct between SPG4 and SPG5 and were not only secondary to WM damage but also associated with disease‐related gene expression.

Clinical trial registration no. NCT04006418.

INTRODUCTION

Hereditary spastic paraplegias (HSPs) are a heterogeneous group of inherited neurodegenerative disorders characterized by progressive spastic gait disturbance and are generally divided into pure and complicated forms according to the Anita Harding classification [1]. The genetic basis of HSPs includes autosomal dominant, autosomal recessive, X‐linked and maternally inherited. Currently, more than 80 genetic subtypes of HSPs have been reported [2].

The molecular mechanisms associated with neurodegeneration are different in HSP subtypes. Recent research has demonstrated that, even for pure forms, the patterns of brain damage in HSPs are genotype‐specific [3, 4]. Amongst the genetic subtypes of HSPs, spastic paraplegia type 4 (SPG4) caused by mutations in the SPAST gene is the most common form of autosomal dominant HSP involving impaired axonal transport, whilst spastic paraplegia type 5 (SPG5) caused by mutations in the CYP7B1 gene is an autosomal recessive form of HSP involving abnormalities in lipid metabolism [1]. Previous neuroimaging studies have reported brain structural alteration in both SPG4 and SPG5 [3, 4, 5, 6, 7, 8, 9, 10, 11]. Although both HSP subtypes were characterized by widespread white matter (WM) involvement, inconsistent results in grey matter (GM) alterations have been reported in these studies [5, 6, 7, 8, 9, 10]. These inconsistent findings may be attributed to differences in methodology or small cohort size.

Previous studies have usually relied on an univariate technique, such as a voxel‐based morphometry (VBM) or region‐of‐interest (ROI) approach, which identify focal differences in clusters of voxels without taking into account the interrelationships between voxels [12, 13]. In contrast to VBM and ROI analysis, source‐based morphometry (SBM) is a data‐driven multivariate approach that performs independent component (IC) analysis on brain images of different magnetic resonance imaging (MRI) modalities to identify covarying patterns amongst brain voxels rather than focusing on each voxel separately. SBM allows the non‐random spatial pattern in GM or WM damage to be captured, for which the conventional VBM or ROI approach may not be sufficiently sensitive [14, 15, 16]. Recently, this approach has been successfully used to explore genetic influences on brain structure and function in other neurological disorders [17].

The prominent genetic heterogeneity of HSPs reflects a variety of pathogenic mechanisms that impede effective progress in HSP treatments. This study aims to explore the similarities and differences in pathological processes amongst HSP subtypes by using SBM analysis. This may provide valuable indicators for disease identification and genotype‐specific therapies. Therefore, this study has three main aims: (a) to identify and compare the brain GM and WM damage patterns in SPG4 and SPG5; (b) to explore the spatial associations between GM atrophy patterns and disease‐associated gene expression in HSP subtypes; (c) to investigate the causal relationship between GM atrophy and WM damage in HSP subtypes.

MATERIALS AND METHODS

This prospective study was conducted between August 2019 and March 2022, and received approval from the ethics committee of the First Affiliated Hospital of Fujian Medical University (MRCTA, ECFAH of FMU [2019]194). All participants signed an informed consent prior to enrolment.

Subjects

The HSP patients were recruited from an HSP cohort (https://clinicaltrials.gov/study/NCT04006418) at the Neurogenetic Diseases Centres in the First Affiliated Hospital of Fujian Medical University at Fuzhou, China. All patients were clinically manifested HSP with a genetically confirmed diagnosis of SPG4 or SPG5. Exclusion criteria for all participants were as follows: (a) other neurological or systemic diseases; (b) substance abusers; (c) other causes of focal or diffuse brain damage at routine MRI sessions; (d) restrictions for MRI scanning. Initially 44 patients (22 SPG4 patients and 22 SPG5 patients) were recruited. Disease severity was assessed by an experienced neurologist (Y. F., with 20 years of neurology experience) using the Spastic Paraplegia Rating Scale (SPRS) [18]. In addition, healthy controls (HCs) respectively matched with the SPG4 and SPG5 patients by age and sex (i.e., HC4 and HC5) were recruited in the present study.

Seventeen SPG5 patients have been previously reported [10]. The prior report evaluated brain and spinal cord microstructural alterations in SPG5 patients. The current study expands on this by including SPG4 patients, using a different approach to identify brain damage and investigating the relationship between gene expression and brain structural damage in HSP subtypes.

Magnetic resonance imaging data acquisition

All participants underwent an examination on a 3 T MR scanner (Magnetom Skyra; Siemens) equipped with a 20‐channel head–neck coil. Sagittal anatomical images were acquired using a three‐dimensional T1‐weighted magnetization‐prepared rapid gradient‐echo (MP‐RAGE) sequence. Diffusion tensor imaging (DTI) data were obtained with a single‐shot, echo‐planar imaging sequence, applying along 30 noncollinear directions (b = 1000 s/mm²) with an acquisition without diffusion weighting (b = 0 s/mm²). In addition, an axial T2‐weighted image (T2WI) sequence and a T2‐fluid attenuated inversion recovery (T2‐FLAIR) sequence were also performed to detect brain abnormalities in participants. The total scanning time was about 15 min 52 s. The parameters are listed in the Supplementary Material, Table S1.

Magnetic resonance imaging data analysis

Image processing

The T1‐weighted images were preprocessed using the CAT12 toolbox of Statistical Parametric Mapping (SPM12, http://www.fil.ion.ucl.ac.uk/spm/software/spm12). Each image was segmented into GM and WM, normalized using the Diffeomorphic Anatomic Registration Through Exponentiated Lie Algebra algorithm (DARTEL) and then modulated with the Jacobian determinant of the deformation. The modulated GM maps were finally smoothed with a 6‐mm full‐width at half‐maximum (FWHM) Gaussian kernel.

Diffusion tensor imaging preprocessing was performed using the FMRIB Software Library (FSL5.0, https://fsl.fmrib.ox.ac.uk/fsl). The preprocessing steps included skull removal, correcting for eddy current distortion and head motion. The fractional anisotropy (FA) images were calculated using FDT (FMRIB Diffusion Toolbox). Finally, all subjects' FA images were registered onto a standard‐space image (FMRIB58_FA) and then smoothed with a 6‐mm FWHM Gaussian kernel.

Source‐based morphometry of GM and FA images

Source‐based morphometry analysis was conducted on the preprocessed GM and FA images by using the GIFT toolbox (https://trendscenter.org/software) to extract GM or WM covariant patterns amongst all participants. The number of ICs was automatically estimated through a neural network algorithm (Infomax). To ensure their reliability, this process was repeated 20 times using the ICASSO algorithm (http://research.ics.aalto.fi/ica/icasso/). Finally, GM or FA images were decomposed into mixing matrix and source matrix. The mixing matrix contained the weights (loading coefficients) of each component for each subject. The source matrix reflected the relationship between the ICs and the voxels. For visualization, the source matrix was reshaped back into a three‐dimensional brain image and scaled to unit standard deviations (Z maps) [13].

Brain gene expression pattern of HSP subtypes

Gene expression microarray data were obtained from the Allen Human Brain Atlas (https://human.brain‐map.org). The Abagen toolbox was used to obtain cortical gene expression of HSP subtypes based on the automated anatomical labelling (AAL) atlas using the following steps [19]: (a) probe information was re‐annotated to discard the probes that could not be matched to genes; (b) to study disease‐specific genes, intensity‐based filtering of microarray probes was skipped by setting the threshold to 0, and the representative probe per gene was chosen based on differential stability; (c) sample‐to‐region matching tolerance was set to 2.0 mm, and the aggregate metric was set to the median; (d) samples were mirrored bilaterally across the hemisphere to increase spatial coverage; (e) expression values were normalized across genes and samples with a scaled robust sigmoid function. Finally, a region × gene expression matrix was obtained and the regional expressions of the disease‐specific genes (SPAST and CYP7B1) were extracted to build brain gene expression maps. In addition, the selected probes were also rechecked to ensure 100% sequence homology with the corresponding gene by using the nucleotide BLAST pipeline (http://blast.ncbi.nlm.nih.gov/Blast.cgi).

Statistical analysis

Statistical analyses were performed using the R software program (R 4.2.0, R Foundation for Statistical Computing, Vienna, Austria). Demographic and clinical findings were compared between groups by using the analysis of variance (ANOVA), Pearson χ ² test, Mann–Whitney U test and the independent samples t test based on the specific characteristics of the data. After using ANOVA analysis, post hoc Tukey's test was used to assess differences between the groups in the data.

Comparisons of the FA and GM loading coefficients between SPG4, SPG5, HC4 and HC5 were performed by using an analysis of covariance (ANCOVA) with age, sex and total intracranial volume (TIV) as nuisance covariates. Results were adjusted with false discovery rate (FDR) correction at p < 0.05 for multiple comparisons. Post hoc Tukey's test was used to assess differences between the groups.

Spearman rank correlation was used to investigate the spatial relationship between gene expression maps of HSP subtypes and the GM component Z‐maps based on the AAL regions. Considering the strong expression differences between subcortical and cortical tissue [20], the present study only includes the AAL cortex regions.

According to the SBM result, a parallel mediation analysis was conducted to examine whether GM atrophy was secondary to WM damage in HSP subtypes by using the PROCESS (model 4) for SPSS 22.0 [21]. Gene states (SPG4 vs. HC4 and SPG5 vs. HC5) were included as independent variable (X), significant GM components as dependent variables (Y), significant WM components as mediators (M), and age, sex and TIV as controlling variables. A 95% bootstrap confidence interval (CI) based on 10,000 bootstrap samples was used to estimate the level of significance.

Partial  Spearman's rank correlation analysis was performed to explore the relationship between clinical measures (SPRS scores and disease duration) and the loading coefficients of GM and WM components in SPG4 patients, SPG5 patients and HSP patients (combining SPG4 patients and SPG5 patients), controlling for age, sex and TIV.

RESULTS

Demographic and clinical characteristics

One SPG4 patient was excluded due to visually detectable movement‐related artifacts in the DTI data, and one SPG5 patient was excluded due to structural abnormalities (arachnoid cyst) in T2WI/T2‐FLAIR. No apparent signal abnormalities were observed in T2WI/T2‐FLAIR sequences in the other subjects. Finally, this study included 21 SPG4 patients (mean age 50.7 years ± 12.0 SD; 15 men), 21 SPG5 patients (mean age 29.1 years ± 13.0 SD; 14 men), 21 controls matched with SPG4 (HC4, mean age 51.1 years ± 11.3 SD; 15 men) and 21 controls matched with SPG5 (HC5, mean age 30.6 years ± 12.1 SD; 14 men). One SPG4 patient was preclinical stage. According to the Harding criteria, based on the clinical phenotype, four SPG5 patients were complicated HSP. The demographic and clinical information for HSP patients and HCs is shown in Table 1. The clinical and genetic features of each HSP patient are shown in the Supplementary Material, Table S2.

TABLE 1.

Demographic and clinical findings in HCs and HSP patients.

	HCs (n = 42)		HSP (n = 42)		Group differences
	HC4 (n = 21)	HC5 (n = 21)	SPG4 (n = 21)	SPG5 (n = 21)
Age (years)*, a	51.1 ± 11.3	30.6 ± 12.1	50.7 ± 12.0	29.1 ± 12.8	HC4 = SPG4 HC5 = SPG5 HC4 > HC5 (p < 0.001) SPG4 > SPG5 (p < 0.001)
Sex (male/female) b	15/6	14/7	15/6	14/7	HC4 = SPG4 HC5 = SPG5 HC4 = HC5 SPG4 = SPG5
Disease duration (years)*, c	–	–	15.9 ± 10.5	15.4 ± 9.8	SPG4 = SPG5
SPRS score*, d	–	–	14.7 ± 10.5	15.2 ± 8.5	SPG4 = SPG5

Abbreviations: ANOVA, analysis of variance; HC, healthy control; HC4, healthy control matched with SPG4 by sex and age; HC5, healthy control matched with SPG5 by sex and age; HSP, hereditary spastic paraplegia; SPG4, spastic paraplegia type 4; SPG5, spastic paraplegia type 5; SPRS, Spastic Paraplegia Rating Scale.

^*Numbers are means ± standard deviations.

^a ANOVA analysis and post hoc Tukey's test (p values shown for when p < 0.05).

^b Pearson χ ² test.

^c Mann–Whitney U test. One SPG4 patient was excluded due to the preclinical stage.

^d Independent samples t test.

Source‐based morphometry of GM images

The GM images were decomposed into six ICs. The ANCOVA revealed significant group differences in two of them (GM component 2 [GM‐IC2] F = 3.50, p = 0.018 and P _FDR = 0.04; GM component 6 [GM‐IC6] F = 3.89, p = 0.013 and P _FDR = 0.04). For GM‐IC2, the post hoc test indicated that the SPG5 group had lower loading coefficients compared with HC5 (P _Tukey = 0.02), whilst no significant difference was found between SPG4 and HC4 (P _Tukey > 0.99), SPG4 and SPG5 (P _Tukey = 0.11), HC4 and HC5 (P _Tukey >0.99). For GM‐IC6, the post hoc test showed that SPG4 had lower loading weights compared with HC4 (P _Tukey = 0.01), whilst no significant differences were found between SPG5 and HC5 (P _Tukey = 0.71), SPG4 and SPG5 (P _Tukey = 0.78), HC4 and HC5 (P _Tukey = 0.85). GM‐IC2 included the bilateral anterior/median cingulate and paracingulate gyrus (ACG/DCG), middle/inferior/superior frontal gyrus (MFG/IFG/SFG), precentral gyrus (PreCG), rectus gyrus and supplementary motor area (SMA). GM‐IC6 included the bilateral inferior parietal lobule (IPL), precuneus, supramarginal gyrus (SMG), superior/middle temporal gyrus (STG/MTG), thalamus, middle/superior occipital gyrus (MOG/SOG), postcentral gyrus (PoCG), posterior cingulate gyrus (PCG), DCG and angular gyrus (Figure 1, Table 2).

FIGURE 1.

Group differences in GM components amongst SPG4, SPG5, HC4 and HC5. (a) GM component map of ICs. The voxels above the threshold of |Z| > 2.5 are shown. (b) The post hoc Tukey's test for GM‐IC2. (c) The post hoc Tukey's test for GM‐IC6. GM, grey matter; HC4, healthy control matched with SPG4 by sex and age; HC5, healthy control matched with SPG5 by sex and age; IC, independent component; SPG4, spastic paraplegia type 4; SPG5, spastic paraplegia type 5.

TABLE 2.

Brain regions for the GM components with significant differences between groups.

Brain regions according to AAL atlas	L/R volume (mL)	L/R max Z‐value	L/R MNI (x, y, z)
GM‐IC2
Anterior cingulate and paracingulate gyrus	16.75/15.50	5.87/6.83	0, 37.5, −6/10.5, 43.5, 12
Middle frontal gyrus	14.30/28.14	5.68/7.04	−39, 9, 34.5/36, 33, 31.5
Median cingulate and paracingulate gyrus	9.95/10.62	4.88/5.71	0, 12, 33/6, 24, 33
Inferior frontal gyrus; triangular part	9.66/4.16	5.06/6.78	−37.5, 42, 10.5/34.5, 33, 30
Inferior frontal gyrus; opercular part	4.66/4.24	5.08/5.92	−37.5, 10.5, 33/40.5, 15, 37.5
Precentral gyrus	4.54/3.45	6.03/5.02	−39, 6, 34.5/40.5, 6, 34.5
Superior frontal gyrus; medial orbital	3.88/5.76	6.09/6.04	0, 28.5, −13.5/1.5, 34.5, −9
Superior frontal gyrus; medial	3.79/0.96	5.00/5.71	−7.5, 45, 19.5/10.5, 46.5, 7.5
Superior frontal gyrus; dorsolateral	2.47/2.40	4.55/4.88	−22.5, 40.5, 34.5/24, 54, −1.5
Rectus gyrus	2.30/2.69	6.06/5.92	0, 27, −15/0, 24, −15
Superior frontal gyrus; orbital part	2.16/2.68	4.76/5.44	−25.5, 57, −4.5/25.5, 55.5, −1.5
Supplementary motor area	1.12/3.00	3.78/4.19	−6, 22.5, 45/7.5, −3, 46.5
GM‐IC6
Inferior parietal lobule	19.98/5.86	7.27/5.73	−30, −51, 43.5/31.5, −36, 51
Precuneus	18.74/17.19	5.60/7.20	0, −60, 40.5/6, −57, 54
Supramarginal gyrus	9.78/7.70	5.88/6.66	−57, −37.5, 33/40.5, −30, 42
Superior temporal gyrus	7.61/1.32	4.64/3.84	−52.5, −39, 24/45, −57, 22.5
Thalamus	4.64/5.29	4.45/4.32	−12, −21, 10.5/15, −19.5, 10.5
Middle occipital gyrus	3.70/4.74	5.36/6.19	−39, −63, 19.5/30, −69, 33
Middle temporal gyrus	3.60/7.14	6.07/6.53	−42, −64.5, 19.5/48, −54, 12
Postcentral gyrus	3.02/11.71	5.61/6.90	−54, −19.5, 33/42, −28.5, 42
Posterior cingulate gyrus	2.37/0.70	4.94/4.41	−7.5, −52.5, 31.5/1.5, −51, 31.5
Superior occipital gyrus	2.34/2.22	4.30/4.74	−16.5, −88.5, 25.5/30, −67.5, 39
Median cingulate and paracingulate gyrus	1.96/2.71	4.65/4.73	0, −43.5, 34.5/7.5, −42, 39
Angular gyrus	0.98/2.02	4.36/4.87	−54, −51, 36/31.5, −45, 40.5

Note: The MNI coordinates indicate the location of the peak vertex. Within each IC, brain regions were ordered by the volume of the left.

Abbreviations: AAL, automated anatomical labelling; GM, grey matter; IC, independent component; L, left; MNI, Montreal Neurological Institute; R, right.

Source‐based morphometry of FA images

The FA images were decomposed into nine ICs. The ANCOVA revealed significant group differences in two of them (WM component 1 [WM‐IC1] F = 49.50, p < 0.001 and P _FDR <0.001; WM component 2 [WM‐IC2] F = 29.73, p < 0.001 and P _FDR <0.001). For these two components, the post hoc test indicated that both SPG4 and SPG5 showed lower loading coefficients compared with HCs, whilst SPG5 had lower loading coefficients compared with SPG4 (WM‐IC1, SPG4 vs. HC4, P _Tukey < 0.001; SPG5 vs. HC5, P _Tukey < 0.001; SPG4 vs. SPG5, P _Tukey < 0.001; HC4 vs. HC5, P _Tukey = 0.12) (WM‐IC2, SPG4 vs. HC4, P _Tukey < 0.001; SPG5 vs. HC5, P _Tukey < 0.001; SPG4 vs. SPG5, P _Tukey = 0.01; HC4 vs. HC5, P _Tukey = 0.32). WM‐IC1 included the superior longitudinal fasciculus (SLF), posterior thalamic radiation (PTR), superior/posterior corona radiata (SCR/PCR), sagittal stratum and the retrolenticular part of the internal capsule. WM‐IC2 included the splenium of the corpus callosum, posterior/anterior limb of the internal capsule, middle cerebellar peduncle (MCP), genu of the corpus callosum, cerebral peduncle (CP), corticospinal tract (CST) and medial lemniscus (Figure 2, Table 3).

FIGURE 2.

Group differences in WM components amongst SPG4, SPG5, HC4 and HC5. (a) WM component map of ICs. The voxels above the threshold of |Z| > 2.5 are shown. (b) The post hoc Tukey's test for WM‐IC1. (c) The post hoc Tukey's test for WM‐IC2. HC4, healthy control matched with SPG4 by sex and age; HC5, healthy control matched with SPG5 by sex and age; IC, independent component; SPG4, spastic paraplegia type 4; SPG5, spastic paraplegia type 5; WM, white matter.

TABLE 3.

Brain regions for the WM components with significant differences between groups.

Brain regions according to JHU white matter labels	L/R volume (mL)	L/R max Z‐value	L/R MNI (x, y, z)
WM IC1
Superior longitudinal fasciculus	2.65/2.33	5.75/4.53	−28, −22, 38/28, −18, 40
Posterior thalamic radiation	2.57/2.71	6.07/5.18	−34, −44, 14/40, −44, −2
Superior corona radiata	1.97/2.88	5.45/5.21	−26, −20, 36/26, −18, 38
Posterior corona radiata	1.05/1.66	4.71/4.24	−32, −46, 20/20, −40, 38
Sagittal stratum	1.02/0.79	5.19/5.23	−40, −42, −4/42, −40, −6
Retrolenticular part of internal capsule	0.42/0.57	4.70/4.52	−40, −38, 0/36, −38, 12
WM IC2
Splenium of corpus callosum	7.06	6.25	−4, −38, 14
Posterior limb of internal capsule	2.59/2.42	5.09/5.03	−20, −16, −4/16, −8, −2
Middle cerebellar peduncle	2.54	5.68	18, −36, −34
Genu of corpus callosum	2.02	5.16	2, 30, 4
Cerebral peduncle	1.41/1.30	5.06/4.91	−18, −16, −6/14, −8, −6
Anterior limb of internal capsule	0.98/1.30	4.97/5.27	−14, 2, 8/18, 0, 10
Corticospinal tract	0.58/0.73	4.37/4.75	−8, −28, −24/8, −28, −24
Medial lemniscus	0.34/0.39	5.02/5.30	−4, −34, −26/6, −34, −26

Note: The MNI coordinates indicate the location of the peak vertex. Within each IC, brain regions were ordered by the volume of the left (if it exists).

Abbreviations: IC, independent component; JHU, Johns Hopkins University; L, left; MNI, Montreal Neurological Institute; R, right; WM, white matter.

The association between cortical gene expression profiles and GM atrophy patterns

Spearman rank correlation analysis revealed significant correlation between the SPAST gene expression profile and the GM‐IC6 pattern (Spearman's ρ = 0.30, p = 0.008), as well as between the CYP7B1 gene expression profile and the GM‐IC2 pattern (Spearman's ρ = 0.40, p < 0.001), indicating that cortical regions with high expression of disease genes have a good spatial overlap with the patterns of GM cortical atrophy in corresponding HSP subtypes (Figure 3).

FIGURE 3.

The association between cortical gene expression profiles and GM components based on the AAL atlas. (a) The SPAST gene expression map (left) and the GM‐IC6 Z‐map (right); (b) correlation between SPAST gene expression and GM‐IC6 Z‐map based on AAL cortical regions; (c) the CYP7B1 gene expression map (left) and the GM‐IC2 Z‐map (right); (d) correlation between the CYP7B1 gene expression and GM‐IC2 Z‐map based on AAL cortical regions. AAL, automated anatomical labelling; GM, grey matter; IC, independent components.

The causal relationship between the GM atrophy patterns and WM microstructural damage

In the mediation analysis, compared to HC4, SPG4 had a significant indirect effect on GM‐IC6 via WM‐IC1 (indirect effect a ₁*b ₁, beta (SE) = −0.29 (0.14), 95% CI −0.55, −0.23; direct effect c', beta (SE) = −0.26 (0.21), 95% CI −0.67, 0.15, p = 0.21; total effect c, beta (SE) = −0.62 (0.14), 95% CI −0.90, −0.34, p < 0.001). In the path analysis, the WM‐IC1 was significantly negatively associated with SPG4 (a ₁ = −0.99, p < 0.001) and positively associated with GM‐IC6 (b ₁ = 0.29, p = 0.02) (Figure 4b). Similarly, compared to HC5, SPG5 also had a significant indirect effect on GM‐IC2 via WM‐IC1 (a ₁*b ₁, beta (SE) = −0.57 (0.24), 95% CI −1.05, −0.10; c', beta (SE) = −0.05 (0.34), 95% CI −0.71, 0.61, p = 0.88; c, beta (SE) = −0.44 (0.17), 95% CI −0.75, −0.13, p = 0.005). In the path analysis, the WM‐IC1 was significantly negatively associated with SPG5 gene mutation (a ₁ = −1.78, p < 0.001) and positively associated with GM‐IC6 (b ₁ = 0.32, p = 0.01) (Figure 4c).

FIGURE 4.

The causal relationship between the GM atrophy patterns and WM microstructural damage: the loading coefficients of GM‐IC2 (a) or GM‐IC6 (b) as the dependent variable (Y), the SPG4 as the independent variable (X) and the loading coefficients of WM‐IC1and WM‐IC2 as the mediator (M); the loading coefficients of GM‐IC2 (c) or GM‐IC6 (d) as the dependent variable (Y), the SPG5 as the independent variable (X) and the loading coefficients of WM‐IC1 and WM‐IC2 as the mediator (M). Note that sex, age and TIV were used as covariates for all changes in GM‐ICs and changes in WM‐ICs. a ₁ and a ₂, effect of HSP on WM‐IC1 and WM‐IC2 compared to HC; b ₁ and b ₂, effect of WM‐IC1 and WM‐IC2 on change in GM‐ICs; c, total effect on GM‐ICs change for HSP compared to HC; c', direct effect on GM‐ICs change for HSP compared to HC; a ₁*b ₁ or a ₂*b ₂, indirect effect on GM‐ICs change via WM‐IC1 or WM‐IC2 for HSP compared to HC. CI, confidence interval; GM, grey matter; HC4, healthy control matched with SPG4 by sex and age; HC5, healthy control matched with SPG5 by sex and age; IC, independent component; SPG4, spastic paraplegia type 4; SPG5, spastic paraplegia type 5; TIV, total intracranial volume; WM, white matter.

Association between clinical measures and GM atrophy and WM microstructural damage

The results of the relationship between clinical measures and GM atrophy and WM microstructural damage in HSP patients are presented in Table 4. In the HSP group, the loading coefficients of WM‐IC1 and WM‐IC2 are negatively correlated with SPRS scores (ρ = −0.52, p < 0.001, ρ = −0.49, p = 0.002, respectively) or disease duration (ρ = −0.51, p = 0.001, ρ = −0.46, p = 0.003, respectively), and the loading coefficients of GM‐IC6 are negatively correlated with disease duration (ρ = −0.38, p = 0.02). In the SPG4 group, the loading coefficients of WM‐IC2 are negatively correlated with disease duration (ρ = −0.51, p = 0.03), and the loading coefficients of GM‐IC6 are negatively correlated with SPRS scores or disease duration (ρ = −0.47, p = 0.047, ρ = −0.57, p = 0.02, respectively). In the SPG5 group, the loading coefficients of WM‐IC1 are negatively correlated with disease duration (ρ = −0.55, p = 0.02).

TABLE 4.

Correlation between GM‐ICs/WM‐ICs and SPRS scores or disease duration in HSP patients, SPG4 patients and SPG5 patients.

	HSP		SPG4		SPG5
	SPRS	Disease duration*	SPRS	Disease duration*	SPRS	Disease duration
GM‐IC2	ρ = 0.02 p = 0.88	ρ = − 0.01 p = 0.94	ρ = 0.24 p = 0.34	ρ = 0.22 p = 0.39	ρ = 0.03 p = 0.90	ρ = 0.03 p = 0.90
GM‐IC6	ρ = − 0.25 p = 0.12	ρ = − 0.38 p = 0.02	ρ = − 0.47 p = 0.04	ρ = − 0.57 p = 0.02	ρ = − 0.43 p = 0.08	ρ = − 0.20 p = 0.43
WM‐IC1	ρ = − 0.52 p < 0.001	ρ = − 0.51 p = 0.001	ρ = − 0.29 p = 0.25	ρ = − 0.29 p = 0.26	ρ = − 0.46 p = 0.05	ρ = − 0.55 p = 0.02
WM‐IC2	ρ = − 0.49 p = 0.002	ρ = − 0.46 p = 0.003	ρ = − 0.51 p = 0.03	ρ = − 0.48 p = 0.05	ρ = − 0.19 p = 0.45	ρ = − 0.32 p = 0.19

Abbreviations: GM, grey matter; HSP, hereditary spastic paraplegia; IC, independent component; SPG4, spastic paraplegia type 4; SPG5, spastic paraplegia type 5; SPRS, Spastic Paraplegia Rating Scale; WM, white matter.

^*One SPG4 patient was excluded due to the preclinical stage.

DISCUSSION

In this work, a novel data‐driven multivariate approach was applied to investigate brain damage pattern in HSP subtypes. It was found that patients with SPG4 and SPG5 shared a similar WM damage pattern relative to HCs, but demonstrated obviously different patterns of GM atrophy. Moreover, GM atrophy patterns in HSP patients were related to WM damage and associated with brain expression of disease‐related genes. Our findings indicated that brain structural damage patterns in HSP were genotype‐specific.

It was found that SPG4 was characterized by lower loading coefficients of GM‐IC6 comprising bilateral IPL, PoCG, PCG, DCG, MTG and thalamus compared to HCs. This GM atrophy pattern coincides well with previous studies based on univariate imaging methods, in which GM volume reductions in these brain regions have been reported in SPG4 patients [3, 5, 6, 7, 8]. At network level, these cortical regions mainly belong to the dorsal attention network [22]. The dorsal attention network is densely connected with the subcortical structure and can modulate the top‐down sensory (auditory, visual and somatosensory) area [23]. The thalamus together with these cortical regions, participating in cortico‐basal ganglia‐thalamic circuits, play a hub role in emotional, cognitive and motor cortical functions [24, 25]. Unlike SPG4, SPG5 exhibited lower loading coefficients in GM‐IC2 involving the prefrontal cortex. Interestingly, the GM atrophy in SPG5 has not been described in previous studies [10], showing that SBM allows subtle GM atrophy to be detected more sensitively than a conventional univariate approach. Amongst these regions, PreCG and SMA were somatomotor cortex, whilst other regions were mainly overlapping with the frontoparietal network which is considered to be important for top–down information integration from motor networks to control motor output [26] and has been related to motor symptoms in neurodegenerative diseases [27]. In addition, non‐motor region involvement (thalamus, IPL and prefrontal cortex) in these components has also been linked with extra‐motor clinical manifestations such as cognitive impairment in HSP patients in previous studies [5, 7, 28].

Using the microarray expression data from the Allen Human Brain Atlas, it was found that the spatial maps of SPAST and CYP7B1 gene expression in the healthy brain were respectively correlated with the cortical GM atrophy pattern in SPG4 and SPG5. These findings are consistent with a recent study in which the close relationship between cerebral gene expression and cerebral structural damage was first demonstrated in different subtypes of HSP [3]. It is speculated that cortical regions with higher gene expression are more susceptible to cause neurodegeneration due to loss or reduction of the corresponding protein function induced by disease‐related gene mutation. Therefore, our findings establish a link between regional disease‐related gene expressions and selective vulnerability of cerebral structures to neurodegeneration in HSP patients. Generally, these genotype‐specific alteration patterns provide valuable information to better understand the pathophysiological mechanisms underlying cortex damage in HSP subtypes.

The SBM approach may more effectively detect WM damage by grouping WM voxels that have similar covariation [14, 29]. In the present study, the SBM of FA data identified two components that showed decreased loading coefficients in both SPG4 and SPG5. WM‐IC2 mainly involved projection fibres below the thalamus such as CST, CP and MCP, whilst WM‐IC1 was composed of projection fibres (PTR, SCR, PCR) and association fibres (SLF) above the thalamus. These findings are consistent with previous DTI studies that have revealed widespread WM damage in HSP patients [3, 4, 6, 7, 8, 9, 10, 11]. The damage in projection fibres, including CST, supports the pathological characteristics of a dysfunction of the long spinal cord axons that is the predominant cause leading to lower extremity spastic weakness in HSP patients. The impairment of extra‐motor WM tracts may be associated with non‐motor symptoms of HSP [30, 31]. In addition, it was also found that SPG5 had lower loading coefficients than SPG4 in WM components. This phenomenon might be related to their different pathological processes. WM impairment in SPG4 is usually characterized by axonal damage due to axonal transport deficits [8], whilst in SPG5 it is characterized by myelin breakdown loss due to the accumulation of 27‐hydroxycholesterol [10]. Thus, it is speculated that myelin destruction might be more likely to result in WM spatial alteration than axonal damage.

The specific causal relationship between GM atrophy and WM damage in HSP patients was further explored. Using mediation analysis, our results confirmed that GM atrophy in HSP patients was influenced by association and projection fibres above the thalamus (WM‐IC1). These findings are not surprising, since the cerebral cortex may ultimately be affected by WM damage due to the ‘dying back’ neuropathy process in HSP. Moreover, these fibres are adjacent to GM components [23, 32].

In HSP patients, GM/WM damage patterns were significantly negatively associated with SPRS scores or disease duration. However, when the SPG4 and SPG5 groups were analysed separately, the correlations between GM/WM damage patterns and SPRS scores or disease duration were inconsistent, which may be limited by the size of the sample. Thus, the correlations between GM/WM damage patterns and SPRS scores or disease duration needs to be further investigated.

Several limitations should be acknowledged in this study. First, the rarity of the disease made it difficult to recruit patients, so the sample size was small. Secondly, the gene expression and MRI data were not obtained from the same individuals [33, 34]. But the correlation analyses between cortical atrophy and gene expression were based on the assumption that brain gene expression across individuals is highly conserved [35, 36]. The current results regarding the relationship between gene expression and MRI data needed more direct biological evidence to confirm in future studies.

In summary, distinct GM atrophy patterns between SPG4 and SPG5 were identified by using multivariate SBM analysis. Our results indicate that GM atrophy in HSP subtypes are not only secondary to WM damage but also associated with disease‐related gene expression in cortical regions. These findings may offer new insights into understanding the pathophysiological mechanisms underlying brain structural abnormality in different HSP subtypes and provide valuable indicators for genotype‐specific therapies.

AUTHOR CONTRIBUTIONS

Jianping Hu: Conceptualization; investigation; resources; writing – review and editing; software. Yuqing Tu: Conceptualization; formal analysis; methodology; visualization; writing – original draft. Ying Liu: Conceptualization; data curation; funding acquisition; project administration; supervision. Shuping Fan: Formal analysis; visualization; writing – review and editing. Jiaqi Weng: Visualization; writing – review and editing. Mengcheng Li: Writing – review and editing. Fan Zhang: Data curation. Ying Fu: Conceptualization; formal analysis; methodology; validation; writing – review and editing.

FUNDING INFORMATION

This work was supported by the Joint Funds for the Innovation of Science and Technology, Fujian Province (No. 2021Y9097) and Natural Science Foundation of Fujian Province (No. 2022J05142).

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no conflict of interest.

Supporting information

Table S1.

Tu Y, Liu Y, Fan S, et al. Relationship between brain white matter damage and grey matter atrophy in hereditary spastic paraplegia types 4 and 5. Eur J Neurol. 2024;31:e16310. doi: 10.1111/ene.16310

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.