This project made use of data collected from a prospective cohort study (ClinicalTrials.gov Identifier: NCT04006418). The study was approved by the ethics committee of the first affiliated hospital of Fujian Medical University, Fujian, China (project approval number MRCTA, ECFAH of FMU [2019]194), and written informed consent was obtained from all participants.

The study design of this HSPs cohort study was detailed in a previous protocol (25). In summary, this cohort study collected comprehensive data from HSPs patients across six key aspects, encompassing clinical signs, genetic spectrum, cognitive competence, MRI features, as well as potential biochemical indicators and nerve electrophysiological factors. The project included 30 HSPs patients, comprising 25 clinically and genetically confirmed cases (11 SPG4, 10 SPG5, 2 SPG11, 1 SPG30 and 1 SPG76) and five clinically suspicious cases (e.g., exhibiting progressive spastic gait disorder, bilateral lower limb spasm and hyperreflexia, among others). Genomic DNA was analyzed through multiple ligation-dependent probe amplification (MLPA) in conjunction with whole exome sequencing (WES). All enrolled patients fulfilled the predefined exclusion criteria, which encompassed subacute combined degeneration, vitamin E deficiency, syphilis, AIDS, tropical spastic paraplegia, spinocerebellar ataxia, Friedreich’s ataxia and any structural abnormalities or severe lesions that could potentially cause spastic paraplegia. The 47 HC included in this study with no previous history of neurological dysfunction and with normal findings on neurological examination were recruited in the same hospital. All participants with psychiatric disorders, other diseases or disorders that may affect cognitive impairment, and those who did not cooperate with the cognitive assessment were excluded.

All participants (30 HSPs and 47 HC) underwent assessment by the same neurologist in a standardized manner. The neurologist conducting the evaluations remained blinded to the clinical conditions of the participants. The assessments were performed during daytime, in a quiet room, and in a fixed order, following the recommendations of the consensus panel.

Total cognitive function was assessed using the MMSE and MoCA. Moreover, we utilized the modified Chinese version of MACFIMS scale to evaluate performance in five cognitive domains (25). Specifically, Verbal memory was gauged through the immediate and delayed recall scores of the California Verbal Learning Test-second Edition (CVLT-II). Visual memory was evaluated using the immediate and delayed recall scores of the Brief Visuospatial Memory Test-revised (BVMT-R). Visuospatial perception was examined with the Benton Judgment of Line Orientation Test (JLO). Executive functions were measured using the Controlled Oral Word Association Test (COWAT). Information processing speed was assessed using the Symbol Digit Modalities Test (SDMT) and the Paced Auditory Serial Addiction Test (PASAT) 3 and 2-sand versions.

Furthermore, the Hamilton Depression Rating Scale (HAMD) was employed to assess depression, the Hamilton Anxiety Rating Scale (HAMA) for anxiety evaluation, and the Fatigue Scale-14 for fatigue assessment. Sleep quality was evaluated using the Pittsburgh Sleep Quality Index (PSQI), and daytime sleepiness was assessed using the Epworth Sleepiness Scale (ESS). The entire battery of tests required 90–100 min of face-to-face testing time.

Patients with test scores deviating over 1.5 SDs from the HC mean were deemed to have abnormal function in the related cognitive domain. Those with abnormalities in two or more domains were categorized as having cognitive impairment (CI) (23), while others were classified as cognitively preserved (CP). Furthermore, to standardize the scores of different cognitive assessment scales across the study and to compare the severity of cognitive impairment in patients with HSPs, we calculated the ratio of various cognitive scores for each HSPs-CI patient to the average HC score decrease of more than 1.5 SDs. A ratio greater than or equal to 1 corresponds to an abnormality in this cognitive domain that is not evident in HSPs patients, and a ratio less than 1 corresponds to the abnormal functioning of this cognitive domain being significant.

Certain subjects underwent MRI examinations using a 3 T MRI Scanner (Siemens; Munich, Germany) equipped with a 20-channel head–neck coil and a 24-channel spine-array coil. The acquired sequences included: (1) Axial T2-weighted imaging (T2WI), T2-fluid-attenuated inversion recovery (FLAIR), and sagittal 3D T1-weighted imaging (T1WI) of the entire brain; (2) Sagittal 3D-T2WI and 3D-T1WI of the cervical and thoracic spinal cord (25).

To quantify the volumes of distinct regions, encompassing grey matter (GM, sum of voxels classified as GM), white matter (WM, sum of voxels classified as WM), and CSF, 3D-T1WI images underwent meticulous examination and processing (9). This analysis was conducted utilizing the Computational Anatomy Toolbox V.12 (CAT V.12, http://www.neuro.uni-jena.de/cat/), an add-on for Statistical Parametric Mapping V.12 (SPM V.12, http://www.fil.ion.ucl.ac.uk/spm), implemented within the MATLAB (R2013b, the MathWorks; Natick) environment. Additionally, the total intracranial volume (TIV) was defined as the sum of all voxels classified as GM, WM, or CSF.

Each cervical-thoracic spinal segment was acquired through the utilization of Compose software on the SIEMENS workstation. The Spinal Cord Toolbox (V.4.01) will be employed for segmenting and extracting the morphological metrics of the spinal cord at various vertebral levels, spanning from C1 to T9. These metrics will encompass CSA and sectional diameter (anterior–posterior AP, right–left RL, and the AP/RL ratio) (9).

Continuous variables are expressed using the median (IQR). Categorical data were expressed as percentages (%). The Kolmogorov–Smirnov test was employed to assess variable normality. Non-parametric continuous variables were compared using the Mann–Whitney U test (for 2 groups), and independent samples t-test were utilized for normally distributed continuous variables. Comparisons between categorical variables were conducted using the χ² test (Fisher’s exact test when expected values are <5). Pearson or Spearman rank correlation coefficients were used to assess the correlation between variables.

Furthermore, linear regression models were employed with BVMT-R DR or MoCA as the dependent variable and spinal cord morphometric indices as independent variables to assess the extent to which spinal measures explained variance in cognition. We then re-estimated these models, separately including BPF and disease duration as covariates, to assess the associations between spinal morphometric indices and BVMT-R DR or MoCA.

Data analysis was performed using SPSS Statistics (version 26; IBM, USA). Results were considered statistically significant at the level of p < 0.05.

A total of 30 HSPs patients and 47 HC were recruited in the cohort. Five clinically suspected cases were excluded owing to lack of genetic diagnosis. One patient was excluded due to an absence of neuropsychological data. Ultimately, 24 genetically confirmed HSPs patients were included in this analysis.

The age of examination of the 24 HSPs patients (10 SPG4, 10 SPG5, 2 SPG11, 1 SPG30, and 1 SPG76) ranged from 28 to 50 years (median 37 years); 8 (33.3%) were women. All patients with HSPs exhibited clinical symptoms, with median age at disease onset was 22 years (IQR 11–40 years) suffered a median disease duration of 12 years (IQR 5–15 years). The age at examination of the 47 HC was similar to that of patients with HSPs, with an age range of 26 to 45 (median 33); 24 (51.1%) were women. There were no significant differences in the age or gender of HSPs and HC.

For each HSPs group, no significant differences were found between SPG5 patients and other-HSPs patients (10 SPG4, 2 SPG11, 1 SPG30 and 1 SPG76) by age. However, the proportion of women (p = 0.002) was higher, the age of onset (p = 0.048) was earlier and the disease duration (p = 0.014) was longer in SPG5 patients compared with other-HSPs patients. The details of demographic and clinical characteristics of SPG5 and other-HSPs are presented in Table 1.

For general cognitive assessment, HSPs patients had lower education levels (p = 0.031) and MoCA scores (p = 0.012) than HC (Table 1). We next investigated the five cognitive domains, including verbal memory (CVLT-II), visual memory (BVMT-R), visuospatial perception (JLO), executive function (COWAT) and information processing speed (SDMT, PASAT 2 s and PASAT 3 s).

We observed that CVLT-II, JLO, COWAT, SDMT and PASAT 3 s scores decreased significantly in patients with HSPs (All p < 0.05), while there was no significant difference in BVMT-R between HSPs patients and HC. Specifically, for the CVLT-II scores of HSPs patients, the scores of short-term storages (I to V-test and total of five learning trials), short-term and long-term delayed free recall (SDFR and LDFR), short-term and long-term delayed cued recall (SDCR and LDCR), and total of delayed recall (CVLT-II DR) were significantly reduced (Table 2).

In general, about 45.8% of HSPs patients showed cognitive impairment. According to the five cognitive domains, 54.2% of patients had abnormal verbal memory function, 29.2% had abnormal visual memory function, 29.2% had abnormal spatial vision, 20.8% had abnormal executive function, and 33.3% had impaired information processing speed (Figure 1A). The CVLT-II DR, JLO and PASAT 3 s scores decreased significantly in patients with HSPs-CI, with JLO and PASAT 3 s showed the most pronounced decline (Figure 1B).

In addition, psychological scales assessed between patients with HSPs and HC, such as HAMA, HAMD, fatigue scale, PSQI, and ESS, did not showed significant differences (Supplementary Table 1).

For cognitive impairment mode of SPG5, there were no significant differences in the years of education, MMSE and MoCA between SPG5 patients and other-HSPs patients (Table 1). Assessment of five cognitive domains finds patients with SPG5 had lower JLO (p = 0.033), COWAT (p = 0.028) and SDMT (p = 0.023) scores compared with HC (Table 2). We did not find difference between SPG5 and HC in other cognitive assessments, including all tests in CVLT-II, BVMT-R, PASAT 2 s and 3 s. Besides, Patients with SPG5 had lower ESS (p = 0.041) scores, compared with HC (Supplementary Table 1).

Regarding the cognitive impairment pattern in individuals with other-HSPs, compared to HC, those with other-HSPs had shorter education duration (p = 0.013) and lower MoCA scores (p = 0.003) (Table 1). Assessment across five cognitive domains revealed that patients with other-HSPs exhibited lower scores in CVLT-II, JLO, COWAT, SDMT, and PASAT 3 s (All p < 0.05) compared to HC (Table 2). Additionally, psychiatric scale assessments indicated higher Fatigue scale scores (p = 0.038) in individuals with other-HSPs compared to HC (Supplementary Table 1).

This study further explored the differences in cognitive impairment patterns between patients with SPG5 and patients with other-HSPs. There was no significant difference in BVMT-R, JLO, COWAT, SDMT and PASAT 3 s or 2 s in SPG5 patients compared with other-HSPs patients. However, it is worth noting that the CVLT-II II-test (p = 0.016) score of SPG5 patients were higher compared to other-HSPs patients (Table 2). No significant differences were observed in psychiatric scale scores between SPG5 patients and other-HSPs patients (Supplementary Table 1).

To evaluate brain structural changes in patients with HSPs-CI. Quantitative brain MRI was used to evaluate seven patients with HSPs-CI (2 SPG4, 3 SPG5 and 2 SPG11) and 42 HC (9). Compared with HC, CSF volume (p = 0.006) increased and brain parenchymal fraction (BPF, i.e., [GMV + WMV]/TIV, p = 0.018) decreased in HSPs -CI patients (Figure 2A). There was no significant difference in TIV, GMV, WMV, GMV/TIV and WMV/TIV between HSPs-CI and HC (Supplementary Figure 1).

Previous research by our team found significant cervical and thoracic spinal cord atrophy in patients with SPG5 compared to HC (9). In this study, we quantified cervical and thoracic spinal cord cross-sectional morphometry in six patients with SPG5. Similarly, the total of CSA, AP and RL in the cervical and thoracic spinal cord of SPG5 patients (All p < 0.01) were significantly smaller than that of HC (Figure 3A).

Subsequently, we evaluated the relationship between CSF volume, BPF, and cognitive scores in HSPs-CI patients (N = 7). CSF volume in HSPs-CI patients was negatively correlated with PASAT 3 s (Pearson’s r = −0.827, p = 0.022). BPF was positively correlated with BVMT-R TL (Pearson’s r = 0.793, p = 0.033), BVMT-R DR (Spearman’s r = 0.917, p = 0.004) and MMSE (Spearman’s r = 0.927, p = 0.003) (Figure 2B).

In addition, we further investigated the correlation between cervical and thoracic spinal cord cross-sectional morphometrics and cognition in patients with SPG5 (N = 6). For the cervical spinal cord, CSA total, AP total, and RL total showed no significant correlations with scores on the modified Chinese version of the MACFIMS, MMSE and MoCA (Figure 3B). For the thoracic cord, AP total was positively correlated with MoCA (Pearson’s r = 0.937, p = 0.006); RL total was positively correlated with BVMT-R DR (Pearson’s r = 0.900, p = 0.014) and MoCA (Pearson’s r = 0.863, p = 0.027) (Figure 3B). Linear regression showed that thoracic spinal cord RL total explained 81% of the total variance in BVMT-R DR scores (p = 0.015), and 74% of the total variance in MoCA scores (p = 0.027). There was no significant difference between the total AP of thoracic spinal cord and MoCA in linear regression.

In SPG5 (N = 6), global brain volumetrics (TIV, GMV, WMV, CSF, BPF) were not significantly associated with BVMT-R DR or MoCA (Supplementary Table 2). With BPF included as a covariate, thoracic RL total remained positively associated with BVMT-R DR (B = 0.218, 95% CI 0.068 to 0.367, p = 0.019) and MoCA (B = 0.329, 95% CI 0.105 to 0.554, p = 0.019), and thoracic CSA total remained associated with MoCA (B = 0.059, 95% CI 0.021 to 0.096, p = 0.015) (Supplementary Table 3). Cervical AP total correlated negatively with disease duration (Pearson’s r = −0.886, p = 0.019) (Supplementary Table 4). In models additionally adjusting for duration, thoracic RL total remained positively associated with BVMT-R DR (B = 0.220, 95% CI 0.109 to 0.332, p = 0.008), while its association with MoCA did not reach significance. By contrast, thoracic CSA total remained positively associated with MoCA after adjustment (B = 0.060, 95% CI 0.020 to 0.100, p = 0.017) (Supplementary Table 3).