Causal neurodevelopmental origins of scoliosis: Multi-omics Mendelian randomization reveals brain–region–specific pathways and protein mediators

Abstract

Background

The causal relationship between early-life central nervous system (CNS) infections and adolescent idiopathic scoliosis (AIS) remains unresolved, with limited evidence on mediating mechanisms linking neuroanatomical changes to spinal deformity.

Method

Through three longitudinal cohorts (n = 57875) and two-step Mendelian randomization (MR) integrating neuroimaging genomics (UK Biobank) and proteomics, we dissected causal pathways from childhood viral encephalitis (VE) to scoliosis. We employed inverse-variance weighted (IVW), MR-Egger regression, and sensitivity analyses (MR-PRESSO) to control for pleiotropy and reverse causation.

Results

Childhood VE was associated with a 3.6-fold increased risk of scoliosis (HR = 3.604, 95 % CI: 3.121–4.163, P = 0.001), with lesions in the corpus callosum and cerebellum showing the strongest effects. MR analysis identified seven imaging-derived phenotypes (IDPs) causally linked to scoliosis, including grey matter volume in the left thalamus (OR = 1.451) and isotropic free water fraction in the cerebellar peduncle (OR = 2.408). Mediation MR revealed that brain protein ERBB4 and cerebrospinal fluid protein LBP mediated 34.9 % of the total effect (β = −0.181), highlighting their role in bridging neuroinflammation to spinal deformity.

Conclusion

This study offers suggestive evidence for a causal pathway from childhood CNS infections to scoliosis, mediated by specific brain-region damage and protein biomarkers.

The translational potential of this article

Clinically, the results support the implementation of scoliosis screening programs in children recovering from CNS infections.

Graphical abstract

1. Introduction

Adolescent idiopathic scoliosis (AIS), a three-dimensional spinal deformity affecting 0.5 %–5 % of adolescents globally, poses significant challenges to public health due to its unclear etiology and limited preventive strategies [1,2]. While genetic predisposition, biomechanical factors, and hormonal imbalances have been implicated in AIS pathogenesis, the role of early-life neurological insults remains poorly understood [3,4].

Recent studies suggest that neuroanatomical abnormalities, such as altered white matter integrity in the corpus callosum and cerebellar dysfunction, are prevalent in AIS patients [5,6]. However, whether these changes are primary drivers or secondary consequences of spinal deformity remains controversial [7,8]. Viral encephalitis (VE), a common childhood central nervous system (CNS) infection, provides a unique model for investigating this question due to its association with localized brain lesions and chronic neurodevelopmental disorders [[9], [10], [11]].

Despite the potential link between neuroanatomical abnormalities and AIS, no large-scale studies have systematically examined the causal relationship or identified mediating mechanisms. Observational studies are prone to confounding and reverse causation, while randomized controlled trials are ethically and logistically unfeasible. Mendelian randomization (MR), a genetic instrumental variable approach, offers a robust framework to infer causality by leveraging genetic variants as proxies for modifiable exposures [12].

Here, we integrate multi-cohort epidemiological data with two-step MR analysis to establish the causal effect of childhood CNS infections on AIS risk and identify specific brain regions and molecular mediators underlying this association. By combining neuroimaging genomics, proteomics, and longitudinal clinical data, we aim to unravel the neurodevelopmental origins of AIS and nominate actionable targets for early intervention.

2. Method

2.1. Hospital patients

We reviewed the medical records, radiographic assessment, and clinical evaluation of all patients aged 3–16 years diagnosed with VE in a large pediatric hospital and two general tertiary hospitals (the Second Affiliated Hospital of Wenzhou Medical University, Ningbo First Hospital and Ruian People's Hospital). All the VE patients were consecutively diagnosed between January 2013 and March 2022. Control patients without VE were matched to cases in a 4:1 ratio by sex and age (tolerance ± 0.1).

Based on International Encephalitis Consortium criteria, viral encephalitis was diagnosed clinically in children who consisted of altered mental status in addition to the presence of 2 or more of the following minor criteria: (1) fever, (2) seizure, (3) focal neurologic findings, (4) cerebrospinal fluid (CSF) white cell count of ≥5 cells/mm³, (5) abnormal brain imaging, and/or (6) electroencephalogram abnormalities [[13], [14], [15]]. Diagnosis of bacterial meningitis was ruled out through CSF microscopy and culture. Fungal infections were detected via CSF culture or detection of specific CSF antigens. Parasitic infections were identified by microscopic examination of brain tissues or CSF antigens. Autoimmune encephalitis was diagnosed based on detectable CSF antibodies without alternative causes and concurrent autoimmune disorders, including anti-NMDA receptor positivity [16]. Exclusion criteria were as follows: incomplete medical records; pediatric VE patients diagnosed with immunodeficiencies, autoimmune disease, metabolic disease, multisystem dysfunction, congenital malformations, hereditary diseases, spinal diseases, musculoskeletal systemic diseases, nervous system diseases, or history of other neurological diseases (including cerebrovascular disease or brain tumors) [17]. 58 VE patients and 138 controls were excluded. Finally, this study included 6634 VE patients and 26398 non-VE children.

VE patients received intensive monitoring and supportive care, including oxygenation, airway protection, circulatory support, and management of fever, cardiac arrhythmias, autonomic instability, antiviral therapy, and immunomodulatory agents [18]. Plain chest radiographs were commonly acquired during the initial clinical workup of hospitalized viral encephalitis patients. Non-VE children who underwent the anteroposterior plain chest X-ray due to respiratory symptoms or a health check-up were enrolled as the control group.

The retrospective cohort study used cross-sectional chest radiographs as the baseline. The hospital longitudinal study used the same recruitment approach and inclusion and exclusion criteria as the cross-sectional study. This longitudinal study was conducted to enroll 7420 children who had undergone a second or more chest radiograph at least 60 days after the baseline date. The incidence of scoliosis in the latest radiographic data was measured for 1452 children with recovered VE and 5608 children without VE who were free of scoliosis (baseline Cobb angle <10°). The primary endpoint was defined as scoliosis (Cobb angle ≥10°) on any radiograph after baseline. Follow-up time was defined from baseline to the earliest of the following: 1. Date of scoliosis diagnosed; 2. Date of last X-ray assessment in those without scoliosis.

The local ethics committees approved this retrospective study and waived the requirement for signed informed consent. (LCKY2020-69-01)

2.2. Zhejiang province scoliosis screening cohort

The scoliosis screening program in Zhejiang Province conducted a comprehensive screening of scoliosis among junior high school and primary school students across the province. Examiners underwent standardized training for scoliosis screening and used identical measurement scoliometer to ensure consistency in the assessments. Some patients in the hospital cohort, having recovered from encephalitis, did not return for follow-up care, resulting in loss to follow-up. The majority of patients in the multi-center hospital cohort are enrolled in schools within Zhejiang province. Therefore, the screening results for this cohort from Zhejiang province will serve as validation for the hospital-based cohort. We linked hospital medical records with database information. Children with a history of recovered encephalitis were identified using hospital diagnostic records. Utilizing the results from this screening, we established a retrospective cohort to assess the prevalence of scoliosis. The incidence of scoliosis was evaluated in 3524 children with recovered VE and 16341 children without VE. All participants were confirmed to be free of scoliosis at baseline. Angle of trunk rotation (ATR) is described as axial rotation measured by scoliometer during the Adams forward-bend test. The primary endpoint was defined as suspected scoliosis, indicated by an ATR ≥5°during government screening (Fig. 1).

Fig. 1.

Study flowchart.

2.3. UK biobank cohort

Due to the limited number of viral encephalitis cases recorded in the UK Biobank, patients with central nervous system (CNS) inflammation before the age of 18 were included in the CNS inflammation group. To maintain consistency with the hospital cohort, the control group consisted of patients who had experienced respiratory system inflammation before the age of 18. Exclusion criteria included congenital CNS disorders, CNS injuries, congenital spinal diseases, and congenital heart disease. The disease diagnosis date was used as the starting point for follow-up, and the onset of scoliosis served as the endpoint. If scoliosis did not develop, the follow-up endpoint was defined by death, loss to follow-up, or the last contact date with the UK Biobank. Body size at age 10 was selected to represent childhood body type. Participants retrospectively reported their perceived body size at age 10 using categories such as "thinner," "average," or "plumper" compared to peers. A total of 1259 individuals with CNS inflammation were included, along with 29691 controls (Supplementary Table 1).

2.4. Radiographic assessment

Radiographs were evaluated using the anteroposterior plain chest X-ray (Siemens 500 mA camera, AGFA CR system, or Philips DR system). Scoliosis was defined as a Cobb angle of greater than 10°. The Cobb angles were measured by 3 of the authors. In addition, 400 randomly selected cases were chosen to determine the intra- and inter-observer variability of measurement and were performed 1 months following initial evaluation. Among the three authors the intra-class correlation coefficient = 0.92(95 % CI 0.89–0.93); the inter-class correlation coefficient = 0.84 (95 % CI 0.81–0.91) [19].

2.5. Clinical evaluation

Hospital recorded the following data for each subject: diagnosis, sex, chronological age, and body mass index (BMI). The Chinese children's BMI curve divided patients into “normal or thin” and “overweight or obesity” [20]. The following data were additionally collected for VE patients: symptoms and duration (e.g., fever, convulsion), days of hospitalization, brain Magnetic Resonance Imaging (MRI), and electroencephalogram (EEG) results. Fever was identified as a body temperature above 37.4 °C. Scalp electrodes were placed by the international 10–20 system, and additional temporal encryption electrodes or sphenoidal electrodes were added based on the patient's illness. 16 or 32 lead recorded EEG data for patients. GE Discovery MR750 3.0T magnetic resonance scanner was used to acquire imaging data, utilizing an 8-channel standard phased-head coil [21,22]. All participants were instructed to remain motionless during the examination. Results from MRI and EEG extracted were acquired from the electronic medical record system, and the information was analyzed by trained physicians. According to the reported EEG, the EEG results were labeled as normal, mild abnormality, or moderate-severe abnormality. MRI results were categorized as a single lesion if an area of the brain displayed abnormalities in signal, intensity, or morphology or as multiple lesions if there was more than one area in the patient.

2.6. Bidirectional MR analyses on scoliosis and neuroimaging variables

The data on scoliosis were sourced from a Genome-Wide Association Study (GWAS) conducted by Kou I on a cohort of 79,211 East Asian individuals, which included 5327 cases of AIS [23]. We utilized GWAS data on brain structure imaging derived phenotypes (IDPs) from an analysis by Stephen M involving 31,968 European individuals [24]. Furthermore, we extracted data on cerebrospinal fluid proteins and brain proteins from a GWAS conducted by Chengran Yang on a European cohort [25]. This study identified genetic data for 832 cerebrospinal fluid proteins from 835 samples and for 1237 brain proteins from 458 samples. GWASs of plasma protein from deCODE on a European cohort [26]. Verification analyses for scoliosis data were based on a GWAS conducted by Khanshour AM on 20,097 European and 13,379 East Asian individuals [27], including 2295 East Asian and 1503 European cases of AIS, along with data from the FinnGen consortium's R10 and R11(https://www.finngen.fi/fi).

MR uses genetic data to test whether there is any causal effect between an exposure, mediators and an outcome variable. To test the bidirectional causal effect between scoliosis and neuroimaging variables, genetic instruments were chosen from the GWAS summary statistics of neuroimaging at a P threshold of 5 × 10⁻⁸. For analysis concerning mediator factors, when setting the data threshold to P = 5 × 10^⁻⁸, we were unable to extract a sufficient number of single nucleotide polymorphism (SNP). Therefore, the threshold was adjusted to P = 5 × 10^⁻⁶. These SNPs were then clumped with a distance of 5000 kb and a maximum LD r² of 0.001. SNP effect data on both the exposure and outcome were then harmonised to match the effect alleles before conducting the MR analyses.

To elucidate the mediating roles of brain protein, cerebrospinal fluid protein, and plasma proteins in the relationship between brain structure and scoliosis, we employed a two-step Mendelian Randomization analysis, utilizing brain structures identified at various thresholds. In this framework, β1 captures the effect of brain structure on the levels of brain proteins, cerebrospinal fluid proteins, and plasma proteins; β2 quantifies the influence of these proteins on the incidence of scoliosis; and β3 reflects the direct effect of brain structure on scoliosis. The percentage of the mediating effect attributable to brain protein, cerebrospinal fluid protein, and plasma protein is derived by calculating the ratio of the indirect effect to the total effect (β1 ∗β2/β3).

The fixed-effect weighted median, MR Egger, inverse variance weighted (IVW) method is the primary analysis to explore the causal effect between neuroimaging and scoliosis. Specifically, the Wald estimator and Delta method were employed to generate a causal estimate and standard deviation for each instrumental variable (IV). Then, the IVW mean of these ratio estimates was calculated as the effect estimate [28]. The results are presented as odds ratios (ORs).

To explore potential unbalanced horizontal pleiotropy, our sensitivity analyses for two-sample MR included assessing between-SNP heterogeneity (determine whether a specific SNP perturbed the results [29]) using Cochran's Q-statistic, leave-one-out analysis (when the number of SNPs exceeds two) and MR-PRESSO. We evaluated the strength of each SNP by employing the F statistic, F = β²/SE², a measure of the magnitude and accuracy of the genetic effect. SNPs with F values lower than 10 were excluded as SNPs with F scores greater than 10 offered sufficient evidence to guarantee their validity [30].

2.7. Statistical analysis

All continuous data which was normally distributed were described as mean ± standard deviation (SD). The median and interquartile range (IQR) were used when the sample did not display normal distribution. Statistical analyses were performed using SPSS 22.0 computer software (SPSS Inc., Chicago, IL, USA).

The t-test, nonparametric test, and Chi-square test were used appropriately to describe the difference in patients’ baseline characteristics. To identify risk factors associated with scoliosis, univariate analysis by logistic regression was performed, using demographic and medical characteristics. Multivariable logistic regression included variables with P ≤ 0.10 in univariate analyses or those of clinical relevance (sex [[31], [32], [33]], convulsion [13,34], and EEG severity [35,36]). ORs were calculated from the multivariable analysis to the quantify association with scoliosis with 95 % CIs. The time course for scoliosis development was analyzed by the Kaplan–Meier method. Cox proportional hazard ratio (HR) and 95 % confidence interval (CI) were used to determine related factors for scoliosis development and progression. Using the multiple correction method of False Discovery Rate (FDR) adjustment. All tests were 2-sided, and P values less than 0.05 were considered to be statistically significant.

3. Result

3.1. Childhood CNS inflammation increases the incidence of scoliosis in the UK biobank cohort

The incidence of scoliosis in the CNS inflammation group was 1.51 %, compared to 0.91 % in the control group (Supplementary Table 2, P = 0.03). Cox regression analysis showed that CNS inflammation increased the new-onset scoliosis by 1.77 times (P = 0.017). Male was identified as a protective factor (HR = 0.479, P = 0.001). Childhood body type is defined as participants’ self-reported perceived body size at age 10 —“thinner,” “about average,” or “plumper” compared to peers. Childhood body type had no impact on the incidence of new scoliosis (Supplementary Table 3). CNS inflammation occurring between ages 0–18 years was a risk factor for scoliosis, with younger children having a higher probability of developing scoliosis after CNS inflammation (P = 0.042). Additionally, we analyzed the effects of encephalitis during adulthood and childhood on scoliosis. The results indicated that only childhood CNS inflammation was associated with an increased incidence of scoliosis (HR = 1.744, P = 0.02), while CNS inflammation in adulthood had no significant effect on the development of new scoliosis (Supplementary Table 4, P = 0.759).

3.2. Childhood viral encephalitis increases scoliosis incidence in multi-center cohort

Viral encephalitis is a quintessential central nervous system disorder. As such, it provides a compelling basis for investigating the relationship between childhood viral encephalitis and scoliosis within hospital-based cohorts.

In hospital cross-section study, After the exclusion criteria, 4036 male and 2598 female VE patients with a mean age of 7.17 ± 2.75 years were included in the cross-sectional study, while the control group consisted of 26398 children with the same age (7.18 ± 2.73, P = 0.983) and sex (age- and sex-matched). Overall baseline and clinical characteristics of patients were well balanced between groups (Table 1). Scoliosis was present in 9.6 % (638/6634) of VE group vs. 4.0 % (1050/26398) of non-VE group (P = 0.001, Table 1). All VE patients had spinal radiographs accessible for review (Supplementary Table 5), 62 had a Cobb angle of 15°–20° (0.93 %), and 9 had a Cobb angle of ≥20°(0.14 %). The non-VE group with a prevalence of 0.42 % for scoliosis of 15°–20° and 0.05 % for scoliosis of ≥20°. There was a higher prevalence of scoliosis and larger curves in VE patients.

Table 1.

Variables associated with scoliosis in all samples and within the VE group.

variable		Cobb<10°		Cobb≥10°		X-squared	P Value
		N	Composition ratio%	N	Composition ratio %
All samples		33032
Sex	Male	19166	95.2 %	960	4.8 %	12.298	<0.001
	Female	12178	94.4 %	728	5.6 %
Age	3-6∗	17161	96.3 %	663	3.7 %	245.360	<0.001
	7-9∗	8987	94.6 %	515	5.4 %
	10-16∗	5196	91.1 %	510	8.9 %
BMIa	Normal or thin	20148	94.1 %	1258	5.9 %	73.721	<0.001
	Overweight or obesity	11196	96.3 %	430	3.7 %
VE	Non-VE	25348	96.0 %	1050	4.0 %	347.731	<0.001
	VE	5996	90.4 %	638	9.6 %
VE Cases		6634
Corticosteroid therapy	No	5100	91.0 %	504	9.0 %	1.755	0.1853
	Yes	924	89.7 %	106	10.3 %
Fever	No	708	85.5 %	120	14.5 %	25.874	<0.001
	Yes	5288	91.1 %	518	8.9 %
Convulsion	No	5507	90.3 %	593	9.7 %	0.946	0.331
	Yes	489	91.6 %	45	8.4 %
EEG severity	Normal	2286	89.9 %	257	10.1 %	7.384	0.025
	Mild abnormality	3153	91.2 %	305	8.8 %
	Moderate-severe abnormality	557	88.0 %	76	12.0 %
Brain lesions	Normal	4928	91.1 %	482	8.9 %	25.961	<0.001
	Solitary lesion	847	88.6 %	109	11.4 %
	Multiple lesions	221	82.5 %	47	17.5 %

EEG:Electroencephalogram.

^aCalculated as weight in kilograms divided by height in meters squared.

Within the VE group, patients with multiple brain lesions and solitary lesion on brain MRI had a higher prevalence of scoliosis than the normal group (17.5 % vs. 11.4 % vs. 8.9 %, P = 0.001, Table 1). The prevalence of scoliosis between the moderate-severe abnormalities and normal group (12.0 % vs. 10.1 %, P = 0.163, Table 1). Afebrile patients had a higher prevalence of scoliosis than febrile patients (14.5 % vs. 8.9 %, P = 0.001, Table 1). Days of hospitalization, which represents disease severity, were likewise not associated with scoliosis (P = 0.263, data not shown). There was no observable association between the utilization of corticosteroids and the incidence of scoliosis (9 % vs. 10.3 %, P = 0.1853, Table 1).

Age, sex, and BMI are the most recognized factors associated with AIS. The study also showed the consistent results. According to the multiple logistic regression analysis (Supplementary Table 7), it was found that being female (OR = 1.227, P = 0.001) and having a higher chronological age (“7–9” vs. “3–6”, OR = 1.465, P = 0.001; “10–17” vs. “3–6”, OR = 2.500, P = 0.001) were identified as risk factors for scoliosis. Conversely, having a higher BMI ("overweight or obesity" vs. "normal or thin"; OR = 0.721, P = 0.001) was found to be protective against scoliosis. After adjusting for age, sex, and BMI, VE was identified as a risk factor for scoliosis (“VE” vs. “non-VE”, OR = 2.561, P = 0.001). Furthermore, it should be noted that regardless of whether the diagnostic criteria for scoliosis were set at 10° or 15°, there was still a significant difference in the prevalence of scoliosis (Cobb≥15°, OR = 1.485, P = 0.014).

Multiple logistic regression analysis in the VE group (Table 2) showed that chronological age (“7–9” vs. “3–6”, OR = 1.325, P = 0.005, “10–17” vs. “3–6”, OR = 2.196, P = 0.001), BMI (“overweight or obesity” vs. “normal or thin”, OR = 0.703, P = 0.001), Fever (“no” vs. “yes”, OR = 0.640, P = 0.001), brain lesions (“multiple lesions” vs. “normal”, OR = 1.924, P = 0.004, “solitary lesion” vs. “normal”, OR = 1.254, P = 0.048),and EEG severity (“moderate-severe abnormality” vs. “normal”, P = 0.018) were statistically related to the prevalence of scoliosis. However, no statistical significance was found in subgroups of convulsion (P = 0.263) and EEG severity (“mild abnormality” vs. “normal”, P = 0.517). When choosing Cobb angle≥15° as the diagnostic criteria for scoliosis, after sex, age, and BMI adjustment, multiple brain lesions also was a risk factor for scoliosis (OR = 4.260, P = 0.001). A higher prevalence of scoliosis showed in children with brain MRI abnormalities in the cerebellum (16.67 %), brainstem (20.00 %), corpus callosum (20.78 %), and basal ganglia (21.31 %) than in other areas of the brain (Fig. 2). The greater the number of abnormal lesions of brain MRI, the higher prevalence of scoliosis within the VE group (“multiple” vs. “single” vs. “normal”, 17.53 % vs. 11.40 % vs. 9.62 %). However, it has been challenging to establish the causality link between specific brain regions and scoliosis development.

Table 2.

Multivariable logistic regression analysis for the prevalence of scoliosis within VE group.

		Cobb≥10°		Cobb≥15°
		OR(95 % CI)	P	OR(95 % CI)	P
Sex	Male	Reference		Reference
	Female	1.171 (0.989–1.386)	0.066	2.487 (1.482–4.174)	0.001
Age	3–6	Reference		Reference
	7–9	1.325 (1.087–1.615)	0.005	2.285 (1.216–4.295)	0.010
	10–16	2.196 (1.779–2.711)	<0.001	4.348 (2.278–8.299)	<0.001
BMI	Normal or thin	Reference		Reference
	Overweight or obesity	0.703 (0.577–0.856)	<0.001	0.913 (0.504–1.653)	0.764
Fever	No	Reference		Reference
	Yes	0.640 (0.514–0. 796)	<0.001	1.331 (0.597–2.968)	0.484
Convulsion	No	Reference		Reference
	Yes	0.831 (0.601–1.149)	0.263	0.691 (0.245–1.951)	0.486
EEG severity	Normal	Reference		Reference
	Mild abnormality	0.943 (0.788–1.127)	0.517	1.601 (0.890–2.878)	0.116
	Moderate-severe abnormality	1.402 (1.058–1.858)	0.018	2.380 (1.063–5.332)	0.035
Brain lesions	Normal	Reference		Reference
	Solitary lesion	1.254 (1.002–1.569)	0.048	1.258 (0.624–2.534)	0.521
	Multiple lesionsa	1.924 (1.375–2.695)	<0.001	4.260 (2.070–8.767)	<0.001

^a(odds ratio, 95 % confidence interval [CI]).

Fig. 2.

Chest radiographs of children and the prevalence of scoliosis in children with different MRI abnormal sites.

Chest plain radiographs (A, B, E, F, I, J, M, N) are routinely obtained. We used electronic tools to adjust the grayscale to allow for adequate spine visualization. Brain MRI of the VE patients (C, D, G, H, K, L, O, P) demonstrates swelling with T2 hyperintensity of abnormalities in the frontal lobe, occipital lobe, thalamus, etc. (Q) The incidence of scoliosis in different brain lesions.

In order to minimize reverse causality, 7060 subjects (1452 VE patients and 5608 non-VE children) without a diagnosis of scoliosis at baseline were enrolled in the hospital cohort study to assess new-onset scoliosis. The baseline characteristics of two groups were summarized in Supplementary Table 8. There were no significant differences in chronological age (P = 0.979) and sex (P = 0.155). However, BMI (P = 0.001) and duration of follow-up (P = 0.001) were discovered to be statistically different between two groups. The incidence of de novo scoliosis in VE group was significantly higher than in non-VE group (25.4 % vs.7.6 %, P = 0.001). As illustrated in Fig. 3A, VE patients had significantly low de novo scoliosis-free survival than non-VE children (log-rank test, P < 0.0001).

Fig. 3.

Scoliosis-free survival in viral encephalitis and fever, brain lesions subgroups within VE group.

(A) Scoliosis-free survival in VE and non-VE. (B) Scoliosis-free survival in fever. (C) Scoliosis-free survival in the different number of brain lesions. (D) Scoliosis-free survival in the different parts of brain lesions. ∗Specific part: Abnormalities in one of the corpus callosum, basal ganglia, brain stem, and cerebellum. (E) The thesis is that neurological changes could have a causal association with scoliosis.

After sex, age, and BMI adjustment, Cox regression analysis revealed that a history of VE was associated with a 3.604-fold (95 % CI, 3.121–4.163, P = 0.001) increased risk for scoliosis development compared with non-VE children (Table 3). The dynamic impact of different time windows after VE on the risk of scoliosis was analyzed to strengthen the causal temporal evidence. Cox regression analysis revealed a HR of 3.048 (95 % CI: 2.366–3.926) for follow-up periods <2 years and 4.034 (95 % CI: 3.374–4.824) for follow-up >2 years. Age (“10–17” vs. “3–6”, hazard ratio (HR) = 1.643, P = 0.001; “7–9” vs. “3–6”, HR = 1.278, P = 0.004) and sex (“male” vs. “female”, HR = 1.362 P = 0.001) were statistically significant factors associated with scoliosis development. However, scoliosis development had no significant association with BMI in our results(P = 0.171).

Table 3.

Cox regression analysis for new-onset scoliosis of scoliosis.

		HR	95 % CI	P
Sex	Male	Reference
	Female	1.362	1.184–1.566	<0.001
Age	3–6	Reference
	7–9	1.278	1.082–1.510	0.004
	10–16	1.643	1.342–2.011	<0.001
BMI	Normal or thin	Reference
	Overweight or obesity	0.901	0.775–1.046	0.171
VE	Non-VE	Reference
	VE	3.604	3.121–4.163	<0.001

The multivariable Cox regression analysis within VE patients showed that the risk of new-onset scoliosis was increased by 1.490 (solitary lesion) to 2.188 (multiple lesions) times in VE patients with brain lesions compared to patients with normal MRI of the brain (Table 4). We did not find significant statistical differences in age, sex, BMI, convulsions, and EEG abnormality grade.

Table 4.

Cox regression analysis of risk factors for new-onset scoliosis in children with VE.

		HR	95 % CI	P
Sex	Male	Reference
	Female	1.341	1.089–1.652	0.006
Age	3–6	Reference
	7–9	1.156	0.910–1.468	0.236
	10–16	1.165	0.847–1.604	0.348
BMI	Normal or thin	Reference
	Overweight or obesity	0.783	0.617–0.993	0.044
Fever	No	Reference
	Yes	0.693	0.505–0.949	0.022
Convulsion	No	Reference
	Yes	0.920	0.629–1.345	0.667
EEG severity	Mild	Reference
	Moderate	0.891	0.715–1.111	0.306
	Severe	0.809	0.543–1.205	0.298
Brain lesions	Normal	Reference
	Solitary lesion	1.490	1.151–1.930	0.002
	Multiple lesions	2.188	1.423–3.366	<0.001

Fever was a protective factor against the new onset of scoliosis (HR = 0.693, P = 0.022, Table 4, log-rank P = 0.015, Fig. 2B). Patients with solitary or multiple lesions on brain MRI during acute viral encephalitis had a higher risk of developing scoliosis during follow-up than patients with normal MRI (P = 0.001, Fig. 2C), again suggesting a dose–response association for exposure to multiple brain lesions. More importantly, patients with abnormalities in one of the corpus callosum, basal ganglia, brainstem, and cerebellum had the highest incidence of scoliosis compared to other parts of the MRI (P = 0.001, Fig. 2D).

In order to exclude an association between pediatric corticosteroid therapy and scoliosis might confound some of the findings. We conducted a new assessment of the type and dose of hormone use in patients with viral encephalitis. There was no observable association between the utilization of corticosteroids and the incidence of scoliosis (Supplementary Table 10, P = 0.750). When immunomodulatory agents use is considered, the multivariable Cox regression analysis within VE patients showed that the risk of new-onset scoliosis was increased by 1.778 (solitary lesion) to 2.445 (multiple lesions) times in VE patients with brain lesions compared to patients with normal MRI of the brain (Supplementary Table 11). The outcomes of univariate COX analysis exhibited no significant difference between budesonide or dexamethasone dose and the development of scoliosis (budesonide P = 0.178, dexamethasone, P = 0.127, data not shown). The utilization of various hormones did not display an association with the onset of scoliosis, nor did it influence the results of other variables (Supplementary Table 12).

3.3. Childhood viral encephalitis increases scoliosis incidence in zhejiang province scoliosis screening cohort

In the Zhejiang province screening cohort, the results show that the incidence of de novo scoliosis in the VE group was significantly higher than that in the non-VE group (15.4 % vs. 4.2 %, P = 0.001, Supplementary Table 13). We also observed that multiple brain lesions were associated with a higher incidence of scoliosis. Furthermore, the incidence was significantly greater in individuals with multiple brain lesions compared to those with single brain lesion. (HR = 1.896 vs. 1.35, P = 0.001, Supplementary Table 14).

3.4. MR on brain imaging phenotypes and scoliosis

Consequently, we conducted a bidirectional two-sample MR analysis to explore the causal relationships between this image-derived phenotypes (IDPs)-scoliosis. We selected 103 IDPs concerning whole-brain volume T1 imaging, 54 IDPs segmented by the subcortical volumetric segmentation (aseg) method, and 13 data measuring the mean orientation dispersion index (ODI) and the mean isotropic or free water volume fraction (ISOVF) of regions of interest, including the brainstem, cerebellum, basal ganglia, and corpus callosum.

Firstly, the analysis revealed causal effects of volume of grey and white brain, volume of grey matter in right supracalcarine cortex, and volume of grey matter in left thalamus from T1 brain image on the incidence of scoliosis. There were causal effects of volume of SubCortGray in the whole brain generated and volume of Caudate in the right hemisphere generated, as determined aseg, on scoliosis. Both OD in the splenium of the corpus callosum and ISOVF in middle cerebellar peduncle exhibited a causal relationship with scoliosis (Table 5, Supplementary Table 12). Multiple testing corrections, the findings did reach statistical significance. The F-statistic for each SNP is demonstrated to be greater than 29, as presented in Supplementary Table 13. Conversely, the directional effect of scoliosis on neuroimaging phenotypes of the specific brain was then tested. No significant reverse effects of scoliosis on IDPs were found (Supplementary Table 14).

Table 5.

Forest plot summarizing the Mendelian randomization estimates the causal effect on scoliosis.

Exposure	P	FDR P	OR (95 %CI)
IDP_T1_SIENAX_brain-unnormalised_volume	7.32E-04	0.0377	1.68(1.24–2.27)
IDP_T1_FAST_ROIs_R_supracalc_cortex	8.62E-04	0.0296	0.31(0.15–0.61)
IDP_T1_FAST_ROIs_L_thalamus	1.80E-03	0.0464	0.64(0.49–0.85)
aseg_global_volume_SubCortGray	8.36E-04	0.0452	0.63(0.48–0.83)
IDP_dMRI_TBSS_OD_Splenium_of_corpus_callosum	1.82E-05	0.0002	0.61(0.48–0.76)
IDP_dMRI_TBSS_ISOVF_Middle_cerebellar_peduncle	3.53E-03	0.0229	2.10(1.28–3.45)
aseg_rh_volume_Caudate	9.11E-04	0.0246	1.49(1.18–1.89)

The scatter plot results for each outcome of interest are listed in Supplementary Fig. 1. Moreover, the MR-Egger regression and showed that there was a low likelihood of horizontal pleiotropy for our estimations (P-values for MR-Egger intercept >0.05) (Supplementary Table 12). Visually, the leave-one-out analysis plot proved that the results were not altered by the removal of any SNP and were quite robust (Supplementary Fig. 2).

The findings of the multi-center study observational findings were consistent with Mendelian randomization analysis, which showed strong evidence of a causal association between changes in specific brain areas and scoliosis. As listed in Supplementary Table 12, there was no heterogeneity in the neuroimaging biomarkers. Also, the Mendelian randomization sensitivity analysis revealed that the results were consistent with the primary outcome.

3.5. Mediation MR analysis linking brain imaging phenotypes and scoliosis

To assess the mediating role of brain proteins, cerebrospinal fluid proteins, and plasma proteins in the relationship between brain structure and scoliosis, we first evaluated the effects of 832 brain proteins, 1237 cerebrospinal fluid proteins, and 4907 plasma proteins on scoliosis. The results indicate a causal relationship between the brain proteins LBP (OR = 0.309 (0.180–0.532), FDR Pval = 0.006), LRPAP1 (OR = 1.915 (1.457–2.517), FDR Pval = 0.001), and ERBB4 (OR = 1.909 (1.461–2.495), FDR Pval = 0.002) and scoliosis (Table 6, Supplementary Table 15). Additionally, LBP (OR = 0.311 (0.181–0.534), FDR Pval = 9.5 × 10⁻³) and LRPAP1 (OR = 1.890 (1.452–2.460),FDR Pval = 1.88 × 10⁻³) in cerebrospinal fluid also show a causal relationship with scoliosis (Table 6, Supplementary Table 16). LGALS8 (OR = 1.492 (1.240–1.795), FDR Pval = 0.037), TPST2(OR = 0.768 (0.677–0.871), FDR Pval = 0.04), CDY1 (OR = 3.960 (2.151–7.292), FDR Pval = 0.024), MPZ (OR = 0.824(0.749–0.907), FDR Pval = 0.047), CASC4(OR = 0.759(0.697–0.826), FDR Pval = 9.18 × 10⁻⁷), FAM20B (OR = 0.664(0.547–0.806), FDR Pval = 0.042), GOLM1(OR = 0.805(0.725–0.895), FDR Pval = 0.046) in plasma also show a causal relationship with scoliosis (Table 6, Supplementary Table 17). The F-statistic for each SNP is presented in Supplementary Tables 18–20.

Table 6.

Mediation MR analysis linking brain proteins, CSF proteins, plasma proteins with scoliosis.

Exposure	Outcome	P	FDR P	OR (95 %CI)
Brain protein
LBP	Scoliosis	2.22E-05	0.0058	0.31(0.18–0.53)
LRPAP1	Scoliosis	3.12E-06	0.0012	1.92(1.46–2.52)
ERBB4	Scoliosis	2.22E-06	0.0017	1.91(1.46–2.50)
Cerebrospinal fluid proteins
LBP	Scoliosis	2.23E-05	0.0095	0.31(0.18–0.53)
LRPAP1	Scoliosis	2.20E-06	0.0019	1.89(1.45–2.46)
Plasma proteins
LGALS8	Scoliosis	2.27E-05	0.037	1.49 (1.24–1.80)
TPST2	Scoliosis	4.11E-05	0.040	0.77(0.68–0.87)
CDY1	Scoliosis	9.92E-06	0.024	3.96(2.15–7.29)
MPZ	Scoliosis	7.59E-05	0.047	0.82(0.75–0.91)
CASC4	Scoliosis	1.87E-10	9.18E-07	0.76(0.70–0.83)
FAM20B	Scoliosis	3.42E-05	0.042	0.66(0.55–0.81)
GOLM1	Scoliosis	5.67E-05	0.046	0.81(0.72–0.89)

We further conducted mediator MR analyses on the identified statistical significance results to evaluate the effect of brain structure on brain proteins, cerebrospinal fluid proteins, plasma proteins. Six positive results were obtained: the volume of the caudate in the right hemisphere, as determined by aseg, was linked to the brain protein ERBB4 (OR = 0.967, P = 3.02 × 10⁻², 95 % CI 0.938–0.997); the combined volume of grey and white matter in the brain was associated with the cerebrospinal fluid protein LBP (OR = 1.168, P = 1.33 × 10⁻², 95 % CI 1.032–1.321); the volume of grey matter in left thalamus from T1 brain image was associated with FAM20B, LGALS8, and TPST2; the volume of SubCortGray in the whole brain was associated with MPZ (OR = 0.886, P = 7.79 × 10⁻³, 95 % CI 0.811–0.969) (Table 7, Supplementary Table 21). The F-statistic for each SNP is presented in Supplementary Table 22.

Table 7.

Mediation MR analysis linking brain proteins, CSF proteins, plasma proteins with brain imaging phenotypes.

Exposure	Outcome	P	OR (95 %CI)
Brain protein
aseg_rh_volume_Caudate	ERBB4	0.030	0.97 (0.94–1.00)
Cerebrospinal fluid proteins
IDP_T1_SIENAX_brain-unnormalised_volume	LBP	0.013	1.17 (1.03–1.32)
Plasma proteins
IDP_T1_FAST_ROIs_L_thalamus	FAM20B	0.050	1.10 (1.00–1.21)
IDP_T1_FAST_ROIs_L_thalamus	LGALS8	0.00063	1.13 (1.05–1.22)
aseg_global_volume_SubCortGray	MPZ	0.0078	0.89 (0.81–0.97)
IDP_T1_FAST_ROIs_L_thalamus	TPST2	0.038	1.14 (1.00–1.28)

Brains protein ERBB4 mediated −5.42 % of the effect of the caudate volume in the right hemisphere, as determined by aseg, on scoliosis (β = −0.022, 95 % CI -0.043–1.16 × 10⁻⁴, P < 0.05). While the cerebrospinal fluid protein LBP mediated −34.93 % of the effect of the volume of grey and white brain on scoliosis (β = −0.181, 95 % CI -0.348 to −0.0151, P < 0.05). Plasma protein FAM20B mediated 8.81 %, while plasma protein TPST2 accounted for 7.62 % of the effect of grey matter volume in the left thalamus on scoliosis incidence, as determined from T1-weighted brain imaging. In contrast, plasma protein LGALS8 mediated −11.43 % of this effect (Supplementary Table 23).

3.6. Cross-validation analysis

We conducted validation analyses using scoliosis data from other sources, examining seven identified brain regions, three brain proteins, and two cerebrospinal fluid proteins, seven plasma proteins. The results indicated a causal relationship between the brain protein ERBB4, plasma proteins CASC4, FAM20B, GOLM1, and LGALS8, and scoliosis, as observed in the data from Khanshour AM. Additionally, a causal relationship between plasma protein TPST2 and scoliosis was identified in the data from FinnGen R11. A causal relationship was observed between the mean ISOVF in the middle cerebellar peduncle and scoliosis (OR = 2.408, P = 2.66E-02, 95 % CI 1.107–5.238), as well as between the volume of grey matter in the left thalamus and scoliosis (OR = 1.451, P = 4.49E-02, 95 % CI 1.009–2.087). In the scoliosis data from the FinnGen R10 release, a causal relationship was again identified between the mean ISOVF in the middle cerebellar peduncle and scoliosis (OR = 0.45, P = 8.75E-03, 95 % CI 0.251–0.819). This relationship was further confirmed in the scoliosis data from the FinnGen R11 release, where the mean ISOVF in the middle cerebellar peduncle also showed a causal relationship with scoliosis (OR = 0.546, P = 3.13E-02, 95 % CI 0.315–0.947) (Supplementary Table 27).

4. Discussion

The main findings of this study indicated that CNS inflammation during early childhood significantly increase the risk of new-onset scoliosis in UK biobank cohort.

Subsequent studies using both the hospital cohort and the Zhejiang Province screening cohort will focus on viral encephalitis, a classic form of central nervous system inflammation, as the primary research subject. It was found that viral encephalitis, along with alterations in specific brain regions, contributes to an increased incidence of scoliosis. The results were replicated in East Asian and European populations. Similar enhancements were observed across other subgroups by fever, the severity of EEG, and MRI brain lesions. The MR analysis identified seven IDPs associated with brain volume, brainstem, basal ganglia, corpus callosum, and cerebellum, which were causally linked to the incidence of scoliosis. Mediation MR analysis further identified three brain proteins, two CSF proteins, and seven plasma proteins as mediators in the pathway linking brain imaging phenotypes to scoliosis incidence. Unlike prior observational studies, our MR approach robustly addresses reverse causation and confounding, identifying the cerebellum and corpus callosum as primary causal drivers.

Central nervous system infections are known to cause brain region damage, which can be detected using various imaging techniques such as MRI and EEG. These methods allow for the assessment of inflammation locations and severity. Brain MRI is the best-visualized technique for acute encephalitis. It was abnormal in 90 % of cases of Herpes simplex virus (HSV) encephalitis [37]. HSV is one of the most common causes of virus encephalitis. Encephalitis typically involves the temporal lobe [38], insular, cingulate, and frontobasal cortex [37]. The MRI imaging findings of Japanese encephalitis have been described as the subcortical grey matter, including the substantia nigra, basal ganglia, cerebellum, and thalamus abnormalities [39,40]. In addition, enteroviral encephalitis may involve deep grey matter, brainstem, and spinal cord [41,42]. MRI findings in viral encephalitis vary by etiology, with lesion location (e.g., temporal lobe, brainstem) correlating with specific neurological deficits. Abnormal EEG results were found more frequently in critically ill children but were non-specific and could be abnormal in some other causes of encephalopathy [43,44]. Similarly, our results showed that EEG abnormality was a risk factor for scoliosis above 15°. Brain lesion was not only a risk factor for scoliosis development but also an independent risk factor associated with the prognosis of VE in retrospective cohort studies.

Previous research on brain abnormalities-spinal deformity associations has focused on neurofunctional alterations in the cerebrum, brainstem, and cerebellum [45]. Changes in the microstructure and intensity of the white matter within the corpus callosum of AIS patients may be associated with altered motor function, as this area connects the motor and premotor cortices of the two cerebral hemispheres [6,46]. Furthermore, researchers observed an increasingly central role of the temporal and occipital cortex and changes in structural connectivity between the cerebral hemispheres in AIS [47]. Overall, AIS patients may exhibit abnormal brain regions related to motor control, vestibular funuction, hemisphere, and interregional connectivity. However, it has not yet been confirmed whether brain changes found in patients with AIS can pathologically affect neurological function. Furthermore, it is unclear whether the detected abnormalities are the cause of AIS or are compensatory consequences of spinal asymmetry.

Interestingly, our cohort study found that patients with abnormalities in one of the corpus callosum, basal ganglia, brainstem, and cerebellum were more likely to develop scoliosis during follow-up. Evidence also supported a dose–response association within VE group for exposure to multiple brain lesions. Therefore, we speculate that specific brain parts, such as corpus callosum and basal ganglia, could indeed play a causal role in the occurrence and development of scoliosis. Moreover, even after VE was cured, the risk of de novo scoliosis remained higher than in healthy children. To our knowledge, there has not been a large-database epidemiologic study on the association between VE and scoliosis in children. For the first time, we have found that even changes in specific brain areas in early childhood may significantly increase the risk of new-onset scoliosis in the future.

Moreover, findings from conventional observational studies could not account for confounding factors. It is critical to explore the putative causal role of IDPs on scoliosis and vice versa. Mendelian randomization analysis shows that there were substantial causal effects of ODI in brainstem on scoliosis. ODI detect the microstructural complexity of brain tissue is neurite orientation dispersion [48]. Previous research has established a correlation between the ODI and various diseases [49]. Specifically, in patients with multiple sclerosis (MS), an elevated ODI in the corpus callosum was associated with diminished performance on the timed walk test [50]. Additionally, another study revealed heightened ODI levels in MS patients compared to controls across multiple brain regions [51]. Notably, impaired set-shifting reaction time, a metric reflecting cognitive flexibility, was observed and demonstrated a correlation with higher ODI in survivors. To sum up, adolescent scoliosis may be caused by alterations in the brain as a result of viral encephalitis. Our findings provide potential strategies for predicting and intervening in scoliosis risk at the brain-imaging level.

Changes in brain regions and cerebrospinal fluid have been associated with the occurrence of scoliosis; however, the specific molecular mechanisms remain unclear. Our MR mediation analysis may provide insights into potential underlying molecular pathways. Studies have shown that ErbB4 is essential for Venezuelan equine encephalitis virus (VEEV) infection. Inhibition of ErbB4 by Lapatinib effectively suppressed the secretion of proinflammatory cytokines in the VEEV-infected gNVU model. This finding is consistent with our Mendelian randomization results, which revealed a positive correlation between brain protein ErbB4 expression and scoliosis incidence. Moreover, ErbB4 acts as a mediator linking brain injury to the development of scoliosis [52].

Viral infections were also associated with region-specific brain volume loss and were found to induce changes in plasma protein levels [53]. Glycosaminoglycans (GAGs) are critical constituents of the extracellular matrix (ECM) in the annulus fibrosus (AF). FAM20B has been identified as a novel xylose kinase that catalyzes the biosynthesis of GAGs. Mice lacking FAM20B exhibit pronounced spinal deformities [54]. Furthermore, mice that overexpress galectin-8, a secreted mammalian lectin, display increased bone turnover and diminished bone mass [55]. Studies have also demonstrated that Golgi tyrosylprotein sulfotransferase-2 (TPST-2) KO mice present with mild to moderate primary hypothyroidism [56]. Reduced levels of thyroid hormones are associated with decreased bone mass, which has a well-established correlation with the incidence of scoliosis.

Our findings could offer a perspective to guide the early-stage prevention, etiological diagnosis, and treatment of scoliosis at the brain-imaging level. General pediatricians, hospitalists, orthopedic surgeons, researchers, and others may find our findings useful to help them to assess how to optimize health recovery in children best. In view of the significant morbidity rates and long-term sequelae still seen with VE, active investigation of improved therapeutic approaches is warranted. Many viral infections of the nervous system occur during infants and childhood, whereas most cases of AIS occur during adolescence. Moreover, The majority of virus infections are asymptomatic [57], with symptomatic-to-asymptomatic ratios ranging from 1:25 to 1:1000 [58]. Asymptomatic individuals are typically unaware of their infection. Therefore, parents should be aware of the importance of the onset of VE in early life and counsel about the increased risk of AIS occurring in the future. Pediatrics should provide an early referral to orthopedic surgeons. On the other hand, the findings of this study can also help orthopedic surgeons explain some of the etiopathogenesis of idiopathic scoliosis; for researchers, there is more evidence of a causal link between specific brain changes and idiopathic scoliosis.

The main advantage of this study lies in a comprehensive analysis, including multi-cohort analysis and 2-step MR analysis. However, we also acknowledge several limitations of this study. First, brain MRI was assessed on a single measure at baseline, and changes during the follow-up may affect risk evaluation. Secondly, the retrospective design is a limitation. Our findings are primarily applicable to East Asian and European populations due to the genetic and environmental homogeneity of the studied cohorts. Generalization to other ethnic or geographic populations requires further validation in independent cohorts. Determining if scoliosis is caused by viral aggression to encephalitis or neurological sequelae can be difficult. Thirdly, MR analysis exposed populations from Europe, while the outcomes were assessed in an Asian cohort. This potential population heterogeneity could introduce confounding factors. However, the results were consistently validated through cross-validation, strengthening the reliability of the findings despite the cross-population differences. Due to the absence of detailed individual‐level data in FinnGen, we were unable to distinguish primary from secondary forms of scoliosis. The number of cases coded specifically as idiopathic scoliosis in the UK Biobank was insufficient. We acknowledge this low count as a limitation; therefore, we expanded our inclusion to the broader M41 category while applying stringent exclusion criteria. We think more prospective randomized studies are required to help fill the gaps of retrospective studies by collecting new data and confirming cause-and-effect relationships.

Declaration of Generative AI in Scientific Writing

This work did not involve the use of generative AI tools or services in its preparation. The authors take full responsibility for the content of the published article.

Author contributions

SZ, CH and XW contributed to the conception and design of the study; SL, SC, XP, DX, LL, MN, AW and CX contributed to the acquisition and analysis of data; SZ, CH, and XZ contributed to drafting the text or preparing the figures.

Role of the funder/sponsor

The sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Funding

This study was supported by the National Natural Science Foundation of China (82302783), "Pioneer" and "Leading Goose" R&D Program of Zhejiang (2022C03144), Clinical Research Fundation of the 2nd Affiliated Hospital of Wenzhou Medical University (SAHoWMU-CR2018-08-222). Wenzhou Science and Technology Bureau Foundation (ZY2023015).

Declaration of competing interest

All authors have no conflicts of interest to disclose, specifically related to this manuscript. Genetic instruments for scoliosis from FinnGen biobank analysis R10 and R11 (https://www.finngen.fi/fi), and the neuroimaging variables were obtained from complete GWAS summary data on brain region as described by Elliott.

Appendix A. Supplementary data

The following are the Supplementary data to this article: