Astragalus membranaceus therapy in patients with spinocerebellar ataxia type 3 Authors: Chungmin Chiu, Wenling Cheng, Tatsung Lin, Huiju Chang, Minhui Tseng, Yujun Chang, Shihli Su, Chiaju Lee, Henhong Chang, Chinsan Liu Published: 2026 ABSTRACT AbstractBackground and aimSpinocerebellar ataxia type 3 (SCA3) is the most common form of hereditary cerebellar ataxia. The insulin/insulin-like growth factor 1 (IGF1) system (IIS) has been proposed as a potential target for disease-modifying therapy. Astragalus membranaceus (AM) may modulate the IIS pathway and reduce neurodegeneration, as indicated by plasma neurofilament light chain (NfL), a biomarker of disease progression.Experimental procedureA randomized, triple-blind, placebo-controlled crossover trial was conducted in 32 patients with SCA3. Participants received AM or placebo for three months, followed by a one-month washout, then crossed over to the alternate treatment. Plasma NfL and IIS-related markers were measured. Twenty-three participants completed the trial.Results and conclusionIntention-to-treat analysis revealed a modest within-treatment reduction in NfL levels (27.6 ± 10.7 vs. 25.6 ± 9.8 pg/mL, P = 0.040) and increased insulin (12.5 ± 15.2 vs. 21.1 ± 15.5 μIU/mL, P = 0.004) after AM. However, the adjusted regression model showed no significant between-treatment difference in NfL (P = 0.127), and changes in NfL were not correlated with insulin. No significant changes were observed in IGF1 or IGF-binding proteins. Exploratory subgroup analyses suggested larger biomarker changes in later-stage patients. No carryover effects or demographic differences were noted. Overall, AM administration was associated with modest within-treatment biomarker changes in NfL and insulin, although the adjusted between-treatment difference in NfL was not statistically significant. These findings should be interpreted cautiously as preliminary and hypothesis-generating observations rather than confirmatory evidence of therapeutic efficacy.Keywords: Astragalus membranaceus, Spinocerebellar ataxia type 3, Neurofilament light chain, Insulin/insulin-like growth factor 1 signaling, Clinical trial AbbreviationsADAlzheimer's diseaseGHgrowth hormoneAMAstragalus membranaceusGOTglutamic oxaloacetic transaminaseAPSastragalus polysaccharidesGPTglutamic pyruvic transaminaseASTastragaloside IVIGF1insulin-like growth factor 1BMIbody mass indexIGFBPIGF binding proteinBUNblood urea nitrogenITTintention-to-treatCAcerebellar ataxiaIISInsulin/insulin-like growth factor 1 systemCBC/DCcomplete blood count with differential countNfLneurofilament light chainCCHChanghua Christian HospitalPolyQpolyglutamineCrcreatinineSARAscale for assessment and rating of ataxiaDCSdirect current stimulationSCAspinocerebellar ataxiaDMdiabetes mellitusSCA3spinocerebellar ataxia type 3DMTdisease-modifying therapyTMStranscranial magnetic stimulationeGFRestimated glomerular filtration rateWBCwhite blood cellLOCFlast observation carried forward8-OHdG8-hydroxy-2-deoxyguanosine Abbreviations Introduction Spinocerebellar ataxias (SCAs) are a group of dominantly inherited neurodegenerative diseases. Among them, Spinocerebellar Ataxia Type 3 (SCA3), also known as Machado-Joseph Disease, is the most common subtype globally and particularly prevalent in Asia.1 The genetic mutation on chromosome 14q32.1 causing an abnormal expansion of CAG repeats and leading to toxic polyglutamine (polyQ) protein aggregation in the nervous system, ultimately causing neuronal cell loss in the brainstem, cerebellum, midbrain, spinal cord, striatum, and thalamus.2 These aggregates disrupt protein homeostasis, impairing cellular clearance mechanisms and triggering apoptosis, leading to progressive motor deficits and neurodegeneration. Clinically, SCA3 manifest as progressive gait imbalance, dizziness, dysarthria, general fatigue, diplopia, dysphagia, pyramidal or extrapyramidal signs.1 Given the challenges and ethical dilemmas associated with disease-specific gene therapy,3 current management of SCA3 leans towards disease-modifying therapies (DMT) aimed at slowing disease progression.4 Non-pharmacological interventions such as intensive rehabilitation or coordination training, as well as emerging techniques like transcranial magnetic stimulation (TMS) and direct current stimulation (DCS), are among the novel intervention strategies.5,6 Recent advancements in clinical pharmaceutical trials in phases II or III with agents such as trehalose, Troriluzole, human mesenchymal stem cells derived from the umbilical cord or adipose tissues.7 These studies have all provided prospects for the treatment of the disease. The therapeutic target niche for SCA3 within the insulin/insulin-like growth factor 1 system In the brain, the insulin/insulin-like growth factor 1 (IGF1) system (IIS) supports neuronal survival and function, playing a crucial role in cognitive processes and neuroprotection. Insulin and IGF1 activate signaling pathways that promote neuronal growth and repair, enhancing synaptic plasticity and facilitating memory formation. Their anti-apoptotic and anti-inflammatory effects mitigate neuronal damage and protect against neurodegeneration, while ensuring energy supply to neurons and maintaining optimal brain function.8,9 Peripheral insulin and IGF1 penetrate the blood-brain barrier via saturable transport systems and are further facilitated in crossing brain vascular endothelial cells.10,11 Although central insulin levels in SCA3 patients are unknown, a study in 2011, indicated that peripheral insulin levels are lower in SCA3 patients compared to healthy individuals.12 Nasal insulin spray, glucagon-like peptide-1 analogue, or PPARγ agonist, has been attempted in Alzheimer's disease (AD) to promote insulin signaling in the brain.13 In the growth hormone (GH) -IGF1 axis, SCA3 84Q transgenic mice treated with GH showed preserved Purkinje cells and reduced ataxin-3 aggregation, while 9-month IGF1 treatment improved locomotor function and minimized cerebellar degeneration, potentially by modulating mitochondrial function, oxidative stress, and autophagy.14,15 However, whether modulation of the IIS pathway translates into meaningful therapeutic effects in human SCA3 remains unclear. Moreover, long-term stimulation of the somatotropic axis may increase the risk of cancer. Therefore, more moderate approaches to influencing IIS activity may be of interest, and extracts from commonly used herbal medicines could be considered as potential candidates for DMT in SCA3. Astragalus membranaceus as a possible IIS inducer Chen et al. have identified several Chinese herbal medicines that target the ubiquitin proteasome pathway for SCA3.16 However, rigorous and randomized research designs are scarce. In Asia, Astragalus membranaceus (AM, 黄耆Huáng Qí) is widely used for tonifying qi, glycemic homeostasis, anti-fatigue, and anti-aging. Its main components include polysaccharides, saponins, and flavonoids. The pharmacological effects of Astragalus polysaccharides (APS) include immune modulation, anti-tumor activity, anti-apoptotic effects, and prevention of abnormal polyQ aggregation.17 APS enhances insulin sensitivity through AKT activation and IGF1 pathway modulation, and it also promotes the proliferation and insulin secretion of pancreatic β cells.18 Astragaloside IV (AST), one of its most important saponins, exhibits even remarkably antioxidant, anti-inflammatory, and neuroprotective.19 AST protects against diabetes mellitus (DM) by triggering a decline in blood glucose levels and a rise in plasma insulin levels,20 via regulating gut microbiota, AMPK/SIRT1 and PI3K/AKT signaling pathways.21 In our previous study, the use of a herbal formula mainly composed of AM has been shown to assist in improving the quality of life of a SCA patient.22 As mentioned above, AM has been reported to influence components of the IIS; nonetheless, whether this manifest as measurable neuroprotective effects in humans remains unclear. Neurofilament light chain as outcome biomarker in SCA3 Cerebrospinal neurofilament light chain (NfL) reflects neural axonal damage, with plasma levels correlating with those in the cerebrospinal fluid. Plasma NfL demonstrates stability, sensitivity, reliability, and predictive value in tracking the course of neurodegenerative diseases such as multiple sclerosis, AD, amyotrophic lateral sclerosis, spinal cord injury, and ataxia-telangiectasia. High-sensitivity assays further indicate that plasma NfL can capture disease-related changes within weeks to a few months,23 making it a suitable biomarker for short-term monitoring. These features support its application as a dynamic outcome measure in clinical trials, particularly when aiming to minimize participant burden, making this approach both justifiable and feasible.23 NfL for assessing SCA3 has emerged as a novel approach in recent years.24 In transgenic mice, alterations in plasma NfL concentration precede even those of mutant ataxin-3.25 The present study will not assess traditional neurological examinations, such as scale for assessment and rating of ataxia (SARA), because changes in SARA for SCA3 patients within a short timeframe are not anticipated. The mean annual progression in SARA was 1.56 points out of a total score of 40 (SE 0.08), and up to 202 patients would be needed to detect a 50% reduction in progression of the SARA in a one-year trial.26 Hence, we will conduct a three-month clinical trial using a randomized triple-blind placebo-controlled crossover design in patients of SCA3 with AM or placebo administration. We aim to observe whether AM could contribute alterations in plasma IIS indices and corresponding NfL levels. Material and method Participants Patients with SCA3 were recruited from the outpatient clinic of Changhua Christian Hospital (CCH) in Taiwan and provided informed consent. Inclusion criteria comprised: (1) genetic diagnosis of SCA3 by a neurologist, (2) aged between 20 and 80 years. Exclusion criteria included: (1) current use of other herbal or traditional Chinese medicine, (2) allergy to herbal or traditional Chinese medicine, (3) pregnancy or lactation, (4) concurrent major illnesses such as cancer, stroke, heart failure, or renal failure, (5) DM. Basic demographic data including gender, age, body mass index (BMI), age at onset, duration of illness, and CAG trinucleotide repeat number were collected from participants. Study design This was a randomized, triple-blind, placebo-controlled crossover study, conducted at CCH from May 31, 2021, to September 26, 2022. It was approved by the CCH Institutional Review Board (No. 200703), and it has also been registered on ClinicalTrials.gov (NCT05038306). SCA3 participants were randomized into Group A or Group B according to the recruitment order, without knowledge of the experimental drug composition. Each group received either Chinese medicine or placebo first. Group A received AM for twelve weeks initially, followed by a four-week washout period, and then placebo for twelve weeks. Group B followed the reverse sequence. Participants were allowed to continue their previous western medication therapy during the study but were restricted from receiving other Chinese herbal medicines. A research assistant regularly monitored medication adherence and side effects via phone calls during the trial period. In addition, we analyzed the characteristics of dropout participants to evaluate whether attrition might bias the interpretation of trial outcomes. Randomization and blinding Sun Ten Pharmaceutical Co., Ltd. designed a randomization list, using block randomization with a block size of four, which is commonly applied in small-sample studies to maintain balanced allocation between groups during recruitment, and provided it to the CCH Chinese Medicine Pharmacy for sequential medication distribution. Allocation was thus concealed: investigators, study staff and participants were blinded to assignment, and the randomization code was held by the manufacturer until study unblinding. Sample size In this crossover study, participant numbers needed were assessed quantitatively to estimate: a significant difference of 0.05% was set, utilizing a two-tailed test, with a power set at 0.8, and calculating the variance in the same patient's standard deviation before and after. If the sample size was set to advance by 50% before and after, the paired number of cases was estimated to be 33 individuals to reject the null hypothesis. Assuming a dropout rate of 25%, the sample size would then require 44 individuals. In a study by Wilke et al. on serum NfL levels, it was suggested that with an effect size of 20%, approximately 15 participants were needed in both the study and control groups.25 Preparation of AM and the placebo AM and placebo were provided by Sun Ten Pharmaceutical Co., Ltd. AM was manufactured at a rate of 3.0 g from every 3.3 g of crude AM. The placebo consisted of starch and maltodextrin (in a 7:3 ratio), with a small amount of caramel coloring to match the appearance, taste, and color of AM powder. Based on a rat model study, the safe dosage of AM for adults is less than 20 g per day without significant nephrotoxicity.27 According to the dosage of AM extract used for stroke recovery and post-stroke fatigue, a daily dosage of 9 g has no significant side effects.28,29 Therefore, combining clinical experience and the above research results, the dosage of AM in this study was set at 3 g per sachet, twice a day. Participants were required to take one sachet before breakfast and dinner for 12 weeks. Adherence was monitored through both returned sachet counts and review of participants’ medication diaries. Participants were instructed to bring back any unused sachets at each follow-up visit. Safety monitoring AM has been documented as a top-grade herb medicine in the Shen Nong's Herbal Classic (Shennong Bencao Jing), indicating its non-toxic, safe, and suitable for long-term consumption. In animal studies, AM has demonstrated hepatorenal protective effects.30,31 In a clinical trial for post-stroke fatigue, one-month administration of AM showed no side effects on peripheral blood levels, liver function, or renal function.28 Nevertheless, we continued to monitor creatinine (Cr), estimated glomerular filtration rate (eGFR), blood urea nitrogen (BUN), glutamic oxaloacetic transaminase (GOT), glutamic pyruvic transaminase (GPT), and complete blood count with differential count (CBC/DC) during each outpatient visit to ensure the safety in patients with SCA3. Outcome measurements The participants were instructed to undergo blood sampling with assistance from a designated assistant. The initial outpatient neurological assessment, SARA, was conducted by a consistent neurologist. Plasma NfL reduction was designated as the pre-specified primary outcome of this trial. Insulin system, IGF1, and associated binding proteins (IGFBPs), exhibit anomalies in SCA3,12 making the IIS a potential therapeutic target.32 Free-form IGF1, represented by the IGF1/IGFBP3 ratio, reflects IGF1 activity, with IGFBP3 predominantly binding to regulate IGF1 activity. An increase in IGFBP1 levels within the brains of transgenic mice has been found to suppress the quantity of myelin axons and oligodendrocytes,33 contributing to the possible pathogenesis of SCA3.34 8-hydroxy-2-deoxyguanosine (8-OHdG), a product of DNA oxidation, serves as an oxidative stress marker in neurodegenerative diseases such as AD and in SCA3 mice.15,35 Our assessments aim to determine whether AM could ameliorate NfL, IIS or oxidative stress in SCA3 patients. Clinical blood chemistry The analyzed laboratory parameters included blood levels of CBC/DC were analyzed by Beckman Coulter DxH 900 hematology analyzer Beckman Coulter DxH 900 (Beckman Coulter, Miami, Florida, USA). The plasma levels of GOT, GPT, BUN, and Cr were measured using an AU5800 Series Clinical Chemistry Analyzers (AU5811; Beckman Coulter, Brea, CA, USA) by Department of Laboratory Medicine, CCH. Determination of 8-OHdG and NfL in plasma The plasma level of 8-OHdG was assessed using the Highly Sensitive 8-OHdG Check ELISA kit (JaICA-#KOG-HS10E). Quantification of NfL levels in plasma was using the Simoa® NF-light™ Advantage Kit (#103186) and an ultra-high sensitivity protein molecular detection instrument (Simoa HD-X, Quanterix, MA, USA), as previously described in the literature.36 Measurement of IIS in plasma Blood samples were drawn into 10 mL BD Vacutainer® EDTA Tubes and centrifuged at 3000 rpm at 4 °C for 15 min to isolate plasma, which was then stored at −80 °C until analysis. The concentrations of glucose, insulin, IGF1, IGFBP1, and IGFBP3 were determined using specific assay kits: Glucose Colorimetric Assay Kit (Cayman Chemical-10009582), Insulin ELISA kit (Mercodia 10-1113-01), Human IGF1/IGF1 Immunoassay ELISA kit (R&D Systems-DG100B), Human IGFBP-1 ELISA Kit (RayBiotech-ELH-IGFBP1-1), and Human IGFBP3 ELISA Kit (Abcam-ab100541) respectively. Statistics We utilized SPSS v.22.0 to analyze variable changes in the study, employing both parametric and non-parametric statistical methods, depending on the distribution of the data. Due to missing cases and a small sample size, we employed intention-to-treat analysis (ITT) to maintain randomization and reflect real-world conditions. Missing values were handled using last observation carried forward (LOCF) under the ITT principle, and baseline characteristics of completers and non-completers were compared to assess potential bias. Carryover effects were assessed using mixed model analysis to evaluate the quality of the randomized crossover design. After considering variables such as age, baseline NfL plasma levels, drug administration sequence (placebo/drug first), treatment type (placebo/drug), intervention order (first/second three months), and treatment-order interaction, the difference in NfL levels was not statistically significant (P = 0.174). Therefore, we analyzed the results using a pooled sequence. We used independent t-tests and Mann-Whitney U tests for between-group comparisons, and paired t-tests or Wilcoxon tests for within-group comparisons. In addition, multiple regression models were applied to adjust for potential confounders of NfL reduction, providing a more robust assessment of treatment effects. Correlation analyses were conducted to explore relationships between changes in plasma NfL levels and changes in IIS-related exploratory biomarkers showing treatment-associated effects. Statistical significance was set at P < 0.05. Results Study flow and demographics This study recruited 32 subjects with SCA3, who were randomly assigned, with 16 patients entering Group A and 16 patients entering Group B (Fig. 1). In Group A, there were no participants who dropped out during the first three months of AM. During the washout period, two participants asked to exit the study, one had experienced a seizure due to discontinuing seizure medication six months before the beginning of the trial, and the other discontinued because of recurrent rhinitis and a desire to reduce medication usage. In the subsequent three months with placebo, one was asked to exit due to a transportation issue. In Group B, there were four participants who dropped out during the first three months of placebo: one due to a transportation issue, one due to cardiac event, and two due to numbness of limbs. During the washout period, two participants asked to exit: one due to a transportation issue and one due to an anxiety disorder attack. In the subsequent three months with AM, no participants dropped out. Adherence was satisfactory, with medication records showing that regularly followed-up participants had fewer than 10 sachets unconsumed during the study period. Flowchart of study recruitment. SCA, spinocerebellar ataxia. CA, cerebellar ataxia. DM, diabetes mellitus. AM, Astragalus membranaceus. In total, 23 participants completed the study, corresponding to an overall dropout rate of 28%. Attrition bias assessment revealed no significant differences between completers and non-completers (Supplementary Table 1). Our study design employed ITT analysis and pooling sequence, encompassing all 32 patients, with 26 receiving AM and 25 receiving placebos. Basic demographic data for Group A and Group B are provided in Table 1. There were no significant differences between the two groups in terms of demographic characteristics, SCA features, and IIS. Table 1Demographic characteristics of participants.Table 1Group A (mean ± SD)Group B (mean ± SD)PBasic featuresCase number1616Sex (male, %)12 (75)8 (50)0.144Age (yr)47.5 ± 13.245.9 ± 14.10.739BMI (m2/kg)22.7 ± 3.821.9 ± 3.10.520SCA featuresAge at onset (yr)36.9 ± 9.837.5 ± 14.50.887Duration (yr)10.6 ± 7.58.4 ± 5.60.344CAG count72.2 ± 3.670.3 ± 5.50.264SARA14.9 ± 10.814.2 ± 6.00.818NfL (pg/mL)28.9 ± 12.627.6 ± 9.10.7488-OHdG (ng/mL)3.37 ± 0.633.95 ± 2.000.955 Insulin/Insulin-like growth factor 1 systemGlucose (mg/dL)97.7 ± 20.0101.6 ± 27.90.792Insulin (μIU/mL)5.7 ± 3.29.2 ± 7.50.187IGF1 (ng/mL)97.1 ± 33.185.0 ± 23.00.239IGFBP1 (pg/mL)441.4 ± 554.11043.0 ± 1974.20.806IGFBP3 (ng/mL)55.5 ± 27.045.0 ± 21.70.258IGF1/IGFBP3 ratio1.96 ± 0.862.18 ± 0.980.366BMI, body mass index. SCA, spinocerebellar ataxia. SARA, scale for the assessment and rating of ataxia. NfL, neurofilament light chain. 8-OHdG, 8-hydroxy-2′-deoxyguanosine. IGF1, insulin-like growth factor 1. IGFBPs, insulin-like growth factor binding proteins. Demographic characteristics of participants. BMI, body mass index. SCA, spinocerebellar ataxia. SARA, scale for the assessment and rating of ataxia. NfL, neurofilament light chain. 8-OHdG, 8-hydroxy-2′-deoxyguanosine. IGF1, insulin-like growth factor 1. IGFBPs, insulin-like growth factor binding proteins. Safety monitoring There were no participants who dropped out during the three months of AM in either Group A or Group B. Following interventions with AM and placebo, most monitored biochemistry parameters in plasma showed no significant changes (Supplementary Table 2). Although some parameters exhibited alterations, they remained within the normal range: Cr significantly decreased in the AM group (P = 0.028), eGFR significantly increased in the AM group (P = 0.016), and white blood cell (WBC) significantly decreased in the placebo group (P = 0.004). Regarding liver and kidney function, there were no obvious changes observed in either group. NfL, 8-OHdG, and IIS after intervention Plasma NfL, the primary outcome of this trial, decreased after AM administration from 27.6 ± 10.7 to 25.6 ± 9.8 pg/mL, representing an approximately 7% reduction (P = 0.040; Fig. 2A) in 26 patients with SCA3, whereas no significant change was observed in the placebo group (25.4 ± 12.5 vs. 25.8 ± 14.3 pg/mL, P = 0.768). However, in the multiple regression model adjusting for potential confounders, the between-treatment difference in NfL reduction did not reach statistical significance (mean difference: −2.372, P = 0.127; Supplementary Table 3). To further address potential bias from missing data, LOCF analysis was applied and confirmed a modest but significant reduction in NfL after AM (P = 0.041), with no significant change in the placebo group (P = 0.767). Pooled analysis of NfL and insulin before and after three months AM (A)/Placebo (B) intervention. The data for the control group were sourced from another study conducted by our team and comprised 36 sex- and age-matched healthy subjects. NfL, neurofilament light chain. AM, Astragalus membranaceus. Regarding insulin and glucose metabolism, glucose levels remained unchanged in both groups, but insulin nearly doubled after AM treatment (12.5 ± 15.2 to 21.1 ± 15.5 μIU/mL, P = 0.004; Fig. 2B), while no significant change was detected in the placebo group (10.7 ± 9.0 vs. 13.6 ± 15.6 μIU/mL, P = 0.990). Importantly, no significant correlation was observed between the magnitude of NfL reduction and insulin elevation (r = 0.165, P = 0.420; Supplementary Fig. 1). Other biomarkers, including 8-OHdG, IGF1, IGFBP1, IGFBP3, and the IGF1/IGFBP3 ratio, did not exhibit significant changes following AM or placebo administration (Table 2). Table 2Plasma biomarkers and IIS before and after 12 weeks of treatment with AM or placebo.Table 2AM (n = 26)Placebo (n = 25)mean ± SDPmean ± SDPNfL (pg/mL)Before27.6 ± 10.70.040∗25.4 ± 12.50.768After25.6 ± 9.825.8 ± 14.38-OHdG (ng/mL)Before3.89 ± 1.880.7703.87 ± 1.660.864After3.78 ± 1.464.01 ± 1.86Glucose (mg/dL)Before103.5 ± 23.50.101104.1 ± 23.00.696After118.5 ± 39.0103.6 ± 22.3Insulin (μIU/mL)Before12.5 ± 15.20.004∗10.7 ± 9.00.990After21.1 ± 15.513.6 ± 15.6IGF1 (ng/mL)Before95.2 ± 32.60.64791.7 ± 29.60.412After97.4 ± 30.395.3 ± 30.0IGFBP1 (pg/mL)Before406.5 ± 592.90.732798.3 ± 1608.90.459After350.9 ± 540.7411.3 ± 520.7IGFBP3 (ng/mL)Before55.4 ± 26.60.84951.1 ± 23.50.989After54.8 ± 25.751.7 ± 24.3IGF1/IGFBP3 ratioBefore1.93 ± 0.840.5172.09 ± 1.030.326After1.99 ± 0.812.18 ± 1.16IIS, insulin/insulin-like growth factor 1 system. AM, Astragalus membranaceus. NfL, neurofilament light chain. 8-OHdG, 8-hydroxy-2′-deoxyguanosine. IGF1, insulin-like growth factor 1. IGFBPs, insulin-like growth factor binding proteins. The P value was calculated for the comparison between before and after treatment. ∗P value less than 0.05. Plasma biomarkers and IIS before and after 12 weeks of treatment with AM or placebo. IIS, insulin/insulin-like growth factor 1 system. AM, Astragalus membranaceus. NfL, neurofilament light chain. 8-OHdG, 8-hydroxy-2′-deoxyguanosine. IGF1, insulin-like growth factor 1. IGFBPs, insulin-like growth factor binding proteins. The P value was calculated for the comparison between before and after treatment. ∗P value less than 0.05. Subgroup analysis of NfL, 8-OHdG, and IIS after intervention Due to variations in the severity of enrolled patients, AM may yield diverse effects. According to the classification derived from the SARA progression scale,37,38 individuals receiving AM with SARA scores <3 points were categorized into the preclinical group (n = 2), those with scores ≤11 were labeled as stage I (n = 9), and those with scores >11 were allocated to stage II (n = 15) (Table 3). For the preclinical subgroup, results are presented as effect sizes due to the very small sample size: plasma NfL decreased by 2.1 pg/mL (16%) after AM, while insulin increased by 21.7 μIU/mL (344%). Other parameters showed variable changes without consistent trends. Stage I results are presented with P values in the table; however, given the limited sample size, these findings should also be considered exploratory. In this group, only a modest reduction in 8-OHdG levels (3.97 ± 1.77 vs. 3.49 ± 1.45, P = 0.028) was observed after AM administration. In contrast, stage II patients exhibited a significant reduction in NfL (32.6 ± 10.8 vs. 30.4 ± 9.3, P = 0.033; Fig. 3A). No change in NfL was observed after the placebo across the two stages (Fig. 3B). Glucose levels remained consistent in each stage. Like NfL, insulin showed significant variation only in stage II patients (8.0 ± 5.6 vs. 19.8 ± 15.3, P = 0.003; Fig. 3C) after AM but not after the placebo (Fig. 3D). However, there was no direct significant correlation between the change in NfL and insulin after AM, possibly due to the complex interaction between the insulin system and the central nervous system. No variation was noticed in IGF1, IGFBPs, and IGF1/IGFBP3 ratio among the two stages. Table 3Subgroup analysis of plasma biomarkers and IIS before and after 12 weeks of treatment in subjects with AM.Table 3Preclinical (n = 2)Stage I (n = 9)Stage II (n = 15)mean ± SDmean ± SDPmean ± SDPNfL (pg/mL)Before13.2 ± 6.722.6 ± 4.80.45832.6 ± 10.80.033∗After11.1 ± 2.920.8 ± 5.330.4 ± 9.38-OHdG (ng/mL)Before3.84 ± 0.873.97 ± 1.770.028∗3.85 ± 2.120.334After3.88 ± 0.253.49 ± 1.453.93 ± 1.60Glucose (mg/dL)Before94.6 ± 1.794.9 ± 16.20.117111.4 ± 26.20.344After92.4 ± 1.1107.2 ± 28.4102.9 ± 19.8Insulin (μIU/mL)Before6.3 ± 6.021.5 ± 22.90.5948.0 ± 5.60.003∗After28.0 ± 30.821.9 ± 14.519.8 ± 15.3IGF1 (ng/mL)Before101.7 ± 21.791.6 ± 37.60.48396.4 ± 31.70.571After115.0 ± 1.599.4 ± 33.893.8 ± 30.5IGFBP1 (pg/mL)Before90.8 ± 24.5353.2 ± 735.90.260480.6 ± 544.30.865After113.8 ± 161.0237.1 ± 509.1450.8 ± 587.5IGFBP3 (ng/mL)Before67.8 ± 40.451.1 ± 26.00.67856.3 ± 27.00.820After74.4 ± 29.050.0 ± 23.755.2 ± 26.9IGF1/IGFBP3 ratioBefore1.99 ± 1.642.01 ± 0.940.1991.87 ± 0.750.955After1.67 ± 0.632.29 ± 1.071.86 ± 0.63IIS, insulin/insulin-like growth factor 1 system. AM, Astragalus membranaceus. NfL, neurofilament light chain. 8-OHdG, 8-hydroxy-2′-deoxyguanosine. IGF1, insulin-like growth factor 1. IGFBPs, insulin-like growth factor binding proteins. The P value was calculated for the comparison between before and after treatment. Due to the very small sample size in preclinical group (n = 2), statistical testing was not performed; values are presented descriptively only. ∗P value less than 0.05. Subgroup analysis of plasma biomarkers and IIS before and after 12 weeks of treatment in subjects with AM. IIS, insulin/insulin-like growth factor 1 system. AM, Astragalus membranaceus. NfL, neurofilament light chain. 8-OHdG, 8-hydroxy-2′-deoxyguanosine. IGF1, insulin-like growth factor 1. IGFBPs, insulin-like growth factor binding proteins. The P value was calculated for the comparison between before and after treatment. Due to the very small sample size in preclinical group (n = 2), statistical testing was not performed; values are presented descriptively only. ∗P value less than 0.05. Subgroup analysis of NfL and insulin before and after three months AM (A and C)/Placebo (B and D) intervention. There were 2 participants in the preclinical stage (SARA <3), 9 in stage I (SARA ≤11), and 15 in stage II (SARA >11). NfL, neurofilament light chain. AM, Astragalus membranaceus. SARA, scale for assessment and rating of ataxia. Patient-reported positive effects and suspected adverse effects In addition to plasma biomarkers, we recorded subjective responses and suspected side effects in patients after medication administration (Supplementary Table 4). Among patient-reported positive effects, three participants reported improved balance and two reported alleviations of fatigue after taking AM, while none reported improved balance and five reported alleviations of fatigue after taking placebo. Regarding suspected adverse effects, four reported dry mouth and three reported insomnias after taking AM, while two reported dry mouth and one reported insomnia after taking placebo. Nevertheless, no significant differences (P > 0.05) were found between the AM and placebo groups in comparison. Discussion Currently, there is no established therapeutic modality for patients with SCA3, underscoring the urgent need for efficacious symptomatic management or DMT. Utilizing plasma NfL levels to monitor changes pre- and post-treatment presents an opportunity to streamline both time and cost efficiency. In this exploratory pilot crossover study, a borderline reduction in plasma NfL was observed within the AM group following administration, along with the notable increase in insulin levels; however, the adjusted between-treatment difference in NfL was not statistically significant. Although the approximately 7% reduction in plasma NfL may represent a potentially relevant biological signal, its clinical significance remains uncertain. Exploratory subgroup analyses numerically suggested larger biomarker changes in stage II patients, although these findings should be interpreted cautiously given the limited sample size after stratification. This study marks the first clinical intervention utilizing NfL assessment in SCA3 and the first triple-blind randomized crossover trial employing traditional Chinese medicine in cerebellar ataxias. Since the insulin rise was not significantly correlated with NfL change and IGF-related markers remained unchanged, the present data do not establish that AM reduced neurodegeneration through IIS signaling, and the observed NfL reduction should not be regarded as confirmatory evidence of efficacy. Rather, the biomarker pattern should be interpreted as a preliminary, hypothesis-generating signal that requires confirmation in larger, adequately powered trials. Accordingly, this study should be considered a pilot biomarker-guided exploratory trial rather than a definitive efficacy study. Central and peripheral insulin play critical roles in regulating metabolic and neuronal processes. Elevated circulating insulin in mice has been shown to modulate hypothalamic gene transcription.39 Lower circulating insulin levels in diabetic mice correlate with decreased gene expression and synthesis of brain cholesterol, which contributes to myelin formation and neuronal membrane function.40 Unlike other neurodegenerative diseases, individuals with SCA3 often present with normal glycemia but relative hypoinsulinemia.12 The decline in peripheral insulin levels might lead to a reduction in cerebral insulin,41 which may be further aggravated by the presence of polyQ in the brain.42 Lower BMI and elevated insulin sensitivity may represent compensatory adaptations to insufficient insulin.12,43 After AM intervention, peripheral insulin levels in SCA3 patients increased, approaching those observed in the normal population. However, as the association between insulin elevation and NfL reduction was not statistically significant, no causal relationship can be inferred, and the present study does not establish whether the observed biomarker changes are mechanistically linked to IIS modulation. The rationale for focusing on IIS in SCA3 is supported by converging evidence linking AM to IIS. Prior studies demonstrate that APS preserves pancreatic β-cell ultrastructure in type 1 DM models with hypoinsulinemia, thereby mitigating islet inflammation.44 APS has also been shown to reduce insulin resistance and improve glucose homeostasis, potentially lowering the risk of AD by addressing metabolic dysfunction at its root.45 Moreover, AST may influence central insulin regulation; in postnatal rat models of schizophrenia, AST promoted serotonin release in the choroid plexus, which may enhance brain insulin signaling.46,47 Additional exploratory preclinical data suggest that in a rat model of olivo-cerebellar degeneration, AM increased serum and cerebellar IGF1 expression and supported neuronal survival in the inferior olive.48 In clinical practice, AM has been widely applied in Chinese communities for type 2 DM, where both animal and clinical studies indicate benefits in improving hyperglycemia, reducing insulin resistance, and enhancing insulin sensitivity.49 Taken together, these prior findings provide a rationale for investigating IIS-related effects of AM in SCA3; nevertheless, the current clinical dataset supports only an association with circulating insulin change and does not demonstrate a peripheral-to-central IIS mechanism. In this study, plasma NfL showed a modest within-treatment fluctuation after AM administration. Considering that NfL reflects axonal injury, this finding may reflect a potential biomarker signal consistent with putative neuroprotective activity. Given that intra-individual biological variation of plasma NfL is in the low single-digit percentage range, the observed ∼7% reduction may reflect a biological signal but remains within the reported variability range.50 Although our initial hypothesis emphasized modulation of the IIS, no significant changes were observed in IGF1 or IGFBPs. Therefore, the observed NfL reduction may involve alternative or complementary pathways. Previous research has explored AM's protective effects against neurodegenerative diseases. An initial mechanism involves APS exerting neuroprotective effects by activating the Klotho-Nrf2 antioxidant pathway, leading to elevated expression of HO-1, NQO1, and TrxR1, which mitigates ROS-induced neuronal damage.51 Additionally, AST demonstrates potent anti-inflammatory and antioxidant effects. In an AD model, it inhibited microglial activation and NADPH oxidase expression, and reduced ROS and pro-inflammatory cytokines (TNF-α, IL-1β, IL-6), thereby attenuating neuroinflammation and neuronal injury.52 Finally, extracts of AM reversed amyloid-induced memory loss and promote axonal regeneration and synaptic reconstruction in mice, further supporting its neuroprotective potential.53 Notably, plasma 8-OHdG levels, a marker of oxidative stress, showed a modest decline in stage I patients; however, this finding should be interpreted cautiously given the limited sample size. Preclinical studies indicate that AM and its components reduce oxidative DNA damage. AST alleviated kidney injury by suppressing oxidative stress and lowering 8-OHdG levels,54 while APS protected cardiomyocytes from high-glucose injury by preserving mitochondria, reducing apoptosis, and decreasing oxidative markers including 8-OHdG.55 These findings support AM's potential to mitigate oxidative stress-related neurodegeneration in SCA3. However, the detailed mechanisms underlying AM intervention in SCA3 neuropathology warrant further investigation. In patients with DM, hyperinsulinemia or insulin supplementation has been shown to stimulate hepatic IGF1 production.56 In our study, although AM increased insulin levels, it did not lead to elevations in total or free IGF1 in SCA3 patients, nor was there improvement in subjective fatigue. Similarly, IGFBP1, a potential disease biomarker in SCA3, was not significantly altered by AM, despite being regulated by insulin.12 Previous studies reported that AM or AM-containing formulas can enhance IGF1 or IGFBP3 signaling, thereby facilitating muscle repair and growth in children.57,58 APS has also been reported to ameliorate insulin resistance in HepG2 cells by restoring glucose uptake and upregulating IGF1/IGF1 receptor expression. This effect was associated with STAT5 and AKT activation, as well as increased IGF1 secretion.18 The initially elevated free IGF1 levels in our cohort were proposed to represent an endogenous compensatory response to insufficient insulin.12 The lack of additional IGF1 increase after AM treatment may be attributable to patients’ normal baseline IGF1 levels, suggesting that they were not deficient and that AM is unlikely to directly boost IGF1 in this neurodegenerative condition. Overall, the precise mechanisms by which AM modulates insulin and IGF1 warrant further confirmation in future studies. Routine biochemical analyses indicated that AM was safe with respect to complete blood count as well as renal and hepatic function. Placebo administration was associated with a minor reduction in WBC levels, which remained within the normal range and likely reflected random variation. In contrast, AM led to a modest decrease in creatinine within the normal range, potentially suggesting a renal-protective effect29 that merits further investigation. AM demonstrated acceptable tolerability, with only a few reports of dry mouth or insomnia, which occurred at similar rates in the placebo group, rendering causality uncertain. Fatigue improvement was likewise observed in both groups, indicating a possible placebo contribution. Prior clinical trials of AM or APS injections reported only mild adverse events such as eczema, febrile reactions, headache, and constipation, none of which exceeding placebo rates, and all were transient and reversible.59,60 Preclinical toxicology studies further support a wide safety margin: rats tolerated astragalol at 150 mg/kg/day for 91 days without toxicity, single doses of AM up to 5000 mg/kg caused no mortality, and subchronic studies in rats and dogs showed no distinct toxicity at up to 70- and 35-fold the human equivalent dose, respectively.27,61 Previous studies suggest that AM extract or its components, such as formononetin, exert dose-dependent effects on neuroprotection and disease modulation, including PI3K/AKT-mediated apoptosis in hepatocellular carcinoma and attenuation of Aβ-induced neuronal toxicity in AD models.62,63 In contrast, our pilot trial employed a single fixed dose of AM, selected for safety and feasibility considerations in this small-sample clinical setting28,29; therefore, potential dose–response effects should be evaluated in future, larger clinical trials. Collectively, these findings indicate that AM is well tolerated, with a safety profile comparable to placebo. Our findings were considered in the context of previous pharmacological trials in SCA3. An interim analysis of an oral trehalose trial showed no improvement in SARA scores, largely attributed to limited recruitment and underpowering, highlighting the challenges of detecting clinical benefits within short trials.64 In an open-label IGF1 study, stabilization of ataxia was reported over 12 months, but the absence of a placebo control precluded definitive causal inference.65 As all pharmacological agents remain investigational, the only currently validated strategies for transiently slowing disease progression are non-pharmacological approaches such as physical rehabilitation and neuromodulation (e.g., TMC and DCS).5 Importantly, these studies typically do not incorporate plasma biomarkers. In contrast, our randomized triple-blind crossover pilot provides short-term evidence of biomarker changes, and future prolonged follow-up will be valuable to determine whether these biological signals translate into durable clinical benefit. An apparent discrepancy between biomarker and clinical or molecular readouts is not unexpected in this pilot study. Plasma NfL was selected as the primary outcome to capture short-term biological change, whereas conventional clinical scales such as SARA generally require longer observation periods and larger cohorts to detect treatment effects. Likewise, the increase in insulin without parallel changes in IGF1, IGFBP1, or IGFBP3 suggests that any biological response, if present, may reflect an incomplete or non-uniform response of the entire IIS axis. Therefore, the current datasets should be interpreted as early-phase signals at different levels of evidence, not as convergent proof of a single mechanism. This study was limited by a 28% dropout rate, primarily related to comorbid conditions and transportation difficulties; however, no participants withdrew during the three months of AM in both groups. Baseline characteristics were comparable between completers and non-completers, suggesting that attrition was unlikely to introduce substantial bias. Although no statistical carryover effect was detected, residual biological effects cannot be completely excluded due to the chronic progressive nature of SCA3 and the relatively short washout period in this crossover design. To mitigate the impact of missing data, we applied LOCF analysis, which confirmed the preliminary signal of AM-associated NfL reduction. Alternative imputation methods (e.g., multiple imputation) could be considered in additional research to further strengthen robustness. Subgroup analyses by disease stage were underpowered, particularly in the preclinical group (n = 2) and stage I (n = 9), and should therefore be regarded as exploratory and hypothesis-generating. Given the pilot nature of this study and the relatively small sample size, the overall statistical power to detect modest treatment effects was restricted, particularly in subgroup analyses; thus, non-significant findings should be interpreted with caution and not as evidence of absence. Moreover, the relatively small sample size and multiple biomarker assessments mean that the possibility of statistical fluctuation or type I error cannot be fully excluded. In our multiple regression analysis, the reduction in NfL with AM was not significantly greater than with placebo. However, this association yielded a lower P-value than other potential confounders such as CAG repeats or BMI, indicating a potential signal that warrants confirmation in larger studies. The intervention period in this study lasted three months, which was likely insufficient to detect changes in SARA scores.26 In addition, the detailed mechanism by which AM influences plasma NfL, including the interplay of insulin or other systems, remains unclear. Furthermore, potential residual bias related to sample size and subgroup stratification should be considered when interpreting the findings. In the subsequent studies, employing purified compounds derived from AM, such as APS or AST, may provide a more targeted means of enhancing IIS compared to the crude extract used in this study. Conclusion In conclusion, this randomized, triple-blind, crossover pilot study suggests that AM administration was associated with measurable short-term biomarker fluctuations in patients with SCA3. A within-treatment reduction in plasma NfL and an increase in circulating insulin were observed; however, the adjusted between-treatment difference in NfL was not statistically significant, and changes in NfL did not correlate with those in insulin. Given the exploratory pilot nature, limited sample size, and attenuation of significance after adjusted modeling, the observed NfL reduction should be interpreted cautiously and not as confirmatory evidence of efficacy. Importantly, this study represents the first clinical intervention in SCA3 to incorporate plasma NfL as an outcome measure and the first triple-blind randomized crossover trial evaluating AM in cerebellar ataxias. These features highlight both the methodological innovation and the translational relevance of the present work. The feasibility of using plasma NfL as a biomarker supports a more efficient and biologically grounded framework for early-phase therapeutic evaluation in rare neurodegenerative diseases. Future studies with larger cohorts, extended follow-up periods, and adequately powered designs are needed to determine whether the observed biomarker changes translate into sustained clinical benefits. Integration of multimodal outcome measures-including quantitative motor assessments, longitudinal functional scales, and neuroimaging biomarkers-may further clarify therapeutic impact. Additional mechanistic studies exploring insulin-related and alternative neuroprotective pathways will be essential to refine therapeutic strategies. Collectively, these results provide a foundation for continued biomarker-guided investigation of AM in SCA3. CRediT authorship contribution statement Chungmin Chiu: Project administration, Conceptualization, Data curation, Writing - original draft. Wenling Cheng: Writing - review & editing, Data curation, Methodology. Tatsung Lin: Data curation, Methodology. Huiju Chang: Data curation, Methodology. Minhui Tseng: Project administration, Methodology. Yujun Chang: Resources, Software. Shihli Su: Validation, Writing - review & editing. Chiaju Lee: Validation. Henhong Chang: Project administration, Funding acquisition, Writing – review. Chinsan Liu: Project administration, Funding acquisition, Writing - review & editing. Ethics statement This randomized, triple-blind, placebo-controlled crossover trial study was conducted with the approval of the Independent Ethics Committee (CCH-IRB approval No:200703) at Changhua Christian Hospital. All participants provided written informed consent. The study was performed according to ethical principles outlined by Belmont Report, Declaration of Helsinki, and the journal's ethical publication guidelines. Declaration of generative AI and AI-assisted technologies in the manuscript preparation process During the preparation of this work, the authors used ChatGPT to assist with English language editing and rephrasing. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article. Funding The research was funded by research grants 109-CCH-IRP-055 and 110-CCHMST-123 from Changhua Christian Hospital, Taiwan. It was also supported by the China Medical University Hospital in Taiwan (DMR-115-006). The funder had no role in the study design, collection, analysis and interpretation of data, writing of the report and decision to submit the article for publication. Sun Ten Pharmaceutical Co., Ltd. did not provide financial support for this study. Study products were commercially purchased by the investigators. Declaration of competing interest The authors declare that there are no conflicts of interest regarding the publication of this manuscript. Sun Ten Pharmaceutical Co., Ltd. assisted in preparation of the randomization list but had no role in study design, data collection, data analysis, manuscript preparation, or the decision to publish. Acknowledgment The authors thank Dr. Tsung-Han Lee for his assistance with IRB document submission and the Health Examination and Management Center of Changhua Christian Hospital for their help in sample collection and patient coordination. This work was also partly supported by National Research Institute of Chinese Medicine, Ministry of Health and Welfare, Taiwan (MOHW114-NRICM-M-124-000002) and the “Chinese Medicine Research Center, China Medical University” from The Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (MOE) in Taiwan (CMRC-CMA-0). Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.jtcme.2026.06.002. Contributor Information Henhong Chang, Email: tcmchh55@gmail.com. Chinsan Liu, Email: liu48111@gmail.com. Supplementary data The following are the Supplementary data to this article: Multimedia component 1 Data availability In adherence to transparency and reproducibility principles, we acknowledge the importance of data availability statements. However, due to patient privacy and confidentiality concerns, we are unable to share the dataset publicly. We are committed to scientific collaboration and will make the data underlying our findings available to qualified researchers upon request. This access will be subject to ethical and legal considerations, along with a data access agreement to ensure participant privacy. Interested researchers can contact Chin-San Liu (liu48111@gmail.com) and Chung-Min Chiu (shongdiah@gmail.com) for further information. References Associated Data Supplementary Materials Multimedia component 1 Data Availability Statement In adherence to transparency and reproducibility principles, we acknowledge the importance of data availability statements. However, due to patient privacy and confidentiality concerns, we are unable to share the dataset publicly. We are committed to scientific collaboration and will make the data underlying our findings available to qualified researchers upon request. This access will be subject to ethical and legal considerations, along with a data access agreement to ensure participant privacy. Interested researchers can contact Chin-San Liu (liu48111@gmail.com) and Chung-Min Chiu (shongdiah@gmail.com) for further information.