Cerebellar Lipid Dysregulation in SCA3: A Comparative Study in Patients and Mice

Abstract

Spinocerebellar ataxia type 3 (SCA3) is the most common dominantly inherited ataxia and belongs to the family of nine diseases caused by a polyglutamine expansion in the disease-causing protein. In SCA3, a polyglutamine expansion in ATXN3 causes neuron loss in disease-vulnerable brain regions, resulting in progressive loss of coordination and ultimately death. There are no disease-modifying or preventative treatments for this uniformly fatal disorder. Recent studies demonstrate prominent white matter atrophy and microstructural alterations in disease-vulnerable brain regions of SCA3 patients and mouse models. However, the major constituent of white matter – lipids – remains understudied in SCA3.

In this study, we conducted the first unbiased investigation of brain lipids in SCA3, focusing on the disease-vulnerable cerebellum of SCA3 postmortem patients and mouse models. Liquid chromatography-mass spectrometry uncovered widespread lipid reductions in patients with SCA3. Lipid downregulation was recapitulated in early- to mid-stage mouse models of SCA3, including transgenic YACQ84 and Knock-in Q300 mice. End-stage Knock-in Q300 mice displayed a progressive reduction in lipid content, highlighting targets that could benefit from early therapeutic intervention. In contrast, Atxn3-Knock-out mice showed mild lipid upregulation, emphasizing a toxic gain-of-function mechanism underlying lipid downregulation in SCA3.

We conclude that lipids are significantly altered in SCA3 and establish a platform for continued exploration of lipids in disease through interactive data visualization websites. Pronounced reductions in myelin-enriched lipids suggest that lipid dysregulation could underlie white matter atrophy in SCA3. This study establishes the basis for future work elucidating the mechanistic, biomarker, and therapeutic potential of lipids in SCA3.

Introduction

Spinocerebellar ataxia type 3 (SCA3), also called Machado-Joseph disease (MJD), belongs to the family of nine neurodegenerative diseases caused by a CAG trinucleotide repeat expansion in the coding disease gene (Bunting et al., 2022; McLoughlin et al., 2020). In SCA3, the disease-causing gene is ATXN3, which translates into the mutant ATXN3 protein containing a polyglutamine (polyQ) repeat expansion. Mutant ATXN3 aggregation in disease-vulnerable brain regions such as the cerebellum and brainstem leads to neuron loss and gliosis, manifesting as symptoms of impaired gait, coordination, and balance (Durr et al., 1996; Paulson et al., 1997; Paulson et al., 2017; Rub et al., 2013). Though SCA3 is the most common dominantly inherited ataxia (Durr, 2010; Gardiner et al., 2019; McLoughlin et al., 2020), no disease-modifying treatments exist, necessitating ongoing research into the mechanisms underlying this universally fatal disease.

Cerebellar white matter atrophy has recently emerged as a prominent feature of SCA3 disease progression (Adanyeguh et al., 2018; Arruda et al., 2020; Chandrasekaran et al., 2022; Faber et al., 2021; Ferreira et al., 2024; Guimaraes et al., 2013; Jao et al., 2019; Kang et al., 2014; Li et al., 2022; Liu et al., 2023; Lukas et al., 2006; Park et al., 2020; Putka et al., 2023; Rezende et al., 2018; Schulz et al., 2010; Wan et al., 2020; Wu et al., 2017). Ferreira et al. (2024) reported severe volumetric reductions in the cerebellar white matter of symptomatic SCA3 patients, with white matter volume inversely correlating with ataxia severity (Arruda et al., 2020; Ferreira et al., 2024). Prior to symptomatic onset of ataxia, SCA3 mutation carriers demonstrate atrophy and alterations in the microstructure of cerebellar white matter tracts (Chandrasekaran et al., 2022; Faber et al., 2021; Rezende et al., 2018; Wu et al., 2017). Measurements of microstructural changes were highly effective at detecting pre-ataxic disease, with receiver operating characteristic analyses demonstrating that a combination of two measures in the cerebellum resulted in nearly complete separation of pre-ataxic patients from controls (Chandrasekaran et al., 2022). Corroborating neuroimaging studies, reduced myelin staining is evident in postmortem SCA3 patient cerebella relative to controls (Costa et al., 2020; Schuster et al., 2023). These studies provide evidence of early, robust white matter deficits in patients with SCA3, necessitating mechanistic investigation in disease models.

Of relevance to preclinical disease modeling, our lab recently demonstrated that white matter alterations are conserved in disease-vulnerable brain regions of SCA3 transgenic mice in the form of thinner myelination, reduced myelin staining, and a reduced number of mature myelinating oligodendrocytes (Schuster et al., 2024; Schuster et al., 2022b). Treating mid-stage symptomatic mice with anti-ATXN3 antisense oligonucleotide gene silencing treatment, previously established to rescue mouse locomotor behavior, disease pathological hallmarks, and irregular neuronal firing (Bushart et al., 2020; McLoughlin et al., 2018; Moore et al., 2017), also rescued the white matter myelination deficit, demonstrating the therapeutic potential of targeting white matter changes (Schuster et al., 2024). However, the mechanism underlying white matter deficits in SCA3 patients and mouse models remains unknown. Determining the molecular underpinnings of this early and robust deficit will improve our understanding of disease mechanisms and potential interventions.

Toward this goal, we considered the composition of white matter—which is highly enriched in lipids (Montani, 2021; Poitelon et al., 2020)—and whether lipid dysregulation could be a feature of SCA3 pathogenesis. Lipid dysregulation has been established in many neurodegenerative diseases (Shamim et al., 2018; Wei et al., 2023), including Alzheimer’s disease (Chew et al., 2020; Kao et al., 2020), Parkinson’s disease (Fanning et al., 2020; Xicoy et al., 2019), Huntington’s disease (Block et al., 2010; Karasinska and Hayden, 2011), and ALS (Abdel-Khalik et al., 2017; Chaves-Filho et al., 2019), but remains understudied in SCA3. The SCA3 studies to date used unbiased techniques to investigate blood biofluid metabolites, meaning no comprehensive studies have been conducted on the brain. For example, symptomatic SCA3 patients were found to have a distinct serum metabolic profile from pre-ataxic patients and healthy controls (Yang et al., 2019). This study highlighted four metabolites – two species of free fatty acids, L-Proline, and L-Tryptophan – as putative biomarkers. Mouse models recapitulate patient data, with unbiased metabolomics showing reduced plasma DL-Tryptophan at two symptomatic timepoints in SCA3 transgenic mice (Toonen et al., 2018). Lipidomic studies in the same mouse plasma samples show increased levels of ceramides and di- and triglycerides in SCA3 mice compared to wild-type controls (Toonen et al., 2018), matching the increased triglycerides observed in patient serum (Pacheco et al., 2013). While these studies establish SCA3-related changes to lipids and metabolites in blood biofluids and suggest novel biomarkers, the brain lipidome remains unexplored. This represents a ripe area for investigation of disease mechanisms, biomarkers, and therapeutic opportunities.

In this study, we conduct the first unbiased evaluation of brain lipids in SCA3, focusing on the disease-vulnerable cerebellum of SCA3 patients and mouse models. We start by characterizing the lipidome of patients with SCA3 relative to controls, finding strong lipid downregulation at end-stage disease. To determine whether human results can be replicated in SCA3 mouse models, we assess a transgenic model overexpressing human ATXN3 with 84 CAG repeats (YACQ84) and a Knock-in model expressing mouse Atxn3 with a hyper-expanded 300 CAG repeats (KIQ300). Both models recapitulate the SCA3 patient signature of lipid downregulation. We further leveraged these models to understand the mechanism and progression of cerebellar lipid dysregulation. For mechanistic insight, we compare the lipidome of SCA3 mice to that of Atxn3-Knock-out (KO) mice and conclude that mutant ATXN3 toxic gain-of-function likely underlies lipid downregulation in disease. To interrogate progression, we compare early- and end-stage Knock-in mice and identify lipid candidates that may benefit from early therapeutic intervention. We also sought to understand how lipidome alterations may affect white matter integrity by profiling the lipid classes enriched in myelin, again finding strong decreases in SCA3 patients and mouse models, but not Atxn3-Knock-out mice. Finally, we created a resource for the field in the form of interactive data visualization websites, encouraging researchers to leverage these findings in future lipid investigations. In sum, we demonstrate robust gain-of-function lipid downregulation in SCA3, shedding light on a possible mechanism for the early white matter loss observed in SCA3 patients and mouse models.

Materials and methods

Animal Procedures

All animal procedures were approved by the University of Michigan Institutional Animal Care and Use Committee and conducted in accordance with the United States Public Health Service’s policy on Humane Care and Use of Laboratory Animals. Animals were housed in a room with standard 12-hour light/dark cycles and food and water provided ad libitum.

Mouse models

Transgenic, Knock-in, and Knock-out mice were used in this study, with the number of female (F) and male (M) mice included in each figure legend. The YACQ84 mouse model (Jax Strain #012705) was originated generated by Cemal et al. (2002) and overexpresses human ATXN3 with 84 CAG repeats. The mice included in this study were homozygous (Q84/Q84) – expressing four copies of human ATXN3 – and were produced by mating hemizygous male and female mice. Of note, these mice still express two copies of mouse Atxn3 with the endogenous 6 CAG repeats. The Knock-in Q300 mouse model was originally generated with 82 CAG repeats (Knock-in Q82) by Ramani et al. (2015, 2017). Over time, breeding of Atxn3^Q300/Q6 females with Atxn3^Q6/Q6 males allowed the expansion of the repeat to 300 CAG repeats. This study uses heterozygous (Atxn3^Q300/Q6) Knock-in mice expressing one copy of mouse Atxn3 with 6 CAG repeats and one copy with a hyper-expanded 300 CAG repeats. Finally, the Atxn3-Knock-out (KO) line was originally generated by Reina et al. (2012). The mice included in this study had complete knock-out of the mouse Atxn3 gene (Atxn3^−/−) and were produced by mating heterozygous male and female mice.

Genotyping

Genotyping was performed using tail biopsy DNA, which was isolated before weaning and confirmed postmortem, as previously described (Schuster et al., 2023; Schuster et al., 2024; Schuster et al., 2022b). Genotyping for the YACQ84 line (Cemal et al., 2002) was completed via qPCR of the human ATXN3 transgene using Taqman primer probes (forward 5′-FAM-ACAGCAGCAAAAGCAGCAA-3′, reverse 5′-CCAAGTGCTCCTGAACTGGT-3′, probe CTGCCATTCTCATCCTC). Genotyping for the Knock-in Q300 (KIQ300) line (Schuster et al., 2023) was performed via PCR amplification with primers flanking the endogenous mouse Atxn3 CAG repeat (KI forward 5′-TTCACGTTTGAATGTTTCAGG-3′, KI reverse 5′-ATAT GAAAGGGGTCCAGGTCG-3′) and gel electrophoresis. Similarly, genotyping for the Atxn3-Knock-out (KO) line (Reina et al., 2012) was accomplished via PCR amplification (ATXN3KO forward 5′-GAGGGAAGTCGTCATAAGAGT-3′, ATXN3KO reverse 5′-TGGGCTACAAGAAATCCTGTC-3′, and ATXN3KO LTRa 5′-AAATGGCGTTACTTAAGCTAG-3′) followed by gel electrophoresis.

Tissue collection

Samples size was determined via power analysis of previously published results (Schuster et al., 2023; Schuster et al., 2022b). Tissue collection occurred at ~16 weeks of age for YACQ84 and Atxn3-KO mice, and ~24 weeks and ~57 weeks for KIQ300 mice. The average age (in weeks) at collection was 15.92 ± 0.05 (SEM) for YACQ84, 15.88 ± 0.04 (SEM) for Atxn3-KO mice, 23.99 ± 0.01 (SEM) for the 24-week KIQ300 timepoint, and 56.57 ± 0.3 (SEM) for the 57-week KIQ300 timepoint. To collect brains for analysis, mice were anesthetized with a lethal dose of ketamine-xylazine and then transcardially perfused with 1X PBS. The cerebellum was macrodissected from the left hemisphere of the brain and flash-frozen on dry ice for subsequent experiments, as previously described (Schuster et al., 2023; Schuster et al., 2024; Schuster et al., 2022b).

Liquid chromatography-mass spectrometry (LC-MS) lipidomics

Human samples

Post-mortem frozen cerebellar tissue from SCA3 patients and control subjects (cause of death not related to the CNS) were acquired from the Michigan Brain Bank (Table 1). The average CAG repeat size of mutant ATXN3 in patients with SCA3 was 69.5 ± 1.44 (SEM).

Table 1.

Human postmortem sample information

Label	BBID	Age (years)	Sex	PMI (hours)	ATXN3 CAG repeat size	Cause of death
CTRL #1	14	65	M	24	N/A	Acute respiratory distress syndrome
CTRL #2	1532	43	M	9	N/A	Acute perforation and septic shock
CTRL #3	729	59	M	12	18/21	Sudden cardiac arrest
SCA3 #1	1547	84	F	20	21/66	SCA3-related
SCA3 #2	1035	59	F	4	21/70	SCA3-related
SCA3 #3	1930	75	F	39	19/69	SCA3-related
SCA3 #4	1832	49	M	48	12/73	SCA3-related

BBID, Michigan Brain Bank ID; Sex, Male (M) or Female (F); PMI, Postmortem interval (hours); N/A, not available.

Lipid extraction

For lipidomic studies, 25 mg of frozen cerebellar tissue from each human postmortem sample or the frozen left cerebellum) from mice was homogenized in 1X PBS buffer with protease (Sigma Aldrich, S8820) and phosphatase (1 mM NaF, 1 mM β- Glycerophosphate, 1 mM PMSF, 1 mM Na3VO3, 2.5 mM Na4(PO4)2) inhibitors. After tissue homogenization, protein concentration was measured using a BCA assay (Thermo Fisher Scientific). Lipids were isolated using the Folch extraction (Folch et al., 1957) method and relative quantification was carried out with positive and negative ion mode liquid chromatography-mass spectrometry (LC-MS). The volume of 100 μg protein equivalent lysate was adjusted to 40 μL using 1X PBS lysis buffer, with 6μl EquiSPLASH^™ LIPIDOMIX^® (Avanti Polar Lipids, 330731) added as the internal standard. Lipids were extracted using 800 μL of 2:1 (v/v) chloroform: methanol mixture. The organic layer was dried using speedVac and resuspended in 100 μL of 1:3 (v/v) chloroform: methanol mixture.

LC-MS instrumentation and data acquisition

Lipid separation was performed using an Agilent 1290 Infinity II UHPLC system equipped with an Agilent Poroshell 120 EC-C18 chromatography column (2.1 × 100 mm, 2.7 μm) with solvent (A) 90:10 (v/v) water: methanol with 10 mM ammonium acetate and solvent (B) 20:30:50 (v/v) acetonitrile: methanol: isopropanol with 10 mM ammonium acetate. Other parameters include column temperature at 60°C and flow rate of 0.3 mL/min. For injection, 1.5 μg protein equivalent and 2.0 μg protein equivalent were used for positive and negative mode, respectively. Data was acquired on an Agilent 6550 quadrupole time-of-flight mass spectrometer scanning m/z range 200 – 1700 in MS only mode. A pooled lipid sample containing equal amounts from all tubes was analyzed using the iterative MS/MS mode with a fixed collision energy of 25 eV.

Lipid identification and analysis

Lipid identification was performed using Agilent Lipid Annotator software to create a library of identified lipids from pooled samples with a set mass deviation of 10ppm and overall spectral match score above 70%. Peak integration of extracted ion chromatograms (EIC) for each m/z value was carried out in Agilent MassHunter Profinder software. The library was utilized for relative quantification and statistical analysis. Lipids were named according to LIPID MAPS classification (Fahy et al., 2005; Fahy et al., 2009; Liebisch et al., 2020; Liebisch et al., 2013; O’Donnell et al., 2019), detecting fatty acids (FA), glycerolipids (GL), glycerophospholipids (GP), and sphingolipids (SP) classes in this study. A full list of lipid abbreviations be found in the Glossary. Data from positive and negative ion modes were combined after normalization to the internal lipid standard. For lipids identified in both modes, the higher raw abundance was used. Log2-fold change was calculated as the average log2-normalized abundance difference between disease/KO samples and WT/Control samples. Significance was determined using a two-sample t-test with a cutoff of p < 0.05 and fold-change (FC) ≥ 1.5 or ≤ −1.5 (or log2-FC ≥ 0.58 or ≤ −0.58). Heatmaps display a class-based analysis, with log2-normalized abundance averaged within each subclass. The proportion of dysregulated lipids was also calculated for each subclass. In Fig. 5, lipid class abundance was computed by summing the log2-normalized abundance across all species within the hexosylceramide (HexCer) and sulfatide (SHexCer) subclasses.

Figure 5. Lipids enriched in myelin are reduced in SCA3 due to toxic gain-of-function.

(A) Myelin is enriched for cholesterol and glycolipids compared to other biological membranes. (B) Cholesterol is significantly reduced in SCA3 patients, 16-week-old YACQ84 mice, and 24-week- and 57-week-old KIQ300 mice compared to controls. Atxn3-KO mice show significantly increased cholesterol. (B) The glycolipid hexosylceramide (HexCer) was reduced in SCA3 patients as well as 24-week and 57-week KIQ300 mice compared to controls. HexCer was unchanged in YACQ84 mice and Atnx3-KO mice. (C) The glycolipid sulfatide (SHexCer) was significantly reduced in SCA3 patients and both timepoints of KIQ300 mice, but not YACQ84 mice. SHexCer was significantly increased in Atxn3-KO mice compared to controls, suggesting that the decreases observed in disease were not due to ATXN3 loss of function. The cholesterol graph depicts normalized cholesterol abundance at m/z = 365.35. The HexCer and SHexCer graphs depict the sum of normalized lipid abundance. Prior to summing, Grubbs’ outlier test with alpha = 0.05 was performed. Graphs display mean ± SEM, where each point is one mouse. Unpaired t-test for normally distributed data by the Shapiro-Wilk test or Mann-Whitney test for data not normally distributed. P-value or significance level denoted on each graph, ns = not significant, * p < 0.05, ** p < 0.01, *** p < 0.001, and **** p < 0.0001.

Lipid pathway analysis

Lipid pathway analysis was conducted using the LION/web online tool (Molenaar et al., 2023; Molenaar et al., 2019; Ni et al., 2023). Lipid names were converted to LION-compatible terms (detailed in the Supplementary Material). However, LION/web did not recognize ACar, DGTS, MGDG, PEtOH, or PMeOH. For each dataset, pathway analysis was conducted via “Target-list mode”, inputting the differentially expressed lipid names relative to all lipids detected. The significant cutoff was an FDR q-value of 0.05. Of note, there were no significantly upregulated pathways in 24-week-old KIQ300 mice, and no significantly downregulated pathways in 16-week-old Atxn3-KO mice. The proportion of lipids significantly altered to those annotated within each pathway is displayed.

Statistics

Statistical analyses were carried out in Microsoft Excel and GraphPad Prism 10.2.2. Agilent MassHunter Profinder software was used to conduct Principal Components Analysis (PCA), generating PCA coordinates. The PCA, Volcano, and dot plots were created using ggplot2 in RStudio. The ellipse on the PCA plot represents the 95% confidence interval for each genotype. Graphs display the mean ± the standard error of the mean (SEM), with significance levels denoted as: ns = not significant, * p < 0.05, ** p < 0.01, *** p < 0.001, or **** p < 0.0001.

Data availability

Shiny apps were created for each of the five datasets in this paper (human SCA3, 16-week YACQ84, 16-week Atxn3-Knock-out, 24-week KIQ300, and 57-week KIQ300) in RStudio (2023.12.1, running R 4.4.3) to provide interactive, accessible visualizations Interactive data visualizations are available for each dataset at the following links: End-stage Human: https://aputka.shinyapps.io/human/; 16-week YACQ84: https://aputka.shinyapps.io/yacq84/ ; 24-week KIQ300: https://aputka.shinyapps.io/kiq300_24w/ ; 57-week KIQ300: https://aputka.shinyapps.io/kiq300_57w/ ; 16-week Atxn3-KO : https://aputka.shinyapps.io/atxn3-ko/. The code to create each app was adapted from publicly available code on GitHub (Hutch, 2021) and is linked at the top of each app. Raw lipid data was uploaded to the MassIVE repository: MassIVE MSV000096520. Further requests for data are available upon reasonable request of the corresponding authors.

Results

Robust lipid downregulation is observed in SCA3 human cerebella

In recent years, white matter atrophy has emerged as a progressive signature of many neurodegenerative diseases, including SCA3 (Casella et al., 2020; Festa et al., 2024; Tang et al., 2024). White matter is enriched for lipids, yet no unbiased studies of brain lipids have been conducted in SCA3. To characterize the lipidome of SCA3 patients compared to controls, we conducted liquid chromatography-mass spectrometry (LC-MS) on the disease-vulnerable cerebellum. Postmortem cerebella from four patients with SCA3 and three controls were acquired from the Michigan Brain Bank. Sex has been considered in this study to emphasize its importance as a biological variable. However, it is noted that sex does not influence the inheritance of SCA3. Data from combined positive and negative ion modes detected 347 unique lipids across four lipid classes, fatty acids (FA), glycerolipids (GL), glycerophospholipids (GP), and sphingolipids (SP), and 27 subclasses. The vast majority of differentially expressed lipids were downregulated in patients with SCA3 relative to controls, with 169 lipids downregulated and 41 lipids upregulated (Fig. 1A). Additionally, the magnitude of change was more drastic in downregulated lipids compared to upregulated lipids. Principal Components Analysis confirmed differing lipid profiles in SCA3 patients relative to controls (Supplementary Fig. 1A, 1B).

Figure 1. Broad lipid dysregulation uncovered in SCA3 postmortem patient cerebella.

(A) LC-MS lipidomics on human postmortem cerebella revealed 169 lipids downregulated and 41 lipids upregulated in SCA3 patients (n=4, 3F/1M) relative to controls (n=3, all M). Dashed lines represent the cutoff for significance, p < 0.05 and log2-fold change (log2-FC) ≥ 1.5 or ≤ −1.5. (B) Heatmap of detected lipids displaying abundance normalized to the average of controls: fatty acids (FA), glycerolipids (GL), glycerophospholipids (GP), and sphingolipids (GP). Depicted to the right of the heatmap is the number of lipid species dysregulated in each subclass (“Dys”) divided by the number detected (“Det”) in that subclass. Across most subclasses, lipids were reduced in SCA3 samples compared to controls (CTRL). (C) LION/web lipid pathway analysis revealed 18 upregulated and 3 downregulated pathways in SCA3 patients relative to controls, of which “ceramides” and “hexosylceramides” were the most significant, respectively. Dashed line represents the cutoff for significance (FDR q-value < 0.05). Size of dot reflects the number of significant lipids as a percentage of the total number of lipids annotated to each term (“Sig/Ann”).

Visualizing the expression of all detected lipids highlights consistent reductions in global lipid abundance across samples and classes (Fig. 1B). Lipids in the GP and SP classes are most affected in disease. For example, 37 of 78 (71%) phosphatidylcholines (PC) detected within the GP class were significantly changed, and 39 of 53 (74%) non-hydroxy-fatty acid sphingosine ceramides (Cer_NS), 9 of 12 (75%) sulfatides (SHexCer), and 34 of 40 (85%) non-hydroxy-fatty acid sphingosine hexosylceramides (HexCer_NS) detected within the SP class were significantly changed. While there are no myelin-specific lipids, myelin is enriched for hexosylceramide and its sulfated version, sulfatide (Montani, 2021; Norton and Poduslo, 1973; Palavicini et al., 2016).

To understand the downstream effects of lipid downregulation in SCA3, we conducted lipid ontology enrichment analysis through LION/web (Fig. 1C) (Molenaar et al., 2023; Molenaar et al., 2019; Ni et al., 2023). While lipid ontology resources are still in nascent stages, this provides a baseline for interpreting lipid alterations. The terms associated with downregulated lipids in SCA3 patients relative to controls included “hexosylceramides,” “headgroup with neutral charge,” and “fatty acid with 24 carbons.” The terms linked to upregulated lipids included “ceramides,” “saturated fatty acids,” and “lipid mediated signaling,” to name a few. The reduction in hexosylceramides (HexCer) matches the heatmap, while the increase in ceramides could be due to the upregulation in ceramide phosphates (CerP) in patients with SCA3. Alterations in headgroup charge (Bruce Alberts, 2002; Corradi et al., 2019; Deol et al., 2004; Israelachvili, 2011; Lira et al., 2021; Pan et al., 2012) and saturated fatty acid content (Bruce Alberts, 2002; Harayama and Riezman, 2018; Kucerka et al., 2011) are known to affect membrane structure and function, suggesting that alterations to the carefully balanced lipidome can have potentially deleterious consequences.

SCA3 mouse models recapitulate lipid downregulation and implicate a toxic gain-of-function mechanism

While human patient postmortem tissue provides critical relevance to human disease, it cannot capture the progression of disease-related changes over time. Therefore, we turned to mouse models to further investigate lipid changes, focusing on mechanistic pathways and progression. In this study, we used two mouse models of SCA3: the YACQ84 transgenic mouse model (Cemal et al., 2002) which overexpresses human ATXN3 with 84 CAG repeats (within the patient pathogenic range), and the Knock-in Q300 (KIQ300) model which carries a hyperexpanded 300 CAG repeats in mouse Atxn3 (Atxn3^Q300/Q6). YACQ84 mice show onset of disease as early as four weeks of age, when homozygous mice (Q84/Q84) have significantly lower weight and total locomotor and rearing activity compared to WT/WT controls (Schuster et al., 2023; Schuster et al., 2024). Initially described as KIQ82 but later selected for higher repeat expression through breeding, the KIQ300 mouse remains less characterized (Ramani et al., 2017; Schuster et al., 2023). To determine disease-relevant timepoints in these mice, we undertook a natural history study, recording the weight and motor behavior of mice as they aged from 4 to 48 weeks (Supplementary Fig. 2). Atxn3^Q300/Q6 mice demonstrated lower weight beginning at 20 weeks and a steep decline in survival at approximately 40 weeks (Supplementary Fig. 2A/B). Motor behavior was impaired beginning at 24 weeks on the open field assay, with reduced total locomotion and rearing activity in diseased mice relative to controls (Supplementary Fig. 2C/D). Motor coordination and balance on the accelerated rotarod were reduced beginning at 36 weeks of age (Supplementary Fig. 2E).

We selected an early- to mid-stage disease timepoint for cerebellar lipid investigations: 24 weeks in KIQ300 mice and 16 weeks in YACQ84 mice, respectively (Schuster et al., 2023; Schuster et al., 2024; Schuster et al., 2022b). To understand the contribution of ATXN3 loss-of-function, and in turn, mutant ATXN3 toxic gain-of-function, to lipid signatures, we also assessed Atxn3-Knock-out (KO) mice at 16 weeks of age (Reina et al., 2012; Schmitt et al., 2007). Atxn3-KO mice and cells derived from them are phenotypically normal (Reina et al., 2012; Schmitt et al., 2007). Studies on Atxn3-KO mouse brains from our lab and others indicate that ATXN3 knockout does not play a role in myelination, motor impairments, or neurodegeneration (Reina et al., 2012; Schuster et al., 2022a; Schuster et al., 2024; Schuster et al., 2022b). Therefore, we hypothesized a gain-of-function mechanism accounted for lipid reductions in SCA3.

In the cerebellum of 16-week-old YACQ84 mice, LC-MS analysis resulted in the identification of 491 unique lipids across 24 subclasses. Similar to the patient lipid profile, there was strong lipid downregulation, with 97 downregulated and 14 upregulated in Q84/Q84 mice relative to WT/WT controls (Fig. 2A). Principal Components Analysis highlighted these significant differences in lipid profiles (Supplementary Fig. 1C, 1D). The glycerophospholipid (GP) and sphingolipid (SP) classes were most significantly affected, with 15 of 41 (37%) phosphotidylserines (PS) and 13 of 50 (62%) phosphatidylgylcerols (PG) significantly altered in the GP class and 8 of 16 (50%) sphingomyelins (SM) significantly changed in the SP class (Fig. 2B). Ganglioside GM3, involved in various cellular processes including oligodendrocyte differentiation (Komura et al., 2017; Yim et al., 1994; Yim et al., 1991), was highly downregulated in Q84/Q84 mice, aligning with findings of reduced mature oligodendrocyte cell counts in Q84/Q84 mice relative to controls (Schuster et al., 2023; Schuster et al., 2022a; Schuster et al., 2024; Schuster et al., 2022b). Lipid ontology analysis revealed that downregulated lipids were associated with “above average lateral diffusion, “below average bilayer thickness,” and “below average transition temperature,” while upregulated lipids were associated with diacylglycerophosphoserines,” “glycerophosphoserines,” and “headgroup with negative charge” (Fig. 2C). Similar to the SCA3 patient results, these alterations suggest membrane destabilization, as would be expected with dramatic alterations to lipid content.

Figure 2. Cerebella from SCA3 transgenic and Knock-in mice recapitulate lipid downregulation and reveal a toxic gain-of-function mechanism.

LC-MS lipidomics was conducted on YACQ84 transgenic mice (A-C), Knock-in Q300 mice (D-F), and Atxn3-Knock-out mice (G-I). (A) Lipidomics on 16-week YACQ84 cerebella revealed 97 downregulated and 14 upregulated lipids in Q84/Q84 mice (n=4, 2F/2M) relative to WT/WT (n=3, 2F/1M). (B) The heatmap of all detected lipids demonstrates lipid downregulation across most subclasses in Q84/Q84 samples compared to WT/WT. (C) LION/web enrichment analysis revealed 16 upregulated and four downregulated pathways in Q84/Q84 mice relative to WT/WT controls, of which “diacylglycerophosphoserines” and “above average lateral diffusion” were the most significant, respectively. (D) Lipidomics on 24-week KIQ300 cerebella revealed 175 downregulated and 8 upregulated lipids in Atxn3^Q300/Q6 (n=6, 3F/3M) mice relative to Atxn3^Q6/Q6 (n=5, 2F/3M). (E) The heatmap of all detected lipids demonstrates that lipids were reduced across most subclasses in Atxn3^Q300/Q6 samples relative to Atxn3^Q6/Q6 samples, similar to the changes observed in YACQ84 mice. (F) Enrichment analysis via LION/web uncovered no upregulated and three downregulated pathways in Atxn3^Q300/Q6 mice compared to controls, of which “monoacylglycerophosphocholines” was the most significant. (G) Lipidomics on 16-week Atxn3-Knock-out cerebella revealed four downregulated and 70 upregulated lipids in Atxn3^−/− (n=5, 2F/3M) mice relative to Atxn3^Q6/Q6 (n=6, 2F/4M). (H) In contrast to the widespread downregulation observed in YACQ84 and KIQ300 mice, the heatmap of all detected lipids demonstrates lipid upregulation across most subclasses in Atxn3^−/− samples relative to Atxn3^Q6/Q6 samples. (I) LION/web enrichment analysis highlighted no downregulated and 11 upregulated pathways, of which “fatty acid with more than 3 double bonds” was the most significant. On the volcano plots, the dashed lines represent the cutoff for significance, p < 0.05 and log2-fold change (log2FC) ≥ 1.5 or ≤ −1.5. Heatmaps display all detected lipids, with the abundance normalized to the average of wildtype controls. Lipids are arranged by class: fatty acids (FA), glycerolipids (GL), glycerophospholipids (GP), and sphingolipids (GP). Depicted to the right of the heatmap is the number of lipid species dysregulated in each subclass (“Dys”) divided by the number detected in that subclass (“Det”). On pathway analysis plots, the dashed line represents the cutoff for significance (FDR q-value < 0.05). Size of dot reflects the number of significant lipids as a percentage of all lipids annotated to each term (“Sig/Ann”).

In the cerebellar lipid profile of 24-week-old KIQ300 mice, we identified 405 unique lipid species, encompassing 26 lipid subclasses. Consistent with human and YACQ84 datasets, downregulation dominated, with 175 lipids downregulated and 8 upregulated in Atxn3^Q300/Q6 mice relative to Atxn3^Q6/Q6 controls (Fig. 2D). Principal Components Analysis showed segregation of genotypes (Supplementary Fig. 1E, 1F). The most affected lipid classes were again GP and SP lipids. In the GP class, 18 of 31 (58%) phosphatidylserines (PS), and 10 of 10 (100%) lysophosphatidylcholines (LPC) were significantly altered. In the SP class, 22 of 32 (69%) hexosylceramides (HexCer_NS), 6 of 7 (86%) sulfatides (SHexCer), and 12 of 16 (75%) sphingomyelins (SM) were dysregulated (Fig. 2E). Pathway analysis revealed no significantly upregulated pathways and three significantly downregulated pathways: “membrane component,” “above average bilayer thickness,” and “monoacylglycerophosphocholines” (PC species) (Fig. 2F), again suggesting broad membrane disruptions in Atxn3^Q300/Q6 mice relative to Atxn3^Q6/Q6 controls.

Following lipidomics in 16-week-old KO mouse cerebella, we quantified 354 unique lipids across 26 lipid subclasses. In contrast to YACQ84 and KIQ300 mice, we observed lipid upregulation in Atxn3^−/− mice relative to Atxn3^Q6/Q6 controls, with 70 lipids upregulated and 4 downregulated (Fig. 2G). The magnitude of lipid change was smaller than SCA3 mouse datasets, with Principal Components Analysis revealing overlapping lipid profiles in Atxn3^Q6/Q6 and Atxn3^−/− mice (Supplementary Fig. 1I, 1J). GP- and SP-class lipids were minimally altered in KO mice (Fig. 2H). Lipid pathway analysis revealed no significantly downregulated pathways. The pathways associated with upregulated lipids included unsaturated fatty acids (“fatty acid with more than three double bonds,” “polyunsaturated fatty acid”), “diacylglycerols” (DG), and low/below average “transition temperature” (Fig 2I). Notably, the upregulation of unsaturated fatty acids in KO mice is interesting when compared to the upregulation of saturated fatty acids in human SCA3 patients. In summary, YACQ84 and KIQ300 mice exhibit reduced lipid content, consistent with observations in tissue from SCA3 patients, whereas KO mice show increased lipid content. Since KO mice are phenotypically normal, a slight increase in lipid content seems to be well-tolerated. These findings suggest that a gain-of-function mechanism of mutant ATXN3 is the primary factor contributing to lipid downregulation in SCA3.

Conserved lipid changes between SCA3 patients and mouse models highlight candidates for mechanistic investigation

The lipids significantly altered in YACQ84, KIQ300, and human datasets and not significantly altered in the KO dataset represent those most relevant to toxic gain-of-function disease pathophysiology. To this end, we investigated the overlap between the SCA3 patient and mouse model datasets. Evaluating the differentially expressed lipids unique to each cohort, we found 81 in Q84/Q84 mice relative to WT/WT, 129 in Atxn3^Q300/Q6 mice relative to Atxn3^Q6/Q6, and 169 in human SCA3 patients relative to controls (Fig. 3A). Of note, this study detected the most highly expressed lipids in the cerebellum, suggesting that future studies probing more lipids may uncover a wider signature of shared lipid dysregulation. While it is difficult to connect individual lipid species with defined cellular and molecular alterations, outlining the altered lipids provides a more granular understanding of lipid vulnerability in SCA3.

Figure 3. Lipid dysregulation conserved across human and mouse datasets reveals gain-of-function candidates for future mechanistic study.

(A) Venn diagram depicting the number of significantly dysregulated lipids unique to each dataset – 81 in Q84/Q84 mice relative to WT/WT, 129 in Atxn3^Q300/Q6 mice relative to Atxn3^Q6/Q6, and 169 in SCA3 patients relative to controls. (B) YACQ84, KIQ300, and human SCA3 patients share five significantly altered lipids, four of which were reduced in all three datasets. If detected, log2-FC values are displayed for Atxn3-KO mice for reference. (C) YACQ84 and KIQ300 mice share 19 significantly altered lipids, 18 of which were reduced in both datasets. These lipids represent mouse-specific changes or those that were altered too early to detect in human samples. Human and Atxn3-KO log2-FC values for each lipid are displayed if detected. (D) KIQ300 mice and SCA3 patients, which both endogenously express mutant ATXN3, share 30 dysregulated lipids, 27 of which were reduced in both datasets. 25 of these displayed a more robust fold-change in patients compared to mice. Log2-FC values for YACQ84 and Atxn3-KO mice are displayed if detected. (E) YACQ84 mice and SCA3 patients, which both express human ATXN3, share six significantly altered lipids, four of which are reduced in both datasets. If detected, log2-FC values are displayed for KIQ300 and Atxn3-KO mice. The shaded area between −0.58 and 0.58 log2-FC represents the region in which lipids would not be significant due to log2-FC abundance (cutoff of 1.5-FC). Note that lipids outside this range may still not be significant if the p-value was greater than 0.05.

Lipid alterations recapitulated by both mouse models represent the most relevant species for future mechanistic investigation. Four of the five lipids differentially expressed in all three datasets were reduced in disease: acylcarnitine (ACar) 18:0, phosphatidylcholine (PC) 42:2, sphingomyelin (SM) d34:1, and SM d42:2 (Fig. 3B). Of these, two were not altered in Atxn3-KO mice (ACar 18:0 and SM d34:1), one was significantly increased in Atxn3-KO mice (PC 42:2), and one (SM d42:2) was not detected in the Atxn3-KO dataset. Since glycerophospholipids and sphingolipids were abundantly detected and dysregulated within this study, PC 42:2 and SM d34:1 are likely the most accessible toxic gain-of-function lipids for future investigation.

Inspecting lipids significantly altered only in SCA3 mouse models provides a longer list of targets, with 19 in total (Fig. 3C). These lipids provide insight on species-specific changes or early-stage disease changes that are not captured in end-stage human tissue. Additionally, they validate the lipid alterations in YACQ84 mice that are not solely due to overexpression of human ATXN3. All the lipids belong to the GP or SP classes. Of the 18 lipids reduced in diseased mice relative to controls, 12 were not detected in the Atxn3-KO dataset. The remaining six were not significantly altered or were significantly increased in Atxn3-KO mice, emphasizing a mutant ATXN3 toxic gain-of-function mechanism underlying downregulation of lysophosphatidylcholine (LPC) 24:0/0:0, phosphatidylcholine (PC) 40:2, phosphatidylethanolamine (PE) 18:1_20:4, phosphatidylglycerol (PG) 16:0_16:0, non-hydroxy-fatty acid sphingosine ceramide (Cer_NS) d18:1_22:2, and sphingomyelin (SM) d32:1.

The YACQ84 and KIQ300 mouse models differ in the expression level and polyQ repeat size of mutant ATXN3. More specifically, YACQ84 mice express four copies of human mutant ATXN3 with 84 CAG repeats, while KIQ300 mice express one copy of mouse Atxn3 with a hyper-expanded 300 CAG repeats. To address these differences, we investigated lipid alterations in systems endogenously expressing one copy of mutant ATXN3 – KIQ300 mice and patients with SCA3. With expression level normalized, we defined the changes shared at two ATXN3 polyQ repeat sizes, 69.5 (average) in patients with SCA3 and 300 in Knock-in mice. Of the 30 lipids significantly altered in both datasets, 27 were reduced in both, with almost all (25) displaying a higher magnitude log2-fold change in patients compared to mice (Fig. 3D). Removing the lipids not detected in the Atxn3-KO dataset (11), the remaining 16 were not altered or were significantly increased in Atxn3-KO mice. These lipids, which span the GL, GP, and SP classes, represent those altered by a toxic gain-of-function mechanism by endogenously expressed mutant ATXN3, irrespective of the size of the repeat expansion.

KIQ300 mice recapitulate another feature of SCA3 disease that YACQ84 mice do not: expression of one copy of normal, non-expanded ATXN3 instead of two. To determine the effect of reduced levels of normal ATXN3 on the SCA3 lipidome, we compared significantly altered lipids in patients with SCA3 and Atxn3-KO and KIQ300 mice (data not shown). We found only two examples of lipids possibly affected by reduced levels of normal ATXN3. Phosphatidylethanolamine (PE) 16:0_18:1 was significantly downregulated in KIQ300 mice and Atxn3-KO mice, and diglyceride (DG) 18:0_22:6 was significantly upregulated in patients with SCA3 and Atxn3-KO mice. There were no lipids significantly altered in the same direction in all three datasets. This suggests that ATXN3 loss-of-function has a limited effect on lipid regulation in SCA3. Finally, to isolate the effect of human ATXN3 on the SCA3 lipidome, we considered the six lipids significantly altered in patients with SCA3 and YACQ84 mice (Fig. 3E). Three were not detected in the Atxn3-KO dataset, and the remaining three were not significantly altered in Atxn3-KO mice. Interestingly, the ganglioside GM3 d36:1 was robustly reduced in YACQ84 mice but slightly increased in patients with SCA3, suggesting that this was a mouse-specific or early-stage disease alteration. All that remains are phosphatidylcholine (PC) 28:0 and phosphatidylethanolamine (PE) 18:0_20:4, which represent conserved lipids reduced due to human mutant ATXN3 expression.

In sum, we can compare the lipids conserved between patients with SCA3 and mouse models to identify the most promising targets for future mechanistic advances, primarily those that are due to a toxic gain-of-function of mutant ATXN3.

Lipid downregulation is exacerbated with disease progression in a Knock-in SCA3 mouse model

To investigate the progression of lipid dysregulation in SCA3 mice, we conducted unbiased LC-MS lipidomics in end-stage Knock-in mice (57 weeks of age) and compared our results to those from early-stage Knock-in mice (24 weeks of age). In 57-week-old mice, we identified 406 unique lipid species spanning 26 lipid subclasses. Similar to the aforementioned SCA3 datasets, we found expansive lipid downregulation in 57-week-old Atxn3^Q300/Q6 mice relative to Atxn3^Q6/Q6 controls. Specifically, 159 lipids were downregulated and only 5 were upregulated (Fig. 4A). Principal Components Analysis maintained clear genotype-specific differences in lipid profiles (Supplementary Fig. 1G, 1H). A closer examination revealed that the GP and SP classes again had the highest proportions of dysregulated lipids (Fig. 4B). For instance, 23 of 32 (72%) of non-hydroxy-fatty acid sphingosine hexosylceramides (HexCer_NS) were significantly altered. Lipid pathway analysis corroborated these findings, showing significant downregulation of the LION terms “diacylglycerophosphocholines” (PC species) and “hexosylceramides” (Fig. 4C). “Ceramides” and “very high transition temperature” were among the upregulated pathways, consistent with the human and YACQ84 datasets, respectively. Therefore, 57-week-old mice corroborate pathway alterations in our other SCA3 datasets.

Figure 4. End-stage KIQ300 mice highlight lipids progressively decreased in SCA3.

(A) LC-MS lipidomics of cerebella from 57-week-old KIQ300 mice revealed 159 downregulated and 5 upregulated lipids in Atxn3^Q300/Q6 (n=6, 3F/3M) mice relative to Atxn3^Q6/Q6 (n=6, 3F/3M). Dashed lines represent the cutoff for significance, p < 0.05 and log2-fold change (log2FC) ≥ 1.5 or ≤ −1.5. (B) Heatmap of all detected lipids, with the abundance normalized to the average of wildtype controls. Lipids are arranged by class: fatty acids (FA), glycerolipids (GL), glycerophospholipids (GP), and sphingolipids (GP). Depicted to the right of the heatmap is the number of lipid species dysregulated in each subclass (“Dys”) divided by the number detected in that subclass (“Det”). As in YACQ84 and 24-week-old KIQ300 mice, lipid downregulation was observed across most lipid subclasses. (C) Lipid enrichment analysis uncovered seven upregulated and four downregulated pathways in Atxn3^Q300/Q6 mice relative to controls, of which “PE 16:0/18:0” and “diacylglycerophosphocholines” were the most significantly altered, respectively. Dashed line represents the cutoff for significance (FDR q-value < 0.05). Size of dot reflects the number of significant lipids as a percentage of all lipids annotated to each term (“Sig/Ann”). (D) Venn diagram depicting significantly dysregulated lipids in 24-week- and 57-week-old KIQ300 mice. Common to both timepoints were 81 downregulated lipids and three lipids that were increased at 24 weeks and decreased at 57 weeks. (E) The three lipids with opposite directionality are PI 18:0_20:4, PS 40:6, and SHexCer d43:2, all significantly reduced in 24-week and significantly increased in 57-week Atxn3^Q300/Q6 compared to Atxn3^Q6/Q6 controls. The shaded area between −0.58 and 0.58 log2-FC represents the region in which lipids would not be significant due to log2-FC abundance (cutoff of 1.5-FC). (F) 81 lipids were significantly reduced at both timepoints, with 52 displaying a higher magnitude log2-FC at 57 weeks compared to 24 weeks. (G) 80 lipids are significantly altered at 57 weeks (75 down, 5 up) but not 24 weeks. Of note, EtherPE 18:2e_18:1 RT:13.137, which was significantly reduced at 57 weeks, was not detected in the 24-week dataset.

To understand the progression of lipid dysregulation, we compared the significantly altered lipids in 24-week and 57-week KIQ300 datasets (Fig. 4D). Shared at both timepoints were 81 downregulated lipids and 3 lipids upregulated at 24 weeks but decreased at 57 weeks, suggesting differential regulation across the disease course (Fig. 4E). Of the 81 lipids downregulated at both timepoints, 52 were exacerbated with disease progression, with a higher-magnitude log2 fold-change at 57 weeks compared to 24 weeks (Fig. 4F). Finally, 80 lipids were altered at 57 weeks only, representing a cascade of lipid dysregulation unique to end-stage disease (Fig. 4G). Taken together, the downregulated lipids in Fig. 4F–G, which mostly belong to the glycerophospholipid and sphingolipid classes, represent potential candidates for early intervention to prevent more severe decline in later disease stages.

To verify the signatures of end-stage disease identified in mice, we compared dysregulated lipids in KIQ300 mice and patients with SCA3 (Supplementary Fig. 3A). Of the 17 significantly altered lipids common to SCA3 patients, 24-week KIQ300 mice, and 57-week KIQ300 mice, we noted that all were reduced in disease, emphasizing a shared signature of reduced lipid content in SCA3 (Supplementary Fig. 3B). Additionally, the conserved lipids appeared to be exacerbated by disease progression, with patients typically displaying the most severe reductions, followed by 57-week-old mice. Of the 17 lipids altered only at late stages of disease (in 57-week-old mice and SCA3 patients), 11 were reduced in both datasets (Supplementary Fig. 3C). The lipids were similarly distributed across the glycerophospholipids and sphingolipid classes, with patients displaying more robust downregulation in 10 of 11 lipids. These shared lipids support the utility of 57-week-old KIQ300 mice for end-stage disease mechanistic investigations.

Lipid signatures provide evidence of myelin alterations in SCA3 patients and mice

Given the well-established white matter disruptions in SCA3 patients and mouse models, we sought to specifically compare the lipids enriched in myelin across our datasets. Lipids comprise approximately 80% the dry weight of myelin (Boggs and Moscarello, 1978; Montani, 2021; O’Brien and Sampson, 1965; Palavicini et al., 2016; Poitelon et al., 2020; Quarles, 1999). As such, myelin is particularly sensitive to lipid dysregulation. We considered the typical composition of myelin, which is enriched for cholesterol and glycolipids. Myelin has a molar ratio of 40% cholesterol: 40% phospholipids: 20% glycolipids, compared to 25%: 65%: 10% in other biological membranes (Fig. 5A) (Montani, 2021; O’Brien, 1965; Poitelon et al., 2020; Schmitt et al., 2015).

While cholesterol was not automatically annotated in our dataset, we manually searched for cholesterol at m/z = 369.35 within the mass spectra and found significantly reduced cholesterol in patients with SCA3 patients and mice compared to controls (Fig. 5B). Atnx3-KO mice showed the opposite trend, with significantly increased cholesterol compared to wildtype controls.

To contextualize the reduction in cholesterol in the human dataset, we conducted absolute quantification of the sterols and CoAs of the cholesterol biosynthesis pathway (Supplementary Table 1). However, we found no significant changes in total (hydrolyzed sample preparation) or free (non-hydrolyzed sample preparation) sterol precursors of cholesterol, acetyl-CoA and HMG-CoA, and 24S-hydroxycholesterol (the main metabolite of cholesterol). Therefore, we did not conduct the same studies in SCA3 mice, instead turning our focus to other lipids enriched in myelin.

The major glycolipids characteristic of myelin are hexosylceramide and sulfatides (Montani, 2021; Palavicini et al., 2016). We found a significant reduction in hexosylceramides (HexCer) in 24-week and 57-week KIQ300 mice compared to controls (Fig. 5C). HexCer was trending down in patients with SCA3 but not statistically significant. Importantly, it was not significantly altered in Atxn3-KO mice, suggesting a mutant ATXN3 toxic gain-of-function mechanism. Sulfatide (SHexCer), a derivative of hexosylceramide, was significantly lower in patients and at both timepoints in KIQ300 mice compared to controls (Fig. 5D). SHexCer showed the opposite trend of significant upregulation in Atxn3-KO mice. Interestingly, HexCer and SHexCer subclasses were not significantly altered in YACQ84 mice, representing a deviation from recapitulating human disease.

In short, we define a mutant ATXN3 toxic gain-of-function mechanism underlying the downregulation of myelin-enriched lipids in SCA3. Since lipid metabolism disorders frequently demonstrate myelin abnormalities (Chrast et al., 2011), our findings suggest that lipid downregulation may be related to SCA3 myelin deficits.

Discussion

Previous work in the SCA3 field establishes progressive cerebellar white matter atrophy in SCA3 patients and preataxic mutation carriers (Arruda et al., 2020; Chandrasekaran et al., 2022; Faber et al., 2021; Ferreira et al., 2024; Rezende et al., 2018; Wu et al., 2017). However, little is known about how lipids, the major constituent of white matter, are altered in the SCA3 brain. Therefore, we conducted the first unbiased study of brain lipids in SCA3, focusing on the disease-vulnerable cerebellum of SCA3 patients and mouse models. We provide evidence of robust and progressive lipid downregulation in SCA3 due to mutant ATXN3 toxic gain-of-function. Pathway analysis revealed lipid changes that are associated with increased bilayer thickness and decreased membrane fluidity in patients with SCA3, transgenic SCA3 mice, and end-stage Knock-in mice. We enumerate the lipids conserved between mice and humans and those that are progressively decreased over the course of disease, establishing candidates for future mechanistic and biomarker identification. Finally, we show an impairment of myelin-enriched lipids including cholesterol and the glycolipids hexosylceramide and sulfatide. This study identifies novel, gain-of-function alterations to lipids in the SCA3 cerebellum, establishing the groundwork for further mechanistic and therapeutic investigations.

In the present study, we uncovered major lipid downregulation in the cerebellum of SCA3 patients and mice, while Atxn3-Knock-out mice displayed a relatively upregulated lipid profile. These opposing trends suggest an interesting mechanism at play that remains to be elucidated. One possibility is that endogenous ATXN3 inhibits a positive regulator of lipid levels, such that when ATXN3 is knocked out, this regulator is disinhibited, and lipid levels increase. In disease, mutant ATXN3 may gain a toxic increased affinity for this positive regulator, further inhibiting lipid homeostasis and resulting in a drop in lipid levels below the basal range. The identity of these positive regulators remains unclear, though a recent study proposed that ATXN3 can regulate the stability of 3 hydroxy-3-methylglutaryl-CoA reductase (HMGCR) (Stahl et al., 2023), the rate-limiting enzyme of cholesterol biosynthesis. In vivo support for this interaction is lacking, necessitating future studies to better understand the role of ATNX3 in lipid homeostasis. Our study importantly establishes that increased lipid levels (at least to the extent we observed) are well tolerated, while reduced lipid levels may be pathogenic.

While this is the first unbiased study of lipids in the SCA3 brain, other studies have conducted metabolomics and lipidomics on patient and mouse blood biofluids. Toonen and colleagues (2018) studied lipids in the plasma of hemizygous YACQ84 (Q84/WT) mice, finding reductions in phosphoserines (PS), sulfatides (SHexCer), and lysophosphatidylinositols (LPI) in aged mice. We established broad downregulation of these subclasses in the cerebellum of patients with SCA3, homozygous YACQ84 mice, and Knock-in mice. In particular, the conserved reduction of sulfatide in the cerebellum is striking. Sulfatide is known to play a role in myelin biogenesis and maintenance (Blomqvist et al., 2021; Montani, 2021; Palavicini et al., 2016), suggesting a link between the sulfatide decreases in the cerebellum and the myelin thinning observed in our prior studies (Schuster et al., 2023; Schuster et al., 2024; Schuster et al., 2022b). However, it remains to be determined if lipid dysfunction is a cause or consequence of cell type-specific deficits in SCA3. For example, one could harness SCA3 conditional mouse models to turn off expression of mutant ATXN3 in cell types of interest, such as oligodendrocytes, and interrogate lipidomic changes. In sum, in conjunction with mechanistic studies, validation of lipid biomarkers in SCA3 may provide additional methods for monitoring disease progression, which is critical for future interventional studies.

Contextualizing SCA3-related lipid changes within the polyglutamine disease field reveals a shared pattern of lipid dysfunction. Among the polyglutamine diseases, lipid deficits are most well-described in Huntington’s disease (HD). Similar to the changes we observed in SCA3, the profile of sphingolipids and phospholipids is altered in disease-vulnerable brain regions of HD animal models and postmortem patient samples (Farzana et al., 2023; Hunter et al., 2018; Phillips et al., 2022; Yilmaz et al., 2025). More specifically, phosphatidylserine (PS) 22:6, a lipid decreased in YACQ84 and 24-week-old KIQ300 mice, was also reduced in the caudate nucleus of postmortem HD patients (Hunter et al., 2021). Similarly, sulfatide (SHexCer) d18:1_24:0, d18:1_24:1, and d36:1, reduced in 24-week-old KIQ300 mice and human patients with SCA3, were also reduced in Huntington’s disease postmortem samples (Hunter et al., 2018; Phillips et al., 2022). Cholesterol alterations have also been implicated in Huntington’s disease, similar to the reductions we reported across SCA3 mouse models and human samples (Kacher et al., 2022). Indeed, a recent review discussed strategies for targeting cholesterol biosynthesis in HD animal models for therapeutic benefit (Valenza et al., 2023). These studies and more suggest a conserved mechanism of polyglutamine-induced lipid dysfunction in SCA3 and HD and highlight potential therapeutic avenues for SCA3.

Altogether, this study characterizes lipid dysregulation in SCA3 brains for the first time and paves the way for future comparative studies via an accessible lipid database. Through interactive data visualization websites (Hutch, 2021), users can easily select lipids of interest and gain information on the fold-change and p-value observed in SCA3 humans and YACQ84, KIQ300, and Atxn3-KO mice. This may prove useful for future comparisons between our datasets and studies of SCA3 biofluids, or even brain and biofluid samples from other neurodegenerative diseases, improving our collective understanding of lipid dysregulation in diseases with white matter abnormalities and stimulating the development of novel therapeutics.

Supplementary Material

Supplementary material is available online.

Glossary

Fatty acids (FA):

ACar

Acylcarnitines

FA

Fatty acids

FAHFA

Fatty acid esters of hydroxy fatty acids

Glycerolipids (GL):

DG

Diacyl/alkylglycerides (diglycerides)

DGTS

Diacylglyceryl-trimethylhomoserine

MGDG

monogalactosyldiacylglycerol

TG

Triacyl/alkylglycerides (triglycerides)

Glycerophospholipids (GP):

BMP

Bis[monoacylglycero]phosphates

CL

Cardiolipins

EtherOxPE

Ether-linked oxidized Phosphatidylethanolamine

EtherPC

Ether-linked Phosphatidylcholine

EtherPE

Ether-linked Phosphatidylethanolamines

LPA

Lysophosphatidic acid

LPC

Lysophosphatidylcholines

LPE

Lysophosphatidylethanolamine

LPG

Lysophosphatidylgylcerols

LPI

Lysophosphatidylinositols

LPS

Lysophosphatidylserines

OxPI

Oxidized phosphatidylinositol

PA

Phosphatidic acids

PC

Phosphatidylcholines

PE

Phosphatidylethanolamines

PG

Phosphatidylglycerols

PI

Phosphatidylinositols

PMeOH

Phosphatidylmethanol

PS

Phosphatidylserines

Sphingolipids (SP):

Cer_NS

Non-hydroxy-fatty acid sphingosine ceramides

CerP

Ceramide-phosphates

GM3

monosialodihexosylganglioside

HexCer_NS

Non-hydroxy-fatty acid sphingosine hexosylceramides

SHexCer

Sulfatides

SM

Sphingomyelins