Cerebral cortical functional hyperconnectivity in a mouse model of spinocerebellar ataxia type 8 (SCA8)

Abstract

Spinocerebellar Ataxia Type 8 (SCA8) is an inherited neurodegenerative disease caused by a bidirectionally expressed CTG•CAG expansion mutation in the ATXN8 and ATXN8OS genes. While SCA8 patients have motor abnormalities, patients may also exhibit psychiatric symptoms and cognitive dysfunction. It is difficult to elucidate how the disease alters brain function in areas with little or no degeneration producing both motor and cognitive symptoms. Using transparent polymer skulls and CNS-wide GCaMP6f expression, we studied neocortical networks throughout SCA8 progression using wide-field Ca²⁺ imaging in a transgenic mouse model of SCA8. Compared to wild-type controls, neocortical networks in SCA8+ mice were hyperconnected globally, which leads to network configurations with increased global efficiency and centrality. At the regional level, significant network changes occurred in nearly all cortical regions, however mainly involved sensory and association cortices. Changes in functional connectivity in anterior motor regions worsened later in the disease. Near perfect decoding of animal genotype was obtained using a generalized linear model based on canonical correlation strengths between activity in cortical regions. The major contributors to decoding were concentrated in the somatosensory, higher visual and retrosplenial cortices and occasionally extended into the motor regions, demonstrating that the areas with the largest network changes are predictive of disease state.

1. Introduction

Nucleotide repeat expansion mutations cause several neurological disorders, including Huntington’s Disease (HD), Myotonic Dystrophy Type 1 (DM1), and many of the spinocerebellar ataxias (SCAs; Castelli et al., 2021). Repeat expansion mutations are variable in size and for most disorders become both pathological and genetically unstable at a disease-specific repeat length threshold (Moseley et al., 2000; Moseley et al., 2006). Spinocerebellar Ataxia Type 8 (SCA8) is an autosomal dominant inherited disease caused by a bidirectionally expressed nucleotide expansion mutation in the ATXN8 and ATXN8OS genes. Onset of SCA8 in humans most frequently occurs in mid-adulthood and symptoms become progressively worse throughout the disease. There is a high degree of reduced penetrance with the SCA8 mutation, although most affected patients have repeat expansions >70 CTG•CAGs (Perez et al., 2021). Clinical imaging reveals cerebellar atrophy and Purkinje cell loss in both the hemispheres and vermis, and mild brain stem atrophy. Neocortical atrophy is found, but only in a minority of cases (Cleary, 2001; Day et al., 2000; Ikeda et al., 2000; Juvonen et al., 2000; Lilja et al., 2005).

At the molecular level, the CTG•CAG mutation is transcribed into both CAG and CUG expansions in RNAs. The CAG expansion RNAs undergo repeat-associated non-ATG (RAN) translation which produce mutant polyserine, polyalanine, and polyglutamine expansion proteins. These homopolymeric expansion proteins accumulate in cells as intranuclear inclusions and alter normal cellular function. Production of these RAN proteins can trigger apoptosis and lead to toxic gain of function effects. The CUG expansion RNAs in SCA8 and DM1 can bind to and sequester muscleblind-like protein 1 (MBNL1), an RNA binding protein critical for RNA splicing regulation of other transcripts (Daughters et al., 2009; Goodwin et al., 2015; Sta Maria et al., 2021). In SCA8, CUG expansion RNAs cause increased expression of CUG RNA binding protein 1 (CUG-BP1) and reduced MBNL1 that regulates the CNS target, GABA-A transporter 4 (GAT4/GABT4). Increased levels of GABT4 are found in SCA8 mice and human patient tissue. These findings have a functional correlate in SCA8 mice of increased cerebellar neural responses to stimulation (Daughters et al., 2009; Moseley et al., 2006), suggestive of decreased GABAergic inhibitory tone. Therefore, RAN protein intranuclear inclusions and RAN protein sequestration gain-of-function profoundly alter brain function.

Phenotypically, SCA8 is characterized by both motor and cognitive symptoms. In the motor domain, symptoms include unstable gait, dysarthria, nystagmus, but additional symptoms can also manifest (Cleary, 2001). In the cognitive domain, studies including self-reported cognitive symptoms described SCA8 patients with personality changes, mood disturbances, anxiety, and depression (Zeman et al., 2004). More recent work studying cognitive decline in SCA8 patients found attention and information processing deficits, including reduced detection of visual targets and reduced performance in the Stroop Color/Word Interference test (Lilja et al., 2005). Additional deficits are found in executive function, and verbal tasks; however, memory was largely unaffected. These findings suggest SCA8 results in brain-wide dysfunction.

To aid in understanding this disorder, a transgenic mouse model of SCA8 expressing a human derived bacterial artificial chromosome (BAC) with the pathogenic SCA8 genes was created. These BAC transgenic mice, which contain 115 CAG•CTG repeats, recapitulate several aspects of the human disease including inheritance pattern, cellular dysfunction, and motor symptoms (Moseley et al., 2006). Importantly, this SCA8 mouse has no obvious neuron loss or gliosis in regions of the neocortex. Also, the mouse model of SCA8, unlike human SCA8, does not exhibit Purkinje cell death (Moseley et al., 2006), possibly due to the shorter lifespan of the mouse. However, both SCA8 patients and the mouse reveal several other types of brain-wide pathologies. For example, RAN protein aggregates are found in the cerebellum, motor cortex, brain stem, and white matter tracts (Ayhan et al., 2018). Abnormalities in the cerebellum of SCA8 mice include vacuole formation, and demyelination of the cerebellar white matter regions and axonal degeneration (Ayhan et al., 2018). Mirroring human disease, axonal degeneration is accompanied by a reduction in mature oligodendrocytes and astrogliosis in the cerebellar white matter. Despite these cellular and molecular abnormalities in the SCA8 mouse, the overall structure and architecture of the brain show no overt abnormalities (Moseley et al., 2006). These molecular pathology and clinical findings suggest that neocortical dysfunction in SCA8 is likely, as observed in other spinocerebellar ataxias (Bares et al., 2011; Bürk et al., 2003; Duarte et al., 2016).

In addition to direct pathology, cortical dysfunction in SCA8 could also arise from disrupted input. The cerebellum has extensive reciprocal connections with the neocortex through the cerebello-thalamo-cortical and cortico-ponto-cerebellar pathways (Fujita et al., 2020; Henschke and Pakan, 2020; Suzuki et al., 2012). This long-range cerebello-cortical loop could provide a neural substrate for propagation of cerebellar pathophysiology in SCA8 to neocortical networks. Furthermore, the cerebellum can have a profound impact on intra-neocortical oscillations and synchrony which can be dependent on behavioral context (Lindeman et al., 2021; Popa et al., 2013). Therefore, cortical changes in SCA8 could manifest as changes in functional connectivity (FC), as observed in other neurological disorders (Bernhardt et al., 2015; Harrington et al., 2015; Paldino et al., 2017; Vasilkovska et al., 2023). While FC has not been well investigated in SCA8, one study of SCA6 and 8 patients found smaller regional activations in the cerebellum and striatum when performing a motor task (Bares et al., 2011). Therefore, we hypothesize that neocortical FC will be abnormal in SCA8 and that the dysfunction can be driven by both cerebellar as well as direct neocortical pathology.

To address this hypothesis, we used neocortex-wide Ca²⁺ imaging to investigate changes in neocortical processing in a mouse model of SCA8. We find that the functional segmentation of the neocortex in SCA8+ mice is largely unchanged compared to non-transgenic controls (Moseley et al., 2006). However, FC analysis revealed hyperconnectivity and stronger connections across atlas regions both before and after symptoms developed in SCA8 transgenic mice. These connectivity changes produced neocortical networks with more clustering, increased efficiency, and more network communities. SCA8 networks showed localized changes in the posterior sensory and sensory integration areas of the neocortex, in addition to the motor regions, which could be used to decode animal genotypes using a generalized linear model. These data suggest that neocortical processing is fundamentally altered prior to symptom onset in SCA8.

2. Methods

2.1. Sex as a biological variable

Our study examined both male and female mice. While a difference in age of onset was noted between sexes, changes in cortical networks for SCA8+ mice were similar, and sexes were combined for analyses.

2.2. Animal model and surgical procedures

Transgenic mice (FVB) expressing the human SCA8 repeat expansion (SCA8+, 5 female, 2 male; Fig. 1A, top) and non-transgenic FVB control animals (NTC; 5 female, 4 male) were used for this wide-field cortical Ca²⁺ imaging study (Moseley et al., 2006). Briefly, mice 76 ± 23 (mean ± SD) days of age were anesthetized with isoflurane (5 % induction; 1–2 % maintenance) and implanted with transparent polymer skulls, “See-Shells”, and titanium headplates as previously described, including analgesia, monitoring, craniotomy, and post-operative care (Cramer et al., 2023; Ghanbari et al., 2019; Nietz et al., 2023; West et al., 2022; West et al., 2024). The craniotomy and polymer skull provided access to a large region of the dorsal neocortex. Images were taken of the craniotomy and cranial landmarks prior to skull removal. Immediately post-surgery, prior to recovery from anesthesia, animals were injected retro-orbitally with virally encoded GCaMP6f (AAV-PHP-eb-hSyn-GCaMP6f; 150 μL; titer: 3.81 × 10¹²–7 × 10¹³ gc/mL) which crosses the blood-brain barrier (Chan et al., 2017; Yardeni et al., 2011). Post-surgery, mice were housed on a 12-h reverse light-dark cycle and allowed 2 or more weeks (28 ± 12 days) recovery for sufficient viral expression and habituation to head-fixation on our treadmill. Post-experimental immunohistology for GCaMP6f expression in selected animals showed pan-neuronal infection of cells across all neocortical regions (Fig. 1B, left; Supplementary Fig. 2) and neocortical cell layers (Fig. 1B, right).

Fig. 1.

Viral GCaMP expression and wide-field Ca²⁺ imaging paradigm in SCA8+ mice. A) Schematic showing the expansion mutation in the human SCA8 transgene (top). Experimental setup showing Ca²⁺ imaging paradigm with animal head-fixed above a freely moving treadmill (bottom). B) Example coronal images of the neocortex showing broad expression of the Ca²⁺ sensor GCaMP6f after retro-orbital viral delivery (left; scale bars 1 mm). Example confocal image stacks showing GCaMP6f expression throughout the neocortical layers (right; scale bars 200 μm). C) Flow chart showing the imaging timeline across SCA8 disease progression and established analysis epochs (top) and a line graph showing animal weight (% maximum) across weeks of imaging used to empirically define SCA8 onset (black line - SCA8+; gray line - NT control; mean ± SD; week 0 - SCA8 onset). D, top) Example image of surgical craniotomy with anatomical landmarks (brown - olfactory bulb base/inferior cerebral vein & sagittal suture; blue - bregma & lambda; ruler ticks/scale bar 1 mm). D, bottom) Image of surgical craniotomy aligned to the Allen CCF using defined anatomical landmarks (bottom; blue - Allen CCF outline; orange - Allen cranial landmarks; purple - aligned surgical cranial landmarks). E) Surgical (left) and neocortical imaging field of view (right) and their associated masks (dark blue) used for atlas alignment. F) Example image showing final alignment of Allen CCF to the Ca²⁺ imaging field of view (E-F scale bars 1 mm).

2.3. Habituation and experimental setup

Mice were habituated to head-fixation and the disc treadmill over three sessions. In each session, mice were allowed to explore the treadmill freely for 5 min followed by increasing head-fixation time (5, 10, and 20 min) over 3 days. If mice still showed signs of stress or discomfort during head-fixation, additional habituation sessions were given. As age of onset can vary considerably in SCA8 (167 ± 47 days; see Supplementary Fig. 1), imaging began around postnatal day 100 (103 ± 25 days) and mice were imaged weekly or biweekly up to 12 weeks post-onset or death (Moseley et al., 2006). Imaging sessions took place during the animal’s dark phase and consisted of ~10–12 trials (5.5 min long) per mouse per day. Imaging sessions were performed with mice head-fixed above the freely moving treadmill which allowed for awake resting, and a variety of motor behaviors (Fig. 1A, bottom). Movement of the disc treadmill was monitored by a rotary encoder and recorded at 1 kHz by an Arduino microcontroller (Arduino Mega 2560; Arduino). As disease progressed, some of the SCA8+ mice were given the hydrating wet food nutritional supplement, Dietgel (ClearH₂O), on the bottom of the cage to make food/water more accessible and prevent early death due to nutritional deficiency and weight loss.

2.4. Mesoscale Ca²⁺ imaging

Ca²⁺ imaging was performed with the animal head-fixed on the disc treadmill under a wide-field epifluorescence microscope (Nikon AZ-100; Nikon). Manual zoom was used to ensure the neocortex window maximally filled the imaging field (~1.5 zoom). The microscope was focused below the brain surface to capture fluorescence from layers II/III of the neocortex (~150–200 μm). Dual wavelength imaging was achieved by strobing LEDs using a switcher (OptoLED; Cairn) and alternating image frames of Ca²⁺-dependent GCaMP6f signal (470 nm) and Ca²⁺-independent signal (405 nm; Cairn) were captured using a high-speed CMOS camera (40 fps, 18 ms exposure; 256 × 256 pixels; Orca Flash4.0; Hamamatsu; ~ 33.8 × 33.8 μm) and Metamorph software (Molecular Devices). Synchronization of the microscope camera as well as the rotary encoder was achieved using TTL pulses from a Power 1401 data acquisition system and Spike2 software (Cambridge Electronic Design).

2.5. Processing of Ca²⁺ imaging data

Individual imaging trials for each mouse were pre-processed as previously described (Nietz et al., 2023). Briefly, 470 and 405 nm images were deinterleaved and the first 30 s of each trial removed due to potential rundown of the Ca²⁺ signal. Ca²⁺-dependent GCaMP signals were corrected for Ca²⁺-independent GCaMP signals in a similar manner as prior studies (Cramer et al., 2023; Jacobs et al., 2020; MacDowell and Buschman, 2020; Nietz et al., 2023). For each experimental session, corrected Ca²⁺ images were saved along with a background (470 nm) image. A reference day was chosen for each mouse, a mask was drawn to separate out the neocortex field-of-view (FOV), and all imaging sessions were aligned to the reference day using standard Matlab registration functions (imregconfig; imregtform; imwarp). Registration of images within and between sessions was done to minimize artifacts due to slight changes in implant position in the imaging field and any motion (Nietz et al., 2023).

As the number of imaging sessions per mouse was large and varied in number, we limited our analyses to three stages of SCA8 disease. The pre-disease phase was defined as the 2–3 imaging sessions (spanning 14 ± 2.2 days) prior to SCA8 onset. Disease onset was empirically defined as a greater than 10 % reduction in maximum body weight in SCA8+ mice (all: 167 ± 47 days; males: 156 ± 46 days; females: 184 ± 47 days; n = 15,6,9 respectively). For SCA8+ mice, having an NTC littermate in the same cohort, onset for the NTC littermate was at the same age as the SCA8+ littermate. For NTC animals without an SCA8+ littermate, onset was near the average onset age for the sex of the mouse (males: ~150 days, females: ~200 days). The onset phase of SCA8 was defined as the 2–3 imaging sessions (spanning 14 ± 1.8 days) after disease onset. The late disease phase of SCA8 was defined as the final 3 imaging sessions (spanning 13.9 ± 2.9 days) prior to death/euthanasia or the final 3 imaging sessions when the mouse reached 3 months post-onset (Fig. 1C). For analyses, only imaging sessions within the 3 defined disease phases were used. The hemodynamic-corrected data for each mouse were spatiotemporally smoothed using a 3D 9×9×9 spatial gaussian filter (function: imgaussfilt3; Matlab 2022a) and concatenated. Signal noise due to LED illumination variability was removed by regressing the signal in the mask FOV against the signal in the neocortical FOV. Analysis was performed on the residuals of the regression. Denoised data were compressed using low rank singular value decomposition (SVD) keeping the first 200 components (Musall et al., 2019; Nietz et al., 2023; Saxena et al., 2020; West et al., 2022).

2.6. Spatial independent component analysis

Spatial independent component analysis (ICA) was performed on the concatenated data set for each mouse, computing 60 independent components (ICs) using the Joint Approximation and Diagonalization of Eigenmatrices (JADE) algorithm that obtains maximally independent source signals from signal mixtures by minimizing mutual information (Cardoso, 1999; Sahonero-Alvarez and Calderon, 2017), as used in previous wide-field Ca²⁺ imaging studies (Makino et al., 2017; Nietz et al., 2023; West et al., 2022). The solutions were multiplied back into the original vector space and z-scored to yield spatial maps of the ICs. Binary masks of the significant areas of the ICs were obtained by setting values between ±2.5 SD to zero and all other values to one. Very small ICs with less than 250 contiguous pixels were excluded. Occasionally, ICs contained more than one region, for example a pair of homotopic regions. In these cases, the IC was separated, so each IC mask consisted of a single region. Remaining ICs were inspected for artifacts and were manually discarded, including areas overlying only vasculature (Musall et al., 2019; Nietz et al., 2023; West et al., 2022).

To aid in results interpretation, the Allen Common Coordinate Framework (CCFv3) was aligned to the cranial window of each animal through a multi-step warping and alignment process using custom code modified from Paninski and colleagues (Saxena et al., 2020; Wang et al., 2020). For each animal, the surgical craniotomy image containing cranial landmarks (inferior cerebral vein near the frontonasal suture; bregma; midline; lambda) and full craniotomy drill path was cropped and rotated to match the orientation of the neocortical window during Ca²⁺ imaging sessions. The surgical image was used to align the craniotomy to the CCFv3 (Saxena et al., 2020; Fig. 1D) and the inverse transform was obtained to align the atlas to the final images. Next, the reference image for the Ca²⁺ imaging sessions was loaded. A mask was drawn around the drill path of the large craniotomy or the frame of the implant which sits directly above the drill path for each image, respectively (Fig. 1E). The surgical craniotomy mask was registered to the implant frame mask (imregconfig; imregtform; imwarp) and the surgical craniotomy images containing cranial landmarks were used to align the craniotomy to the CCFv3 (Saxena et al., 2020; Wang et al., 2020). The resulting transformations were applied to the atlas aligning it to the reference image of the implant for each mouse. Once aligned, we were able to visualize ~15 atlas regions in each hemisphere (30 total) of the mouse cortex (Fig. 1F).

2.7. Functional connectivity analysis

Using the aligned Allen CCF for each mouse, ICs were assigned to an atlas region based on position of their centers; or if on a border, which atlas region the IC overlapped with most. Functional connectivity (FC) adjacency matrices were computed across atlas regions using the average ΔF/F signal from ICs on a trial basis using canonical correlation (canoncorr), which gives a weighted correlation between sets of variables (in this case sets of ICs within two atlas regions). Only the first canonical correlation value was used in the FC matrices. To aid in comparison across genotypes, adjacency matrices for each mouse were expanded to include all possible atlas regions seen across mice (30 areas). When a particular atlas region was not represented for a mouse (did not have an IC assigned), zeros were placed in the adjacency matrix for that atlas region. To determine network structure and properties, expanded adjacency matrices were thresholded at canonical correlation values ≥0.5. FC graphs were plotted using the centroids of atlas regions as nodes and thresholded canonical correlation values as edges.

The Brain Connectivity Toolbox was used to calculate FC graph properties at a trial level using thresholded FC matrices containing all 30 possible atlas areas, excluding strength, matching index, and community structure (Rubinov and Sporns, 2010). Strength and matching index were calculated using unthresholded FC matrices.

To compute global network properties for each node in the graph (such as strength and eigenvector centrality), we averaged across nodes for each trial-based graph to obtain an average metric per trial. To calculate community structure (Louvain-communities), adjacency matrices containing only the represented nodes (atlas regions) for a mouse were used and the measure was subsequently normalized to the number of nodes in the graph.

To calculate region-based changes in network properties, measures calculated at the node level (strength, centrality, degree, matching) were averaged for each node pairing (i.e. all measures comparing the relevant node to other network nodes) for each disease phase. While we were able to calculate some of the global measures at the node level (strength, eigenvector centrality), others are strictly global (density, global efficiency). To obtain similar measures at the node level, we determined nodal degree (as a proxy for connection density) and nodal matching index (to assess connectivity overlap between two nodes, with the assumption that more overlap suggests greater efficiency). Microsoft Excel (Microsoft Corporation, 2016), GraphPad Prism (Graphpad, 2024, Boston MA), and JMP Pro software (JMP Statistical Discovery LLC, 2024, Cary NC) were used to compare network properties across the three disease phases for SCA8+ and NTC mice.

2.8. GCaMP6f Immunohistochemistry and cortical expression

A selected set of retro-orbitally injected animals (both genotypes) were used for histology to verify the efficacy of viral expression. Mice were anesthetized with isoflurane (5 %) and either perfused intracardially with 0.1 M phosphate buffered saline (PBS; pH ~ 7.2–7.4) followed by paraformaldehyde (PFA; 4 %) or injected intracardially with 0.3 mL Euthasol solution (Virbac). Brains were extracted and post-fixed for 1–3 days, then sectioned (50 μm sections; coronal) on either a cryostat or vibratome. For cryostat sectioning, brains were submerged in 30 % sucrose solution 1–2 days prior to sectioning for dehydration. After sectioning, tissue was kept in antifreeze solution (30 % glycerol; 30 % ethylene glycol; 40 % PBS) until needed for histology to prevent tissue degradation. At time of histology, tissue sections were washed with PBS (0.1 M; 3 times for 10 min) and blocked in a PBS or Tris Buffered Saline (TBS) solution containing 0.5 % Triton-X 100 and 10 % normal donkey serum (NDS; Sigma-Aldrich D9663) on an orbital shaker for 1 h at room temperature. Tissue was incubated overnight at room temperature with a primary rabbit anti-GFP antibody (1:1000; A-6455 ThermoFisher). The following day, tissue was rinsed with PBS (0.1 M; 3 times for 10 min) and incubated with a donkey-anti-rabbit Alexa Fluor 555 (1:500; A-31572, ThermoFisher) or donkey-anti-rabbit Alex Fluor 488 secondary antibody (1:500; A-32790, ThermoFisher) for two hours at room temperature. Tissue was subsequently rinsed with PBS (0.1 M; 3 times for 10 min) and mounted with Invitrogen ProLong Diamond Antifade Mountant with DAPI (P36962, Invitrogen, ThermoFisher).

Tissue sections were imaged using a Leica Stellaris confocal microscope (Fig. 1) or a Leica Thunder epifluorescence microscope (Supplementary Fig. 2). Single plane tiled images of tissue sections were taken using a 20× objective lens (1024 × 1024 resolution confocal; 2048 × 2048 resolution epifluorescence). The confocal was equipped with a tunable excitation laser at 521 nm and an emission filter in the range of the Alexa Fluor 555 fluorescence peak (540–625 nm). The epifluorescence microscope was equipped with a Sola Light Engine white light source and excitation/emission filters in the range of the Alexa Fluor 488 spectra. Coronal neocortical images were collected, stitched, and merged using Leica LAS X software (Leica Microsystems). Cortical z-stacks were also acquired using a 20× objective with 2–3 frame averaging and an optical z-step of 1 μm. Images were processed using FiJi software (Schindelin et al., 2012). Tiled images or confocal stack maximum projections were background subtracted, denoised and contrast was enhanced for visualization.

2.9. Statistical analysis

Data were analyzed and plotted using custom-written Matlab (Matlab 2022a) code and the Brain Connectivity Toolbox (Rubinov and Sporns, 2010). For data presented in bar and line graph format as well as all tables, descriptive statistics shown are mean ± standard deviation. For data presented in violin plot format, descriptive statistics shown are median ± upper/lower quartile. Atlas maps showing average change in network measures per atlas area are presented in a colorimetric manner with increases in red and decreases in blue (hue indicates change magnitude). Atlas areas in gray indicate no statistically significant change. Statistical analysis was performed using GraphPad Prism and JMP. Comparison of IC numbers and cortical coverage between genotypes were evaluated using non-parametric Mann-Whitney (MW) tests. Comparisons of global network measures across time and genotype were done at the trial level (~450 trials per time point) using repeated measures mixed-effects two-way ANOVAs with post-hoc Bonferroni corrected pairwise comparisons.

Comparison of region-specific network measures was performed in JMP statistical software. A repeated measures mixed model 3-way ANOVA was done by fitting a random effects linear model to the data. Missing data and measurements with a value of zero were excluded from analysis, as including zero values skewed the data distribution. Due to the size of the dataset at the trial level, each node was assigned to one of 6 major anatomical regions (primary motor cortex – MC primary, secondary motor cortex – MC secondary, barrel field cortex - Bfd, somatosensory cortices - SSp, retrosplenial cortex - RSP, visual and accessory visual cortices - VIS). Network property measurements per node were also grouped into the 6 major anatomical parcellations and used to analyze regional differences. The average magnitude of the change in network properties across the major anatomical regions were plotted over the Allen CCF to show the additional subregions of the major cortical areas. Post-hoc Tukey tests were used for pairwise comparisons between network measures in SCA8+ and NTC mice in previously defined major atlas regions.

Study Approval:

All experimental procedures were approved by the Institutional Care and Use Committee of the University of Minnesota.

Data and Code Availability:

The raw database of Ca²⁺ recordings in these animals consists of several terabytes of data. As such, the raw and compressed data will only be available upon request. Custom code will also be available upon request.

3. Results

3.1. Database and experimental paradigm

SCA8 transgenic mice (SCA8+; n = 7) and non-transgenic controls (NTC; n = 9) were imaged on a freely moving disc treadmill allowing for spontaneous rest and locomotion for ~1 h per session (~10 trials; 5.5 mins per trial) throughout disease progression (see Methods; Fig. 1A). Cortex-wide GCaMP6f expression was achieved using a retro-orbital injection and verified using post-hoc immunohistochemistry (Fig. 1B; Supplemental Fig. 2). Prior studies have shown that retro-orbital injection of AAV-PHP-eb viruses result in wide-spread efficient sensor expression throughout the CNS (Bedbrook et al., 2018; Chan et al., 2017). Neural activity was monitored in these mice throughout disease progression with disease onset defined as a 10 % drop in maximal weight (Fig. 1C). Analysis was done using defined chronological epochs (see Methods; Fig. 1C). Implanted transparent polymer skulls were aligned to the atlas using a multistep registration process. First Allen CCF landmarks were aligned to pre-craniotomy landmarks (Fig. 1D). Next, the craniotomy path and implant border were aligned (Fig. 1E). Finally, the Allen atlas was back transformed to the implanted window. The aligned implanted windows allowed visualization of layer II/III neocortical activity spanning the secondary motor cortex to the visual cortex in the anterior-posterior direction and the retrosplenial cortex to the medial edge of the barrel fields in the medial-lateral direction (Fig. 1F).

3.2. SCA8 transgenic mice and controls show similar IC coverage across functional atlas areas

The first question addressed is whether functional segmentation of the cortex differed between SCA8+ and NTC animals. Spatial independent component analysis (ICA) was run on a mouse level by concatenating data from the pre-disease, onset, and late phases (see Methods; Nietz et al., 2023) yielding a single set of independent components (ICs; Fig. 2A). SCA8+ and NTC animals have similar numbers (Fig. 2B; SCA8+: 55.3 ± 5.9; NTC: 50 ± 8.5; MW: U = 17.50, p = 0.15, n = 7,9) and coverage of the ICs (SCA8+: 78.4 ± 3.7 %; NTC: 74.2 ± 5.7 %; MW: U = 18, p = 0.17; n = 7,9). The ICs were assigned to CCF atlas regions based on their center positions (Fig. 2C). Similar to the overall IC values between SCA8+ and NTCs, each atlas region contains similar numbers of ICs (Fig. 2D; for statistical details see Table 1). The numbers and spatial distribution of ICs imply that the functional segmentation remains intact in SCA8+ mice. For each IC, the hemodynamic-corrected ΔF/F time series was extracted for both SCA8+ and NTCs at each phase of disease progression. All phases of disease in both genotypes show GCaMP fluorescence transients of varying amplitudes, indicative of neuronal activity; similar to previous reports (Cramer et al., 2023; West et al., 2022; West et al., 2024). Example time series from select ICs show the expected fluorescence modulation with comparable levels (Fig. 2E). Qualitatively, the fluorescence signals suggest stronger correlations between cross regional ICs in the SCA8+ mice compared to the NTCs. The subsequent functional connectivity (FC) analyses address how these patterns of correlation across the neocortex are organized and how they differ between genotypes.

Fig. 2.

SCA8+ and NTC mice show similar neocortical functional segmentations using spatial ICA. A) Schematic showing chronological concatenation of data over 3 defined SCA8 disease phases (top) and the resulting neocortical functional segmentations for an SCA8+ (middle) and NTC mouse (bottom; different colors denote individual ICs). B) Scatter bar plots showing the average number of ICs (top) per SCA8+ (black) or NTC (gray) mouse after spatial ICA processing and average percentage of cortical coverage by ICs (bottom). C) Example neocortical images showing aligned pseudo-colored Allen CCF overlay and IC centroids (left) used to assign and color code ICs to functional atlas regions (right). Scale bars 1 mm. D) Scatter bar plot showing the average number of ICs assigned to each functional atlas region for SCA8+ (black) and NTC (gray) animals. MC - Motor Cortex; SS - Somatosensory Cortex; BF - Barrel Field Cortex; RSP - Retrosplenial Cortex; VIS - Visual Cortex; R - right; L - left. E) Images showing select ICs (colored areas) overlayed on the neocortex for an SCA8+ (left) and NTC (right) mouse and examples of their corresponding Ca²⁺ traces for each of the three disease phases (bottom).

Table 1.

Numerical comparison of ICs per CCF functional area.

Atlas Area	# ICs per region (mean ± SD)		MW U statistic	MW p-value (SCA8+ vs. NTC)
MC_L	SCA8+: NTC:	7.14 ± 3.18 7.33 ± 2.87	U = 29.5	p > 0.99
MC_R	SCA8+: NTC:	8.29 ± 3.15 8.44 ± 3.43	U = 30.5	p > 0.99
SS_L	SCA8+: NTC:	8.29 ± 1.6 7.67 ± 1.73	U = 24.0	p > 0.99
SS_R	SCA8+: NTC:	7.29 ± 1.38 6.67 ± 1.32	U = 27.0	p > 0.99
BF_L	SCA8+: NTC:	2.00 ± 1.00 1.22 ± 1.48	U = 20.0	p = 0.90
BF_R	SCA8+: NTC:	1.71 ± 1.25 1.22 ± 1.30	U = 23.5	p > 0.99
RSP_L	SCA8+: NTC:	2.43 ± 0.98 3.44 ± 1.59	U = 19.0	p = 0.90
RSP_R	SCA8+: NTC:	5.00 ± 1.83 2.89 ± 1.05	U = 10.5	p = 0.21
VIS_L	SCA8+: NTC:	6.57 ± 1.72 6.22 ± 3.60	U = 26.0	p > 0.99
VIS_R	SCA8+: NTC:	6.29 ± 3.04 5.00 ± 1.22	U = 26.0	p > 0.99

Average number of ICs (± SD) and statistical comparisons per CCF atlas region between SCA8+ and NT control mice. SCA8 - SCA8+; NTC - NT control; MW - Mann- Whitney test; MC - motor cortex; SS - somatosensory; BF - barrel fields; RSP - retrosplenial cortex; VIS - visual cortex; L - left; R - right.

3.3. Global network functional connectivity is disrupted in SCA8+ mice

For each disease phase, FC was assessed by correlating IC average fluorescence signals within each atlas region to other atlas regions using canonical correlation analysis (CCA; Saxena et al., 2020; Wang et al., 2020). The CCA matrices were thresholded and all values <0.5 were set to zero and graphed using Matlab (graph, plot). The FC graphs were plotted over brain images with the nodes at the center of each atlas area and node size signifying relative connection density. The edges are CCA values ≥0.5 with edge color signifying the weight for each connection. Examples of atlas assigned ICs from individual SCA8+ animals (Fig. 3A-B) and NTC animals (Fig. 3D-E) show the larger atlas cortical areas (e.g., motor, somatosensory, visual, retrosplenial) are represented in both genotypes. Additionally, the size, number, and spatial distribution of ICs identified in both genotypes show that functional parcellation of the neocortex is not as rigidly defined as suggested by the CCF. This agrees with previous studies (Makino et al., 2017; Nietz et al., 2023; West et al., 2022) and is not unexpected as the ICs are generated from underlying cortical activity from each mouse. Taken together, these data show that the structure and organization of the functional parcellation is not fundamentally altered in SCA8+ mice.

Fig. 3.

SCA8+ mice have hyperconnected neocortical networks. A,D) Example ICs from an SCA8+ (A) or NTC (D) mouse color-coded to their assigned atlas regions. B,E) Numerical IDs for broad Allen CCF atlas regions corresponding to the matrices shown in C and F, respectively. C,F) Average thresholded (≥0.5) canonical correlation analysis (CCA) matrices (top) for the atlas regions defined in B or E for each disease phase in an SCA8+ (C) and NTC (E) mouse. Canonical correlations are calculated using the Ca²⁺ signals of the ICs belonging to each atlas region keeping only the first canonical correlate for each comparison. Black lines delineate major CCF functional divisions. Corresponding neocortical network graphs (bottom) for the CCA matrices shown above. Nodes represent the CCF atlas subregions and nodes are color coded according to the major atlas region to which they belong. Edges represent functional connections assessed by CCA. Node size indicates relative number of connections and edge color indicates relative connection strength.

Individual SCA8+ animals show a global increase in the canonical correlation between Ca²⁺ signals in the pre-disease phase that persists throughout disease progression (Fig. 3C, top) compared to NTCs (Fig. 3F, top). These global increases in correlation produce an increased number and strength of connections in network graphs (Fig. 3C/F; bottom), altering the terrain of the FC maps, with the increased connectivity observed across SCA8+ animals compared to NTCs (Fig. 4A). The canonical correlations of IC fluorescence signals in SCA8+ animals are greater both within and between atlas regions (Fig. 4B, top) compared to NTCs (Fig. 4C, bottom) and the increase is widespread. Therefore, the SCA8+ mice have greater intra- and inter-nodal FC than NTCs.

Fig. 4.

Neocortical networks are globally hyperconnected across SCA8+ mice. A) Pseudo-colored Allen CCF showing atlas subregions visible across all subjects (top, 30 total regions) and the centroid of each atlas region (middle; centroids denoted by black dots). Atlas subregions are color coded according to which major atlas area they belong to. Bottom panel shows color-coded legend for visible atlas subregions above and their numerical IDs corresponding to the matrices shown in B and C. B-C) Average CCA matrices (top) across all SCA8+ (B) and all NTC (C) mice for all visible atlas regions in A at each disease phase (gray squares are connections below the CCA threshold of 0.5) and their associated network graphs (bottom). Black lines denote major CCF functional divisions. Corresponding neocortical network graphs (bottom) for the CCA matrices shown above. Nodes and node colors represent CCF atlas areas and edges represent functional connections assessed by CCA. Node size indicates relative number of connections and edge color indicates relative connection strength. Missing nodes denote regions that had no correlations above the set threshold.

We quantified these network-wide changes in FC using the Brain Connectivity Toolbox (Fornito et al., 2016; Rubinov and Sporns, 2010). As observed in the canonical correlation matrices (Figs. 3 and 4), the average connection density and global nodal strength are significantly higher in the SCA8+ animals compared to NTC animals during all three disease phases (Fig. 5A left, middle; for statistical details see Table 2), suggesting a hyperconnectivity and strongly coupled cross-regional fluorescence signals. In all three disease phases, the average global efficiency, global eigenvector centrality, and number of community partitions are also increased in SCA8+ animals (Fig. 5A right; Fig. 5B left, middle; also see Table 2) compared to NTCs. Global efficiency measures the ease of information exchange across the network and eigenvector centrality measures the magnitude of influence a node has over network processing. Interestingly, the average global transitivity (an analog of the clustering coefficient which measures connectivity of a node to its neighbors) is significantly increased during the pre-disease and late disease phases but not during disease onset (Fig. 5B, right; also see Table 2). As several global connectivity and network topology measures are altered in SCA8+ animals compared to NTCs prior to our empirical definition of SCA8 onset, this suggests that neocortical alterations happen early-on in disease before the appearance of overt physical symptoms.

Fig. 5.

Network configuration is altered across disease phases in SCA8+ mice. A) Violin plots showing the average network density (left), average strength per node (middle), and global network efficiency (right) per trial in SCA8+ (black) and NTC (blue) animals at each disease phase. B) Violin plots showing average eigenvector centrality per node (left), average number of communities (middle), and average network transitivity (right) per trial in SCA8+ and NTC animals. Red bars denote median and orange bars denote the quartiles. Community partitions were calculated using only regions present in each mouse and normalized to the number of nodes in each animal’s network to allow equal comparison. Black bars denote statistically significant differences between SCA8+ and NTC mice according to a 2-way repeated measures mixed-model ANOVA with Bonferroni post-hoc comparisons matching the statistics presented in Table 2. *** p < 0.0001; * p < 0.05.

Table 2.

Network configuration metrics and statistical comparisons in NTC and SCA8 mice.

Network Measure	Mean ± SD	Statistical test (post-hoc test)	Test statistic + post-hoc comparisons	p-value	# trials (# animals)
Density	Pre-disease SCA8: 0.51 ± 0.14 NTC: 0.44 ± 0.12 Onset SCA8: 0.50 ± 0.14 NTC: 0.45 ± 0.12 Late SCA8: 0.51 ± 0.13 NTC: 0.42 ± 0.11	2-way repeated measures ANOVA (Bonferroni Correction)	Time: F (1.630, 709.8) = 0.084 Genotype: F (1, 481) = 31.63 Interaction: F (2, 871) = 0.66 NTC vs. SCA8 pre-disease NTC vs SCA8 onset NTC vs. SCA8 late NTC pre-disease vs. onset NTC pre-disease vs. late NTC onset vs. late SCA8 pre-disease vs. onset SCA8 pre-disease vs. late SCA8 onset vs. late	0.88 <0.0001 0.51 <0.0001 < 0.0001 < 0.0001 > 0.999 0.0002 < 0.0001 > 0.999 > 0.999 0.9721	Pre-disease SCA8: 204 (7) NTC: 256 (9) Onset SCA8: 208 (7) NTC: 250 (9) Late SCA8: 205 (7) NTC: 235 (9)
Strength	Pre-disease SCA8: 13.18 ± 3.20 NTC: 11.05 ± 2.40 Onset SCA8: 13.02 ± 3.05 NTC: 11.33 ± 2.85 Late SCA8: 13.19 ± 2.90 NTC: 10.53 ± 2.35	2-way repeated measures ANOVA (Bonferroni Correction)	Time: F (1.716, 747.5) = 0.48 Genotype: F (1, 481) = 63.54 Interaction: F (2, 871) = 4.87 NTC vs. SCA8 pre-disease NTC vs. SCA8 onset NTC vs. SCA8 late NTC pre-disease vs. onset NTC pre-disease vs. late NTC onset vs. late SCA8 pre-disease vs. onset SCA8 pre-disease vs. late SCA8 onset vs. late	0.59 < 0.0001 0.0079 < 0.0001 < 0.0001 <0.0001 0.009 0.0004 < 0.0001 0.59 > 0.99 0.4534	Pre-disease SCA8: 204 (7) NTC: 256 (9) Onset SCA8: 208 (7) NTC: 250 (9) Late SCA8: 205 (7) NTC: 235 (9)
Global Efficiency	Pre-disease SCA8: 0.46 ± 0.11 NTC: 0.38 ± 0.08 Onset SCA8: 0.45 ± 0.10 NTC: 0.39 ± 0.10 Late SCA8: 0.46 ± 0.10 NTC: 0.37 ± 0.08	2-way repeated measures ANOVA (Bonferroni Correction)	Time: F (1.681, 732.1) = 0.76 Genotype: F (1, 481) = 66.47 Interaction: F (2, 871) = 4.64 NTC vs. SCA8 pre-disease NTC vs. SCA8 onset NTC vs. SCA8 late NTC pre-disease vs. onset NTC pre-disease vs. late NTC onset vs. late SCA8 pre-disease vs. onset SCA8 pre-disease vs. late SCA8 onset vs. late	0.45 < 0.0001 0.0099 < 0.0001 < 0.0001 < 0.0001 0.0052 0.0002 < 0.0001 0.77 > 0.999 0.5515	Pre-disease SCA8: 204 (7) NTC: 256 (9) Onset SCA8: 208 (7) NTC: 250 (9) Late SCA8: 205 (7) NTC: 235 (9)
Eigenvector Centrality	Pre-disease SCA8: 0.16 ± 0.010 NTC: 0.15 ± 0.008 Onset SCA8: 0.15 ± 0.010 NTC: 0.15 ± 0.009 Late SCA8: 0.16 ± 0.009 NTC: 0.15 ± 0.008	2-way repeated measures ANOVA (Bonferroni Correction)	Time: F (1.744, 759.7) = 0.49 Genotype: F (1, 481) = 33.55 Interaction: F (2, 871) = 1.537 NTC vs. SCA8 pre-disease NTC vs. SCA8 onset NTC vs. SCA8 late NTC pre-disease vs. onset NTC pre-disease vs. late NTC onset vs. late SCA8 pre-disease vs. onset SCA8 pre-disease vs. late SCA8 onset vs. late	0.59 < 0.0001 0.22 < 0.0001 < 0.0001 < 0.0001 > 0.999 < 0.0001 < 0.0001 0.4423 > 0.999 0.6792	Pre-disease SCA8: 204 (7) NTC: 256 (9) Onset SCA8: 208 (7) NTC: 250 (9) Late SCA8: 205 (7) NTC: 235 (9)
Community #	Pre-disease SCA8: 0.38 ± 0.23 NTC: 0.29 ± 0.17 Onset SCA8: 0.37 ± 0.22 NTC: 0.32 ± 0.18 Late SCA8: 0.39 ± 0.22 NTC: 0.28 ± 0.15	2-way repeated measures ANOVA (Bonferroni Correction)	Time: F (1.965, 855.7) = 0.464 Genotype: F (1, 481) = 33.35 Interaction: F (2, 871) = 5.694 NTC vs. SCA8 pre-disease NTC vs. SCA8 onset NTC vs. SCA8 late NTC pre-disease vs. onset NTC pre-disease vs. late NTC onset vs. late SCA8 pre-disease vs. onset SCA8 pre-disease vs. late SCA8 onset vs. late	0.63 < 0.0001 0.0035 < 0.0001 0.0074 < 0.0001 0.0466 0.7561 0.0025 > 0.999 > 0.999 0.4949	Pre-disease SCA8: 204 (7) NTC: 256 (9) Onset SCA8: 208 (7) NTC: 250 (9) Late SCA8: 205 (7) NTC: 235 (9)
Transitivity	Pre-disease SCA8: 0.79 ± 0.13 NTC: 0.77 ± 0.11 Onset SCA8: 0.79 ± 0.13 NTC: 0.78 ± 0.12 Late SCA8: 0.80 ± 0.13 NTC: 0.75 ± 0.11	2-way repeated measures ANOVA (Bonferroni Correction)	Time: F (1.969, 857.4) = 1.022 Genotype: F (1, 481) = 6.913 Interaction: F (2, 871) = 4.471 NTC vs. SCA8 pre-disease NTC vs. SCA8 onset NTC vs. SCA8 late NTC pre-disease vs. onset NTC pre-disease vs. late NTC onset vs. late SCA8 pre-disease vs. onset SCA8 pre-disease vs. late SCA8 onset vs. late	0.3595 0.0088 0.0117 0.0175 0.3323 < 0.0001 0.067 0.2836 0.0030 > 0.999 0.9024 0.5126	Pre-disease SCA8: 204 (7) NTC: 256 (9) Onset SCA8: 208 (7) NTC: 250 (9) Late SCA8: 205 (7) NTC: 235 (9))

Average values (± SD) across all trials and mice for each network measure shown in Fig. 5 at each timeframe (italicized cell headings) in SCA8+ and NT control mice and corresponding statistical comparisons. Both main effects and interaction test statistics and p-values are reported for 2-way repeated measures ANOVAs. All p-values are reported for Bonferroni corrected post-hoc comparisons (italicized comparisons). Significant p-values are bolded. SCA8 - SCA8+; NTC - NT control.

3.4. Area specific alterations in functional connectivity in SCA8

To determine if there are location-specific alterations in the SCA8+ networks, node-specific analyses were performed using six major anatomical regions (Motor Cortex – MC secondary and MC primary; Barrel Fields – Bfd; Somatosensory Cortices – SSp; Retrosplenial Cortex – RSP; Visual Cortices – VIS; see Supplementary Tables 1–3). For analyses, FC measures were determined for all CCF subregions belonging to the defined major regions bilaterally and assigned to their major region. Statistical comparisons were made between major regions at each disease phase between genotypes using a 3-way repeated measures mixed-model ANOVA. FC measures between major regions were considered significantly different if the Tukey-adjusted p-value was <0.05. Changes were common throughout the cortex, including the visual and retrosplenial areas, in addition to the sensorimotor cortices. Nodal degree is increased in the secondary motor, retrosplenial, and the visual cortices (Fig. 6A; see Supplementary Tables 1–3). These increases in nodal degree persist from the pre-disease phase to the late disease phase and progress into the somatosensory cortices in late disease. Most areas of the neocortex showed increases in connection strength with other cortical regions (Fig. 6B; see Supplementary Tables 1–3). The increase in strength becomes more pronounced in late disease as SCA8 symptoms become severe.

Fig. 6.

Region-specific changes in network topology drive global differences in SCA8 networks. Pseudo-colored Allen CCFs showing statistically significant changes in nodal degree (A), strength (B), matching index (C), and eigenvector centrality (D) between SCA8+ mice and NTC mice for each disease phase. The magnitude of change in network measures for pseudo-coloring and scaling is calculated by subtracting the average measure across NTC animals for each of six major atlas regions from the average across SCA8+ animals. If the network property change was significant, all atlas subregions belonging to the major areas received the same color-coding, if no significant changes occurred, areas were colored gray. Positive changes in magnitude (red) indicate the network measure is increased SCA8+ animals compared to NTCs. Negative changes in magnitude (blue) indicate the network measure is reduced in SCA8+ mice compared to NTC. All colored regions were significantly different between genotypes for the major atlas area they belong to (p < 0.05) with Tukey HSD post-hoc comparisons following a 3-way repeated measures mixed model ANOVA. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Next, we examined whether specific nodes in the SCA8+ networks were more functionally similar to one another compared to NTC networks. To do this, we calculated the matching index for each node which measures connectivity overlap between two nodes (or redundancy of connections) which are not connected to one another (Koutrouli et al., 2020). We found the networks in SCA8+ animals have an increased matching index compared to NTCs in the somatosensory cortices across disease phases (Fig. 6C; see Supplementary Tables 1–3). Matching index is selectively increased in the primary motor, retrosplenial, and visual areas of SCA8+ mice in the onset phase. These data reinforce the findings that the sensory cortices and integrational areas contain many of the changes in FC in SCA8+ compared to NTC animals. Finally, we assessed nodal changes in eigenvector centrality in SCA8+ and NTC networks (Fig. 6D; see Supplementary Tables 1–3). Centrality is selectively reduced in the primary motor cortex and major somatosensory areas. In contrast, centrality is increased in the retrosplenial areas during the pre-disease and late phases. While this contrasts with the increase in global network centrality in SCA8+ animals, we hypothesize that the increase in global network centrality is driven by low magnitude increases in centrality across nodes which, when examined at a regional level, do not result in region specific significant centrality changes. These node specific alterations suggest that cortical areas with high sensory integration have considerable control over network processing in these SCA8 mice.

3.5. Predicting SCA8 genotype using functional connectivity

Finally, we utilized CCA-based FC matrices to determine whether FC network properties can be used to accurately decode the genotype of animals using a stepwise generalized linear model (GLM). In our GLM, Bayesian information criterion (BIC) was used to determine the best-fit model and the minimum number of functional connections needed to predict animal genotype. At each phase of disease, the GLM decodes genotype with >98 % accuracy, >98 % precision, and > 98.5 % recall (Fig. 7A; n = 10 cross-validations). When the model was trained using shuffled genotypes, model accuracy dropped significantly with <53 % accuracy, <46 % precision, and < 22 % recall (Fig. 7B; total n = 5000; iterations - 1000; cross-validations/shuffle - 5). These data show that FC can be used as a distinguishing feature for SCA8 disease prior to symptom onset in our model.

Fig. 7.

Functional connectivity can be used to decode genotype in SCA8 animals. A) Summative confusion charts (top) for each disease phase showing mouse genotype classification after training a stepwise GLM with ten-fold cross-validation and the average performance metrics associated with each GLM (mean ± SD). B) Summative confusion charts (top) for disease phase showing mouse genotype classification after training a stepwise GLM with shuffled classification labels (1000 permutations each with 5-fold cross-validation) and their corresponding performance metrics (bottom). C) Allen CCF outlines and network nodes (color-coded to major CCF atlas region) showing connections important for discriminating between genotypes in each GLM at each disease phase. Edges show important connections and color shows each connection’s probability of use across the 10 cross-validations of the GLM. Note that most of these areas had significant network measure changes in Fig. 6.

Next, the GLM was used to determine which predictors (FC between atlas areas) were most effective at discriminating between SCA8+ mice and NTCs. The predictors repeatedly selected by the stepwise GLM algorithm across cross-validations provided the most predictive power. The most powerful predictive interactions are located primarily in the posterior sensory cortices and integration areas across all disease phases, particularly between the barrel and somatosensory cortices as well as the visual and retrosplenial cortices (Fig. 7C). In the pre-disease and late disease stages, interactions between the motor cortices and retrosplenial areas contribute to the prediction of disease state. Interestingly, the areas most effective at predicting animal genotype are the same areas with specific changes in network connectivity metrics. These data show that neocortical network topology and sensorimotor integrational FC is fundamentally impaired throughout the lifespan of SCA8+ mice, and the altered topology provides highly robust information on the genotype.

4. Discussion

We performed longitudinal neocortex-wide Ca²⁺ imaging in mice expressing the human SCA8 expansion mutation as well as non-transgenic control animals. To assess the FC between cortical areas, we used spatial ICA to functionally segment the cortex in an unbiased, data-driven manner and mapped the ICs onto the CCF. This was followed by CCA between atlas regions as the basis for determining differences in cortical network FC. Compared to the control animals, SCA8+ mice exhibit several global differences, including hyperconnectivity and increased connection magnitude as well as increases in efficiency, global eigenvector centrality, and number of communities. Interestingly, these changes are evident prior to our definition of the disease onset, a 10 % loss of maximal body weight. These global changes suggest that in the SCA8+ neocortex information processing is more fragmented and the specificity of information transfer across the network is impaired. At the regional (nodal) level, FC changes were evident not only in the motor cortices but also the posterior sensory and sensory integration regions. The most prominent changes over time are connection density and strength, with subtler changes in nodal matching index and nodal eigenvector centrality. Additionally, the region-specific changes increased, both spatially and in magnitude, including more sensory and higher visual regions as the disease progressed. The information in these region-specific changes could be used to perform near perfect decoding of animal genotypes using a generalized linear model.

4.1. Neocortical functional segmentation is preserved in SCA8

Many studies of FC utilize canonical neocortical segmentations such as the Allen CCF or pool segmentations across animals (Cramer et al., 2019; White et al., 2011). This includes using the same segmentation for both control and models of disease (Bero et al., 2012; Cramer et al., 2019; Vasilkovska et al., 2023). While common segmentations make interpretation easier, inter-subject variability and disease-related changes are not available. As neurodegenerative diseases can dramatically alter both brain structure and function, including in SCA8 (Ayhan et al., 2018; Kumar and Miller, 2008; Moseley et al., 2006), we could not presume that neocortical functional segmentation remains stable. Therefore, we used spatial ICA to determine if abnormalities exist in the functional segmentation of SCA8+ mice. We found that neocortical functional segmentation does not differ between SCA8+ and NTCs, with similar numbers and spatial distributions of ICs within the canonical CCF. The present finding of a stable functional segmentation is consistent with our previous results using a mouse model of mild traumatic brain injury (Cramer et al., 2023) and that activity-dependent segmentations from spatial ICA are largely stable within subjects across time and behavior (Nietz et al., 2023). These data suggest that neocortical functional segmentation remains stable even in this mouse model of cerebellar dysfunction.

4.2. Neocortical networks are hyperconnected in SCA8+ mice

Using the Brain Connectivity Toolbox (Rubinov and Sporns, 2010), we constructed FC networks and found cortical networks are globally hyperconnected in SCA8+ animals compared to NTCs, including increased connection density and strength. The network hyperconnectivity preceded our definition of disease onset and presentation of phenotypic symptoms. As the SCA8 mutation is constitutively expressed, pre-symptomatic effects on brain function are not necessarily unexpected. As disease progressed and motor symptoms became severe, global FC in the neocortex did not change appreciably between SCA8+ and NTCs. These findings suggest that FC alterations are not simply due to increasing behavioral deficits or health decline, but instead to early and persistent changes in SCA8+ brain function.

While we are not aware of comparable FC studies in SCA8 mouse models or patients, pre-symptomatic altered FC occurs in other neurodegenerative diseases. For example, several AD mouse models exhibit hyperexcitability in cortical and hippocampal networks with reduced FC between regions and increased seizure susceptibility prior to the manifestation of memory deficits (Busche et al., 2015; Busche and Konnerth, 2016; Kazim et al., 2017). Frontal-cerebellar FC is reduced in Fredrich’s Ataxia patients whereas FC between cortical regions is enhanced (Cocozza et al., 2018). In SCA3 patients, cerebello-cerebral FC is disrupted and correlates with trinucleotide repeat length (Guo et al., 2023). Here, we show for the first time functional hyperconnectivity in a mouse model of SCA8.

4.3. Cortical network topology is altered in SCA8

Investigation into network structure in SCA8+ and NTCs revealed different network configurations. Network efficiency and nodal centrality were increased in SCA8+ mice relative to NTCs and the network was partitioned into more, smaller communities with more clustered connectivity. Other disease states, like temporal lobe epilepsy, show similarly altered cortical networks (Bernhardt et al., 2015; Paldino et al., 2017). Epileptic children display hyper-clustered, highly efficient network configurations, whereas adults display hyper-clustered but less efficient networks compared to non-epileptic individuals. In human stroke patients and two mouse stroke models, networks show hyperconnectivity, increased nodal strength, increased clustering of nodes in the network, and shorter characteristic path length (indicative of increased efficiency; Blaschke et al., 2021). Individuals with Aβ deposits, suggestive of early AD, show brain area dependent increases or decreases in eigenvector centrality (Lorenzini et al., 2023). In Huntington’s Disease (HD), resting state networks shift toward within-network hyperconnectivity but reduced connectivity between networks (Werner et al., 2014), and area-specific changes in FC are apparent with concurrent changes in network configuration (Harrington et al., 2015). Finally, a zebrafish model of depression shows increased network modularity with more anatomically distributed communities (Burgstaller et al., 2019), suggesting both a less structured organization and reduced long-range information transfer (Nelson and Bonner, 2021). In our SCA8+ mice, we propose that the increased network efficiency and global centrality together with the larger community number and clustering produce a more randomly distributed network where information flow is less segregated and causes disintegration of network processing.

4.4. Region-specific hubs drive global network alterations in SCA8

Next, we evaluated whether the FC changes were uniform or area specific. Surprisingly, we found that in the pre-disease and onset disease phases, nodal degree and strength were increased in the secondary motor, visual, and retrosplenial cortices. In late disease, changes progressed to include the barrel, primary, motor, and major somatosensory cortices. The nodal matching index (a measure of connection redundancy) was increased in the primary motor area and somatosensory areas. Increased matching index extended into retrosplenial and visual areas specifically in the onset phase. Interestingly, nodal eigenvector centrality was selectively reduced in the primary motor and somatosensory cortices throughout disease progression and increased in the retrosplenial cortex in the pre-disease and late phases.

These area-specific changes were expected as other disease models have shown area specific alterations in network configuration (Harrington et al., 2015; Lorenzini et al., 2023; Wang et al., 2013). What was unexpected was where the changes occurred, as we initially hypothesized that most network alterations would be concentrated in the sensorimotor regions as SCA8 in both patients and this mouse model has a significant motor component (Cleary, 2001; Daughters et al., 2009). While many motor and sensory regions did show changes in network connectivity, major changes also occurred beyond the sensorimotor cortices in the visual and retrosplenial areas. The cerebellum has coherent activity with several neocortical regions including the barrel (O’Connor et al., 2002), prefrontal (Watson et al., 2014), primary motor, and somatosensory cortices (Heck et al., 2023; Proville et al., 2014) which may explain the wide array of changes. While these area-specific FC changes may result from the pathological and physiological changes in these mice, an alternative explanation is that the network changes are compensatory to preserve normal function. The increased connection strength and matching index may be a mechanism to maintain normal neural function in the sensorimotor system. While the mechanisms of cerebellar control of neocortical processing as well as SCA8’s impact on them remain unclear, our results demonstrate that profound alterations in neocortical processing occur in this canonically cerebellar disease.

4.5. SCA8 genotype can be decoded using neocortical functional network alterations

Functional connectivity is being used as a biomarker for neurological diseases and diagnostic tool for disease progression (Harrington et al., 2015; Paldino et al., 2017), including identifying early changes in pre-symptomatic individuals genetically positive for neurodegenerative diseases (Rizk-Jackson et al., 2011). Based on fMRI, FC has been used with machine learning to successfully classify individuals with major depressive disorder and show that alterations in the dorsal cingulate, prefrontal, and parietal cortices were the most influential areas for classification (Geng et al., 2018). Combined MRI imaging modalities can be used to distinguish pre-onset Huntington’s disease individuals from controls and predict years to disease onset (Rizk-Jackson et al., 2011). In childhood epilepsy, FC network metrics can be used to predict epilepsy duration using a machine learning algorithm (Paldino et al., 2017).

Here, a stepwise GLM was used to predict animal genotype (SCA8+ or NTC) from the CCA matrices used to construct neural networks. The GLM decodes animal genotypes with >95 % accuracy at all disease phases and when genotype data is shuffled, the predictive power of the GLM is lost. Furthermore, only a subset of connections is required for optimal decoding in each disease phase, most concentrated in the primary and accessory sensory cortices and higher visual integration areas. These data suggest that network integration is altered within and across sensory modalities that in turn contributes to motor dysfunction in SCA8 as mice are not able to efficiently process sensory information. Therefore, robust decoding results suggest network topology could be used to stage cognitive impairment and evaluate the efficacy of treatments in SCA8.

4.6. Experimental limitations

While we show that wide-field Ca²⁺ imaging can be utilized to investigate cortex-wide processing changes in health and disease, our study of SCA8 has several limitations. First, due to the mice needing to be fully grown at time of window implantation, we are limited by the age imaging can begin. Due to this and the variability of SCA8+ disease onset, it was difficult to obtain more than 4–6 weeks of imaging data prior to disease onset. Therefore, we were unable to image at earlier time points and assess if changes in neocortical FC precede our empirical definition of the pre-disease phase. Second, due to our small sample size and the length of the experimental protocol, we are unable to investigate sex differences in SCA8 in our current study. However, we do note a difference in the age of SCA8 onset between males and females and see this as a path of inquiry worth investigating in the future.

Lastly, we were unable to analyze the data when segmented by behavior as the SCA8+ animals walked significantly less than NTC animals which skewed behavioral data (data not shown). As disease progressed, the SCA8+ mice showed fewer bouts of spontaneous locomotion. While we acknowledge this as a potential caveat, it would not fully explain the hyperconnectivity in the SCA8+ neocortex. In healthy animals, cortical networks during locomotion show increased functional connectivity and activation globally, that is particularly marked at the onset and end of locomotion (see Pinto et al., 2019; West et al., 2022) compared to the resting state, in sharp contrast with the hyperconnectivity in SCA8+ mice, despite being mostly at rest. This suggests that the changes in FC are a result of SCA8 pathology rather than due to behavior. Even with these limitations, these data provide compelling evidence for neocortical dysfunction in SCA8 that needs to be studied further.

4.7. Implications for network processing in SCA8

Any disruption of network homeostasis can lead to ill-configured information processing that correlates with disease state (Paldino et al., 2017; Werner et al., 2014). One explanation for SCA8+ FC shifting toward a globally hyperconnected state, with increases in clustering of spatially distributed modules and efficiency, is that the network has shifted to a more random topology where cross-regional integration has been diminished as hypothesized for HD and epilepsy (Harrington et al., 2015; Paldino et al., 2017; Van Den Heuvel and Sporns, 2011). This explanation fits with the observed increases in global connection density and strength. This suggests an increase in the number of network shortcuts between distant regions which is also supported by the global increases in efficiency, centrality, transitivity, and number of communities. The result is a noisier network containing anatomically distributed communities resulting in degradation of information transfer and local processing specificity.

Our results suggesting an ill-configured network in SCA8 are consistent with disinhibition data obtained from computational models and bioengineered in vitro neural networks. In healthy macaque and cat cortical networks small-world topology, where spatially localized cliques of nodes/communities are densely connected within the grouping and more sparsely connected to other groups, dominates (Achard and Bullmore, 2007; Sporns et al., 2000). This network configuration allows efficient information transfer with minimal structural wiring costs and easier reprogramming by gating information transfer both within and between modules or communities. In vitro experiments demonstrate that modular networks are supportive of non-uniform conditional information transmission and dynamic transmission timing with inhibition providing powerful gating to transmission (Shein-Idelson et al., 2016). Blockade of inhibition in in vitro networks (producing an aberrant state) dramatically increases information propagation and synchronization across network modules/communities causing the network to function uniformly as a single entity (Shein-Idelson et al., 2016). The SCA8+ mice show properties similar to a disinhibited, more random network having global hyperconnectivity with increased connection strengths. As a result, it is likely that aberrant pathways of information transfer which are not functionally relevant are being potentiated leading to higher efficiency and an ill-configured anatomically distributed community structure.

4.8. Toward a contributing mechanism to SCA8 pathogenesis

This study is the first showing long-range FC alterations in a mouse model of SCA8 at the mesoscale, a level between the microscopic and macroscopic. Both mouse models and human post-mortem tissue studies report cellular level alterations in SCA8 (Ayhan et al., 2018; Daughters et al., 2009; He et al., 2006; Moseley et al., 2006; Zu et al., 2011). These changes are likely due to RAN translation of pathological SCA8 RNAs and sequestration of proteins such as MNBL1 (Daughters et al., 2009) in addition to abnormal protein aggregation in the cerebellum and other brain areas (Ayhan et al., 2018; Moseley et al., 2006; Zu et al., 2011). Expression of pathogenic SCA8 alleles causes upregulation of GABT4, likely through reduced MBNL1 or increased CUGBP1 activity. As a result, the SCA8 cerebellum exhibits hyperexcitability due to increased reuptake of synaptic GABA and reduced inhibitory tone (Daughters et al., 2009; Moseley et al., 2006). As many of the long-range connections between the cerebellar nuclei and other brain regions are glutamatergic, we propose that reduced inhibitory tone would increase excitatory cerebellar outputs and alter neocortical networks. Both our previous and current results support this hypothesis of global GABAergic dysfunction. A case study proposed a similar hypothesis after observing glucose hypometabolism in the cerebellum and neocortex, as well as reduced [11C]-fluzamenil binding brain-wide in an SCA8 patient (Terada et al., 2013). These data suggest a globally altered GABAergic system contributes to SCA8 pathology and contributes to both cerebellar and extra-cerebellar symptoms of SCA8.

5. Conclusions

Taken together our results show that SCA8 has profound extra-cerebellar effects on neural processing. The SCA8-related changes in neocortical processing begin prior to symptom onset, producing a hyperconnected neocortical network with less flexible and less segregated information processing. These changes to neocortical FC could be caused by direct neocortical alterations and/or changes to cerebellar-neocortical communication. Importantly, neocortical changes were able to predict the presence of SCA8 disease prior to symptoms highlighting the importance of brain-wide examinations of function in neurological disease.

Data availability

Data will be made available on request.