Mouse Ataxin-2 Expansion Downregulates CamKII and Other Calcium Signaling Factors, Impairing Granule—Purkinje Neuron Synaptic Strength

Spinocerebellar ataxia type 2 (SCA2) is caused by polyglutamine expansion in Ataxin-2 (ATXN2). This factor binds RNA/proteins to modify metabolism after stress, and to control calcium (Ca2+) homeostasis after stimuli. Cerebellar ataxias and corticospinal motor neuron degeneration are determined by gain/loss in ATXN2 function, so we aimed to identify key molecules in this atrophic process, as potential disease progression markers. Our Atxn2-CAG100-Knock-In mouse faithfully models features observed in patients at pre-onset, early and terminal stages. Here, its cerebellar global RNA profiling revealed downregulation of signaling cascades to precede motor deficits. Validation work at mRNA/protein level defined alterations that were independent of constant physiological ATXN2 functions, but specific for RNA/aggregation toxicity, and progressive across the short lifespan. The earliest changes were detected at three months among Ca2+ channels/transporters (Itpr1, Ryr3, Atp2a2, Atp2a3, Trpc3), IP3 metabolism (Plcg1, Inpp5a, Itpka), and Ca2+-Calmodulin dependent kinases (Camk2a, Camk4). CaMKIV–Sam68 control over alternative splicing of Nrxn1, an adhesion component of glutamatergic synapses between granule and Purkinje neurons, was found to be affected. Systematic screening of pre/post-synapse components, with dendrite morphology assessment, suggested early impairment of CamKIIα abundance together with the weakening of parallel fiber connectivity. These data reveal molecular changes due to ATXN2 pathology, primarily impacting excitability and communication.

. Global transcriptome profile of Atxn2-CAG100-KIN mouse cerebellum at pre-onset stage. (A) Volcano plot analysis of all the quantified mRNAs showing significant up-or downregulations beyond 20% fold-change in red and green, respectively. Highly significant and cerebellar pathologyrelevant transcripts among downregulations are colored in magenta. Transcriptome analysis was performed with Affymetrix Clariom D oligonucleotide microarray technology, comparing 3 Atxn2-CAG100-KIN cerebellum samples with 3 age-and sex-matched WT controls. Statistical assessment of the expression data was done using the Affymetrix Transcriptome Analysis Console; (B) Representative diagrams depicting the numbers of all mRNA transcripts measured (20,299), versus those that are dysregulated more than 35%, regardless of statistical significance (253 down, 147 up), with the number of significant alterations among all (2162), and the numbers of nominally significant (p < 0.05) dysregulations in either direction more than 35% (209 down, 94 up); (C) Protein interaction network of the significantly downregulated transcripts (<−1.35 fold) generated with STRING database (https://string-db.org/) revealed a core network with two central nodes of Itpr1 and Camk2a. Many factors in the network contributed to significantly altered pathways and cellular processes identified by Functional Enrichment Analysis function of STRING, and are highlighted in different colors corresponding to the colored bars in the subsequent panel D; (D) Functional Enrichment Analysis results of STRING utilizing GO terms Biological Process, KEGG Pathways, Reactome Pathways, UniProt Keywords and GO terms Molecular Function. Top ten dysregulated pathways and cellular processes are depicted per database for clarity, and complete lists are available in Supplementary  Table S2. Colored bars represent pathways corresponding to similarly colored proteins in panel C. showing significant up-or downregulations beyond 20% fold-change in red and green, respectively. Highly significant and cerebellar pathology-relevant transcripts among downregulations are colored in magenta. Transcriptome analysis was performed with Affymetrix Clariom D oligonucleotide microarray technology, comparing 3 Atxn2-CAG100-KIN cerebellum samples with 3 age-and sex-matched WT controls. Statistical assessment of the expression data was done using the Affymetrix Transcriptome Analysis Console; (B) Representative diagrams depicting the numbers of all mRNA transcripts measured (20,299), versus those that are dysregulated more than 35%, regardless of statistical significance (253 down, 147 up), with the number of significant alterations among all (2162), and the numbers of nominally significant (p < 0.05) dysregulations in either direction more than 35% (209 down, 94 up); (C) Protein interaction network of the significantly downregulated transcripts (<−1.35 fold) generated with STRING database (https://string-db.org/) revealed a core network with two central nodes of Itpr1 and Camk2a. Many factors in the network contributed to significantly altered pathways and cellular processes identified by Functional Enrichment Analysis function of STRING, and are highlighted in different colors corresponding to the colored bars in the subsequent panel D; (D) Functional Enrichment Analysis results of STRING utilizing GO terms Biological Process, KEGG Pathways, Reactome Pathways, UniProt Keywords and GO terms Molecular Function. Top ten dysregulated pathways and cellular processes are depicted per database for clarity, and complete lists are available in Supplementary Table S2. Colored bars represent pathways corresponding to similarly colored proteins in panel C.

Ataxin-2 Pathology Alters the Expression of Ca 2+ Channels and Transporters
We aimed to validate the high-throughput transcriptome data, focusing on Ca 2+ -associated transcripts and their role in ATXN2 expansion-driven disease and in complete ATXN2 loss. This work also extends previous findings that report similar dysregulations of Ca 2+ -associated transcripts in other mouse mutants [59,60]. Targeted expression analyses of the most relevant candidates were performed by reverse-transcriptase quantitative polymerase chain reaction (RT-qPCR). Protein abundance was assessed for promising factors, where specific and sensitive commercial antibodies were available. The levels of various Ca 2+ channels and transporters were quantified in Atxn2-CAG100-KIN cerebella at different stages of disease progression (3 and 14 mo) and were also tested in Atxn2-KO cerebella, in order to distinguish partial loss-of-function effects from progressive aggregation pathology.
Depending on their expression profile at pre-onset and late disease stages, we categorized the transcript changes as "early" (present at 3 mo and persisting at 14 mo), and "late/secondary" changes (occurring at 14 mo). Early changes occurred for ER-membrane channels Itpr1 (encoding the protein IP3R) and Ryr3 (RYR), transporters Atp2a2 (SERCA2) and Atp2a3 (SERCA3), and plasma membrane channel Trpc3 (TRPC3). In contrast, ER-membrane factors Ryr1 (RYR), Stim1 (STIM1), Atp2a1 (SERCA1), plasma membrane carrier Slc8a2 (NCX) and transporter Atp2b2 (PMCA) showed rather late dysregulations in disease process ( Figure 2A). Among the early and late changes, Ryr1, Atp2a1, Atp2a3 and Trpc3 also showed similar dysregulations in Atxn2-KO cerebellum, suggesting a physiological regulation of these transcripts by ATXN2 protein. Plasma membrane channels Orai1 (ORAI1) and Cacna1a (CaV2.1) were found unaltered throughout the disease course or in Atxn2-KO cerebellum ( Figure 2A). Two factors of special interest, ITPR1/IP3R and ATP2A2/SERCA2, were further analyzed at the protein level in cytosolic (extracted by RIPA buffer) and insoluble/membrane-associated (pellet subsequently treated with 2x SDS buffer) fractions of Atxn2-CAG100-KIN mouse cerebellum ( Figure 2B). At the pre-onset stage, both proteins were unaltered in the cytosolic soluble protein fraction, and showed subtle reductions with disease progression at 14 mo. In the more insoluble protein fraction, ITPR1 showed a strong decrease at 3 mo, persisting throughout disease progression until the terminal stage of 14 mo ( Figure 2B, individual fold changes and p-values are available in Supplementary Table  S3). Interestingly, ITPR1 was previously reported to interact exclusively with expanded ATXN2, and to have increased abundance in the insoluble protein fraction of aged Atxn2-CAG42-KIN cerebellum [59]. Similar to ITPR1, ATP2A2 also showed decreased abundance in the insoluble protein fraction at both 3 mo and 14 mo of age ( Figure 2B, Supplementary Table S3).
Overall, the expression data highlight Itpr1, Atp2a2 and Atp2a3 as the early-onset downregulations, which worsen over time in parallel to the expansion-driven aggregation process and are absent in Atxn2-KO cerebellum.
Overall, the expression data highlight Itpr1, Atp2a2 and Atp2a3 as the early-onset downregulations, which worsen over time in parallel to the expansion-driven aggregation process and are absent in Atxn2-KO cerebellum.

Subcellular Ca 2+ Imbalance Promotes Ataxin-2 Relocalization into Stress Granules
Potent dysregulations observed in various ER and plasma membrane associated Ca 2+ factors at early disease stage could lead to the mislocalization of Ca 2+ ions among the subcellular organelles. It is known that Ca 2+ ions are deliberately kept at very high concentrations in the ER for proper chaperone function in the ER lumen and for structural integrity of the cytosolic proteins, which tend to expose their internal hydrophobic residues at high Ca 2+ concentration and start forming aggregates [61]. Thus, it is intriguing to investigate whether an aggregation-prone protein such as ATXN2 is affected by high cytosolic Ca 2+ concentration and changes its subcellular distribution or expression. In order to test this notion, we treated WT and Atxn2-KO primary mouse embryonal fibroblasts (MEFs) with thapsigargin (TG), an ER-stress inducer known to increase Ca 2+ concentration in the Expression analyses of Ca 2+ channels, transporters and associated factors in Atxn2-CAG100-KIN cerebellum throughout the disease course. (A) Transcript levels of various plasma membrane and ER resident Ca 2+ -associated factors in Atxn2-KO (blue bars) and Atxn2-CAG100-KIN versus WT mouse cerebellum at pre-onset stage of 3 mo (pink bars) and terminal stage of 14 mo (red bars) were measured by RT-qPCR; (B) Protein levels of ER resident Ca 2+ channel ITPR1 and transporter ATP2A2 in soluble/cytosolic (RIPA) and insoluble/membrane-associated (SDS) protein fractions of Atxn2-CAG100-KIN mouse cerebellum at pre-onset and final disease stages were determined by quantitative immunoblots. Student's t-test with Welch's correction; 0.05 < p < 0.1 T , p < 0.05 *, p < 0.01 **, p < 0.001 ***. Further information regarding individual fold changes and p-values can be found in Supplementary Table S3.

Subcellular Ca 2+ Imbalance Promotes Ataxin-2 Relocalization into Stress Granules
Potent dysregulations observed in various ER and plasma membrane associated Ca 2+ factors at early disease stage could lead to the mislocalization of Ca 2+ ions among the subcellular organelles. It is known that Ca 2+ ions are deliberately kept at very high concentrations in the ER for proper chaperone function in the ER lumen and for structural integrity of the cytosolic proteins, which tend to expose their internal hydrophobic residues at high Ca 2+ concentration and start forming aggregates [61]. Thus, it is intriguing to investigate whether an aggregation-prone protein such as ATXN2 is affected by high cytosolic Ca 2+ concentration and changes its subcellular distribution or expression. In order to test this notion, we treated WT and Atxn2-KO primary mouse embryonal fibroblasts (MEFs) with thapsigargin (TG), an ER-stress inducer known to increase Ca 2+ concentration in the cytosol by blocking the ATP2A2 (SERCA2) transporter [61]. As hypothesized, TG-mediated high cytosolic Ca 2+ concentration triggered ATXN2 protein to redistribute into PABP-positive SGs ( Figure 3A). The complete loss of ATXN2 did not alter SG dynamics under TG treatment, in agreement with the previous findings under sodium arsenite-driven oxidative stress [34]. In order to ensure that SG formation and ATXN2 relocalization are induced by increased cytosolic Ca 2+ levels per se, and not due to ER-stress response in general, we performed tunicamycin (TM) treatment in parallel, which induces ER-stress via the blockage of N-linked protein glycosylation and accumulation of the un-or misfolded proteins in the ER lumen [61]. This arm of ER-stress induction did not lead to SG formation or any relocalization of ATXN2 ( Figure 3A). In conclusion, the subcellular localization of ATXN2 was found to be specifically modulated by cytosolic Ca 2+ levels, but not by ER stress in general.
As a stress-response factor, ATXN2 is often regulated transcriptionally under stress conditions, mostly towards upregulation [34,36]. Therefore, we analyzed the regulation of Atxn2 expression under increased cytosolic Ca 2+ concentration in a time-and dose-dependent manner. While no change was observed in Atxn2 levels after 1 h of TG treatment, prolonged Ca 2+ imbalance over 6 h led to a decrease in Atxn2 abundance, independent of the drug dosage ( Figure 3B), which is compatible with the notion of delayed protein turnover due to the aggregation of ATXN2 leading to diminished transcriptional resynthesis. The expression of ER stress markers was also quantified under TG treatment to validate successful stress induction and to observe the potential effects of ATXN2 loss in ER-stress response. ER lumen chaperone Bip and further downstream apoptosis promoting factor Chop were readily upregulated upon 1 h of TG treatment, with further inductions at 6 h ( Figure 3B). Splicing of Xbp1 transcript by ER membrane-associated RNase IRE1 is another hallmark of ER stress. Quantification of the unspliced isoform (Xbp1u) revealed a time-dependent decrease, while the spliced isoform (Xbp1s) showed a time-dependent increase in abundance under TG treatment ( Figure 3B). Interestingly, lower dosage of TG at 6 h seemed to induce more of a stress response than the higher dosage suggested by Bip, Chop and Xbp1s levels in WT cells. Quantification of the same transcripts in Atxn2-KO MEFs revealed no initial dysregulation in untreated cells due to ATXN2 loss, although a subtle increase was observed in Xbp1u levels without significance. At 6 h under lower dosage of TG, Bip, Chop and Xbp1s levels showed a significant induction deficit in Atxn2-KO cells, suggesting that ATXN2 is necessary for a maximal ER stress response when Ca 2+ homeostasis is abnormal.
Recent work on ATXN2 revealed its role in mitochondrial dynamics and proteostasis during stress through the modulation of PINK1 and OPA1 levels [45], regulation of mitochondrial protein import [54], and affecting metabolic enzymes involved in TCA cycle, amino acid and fatty acid catalysis [36,47]. In addition, ATXN2 was found to localize at the rER, and regulate two important ER-associated aspects: protein translation and Ca 2+ homeostasis [37,59]. Therefore, we asked whether the contact sites between mitochondria and ER, namely mitochondria-associated membrane (MAM) complex, could be affected from ATXN2 expansion, and somehow initiate the pathology targeting two organelles at once. Expression analyses of important MAM complex components Mcu, Micu1, Micu2, Micu3, Smdt1, Vdac1, Grp75, Mfn1, Mfn2 and Sigmar1 in Atxn2-CAG100-KIN cerebellum revealed no major alteration of this complex, even at the terminal disease phase (Supplementary Figure S1A). Yet, subtle downregulations were observed for many components, which could downgrade the collective activity of the complex. The previously mentioned Ca 2+ channel ITPR1 (Figure 2A,B) also belongs to the MAM complex. Thus, the biggest dysregulation among all MAM complex members at the terminal phase was found to occur for Itpr1 with an early onset expansion-associated downregulation, while the subtle changes of the other members could represent a secondary coping mechanism. The effect of subcellular Ca 2+ imbalance on ATXN2 localization and expression in mouse embryonal fibroblasts. (A) Immunocytochemical assessment of ATXN2 and SG marker PABP in WT and Atxn2-KO MEFs under thapsigargin-(TG, 5 μM, 6 h) or tunicamycin-induced (TM, 10 μg/mL, 6 h) ER stress. ATXN2 relocalization into PABP-positive SGs was observed, solely upon cytosolic Ca 2+ imbalance driven by TG, but not upon blockage of N-glycosylation by TM. Atxn2-KO MEFs did not show a difference in SG formation upon TG treatment. (B) Transcriptional regulation of Atxn2 and ER stress markers under TG treatment, in a time-and dosage-dependent setup. Three different clones of WT and Atxn2-KO MEF pairs were treated simultaneously with 1 μM or 5 μM TG for 1 h or 6 h. Stress response was already visible at 1 h for both TG dosages, and further increased at 6 h. While Atxn2 showed a significant downregulation under TG treatment, ER stress markers Bip, Chop and Xbp1s showed a suppressed induction in the absence of ATXN2 in KO cells (blue bars). Expression data obtained by RT-qPCR; (C) Colorimetric Ca 2+ concentration measurement in cytosolic and organelle-enriched fractions of WT and Atxn2-CAG100-KIN cerebellum at three mo of age. Higher Ca 2+ concentrations were observed in the membrane-encapsulated organelle fraction with no difference between WT and KIN animals. Statistical assessment of the cell culture data was done using 2-way ANOVA with multiple testing corrections. Statistical assessment of the cerebellar Ca 2+ measurement was done using Student's t-test with Welch's correction; p < 0.05 *, p < 0.01 **, p < 0.001 ***, p < 0.0001 ****. Hashtag (#) indicates comparison of KO cells with untreated KO cells, with p < 0.05 Expression data obtained by RT-qPCR; (C) Colorimetric Ca 2+ concentration measurement in cytosolic and organelle-enriched fractions of WT and Atxn2-CAG100-KIN cerebellum at three mo of age. Higher Ca 2+ concentrations were observed in the membrane-encapsulated organelle fraction with no difference between WT and KIN animals. Statistical assessment of the cell culture data was done using 2-way ANOVA with multiple testing corrections. Statistical assessment of the cerebellar Ca 2+ measurement was done using Student's t-test with Welch's correction; p < 0.05 *, p < 0.01 **, p < 0.001 ***, p < 0.0001 ****. Hashtag (#) indicates comparison of KO cells with untreated KO cells, with p < 0.05 #, p < 0.001 ###, p < 0.0001 ####. Further information regarding individual fold changes and p-values can be found in Supplementary Table S3. Taking these findings into account, we questioned the extent of ER stress and associated transcript inductions in Atxn2-CAG100-KIN mouse cerebellum at terminal stage. Expression analyses of ER-resident primary unfolded protein response (UPR) mediators Perk, Ire1 and Atf6 interestingly revealed no transcriptional induction, but decreased abundance (Supplementary Figure  S1B). Their luminal regulator Bip, and direct downstream effectors Xbp1u and Xbp1s were found unaltered. The further downstream effector Chop was found subtly upregulated, together with its transcriptional activator Atf4, which could have arisen from a plethora of stress inputs of different origins (Supplementary Figure S1B). This set of data indicates that, although ATXN2 might acutely modulate ER stress response and the induction of apoptosis, chronic UPR activation is not a prominent aspect of disease pathology in the brain, showing only mild alterations at the terminal stage when numerous stress stimuli converge and cross-activate each other.
Finally, after establishing the effect of cytosolic Ca 2+ concentration on ATXN2 distribution, and the lack of significant ER stress induction at terminal stage, we aimed to see if there is overall subcellular Ca 2+ mislocalization in the intact cerebellum at pre-onset stage. Fresh cerebella were fractionated into cytosolic and membrane-encapsulated organelle rich fragments, and Ca 2+ concentrations of these lysates were measured with a commercially available colorimetric kit. As expected, organelle rich fraction showed a significantly higher Ca 2+ content in comparison to cytosol in both WT and Atxn2-CAG100-KIN samples ( Figure 3C). However, no change was observed by this steady-state assay in Ca 2+ levels or subcellular distribution in KIN samples compared to WT animals, indicating that chronic Ca 2+ mislocalization is not an initial cause of the disease, and that the early onset alterations of several ER, and plasma membrane channels and transporters are able to maintain the balance at this stage.

Pre-Onset Dysregulation of the Ca 2+ /Calmodulin-Dependent Protein Kinase Pathway
Aside from the role of Ca 2+ in the maintenance of ER dynamics and proteostasis, it is also an important secondary messenger in signaling cascades, especially for excitable cells, such as neurons and myocytes. While low cytosolic concentrations normally prevail, bursts of Ca 2+ increase, for instance, due to synaptic impulse, and are recognized as specific input, and activate the Ca 2+ /Calmodulin-dependent protein kinase (CaMK) pathway. CaMKs are involved in a broad spectrum of cellular processes, such as regulation of phosphorylation cascades, gene expression, mRNA splicing, metabolism and cell survival/death [62].
We observed several members of the CaMK family and associated pathway components among the significantly dysregulated transcripts in the global transcriptome data, such as Camk2a, Camkk2, Gria3, Grm4, Plcb3, Inpp5a and Pcp4 (Supplementary Table S1). Targeted expression analyses of these transcripts together with other pathway components were performed in 3 mo and 14 mo Atxn2-CAG100-KIN cerebellum in parallel to Atxn2-KO samples.
Upon extracellular activation of receptor tyrosine kinases (RTKs) or ionotropic glutamate receptors (e.g., GluAs) or metabotropic glutamate receptors (mGluRs), the phospholipase C (PLC) isoforms are activated at the intracellular plasma membrane leaflet and catalyze the cleavage of PIP 2 (phosphatidylinositol-4,5-bisphosphate) into IP 3 (inositol 1,4,5-trisphosphate), as well as DAG (diacylglycerol) [63,64]. IP 3 may directly bind and activate IP3R leading to Ca 2+ flux from ER to cytosol; alternatively, it can be phosphorylated by ITPKA or de-phosphorylated by INPP5A for recycling [65]. On the other hand, DAG signals either directly to plasma membrane Ca 2+ channel TRPC3 or indirectly via activating protein kinase C (PKC) [63]. The ionotropic AMPA receptor subunit Gria3 (encoding GluA3) showed an early and strongly progressive downregulation, and the two mGluR isoforms Grm1 (mGluR1) and Grm4 (mGluR4) showed a subtle yet significant downregulation in Atxn2-CAG100-KIN cerebella already at three mo of age ( Figure 4A). In view of the relevance of Gria3 dysregulation as an early risk marker and later progression marker, these changes were assessed at the protein level. At the pre-onset stage of three mo, GluA3 abundance showed a significant decrease to 74% in the soluble but not the insoluble/membrane-bound protein fraction. At the terminal stage of 14 mo, GluA3 abundance was similarly decreased in the soluble fraction (to 72%), but now also in the membrane-bound fraction to 53% ( Figure 4B).  were quantified by RT-qPCR in Atxn2-KO (blue bars) and Atxn2-CAG100-KIN mouse cerebellum at pre-onset (pink bars) and terminal (red bars) disease stages; (B) Protein levels of GluA3, CaMKIIα, CaMKIV and Sam68 in soluble (RIPA) and insoluble (SDS) fractions of Atxn2-CAG100-KIN mouse cerebellum at pre-onset and final disease stages were determined by quantitative immunoblots. Student's t-test with Welch's correction; 0.05 < p < 0.1 T , p < 0.05 *, p < 0.01 **, p < 0.001 ***, p < 0.0001 ****. Further information regarding individual fold changes and p-values can be found in Supplementary  Table S3. As important downstream membrane-associated signaling factors, the PLC isoforms β3 and β4 (Plcb3 and Plcb4) also showed a subtle downregulation, which became more prominent later in disease pathology. The presence of a similar dysregulation of both transcripts in Atxn2-KO cerebellum indicates their potential modulation by native ATXN2 function ( Figure 4A). PLC isoform γ1 (Plcg1) showed a significant downregulation at three mo of age, and interestingly, no alteration afterwards or in Atxn2-KO samples, which represents a very specific early-pathology marker. Inpp5a and Itpka expression were found significantly downregulated at pre-onset phase, progressively decreasing with disease pathology at 14 mo, and diminished in Atxn2-KO cerebella, representing ATXN2 native function-dependent progression markers. Downstream of DAG, however, PKC subunit δ (Prkcd) showed an expansion pathology-specific downregulation in Atxn2-CAG100-KIN cerebella, starting from pre-onset stage ( Figure 4A). This arm of the signaling pathway revealed that the polyQ expansion influence starts from the plasma membrane with reduced glutamate receptor transcript levels, extends to altered conversion of membrane lipids to DAG and IP 3 , and reaches IP 3 recycling that is probably diminished.
During further signal transduction in the cytoplasm, CaMK is activated upon Calmodulin (CaM) association with four Ca 2+ ions. PCP4 (or PEP-19) was shown to be a regulator of Ca 2+ /CaM association and dissociation dynamics [66]. Activated CaM further initiates a phosphorylation cascade involving CaMK-kinases (CaMKKs), CaMKs and downstream targets. In Atxn2-CAG100-KIN cerebellum, Pcp4 showed a prominent downregulation, starting early and progressing with the disease. Lack of its dysregulation in Atxn2-KO cerebellum indicates an expansion-driven alteration of Pcp4 expression ( Figure 4A). Both CaMKKs, Camkk1 and Camkk2, showed an ATXN2-dependent downregulation to 80% in Atxn2-KO cerebellum, and a decrease to 60% at the late phase of the disease. In addition, Camkk2 showed a significant downregulation at the pre-onset phase, showing an earlier alteration than Camkk1 ( Figure 4A). Due to their thoroughly investigated roles in dendritic spine morphology and synaptic integrity maintenance, we further examined the dysregulations in CaMKII isoforms and CaMKIV expression. Both Camk2a and Camk2b isoforms showed a progressive decrease in expression parallel to the disease pathology, with Camk2a also showing a mild downregulation in Atxn2-KO cerebellum. Camk2g showed downregulation later in disease and Atxn2-KO cerebellum, whereas no significant change was observed in Camk2d levels at any age or in Atxn2-KO ( Figure 4A). Similar to the expression profile of Camk2a, Camk4 also showed prominent downregulation at the pre-onset phase, maintained throughout the disease until the terminal phase, and also mildly downregulated in Atxn2-KO cerebellum ( Figure 4A). Both transcripts represent potent markers of early pathology and disease progression, and were therefore also validated at the protein level. At the pre-onset stage of 3 mo, CaMKIIα showed no dysregulation in the soluble protein fraction, but showed a significant downregulation to 57% in the insoluble/membrane-bound protein fraction. At the terminal stage of 14 mo, CaMKIIα abundance had progressively decreased in both fractions ( Figure 4B). Similarly, CaMKIV abundance was unaltered at the pre-onset phase in soluble fraction, with a significant decrease to 77% in the insoluble fraction. At 14 mo, CaMKIV protein was also found significantly downregulated in both fractions, a decrease in parallel to CaMKIIα and disease progression ( Figure 4B).
Overall, these data suggest that ATXN2 loss modulates intracellular Ca 2+ signaling component levels, especially that of PLCβ and IP 3 recycling enzymes, but does not alter glutamate receptor levels (as indicated by mRNA changes in Atxn2-KO cerebellum). The ATXN2 expansion at 14 mo triggers a stronger dysregulation in the same direction, when compared to the Atxn2-KO, for these Ca 2+ signaling components (namely Plcb3, Plcb4, Inpp4a, Itpka, Camkk1, Camk2a, Camk4), and in addition, it affects the ionotropic and metabotropic glutamate receptors.

Impact of Ataxin-2 Pathology on the CamKIV-Modulated RNA Splicing Factor Khdrbs1/Sam68
Initiation of Ca 2+ /CaM-dependent signaling cascade in cerebellar granule neurons upon frequent stimulation activates CaMKIV and leads to the phosphorylation of its substrates, one of which is the K-homology domain, containing RNA-binding protein Sam68 (gene symbol Khdrbs1). Phosphorylated Sam68 detaches from its target mRNAs, thus differentially regulating their alternative splicing [67]. Thus, any loss of CaMKIV-dependent phosphorylation might influence the nuclear distribution of Sam68, and its association with ribonucleoprotein (RNP) granules, known for their poor solubility in phase separation. One of the Sam68 target transcripts is Nrxn1 pre-mRNA, encoding different isoforms of the pre-synaptic Neurexin-1 protein involved in structural synapse maintenance [68]. Quantitative assessment of Sam68 (Khdrbs1) expression revealed an expansion-specific downregulation at late disease stage in Atxn2-CAG100-KIN cerebellum, without an alteration at the pre-onset stage or in Atxn2-KO ( Figure 4A). Likewise, the two closely related family members of Sam68, namely Slm1 and Slm2 (encoded by Khdrbs2 and Khdrbs3, respectively), also showed no dysregulation in Atxn2-KO cerebella. Only Slm2 (Khdrbs3) showed an expansion-driven downregulation at the terminal disease stage, while Slm1 (Khdrbs2) showed no dysregulation at the transcript level throughout the disease course ( Figure 4A).
At the protein level, Sam68 abundance in the soluble fraction showed no alteration at the early or late disease stages. However, in the SDS fraction that is enriched for insoluble cytosolic aggregates or membrane encapsulated organelles including nuclei, Sam68 abundance was found to be subtly decreased at the age of three mo, but significantly increased up to 150% at the terminal stage of 14 mo ( Figure 4B). This observation is consistent with the normally nuclear localization of Sam68, its potential redistribution into RNP granules and potential interaction with ATXN2. We therefore questioned whether Sam68 could be trapped in cytosolic ATXN2 aggregates like many other nuclear RNA processing proteins such as TDP-43, FUS, TIA-1 [52,69], which would explain its increased abundance at the terminal stage. Immunohistochemical staining of Sam68 and ATXN2 confirmed the nuclear localization of Sam68 in Purkinje and granule neurons of the WT cerebellum at 14 mo of age, while ATXN2 showed a diffuse cytosolic localization (Supplementary Figure S2). In the Purkinje neurons of Atxn2-CAG100-KIN cerebellum, ATXN2 was found clumped into a single large aggregate localized at the entrance of the dendritic arbor, in agreement with a previous report [23]. Sam68 immunostaining in Atxn2-CAG100-KIN cerebellum did not seem to co-localize with these aggregates and was solely detected in the nuclei (Supplementary Figure S2). However, the nuclear Sam68 signal in both Purkinje and granule neurons of Atxn2-CAG100-KIN cerebellum was stronger and more punctate compared to WT (Supplementary Figure S2), indicating the higher nuclear abundance of Sam68, which explains its upregulation in the SDS fraction at this age.
In short, detailed investigation of the glutamate-dependent synaptic signaling and Ca 2+ /CaM-dependent kinase signaling revealed an early defect in Ca 2+ association dynamics of CaM, and reduced the expression of many cascade components at both transcript and protein levels, including GluA3, CaMKIIα and CaMKIV as early risk and progression markers of the disease. Sam68, a downstream target of CaMKIV, was found to be affected later in disease progression with increased nuclear abundance and granular redistribution, but without detectable sequestration into cytosolic ATXN2 aggregates.

Alternative Splicing Profile of Nrxn1 in Ataxin-2 Pathology
It is well known that expanded ATXN2 interacts with other RNA-binding proteins, and prominently with the splice modulator TDP-43. To investigate the potential downstream effects of CaMKIV-Sam68 pathway alterations observed in Atxn2-CAG100-KIN mouse cerebellum starting before disease onset, we focused on the alternative splicing of Neurexin-1 (encoded by Nrxn1), a well-studied structural synapse component. Nrxn1 transcript has six alternative splice (AS) sites, AS1-6 ( Figure 5A), generating a plethora of mature mRNA and protein isoforms exhibiting differential interactions with synaptic cleft or post-synaptic membrane proteins, such as cerebellins, neuroligins, leucine-rich repeat proteins (LRRTMs), dystroglycan and latrophilins [68,70]. The neuronal activity-dependent splicing of Nrxn1, especially at AS4, is governed by Sam68, Slm1 or Slm2 in distinct neuronal populations [71]. In cerebellar granule neurons, Sam68 is the dominant regulator, and enhances the Nrxn1 splicing towards AS4(−) isoform lacking the alternatively spliced exon at this site. Loss of Sam68 activity in cerebellum, or that of Slm1 and Slm2 in different CNS regions, has been shown to significantly diminish splicing activity at this site, therefore reducing the abundance of AS4(−) and increasing AS4(+) isoform [67,72,73]. In parallel to the widely studied AS4 site, we investigated the splicing activity at all Nrxn1 AS sites in Atxn2-CAG100-KIN cerebellum at terminal disease stage, when Khdrbs1 (Sam68) and Khdrbs3 (Slm2) are significantly decreased. (C) Splicing activity ratio at AS2-6 sites of Nrxn1 reveal the missplicing of AS2 and AS3 sites in Atxn2-CAG100-KIN cerebellum at 14 mo. The ratio between spliced (−) to unspliced (+) variants shown in panel B was calculated for each AS site to assess splicing activity at each site. Significantly decreased activity at AS2, and significantly increased splicing at AS3 was observed; (D) Total levels of Nrxn1-3 transcripts measured by RT-qPCR amplifying constitutive exons showed no dysregulation in Atxn2-CAG100-KIN cerebellum at the terminal disease stage. Student's t-test with Welch's correction; p < 0.05 *, p < 0.01 **, p < 0.001 ***. Further information regarding individual fold changes and p-values can be found in Supplementary Table S3. Investigation of total Nrxn1 level and AS4 splicing pattern, as the most studied Nrxn1 splicing event, in Atxn2-KO and pre-onset Atxn2-CAG100-KIN cerebella showed no significant dysregulations (Supplementary Figure S3A), indicating that the splicing alterations emerge and contribute to disease pathogenesis in an expansion-driven manner and rather late in progression.

Morphological Assessment of Purkinje Cell Spines
Having established the early influence of ATXN2 expansion on CaMKIIα and CaMKIV signaling, and its downstream effects on the alternative splicing of presynaptic Neurexin-1 protein, we assessed whether the deficit of these crucial factors for synaptic plasticity and adhesion triggered  [70]; (B) Transcript levels of spliced (−) or unspliced (+) variants of Nrxn1 at AS1-6 show altered splicing in Atxn2-CAG100-KIN cerebellum at 14 mo. Site-specific primers were designed to selectively amplify exon inclusion or excision at a given AS site by RT-qPCR. At AS1 site, only AS1(−) isoform lacking all intermediate exons could be quantified with this method, due to the structural complexity of the region and impossibility of proper primer design for all splice variants; (C) Splicing activity ratio at AS2-6 sites of Nrxn1 reveal the missplicing of AS2 and AS3 sites in Atxn2-CAG100-KIN cerebellum at 14 mo. The ratio between spliced (−) to unspliced (+) variants shown in panel B was calculated for each AS site to assess splicing activity at each site. Significantly decreased activity at AS2, and significantly increased splicing at AS3 was observed; (D) Total levels of Nrxn1-3 transcripts measured by RT-qPCR amplifying constitutive exons showed no dysregulation in Atxn2-CAG100-KIN cerebellum at the terminal disease stage. Student's t-test with Welch's correction; p < 0.05 *, p < 0.01 **, p < 0.001 ***. Further information regarding individual fold changes and p-values can be found in Supplementary Table S3. Excision of the alternatively spliced exon at each AS site produces the "spliced" or (−) isoform, whereas retaining the exon generates the "unspliced" or (+) isoform. Due to high structural complexity of the AS1 site, only the AS1(−) variant lacking all intermediate exons was successfully quantified with RT-qPCR, showing a significant increase in Atxn2-CAG100-KIN cerebellum compared to WT animals, indicative of higher exon excision rate at AS1 ( Figure 5B). While the AS2(−) isoform showed no significant alteration, AS2(+) isoform was found increased in Atxn2-CAG100-KIN animals indicating a tendency towards exon retention at this site. In contrast, exon excision at AS3 site was found to be increased, as supported by significantly high abundance of AS3(−) isoform in Atxn2-CAG100-KIN samples, without a change in AS3(+) levels. All the other splice sites located downstream, namely AS4, AS5 and AS6, showed the increased abundance of all splice variants in Atxn2-CAG100-KIN cerebellum, without a selective preference for (+) or (−) isoforms ( Figure 5B). The ratio of spliced-to-unspliced variant abundance for each AS site reveals the splicing activity (i.e., exon excision rate) at this region. The assessment of excision rates at AS2-6 revealed significantly lower splicing activity at AS2 and significantly higher activity at AS3 in Atxn2-CAG100-KIN cerebellum at terminal disease stage, whereas no alteration of splicing rate was observed at AS4-6 ( Figure 5C). The AS2(−) isoform of Nrxn1 protein was reported to selectively interact with the postsynaptic dystroglycan/dystrophin complex [74,75], and currently there are no studies revealing a selective interaction of Nrxn1 based on AS3 site splicing. Quantification of the total levels of Nrxn1, together with Nrxn2 and Nrxn3, revealed no alteration in Atxn2-CAG100-KIN cerebellum at terminal disease stage ( Figure 5D).
Altogether, these results suggest synaptic connections between granule and Purkinje neurons in Atxn2-CAG100-KIN cerebellum to be affected by abnormal splicing of selective Nrxn1 isoforms, rather than by the steady-state abundance change of all Neurexins.
Investigation of total Nrxn1 level and AS4 splicing pattern, as the most studied Nrxn1 splicing event, in Atxn2-KO and pre-onset Atxn2-CAG100-KIN cerebella showed no significant dysregulations (Supplementary Figure S3A), indicating that the splicing alterations emerge and contribute to disease pathogenesis in an expansion-driven manner and rather late in progression.

Morphological Assessment of Purkinje Cell Spines
Having established the early influence of ATXN2 expansion on CaMKIIα and CaMKIV signaling, and its downstream effects on the alternative splicing of presynaptic Neurexin-1 protein, we assessed whether the deficit of these crucial factors for synaptic plasticity and adhesion triggered morphological correlates for the synapses between granule and Purkinje neurons. Golgi silver impregnation of Atxn2-CAG100-KIN cerebella showed a significant reduction in spine length and spine density at the age of 9 mo, at a stage when motor incoordination had become apparent in each individual mouse ( Figure 6).
Overall, both poles of the parallel fiber synapse onto Purkinje spines were found to be influenced by the ATXN2 expansion; this was prominently reflected by CaMKIV-dependent alternative splicing anomalies in the presynaptic granule neuron, and by CaMKIIα-driven dendritic spine collapse in the postsynaptic Purkinje neuron. Scale bars indicate 20 μm (upper row) and 10 μm (lower row); in order to be able to appreciate differences in spine length and spine density, an inlet was added showing a dendritic segment at higher magnification. This image is a digital zoom-in at a factor of 2. It can be clearly seen that the density is much less in KIN mice, while the spine length is a little smaller. (B) Significant reductions of spine number and length were observed in Atxn2-CAG100-KIN Purkinje dendrites compared to WT (n = 13 WT vs. 14 KIN Purkinje cells, from 4 WT vs. 3 KIN animals). Levene's test was used for evaluating equal data distribution, and ANOVA equals t-test was used to compare WT vs. KIN cells; p < 0.01 **, p < 0.001 ***. Further information regarding individual fold changes and pvalues can be found in Supplementary Table S3. Overall, both poles of the parallel fiber synapse onto Purkinje spines were found to be influenced by the ATXN2 expansion; this was prominently reflected by CaMKIV-dependent alternative splicing anomalies in the presynaptic granule neuron, and by CaMKIIα-driven dendritic spine collapse in the postsynaptic Purkinje neuron.

Molecular Assessment of Glutamatergic Synapse Strength and Adhesion
The morphological atrophy of Purkinje dendrite spines and the altered Neurexin-1 splicing suggest impaired synaptic strength and adhesion. Parallel fiber activity via glutamatergic excitation determines the growth of Purkinje spine postsynaptic compartments, the facilitation of synaptic plasticity, and the interaction stability between pre-and post-synaptic structures. Differentially spliced isoforms of Nrxn1 were previously shown to interact with postsynaptic ionotrophic Glutamate receptor δ2 (Grid2 encoding GluD2) via extracellular cerebellins (Cbln1-4), neuroligins (Nlgn1-3), dystroglycan, LRRTMs and latrophilin, to regulate the long-term potentiation and depression features of the synapse [68]. Systematic analyses of cerebellin isoforms revealed an early dysregulation of Cbln4, with progressive decrease throughout the disease course, and late-onset Scale bars indicate 20 µm (upper row) and 10 µm (lower row); in order to be able to appreciate differences in spine length and spine density, an inlet was added showing a dendritic segment at higher magnification. This image is a digital zoom-in at a factor of 2. It can be clearly seen that the density is much less in KIN mice, while the spine length is a little smaller. (B) Significant reductions of spine number and length were observed in Atxn2-CAG100-KIN Purkinje dendrites compared to WT (n = 13 WT vs. 14 KIN Purkinje cells, from 4 WT vs. 3 KIN animals). Levene's test was used for evaluating equal data distribution, and ANOVA equals t-test was used to compare WT vs. KIN cells; p < 0.01 **, p < 0.001 ***. Further information regarding individual fold changes and p-values can be found in Supplementary Table S3.

Molecular Assessment of Glutamatergic Synapse Strength and Adhesion
The morphological atrophy of Purkinje dendrite spines and the altered Neurexin-1 splicing suggest impaired synaptic strength and adhesion. Parallel fiber activity via glutamatergic excitation determines the growth of Purkinje spine postsynaptic compartments, the facilitation of synaptic plasticity, and the interaction stability between pre-and post-synaptic structures. Differentially spliced isoforms of Nrxn1 were previously shown to interact with postsynaptic ionotrophic Glutamate receptor δ2 (Grid2 encoding GluD2) via extracellular cerebellins (Cbln1-4), neuroligins (Nlgn1-3), dystroglycan, LRRTMs and latrophilin, to regulate the long-term potentiation and depression features of the synapse [68]. Systematic analyses of cerebellin isoforms revealed an early dysregulation of Cbln4, with progressive decrease throughout the disease course, and late-onset downregulations of Cbln2 and Cbln3 (Figure 7). Transcript levels for the postsynaptic effector of NRXN1 and Cerebellins, Grid2, also showed a significant downregulation, which started at the pre-onset phase and further decreased with disease progression (Figure 7). Another group of postsynaptic NRXN1 interactors, Neuroligins, only showed a significant dysregulation of the Nlgn2 isoform in parallel with the disease progression (Figure 7). A glutamatergic trans-synaptic adhesion protein complex, which consists of membrane-associated Adam22, extracellular Lgi1 and Lgi3, and membrane-associated Adam23 on the opposite side of the synaptic cleft, serves also as a clustering platform for other synaptic entities such as K + channels or AMPA receptors [76]. Expression analyses of these complex components revealed an early dysregulation of Adam23 and Lgi1 at 3 mo, with a progressive decrease of Adam23 at 14 mo. Interestingly, Lgi1 showed normal expression at the terminal disease phase in Atxn2-CAG100-KIN cerebellum (Figure 7). Finally, transcript levels of ionotropic glutamate receptor NMDA type 1 (Grin1) and of the IP3R-associated postsynaptic structural component Shank (Shank1, Shank2) showed a significant dysregulation at 14 mo. This is in line with the previous observation of dendritic spine pathology in Purkinje cells at 9 mo. Shank2 also showed an earlier trend towards downregulation at three mo, and a milder dysregulation in Atxn2-KO cerebellum. The postsynaptic receptor scaffold PSD95 (Dlg4) did not show altered expression throughout the disease course (Figure 7), in good agreement with a previous report that PSD95 binding and relocalization are under the control of CamKII [77]. Together, the data highlight extracellular intermediates Cbln2 and Cbln4, and post-synaptic Grid2, Nlgn2, Adam23 and Shank2 transcripts as important targets of ATXN2 aggregation pathology, among many other structural synapse components. A schematic representation of all the signaling cascades and transcripts/proteins studied in the context of this work is depicted in Figure 8.
downregulations of Cbln2 and Cbln3 (Figure 7). Transcript levels for the postsynaptic effector of NRXN1 and Cerebellins, Grid2, also showed a significant downregulation, which started at the preonset phase and further decreased with disease progression (Figure 7). Another group of postsynaptic NRXN1 interactors, Neuroligins, only showed a significant dysregulation of the Nlgn2 isoform in parallel with the disease progression (Figure 7). A glutamatergic trans-synaptic adhesion protein complex, which consists of membrane-associated Adam22, extracellular Lgi1 and Lgi3, and membrane-associated Adam23 on the opposite side of the synaptic cleft, serves also as a clustering platform for other synaptic entities such as K + channels or AMPA receptors [76]. Expression analyses of these complex components revealed an early dysregulation of Adam23 and Lgi1 at 3 mo, with a progressive decrease of Adam23 at 14 mo. Interestingly, Lgi1 showed normal expression at the terminal disease phase in Atxn2-CAG100-KIN cerebellum (Figure 7). Finally, transcript levels of ionotropic glutamate receptor NMDA type 1 (Grin1) and of the IP3R-associated postsynaptic structural component Shank (Shank1, Shank2) showed a significant dysregulation at 14 mo. This is in line with the previous observation of dendritic spine pathology in Purkinje cells at 9 mo. Shank2 also showed an earlier trend towards downregulation at three mo, and a milder dysregulation in Atxn2-KO cerebellum. The postsynaptic receptor scaffold PSD95 (Dlg4) did not show altered expression throughout the disease course (Figure 7), in good agreement with a previous report that PSD95 binding and relocalization are under the control of CamKII [77]. Together, the data highlight extracellular intermediates Cbln2 and Cbln4, and post-synaptic Grid2, Nlgn2, Adam23 and Shank2 transcripts as important targets of ATXN2 aggregation pathology, among many other structural synapse components. A schematic representation of all the signaling cascades and transcripts/proteins studied in the context of this work is depicted in Figure 8.  In Atxn2-CAG100-KIN cerebellum throughout disease course at 3 mo and 14 mo of age and in Atxn2-KO cerebellum, transcript levels of Cerebellin isoforms (Cbln1-4), ionotrophic glutamate receptor δ2 (Grid2) and Neuroligin isoforms (Nlgn1-3) were examined as extracellular and postsynaptic interactors of Neurexins in maintaining synaptic integrity. The structural bridge of glutamatergic synapses consisting of Adam22, Adam23, Lgi1 and Lgi3, together with ionotrophic glutamate receptor NMDA type 1 (Grin1) involved in synaptic transmission, post-synaptic density markers PSD95 (Dlg4) and Shank isoforms (Shank1-2) were also quantified throughout disease course by RT-qPCR. Student's t-test with Welch's correction; 0.05 < p < 0.1 T , p < 0.05 *, p < 0.01 **, p < 0.001 ***, p < 0.0001 ****. Further information regarding individual fold changes and p-values can be found in Supplementary Table S3. NMDA type 1 (Grin1) involved in synaptic transmission, post-synaptic density markers PSD95 (Dlg4) and Shank isoforms (Shank1-2) were also quantified throughout disease course by RT-qPCR. Student's t-test with Welch's correction; 0.05 < p < 0.1 T , p < 0.05 *, p < 0.01 **, p < 0.001 ***, p < 0.0001 ****. Further information regarding individual fold changes and p-values can be found in Supplementary Table S3.

Dissecting the Molecular Signature of SCA2 Pathology to Define Purkinje Neuron Contribution
In order to clarify which dysregulated molecules in the Atxn2-CAG100-KIN cerebellum at pre-onset and terminal disease stages are due to Purkinje pathology rather than granule neuron affection, we evaluated their cerebellar in situ hybridization data from the public Allen Mouse Brain Atlas. Prominent expression in Purkinje neurons was observed for most calcium homeostasis channels/transporter (see compilation in Supplementary Figure S3), the glutamate receptors Grm1 and Grid2, Plcb4, Camk2a, Khdrbs1, as well as the adhesion factors Adam23 and Nlgn2-3 ( Supplementary  Figures S4 and S5). Prominent expression in granule neurons was detected for Plcg1, Camkk2, Camk4, the adhesion factors Cbln1-4 and Nlgn1 (Supplementary Figures S4 and S5).
It is relevant to understand whether the postsynaptic pathology is secondary to abnormal presynaptic input, or cell-autonomous. To address this question, we examined several prominent changes that progressed throughout the disease, using another mouse model of SCA2 with human ATXN2-Q58 transgenic overexpression under control of the Purkinje cell specific Pcp2 promoter [78], shortly named Q58-Tg here. Two stages of disease in the Q58-Tg model were investigated, using 16-week (wk) and 46-wk-old cerebella, in accordance with the behavioral and neuropathology data reported earlier [78]. Among the investigated Ca 2+ transporters, only Itpr1 showed a rather noteworthy downregulation at early disease stage, with no alteration at late stage. Ryr1, Inpp5a, Camk2a, Camk4, Khdrbs1, Nrxn1, Nrxn2, Nrxn3, Nrxn1 AS4 splicing, Grid2 or Adam23 were not found dysregulated at any point in disease duration (Supplementary Figure S6). The only other significant finding was the upregulation of the postsynaptic adhesion factor Nlgn2 at younger age in Q58-Tg cerebellum, which again was not observed in old cerebella. Thus, almost all of the many dysregulations observed in the Atxn2-CAG100-KIN mouse cerebellum appear to occur in granule neurons or to occur in Purkinje neurons due to altered glutamatergic input from granule neurons. Only two of the dysregulations could be attributed to Purkinje neuron events in a cell-autonomous manner.

Discussion
Generation of the Atxn2-CAG100-KIN mouse allowed us to employ cerebellum from this authentic SCA2 model, to define molecular markers of pathology at the pre-onset stage and the terminal stage. The first survey of unbiased global transcriptome profiles and its systematic targeted validation defined a physiological role of ATXN2 for calcium homeostasis and calcium-dependent signaling, while the neurotoxic progressive aggregation of ATXN2 has a stronger impact and affects signaling cascades more broadly, including synaptic strength via glutamate receptors, inositol second messengers, stimulus-dependent allelic splicing, and adhesion factors. The downregulation of about 50 factors in these pathways was assessed regarding progression across lifespan, determining the impact of gain-of-toxic-function versus loss-of-physiological-function of ATXN2 in each case. Prominent disease-associated mRNA markers with significant early downregulation and strongly progressive deficits included Atp2a2, Trpc3, Gria3, Inpp5a, Itpka, Camkk2, Camk2a, Camk2b, Camk4, Cbln4, Grid2 and Adam23. Overall, the emerging scenario indicates that the marked Purkinje neuron degeneration to occur largely within the framework of impaired connectivity and stimuli from the glutamatergic granule neurons in the cerebellum.
Itpr1 mRNA downregulation was significant, but not as strong as expected from the Q58-Tg mouse model, in view of the reported abnormal interaction between expanded ATXN2 and the ITPR1 protein as a potential cause of pathology in Purkinje neurons [79]. Even in the Q58-Tg mouse, the Itpr1 downregulation was not progressive with age. Of course, the disease mechanisms may result from a combination of pathways and protein interactions, the role of key molecules is not mutually exclusive, and additional abnormal interactions of ATXN2 beyond ITPR1 may contribute. Interestingly, two of the other progression markers have also been implicated among the primary causal events in the pathogenesis of SCA2.
Firstly, Sam68, as a K-homology domain-containing RNA-binding signal transduction factor, which is under the control of CaMKIV and of the insulin receptor, was found to co-immunopurify with Ataxin-2 in several species [80]. It was therefore noteworthy that Sam68 protein was elevated in a tissue where ATXN2 aggregates accumulate progressively, and might reflect sequestration into the cytosolic inclusion bodies. Although our immunohistochemical studies observed a granular nuclear redistribution instead, it cannot be excluded that Sam68 is also relocalized to cytosolic stress granules by the expanded ATXN2, where its epitopes may be masked from detection. The sequestration of Sam68 into intranuclear RBP aggregates was shown to be an early disease event and crucial in fragile-X-tremor-ataxia-syndrome (FXTAS), impacting the mRNA splicing/metabolism in dendrites and thus determining the number of glutamatergic synapses [81,82].
Secondly, the Camk2a mRNA may have a crucial role in initial pathogenesis, given that Drosophila melanogaster studies showed Ataxin-2 to modulate olfactory habituation via CaMKIIα, and demonstrated the Camk2a mRNA to copurify with Ataxin-2 in fly head extracts, so it might be a direct mRNA target of Ataxin-2 binding and regulation effects [83]. The impact of fly Ataxin-2 as well as mouse Ataxin-2 on CaMKIIα is phylogenetically conserved, and it may explain the selective atrophy of nervous tissue better than the conserved effect of Ataxin-2 on mTORC1. CaMKIIα is a central hub for the regulation of the electrophysiological long-term potentiation in glutamatergic synapses, for the subsequent biochemical and morphological adaptations of synaptic strength known as plasticity, and for functional consequences such as motor learning [84][85][86][87]. Its progressive deficiency in the Atxn2-CAG100-KIN cerebellum is expected to cause a visible atrophy of the dendritic spines in Purkinje neurons, a process that became conspicuous and statistically significant at the age of 9 mo. In the global transcriptome profile of 3-month-old Atxn2-CAG100-KIN cerebellum, the strongly significant and massive downregulation of several growth-associated factors such as Igfbp5 and Prkcd may represent very early events in this process of synaptic atrophy.
Alternative splicing is a known downstream effect of several known ATXN2 protein interactors, with a contributing role to the neurodegenerative process, such as TDP-43, RBFOX1, and Sam68. In the disease process of SCA2, it is unclear at present what the crucial mRNA targets of such splice anomalies can be. Neurexin-1 (Nrxn1) splicing is under the control of Sam68, its role as presynaptic receptor for the Neuroligins (Nlgn1-3) and the Cerebellins (Cbln1-4), with subsequent effects on postsynaptic GluD2 and Adam23 [88,89], may be responsible for the downregulation of these factors, and all these anomalies together suggest that synaptic adhesion is impaired. Of course, the Camk2a mRNA itself can also be alternatively spliced, and numerous other excitability factors in neural tissue are regulated in dependence on trophic stimuli versus stress, via alternative splicing or by alternative polyadenylation, so extensive studies at genome-wide levels will be necessary to obtain a systematic overview.
A question that remains unclear is the role of calcium homeostasis in this process-is its disruption only a consequence of the altered RNA processing and protein interactions in SCA2, or does it have a causal role promoting the aggregation of expanded ATXN2 protein? Several lines of evidence indicate that ATXN2 expansion affects one of the most important neuronal processes, namely the regulation of calcium (Ca 2+ ) flux and homeostasis, probably at the ER and at mitochondria. Studies in a transgenic mouse model with overexpression of Q58-expanded human ATXN2 in cerebellar Purkinje neurons showed that mutant ATXN2 physically interacts with inositol-1,4,5-trisphosphate (IP 3 ) receptor (IP3R), which was not observed in healthy animals. Mice with transgenic overexpression of Q127-expanded ATXN2 show decreased firing rates of Purkinje neurons that precede motor deficits [90]. Direct interaction of ATXN2 was proposed to increase IP3R sensitivity to activation by IP 3 , and leads to an enhanced cytosolic Ca 2+ burst upon glutamatergic stimulation in primary Purkinje cell culture [79]. Overexpression of the IP 3 phosphatase INPP5A, thus reducing IP3R activation in mutant Purkinje cells, was reported to decrease Purkinje cell death, regulate firing patterns and alleviate motor incoordination [91]. In addition to several other transgenic overexpression models, an Atxn2-CAG42 knock-in (KIN) mouse model was generated, which showed very mild neurological disease signs towards the end of normal mouse lifespan without obvious metabolic alterations [22]. Investigation of Atxn2-CAG42-KIN mouse cerebellum and its comparison to Atxn2-KO mice revealed similar downregulations of Ca 2+ homeostasis pathway components such as Itpr1, Atp2a2 and Inpp5a [59].
Moreover, Atxn2-CAG42-KIN cerebella showed increased IP3R levels in the insoluble protein fraction, suggesting its accumulation in the aggregates. However, co-immunoprecipitation experiments did not show a direct interaction of IP3R with expanded ATXN2, contrasting with previous observations in overexpression mutants [59,79]. Machado-Joseph disease (also known as Spinocerebellar ataxia type 3) is caused by polyQ expansions of Ataxin-3, and it was shown that the mutant disease protein does not undergo aggregation in fibroblasts, induced pluripotent stem cells, glia cells, but only in neurons after stimulation by glutamate in processes that depended on Ca 2+ [92]. In the case of SCA2 pathogenesis, deficient or abnormal stimulation of Purkinje neurons by parallel fibers and erroneous retrograde feedback within the circuit might lead to non-physiological release and reuptake of Ca 2+ within synapses and neuronal soma, thus promoting the aggregation tendency of expanded ATXN2. This notion is supported by our observation that the Ca 2+ modulator thapsigargin, but not the glycosylation modulator tunicamycin enhance the interaction of RNA-binding proteins such as ATXN2 and PABP within granular cytosolic structures, as an unspecific effect on proteins and RNAs that is not blocked by the genetic ablation of Ataxin-2. Although the thapsigargin treatment also caused ER stress features in cultured cells, in brain tissue, the ATXN2 aggregation is clearly progressive, while the ER stress at terminal stages of disease is not prominent. Overall, the availability of drugs that enhance or inhibit glutamate stimulation, inositol signals and calcium homeostasis, combined with the generation of authentic SCA2 animal models, will help us to distinguish primary events of SCA2 pathogenesis from the unspecific downstream affection of cerebellar circuit efficiency.
A final consideration should focus on the physiological function of ATXN2 for these signaling pathways. Given that the genetic ablation or knockdown of Ataxin-2 via the suppression of its physiological roles is able to mitigate and postpone the neurodegenerative process in several disorders (spinocerebellar ataxia type 1, known as SCA1, fronto-temporal lobar degeneration/dementia known as FTLD, amyotrophic lateral sclerosis known as ALS, tauopathies) [55,93,94], it is critical to ask which of the above factors are also modulated in the Atxn2-KO mouse cerebellum. Among the promising markers of disease progression summarized in the first Discussion paragraph, only Trpc3, Inpp5a, Itpka, Camk2a and Camk4 show significant impact by the ATXN2 deficiency, and may therefore be core elements of the neurodegenerative process, while the glutamate receptors and the adhesion factors might represent secondary players. Interestingly, the downregulated expression of these five factors in the Atxn2-CAG100-KIN cerebellum at pre-onset stage is on average as strong as in the Atxn2-KO cerebellum, but at terminal stage, about one third stronger than the Atxn2-KO. This observation suggests that a cerebellar phenotype does not appear in the Atxn2-KO because the deficit is non-progressive and compensated by other molecules within the same pathway, while the Atxn2-CAG100-KIN develops a cerebellar phenotype when the aggregation process sequestrates ATXN2 interactor molecules within the relevant pathway into insolubility, progressively eliminating the compensatory players and challenging multiple stress response pathways (i.e., ERAD, autophagy, mitochondrial dysfunction, energy deprivation) to be activated simultaneously. These mouse data on the role of Ataxin-2 in signal transduction and integration are compatible with the previous fly observations that Ataxin-2 modulates the habituation of sensory input in the olfactory system [83,95]. Sensory habitation was shown to depend also on other RNA-binding proteins, such as FMR1, both in flies and rodents [96,97].
Jointly, the understanding of molecular expression profiles at different ages of Atxn2-CAG100-KIN cerebellum and its comparison to that of Atxn2-KO is essential to elucidate the molecular chain of causality in the onset and progression of the pathology. The precise identification of native ATXN2 target pathways vs. expansion/aggregation effects is crucial. This approach identifies useful molecular markers to assess the benefit of neuroprotective treatment approaches, at a time when the utilization of ATXN2-ASOs for the treatment of SCA2 and ALS in clinical trials is imminent. The data presented here will also help to understand potential side effects of an ATXN2 knockdown approach.

Animals and Genotyping
The generation, housing and genotyping of both Atxn2-CAG100-knock in (KIN) and Atxn2-knock out (KO) lines were described before [23,48]. The generation of ATXN2-Q58-Tg line was described earlier [78], and the genotyping was done according to the published protocol. All animals were housed in individually ventilated cages with 12 h light/dark cycle at the Central Animal Facility (ZFE) of the Goethe University Medical School in Frankfurt am Main, Germany. All animal experiments were performed in accordance with the German Animal Welfare Act, the Council Directive of 24th November 1986 (86/609/EWG) with Annex II, and the ETS123 (European Convention for the Protection of Vertebrate Animals), and were reviewed by the Regierungspräsidium Darmstadt with approval code V54-19c20/15-FK/1083 (approved on 27th of March, 2017).

Transcriptome Screening
Single-stranded cDNA library was generated from DNase treated 1 µg of total RNA (3 WT vs. 3 Atxn2-CAG100-KIN cerebella), using GeneChipTM WT PLUS Reagent Kit (Applied Biosystems, Thermo Scientific, Waltham, MA, USA). Fragmentation and labeling of the cDNA library was performed immediately before hybridization to a Clariom D Array (Thermo Scientific, Waltham, MA, USA). The microarrays were scanned with the Affymetrix GeneChip Scanner, and data analysis was done with the Transcriptome Analysis Console (TAC) 4.0.1 (Applied Biosystems, Thermo Scientific, Waltham, MA, USA) using default parameters. For STRING interaction and pathway enrichment analysis, default parameters were employed.

Cell Culture and Treatments
Different clones of WT and Atxn2-KO murine embryonal fibroblasts (MEFs) were generated and cultured as described before [34], in growth medium consisting of DMEM  Table 1. Each sample was measured in duplicate with the following cycling conditions: 50 • C for 2 min, 95 • C for 10 min, followed by 40 cycles of 95 • C for 15 s and 60 • C for 1 min. Expression data were analyzed using 2 −∆∆Ct method [98] with Tbp as housekeeping gene.

Transcript
Assay ID Transcript Assay ID Transcript Assay ID

Adam22
Mm01316488_m1 For quantification with SYBR ® Green primers (Merck, Darmstadt, Germany), each qPCR reaction consisted of cDNA from 25 ng total RNA, 5 pmol/µL primers, 10 µL qPCR Mastermix Plus for SYBR Green I (Eurogentec, Liège, Belgium) and ddH 2 O up to 20 µL of total volume. All SYBR ® Green primers utilized in this study are listed in Table 2. Each sample was measured in duplicates with the following cycling conditions: 95 • C for 10 min, followed by 40 cycles of 95 • C for 15 s and 60 • C for 1 min, and a melt curve stage of 95 • C for 15 s, 60 • C for 1 min and 95 • C for 15 s. Expression data were analyzed using 2 −∆∆Ct method [98] with Actb as housekeeping gene.

Colorimetric Ca 2+ Measurement
Colorimetric Calcium Detection Assay Kit (Abcam, Cambridge, UK) was utilized for the measurement of total Ca 2+ concentrations in cytosolic and membrane-encapsulated organelle-rich fractions of mouse cerebellum following manufacturer's protocol with minor modifications in the sample lysis step. Fresh cerebellar tissue from 7 WT vs. 6 homozygous Atxn2-CAG100-KIN animals at the age of 3 mo were homogenized with a motor pestle in 5-10x weight/volume amount of low detergent PN Buffer (1x PBS, 1% NP-40, 150 mM NaCl) immediately after dissection. Following centrifugation at 13,000× g for 15 min at 4 • C, supernatant was transferred to a fresh tube, and the pellet was homogenized in high-detergent Urea Lysis Buffer (8 M Urea, 10 mM Tris(2-carboxyethyl)phosphine, 40 mM 2-Chloroacetamide, 100 mM Tris) with a motor pestle, after which samples were sonicated with an ultrasonic homogenizer (Bandelin, Berlin, Germany). Colorimetric Calcium Detection Assay Kit was used according to user manual from this point on, each sample being measured in duplicates. The Ca 2+ standard provided in the kit was used to generate a standard curve and determine the Ca 2+ concentrations of unknown samples. Densitometric measurements were performed with Spark ® multimode microplate reader (Tecan Technologies, Zürich, Switzerland) by measuring the absorbance value at 575 nm. Final concentration calculations were done in Microsoft Excel software following the instructions of the manufacturer.

Silver Impregnation
Silver impregnation was done as described before [100]. Since Purkinje cell morphology differs between lobes, only cells from lobules 1 to 6 have been evaluated. In addition, only cells were selected for analysis, which were stained throughout most of the cell and which were positioned isolated from other stained cells so that the whole dendritic profile was visible. Images were taken with a confocal microscope (TCS SP2, Leica, Wetzlar, Germany), having a 40× oil immersion objective (NA = 1.25) using a zoom factor of 1.8 (lower magnifications) and 6.0 (higher magnifications). Voxel sizes have been set to 0.2 µm (x and y) and 2.0 µm (z) for lower magnifications, and 0.06 (x and y) and 0.5 (z) for higher magnifications. Z stacks were imported into ImageJ (v 1.53c) and an extended depth of field calculation was done using ImageJ Plugin of Forster et al. [101]. Three animals have been prepared from each genotype, at least 10 sections from the vermal region have been stained and 1-2 Purkinje cells of each slice have been selected and photographed. For each Purkinje cell 3-4 dendritic regions have been selected and spines were measured along a segment which was defined by the following criteria: (i) it is part of the terminal endings, (ii) it is visible along a length of at least 5 µm, and (iii) the segment was visible within 10 z planes. Using the segmented line tool in ImageJ, the lengths of spines were measured and combined for each Purkinje cell. In order to validate measurement, images of unknown genotype to the investigator were analyzed by three different experimentators (NC, SLB and NES) and results were compared. There was a 95% confidence of obtaining the same results. For statistical evaluation, equal data distribution was assessed with Levene's test, and one-way ANOVA was used to compare genotype dependent spine number and length differences.

Statistical Analyses
Unless specified otherwise, all statistical tests for comparisons between WT and mutant mice (except for silver impregnation analyses) were performed using unpaired Student's t-test with Welch's correction, and for cell culture experiments using 2-way ANOVA with multiple testing corrections on GraphPad Prism software version 7. The fold-change differences and p-values of all the expression analyses performed in mouse tissue and cell culture are listed in Supplementary Table S3. Graphs display mean values with standard error of the mean (s.e.m.). Values p < 0.05 were considered significant and marked with asterisks p < 0.05 *, p < 0.01 **, p < 0.001 ***, p < 0.0001 ****. T indicates a trend towards statistical significance (0.05 < p < 0.1).

Conclusions
The Atxn2-CAG100-KIN mouse holds the advantage of modelling ATXN2-dependent neurological disease, not only in a targeted neuron population, but also in the whole organism. Studying different stages of disease progression in this SCA2 model, and especially the primary pathogenesis events at pre-onset phase, is crucial to dissect the causal chains of molecular dysregulations as the crucial effectors of pathology, and to identify potential targets for preventive therapeutic interventions. The global transcriptome profile of Atxn2-CAG100-KIN cerebellum at pre-onset phase highlights Ca 2+ homeostasis and associated downstream effectors, such as CaMK signaling and glutamatergic neurotransmission, as the prominent targets of early-stage pathology. Progressively altered expression levels of various Ca 2+ channels and transporters indicate an imbalanced Ca 2+ localization between cytosol and ER. Normally diffuse cytosolic ATXN2 protein was found to relocalize into stress granules upon thapsigargin-triggered ER stress via enforced Ca 2+ imbalance, but not upon ER stress via blocking glycosylation. In accordance with their dependence on Ca 2+ homeostasis, CaMKIIα and CaMKIV signaling pathways and their molecular outcomes, i.e., post-synaptic dendritic spine morphology in Purkinje neurons and pre-synaptic alternative splicing of Neurexins in granule neurons, were found to be affected by ATXN2 pathology. These initial findings were further supported by subsequent dysregulations of numerous pre-and post-synaptic adhesion factors at the granule cell-Purkinje neuron interface, suggesting a simultaneous onset and progression of pathology in different neuron types in the cerebellum.

Acknowledgments:
We are grateful for the excellent experimental assistance of Gabriele Köpf, and the kind technical help provided by Elif Fidan and Annett Wilken-Schmidt.

Conflicts of Interest:
The authors declare no conflict of interest. Activating Transcription Factor 4 Atf6