Single cell spatial transcriptomic and translatomic profiling of dopaminergic neurons in health, ageing and disease

1. Oxford Parkinson's Disease Centre and Department of Physiology, Anatomy and Genetics, University of Oxford, Dorothy Crowfoot Hodgkin Building, South Parks Road, Oxford, OX1 3QU, UK. 2. Kavli Institute for Nanoscience Discovery, University of Oxford, Dorothy Crowfoot Hodgkin Building, South Parks Road, Oxford, OX1 3QU, UK. 3. Aligning Science Across Parkinson's (ASAP) Collaborative Research Network, Chevy Chase, MD, USA. 4. BGI-Shenzhen, Shenzhen 518083, China 5. Lead contact


Summary
The brain is spatially organized and contains unique cell types, performing diverse functions, and exhibiting differential susceptibility to neurodegeneration. This is exemplified in Parkinson's disease (PD) with the preferential loss of dopamine (DA) neurons of the substantia nigra pars compacta.
Using a PD transgenic model, we conducted a single-cell spatial transcriptomic and DA neuron translatomic analysis of young and old mouse brains. Through the high resolving capacity of single-cell spatial transcriptomics, we provide a deep characterization of the expression features of DA neurons within their spatial context, identifying markers of healthy and aging cells, spanning PD-relevant pathways. We integrate marker enrichment and GWAS data to prioritize putative causative genes for disease investigation; identifying CASR as a novel regulator of DA neuron calcium handling. These datasets (accessible at spatialbrain.org) represent an invaluable resource for the investigation into spatially governed gene expression in brain cells in health, aging and disease.

Introduction
In Parkinson's disease (PD) there is preferential loss of dopaminergic (DA) neurons of the substantia nigra (SN) pars compacta 1,2 and intracellular accumulation of ⍺-synuclein. Age is the biggest risk factor for PD and SN DA neurons may also be lost in healthy aged individuals [3][4][5] . Overexpression of ⍺-synuclein through locus multiplication causes PD and in vivo overexpression of human ⍺-synuclein in the SNCA-OVX mouse model recapitulates DA neuron loss 6,7 .
Single-cell RNA-sequencing has advanced our understanding of cell-specific expression in complex tissues, such as brain [8][9][10][11][12][13] . To isolate individual cells, the tissue is dissociated, resulting in destruction of the tissue architecture and gene expression artifacts 14 . Spatial transcriptomics preserves this architecture, however current technologies do not consistently achieve singlecell or subcellular resolution at high throughput 15 . Stereo-seq (spatial enhanced resolution omics-sequencing) offers nanoscale resolution spatial expression data, detecting thousands of genes simultaneously 16 .
Single-cell and spatial transcriptomics capture minute quantities of RNA, resulting in lower measurement accuracy compared to bulk RNA-sequencing 17 . Translating ribosome affinity purification (TRAP) enables cell type-specific sequencing with measurement sensitivity comparable to bulk RNA [18][19][20][21][22][23][24] . TRAP mRNA is ribosome-bound and engaged in translation, providing a more accurate indicator of protein abundance 25 .
In this study, we combined the advantages of Stereo-seq and TRAP to characterize the spatial expression signature of individual cells in mouse brain, with a focus on DA neurons. We developed a protocol to segment single cells from Stereo-seq data and subsequently identified 21 cell types in each brain section. By considering the location of each cell, we identified genes . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint with spatially variable expression, such as in SN and ventral tegmental area (VTA) DA neurons. By contrasting DA neuron gene expression with other cell types, we identified putative novel markers, including Slc10a4 and Cpne7. Using long-read sequencing of translating mRNA captured by TRAP, we discovered splice variants specific to DA neurons.
We further investigated whether DA axons harbor actively translating ribosomes. To aid the investigation of novel causative genes in PD, we demonstrated how measures of expression specificity from Stereo-seq and TRAP can be used to prioritize candidate genes of interest from GWAS regions. By this process, we identified a novel role for CASR in regulating intracellular calcium handling in DA neurons. We finally compared aged and young brains, revealing a SNspecific loss of DA neurons and expansion of activated microglia with age. Further, we identified a range of age-related expression changes in DA neurons, spanning multiple PDrelevant pathways.
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint

Integrated transcriptomic profiling in the brain
To combine the advantages of spatial resolution with the sensitivity of the TRAP platform, we generated Rosa26 fsTRAP ::DAT IREScre (DAT-TRAP) mice, which express eGFP-L10a in DATexpressing cells ( Figure 1A) 26,27 . DAT-TRAP mice were crossed with SNCA-OVX mice and aged to 18 months to investigate the effects of overexpression of human alpha-synuclein and aging on DA neuron gene expression ( Figure 1A-B). TRAP samples were prepared from 56 mice, by dissecting the ventral midbrain, dorsal and ventral striatum and incubating each homogenate in paramagnetic beads coated in anti-eGFP antibody (Methods, Figure 1H).
Stereo-seq samples were prepared from 7 mice, by cutting 10 µm cryopreserved sections from fresh frozen brains (Methods).

Stereo-seq single-cell spatial transcriptomic profiling enables the identification and annotation of distinct cell types in the brain
To generate in situ transcriptomic data, cryopreserved mouse brain sections were mounted onto DNA nanoball-patterned arrays for library preparation. All sections were analyzed as a single group (encompassing both age groups and genotypes) to validate Stereo-seq analysis methodology ( Figure 1C, Methods). We produced a spatial map of transcript detection, to visualize the intensity of RNA capture across each brain ( Figure 1C). In each map, we observed distinct anatomical compartments and cell boundaries ( Figure 1C-D). We developed a custom image processing pipeline to segment individual cells from Stereo-seq brain sections ( Figure   1D, Methods and Supplementary Figure 1 for a detailed example). In the first stage, cells were filtered based on the number of detected genes (between 200 and 3000 per cell). We isolated 197,698 high-quality transcriptomes with spatial coordinates from seven mouse brains. 19,240 . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint genes were detected across all brains, and a median of 903 genes were detected per cell. A summary of the quality control workflow and outcomes is shown in Supplementary Figure 2.
Spatial transcriptomics enables the identification and annotation of distinct cell types using both expression and location features. We divided the entire dataset into training and testing subsets (50/50) and performed unsupervised clustering using expression features (Methods).
Twenty-one distinct cell types were identified and could be predicted with robust accuracy in the testing subset (F1 > 0.8, Figure 1E, summary of all identified cell types in Supplementary   Table 1). Cell types were labelled according to both their anatomical localization and the genes most distinctively expressed (marker genes). For example, neurons were distinguished from glia by the expression of Snap25 and by patterned localization in regions, such as the hippocampus or thalamus ( Figure 1F-G). Distinct subpopulations of cells could be visualized, such as neurons of the CA1, CA3, dentate gyrus, and subiculum in the hippocampal region, or GABAergic nuclei within the midbrain. Oligodendrocytes, astrocytes, microglia, and erythrocytes were readily identifiable by marker expression (Oligodendrocytes: Olig1, Mbp, Sox10, Mog; Astrocytes: Gfap, Slc1a3, Atp1a2, Mt3; Microglia: Tyrobp/Dap12, Ftl1, Cts(a/b/d/f/h/l/s/z), Aif1, Tmem119, Cd68; Erythrocytes: Hba-and Hbb-genes). Mapping the identity of every cell to its spatial position of origin gives confidence in ascribing greater annotation detail than with expression data alone, e.g. labelling CA1 vs CA3 hippocampal neurons.

TRAP generates highly sensitive 'translatomic' profile of dopaminergic neurons
TRAP complements Stereo-seq by generating a highly sensitive measure of gene expression in a target cell type. In addition, mRNA captured by TRAP is engaged in translation, providing a more accurate readout of gene expression, compared to conventional transcriptomic technologies 25 . For deep characterization of the DA neuron translatome, short-and long-read . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint sequencing were performed on DAT-TRAP samples ( Figure 1H). The dissected ventral midbrain and striatum were processed, to enrich for cell body-and putative axon-localized transcripts, respectively. Specific expression of the TRAP transgene was confirmed by immunohistochemical staining for eGFP, which showed distinct colocalization to TH; a marker of DA neurons ( Figure 1I). In DAT-TRAP samples, canonical markers of DA neurons were robustly enriched, relative to RNA from bulk tissue homogenate, while markers of other cell types present in the ventral midbrain were depleted ( Figure 1J and Supplementary Figure 2D neuronal analyses could be focused to those most relevant to PD. We sought to identify DA neurons in our Stereo-seq data and to characterize spatially dependent changes in their expression. In total, 2,332 DA neurons were robustly detected across all seven brains ( Figure   2A). Canonical marker genes, Th, Slc6a3 (DAT), Ddc (Dopa decarboxylase) and Slc18a2 (VMAT2) were strongly enriched in DA neurons, relative to other cell types ( Figure 2B).
Highly specific markers were also identified with an underreported role in DA function (e.g. Slc10a4, Cpne7): SLC10A4 is a member of the bile acid transporter family and regulates vesicular uptake of dopamine 28 . CPNE7 is a calcium-dependent phospholipid binding protein that regulates autophagy and axonal/dendritic extension in other cells 29 . Of the top 100 DA . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023.

TRAP reveals the specificity of transcript expression in dopaminergic neurons
To focus on genes actively translated in DA neurons, a measure of gene enrichment was calculated by comparing the abundance of transcripts in DAT-TRAP mRNA and bulk tissue . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is  Figure 3C). We leveraged the combination of short-and long-read sequencing technology to profile the specific splice variants that define DA neurons: 1,617 alternatively spliced genes were detected, relative to ventral midbrain RNA (Supplementary Figure 3E).
Interestingly, splicing was not restricted to genes enriched in DAT-TRAP samples: 817 genes demonstrated evidence of differential transcript usage without gene-level enrichment, suggesting that a substantial component of cell type-specific function could be conferred by splicing and not relative gene-level abundance.
We hypothesized that DA axons locally translate mRNA, due to their extensive projection length into the striatum. We used TRAP to capture putative axonal mRNA from the dorsal and ventral striatum ( Figure 1H). In striatal DAT-TRAP samples, we observed an enrichment of 1,803 genes (FDR-adjusted P < 0.01), including canonical DA markers, Th and Slc6a3 (DAT) (Supplementary Figure 5A). We compared our enrichment data with a previously reported proteomic characterization of the striatal DA axonal compartment and found significant overlap between enriched genes/proteins (Hypergeometric test, P = 1.71e -29 ). The abundance of DA neuron marker genes in striatal DAT-TRAP samples was substantially lower than in midbrain-derived samples, however. We also observed the enrichment of markers of cell types other than DA neurons (e.g. Gfap, Gad1, Gad2). We performed immunohistochemical staining for TH and GFP in mouse brain sections at the level of the striatum (Supplementary Figure   . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint 5B). GFP puncta could be identified that colocalized with TH, however the overall signal was sparse.

Heritability enrichment analysis identifies CASR as a novel regulator of intracellular calcium handling in dopaminergic neurons
Using Stereo-seq and TRAP data, we designed an approach to prioritize candidate genes for sporadic PD investigation. We reasoned that enrichment and specificity measures could be used to partition PD heritability according to causative cell types, as demonstrated previously 33,34 . We measured the cell type-specific enrichment and specificity index (Methods) of genes containing SNPs at an r 2 > 0.5 and located within ±1 Mb of 107 common risk variants for sporadic PD 35 . 248 out of 303 genes were considered, after retaining genes with mouse homologs (Supplementary Table 3). We observed a broadly neuronal pattern of candidate gene enrichment in Stereo-seq data ( Figure 3A), with SN DA neurons demonstrating the greatest median enrichment. Candidate genes were generally depleted in glial cell types, however Ctsb, Dpm3, Inpp5f, Rps12, Sbds, Scarb2 and Stx4a were commonly enriched between glia and DA neurons ( Figure 3A). In TRAP data, DA neurons and oligodendrocytes were jointly found to specifically express the greatest number of candidate genes ( Figure 3B). Together, our results indicate a primary role for DA neurons in conferring genetic risk of sporadic PD, however the common enrichment of a minority of genes across distinct cell types also supports cell typeagnostic disease processes, as reported previously 36 .
We sought to demonstrate how cell type-specific gene expression could be used to prioritize candidate genes containing variation in linkage disequilibrium with lead PD SNPs. We observed that for 63 out of 91 testable PD GWAS loci, the lead SNP localized to a gene not considered specific or enriched in DAT-TRAP samples. Notable exceptions included rs356182, rs356203, rs256228, rs5019538 (SNCA), rs620513 (FGF20), rs11158026 (GCH1) . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint and rs649339 (SYT17), in which the most proximal gene was also the most significantly enriched and specifically expressed. We focused on risk variant rs55961674, an intronic variant of KPNA1. In DAT-TRAP samples, Kpna1 was depleted and in low abundance, relative to bulk ventral midbrain homogenate RNA, indicating low, nonspecific expression in DA neurons.
Three candidate genes within the rs55961674 search window were significantly enriched by TRAP, specifically expressed (compared to other TRAP/RiboTag datasets) and contained variants in linkage disequilibrium with the lead SNP ( Figure 3C). Casr, encoding the calcium sensing receptor, was selected for further investigation, based on demonstrating the most specific expression to DA neurons.
We first validated specific Casr protein expression in DA neurons of the mouse ventral midbrain ( Figure 3D): Intriguingly, we observed a distinctly cytoplasmic pattern of expression specifically in DA neurons, while neighboring cells showed depleted cytoplasmic signal and intense nuclear staining. To confirm this difference, we compared the ratio of cytoplasmic to nuclear Casr intensity and found significantly higher cytoplasmic expression in TH-positive than TH-negative cells of the ventral midbrain ( Figure 3D, Methods).
We next sought to investigate CASR expression and function in human pluripotent stem cell (iPSC)-derived DA neurons. We observed that CASR protein was prominently expressed in TH-positive neurons, with a diffuse cytoplasmic signal ( Figure 3E). To evaluate whether CASR protein was functional and able to modulate intracellular calcium levels and dynamics in iPSC-derived DA neurons generated from a Parkinson's patient carrying an SNCA triplication mutation, we measured cytoplasmic calcium in cells treated for one hour with R568 (10 μM), a positive allosteric modulator of CASR 37 . To estimate the levels of calcium stored in the intracellular compartments, we stimulated the cells with ionomycin (5 μM) and measured the increase in cytoplasmatic calcium using Fura-2AM 38 . The ionomycin-evoked calcium . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint release in the cytoplasm was significantly increased in the iPSC-derived DA neurons from Parkinson's patients and controls acutely treated with R568 ( Figure 3F).
By integrating expression specificity data, GWAS summary statistics and the ability to functionally study gene function in iPSC-derived DA models, we can identify genes with novel function with a putative disease role in DA neurons.

Stereo-seq captures the loss of nigral dopaminergic neurons and a neuroinflammatory expansion of microglia with age
Age remains the most important risk factor for neurodegeneration. We therefore sought to assess aging-related changes in cell number, by comparing the abundance of annotated spatial cell types between old and young brains ( Figure 4A). SN DA neurons were most significantly depleted (FDR-adjusted P = 8.01e -7 ), while an activated population of microglia and oligodendrocytes were most significantly enriched (FDR-adjusted P = 1.37e -8 and 1. 37e -6 respectively). The expansion of activated microglia in aged brains was restricted to the midbrain, corpus callosum and external capsule ( Figure 4B). Immunohistochemical staining for Tyrobp/Dap12 (the most enriched Stereo-seq-derived marker of this microglial population), confirmed an increase in the number of Tyrobp/Dap12-positive cells with a microglial morphology ( Figure 4B). We next tested for differential gene expression between aged groups in each cell type ( Figure   4C). Glial cell types exhibited a more pronounced aging expression change, with 50 genes found to be significantly differentially expressed, compared to 20 in neurons (FDR-adjusted P < 0.05). Furthermore, the absolute magnitude of changes was greater in glia (median absolute log2 fold change: glia, 0.183; neurons, 0.0991). Ribosome-related genes (Apod, Rpl13a, Rps17, Rpl34, Rps21, Rpl37a) were upregulated, in agreement with previous studies of aging changes . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023.

TRAP reveals the extent of age-induced expression changes in dopaminergic neurons
We used TRAP to provide a deeper focus on DA neuron differential gene expression with aging. Greater measurement sensitivity and cohort size led to the discovery of 667 genes with altered expression in aged TRAP samples (399 upregulated, 268 downregulated, FDR-adjusted P < 0.01) ( Figure 4D). Thirteen candidate PD GWAS genes were among the differentially expressed genes, including Gpnmb, also found to be enriched in both DA neurons and oligodendrocytes, and in agreement with previous age-related neuronal findings 8 . We used 1,931 high-confidence protein-protein interactions to subdivide differentially expressed genes into functionally related clusters ( Figure 4E, Methods). Each cluster of genes was distinctly enriched for terms related to neuronal function, with particular importance in PD (e.g. Lysosomal V-ATPase activity, locomotory behavior and synaptic endocytosis). By taking the directionality of expression change into account, gene set enrichment analysis further indicated . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint an upregulation of synaptic and lysosomal-related genes (e.g. Syt1, Syt11, Sv2a, Ap2b1, Atp6v0a1, Atp6v0d1, Atp6v1e1, and Atp6ap1) and a downregulation of mitochondrial and PDrelated genes (mt-Co1, mt-Nd1, mt-Nd4, mt-Nd5, mt-Cyb).

Discussion
In this study, we generated a single cell-level spatial transcriptomic map of gene expression in the adult mouse brain and produced a high-fidelity translatome-level profile of DA neuron gene expression. By integrating these two data modalities, we characterized the distinctive expression features of DA neurons, demonstrated how expression specificity can be used to prioritize candidate causal genes in PD, and examined the changes that occur in the brain and DA neurons specifically with age.
We identified 21 distinct cell types by unsupervised clustering of cells, based on their expression properties. The spatial compartmentalization of distinct cell types could be mapped to anatomical regions, as in the case of neuronal populations of the hippocampus, thalamus, and midbrain. The ability to spatially visualize each cluster aided the annotation of cell types, however as methods of spatial transcriptomic analysis develop, we anticipate the integration of spatial information directly into the clustering process, wherein cells with close spatial proximity or patterned localization (e.g. cortical, intestinal and dermal layers) and matched expression are more considered more similar.
The capture area of each Stereo-seq array is 100 mm 2 in this study; larger than existing available spatial transcriptomic technologies (Visium, 42.25 mm 2 ; Slide-Seq 7.1 mm 2 ) 39-41 , and each Stereo-seq array contains 40 billion capture spots (Visium, 5,000; Slide-Seq, tens of thousands). We leveraged this greater size and density of information to spatially resolve . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint individual cells across the brain. By achieving nanoscale resolution, single-cell spatial data were generated without the need for deconvolution-based methods. We also did not require matched single-cell RNA-seq samples for cell identity annotation.
By integrating Stereo-seq and TRAP data, we identified the expression features most specific to DA neurons in comparison to neighboring cell types. While Stereo-seq provided spatial context, single-cell resolution, and a greater number of unique cell types for comparison, TRAP demonstrated greater measurement sensitivity. By integrating short-and long-read sequencing data from TRAP RNA, we have been able to profile the state of splicing in DA neurons, revealing differential transcript usage across more than a thousand genes. We detected 817 instances of alternative splicing in which individual isoforms were enriched by TRAP, but the overall gene-level count was not. This finding indicates that mouse DA neurons actively translate a larger number of genes than are detectable from only gene-level count data. Our understanding of cell type-specific expression is strengthened by considering transcript-level expression data.
The single-cell resolution of Stereo-seq enabled the distinction of SN and VTA DA neurons in ventral midbrain. Although we detected subtypes within both populations, we have not reported them here, as their identity could not be accurately predicted by gene expression in testing data, according to our stringent F1 score requirement. We expect the addition of further Stereo-seq samples, coupled with improved spatial clustering method development will lead to the robust identification of further DA neuron subtypes in our data.
We used measures of expression specificity from Stereo-seq and TRAP to demonstrate how genes within disease risk loci can be prioritized for investigation. We showed that candidate . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint causal gene expression was most specific to SN DA neurons, supporting the hypothesis that selective vulnerability may be in part conferred by the selective expression of causative genes 33 . We focused on a window surrounding rs55961674 for study, as this SNP falls within the intronic region of KPNA1; a gene considered to not be expressed abundantly or with specificity in DA neurons. By leveraging the sensitivity of TRAP and the diversity of cell types detected in Stereo-seq, we could rank linkage-associated genes surrounding rs55961674 by their expression specificity. The specificity of cytoplasmic Casr expression to DA neurons and demonstration of its functional role in regulating intracellular calcium handling indicates that variation in this gene could contribute to PD disease risk. Long-term functional study of this gene is of interest, such as by generating a DA neuron-selective knockout Casr mouse model.
The large capture area of the Stereo-seq array, combined with single cell-level resolution, enabled parallel region-dependent and cell type-dependent comparison. We identified an agedependent reduction in SN DA neuron cell number. This supports previous findings that DA neuron number declines with age 1,2 and demonstrates the ability of Stereo-seq to identify a subtype of DA neuron that is most vulnerable to aging. In addition, we detected and confirmed the expansion of microglia with age, expressing a variety of pro-inflammatory markers.
Previous investigations into changes in microglial number with age have reported disparate findings, depending on the species and brain region investigated 8,42,43 . However, age-related neuroinflammation has been shown previously to significantly contribute to degeneration in PD 44 .
A surprising finding in this study was the absence of detectable gene expression changes due to SNCA overexpression. The age-dependent pattern of SN DA neuronal loss in SNCA-OVX mice is considered to recapitulate the slow progression of PD pathology in patients. We would . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is We leveraged the sensitivity of TRAP to assess the evidence for axonal translation in DA neurons. We observed enrichment of DA markers, Th, Slc6a3 (DAT), and Slc18a2 (VMAT2) in striatal TRAP samples, however we also observed enrichment of markers of other cell types.
Furthermore, the abundance of DA markers in striatal TRAP samples was markedly lower than in midbrain-derived samples. We detected GFP puncta in striatum that colocalized with TH, although overall signal was sparse.
Hobson et al. previously used RiboTag, a similar technology to TRAP, to study axonal translation in DA neurons and concluded that there was no evidence for the process occurring.
We conclude that we do see evidence for axonal translation in DA neurons by TRAP, although we suggest that the scale of activity is substantially lower than at the level of the cell body. The Together, our spatial transcriptomic and translatome profiling of DA neurons represent a valuable resource to the neuroscience community. Our spatial data can be used to prioritize candidate causal genes involved in conferring genetic risk of brain-related diseases other than PD. In addition, these data can be used as a reference for the development of novel analytical approaches to spatial research. Our TRAP data provides a reference for the querying of genes or isoforms specific to DA neurons in health and with age. We have combined all results from analyses in this study into a database for public access: spatialbrain.org    (Thermo Fisher) and Agilent 2100 Bioanalyzer, respectively.

Long-read sequencing and data processing
Twelve TRAP samples were sequenced using the Oxford Nanopore Technologies MinION platform. TRAP samples were equally divided by age and genotype (N = 3 per age:genotype).

Stereo-seq library preparation and sequencing
Stereo-seq libraries were prepared as previously described by Chen et al. (2022). In brief, Stereo-seq samples were first prepared by collecting postmortem mouse brains and flashfreezing at -80°C. 10 μm tissue sections were collected -3.5 mm from bregma using a Leica   For the analysis, the 340/380 ratio was computed, and then the baseline was subtracted from all the timepoints, to obtain a normalized trace for each well. The maximum intensity (peak amplitude) of the normalized trace was found and the area under the curve (AUC) was calculated using the left rectangular approximation method.
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint A uniform manifold projection (UMAP) was generated for cells from each mouse brain and overlaid, showing a similar pattern of separation. Count data were therefore concatenated into a single expression object, without additional batch correction or integration methods.

Processing of raw Stereo-seq data, quality control and cell type identification
For cell type identification, Leiden clustering was performed at iteratively greater resolutions 54 .
At each resolution, a logistic regression model was fitted on a training subset of data (50 %).
The ability to predict cluster identity within the testing subset (50 %) was assessed by F1-score.
Clusters that could be predicted with an F1-score ≥ 0.8 were considered for further subclustering. Clusters that could not be predicted accurately (F1 < 0.8) were discarded and the . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint identity of included cells was annotated as the parent cluster. If a cluster contained fewer than 200 cells, it was considered final and no further subclustering was performed.
Marker genes for each cluster were identified using both the Wilcoxon rank-sum test and MAST mixed-effect hurdle model, providing two-sided P values. The identity of each cell type was annotated by integrating marker gene data with previous literature and by confirming the spatial distribution of clustered cells.
A second round of cell filtering was performed after clustering to remove cells that were enriched in both Snap25 and Plp1. We suspected that these cells represent neuronal/glial contaminated mixtures, as previously reported in single cell data 55 . After second-round filtering, 197,698 cells remained.

Spatially variable gene detection
Spatially variable genes were identified using SpatialDE2 with default settings. Coordinates and full expression matrices of each brain were supplied, divided by cell type. To combine results from separate samples, a meta-analysis was performed on the raw P values from each sample, using Fisher's method. The Benjamini & Hochberg correction for multiple comparisons was subsequently used and genes were considered spatially variable with an adjusted P value of < 0.01.

Cell type abundance analysis in Stereo-seq data
To test for differential abundance of cell types between age groups, we used mixed effects modelling of associations of single cells (MASC). MASC tests whether cell type membership of individual cells is influenced by an experimental covariate of interest, while accounting for technical covariates and biological variation. We specified mouse genotype as a fixed covariate . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint and mouse of origin as a random effect in the generalized, mixed-effect model. Changes in cell type abundance were considered significant at FDR-P < 0.01.

Differential gene expression analysis
Differential gene expression analysis in TRAP samples (including calculation of gene enrichment and depletion, relative to tissue homogenate RNA) was performed using DESeq2

Specificity index calculation in TRAP/RiboTag samples
To calculate the specificity index of genes detected within TRAP/RiboTag datasets of dopaminergic neurons and other cell types within midbrain, pSI (v1.1) was used with default settings 20, 22,[30][31][32]57 . Genes were considered significantly specifically expressed with an FDRadjusted P value < 0.01.

GWAS prioritization analysis
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is

Statistics and reproducibility
No statistical methods were used to predetermine sample sizes, but our sample sizes for TRAP analyses surpass those reported in previous publications 21,62,63 and our sample sizes for Stereoseq samples are comparable with those of similar spatial transcriptomic datasets 9 . All statistical analyses were performed with R (v4.2.1) and Python (v3.9). All P values were modified to an FDR of 1 % or 5 % as described in the text with the Benjamini & Hochberg method.

Data and code availability
All raw data from TRAP experiment have been uploaded to the Gene Expression Omnibus (GEO accession number: GSE215276). The Stereo-seq raw data that supports the findings of . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is           to a binary mask. 6. "Holes" were filled to improve the masking of cellular regions with lower signal. 7. Watershed to segment each masked region into distinct objects. 8 -9. Inversion of the binary mask and object labelling. 10. Aggregation of gene counts to the cell-level.

Supplementary Figure 2: Stereo-seq and TRAP quality control
A) Histograms of the total counts, number of genes detected, % of counts assigned to the top 100 most expressed genes, % of counts assigned to ribosomal genes and % of counts assigned to mitochondrial genes per brain. Cells were filtered to exclude those containing fewer than 200 and greater than 3000 genes each. B) % of total counts assigned to the top 50 expressed genes across the entire dataset. C) The number of cells retained at each filtering step. During . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint unsupervised clustering, a subset of cells contained both neuronal and glial markers (e.g. Snap25/Plp1). These cells were excluded from further analysis, as their identity could not be confirmed. D) Enrichment score (-log10 P multiplied by log2 fold-change) for cell types likely to be resident within dissected midbrain tissue for TRAP. DA neurons were the only cell type, Volcano plots for all genes significantly differentially expressed due to SNCA overexpression across cell types of Stereo-seq samples.
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted June 4, 2023. ; https://doi.org/10.1101/2023.04.20.537553 doi: bioRxiv preprint