Spatial transcriptomics identifies spatially dysregulated expression of GRM3 and USP47 in amyotrophic lateral sclerosis

The mechanisms underlying the selective degeneration of motor neurones in amyotrophic lateral sclerosis (ALS) are poorly understood. The aim of this study was to implement spatially resolved RNA sequencing in human post mortem cortical tissue from an ALS patient harbouring the C9orf72 hexanucleotide repeat expansion to identify dysregulated transcripts that may account for differential vulnerabilities of distinct (i) cell types and (ii) brain regions in the pathogenesis of ALS.


Introduction
Genetics can influence the development of amyotrophic lateral sclerosis (ALS), with both causative genes in familial cases (fALS) and an increasing number of susceptibility genes. fALS cases make up approximately 10-20% of all cases of ALS and over a dozen causative genes have been described [1,2]. The recently identified intronic hexanucleotide repeat expansion in chromosome 9 open reading frame 72 (C9orf72) accounts for 40-50% of fALS [3,4]. Carrying and ubiquitously 1 expressing this repeat expansion is thought to result in neurotoxicity and cell death contributing to the clinical manifestations of ALS through three putative mechanisms: (i) repeat-associated non-AUG (RAN)-translation of the hexanucleotide repeat, resulting in five highly aggregation-prone dipeptide repeat proteins, whose pathological accumulation is thought to cause proteotoxic stress, cytotoxicity and may lead to the sequestration of other aggregation-prone proteins [5], (ii) pathological formation of RNA foci, potentially sequestering functional RNA transcripts and RNA/DNA-binding proteins preventing their normal function [6] and (iii) loss of normal C9orf72 protein function due to haploinsufficiency resulting from the repeat expansion inhibiting normal protein production [7]. However, there is considerable heterogeneity in the penetrance and clinical manifestations of the C9orf72 repeat expansion mutation, where patients who carry the same mutation present with diverse symptoms that span an ALS-frontotemporal dementia (FTD) spectrum [8]. This considerable clinical heterogeneity presents a substantial diagnostic challenge and makes stratification of ALS patients in clinical trials very difficult. However, despite this clinical heterogeneity there are shared pathological pathways in ALS pathogenesis [9].
TDP-43 is an RNA/DNA-binding protein involved in transcriptional regulation and is found at post mortem in the majority of ALS cases to be misfolded and accumulated as cytoplasmic aggregates [10]. The pathological misfolding of TDP-43 is thought to contribute to neurotoxicity through two major pathways: (i) gain of function, due to the presence of toxic aggregates in the cytoplasm that have the ability to sequester other aggregation-prone proteins and impair the balance of proteostasis in the cell and (ii) loss of function, whereby the normally nuclear TDP-43 is sequestered in cytoplasmic aggregates and is no longer able to participate in transcriptional regulation [9]. Clearly, given the common neuropathological finding of TDP-43 aggregates in central nervous system (CNS) tissue at post mortem in a majority of ALS cases, transcriptional dysregulation resulting from TDP-43 loss of function is an important mechanism warranting further investigation. Importantly, once thought of as purely a disorder of motor neurones, ALS is now recognized as showing a complex interplay between neurones and glia [11]. Indeed, glial TDP-43 inclusions are seen in cases of ALS and FTD, and there is evidence supporting active neuroanatomical propagation of molecular processes [12].
RNA sequencing allows transcriptional information to be analysed from ALS patients and compared with control tissue to highlight key pathways involved in the pathological disease process. Currently, RNA sequencing studies use whole-tissue homogenates of specific brain regions and cannot provide in situ information at a cellular level regarding the complex relationship between different cell types due to the lack of spatial resolution. A recently developed technique, called spatial transcriptomics (ST), has been developed that modifies RNA sequencing methodology to allow interrogation of the entire transcriptome while retaining in situ spatial information [13]. This technique has been successfully implemented in ALS mouse models and human spinal cord tissue [14].
Post mortem CNS tissue is subject to high levels of autolysis and ALS CNS tissue is particularly vulnerable to these post mortem changes as patients often die in an acidotic state due to agonal respiratory failure. We therefore chose to examine an area of the brain, the cerebellum, that is better preserved at post mortem compared to other brain regions [15] and only selected tissue with an RNA integrity number (RIN) of greater than 6. The additional benefit of examining the cerebellum is that, in ALS patients carrying the C9orf72 hexanucleotide repeat, there is a high burden of protein misfolding without the corresponding cell death that is seen in other brain areas [16]. This is particularly useful when examining transcriptional dysregulation because in brain areas with substantial cell death, for example in the motor cortex, there will be a number of transcripts whose dysregulation is simply related to cell death and is not related to active disease processes. Our aim in this study is therefore to initially examine the cerebellum to circumvent this 'dying-cell artefact' and then validate any potential candidate transcripts in other brain regions using a complementary technique, BaseScope. By assessing only the granule cell layer of the cerebellum (substantial pathology with no associated cell death), this approach enables us to focus on transcripts associated with active disease rather than artefacts of dying tissue. This is, to our knowledge, the first known application of this technology to deeply phenotyped human C9orf72 ALS cortical tissue with a unique approach to circumvent dying-cell artefact. Furthermore, due to limitations in the spatial resolution of ST (100 lm), we implemented, in parallel, a complementary technique called BaseScope, as a targeted approach with singletranscript, single-cell resolution [17]. Taken together these techniques have allowed us to interrogate the previously unexplored, spatial transcriptome of human post mortem ALS CNS tissue.
The aim of this study was to detect and spatially resolve transcriptional differences between brain regions and their composite cell types in ALS patient neural tissue compared to control tissue, to identify underlying differential cell vulnerabilities to the disease process in ALS, ultimately to inform therapeutic target selection and biomarker development in this field.

Case and tissue selection
All clinical data were collected as part of the Scottish Motor Neurone Disease register (ethics approval from Scotland A Research Ethics Committee 10/MRE00/78 and 15/SS/0216) and all patients consented to the use of their data during life. All post mortem tissue was collected via the MRC Edinburgh Brain Bank (East of Scotland Research Ethics Services [EoSRES] LR/16/ES/ 0084). Use of human tissue for post mortem studies has been reviewed and approved by the Edinburgh Brain Bank ethics committee and the Academic and Clinical Central Office for Research and Development (ACCORD) medical research ethics committee (AMREC). Cases had all undergone whole genome sequencing and repeatprime PCR (for C9orf72 hexanucleotide repeat expansion) to identify genetic diagnoses.
Control cases were age-matched and had no neurological disease during life and no significant neuropathology identified at post mortem examination.
The RNA integrity number (RIN) of the tissues used was ALS: 6.2 and Control: 6.8. A block of brain tissue was cut to an appropriate size, approximately 5-7 mm 2 and sections were cut on a cryostat set to À20°C at 10 lm thickness on to the commercially available library preparation ST slides within the prespecified areas containing the mRNA probes (six areas per slide). One area was left blank as a negative control. Once cut, slides and sectioned tissue were stored at À80°C. All subsequent work was carried out inside a clean laminar flow hood. The slide was heated for 1 min at 37°C. Nine hundred microlitres of freshly diluted 4% paraformaldehyde was added to the surface of the slide ensuring all tissue was covered and incubated at room temperature for 10 min. The slide was then washed by dipping in to 19 PBS (phosphate buffered saline) five times followed by incubation in 500 ll isopropanol for 1 min and the slide was then air-dried. Haematoxylin and eosin (H&E) staining was performed by firstly incubating sections in 500 ll Harris haematoxylin (ThermoFisher) for 4 min and washed by dipping 10 times in RNAase/DNAase free water and 1 dip in acid alcohol (1 % HCl in 70 % ethanol) followed by 10 further dips in RNAase/DNAase free water. Sections were then incubated in 500 ll of bluing buffer (Dako) for 30 sec followed by dipping 10 times in RNAase/DNAase free water. Sections were then incubated in 500 ll of Eosin Y (ThermoFisher) for 4 min followed by dipping 10 times in RNAase/DNAase free water. Slides were then air-dried and incubated at 37°C for 5 min and mounted with a coverslip using 100 ll of 85% glycerol. The slide was then imaged at 409 magnification. The remainder of the procedure was done as previously reported [13]. In brief, following removal of the coverslip, prepermeabilization (5min incubation) and permeabilization (10-min incubation) was performed using pepsin and collagenase, followed by cDNA synthesis and probe cleavage on the slide. Samples collected following probe cleavage were then processed as per library preparation protocol published previously [13] to prepare the probe-mRNA construct for sequencing (involving the addition of the Illumina adapter (read 2) and the third unique molecular identifier (UMI), the array ID). The finished libraries then underwent quality control by assessing integrity and concentration by bioanalyser analysis (high sensitivity chip) and qubit analysis. Following this, four libraries (2 9 ALS and 2 9 control) were submitted to Edinburgh Genomics for 75 paired end sequencing over 1 lane of a flow cell as 1 pooled library on the HiSeq 4000 platform. Sequences were analysed as detailed in the ST pipeline ( [18]; an automated pipeline for spatial mapping of unique transcripts Oxford BioInformatics 10.1093/bioinformatics/btx211). In brief, fastq files for paired reads (read 1 and read 2) were decoupled using their unique well IDs. Analysis firstly involved quality trimming by removing reads with low quality bases and low quality UMIs, removing artefacts and checking AT and GC content. The reads were then mapped to the human reference genome version 89 using STAR software, demultiplexed (removing duplicate reads) using Taggd software and then individual reads were annotated using htseq-count and group annotated by barcode (spot position), gene and genomic location to obtain the read count [18]. The final output included the gene name, read count and spot position in a format that was directly uploaded to the ST viewer software [18] for integration with the two high-resolution images (i) H&E (tissue morphology) and (ii) Cy3 spots (location of transcripts) for spatial analysis. Following this, spatial analysis was performed using the ST viewer software according to the user manual [18]. Spots underlying the granule cell layer were selected based on cell morphology determined by H&E image (spots highlighted by white circles in Figure 1B), spots on the edge of each region were excluded from the analysis, to reduce bias. Analysis was performed in ST viewer comparing the most differentially expressed transcripts based on predetermined spot selections. NB: we only selected spots directly underlying tissue and QC data are presented in Table S2.
Immunohistochemistry p62 staining was performed on serial sections of frozen brain tissue cut at the same time, same temperature and same thickness as ST sections. Sections were placed in a slide holder in running tap water to remove OCT and washed in Tris buffered saline (TBS) for 5 min followed by incubation with mouse-anti-p62 primary antibody (BDbio-610833 150 ug LCK ligand) at a 1 in 1000 dilution in TBS overnight at room temperature. Antibody detection was then performed using Novolink max polymer detection kit. A 5-min wash with TBS was followed by a post primary blocking step for 30 min at room temperature, followed by a 5-min wash with TBS and incubation with Novolink polymer for 30 min at room temperature. A TBS wash for 5 min was then followed by incubation with 3, 3 0 diaminobenzidine tetrahydrochloride (DAB) (50 l chromogen + 1 ml DAB substrate buffer) for 5 min followed by washing in running water for a further 2 min. The sections were then counterstained with Harris haematoxylin for 1 min followed by dipping in lithium carbonate for 30 sec, then a dip in each of 70% ethanol and xylene and mounted with a 24 9 50 mm coverslip using two drops of VectaMount mounting medium. Sections were then imaged at 20 9 magnification on a NanoZoomer.

BaseScope
For a detailed explanation of how the BaseScope technology works please see the manufacturer's website: https://acdbio.com/science/how-it-works.
Brain tissue was taken at post mortem from standardized Brodmann areas (BA) and fixed in 10% formalin for a minimum of 24 h. Tissue was dehydrated in an ascending alcohol series (70-100%) followed by three successive 4-h washes in xylene. Three successive 5-h paraffin wax embedding stages were performed followed by cooling and sectioning of the FFPE (formalin-fixed paraffin embedded) tissue on a Leica microtome in 4 lm sections on to a superfrost microscope slide. BaseScope reagents (Advanced Cell Diagnostics) were used as per manufacturer's guidelines according to the original protocol [17]. In brief, following deparafinization, tissue sections were incubated with hydrogen peroxide for 10 min at room temperature and target antigen retrieval was performed by submerging slides in BaseScope 1X target retrieval reagent at 99°C in a Braun Multiquick FS 20 steamer for 15 min. The tissue was then permeabilized using BaseScope protease III at 40°C for 30 min. Probe hybridization was then performed by incubating the slides with four drops of custom designed BaseScope probe, negative control probe (DapB) or positive control (PPIB) probe for 2 h at 40°C. Following successive probe amplification steps, transcripts were detected using the BaseScope RED detection kit and slides were counterstained using haematoxylin and lithium carbonate. The slides were then cleared in xylene and mounted with a 24 9 50 mm coverslip using two drops of VectaMount mounting medium. Sections were then imaged at 20 9 magnification on a NanoZoomer slide scanner and the number of transcripts per cell was quantified by two independent pathologists.

Statistics
Differential expression was defined as >3-fold difference in expression (median number of transcripts identified per feature) between ALS and control tissue, but not between technical replicates of control tissue. Any transcripts where the overall read count was zero were excluded, to limit zero-inflation, as it is not possible to say whether this was because (i) there was no expression of that gene in that selection, or (ii) whether there was insufficient coverage to detect it or (iii) whether there has been degradation of the transcript. Therefore, transcripts were included as potential hits if they (i) reached statistical significance (compared to control expression), (ii) had a median number of transcripts per feature not equal to zero and (iii) had a greater than 3-fold difference between case and control expression. Due to the extensive nature of our planned validation cohort (multiple brain regions and multiple ALS cases including sporadic ALS (sALS), SOD1 and additional C9orf72 cases), this strict filtering process was performed to enable the greatest confidence in selecting a manageable number of high-value candidate transcripts. Median number of transcripts per feature was calculated by the ST software as stated above, data were not normally distributed and therefore significance was calculated by nonparametric testing using Mann-Whitney U test.
For BaseScope analysis, gene expression was calculated by counting the number of transcripts per cell in three randomly generated fields of view (1 mm 2 ) and then averaged. This was done for each of the twelve cases (three C9orf72 cases, three control cases, three sALS cases and three SOD1 cases) and the mean and standard deviation was plotted for each of the groups. Therefore, each group (Control, C9orf72, sALS and SOD1) plotted on the graph is the mean and standard deviation of three fields of view in three individuals. Differences in expression (number of dots/transcripts per cell; measured by BaseScope) between ALS case and control in Figure 1 were assessed by two-tailed unpaired t-tests following log transformation (x' = log  Table summarizing the clinical data associated with the C9orf72 case selected for spatial transcriptomics evaluation.) Graphs quantifying expression of transcripts (measured by BaseScope) in three additional C9orf72 cases and three additional control cases, demonstrating that only USP47 (A) and GRM3 (B) are consistently dysregulated in all cases, whereas spatial differences in YWHAE expression do not reach statistical significance (C). Error bars indicate standard deviation.
x + 1). Differences in expression in Figures 2-4 and Figure S2 were analysed using 1-way ANOVA (following log transformation), compared to control expression using a Bonferroni post-test correction for multiple testing.

Results
ST analysis identified 16 candidate transcripts with spatially dysregulated expression Spatial transcriptomics analysis was performed on frozen sections of cerebellum from one ALS and one ageand sex-matched control case ( Figure 1A and B). A summary of the clinical data of this case is included in Table 1. Serial sections were stained for p62 to demonstrate that there was evidence of substantial protein aggregation in the granule cell layer of the cerebellum of the ALS case, but not in the control case ( Figure 1A; right panel). Initially, as a quality control (QC) exercise to test the ST analysis pipeline and software, to ensure that the images and transcripts were correctly aligned and that ST analysis can distinguish between different regions of the cerebellum, two QC experiments were performed. The first assessed four spatially conserved transcripts, known to be differentially expressed in each of the granule cell layer, molecular cell layer, Purkinje cells and white matter, were analysed and visualized as a heat map of expression. Spatial transcript expression analysis in this way demonstrated accurate regional specificity ( Figure S1). The second QC experiment involved simulation of a whole-tissue RNA sequencing experiment and subsequent comparison of our data to a previously published whole-tissue data set. A transcriptomic analysis of post mortem whole-tissue from the cerebellum of patients carrying the C9orf72 hexanucleotide repeat expansion was conducted previously, identifying 15 genes that were dysregulated between C9orf72 tissue and control tissue [19]. Comparison of the entire section of our tissue, simulating a whole-tissue experiment, revealed that five of the 15 dysregulated transcripts detected in this previous study were replicated in our analysis (ALAS2, HSPH1, DNAJB1, HSPA1A and HSPA1B; Table S1).
Following these QC experiments, ST analysis was performed comparing control and ALS granule cell layers to identify candidate transcripts that may be differentially expressed accounting for differential vulnerabilities to disease. Only the spots underlying the granule cell layer (spots selected with a white circle; Figure 1B) were included in the analysis. The number of expressed genes identified in each of the sections ranged from 1123 in the ALS tissue and up to 2588 in the control tissue (summary of QC data is presented in Table S2). The number of genes identified was lower than expected, however is in line with other transcriptome analyses done in post mortem ALS CNS tissue [14]. This has been hypothesized to be a result of post mortem autolysis, which primarily affects 3' and 5' ends of the mRNA, affecting the 3' poly-A tag, crucial to the mRNA capture prior to sequencing. Spatial analysis followed by the strict filtering process resulted in the identification of 16 differentially expressed candidate transcripts ( [13] with a higher expression in ALS and 3 with a lower expression) in the granule cell layer of post mortem ALS patient cerebellum (Table 2). Gene ontology (GO) analysis was performed using pantherdb.org to identify common disease pathways between the identified transcripts. This analysis identified six prominent pathways: (i) proteostasis, (ii) RNA/DNA-binding proteins, (iii) mitochondrial metabolism, (iv) extracellular matrix, (v) excitotoxicity and (vi) apoptosis.

BaseScope analysis validated eight candidate transcripts and demonstrated highly specific spatially dysregulated expression
Given the limitations, primarily in coverage, of ST in our tissue we sought to validate the ST data using a complementary technique, BaseScope, which has high sensitivity and specificity for detection of transcripts on a single-cell level in histological sections. Using this technique, we assessed whether any of the sixteen transcripts identified by ST were dysregulated in the same case and control tissues as those used in the ST experiment. A summary of the clinical data of all cases used in for BaseScope validation is included in Table 3. BaseScope analysis showed independent evidence of differential expression for eight of the sixteen transcripts (USP47, C5orf63, COL27A1, CELF1, MTRNR2, UBR7, GRM3 and YWHAE; Figure 1C-E). Differential expression was not confirmed for the remaining eight transcripts ( Figure S2). Furthermore, the added value of retaining spatial resolution was clearly highlighted by the identification of two transcripts (GRM3 and YWHAE), whose expression was highly specifically dysregulated in C9orf72 cerebellum compared to control cerebellum ( Figure 1D and E). The expression of GRM3 is restricted to the interneurones of the granule cell layer and is markedly increased in ALS granule cell interneurones compared to control interneurones in this layer (Figures 1D and 2B). The expression of YWHAE is also spatially dysregulated. Its expression is extensive within the cerebellum, however in control tissue it is predominantly expressed in the granule cells and the Purkinje cells, whereas in the ALS C9orf72 case its expression is strikingly redistributed from the Purkinje cells to the interneurons of the granule cell layer ( Figure 1E). It is highly unlikely that these transcriptional variations would have been identified using a whole-tissue approach or indeed without the retention of spatial resolution.

Spatial dysregulation of GRM3 and USP47 in cerebellum of an extended cohort of ALS C9orf72 patients
Due to cost of the existing ST technologies we had assessed only a single C9orf72 case against a control case, but we sought to validate our observations across an increased number of C9orf72 cases by using BaseScope. We validated the eight candidate transcripts across an additional three C9orf72 cases with an additional three appropriately matched control cases. This analysis revealed that there were two candidate transcripts whose expression was consistently dysregulated in the cerebellum of all three C9orf72 cases compared to the controls: USP47 (Figure 2A) and GRM3 (Figure 2B). The mean expression of USP47 in the granule cells of the cerebellum of C9orf72 patients was statistically significantly higher than that of control cases (P = 0.005; t = 5.459; df = 4). Furthermore, expression of GRM3 in the cerebellar interneurones of ALS C9orf72 cases was found to be over three times that of control interneurones (P = 0.0204; t = 3.724; df = 4). However, when the expression of YWHAE was examined across an extended cohort of C9orf72 and control patients, the difference between C9orf72 patients and control cases did not reach statistical significance ( Figure 2C).

Extended CNS analysis reveals spatial dysregulation of GRM3 and USP47 in the cortex of diverse cohort of ALS patients
Given the consistent validation of GRM3 and USP47 in an extended cohort of ALS C9orf72 patients with two independent techniques (ST and BaseScope) we next sought to extend our transcriptional analysis of these two transcripts to additional brain regions and to other cases of ALS including three cases of SOD1-ALS and three cases of sALS compared to the three C9orf72 cases and three controls. We assessed expression in the spinal cord, motor cortex, prefrontal cortex and cerebellum of all twelve cases (Figure 3). There was no expression of GRM3 in the spinal cord of any of the cases (neither control nor ALS; Figure 3A). However, whilst the transcriptional dysregulation observed in the interneurons of the granule cell layer of the cerebellum was restricted to C9orf72 cases, dysregulation of GRM3 across the cerebral cortex was a phenomenon observed in all ALS cases ( Figure 3A). Expression of GRM3 in the motor cortex and prefrontal cortex was statistically significantly lower in the C9orf72, sALS and SOD1 cases compared to controls ( Figure 3B), despite showing no significant difference in positive and negative control genes ( Figure S3). The expression of USP47 was also found to be highly spatially dysregulated. Whilst the transcriptional dysregulation observed in the granule cells of the cerebellum was restricted to C9orf72 cases, dysregulation of USP47 across the cerebral cortex and spinal cord was observed in all ALS cases ( Figure 4A). Expression of USP47 in the spinal cord, motor cortex and prefrontal cortex was statistically significantly lower in the C9orf72, sALS and SOD1 cases compared to controls ( Figure 4B).

Discussion
Spatial transcriptomics is a powerful tool in the analysis of (i) spatially heterogeneous tissues like the brain and (ii) spatially heterogeneous diseases, like ALS. Our ST analysis identified sixteen candidate dysregulated transcripts from six relevant disease-related pathways. Furthermore, the dysregulation of two of these transcripts, GRM3 and USP47, was confirmed in an extensive cohort of ALS patients, including three sALS cases and three SOD1 ALS cases (all harbouring the p.I114T Scottish founder mutation) in addition to three C9orf72 ALS cases, all compared to three control cases, using a complementary technique, BaseScope.

GRM3 as a potential therapeutic target in ALS
Data in this study demonstrate that gene expression of GRM3 is consistently lower in the prefrontal and motor cortex in C9orf72 repeat expansion, mutant SOD1 and sALS compared to controls. Furthermore, our data indicate divergence of GRM3 expression between ALS subtypes, where specifically GRM3 expression is higher in interneurones in the granule cell layer of the cerebellum in C9orf72 repeat expansion patients.
The GRM3 gene encodes mGlu 3 , a metabotropic glutamate receptor known to modulate glutamate neurotransmission in the central nervous system. Data obtained from control tissue by this study are consistent with previous descriptions of the mGlu 3 receptor in human tissue; GRM3 is expressed in neurons within the cerebral cortex, interneurons/Golgi cells in the cerebellum, but not in spinal cord neurons (reviewed in [20]). Neuronal mGlu 3 receptor expression is largely found at presynaptic terminals, though at extrasynaptic sites not associated with active synaptic zones. This appears to be inherently related to the physiological role of mGlu 3 receptors, where activation by extrasynaptic glutamate overspill negatively regulates presynaptic vesicular release of glutamate by modulation of presynaptic ion channel activity through G-protein signalling cascades [21][22][23][24]. The inhibition or decreased expression of mGlu 3 receptors is therefore consistent with an increased glutamate transmission and excitotoxic scenarios [25,26]. mGlu 3 receptors are also expressed in astrocytes and activation via synaptic glutamate or mGlu 3 agonists generate the release of neuroprotective neurotrophic factors in a rodent mutant SOD1 ALS model [26]. Of note, the reduction of mGlu 3 receptor expression in the prefrontal cortex is more commonly associated with neuropsychiatric diseases including schizophrenia and bipolar disorder [27,28]. These studies suggest there may be shared mechanistic features between schizophrenia and ALS [29]. In this regard, a common convergence of reduced mGlu 3 receptors in ALS and neuropsychiatric disease is potentially highlighted by the fact the mGlu 3 receptor knock out mouse was initially generated to understand neuropsychiatric disease, however, also displays locomotor dysfunction [30,31]. The development of selective pharmacology targeting mGlu 3 receptors in  neuropsychiatric disease may also therefore be relevant to motor aspects of ALS.

USP47 and de-ubiquitinating proteins
As the granule cell layer in C9orf72 post mortem cerebellar tissue is subject to a high burden of protein misfolding [32], it is understandable that the pathway common to many of the candidate transcripts is proteostasis. The upregulation of proteostatic mechanisms such as chaperone proteins and the ubiquitinproteasome system (UPS) to cope with the high levels of protein misfolding may enable these cells to survive despite a high burden of cytotoxic protein misfolding. The UPS is a pathway whereby cells can target unwanted or misfolded proteins for degradation by covalently linking them with ubiquitin molecules. This ubiquitin chain targets the client protein to the proteasome, where the protein is degraded and the ubiquitin molecules recycled. There is evidence to suggest that mechanisms exist to prioritize the degradation of certain proteins over others by increasing the number of ubiquitin molecules attached to the protein [33]. The ability to either (i) reverse the fate of these ubiquitinated proteins, (ii) recycle ubiquitin molecules or (iii) prioritize specific proteins for degradation, relies upon the function of de-ubiquitinating proteins. USP47, found to exhibit lower expression in the cortex of our ALS cohort patients compared to controls, is one such de-ubiquitinating protein. De-ubiquitinating proteins can either fully remove a ubiquitin motif from a protein to prevent degradation, or remove only a part of it, to enable recycling of ubiquitin molecules and/or to allow for the prioritization of certain proteins by removing ubiquitin molecules from proteins with a less urgent need for degradation [33].
Expression of USP47 was also found to be specifically increased in the granule cell layer of the cerebellum of patients with the C9orf72 hexanucleotide repeat expansion compared to controls, but not in the cerebellum of patients with other causes of ALS (sporadic or SOD1 mutation). The cerebellum of C9orf72 ALS patients is also found to have a high burden of protein misfolding, particularly in the granule cell layer, where expression of USP47 is found to be the greatest. There are, however, no clinical manifestations associated with this high burden of protein misfolding. This implies, that whilst there is clearly an underlying pathology, the cells are somehow able to cope without evidence of neurotoxicity seen in other similarly affected areas, such as the motor cortex. The finding that USP47 is increased in these cells could be associated with an increased capacity to recycle ubiquitin and prioritize toxic proteins, like TDP-43, for degradation. Indeed, the cells of the granule cell layer in C9orf72 patients, whilst containing abundant p62 positive protein aggregates, rarely exhibit TDP-43 aggregation [32]. TDP-43 aggregation is seen in other brain regions and has been associated, through multiple neuropathological, in vitro and in vivo studies, with neurotoxicity therein [9]. It is therefore possible that increasing the expression of deubiquitinating proteins, such as USP47, may be a potential therapeutic strategy in ALS patients.
USP47 has also been shown to be directly involved in promoting cell survival through an anti-apoptotic mechanism mediated through its interaction with beta- transducin repeats-containing protein (beta-TRCP; [34]). The observation, therefore, that USP47 is upregulated in the cerebellum, where there is comparatively substantially less cell death, may be unrelated to its role in proteostasis, and may instead be related to its protective interaction with beta-TRCP. Furthermore, USP47 has been shown to play a crucial, regulatory role in axonal growth during development [35], therefore its reduced expression in the cortex could be associated with axonal dysfunction. Further mechanistic studies are clearly warranted to probe these regional differences further. However, de-ubiquitinating proteins are important drug targets in the oncology field, demonstrating good safety and efficacy profiles [33]. It is therefore possible that de-ubiquitinating enzymes (DUBs) could provide a potential therapeutic target in ALS.

Retaining spatial resolution
Our experiments demonstrated specific spatial localization of regionally distinct transcripts ( Figure S1). Furthermore, our simulated whole-tissue evaluation was able to validate five out of fifteen previously identified dysregulated transcripts, that were independently validated by qPCR, from a whole-tissue analysis of C9orf72 patient cerebellar tissue (Table S1; [36]). This implies that spatial transcriptomics is not only able to detect and replicate differences found in whole-tissue experiments, it also has the added benefit of being able to detect, more targeted, hypothesis driven and spatially resolved questions as implemented in our study. Using this approach, we were able to identify two spatially dysregulated transcripts that have previously eluded classical whole-tissue approaches. Moreover, by firstly assessing cerebellar tissue, we were able to overcome previously identified issues regarding death and dying-cell artefact. In motor cortex and/or spinal cord, many transcripts are identified that may be a generic signature of dying cells, rather than a reflection of underlying pathology that could be driving the disease. By initially examining cerebellar tissue, we were able to analyse the transcriptome of cells that clearly have cellular pathology, including cytoplasmic protein inclusions, but are not undergoing cell death, and then validate candidate transcripts in other brain regions using a more targeted approach. Using this approach, we were able to demonstrate differential dysregulation of transcripts across distinct CNS regions, that may account for differential susceptibilities to the underlying disease process.

Study limitations
ST is a new technology that, to our knowledge, has never been used to interrogate human post mortem brain tissue. As with any novel technology there are limitations that must be taken into account when assessing the generated data. Post mortem brain tissue is subject to particularly high levels of autolysis compared to other tissues, which may compromise its RNA integrity. This technique is best implemented as a nonbiased assessment of the spatial transcriptome. Clearly with tissue demonstrating substantial RNA degradation, there could be a bias towards the identification of more robust transcripts potentially skewing the data set. However, given these limitations, we implemented an extensive validation of potential hits using a much larger sample size (including both technical and biological replicates) and a technique that is not subject to the same degradation bias, BaseScope, which (i) can detect partially degraded mRNA and (ii) can be implemented on FFPE tissue, with better mRNA preservation compared to frozen tissue. New refinements in ST technology, such as sequential approaches and/or compilation of larger aligned data sets, may further extend the applications of this technology to the assessment of human neurodegenerative conditions [14,37].

Conclusions
Taken together these data indicate that through maintaining spatial resolution when assessing transcriptional dysregulation, it is possible to identify genes whose dysregulation may be contributing to the differential regional and cell-type specific susceptibility to disease and may shed light upon previously unidentified pathways contributing to the pathogenesis of ALS. drafting process. The authors would also like to thank (i) the MRC Edinburgh Brain Bank for supplying all post mortem brain material and the Scottish MND Register/CARE-MND Consortium for all clinical and demographic data. (ii) The Scottish MND Clinical Specialist, team in discussing and obtaining consent from MND patients for inclusion in these resources. (iii) MND Scotland and the Sylvia Aitken Charitable Trust for funding CS to help to establish the MND Tissue bank (iv) we acknowledge funding from the UK Medical Research Council and Engineering and Physical Sciences Research Council to the Edinburgh/St Andrews Molecular Pathology Node and funding from BioGen to JMG. JMG is also funded by a starter grant for clinical lecturers from the AMS (210JMG 3102 R45620) and brain bank funding is from the MRC (MR/L016400/1).

Author contributions
JG, MRL, SC, CS and TA contributed to (i) the conception and design of the study, (ii) acquisition and analysis of data and (iii) drafting a significant portion of the manuscript or figures. KM, IC and SMP contributed to (i) the conception and design of the study, (ii) acquisition and analysis of data. Table S2. Summary of QC data from spatial transcriptomics sequencing. Table summarizing the QC data from the spatial transcriptomics sequencing. CF1 and CF2 refer to the control sections and ME2 refers to the motor neurone disease section Figure S1. Expression heat map shows localized expression of candidate transcripts demonstrating regional specificity of ST. Figure S2. Six transcripts identified as differentially expressed by ST showed no difference when assessed by BaseScope in the same ALS case and control. Figure S3. Expression of positive (PPIB) and negative (DapB) control probes across all cases shows no statistically significant difference in RNA quantity or quality between cases assessed in our cohort