Retinal regions shape human and murine Müller cell proteome profile and functionality

The human macula is a highly specialized retinal region with pit‐like morphology and rich in cones. How Müller cells, the principal glial cell type in the retina, are adapted to this environment is still poorly understood. We compared proteomic data from cone‐ and rod‐rich retinae from human and mice and identified different expression profiles of cone‐ and rod‐associated Müller cells that converged on pathways representing extracellular matrix and cell adhesion. In particular, epiplakin (EPPK1), which is thought to play a role in intermediate filament organization, was highly expressed in macular Müller cells. Furthermore, EPPK1 knockout in a human Müller cell‐derived cell line led to a decrease in traction forces as well as to changes in cell size, shape, and filopodia characteristics. We here identified EPPK1 as a central molecular player in the region‐specific architecture of the human retina, which likely enables specific functions under the immense mechanical loads in vivo.

K E Y W O R D S retina, macula, Müller cells, glial heterogeneity, EPPK1

| INTRODUCTION
A healthy retina is our most important gateway to the outside world, providing us with a major part of our sensory input to orient and interact with our environment quickly and efficiently (Hutmacher, 2019).
For sharp vision in humans, most of the image information is focused on a tiny spot of the retina, the macula. Therefore, any damage in this area has catastrophic effects. The macula with its central fovea is characterized by a pit-like depression in which the somata of the inner retinal cells are displaced laterally so that light can strike the photoreceptors unimpeded. The outer nuclear layer in this region is increasingly dominated by cones. This culminates in the foveola, which contains almost exclusively cones, surrounded and supplied only by processes of a z-shaped subpopulation of Müller cells, the major macroglia of the retina. Finally, the retinal vasculature is completely absent from the fovea  and increased light exposure leads to a higher turn-over in metabolites and the production of reactive oxygen species (Handa, 2012).
Many known pathologies leading to visual impairment are caused by defects in photoreceptor, vasculature, or the retinal pigment epithelium (RPE) functions, which have been intensively studied in recent decades (Bhutto & Lutty, 2012;Lenis et al., 2018;Verbakel et al., 2018). In contrast, despite their numerous important functions, Müller cells are still poorly understood. Originally thought to provide mainly structural support for retinal neurons, it has later been proposed that Müller cells shuttle not only metabolites like pyruvate and lactate, but are, to name just a few examples, also involved in glutamate recycling-similar to brain astrocytes (Hurley et al., 2015;Lu et al., 2006;Toft-Kehler et al., 2018)-or even glutamate release to feedback on neurons (Slezak et al., 2012). Such functions are mediated by a myriad of Müller cell processes contacting all other retinal cell types. In addition, Müller glia, which extend across the entire thickness of the retina, are thought to conduct light through all cell layers to the photoreceptors on the light-averted side of the retina (Franze et al., 2007) and to be responsible not only for the biomechanical stability of the tissue, but also for the formation of the foveal pit Bringmann et al., 2020;MacDonald et al., 2015). In zebrafish, it has been shown that Müller cells can acquire stem cell properties after injury, leading to complete tissue regeneration (Goldman, 2014;Wan & Goldman, 2016). Finally, while the canonical visual cycle involves the multistep enzymatic conversion of all-trans-retinol to 11-cis-retinal in the RPE, an alternative, cone-specific and Müller celldependent pathway has been described recently (Wang & Kefalov, 2011).
Although it is conceivable that the unique conditions of the human macula pose challenges for both photoreceptors and Müller cells, many of these aspects have been studied in model systems that do not have the specifics of this region. This raises the question of how glial metabolism and metabolite exchange adapt to increased energy consumption, increased cone density, and lack of vascularization in the macula. Alternatively, what factors are necessary to provide the extremely long, Z-shaped Müller cells of the macula with the biomechanical properties needed to withstand the high mechanical stresses and tensile forces imposed by the vitreous to preserve the fragile tissue structure of this retinal region (Bringmann et al., 2021).
To approach these questions at a broad molecular level, transcriptomic and especially proteomic profiling of human Müller cells are required. There have been efforts to use bulk RNA sequencing (RNAseq) (Whitmore et al., 2014), which provides excellent depth but is unable to distinguish between cell types, so that only general interregional differences are represented, an obstacle that has been solved by the advent of single-cell (sc)RNAseq (Chambers et al., 2019;Voigt et al., 2021). However, it is important to bear in mind that the final products in the cell are, in most cases, proteins that coordinate cellular metabolism, structural integrity, and intercellular communication.
Many steps influence how much of the final functional protein results from a specific mRNA. These include regulation of mRNA translation, post-translational modifications, oligomerization, and stability issues of both the mRNA and the protein. However, whole retina mass proteomic analysis suffers from similar limitations as bulk RNAseq and single-cell proteomics is still under development (Brunner et al., 2022;Kelly, 2020). To tackle these challenges, we previously established a method to isolate pure, morphologically intact Müller cells from murine (Grosche et al., 2016) or human retina. We compared the proteomes of Müller cells isolated from the R91W/Nrl À/À mice, representing Müller cells in a cone-rich environment similar to the macula, with the proteomes of control mice, representing Müller cells in the peripheral human retina, and of human Müller cells isolated from the macula and periphery. We found differentially expressed proteins converging on specific molecular pathways-a number of which were the same in both species. EPPK1, a selected novel protein candidate, was functionally characterized by studying a respective knockout in

| Mouse lines and donor tissue
The all-cone Rpe65R91W;Nrl À/À (R91W;Nrl) mouse model (Samardzija et al., 2014) and the Rpe65R91W single mutant mice (Samardzija et al., 2008) were bred at a specific pathogen-free barrier animal facility of the Helmholtz Center Munich in accordance with and with allowance by institutional as well as state and federal guidelines (5.1-568-Gas: Allowance to breed and kill animals for scientific purposes).
Eppk1 knockout mice of the C57BL/6 background (Spazierer et al., 2006;Szabo et al., 2015) and C57BL/6 wild-type mice were kept at the pathogen-free mouse facility of the Max Perutz Labs animal facility, Vienna, Austria, in accordance with Austrian Federal Government laws.
Eyes from adult male and female mice (3 to 6 months of age) were collected after animals were sacrificed via cervical dislocation.
Eyes fixed in 4% PFA from Gfap/Vim double knockout mice (3 months of age) were kindly provided by Maria-Theresa Perez (Lund University, Lund, Sweden).
Samples for proteome profiling of human Müller cells were isolated from a set of five donor eyes. The Institutional Review Board at University of Regensburg approved the use of human tissues for this purpose. Five eyes from four non-diabetic Caucasian donors 58-89 years of age (2 males, 1 female, 1 of unknown gender) at a death-to-experimentation interval of <30 h were included in this analysis ( Figure S1, Table S1). Ocular health histories were not available.
Eyes were opened by eye bank recovery personnel using an 18 mm diameter corneal trephine and stored on ice for transfer to the laboratory for further processing.
To stain for EPPK1 in the macular and peripheral retina, human eyes (postmortem time <8-24 h, two eyes from two donors,  (Grosche et al., 2016), different retinal cell types were isolated either from whole murine retina or from 6 mm punches (from macula and from periphery) of human retina using magnetic activated cell sorting. First, the tissue was treated with papain (0.2 mg/ml; Roche) for 30 min at 37 C in PBS/Glucose (12 mM), washed and incubated in DNase I (200 U/ml in PBS/Glucose) for 4 min at RT. PBS/Glucose was removed and substituted with extracellular solution (ECS, 136 mM NaCl, 3 mM KCl, 10 mM HEPES, 11 mM glucose, 1 mM MgCl 2 and 2 mM CaCl 2 , pH 7.4) before dissociating the tissue using a firepolished glass Pasteur pipette. The cell suspension was sequentially depleted of microglia and vascular cells by incubating (15 min, 4 C) with anti-mouse/human CD11b and CD31 microbeads (Miltenyi Biotec), respectively, and passing through LS-columns according to manufacturer's protocol (Miltenyi Biotec).
The resulting suspension was incubated (15 min, 4 C) with anti-CD29 biotinylated antibodies (0.1 mg/ml, Miltenyi Biotec), spun down, washed and the pellet resuspended in ECS containing anti-biotin ultra-pure MicroBeads (1:5; Miltenyi Biotec). The suspension was passed through a LS-column resulting in a neuron-rich flowthrough (CD29 À ), before the bound CD29 + Müller cells were finally eluted from the column. To prepare samples for immunostaining, 100 μl of each of the CD29 + and CD29 À fractions were fixed for 15 min in 4% paraformaldehyde (PFA) at RT, spun down, resuspended in 50 μl PBS and dropped onto a microscope slide.

| LC-MS/MS mass spectrometry analysis
Proteins were proteolysed with LysC and trypsin with filter-aided sample preparation procedure (FASP) as described (Grosche et al., 2016;Wi sniewski et al., 2009). Acidified eluted peptides were analyzed on a QExactive HF or HF-X mass spectrometer (Thermo Fisher Scientific) online coupled to a UItimate 3000 RSLC nano-HPLC (Dionex) as described (Grosche et al., 2016). Briefly, samples were automatically injected and loaded onto the C18 trap cartridge and after 5 min eluted and separated on the C18 analytical column (nanoEase MZ HSS T3, 100 Å, 1.8 μm, 75 μm Â 250 mm; Waters) by a 95 min nonlinear acetonitrile gradient at a flow rate of 250 nl/min. MS spectra were recorded at a resolution of 60,000 with an automatic gain control (AGC) target of 3e6 and a maximum injection time of 30 or 50 ms from 300 to 1500 m/z. From the MS scan, the 10 or 15 most abundant peptide ions were selected for fragmentation via HCD with a normalized collision energy of 27 or 28, an isolation window of 1.6 m/z, and a dynamic exclusion of 30 s. MS/MS spectra were recorded at a resolution of 15,000 with a AGC target of 1e5 and a maximum injection time of 50 ms. Unassigned charges, and charges of +1 and > +8 were excluded from precursor selection.
Acquired raw data were analyzed in the Proteome Discoverer software (versions 2.2 or 2.4, Thermo Fisher Scientific) for peptide and protein identification via a database search (Sequest HT search engine) against the SwissProt Mouse database (Release 2020_02, 17,061 sequences) or the SwissProt Human database (Release 2020_02, 20,435 sequences), considering full tryptic specificity, allowing for up to one missed tryptic cleavage site, precursor mass tolerance 10 ppm, fragment mass tolerance 0.02 Da. Carbamidomethylation of cysteine was set as a static modification. Dynamic modifications included deamidation of asparagine and glutamine, oxidation of methionine, and a combination of methionine loss with acetylation on protein N-terminus.
The Percolator algorithm (Käll et al., 2007) was used for validating peptide spectrum matches and peptides. Only top-scoring identifications for each spectrum were accepted, additionally satisfying a false discovery rate < 1% (high confidence). The final list of proteins satisfying the strict parsimony principle included only protein groups passing an additional protein confidence false discovery rate < 5% (target/decoy concatenated search validation).
Quantification of proteins, after precursor recalibration, was based on intensity values (at RT apex) for all unique peptides per protein. Peptide abundance values were normalized on total peptide amount. The protein abundances were calculated summing the abundance values for admissible peptides. The final protein ratio was calculated using median abundance values of five biological replicates each. The statistical significance of the ratio change was ascertained with ANOVA. For the MIO-M1 data sets, the statistical significance of the ratio change was ascertained employing the t test approach described in (Navarro et al., 2014) which is based on the presumption that we look for expression changes for proteins that are just a few in comparison to the number of total proteins being quantified. The quantification variability of the non-changing "background" proteins can be used to infer which proteins change their expression in a statistically significant manner.

| EV isolation from cell culture media and NTA analysis
After 72 h incubation, serum-free media was collected from both WT and EPPK1 knockout MIO-M1 cells and immediately centrifuged at 300 Â g, for 10 min at 4 C. The resulting supernatant was then centrifuged at 2000 Â g, for 10 min at 4 C. The resulting supernatant was further centrifuged at 10.000 Â g, for 30 min at 4 C. Lastly, the 10.000 Â g supernatant was spun at 100.000 Â g (24.000 rpm, sw40ti rotor) for 70 min at 4 C. The final pellet was resuspended in 30 μl PBS with protease inhibitors (Roche). The cell numbers were assessed for each replicate well.
The final EVs suspension was diluted either 1:300 (WT) or 1:100 (KO cell lines). The measurement was done with the LM10 unit (Nanosight). The diluted samples were recorded with 10 videos, each 10 s long. Data analysis with NTA 3.0 software (Nanosight) was performed with the following settings: detection threshold = 6, screen gain = 2. Particle numbers were adjusted for dilution as well as cell number.

| Immunofluorescence staining and microscopy
Mouse eyes were fixed in 4% PFA for 1 h, cryoprotected, embedded in OCT compound and cut into sections of 20 μm thickness using a cryostat. Human donor eyes were immersion-fixated with 4% paraformaldehyde (PFA) for 48 h. Thereafter, the central part of the eye cup containing the optic nerve head and the macula including the underlying RPE, choroid, and sclera was dissected. The tissue was submitted to cryoprotection, embedded in OCT and cut into 20 μm thick sections. Retinal detachment from the RPE is an artifact commonly observed in cryosections.
Sections were washed (1% bovine serum albumin [BSA] in PBS) and incubated with secondary antibodies (2 h at room temperature;

| qPCR
Total RNA was isolated from peripheral retinal samples of three donors and from wild type MIO-M1 cells using PureLink™ RNA Mini Kit (ThermoFisher, 12183018A) following manufacturer's instructions; 50 ng of total RNA per sample were reverse transcribed with Rever-tAid Reverse Transcriptase (ThermoFisher, EP0441) with the help of random hexamer primers. Primers for qPCR were designed using the Universal ProbeLibrary Assay Design Center (Roche) to be used with the corresponding probes (Table 2). Since EPPK1 mRNA consists of only one translated exon, an exon spanning assay was not possible.
Final expression values were calculated via the ΔCt method by taking the difference between EPPK1 and a house keeper's (PDHB) Ct values and using the result as the power of two.

| Western blot analysis
A 100% confluent 10 cm cell culture dish was used for isolating protein lysates. The cells were washed two times with PBS and subsequently overlaid with 500 μl of ice-cold protein lysis buffer (50 mM HEPES pH 7, 100 mM NaCl, 5 mM MgCl₂, 1 mM EGTA or EDTA, 2.5% Triton X-100, 100 nM DTT, 0.5 mg/ml DNaseI, 0.2 mg/ml RNase A, 1 mM PMSF, protease inhibitors [cOmplete ULTRA tablets, Roche]) before scraping them off. The solution was homogenized, transferred into an Eppendorf tube, and incubated for 5 min at room temperature and lastly sheared via a 27-gauge needle. For retinal samples, the tissue was snap frozen in liquid nitrogen and per 10 mg of tissue 200 μl of tissue protein lysate buffer (Tris pH 7.5; 10 mM NaCl; 150 mM EDTA; 5 mM Triton X-100; 1% SDS; 0.1% NP-40; 1% 1% phosphatase inhibitor cocktails 2 and 3 [Sigma-Aldrich, cat no. P5726 and P0044]; 100 μg/ml DNAse I; 100 μg/ml RNAse; 1 mM PMSF) was added. The tissue was homogenized using an IKA ® ULTRA-TUR-RAX ® disperser tool (IKA T10 basic), until a homogenized solution was observed. The suspension was incubated for 10 min at RT with periodic mixing to guarantee efficient cell lysis and ribonuclease digestion.
Samples were combined with SDS loading dye (390 mM Tris-HCl pH 6.8, 485 mM DTT, 10% SDS, 0.1% Bromophenol-Blue, 50% glycerol) and incubated at 95 C for 5 min before clearing the samples by centrifugation at 13800 g. Equal sample volumes were run on a 12% gel and stained with Coomassie dye to account for loading differences ( Figure S2). Next, samples were loaded onto a combination of a stacking (4%) and resolving (6%) SDS polyacrylamide gel. The electrophoresis was run at 20 mA per gel until 1 h after the dye front ran out and the 250 kDa band of the protein ladder reached the bottom of the gel. Protein was transferred to a nitrocellulose membrane using a Mini  (Spazierer et al., 2003). 2.7 | Cell culture and CRISPR approach to generate a EPPK1 knockout MIO-M1 cells (Limb et al., 2002) were cultured in FBS containing medium (DMEM, high glucose, GlutaMAX™ Supplement, HEPES, 10% FBS, 1:100 Penicillin/Streptomycin; Gibco) at 37 C, 5% CO 2 unless stated otherwise.
We generated an EPPK1 knockout in MIO-M1 cells using pSpCas9(BB)-2A-Puro (PX459) V2.0 plasmid supplied by Addgene (plasmid # 6298) as suggested by the authors that deposited the plasmid (Ran et al., 2013). We first designed gRNAs using various tools (Benchling, (2021), CRISPRdirect (Naito et al., 2015)) targeted at a 1000 bp long region in the beginning of the translated exon and chose two guides each with minimal off-site reactivity (Table 3) For the transfection, $300,000 cells were seeded per well of a 6-well plate to receive a subconfluent culture on the next day. Cells were transfected with an equimolar mix of all four vectors using 200 μl of jetOPTIMUS buffer, which then was supplemented with 2.5 μl jetOPTIMUS reagent and incubated for 10 min. This transfection mix was added to the cells and incubated for 4 hours before exchanging the medium. Cells were allowed to recover for 48 h before moving to puromycin (3 μg/ml) containing selection medium.
After 4 days the remaining transfected cells were harvested and seeded in a 96-well plate at a density of 5 cells/ml (100 μl containing 0.5 cells per well) ensuring that most wells would contain either zero or one cell. Thus, colonies grown in such wells would be originating from a single cell and yield monoclonal cell lines. After $3 weeks, we saw colonies big enough to be harvested and further expanded two of them in bigger scale thereby generating the final monoclonal cell lines F7 and C9. Genomic DNA was isolated from these lines and used as input for a PCR with primers amplifying the 1000 bp target region in order to confirm the knockout on genomic level via sequencing (Table 4).

| Traction force microscopy
To perform traction force microscopy, we first produced acrylamide-  To calculate the traction stress and generate traction force maps we used several plugins for the ImageJ (Schindelin et al., 2012) software as well as custom written macros. Images of the beads were first enhanced in contrast, and then combined into a stack of which the background was subtracted using the rolling ball algorithm. The Linear Stack Alignment with SIFT (Lowe, 2004) plugin was then used to align both images of the stack to account for the x,y-drift when revisiting the cell positions. Next, each stack was processed with a plugin for particle image velocimetry (PIV) (Tseng, 2011;Tseng et al., 2012) using the template matching method with advanced settings. Traction forces and corresponding heat maps were calculated from the PIV vector matrices using a plugin for Fourier-transform traction cytometry (FTTC) (Tseng, 2011;Tseng et al., 2012). In some cases, the x,ydrift caused excessively high forces to be detected at the edges of a traction map, which is why we excluded all values in a 335-pixel wide frame. From the remaining values we calculated the average traction stress as the mean of all forces above a threshold set at 30% of the maximum value (peak traction stress) for each image analogous to the protocol developed by Bollmann et al. (2015). The custom ImageJ macros including the specific settings for the SIFT, PIV, and FTTC plugins are available upon request.

| Bioinformatic and statistical analyses
All statistical analyses were performed using the R programming language unless stated otherwise.
For differential protein expression analysis, we first excluded proteins that had missing values in more than two out of five Müller cell samples per group. Then, we chose only proteins that were significantly enriched in Müller cells in at least one group (RMG/neuron ratio >1, adjusted p-value <.05) yielding the Müller cell-specific proteins. The normalized abundance values were log transformed and used as an input to calculate differential protein expression with the limma package (Ritchie et al., 2015). PCA coordinates were calculated on log transformed normalized abundance values as input of the prcomp function and subsequently visualized via the factoextra (Kassambara & Mundt, 2020) package. Heat maps were produced on the basis of median centered, log transformed normalized abundances using the pheatmap (Kolde, 2019) package. Other plots were created with ggplot2 (Data S1).
Single-cell RNA sequencing data sets from Voigt (Voigt et al., 2019) and Cowan (Cowan et al., 2020) were downloaded as count matrices from Gene Expression Omnibus with the accession number GSE130636 or from https://data.iob.ch/, respectively. We We used the CytoScape (Shannon et al., 2003) software to create and explore gene/protein networks and perform pathway enrichment analysis via the StringApp plugin (Doncheva et al., 2019). Analysis scripts as well as CellProfiler pipelines are available upon request.

| RESULTS
3.1 | All-cone mice show generally healthy retinal layering with minor abnormalities in the outer retina Grimm and colleagues described the generation and characterization of the R91W/Nrl À/À (from here on just "all-cone") mouse retina among others, in terms of structure, visual acuity and photoreceptor degeneration over time (Samardzija et al., 2014). To gauge its use as a model for the human macula, we wanted to test for possible differences between the all-cone retinal cytoarchitecture and its respective control carrying only the R91W mutation (from here on regarded as "control" in comparisons with all-cone samples). For this, we stained for cell type-specific markers, quantified the nuclei in the nuclear, and between control and all-cone mice, corroborating their distinct photoreceptor identity (rod-vs. cone-dominated composition) (Figure 2b).  (Table S3). Notably, we found that only six proteins  Figure S3B). Apart from those, novel interesting human-specific candidate proteins could be identified. For example 11-beta-hydroxysteroid dehydrogenase 1 (HSD11B1), an enzyme involved in glucocorticoid metabolism was expressed at significantly higher levels in the macular Müller cell subpopulation (Figure 3f). In the context of glucocorticoid signaling in diabetic retinopathy (Ghaseminejad et al., 2020), it might be interesting to follow up in future studies. EPPK1, even though not reaching significance in this data set as in cone-rich mouse retina, was among the proteins expressed at higher levels in macular Müller cells (Figure 3f).
The list of these 217 differentially expressed proteins (DEP) was subjected to evaluation using the STRING database in combination with the CytoScape network analysis tool to identify functional con- (b-f) Label-free mass spectrometric analysis was performed on cell popluations purified from retinae of 5 control and 5 all-cone mice, respectively regional differences in the fovea and retinal periphery. We first identified genes that were enriched in Müller cells compared to neurons including cones, rods, bipolar, amacrine, and ganglion cells. The resulting genes were then subjected to differential expression analysis by comparing Müller cells of foveal/macular origin with their peripheral counterparts resulting in 477 and 742 differentially expressed F I G U R E 3 Legend on next page. genes for the Voigt and Cowan studies, respectively. 33 and 68 of them, respectively, were also detected in our proteomic analysis, of which, importantly, the absolute majority (30 and 67, respectively) showed the same regionally distinct expression pattern (Figure 4a).
Finally, a total of 29 genes with a congruent expression profile (11 up-, 18 downregulated in macula), were identified in all three data sets (Figure 4b).
Transcripts for EPPK1, our candidate with consistent expression profiles in human and mouse Müller cell proteomes, were identified only in Voigt et al. (2019). Because some transcripts were also detected in ganglion cells, they did not meet the above criteria for cross-validation of proteomes across scRNAseq data sets. However, a consistent trend toward higher EPPK1 transcript levels was observed in macular Müller cells compared with their peripheral counterparts ( Figure S4).

| EPPK1 is specifically enriched in coneassociated Müller cells
Our study revealed high expression of EPPK1 protein in both mouse and human Müller cells (Figures 2 and 3). EPPK1, with a mass of around 500-700 kDa is a huge protein belonging to a family of cytoskeletal linkers like desmoplakin or plectin Sonnenberg & Liem, 2007) and ). While the interregional difference of EPPK1 protein expression was close to, but did not reach statistical significance in our human data set (one-tailed paired t-test p-value: 0.09), immunostaining on cryosections of human retina delineated a clear regional difference in the EPPK1 staining pattern, confirming its localization in macular Müller cells (Figure 5a). In line with this finding, the EPPK1 staining was confined to Müller cell inner and outer stem processes in all-cone and control mice with a beads-on-a-string to fibrillar appearance ( Figure 5a). The higher staining intensity for EPPK1 in Müller cells from all-cone mice is consistent with our findings from mass spectrometric analysis (Figure 5a). Moreover, Western blot analysis done on retinal extracts from control and all-cone mice provided additional proof that EPPK1 levels are higher in cone-rich retina ( Figure 5b). Co-blotting of retinal samples from wild type and Eppk1 knockout (Eppk1 À/À ) mice demonstrated specificity of the antibody, as no specific signal was detected in EPPK1-deficient retina ( Figure 5b).
To assess whether EPPK1 is mandatory for retinal integrity, we studied the retinal architecture in Eppk1 À/À mice. Notably, previous work showed that a knockout of Eppk1 in mice did not lead to major phenotypic differences, even in tissues with high EPPK1 physiological expression levels, such as the skin (Goto et al., 2006;Spazierer et al., 2006), or to a disturbed keratin organization. A knockout in keratin 8, a close interaction partner of EPPK1, on the other hand, completely abolished the EPPK1 localization to the cellular periphery of wild-type hepatocytes demonstrating the dependence of a proper subcellular EPPK1 localization on keratin intermediate filaments . Similarly, no obvious disturbance in retinal or Müller cell morphology was found in Eppk1 À/À mice (Figure 5c Genotyping by PCR confirmed that these two cell lines, named C9 and F7, carried a deletion in their EPPK1 gene (Figure 6d). Both  (Cowan et al., 2020;Voigt et al., 2019) we identified genes that were specifically expressed in Müller cell clusters and showed significant difference between central and peripheral cells. Comparison of these differentially expressed genes with the proteomic data of the current study showed some degree of overlap with 33 genes from Voigt and 68 from Cowan being detected in both approaches. Genes/proteins that show a fold change (FC) of at least two on protein as well as on transcript level are colored, with blue indicating upregulation in macula and yellow upregulation in periphery. Most of the genes/proteins that were found to be differentially expressed in two methods also showed to be regulated in the same regional pattern. (b) Heat map showing a total of 29 genes/proteins differentially expressed in Müller cells (macula vs. peripheral region) that are shared among the three data sets confirmed by qPCR at transcript (Figure 6e) and by Western blot at protein level (Figure 6f).
We then used proteomic profiling to assess possible molecular perturbations of EPPK1 knockout Müller cells and/or their secretome.
We harvested the cell lysates and conditioned medium from mutant and wild type cells and performed tandem mass spectrometric analysis. Checking for the cell lysates first, we found a downregulation of cell adhesion and extracellular matrix proteins like collagens, aggrecan core protein, and fibronectin (Table S4) (Table S5). Accordingly, pathway enrichment analysis clearly points toward a reduced secretion of components of the extracellular matrix via extracellular vesicles (Figure 6g).
F I G U R E 5 EPPK1 expression and localization in human retina and various genetic mouse models. (a) Immunostaining of EPPK1 in human retina showed signal in macular Müller cells, primarily in their endfeet (arrows), while such staining was almost completely absent in the periphery. A comparable staining pattern was observed in mouse retina, where EPPK1 also has a specific localization in Müller cells extending from their basal endfeet (arrows) to their stem processes extending into the outer retina. Co-labeling of vimentin, a classical Müller cell marker, and EPPK1 in mouse retinal sections shows that EPPK1 is mainly localized in Müller cell processes. Scale bars, 50 μm. (b) The implemented anti-EPPK1 antibody shows protein specific binding in Western blot analysis with several bands in wild type due to protein degradation as frequently observed in protein lysates derived from mouse tissues. However, no immune-reactive bands were detected in samples from Eppk1 knockout (Eppk1 À/À ) retina. Increased EPPK1 protein expression was confirmed for all-cone versus control retina. Quantification was performed on the upper two bands (arrows) showing molecular weights expected for intact mouse EPPK1 protein variants (Ueo et al., 2021). (e) EPPK1 seemed to be upregulated in a retinal pathology model (transient ischemia) and can be seen to co-localize spot-like (arrowheads) with GFAP-positive Müller cell stem processes visualized by STED microscopy. Scale bar, 10 μm (f) high intraocular pressure led to reactive gliosis of Müller cells and thus to an upregulation of GFAP (right), while it is usually not detectable in healthy cells (left). GCL, ganglion cell layer; INL, inner nuclear layer; ONL, outer nuclear layer Next, we characterized potential effects of the EPPK1 knockout on cell morphology by staining the actin cytoskeleton using phalloidin ( Figure 7a). We noticed a reduction in cell size as well as an increased number of fine filopodia (Figure 7a,b). The outlines of the cells were segmented using CellProfiler software and various parameters such as cell area, solidity, elongation, form factor, and others were quantified.
F I G U R E 6 Legend on next page.
Most notably, we identified a significant and consistent decrease of the overall cell area in both knockout lines of around 35% (Figure 7b).
Another prominent parameter was solidity, which is the ratio between the area of the cell and its convex hull which is spread out by its protrusions. It can thus be used as an indirect measure of the cell's morphology, e.g., their complexity, with values closer to 1 indicating roundish, regularly shaped cells and values closer to 0 indicating cells with many complex processes (Janssen et al., 2022;Lobo et al., 2016).
Although we observed some variation in the shape of the cells of all genotypes, C9 and F7 exhibited a more regular morphology with fewer large projections. This was reflected in their significantly higher solidity ( Figure 7b).
Additionally, through an image-processing pipeline that included segmentation, erosion, masking, and skeletonization, we were able to robustly segment filopodia of individual cells (Figure 7a). This allowed us to count the number of filopodia per cell and calculate their average length. EPPK1 knockout resulted in an increase in the number of filopodia with a simultaneous decrease in filopodia length ( Figure 7c).
As cell morphology and filopodial dynamics are mainly regulated through the cytoskeleton, which also generates forces critical for cell function and retinal tissue integrity (MacDonald et al., 2015), we investigated the dependence of force generation by Müller cells on EPPK1 using traction force microscopy. For this, we grew cells on polyacrylamide substrates with a shear modulus of G' = 1000 Pa ($ 3000 Pa Young's modulus), which is in the range of reported moduli for retina (Ferrara et al., 2021), but which is much softer than tissue culture plastics (G' $ GPa) (Akhmanova et al., 2015) and found cells of various morphologies in both mutant cell lines and controls. While about half of the cells (49%-59%) displayed a spherical morphology and a smaller part (6%-27%) showed more arborized structure, we focused on bipolar cells (24%-34%) as shown in Figure 7d for our measurements. We analyzed the displacement of the embedded fluorescent beads after removal of the cells and subsequent relaxation of the gels to calculate the forces required for the initial deformation.
The highest contractile forces were generated at the cell poles.
EPPK1-deficient cells exerted significantly lower forces on their substrate than control cells, with a reduction of about 30% to 40% ( Figure 7d).
Given the morphological and biomechanical changes in the absence of EPPK1 in MIO-M1 cells and the implication of exocytotic pathways (Figure 6g), we wanted to investigate whether knockout of EPPK1 affects extracellular vesicle transport. There is evidence that intracellular exosome trafficking requires interaction with intermediate filaments in glial cells (Margiotta & Bucci, 2016;Potokar et al., 2007). Interestingly, tetraspanin CD9, a common vesicleassociated protein (Escola et al., 1998;Théry et al., 1999;Théry et al., 2018), was specifically expressed by Müller cells and enriched in the Müller cell subpopulation of the macular region ( Figure S3)-a pattern that was also consistent in two scRNAseq data sets (Figure 4b).
Immunostaining of MIO-M1 knockout cells for CD9 revealed that Finally, we wondered whether the difference in intracellular CD9 expression depends primarily on a decreased protein synthesis or on a general shift in exocytosis. While we saw no difference in CD9 protein amounts in the secretome of EPPK1 knockout and WT (Table S5), nanoparticle tracking analysis of conditioned medium showed a decrease in the secretion of small extracellular vesicles (<200 nm) from both knockout cell lines, with C9 cells releasing about three times less particles into the medium than the control (Figure 7g).

| DISCUSSION
The human macula is a peculiar structure essential for sharp vision, but unfortunately, also very susceptible to diseases like age-related macular degeneration (AMD), diabetic macular edema or macular telangiectasia type 2 (MacTel2). To date, many studies have focused on the role of different retinal cell types like microglia (Altmann & Schmidt, 2018) or the vasculature (Yeo et al., 2019)  Sequencing of the mutant PCR fragments revealed cuts by gRNA3, gRNA1, and gRNA3, gRNA4 leading to a deletion of 199 bp and 678 bp, respectively (c, bottom). Quantitative PCR ((e), mean of two technical replicates) and Western blot analysis (f) confirmed successful knockout of EPPK1 in the mutant cell lines. Wild type MIO-M1 cells showed three high molecular bands, maybe due to the triploidity of chromosome 8 (Limb et al., 2002), on which EPPK1 is encoded. a fourth band, visible in the wild type (asterisk), might be a degradation artifact. The HaCat wild-type cell line (keratinocyte origin) was used as a positive control showing two immune-reactive bands at the expected molecular weight (789 and 672 k Da, respectively). The respective EPPK1 knockout lines (C9, F7) lacked any immune-reactive bands. Size differences of EPPK1 may occur between samples originating from non-isogenic individuals (Ishikawa et al., 2018). (g) the plot shows GO terms (cellular components) for proteins found at different levels in cell lysate or the secretome of control and EPPK1-deficient MIO-M1 lines. ER, endoplasmatic reticulum Kefalov, 2009;Zhang et al., 2019), there is no deeper understanding of molecular protein pathways involved in regional Müller cell heterogeneity. In this study, we used multiple models in a multiomic approach to start closing this knowledge gap and to identify specific proteins that shape functional differences between central and peripheral Müller cells. Using the all-cone mice, we first studied differences between Müller glia in a cone-rich retina, compared to rod-rich normal controls in order to characterize this aspect of regional differences in photoreceptor subtype distribution in the human macular and peripheral retina, respectively. It has been shown that all-cone mice are capable of functional vision and have normal retinal layering (Samardzija et al., 2014). Nevertheless, we performed additional indepth characterization of the cellular composition as well as morphometric quantification of all retinal layers, in order to exclude possible F I G U R E 7 Legend on next page.
confounding factors from cone-independent retinal changes, like inner neuron composition. Complementary to the morphological analysis performed by Samardzija et al. (2014), we found minor alterations of the outer nuclear layer, with a lower number of nuclei in the all-cone retina. It is unclear whether this discrepancy develops during embryonal stages, maybe due to inefficient cone differentiation, or whether it is a sign of photoreceptor degeneration after birth as a result of the double mutation (Samardzija et al., 2014).
The increased thickness of the outer plexiform layer might be explained by cone pedicles, the synaptic structures relaying signals to the neurons of the inner nuclear layer, being bigger than the corresponding rod spherules (Hoon et al., 2014;Kolb, 1995;Wässle et al., 2002). Comparing our proteome data with two published single cell RNAseq datasets, we found several differentially expressed genes consistently detected in both transcriptomic and proteomic profiles.
Although there is increasing evidence that protein levels can be uncoupled from the synthesis of their transcript by a variety of mechanisms (Liu et al., 2016;Noya et al., 2019;Sharma et al., 2015;Vogel & Marcotte, 2012), we found that genes that overlap in the datasets analyzed here were mostly regulated in the same manner-29 genes showed a consistent regulatory pattern across all three data sets. These included CD9, a member of the tetraspanin family involved in the biogenesis of extracellular vesicles, as well as RDH10 and RLBP1. Unexpectedly, RDH10, RLBP1, RBP1 and DHRS3, all bona fide proteins of the visual cycle, were higher expressed in peripheral Müller cells. This seems counterintuitive, as studies report that cones, which are present in exceptionally high number in the macular retina, specifically rely on the function of these Müller cell-specific genes/ proteins for the regeneration of their photopigment (Kaylor et al., 2013;Wang & Kefalov, 2009;Xue et al., 2015). However, because there are no rods with long outer segments in the fovea to keep the RPE at a distance from the cones and also because the outer segments of the cones are exceptionally longer in the fovea (Domdei et al., 2021;Tschulakow et al., 2018;Yuodelis & Hendrickson, 1986) than in the peripheral retina, the RPE can potentially take care of the cones better in the fovea than in the periphery. Notably, DHRS3, an enzyme that opposes RDH10 in the synthesis of retinaldehyde (Adams et al., 2014;Belyaeva et al., 2019), was one of eight proteins consistently lower in cone-rich samples of both, human and mouse Müller cells. An ablation of Dhrs3 during development was shown to lead to an accumulation of all-trans retinoic acid (ATRA) with a concurrent reduction of the retinol pool resulting in embryonic lethality (Billings et al., 2013 (Belyaeva et al., 2019) and tends to be higher expressed in foveal Müller cells ( Figure S3).
Which isomerization and/or oxidation steps of the cone visual cycle actually occur in Müller cells and which proteins catalyze these reactions have not yet been completely elucidated, but the data provided in this study can serve as a starting point.
As pointed out before, hundreds of Müller cell-specific proteins were found to be differentially regulated. EPPK1 showed one of the highest expression differences in the all-cone mouse model. Immunofluorescence staining for EPPK1 confirmed its Müller cell-specific expression pattern as well as a beads-on-a-string-like co-localization with the glial intermediate filaments vimentin and GFAP. Previously, EPPK1 was shown to not only co-localize with a variety of keratins and vimentin, but also to be able to physically bind to them in in vitro studies (Jang et al., 2005;Spazierer et al., 2008;Szabo et al., 2015;Wang et al., 2006). there seemed to be no effect of Eppk1 knockout on keratin 8 . But what might this mean for primarily cone-associated Müller cells in the retina where EPPK1 showed an increased expression?
Although EPPK1 transcripts seem to be rare and are detected only at very low levels in Müller cells and to some extent also in ganglion cells in the scRNAseq datasets of human donor retinas (Voigt et al., 2019), we confirmed by immunostaining that the protein is specifically expressed in human Müller cells, whereas a low level of expression in ganglion cell axons cannot be completely excluded.
Additionally, qPCR, mass spectrometry, and Western blot analysis Moreover, it has even been speculated that Müller cells help maintain or even form the foveal pit . Here, we found evidence that EPPK1 might be involved in these functions, as its knockout in MIO-M1 cells led to (i) smaller cells with fewer protrusions and (ii) weaker traction forces. This, in turn, indicates that increased EPPK1 expression in macular Müller cells may contribute to the formation of their larger, more complex morphology as well as to the ability to exert the increased mechanical forces required to shape the fovea. In line with the discussed theory, we also found an increase in extracellular matrix proteins and adhesion molecules in macular  (Margiotta & Bucci, 2016;Potokar et al., 2007Potokar et al., , 2020 and EPPK1 was shown to bind directly to several intermediate filaments, including vimentin (Jang et al., 2005;Wang et al., 2006). Therefore, EPPK1 might affect vesicle transport either directly, by promoting the binding of intermediate filaments and vesicles, or indirectly, by modifying the rigidity/density of the cytoskeletal meshwork. In contrast to the cell lysate, we did not find a difference in CD9 abundance in the secretome of EPPK1-ablated cells, but an overall decrease in the number of released small extracellular vesicles (EVs). We thus hypothesize that most vesicles shed from the Müller cell line carry CD9 making it an interesting candidate to follow up on mechanisms of Müller cell secretion, while CD9-dependent EV biogenesis seems EPPK1-independent.
In contrast, intracellular trafficking of structures relevant for EV shedding were disrupted, as we observed less EVs in the supernatant.
Together with the finding that proteins of the ECM (collagens, fibronectin, and chondroitin sulfate proteins) were highly abundant in the secretome of wild type cells, but significantly downregulated in both EPPK1-deficient MIO-M1 lines, this could imply that EPPK1 does not only confer mechanical stability to those huge z-shaped macular Müller cells, but also coordinates transport and release of proteins across long distances in those Müller cells. The pronounced deposi-

| CONCLUSION
Previous studies showed that the central human retina not only presents a microenvironment with specific characteristics that produces challenges to the inhabitant cells different to the periphery, but that it is also highly susceptible to debilitating diseases. Here, we focused on Müller cells and were able to uncover differentially regulated pathways mainly, but not exclusively, linked to secretory and cell adhesion systems. Furthermore, we identified EPPK1 and CD9 to be enriched in macular Müller cells, implicating a role in the cells' biophysical properties as well as intracellular vesicle trafficking and their release. Nevertheless, future studies need to clarify the exact mode of interaction between EPPK1 and CD9 and their potential role in macular pathological processes.