Comparative lipidomic analysis of mammalian retinal ganglion cells and Müller glia in situ and in vitro using High-Resolution Imaging Mass Spectrometry

In order to better understand retinal physiology, alterations to which underlie some ocular diseases, we set out to establish the lipid signature of two fundamental cell types in the retina, Müller Glia and Retinal Ganglion Cells (RGCs). Moreover, we compared the lipid signature of these cells in sections (in situ), as well as after culturing the cells and isolating their cell membranes (in vitro). The lipidome of Müller glia and RGCs was analyzed in porcine retinal sections using Matrix Assisted Laser Desorption Ionization Imaging Mass Spectrometry (MALDI-IMS). Isolated membranes, as well as whole cells from primary cell cultures of RGCs and Müller glia, were printed onto glass slides using a non-contact microarrayer (Nano Plotter), and a LTQ-Orbitrap XL analyzer was used to scan the samples in negative ion mode, thereafter identifying the RGCs and Müller cells immunohistochemically. The spectra acquired were aligned and normalized against the total ion current, and a statistical analysis was carried out to select the lipids specific to each cell type in the retinal sections and microarrays. The peaks of interest were identified by MS/MS analysis. A cluster analysis of the MS spectra obtained from the retinal sections identified regions containing RGCs and Müller glia, as confirmed by immunohistochemistry in the same sections. The relative density of certain lipids differed significantly (p-value ≤ 0.05) between the areas containing Müller glia and RGCs. Likewise, different densities of lipids were evident between the RGC and Müller glia cultures in vitro. Finally, a comparative analysis of the lipid profiles in the retinal sections and microarrays identified six peaks that corresponded to a collection of 10 lipids characteristic of retinal cells. These lipids were identified by MS/MS. The analyses performed on the RGC layer of the retina, on RGCs in culture and using cell membrane microarrays of RGCs indicate that the lipid composition of the retina detected in sections is preserved in primary cell cultures. Specific lipid species were found in RGCs and Müller glia, allowing both cell types to be identified by a lipid fingerprint. Further studies into these specific lipids and of their behavior in pathological conditions may well help identify novel therapeutic targets for ocular diseases.


Results
In order to identify a lipid signature for RGCs and Müller cells, a strategy was developed that integrates in situ and in vitro experiments within a lipidomics workflow (Fig. 1). Prior to performing the lipidomic analysis, the purity and quality of the cultures established was verified by immunocytochemistry using antibodies that specifically recognized RGCs and Müller glia. After 7 days in vitro (div), RGCs had grown and on the whole, regenerated their neurites ( Fig. 2A), while Müller cell cultures were confluent (Fig. 2B). Moreover, the Müller cells maintained their viability and normal function, expressing specific markers like p75NTR, CRALBP and Glutamine synthetase ( Supplementary Fig. 1).
The data from the MS spectral analysis of the microarrays established from the Müller cell and RGC cultures were compared, highlighting significant differences in the relative lipid densities between these two cell types. Two different microarrays were prepared from the cultures for the MS spectral analysis, one from whole cells and the other from membrane preparations. The MS spectral analysis highlighted the differential negative ions (m/z) and the normalized relative intensities in whole cells or membrane preparations of the cultured RGCs and Müller cells (Table 1). Similar results were obtained for the whole cell and membrane preparations for each cell type, highlighting the consistency of the data. www.nature.com/scientificreports/ Following the MALDI-IMS analysis of retinal sections, a cluster analysis of the MS spectra identified regions that corresponded to areas of RGCs and Müller glia, as confirmed by immunohistochemistry of these previously scanned sections (Fig. 3). The cluster analysis successfully identified the area that corresponded to RGCs (the ganglion cell layer, GCL) and the rest of the layers in the retina. Müller cells span the entire thickness of the retina, sharing their space with other cells and with RGCs. Therefore, the GCL area that corresponded to the RGC population (red) was compared with the area where Müller cells overlap with other cell bodies, the area that mostly corresponds to the inner plexiform layer (IPL, green). The raw spectra obtained from both these clusters or areas can be seen in the supplementary information (Supplementary Data 1). There was a significant difference in the relative density of certain lipids identified in the MS/MS analysis between areas containing RGCs and those of Müller glia (p-value ≤ 0.05, Table 2).
To compare the results obtained from the Müller cells and RGCs in situ and in vitro, a correlation was made between the intensity of selected peaks obtained from retinal sections and their intensity in microarrays of membrane preparations. These data confirm that the results obtained in the sections were significantly correlated with those from the in vitro experiments (p < 0.0001, Fig. 4). A linear regression between the spectra from the RGC (Supplementary Fig. 2A) and Müller cells ( Supplementary Fig. 2B) in the sections, and the in vitro spectra of RGCs and Müller cells always showed a stronger correlation between the spectra of the same cell type, irrespective of its origin (i.e.: sections or in vitro cultures). It was notable that the membrane correlation coefficient was stronger than that of the whole cells for both RGCs and Müller cells, such that the peaks obtained from the membrane preparations were more similar to the peaks obtained from the retinal sections ( Supplementary  Fig. 3). The distribution from the membranes in the microarrays was far more homogenous than that from the whole cells. This explained the weaker sensitivity and weaker signal/noise ratio in cells, due to the presence of fewer lipids per unit area.
Finally, a comparison of the lipid profiles obtained from the retinal sections and microarrays suggested a collection of six peaks ( Table 3) that were characteristic of RGCs or Müller glia in both retinal sections and cultures. These lipids were identified by MS/MS, detecting characteristic fragmentation spectra for all the peaks (an example of the MALDI MS/MS assignment procedure is presented in Supplementary Fig. 4).

Discussion
Lipid biochemistry in the vertebrate retina is remarkable from many points of view, not least due to the specific accumulation of DHA (22:6) that is then esterified to different phospholipids and molecular species within the retina. The lipids of retinal membranes have been studied extensively and much information has been obtained 7 , however, this is the first time that in situ and in vitro techniques have been combined to evaluate the specific lipids accumulated by Müller glia and RGCs. There is evidence that certain lipids are distributed specifically in different retinal layers 33,36 . Given the importance of each of the different cell types in the retina, defining the lipids that might be specific to these individual cell types may provide unique information regarding the homeostatic state of this organ and the development of diseased states. This may be particularly true of the RGCs and Müller glia studied here, given their fundamental roles in preserving the homeostasis of the retina and in the transmission of visual information to the brain. After a successful correlation analysis whereby the results of in vitro experiments could be readily extrapolated to in situ results, a group of lipids were identified in which four peaks were more strongly represented in RGCs and two were more prevalent in Müller cells, both in sections and in microarrays. The principal families of lipids in the rat retina are phosphatidylcholine (PC, ca. 40-50%), phosphatidylethanolamine (PE, ca. 30-35%), phosphatidylserine (PS, ca. 5-10%) and phosphatidylinositol (PI, ca. 3-6%), accounting for 85-90% of the total retinal phospholipids 42 . Therefore, it is not surprising that the peaks selected from the retinal samples were PCs, PEs and PIs. PCs, are usually the most abundant glycerophospholipids (GPLs) and the key building blocks of membrane bilayers. They have a small head group that can form hydrogen bonds through their ionizable amine group, which acts as a "chaperone" during membrane protein assembly and guides the folding of associated proteins. Four PCs were previously characterized in three distinct layers of a mouse retina section using MALDI-IMS, although none of them were specific to only one retinal layer: PEs are the second most abundant phospholipids in mammalian cells 43 , enriched in the inner side of membranes and particularly in the inner mitochondrial membrane. PEs are also enriched in arachidonic acid 44 and they have quite notable activities: as a chaperone aiding membrane protein folding; in respiratory complexes; and in the initiation of autophagy 45 . Our results confirm that certain PEs are more abundant in RGCs than in Müller cells (PE P-16:0/20:4, PE 18:1/20:4 and PE 16:0/22:5). PE P-16:0/20:4 has been found in the mouse, rat and human brain 46 and significantly, GPL signatures of retinal cells are similar to those of brain tissue 47 . Here, Table 1. Summary of the differential negative ions (m/z), and the normalized relative intensity obtained after MS spectral analysis of the microarrays prepared from whole cell or membrane preparations of RGC and Müller cell cultures.  48 . According to our data and as found previously, PE 18:0/22:6 is more abundant in Müller glia and it is specifically located in the outer retinal layers where Müller cells may be found 33,49 . PIs participate in essential metabolic processes, and they are primary sources of arachidonic acid and diacylglycerol. These second messengers serve as signaling molecules that regulate the activity of a group of related enzymes, known as protein kinase C, which regulates many crucial cell functions like proliferation, differentiation, metabolism and apoptosis. We detected two peaks (m/z 885.55 and 909.55) that correspond to three PIs more abundant in RGCs than in Müller cells, both in sections and microarrays. It is known that PIs are also main regulators of many ion channels and transporters, which are involved in neuronal excitability and synaptic transmission 50 . Thus, the more common representation of these lipids in RGCs than in Müller cells could be related to their neuronal activity. The basal peak at m/z 885.5 corresponded to PI 18:0/20:4, found in the nerve fiber/GC layer (by MALDI-IMS) and in the inner nuclear layer (INL) of the mouse and human retina 49 , and spreading into the outer plexiform layer (OPL) 36 as well as the optic nerve, retina and sclera 33 . The m/z 909.5504 peak was identified as PI 18:0/22:6 and PI 20:2/20:4, PIs that are more commonly found in RGCs than Müller cells. However, in literature these lipids are not as common as PI 18:0/20:4 and to date, PI 18:0/22:6 has been found only in the cod retina 51 .
In summary, negative ion-mode imaging can be used to define the spatial distribution of a number of lipid species, including PEs, PCs and PIs, enabling us to carry out the first comparative study between in situ and in vitro assays. Combining different techniques that provided sufficiently high spatial resolution, distinguishing specific retinal cell layers, enabled the distributions of specific lipid to be defined. The fact that some lipids from the most relevant lipid families are more characteristic of RGCs or Müller cells suggests that they could fulfill roles in different cell activities. Interestingly, this technology could be used to compare healthy retinal tissue with pathological tissue in order to identify disease-related lipidomic changes in specific regions, such as advanced glycation and lipoxidation end products (AGEs and ALEs). Thus, further studies will provide more information on the implications of lipids in retinal diseases, identifying new therapeutic targets to slow or prevent disease progression.

Methods
Animals. Adult porcine eyes were obtained from a local abattoir and transported to the laboratory in cold CO 2 -independent Dulbecco's modified Eagle's medium (DMEM-CO 2 : Gibco-Life Technologies). The time between sacrifice and processing the eyes was 1 h. This study was carried out in strict accordance with the Guidelines for the Care and Use of Laboratory Animals from National Research Council (US). Moreover, all

Immunocytochemistry of cultures.
In order to check the quality of the cultures, at least 3 wells of each culture were fixed in cold methanol and washed with PBS. After blocking non-specific antigens with blocking buffer (3% BSA and 0.1% Triton X-100 in PBS), the cells were incubated with the following primary antibodies at a dilution of 1:2000: a mouse anti-βIII-tubulin antibody as a specific RGC marker (RRID: AB_430874, Promega Madison, WI, USA); and as specific markers of Müller glia, a rabbit anti-Vimentin antiserum (RRID: AB_1524552, Abcam, Cambridge, England), a rabbit anti-p75NTR antiserum (RRID: AB_2728801, Abcam, Cambridge, England), a rabbit anti-CRALBP antiserum (RRID: AB_1658655, Abcam, Cambridge, England) and a rabbit anti-Glutamine synthetase antiserum (RRID: AB_302521, Abcam, Cambridge, England). After washing the cells, antibody binding was detected with an anti-mouse Alexa Fluor 488 or an anti-rabbit Alexa Fluor 555 (Life Technologies, Carlsbad, AC, USA) secondary antibody, diluted 1:1000. In addition, the cell nuclei were labeled with DAPI (Life Technologies, Carlsbad, AC, USA) at a dilution of 1:10,000.
Cell membrane isolation. The cells isolated from pig Müller glia or RGC cultures were homogenized using a Teflon-glass grinder (Heidolph RZR 2020) and a disperser (Ultra-Turrax ® T10 basic, IKA) in 50 mM Tris buffer (TB) supplemented with 1 mM EGTA, 3 mM MgCl 2 and 250 mM sucrose. The crude homogenates were centrifuged at 40 × g for 5 min and the resulting supernatants were centrifuged again at 18,000 × g for 15 min at 4 °C (Microfuge ® 22R centrifuge, Beckman Coulter). The pellets were washed in 20 volumes of 50 mM TB and re-centrifuged under the same conditions. The protein concentration of each sample was measured using the Bradford method and aliquots of the homogenate were stored at − 80 °C until use.  printed onto glass slides using a non-contact microarrayer (Nano_plotter NP 2.1) and a printing solution 55 . The piezoelectric tips dispense 4 nl per spot, printing each sample in triplicate. These microarrays were stored at − 20 °C until use.
Tissue collection for retinal sections. Adult eyes were enucleated, removing the cornea, crystalline lens and vitreous humor. The retina was then carefully separated from the rest of the eye, cut into small rectangular pieces and folded back over itself like a closed book or "sandwich" to maintain the RGCs internally when the tissue was then frozen at -20 °C. The use of this "sandwich" aimed to protect the layer of ganglion cells at the edge of the section and to avoid interactions of the OCT with the matrix used. In this way, we could also visualize a duplicate image of the retina in each section (Fig. 3). The retinal "sandwich" was then partially embedded in OCT medium to obtain cryosections of the retinal aspect and not of the OCT embedded tissue (14 μm thick). These sections were then stored at − 80 °C in an atmosphere of N 2 to protect them from oxygen and moisture degradation.
Sample preparation for IMS. Retinal sections and microarrays were thawed, and DAN (1,5-Diaminonaphthalene) 57 was deposited on them using a glass sublimator (Ace Glass 8023) 58 . The matrix formed a uniform thin layer that enables retinal sections and microarrays to be scanned in negative ion mode for several hours. Only negative-ion mode was employed in this study as a wider variety of lipid species can be detected in this way, whereas PC species are those predominantly detected in positive-ion mode. For example, plasmalogens are usually obscured by PCs in positive-ion mode. Likewise, sulfatides are detected in negative-ion mode, and any interference between PC and PE is avoided. This analysis was performed as described previously 59 , whereby a LTQ-Orbitrap XL analyzer (ThermoFisher, San Jose, CA, USA) equipped with an N2 laser (100 µJ max power, elliptical spot, 60 Hz repetition rate) scanned the material in negative ion mode at a spatial resolution between 15 and 25 µm. A mass resolution of 60,000 and 100,000 was used to record the data from the full scan spectra, and that of around 2000 for MS/MS and MS 3 due to the higher sensitivity of the Ion Trap (IT). The scanning range in full scan spectra was 550-1200.
Immunohistochemistry of retinal sections. After the sections were scanned they were washed with methanol at room temperature (RT) for 5 min to remove the DAN and they were then fixed with 4% PFA (paraformaldehyde) at RT for 2 min. The sections were immunostained as described previously 38  Data and statistical analysis. The spectra acquired were aligned to maximize the correlation with the overall averaged spectrum and normalized using dedicated software (MSI Analyst, Noray Bioinformatics S.L.). During parsing, the size of the data was reduced to eliminate all the peaks whose intensity was lower than 0.5% of the strongest peak on the spectrum and the spectra were normalized using a total ion current algorithm 60 . The spectra were also aligned using the Xiong method 61 assuming a maximum misalignment of 0.02 a.m.u., very conservative for an orbitrap analyzer. For graphical representation, no interpolation or smoothing algorithms and no de-noising procedures were used, always trying to maintain the original aspect of the data. Re-scaling was required for some images reflecting the distribution of the very weak peaks. Statistical analysis to identify the different areas in the sections was carried out using MATLAB (MathWorks, Natick, USA), employing divisive hierarchical clustering with Rank Compete 62 and k-means 63 . After determining the cluster of Müller cells and RGCs, a principal component analysis (PCA) 64 and ANOVA were used to identify the specific lipids for each cell type in the retinal sections and microarrays.
Peak assignment. The peaks of interest were identified directly by MALDI-MS/MS and MS 3 on the tissue samples in order to discriminate chemical variants of the lipids with identical numbers of acyclic carbons and double bonds. To perform fragmentation we used collision-induced dissociation (CID) and an ion trap to detect the fragments, a set-up that improves the sensitivity for the fragments with little cost of losing mass resolution. The mass window of the ion trap allows is approximately one Da wide and thus, we could identify fragments that belonged to our parent masses and those belonging to other species close to the parent masses (see Supplementary Fig. 4 and Supplementary Table 1). The assignment of lipid species was facilitated by a database created using tissue-dependent lipid compositions. The database was calculated in silico by combining fatty acids and head groups, and with the aid of the Lipid Maps database (https ://www.lipid maps.org/). The experimental values of the precursor molecules, and of the fragments after MS/MS and MS 3 analysis were compared with this database, and with the Lipid MAPS and Madison Metabolomics Consortium (https ://mmcd.nmrfa m.wisc.edu/) databases, using 0.005 Da as a window of tolerance for the full scan precursors in order to identify each mass channel.

Scientific Reports
| (2020) 10:20053 | https://doi.org/10.1038/s41598-020-77087-x www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.