Histology-guided spatial lipidomics and proteomics of the trisynaptic circuit in the human hippocampus

Abstract

The human hippocampal trisynaptic circuit activity is essential for learning and memory. This canonical circuit has spatially distinct populations of neurons, but their unique contributions to neurodevelopment, as well as to dysfunction in neurodegenerative disorders, are missed when analyzing bulk tissue homogenates. Using matrix-assisted laser desorption/ionization (MALDI) mass spectrometry imaging (MSI) to guide laser capture microdissection (LCMD) of regions of interest for spatial multimodal analyses is a relatively new approach to study topographically distinct neuronal cell populations in heterogenous tissues. However, MALDI-MSI may not identify region-defining molecular mass-to-charge ions. Here, we apply a multimodal approach of MALDI-MSI-LCMD-lipidomic and proteomic analysis to the trisynaptic circuit. Our MALDI-MSI of the hippocampus revealed that the of mass-to-charge ions of the cornu ammonis 1 (CA1) and cornu ammonis 3 (CA3) did not segment from the surrounding tissue. Thus, we developed a novel histology-guided MALDI-MS imaging-LCMD-spatial lipidomic/proteomic pipeline with four steps which does not rely on segmentation analysis to determine and co-register regions of interest in tissue sections. Our pipeline allows MALDI imaging, LCMD, lipidomic and proteomic analysis from the same tissue section and does not require co-registration across serial sections. In addition, poly-L-lysine coating for improving tissue/cell adherence on indium-tin-oxide microscopy slides did not impact MALDI-MSI or spatial proteomics. We show that the human trisynaptic circuit proteomes of CA1 and CA3 pyramidal neurons are more similar to each other than those of the dentate gyrus (DG), which is consistent with previously reported transcriptomics studies. The spatial distributions of several phospholipids and proteins, however, were significantly different in cell bodies from the CA1, CA3 and DG regions, and these lipids correlated with some lipid metabolizing enzymes in those regions. As little is known about lipid metabolism in the hippocampus, our pipeline provides an initial step in studying the combined and differential spatial regulation of the lipids and proteins within the trisynaptic circuit that will provide insights into the development and disease-related molecular changes in these important hippocampal regions.

GRAPHICAL ABSTRACT

1. Introduction

The hippocampal formation (Hi) in the human brain is central to learning, as well as memory consolidation and retrieval, both in the long and short term. Among deep brain nuclei, Hi is involved in virtually all aspects of cognition across the human lifespan [1,2]. Anatomically, the Hi extends for most of the antero-posterior axis of the temporal lobe. It contains several intra- and inter-neuronal connections between topographically distinct neuronal cell populations. The trisynaptic circuit is a canonical Hi circuit, which includes input from the entorhinal cortex layer II to the dentate gyrus (DG) to the cornu ammonis 3 (CA3) leading to output from the cornu ammonis 1 (CA1). This trisynaptic circuit is central to activity-dependent long-term potentiation, a critical cellular mechanism underlying learning and memory in the human brain [3]. The spatially heterogenous anatomy of the Hi makes it imperative to analyze specific, spatially distinct neuronal populations to decipher their unique contributions to neurodevelopment, aging, and psychiatric and neurodegenerative brain disorders, which are often missed in Hi studies analyzing bulk tissue homogenates.

Molecular signatures of specific Hi cell-types present in DG, CA3 and CA1 have been identified in single nuclear RNA-sequencing studies [4]. However, spatial transcriptomics approaches frequently lack corresponding information on the proteins expressed by the transcripts, which represent the gene end-products and, therefore, phenotypic properties of the Hi cells. Moreover, lipids are not addressed by either transcriptomics or proteomics and remain another highly diverse molecular species understudied in human Hi. Lipids make up more than 40 % of human brain weight and are critical for membrane integrity, cellular signaling, trafficking, and metabolism [5–8]. There is a clear need to link lipid composition and lipid levels to the corresponding metabolizing and regulatory enzymes, which may harbor critical changes in lipid metabolism that drive aging [9] and cognitive diseases in the brain [10].

Brain tissue has historically been a favorite organ for mass spectrometry imaging due to the abundance of macrostructures as well as its relevance to human health. While the vast majority of these studies are in mouse and rat brain, there are a significant number of studies in post-mortem human brain tissue, especially for brain diseases such as Alzheimer’s Disease, brain cancer, and epilepsy [11]. One landmark study using fresh frozen post-mortem brain tissue, demonstrated that both positively and negatively charged lipid abundances were altered in CA1 and DG as a result of Alzheimer’s Disease [12]. Eberlin et al. have also demonstrated the power of lipidomic mass spectrometry imaging to detect human brain cancers using mass spectrometry imaging [13]. In the field of epilepsy, mass spectrometry imaging has been used to detect changes in neuropeptides in DG in epileptic patients [14]. While a significant amount of work has been done, there is still an extensive amount of work necessary to understand normal brain function, which is why we have started to explore the lipidome and proteome of the trisynaptic circuit in normal human brain.

The combination of matrix-assisted laser desorption/ionization (MALDI) mass spectrometry imaging (MSI), laser capture microdissection (LCMD), and mass spectrometry (MS)-based proteomics is a relatively new pipeline which, prior to our study, has not been applied to the human brain [15,16]. This pipeline allows for spatially resolved analysis of the lipidome and proteome from the same tissue [15,16]. Combining LCMD with proteomics enables proteomic analysis at single cell resolution by isolating single cells from heterogenous tissues [17]. The addition of MALDI-MSI to this workflow was first demonstrated in 2021, in which metabolites were measured from formalin-fixed paraffin-embedded (FFPE) mouse heart tissue prior to LCMD and proteomics [15]. More recently, this method has been used to examine human prostate cancer tissue lipidomics and proteomics in a spatially resolved manner from fresh-frozen tissue sections [16]. Both these studies utilized the MSI data set to guide the LCMD based dissection, and in many cases the selected molecular class for determining where to dissect reflects gross pathology and anatomy. However, scenarios where MSI does not identify a region-defining molecular class, LCMD must rely on histological differences. In this paper, we describe a workflow for combined MALDI-MSI, LCMD, and mass spectrometry-based proteomics using a histology-guided method to characterize lipids and proteins from specific cell populations within the Hi trisynaptic circuit. We report a novel method for co-registering LCMD extracted cells in tissue sections with MSI data from the same section. We show that spatial distributions of several phospholipids and proteins were significantly different in cell bodies from the CA1, CA3 and DG regions, and the correlation between specific lipid abundances and lipid metabolizing enzymes in those regions.

2. Materials and methods

2.1. Reagents

Reagents were purchased from Sigma Aldrich (St. Louis, MO, USA) at HPLC grade or higher and used without further purification unless otherwise noted.

2.2. Human postmortem brain tissue acquisition and dissection

Postmortem human brain tissue was obtained as previously described [12]. All brain samples in this study were collected at the Lieber Institute for Brain Development (LIBD) via audiotaped witnessed informed consent with legal next-of-kin at the time of autopsy, at the Office of the Chief Medical Examiner of the state of Maryland under Maryland Department of Health internal review board (IRB) protocol #12–24. Audiotaped informed consent to study brain tissue was obtained from the legal next of kin on every case collected at LIBD. Details of the donation process and specimen handling have been described previously [12]. After next of kin provided audiotaped informed consent to brain donation, a standardized 36-item telephone screening interview was conducted (the LIBD autopsy questionnaire) to gather additional demographic, clinical, psychiatric, substance abuse, treatment, medical and social history. A psychiatric narrative summary was written for every donor to include data from multiple sources, including the autopsy questionnaire, medical examiner documents (investigative reports, autopsy reports and toxicology testing), macroscopic and microscopic neuropathological examinations of the brain and extensive psychiatric, detoxification and medical record reviews and/or supplemental family informant interviews using the mini-international neuropsychiatric interview. Two board-certified psychiatrists independently reviewed every case to arrive at Diagnostic and Statistical Manual of Mental Disorders (DSM)-5 lifetime psychiatric and substance use disorder diagnoses, including schizophrenia and bipolar disorder, as well as substance use disorders, and if for any reason agreement was not reached between the two reviewers, a third board-certified psychiatrist was consulted. All donors in this study were free from psychiatric diagnosis, significant neuropathology, including cerebrovascular accidents and neurodegenerative diseases. Available postmortem samples were selected based on RNA quality (RIN ≥6.5) to assess tissue quality. A toxicological analysis was performed in each case. The non-psychiatric non-neurological neurotypical individuals have no known history of significant psychiatric or neurological illnesses, including substance abuse. Positive toxicology was exclusionary for neurotypical individuals.

2.3. Human postmortem brain processing, dissections, and donor subject details

Postmortem fresh human brain dissections and freezing were performed as described previously [12]. Anatomically, a single frozen block containing CA1, CA2, CA3, and dentate gyrus (DG) granule cell layer of the human hippocampal formation at the level of the lateral geniculate nucleus was dissected out from the neurotypical control postmortem brain subject. The block was dissected from previously fresh frozen coronal human brain slices using a hand-held dental drill (Cat# UP500-UG33, Brasseler, Savannah, GA, USA) as described before [12]. Specific gross anatomical landmarks above were matched to corresponding coronal brain sections from the Allen Human Brain Reference Atlas (https://atlas.brain-map.org) for the subject to ensure regional and anatomical accuracy.

2.4. Slide preparation and cryo-sectioning

Indium tin oxide (ITO) slides (Delta Technologies, Loveland, CO, USA) were washed by sonicating for 10 min in hexane and then 10 min in ethanol. Slides were allowed to dry overnight at room temperature before coating. Noted slides were coated with poly-l-lysine using previously described methods [18]. Three major projection neurons of the hippocampal trisynaptic circuit – dentate gyrus granular cell layer (DG-GCL), as well as pyramidal neurons of CA3 and CA1 – were isolated from neighboring cellular layers of the hippocampal trisynaptic allocortex. Whole middle hippocampal formation blocks at the level of the dorsomedial thalamus were dissected from coronally sectioned, fresh frozen brain slabs using a dental drill as described previously [19]. Blocks were then sectioned into 25-μm thick sections using a Leica 350S cryostat (Leica, Nussloch, Germany) at −20 °C onto ITO slides. Readily cut tissue samples on slides were vacuum-sealed and stored at −80 °C until imaging. Slides were cut in replicates of five and three technical replicates were run.

2.5. MALDI mass spectrometry imaging

Slides were warmed to room temperature in a vacuum desiccator prior to matrix application. Slides were sprayed with an HTX M5 sprayer using α-cyano-4-hydroxycinnamic acid matrix (CHCA) matrix (5 mg/mL) in 50 % acetonitrile and water with 0.2 % trifluoroacetic acid (TFA) using the following spray parameters: 30 °C nozzle temperature, 4 passes, 0,1 mL/min flow rate, 750 mm/min velocity, 2 mm track spacing, criss cross pattern, 10 psi pressure, 3 L/min gas flow rate, and no drying time. MALDI-MSI was performed on a Bruker Rapiflex TOF/TOF instrument (Bruker, Billerica, MA, USA) in reflectron positive mode. A 50-μm M5 M5 laser ablation pattern and 50-μm raster was used with 200 laser shots per pixel.

2.6. Tandem mass spectrometry experiments

Tandem MS experiments were performed on the same instrument in profiling mode with argon as a collision gas for collision induced dissociation (CID). A single beam laser setting with a spot size of 50 μm was used for MS/MS analysis. An isolation window of ±1 Da was used for all experiments. Data was peak picked in FlexAnalysis 4.0 (Bruker Daltonics, Billerica, MA) using the centroid method. MS1 peaks were searched in Lipid Maps LMSD database. [M+H]⁺ and [M+Na]⁺ were selected for the search with a mass tolerance of ±0.2 Da. Fatty acyls, glycerolipids, glycerophospholipids, and sphingolipids were selected as potential classes of lipids. Structures were downloaded from the Lipid Maps search and fragmentation was solved using ChemDraw 20.0 (PerkinElmer Informatics, Waltham, MA). Once MALDI imaging was completed, slides were washed with 100 % ethanol for 24 h to remove the CHCA matrix. The same slides were then used for LCMD and proteomics.

2.7. Laser capture microdissection

To facilitate better morphological distinction from other cell types within the hippocampal formation, the same sections which had already undergone MALDI imaging and matrix removal, were stained with the Arcturus Histogene LCMD frozen section staining kit reagents as outlined below (Arcturus Catalog number KIT0401, Thermo Fisher, Waltham, MA, USA). Briefly, the sections were immersed in 70 % ethanol and deionized water for 30 s each, stained for 20 s with Histogene staining solution, followed by ethanol dehydration with 70 %, 95 %, and absolute ethanol for 30 s each. Sections were then air dried for 2 min in a fume hood before laser microdissection. Specific neuronal populations were selected in reference to the Allen Brain Atlas for human brain [20], cut using the Zeiss PALM UV laser at 335 nm (Zeiss, Jena, Germany), and catapulted onto adhesive caps of 0.6 mL opaque Zeiss tubes (Zeiss Catalog number 415190–9201-000) with high precision enrichment from surrounding cells in the hippocampal formation (Zeiss PALM v4.9) (Fig. S1). A total of 500 neuronal cells were collected per sample, in triplicate, for each of the neuronal cell-types. Adhesive caps were immediately loaded with 7 μL of lysis buffer (Qiazol lysis reagent no. 79306, Hilden, Germany) and frozen at −80 °C on dry ice for proteomics.

An example showing the dentate gyrus granule cell layer before and after LCMD is shown in Fig. S1 in the Supporting Information. The presented image has been taken on the Zeiss Palm system with the 10X objective. For the sake of clarity, the image has been zoomed in to the slide containing the entire human hippocampal formation.

2.8. Proteomics

Proteolytic Digestion and Labeling:

Protein samples in Qiazol lysis buffer were reduced with 50 mM Dithiothreitol in 10 mM triethylammonium bicarbonate (TEAB) at room temperature for 1 h followed by alkylating with 100 mM chloroacetamide in 10 mM TEAB at room temperature in the dark for 15 min. Qiazol and other non-protein contaminants were removed using a mixture of Sera-Mag SpeedBeads (GE Healthcare, cat. no. 45152105050250 and 65152105050250, Chicago, Il, USA) in a single pot, solid phase sample preparation protocol (SP3) [21]. Protein bound beads were resuspended in 50 μL of 100 mM TEAB and digested with 0.33 μg of trypsin/LysC (Pierce, cat. No. A41007, Thermo Fisher, Waltham, MA, USA) overnight at 37 °C while shaking. The digested peptides were labeled with a unique tandem mass tag (TMT) TMTpro 16-plex reagent (Thermo Fisher, LOT # WK338750, Waltham, MA, USA) using 50 μg of label in 10 μL anhydrous acetonitrile directly to the peptide/bead mixture. Labeling took place while shaking at room temperature for 1 h and quenched with 2 μL 5 % hydroxylamine for 15 min. All TMT labeled peptide samples were removed from the beads by magnetic separation, combined and dried by vacuum centrifugation.

Peptide Fractionation:

The combined TMT-labeled peptides (1.5 μg estimate) were re-constituted in 100 μL 10 mM TEAB buffer, bound to resin on Oasis HLB uElution plates (Waters, Milford, MA, USA) equilibrated in 10 mM TEAB buffer and step-fractionated by basic reverse phase (bRP) chromatography with 15 %, 25 %, 45 % and 75 % acetonitrile in 10 mM TEAB. Peptide fractions were dried by vacuum centrifugation.

Mass Spectrometry Analysis:

TMT-labeled peptides in each of the 4 fractions (50 %) were analyzed by nanoflow liquid chromatography tandem mass spectrometry on an Orbitrap-Fusion Lumos-IC (Thermo Fisher Scientific, Waltham, MA, USA) interfaced with an Vanquish NEO UHPLC. Peptides were separated by reversed-phase chromatography using a 2 %–90 % acetonitrile in 0.1 % formic acid gradient over 125 min at 300 nL/min on a 75 μm × 25 cm in-house packed column with 3 μm, 120 Å ReproSIL-Pur-120-C18-AQ resin (Dr. Maisch, ESI Source Solutions). Eluting peptides were sprayed into the mass spectrometer at 2.5 kV. Survey scans of precursor ions were acquired from 350 to 1800 m/z at 120,000 resolution at 200 m/z with automatic gain control (AGC) at 4e5 and a 60 ms maximum injection time (IT). Precursor ions were individually isolated within 0.7 m/z by data dependent monitoring and 10 s dynamic exclusion, and fragmented using a higher-energy collisional dissociation (HCD) activation energy of 36. Product ion spectra of TMT labeled peptides were acquired using a 5e4 AGC and 86 ms maximum IT at 50,000 resolution.

Data Analysis:

Product ion spectra were processed by Proteome Discoverer v2.5 (PD2.5, ThermoFisher Scientific, Waltham, MA, USA) and searched with Mascot v.2.8.2 (Matrix Science, London, UK) against RefSeq2021_204_H_sapiens database. Search criteria included trypsin as the enzyme with one allowed missed cleavage, 3 ppm mass tolerance for precursor and 0.01 Da for fragments, TMTpro on N-terminus and carbamidomethylation on C as fixed, and TMTpro on K, oxidation on M, phosphorylation of STY, and deamidation on N or Q as variable modifications. Peptide identifications from the Mascot searches were processed within PD2.5 using Percolator at a 1 % False Discovery Rate confidence threshold, based on an auto-concatenated decoy database search. Peptide spectral matches (PSMs) were filtered for Isolation Interference <30 %. Relative protein abundances of identified proteins were determined in PD2.5 from the normalized median ratio of TMT reporter ions from all PSMs excluding phosphorylated peptides from the normalization. Technical variation in ratios from our mass spectrometry analysis was less than 10 % as previously reported [22].

2.9. Co-registration

The slides with tissue sections were scanned on an Epson Perfection V850 Pro desktop scanner at 4800 dots per inch (dpi) as.tif images prior to MALDI imaging, as well as following LCMD, which displayed the holes in the tissue from LCMD cutting. Images were co-registered in FlexImaging (v. 6.0) using three registration points. Regions of interest (ROIs) were drawn in FlexImaging based on the LDMD dissection holes in the tissue for DG, CA1, and CA3 and imported into SCiLS lab for analysis.

2.10. MALDI-MSI data analysis

Data was imported into SCiLS Lab (v. 2023b Pro, Bruker Daltonics, Billerica, MA, USA) using the default settings. Data was normalized to total ion count (TIC) and segmented using the bisecting k-means clustering method with a correlation distance metric. Receiver operating characteristic (ROC) curve analysis was performed using all the spectra in each ROI, comparing CA1 versus DG, CA3 versus DG, and CA1 versus CA3. Significance was determined using an area under the curve (AUC) of >0.75 or <0.25 for ROC analysis.

2.11. Comparative Lipidomic–Proteomic data analysis based on common lipid metabolic pathways

For analyzing the lipidomic and proteomic data comparatively, we first obtained significantly different m/z species by MALDI imaging ROC analyses between CA1 versus DG, CA3 versus DG, and CA1 versus CA3. Significantly different m/z’s of interest were identified by MS/MS fragmentation directly from tissue. Proteomic data which were similarly significantly different for comparisons across DG, CA1, and CA3, were searched for proteins and enzymes in prominent, well known lipid metabolic pathways. These were then matched to significantly different lipids from the lipidomic data in the same pathways whose levels changed for the same comparison across DG, CA1, and CA3.

3. Results

Our workflow has four major steps as shown in Fig. 1. Tissue sections on ITO slides are imaged using MALDI-MSI-based lipidomics, focusing on phospholipids, including phosphatidylcholine (PC), phosphatidylethanolamine (PE), and sphingolipids (SM) (Fig. 1A). The slide is washed and stained for a trained neuropathologist to identify brain regions and cell types of interest. Cell bodies from regions of interest (ROIs) are excised from the same MALDI-imaged slide by LCMD and captured (Fig. 1B). Proteins from each ROI are extracted and their abundances are compared by relative quantification using tandem mass tag (TMT) based mass spectrometry (Fig. 1C). In parallel, following LCMD, the same brain sections are optically re-imaged (Fig. 1D) to spatially co-register the micro-dissected ROIs from LCMD with the MALDI-MS images to relate to the resulting lipidomic and proteomic data from the same ROIs (Fig. 1E). Using this approach, segmentation analysis of the optically scanned (Fig. 2A) and MALDI MS imaged tissue section was performed to determine ROIs (Fig. 2B) as in similar pipelines on other tissue types [15,16]. Although DG clearly segmented from the surrounding tissue, the CA1 and CA3 did not clearly segment (Fig. 2B). Due to this lack of differentiation for CA1 and CA3 from the surrounding tissue, we chose a histology-guided approach to guide the LCMD. Following LCMD, the tissue section was optically scanned a second time (Fig. 2C) using the same scanner as for the first optical image. The tissue features in this re-imaged section after LCMD were used to co-register the micro-dissected tissue with the MALDI imaging data and outline the micro-dissected ROIs for CA1, CA3, and DG on the MALDI and optical images (Figs. 2B and 3D). ROIs for CA1, CA3 and DG were outlined as shown in Fig. 2D.

Fig. 1.

The workflow for spatial multi-omic analysis of the trisynaptic circuit in the human hippocamus. (A) MALDI lipidomic imaging is performed, (B) followed by laser capture microdissection. (C) The dissected tissue is collected for TMT proteomics. (D) The dissected tissue section is re-scanned and (E) co-registered with the MALDI imaging dataset to generate regions of interest for combined proteomic-lipidomic data analysis.

Fig. 2.

Co-registration of optical images to generate ROIs from laser capture microdissection. (A) The original optical image taken for the MALDI-MSI experiment. (B) The tissue is re-scanned using the same scanner to produce a second optical image showing the areas cut by LCMD. (C) The MALDI-MSI segmentation map with rough outlines of CA1, CA3, and DG overlayed on the merged images from A and B. (D) Outlines of the ROIs generated by co-registering the optical images obtained prior to MALDI imaging in A and following LCMD in B showing the holes in the tissue where LCMD has excised tissue pieces and cell bodies, and drawing ROIs precisely around the tissue pieces and cell bodies excised with LCMD.

Fig. 3.

Expanded regions of interest from images of Fig. 2C–D with ROIs outlined, where in the scanned image (A) CA1 is in blue, CA3 is in orange, and DG is in white. The segmentation image with the various ROIs outlined in white. All ROI outlines are oversimplified to generally circle these ROIs for the purpose of visibility of ion images. The analyzed outlines of LCMD holes for each ROI are shown in Fig. 2D. Ion intensity images of two phosphatidylcholine species identified as (C) LPC O-16:1 [M+H]⁺ from m/z 480.3 and (E) PC 40:0 [M+H]⁺ from m/z 846.8. Corresponding ion intensity box-and-whisker plots of (D) m/z 480.3 and (F) m/z 846.8 with significant AUC values. ROC plots for all m/z’s can be found in the supporting information. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Receiver operating characteristic (ROC) analysis was performed on MALDI-MSI data to compare the abundances of analytes in the CA1, CA3, and DG ROIs (Table S1). Table S1 gives the average ROI peak intensity and ROC area under the curve (AUC) value for all 192 m/z’s in the peak list generated from the imaging data set. Comparing CA1 to CA3, 77 analytes were more abundant and 3 analyte less abundant. Comparing CA1 to DG, 18 analytes were more abundant and 16 analytes were less abundant, while comparing CA3 to DG, 2 analytes were more abundant and 82 analytes were less abundant. Although segmentation did not differentiate CA1 and CA3 from the surrounding tissue, MALDI-MSI documented differences between these three ROIs.

Using on-tissue tandem mass spectrometry, we chose 8 well resolved analytes to identify and quantify. These 8 analytes represent 7 known classes of lipids (Figs. 3, 4, S18–S27) We observed significant differences in the relative abundance of 6 lipid analytes in one or more of the three trisynatic circuit ROIs CA1, CA3, and DG. The abundance of lyso-phosphatidylcholine (LPC) O-16:1 was highest in CA1 (Fig. 3C and D), whereas PC 40:0 was highest in the DG (Fig. 3E and F). Carnitine (Car) 22:4, LPE 20:5, LPE 20:3, PC 34:0, and SM 38:3; O₃ were highest in CA1 (Fig. 4-F, I–L). PA 30:3; O₃ had similar abundances in all three ROIs (Fig. 4G and H).

Fig. 4.

(A) Ion intensity image of m/z 476.6 identified as Car 22:4 [M+H]⁺. (B) Ion intensity box-and-whisker plot of m/z 476.6 with significant AUC values. (C) Ion intensity image of m/z 500,4 identified as LPE 20:5 [M+H]⁺. (D) Ion intensity box-and-whisker plot of m/z 500.4 with significant AUC values. (E) Ion intensity image of m/z 504.5 identified as LPE 20:3 [M+H]⁺. (F) Ion intensity box-and-whisker plot of m/z 504.5 with significant AUC values. (G) Ion intensity image of m/z 657.3 identified as PA 30:3; O₃ [M+H]⁺. (H) Ion intensity box-and-whisker plot of m/z 657.3. No significant AUC values. (I) Ion intensity image of m/z 762.8 identified as PC 34:0 [M+H]⁺. (J) Ion intensity box-and-whisker plot of m/z 762.8 with significant AUC values. (K) Ion intensity image of m/z 777.5 identified as SM 38:3; O₂ [M+Na]⁺. (L) Ion intensity box-and-whisker plot of m/z 777.5 with significant AUC values. All ROI outlines are oversimplified to generally circle these ROIs for the purpose of visibility of ion images. The analyzed outlines of LCMD holes for each ROI are shown in Fig. 2D.

Using TMT based mass spectrometry, we directly compared the relative abundance of proteins extracted from approximately 500 cells in each of the three micro-dissected ROIs of CA1, CA3 and DG, and from each of three serial hippocampal tissue sections following MALDI imaging (Figs. 5 and 6). Globally, 3411 proteins groups were identified, 83 % or 2823 proteins of which were quantified from a total of 20,328 peptides detected in 49,030 peptide spectral matches (PSMs). However, 931 proteins had p-values <0.05 for at least one of the three abundance ratios, CA1/DG, CA3/DG and CA1/CA3. The distribution of these normalized proteins abundances, based on the sum of the reporter ion signal-to-noise ratios for all peptides from the same protein, were virtually the same across the three replicate hippocampal ROIs within and across the three serial sections (Fig. 5A). However, both unsupervised principal component (PC) analysis (Fig. 5B), accounting for 80.6 % of the variation, and the heat map (Fig. 5C), based on hierarchical clustering based on Euclidean distance and complete (furthest neighbors) linkage, showed that the protein abundances fall into three distinct groupings based on the three hippocampal regions, CA1, CA3 and DG. Moreover, heat map analysis indicated that DG protein abundances clustered further from the protein abundances in the CA1 or CA3 regions (Fig. 5C), suggesting that specific protein abundances are more similar between the CA1 and CA3 regions than with the DG.

Fig. 5.

Distribution of normalized protein abundances filtered for proteins with abundance ratio p-values ≤0.05 in any ratio of three human hippocampal regions. (A) Distribution of abundances in each ROI of biological triplicate from the CA1, CA2 and DG. (B) Clustering of abundances in unsupervised principal component analysis. (C) Hierarchical clustering abundances based on Euclidean distance and complete (furthest neighbors) linkage.

Fig. 6.

Pairwise comparisons of individual protein normalized abundance ratios in the three hippocampal regions. Protein relative abundance ratios ≥20 % (log2 = 0.26) up (pink area) or down (green area) with p-values <0.05 (−log10 = 1.3) shown in boxed areas. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Accordingly, pairwise comparisons of individual protein abundances in these three hippocampal regions revealed slightly more protein abundance differences between DG and CA1 or CA3 than between CA1 and CA3 (Fig. 6). At ≥20 % abundance change with p-values <0.05, there were 720 or 467 protein abundance differences between DG and CA1 or CA3, respectively, and 312 protein abundance differences between CA1 and CA3. Membrane or vesicle associated proteins had similar abundances in CA1 and CA3, but lower abundance in DG by 1.7–2.6-fold at p-values <0.05 and are listed in Table 1. Mostly nuclear proteins had similar abundances in CA1 and CA3, but higher abundances in DG by 3.2–7.9-fold at p-values <0.009 and are listed in Table 2. Proteins that are higher in abundance by > 2 fold (p-value <0.05) specifically in CA1 relative to CA3 or DG are listed in Table 3. Proteins that are higher in abundance by > 2 fold (p-value <0.05) specifically in CA3 relative to CA1 or DG are listed in Table 4.

Table 1.

Membrane or vesicle associated proteins with similar abundances in CA1 and CA3, but lower abundance in DG by 1.7–2.6-fold at p-values <0.05.

Accession	Description	Abundance Ratio: (CA1)/ (CA3)	Abundance Ratio: (CA1)/ (DG)	Abundance Ratio: (CA3)/ (DG)	Abundance Ratio P-Value: (CA1)/ (CA3)	Abundance Ratio P-Value: (CA1)/ (DG)	Abundance Ratio P-Value: (CA3)/ (DG)
XP_011544017.1	proline-rich transmembrane protein 2 isoform X1 [Homo sapiens]	0.81	2.15	2.64	0.6790	0.0460	0.0167
NP_001123475.1	synaptoporin isoform 1 [Homo sapiens]	0.83	1.84	2.21	0.5005	0.0023	0.0009
NP_443106.1	syntaxin-1B [Homo sapiens]	0.88	1.75	1.99	0.5056	0.0076	0.0025
NP_598006.1	synapsin-1 isoform Ib [Homo sapiens]	0.89	1.77	1.99	0.2244	0.0008	0.0002
NP_114101.4	voltage-dependent calcium channel gamma-8 subunit [Homo sapiens]	1.07	2.05	1.91	0.2767	0.0001	0.0002
XP_011537012.1	synaptotagmin-1 isoform X1 [Homo sapiens]	0.93	1.76	1.90	0.9800	0.0005	0.0004
XP_005260865.1	synaptosomal-associated protein 25 isoform X1 [Homo sapiens]	0.97	1.77	1.83	0.9961	0.0004	0.0004
NP_001308032.1	sodium bicarbonate cotransporter 3 isoform d [Homo sapiens]	0.94	1.70	1.82	0.9911	0.0229	0.0198
NP_001365261.1	protein NDRG4 isoform 8 [Homo sapiens]	1.06	1.84	1.74	0.5614	0.0027	0.0073
XP_016859718.1	anoctamin-7 isoform X3 [Homo sapiens]	1.12	1.82	1.62	0.9576	0.0179	0.0246
NP_444252.1	profilin-2 isoform a [Homo sapiens]	1.20	1.93	1.61	0.1693	0.0025	0.0208
NP_077305.2	EF-hand domain-containing protein D2 [Homo sapiens]	1.14	1.78	1.56	0.9990	0.0194	0.0204
NP_055446.2	SH3 and PX domain-containing protein 2A isoform 1 [Homo sapiens]	0.87	1.50	1.73	0.5882	0.1052	0.0223
XP_016874261.1	arf-GAP with GTPase, ANK repeat and PH domain-containing protein 2 isoform X1 [Homo sapiens]	0.93	1.61	1.72	0.9993	0.0287	0.0275
NP_114430.2	neurocalcin-delta [Homo sapiens]	0.93	1.60	1.72	0.9991	0.0040	0.0038

Table 2.

Mostly nuclear proteins with similar abundances in CA1 and CA3, but higher abundances in DG by 3.2–7.9-fold at p-values <0.009.

Accession	Description	Abundance Ratio: (CA1)/ (CA3)	Abundance Ratio: (CA1)/ (DG)	Abundance Ratio: (CA3)/ (DG)	Abundance Ratio P-Value: (CA1)/ (CA3)	Abundance Ratio P-Value: (CA1)/ (DG)	Abundance Ratio P-Value: (CA3)/ (DG)
NP_001073027.1	heterogeneous nuclear ribonucleoprotein U-like protein 2 [Homo sapiens]	0.83	0.31	0.37	0.2295	0.0001	0.0003
NP_066018.1	nuclear receptor coactivator 5 isoform 1 [Homo sapiens]	0.81	0.30	0.37	0.6698	0.0055	0.0136
NP_110517.2	beta-catenin-like protein 1 isoform 1 [Homo sapiens]	0.87	0.31	0.36	0.0707	0.0000	0.0000
NP_001120700.1	chromobox protein homolog 1 [Homo sapiens]	0.85	0.30	0.35	0.3266	0.0017	0.0072
XP_016856352.1	TAR DNA-binding protein 43 isoform X1 [Homo sapiens]	0.82	0.29	0.35	0.1946	0.0001	0.0003
NP_001460.1	X-ray repair cross-complementing protein 6 isoform 1 [Homo sapiens]	0.87	0.30	0.35	0.1153	0.0000	0.0000
NP_005959.2	heterogeneous nuclear ribonucleoprotein M isoform a [Homo sapiens]	0.83	0.29	0.35	0.1516	0.0000	0.0001
NP_001358848.1	histone H3.X [Homo sapiens]	0.89	0.30	0.34	0.5628	0.0009	0.0022
NP_116126.3	lamin-B2 [Homo sapiens]	0.80	0.27	0.34	0.0016	0.0000	0.0000
XP_011514658.1	condensin-2 complex subunit G2 isoform X1 [Homo sapiens]	0.88	0.29	0.34	0.5703	0.0015	0.0037
NP_006816.2	cytoskeleton-associated protein 4 [Homo sapiens]	0.81	0.27	0.33	0.2456	0.0001	0.0002
NP_003086.1	small nuclear ribonucleoprotein F [Homo sapiens]	0.95	0.30	0.32	0.5993	0.0003	0.0005
NP_005564.1	lamin-B1 isoform 1 [Homo sapiens]	1.01	0.31	0.31	0.9624	0.0001	0.0001
NP_006436.3	pre-mRNA-processing-splicing factor 8 [Homo sapiens]	0.93	0.28	0.31	0.2459	0.0000	0.0000
NP_112738.1	heterogeneous nuclear ribonucleoprotein D0 isoform a [Homo sapiens]	0.95	0.29	0.30	0.7644	0.0003	0.0006
NP_778225.1	histone H2B type 3-B [Homo sapiens]	0.89	0.27	0.30	0.7712	0.0267	0.0625
XP_011542031.1	core histone macro-H2A.1 isoform X1 [Homo sapiens]	0.93	0.28	0.30	0.1409	0.0000	0.0000
NP_055792.1	apoptotic chromatin condensation inducer in the nucleus isoform 1 [Homo sapiens]	0.94	0.28	0.30	0.9090	0.0000	0.0000
NP_001342338.1	histone H3.2 [Homo sapiens]	1.00	0.30	0.30	0.9958	0.0001	0.0001
XP_005268923.1	keratin, type II cuticular Hb6 isoform X1 [Homo sapiens]	1.03	0.30	0.29	0.8158	0.0223	0.0464
NP_114032.2	heterogeneous nuclear ribonucleoprotein U isoform a [Homo sapiens]	0.82	0.23	0.28	0.7957	0.0013	0.0023
NP_002477.1	nuclear cap-binding protein subunit 1 isoform 1 [Homo sapiens]	1.12	0.31	0.28	0.8630	0.0035	0.0022
NP_003539.1	histone H4 [Homo sapiens]	0.88	0.23	0.26	0.9658	0.0074	0.0096

Table 3.

Proteins that are higher in abundance by > 2 fold (p-value <0.05) specifically in CA1 relative to CA3 or DG.

Accession	Description	Abundance Ratio: (CA1)/ (CA3)	Abundance Ratio: (CA1)/ (DG)	Abundance Ratio: (CA3)/ (DG)	Abundance Ratio P-Value: (CA1)/ (CA3)	Abundance Ratio P-Value: (CA1)/ (DG)	Abundance Ratio P-Value: (CA3)/ (DG)
XP_016883895.1	chloride intracellular channel protein 6 isoform X1 [Homo sapiens]	2.65	2.87	1.08	0.0197	0.0273	0.9575
NP_002211.1	inositol-trisphosphate 3-kinase A [Homo sapiens]	2.61	2.14	0.82	0.0005	0.0018	0.2701
NP_006530.1	voltage-dependent calcium channel gamma-3 subunit [Homo sapiens]	2.06	2.41	1.17	0.0086	0.0016	0.2331
NP_001339115.1	ankyrin repeat and sterile alpha motif domain-containing protein 1B isoform m [Homo sapiens]	1.96	2.32	1.18	0.0021	0.0004	0.1510
NP_001295083.1	ras-related protein Rab-15 isoform 2 [Homo sapiens]	1.81	2.14	1.19	0.0175	0.0027	0.2223

Table 4.

Proteins that are higher in abundance by > 2 fold (p-value <0.05) specifically in CA3 relative to CA1 or DG.

Accession	Description	Abundance Ratio: (CA1)/ (CA3)	Abundance Ratio: (CA1)/ (DG)	Abundance Ratio: (CA3)/ (DG)	Abundance Ratio P-Value: (CA1)/ (CA3)	Abundance Ratio P-Value: (CA1)/ (DG)	Abundance Ratio P-Value: (CA3)/ (DG)
NP_005373.2	neurofilament medium polypeptide isoform 1 [Homo sapiens]	0.334	0.887	2.661	0.0155	0.8755	0.0091
NP_001359503.1	ataxin-2 isoform 4 [Homo sapiens]	0.454	1.015	2.235	0.0330	0.9206	0.0530
NP_001161419.1	mitochondrial import inner membrane translocase subunit Tim17-B isoform 1 [Homo sapiens]	0.456	1.149	2.517	0.0420	0.4362	0.0095
NP_000953.2	prostaglandin G/H synthase 1 isoform 1 precursor [Homo sapiens]	0.477	1.152	2.416	0.0003	0.2410	0.0001
XP_011528502.1	neurofilament heavy polypeptide isoform X1 [Homo sapiens]	0.491	1.037	2.113	0.0552	0.6493	0.0186
NP_597680.2	asparagine synthetase [glutamine-hydrolyzing] isoform a [Homo sapiens]	0.526	1.146	2.181	0.0517	0.6668	0.0181
NP_001352542.1	ribosome-binding protein 1 isoform 1 [Homo sapiens]	0.437	0.824	1.885	0.0033	0.5532	0.0094
NP_001012267.1	centromere protein P isoform a [Homo sapiens]	0.475	0.831	1.747	0.0397	0.5965	0.1415

Our comparative analysis of region-specific lipidomic and proteomic data showed that two specific lipid species matched the protein changes observed between the three hippocampal regions of CA1, CA3 and DG. Lysophosphatidic acid phosphatase type 6 (ACP6) protein, which hydrolyzes lysophosphatidic acid lipids, including LPA 20:1, to the corresponding monoacyglycerol [23], was significantly more abundant in CA3 versus DG (Fig. 7C), which is consisted with higher enzymatic activity resulting in lower abundance of LPA 20:1 substrate in CA3 versus DG (Fig. 7A and B). However, the observed minor differences in LPA 20:1 abundance in CA1 versus DG are not explained by 2-fold more abundance of ACP6 in the CA1 versus DG. Lysophosphatidylcholine acyltransferase 4 (LPCAT4) converts a subset of lysophospholipids to phospholipids, which includes the conversion of lysophosphatidylcholines to phosphatidycholines [24]. Differences in LPCAT4 protein levels (Fig. 7F) did not explain its lipid substrate levels, PC 26:0 (Fig. 7D–F), in either CA3 or CA1 versus DG, suggesting different lipid/protein regulation in these ROIs.

Fig. 7.

(A) The ion intensity image of m/z 465.3 identified as LPA 20:1 [M+H]⁺. (B) Ion intensity box-and-whisker plot of m/z 465.3 with significant AUC values. (C) Normalized abundance of lysophosphatidic acid phosphatase type 6 isoform 1 [Homo sapiens] (ACP6) with significant p values. (D) The ion intensity image of m/z 650.4 identified as PC 26:0 [M+H]⁺. (E) Ion intensity box-and-whisker plot of m/z 650.4 with significant AUC values. (F) Normalized abundance of lysophospholipid acyltransferase LPCAT4 [Homo sapiens] (LPCAT4) with significant p values. All ROI outlines are oversimplified to generally circle these ROIs for the purpose of visibility of ion images. The analyzed outlines of LCMD holes for each ROI are shown in Fig. 2D.

4. Discussion

MALDI-MSI-based spatial-omic analysis can be adapted to imaging of a wide range of compounds, including metabolites [18,25–27], neurotransmitters [28], drugs [29], N-glycans [30–32], MALDI immunohistochemistry (IHC) [33] or peptides [25,30,34] for co-registering with LCMD and MS-based proteomics to detect protein pathways in the same regions of interest (ROIs). There is increasing interest in the biomedical research community to collect multi-omics data sets by MALDI-MSI [30,33]. A pipeline using multiple MALDI imaging data sets followed by LCMD and spatial proteomics provides an unprecedented amount of molecular information from a single tissue section. Because of its well-characterized histological anatomy and physiology, the hippocampal trisynaptic circuit of the human brain is ideal to test our novel MALDI-MSI lipidomics-LCMD-MS-proteomics pipeline. While MSI approaches for spatially resolved lipidomic and proteomic approaches have been applied to study the molecular mechanisms of neurodegeneration, including focal cortical dysplasia [35], Alzheimer’s disease [36], glioblastoma [37], traumatic brain injury [38] and depression [39], the human trisynaptic circuit for learning and memory has not yet been studied in the spatial-omics space. However, MSI data sets to guide the LCMD is not always possible even in brain tissue sections with clear histological anatomy. Indeed, even though the DG, CA3 and CA1 regions of the trisynaptic circuit are histologically distinct, our MALDI MS imaging found molecular classes defining only the DG. The full CA3 or CA1 regions were not molecularly distinct from the surrounding tissue by MALDI-MSI lipidomics.

We, therefore, took a histological approach for guiding LCMD. Our pipeline, thus, does not depend on MALDI-MSI data for ROI analysis to determine the LCMD regions for microdissection. This is a distinction from previously published studies which utilized segmentation analysis of MALDI-MSI data to generate ROIs to guide LCMD [15,16,40]. Our pipeline is ideal for any tissue with physiologically distinct regions which do not display clear MALDI-MSI based segmentation from the surrounding tissue. Moreover, because histology stains are commonly used on the same tissue after MALDI imaging to support tissue and cell contrast for LCMD, a histology-guided approach can be effectively implemented for many tissues. The presented pipeline used the Histogene staining solution as a counter stain. However, any histology stain, including hematoxylin and eosin (H&E), could work equally well.

In addition to being histology stain compatible, we found that cells could be excised by LCMD from both uncoated ITO and poly-l-lysine coated ITO slides without compromising the MS-based proteomics data (Fig. 5). Previous studies have shown that cutting cells from tissue on polyethylene naphthalate (PEN) membrane coated slides generally improves the number of proteins identified from a sample as compared to an unmodified ITO slide [15] but compromises the MALDI imaging [16]. To our knowledge, no one has reported testing poly-l-lysine coated slides which is a common modification used to increase tissue/cell adherence to ITO slides, especially in the case of fatty tissues or cell imaging. We found that, unlike the PEN membrane slides, the poly-l-lysine coated slides not only do not impact MALDI imaging or LCMD cell collection but also do not reduce the number of proteins identified from downstream proteomic analysis (Fig. 5).

Most notably, we pioneered the novel approach of co-registering optical images from the same tissue section prior to MALDI imaging and following LCMD. This new way of outlining ROIs on MALDI imaging data uses the holes in the tissue, created by LCMD, as outlines for the ROIs in the MALDI imaging data analysis. In this way, the MALDI imaging data, acquired in the beginning of the pipeline, is co-registered with the micro-dissected regions for proteomics analysis, acquired at the end of the pipeline. This a simple and straightforward process because the optical image obtained prior to MALDI imaging and after LCMD are from the same tissue section, creating inherently co-registered MALDI imaging data with the corresponding optical image, and with the spatial lipidomic and proteomic data.

Despite the limited sample size presented in this study, we can tentatively draw some conclusions from the MALDI imaging and spatial proteomics data sets. Notably, lipidomics from MALDI imaging shows that the CA1 and CA3 do not readily segment out from the surrounding tissue, however, DG does. Proteomics data sets show that the CA1 and CA3 regions display proteomes which are more similar to each other than to DG. The proteomics is in good agreement with recent findings from single cell RNA-sequencing studies of the hippocampus, where a similar transcriptome pattern was observed between the CA1 and the CA3, but both were more different from the DG transcriptome [41,42]. Anatomically, the entorhinal cortex projects to the dentate gyrus of the core hippocampal formation, which has been a location for adult mammalian neurogenesis [43,44]. Both CA3 and CA1 pyramidal neurons are involved in different hippocampal circuitry, with CA1 being a principal output center for the hippocampal formation [45,46]. Taken together, these neuroanatomical differences are reflected at the molecular level in transcriptome and as now shown here, in the lipidome and proteome.

Most of the proteins detected in our study have previously been associated with the human hippocampus in studies of gene expression of bulk human hippocampal tissue. Here we report, for the first time, their differential protein expression in the CA1, CA3, and DG. Our comparative analysis of region-specific lipidomic and proteomic data highlighted two important enzymes in phospholipid metabolism, ACP6 [23] and LPCAT4 [24]. ACP6 was significantly more abundant in CA3 than in DG, which portends that the corresponding lower abundance of the ACP6 substrate, LPA 20:1, in CA3 was due to a higher ACP6 enzyme activity in CA3. Conversely, protein abundance of LPCAT4 alone was not sufficient to explain the relative abundance of its substrate abundance, PC 26:0, in CA3 versus DG, because both enzyme and substrate are lower in the CA3 than DG. Little is known about lipid metabolism in the hippocampus, including ACP6 and LPCAT4 expression levels in different regions of the hippocampus. Our pipeline provides an initial step in studying the combined and differential spatial regulation of the lipidome and proteome within the trisynaptic circuit and how they impact human hippocampal functions.

5. Conclusions

We have demonstrated a novel histology-guided approach to combined MALDI imaging, LCMD, and spatial proteomics which does not rely on segmentation analysis to determine ROIs. Rather, tissue holes from LCMD are registered onto the corresponding MALDI imaging data for matching ROIs from the same tissue section. Further, we determined that the addition of a poly-l-lysine coating on ITO slides does not impact LCMD or spatial proteomics. Finally, we have demonstrated that both the lipidomes and the proteomes of CA1 and CA3 pyramidal neurons are more similar to each other than those of the DG in the trisynaptic circuit of the human brain, which is consistent with previously reported transcriptomic analyses [4]. Future studies using this newly developed pipeline will reveal molecular determinants of neurodevelopment, aging, and disease states, and improve insights into the development of disease-related molecular changes in these important hippocampal regions.

HIGHLIGHTS.

• We developed a spatial lipidomics and proteomics pipeline using coregistration of the laser capture microdissected tissue section to generate regions of interest rather than segmentation analysis.

• We applied this pipeline to the human brain trisynaptic circuit to gain a deeper understanding of the native lipidome and proteome

• We demonstrate differential expression of lipids and proteins in the trisynaptic circuit consisting of CA1, CA3, and the dentate gyrus.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.aca.2026.345110.

Data availability

Data will be made available on request.