High-throughput fluorescence-guided sequential single-cell MALDI-ICC mass spectrometry

Abstract

Lipids are a diverse class of biomolecules essential for brain function, yet their cell-type-specific distributions remain underexplored, presenting significant knowledge gaps in the era of single-cell biology. Traditional bulk measurements provide valuable insights into lipid composition across brain regions but lack the resolution to distinguish lipid profiles at the single-cell level. To address this, we introduce fluorescence-guided sequential single-cell mass spectrometry (SSMS), an automated workflow combining untargeted lipid profiling with antibody-targeted protein detection via photocleavable mass tags, enabling neurolipidomic classification of cell types and cell states. We applied this approach to rodent hippocampal cells, analyzing over a thousand single cells and annotating more than a hundred lipid species with complementary LC-MS/MS measurements. Our findings show that phosphatidylcholine (PC) species are predominantly enriched in oligodendrocytes and neurons compared to astrocytes, while hexosylceramide (HexCer) species are differentially expressed across these cell types. Furthermore, neuronal state analysis revealed an enrichment of phosphatidylethanolamines (PEs) in presynaptic neurons, while non-presynaptic neurons exhibited a more diverse lipid composition, including HexCer, PC, sphingomyelin, triacylglycerol, and PE. Our findings provide new insights into brain lipid heterogeneity with cell-type and cell-state specificity and extend capabilities of next-generation single-cell mass spectrometry to map brain biochemistry.

Graphical Abstract:

INTRODUCTION

The mammalian brain consists of a heterogeneous population of cells, each playing a critical role in maintaining brain function and health. Cell-to-cell communication, such as interactions between glial cells and neurons, facilitates biochemical exchanges essential for neuronal survival.^1–3 Among these biochemicals, lipids are one of the most abundant in the brain, with over 100,000 distinct species estimated in the human brain that contribute to cell-to-cell signaling, energy metabolism, and membrane structure.^4–7 Lipid alterations have been implicated in numerous neurodegenerative diseases, yet identifying disease-specific lipid signatures remains challenging due to the vast diversity of lipid species.^8–13 Many of these lipid changes are cell-type specific, making it essential to investigate neurolipidomics at the single-cell level to identify lipid features associated with cell types, which may have relevance for lipid-targeted therapeutics and drug delivery.¹⁴

Mass spectrometry is a robust technique for single-cell studies, capable of detecting a wide range of analytes, including small metabolites, peptides, proteins, and lipids.^15–21 Probe-based MS techniques, such as matrix-assisted laser desorption/ionization (MALDI), enable comprehensive single-cell biochemical profiling without the need for bulk sample homogenization. Moreover, single-cell MALDI preserves the native molecular state within its cellular context, providing insights into molecular interactions that liquid chromatography-mass spectrometry (LC-MS/MS) techniques cannot capture.^22,23 Additionally, MALDI-MS has been performed in conjunction with fluorescence-based immunocytochemistry (ICC) to associate metabolic profiles to canonical cell types.^24,25 However, traditional fluorescence-based ICC is inherently limited by spectral overlap, restricting cell-type specificity analysis and making it challenging to use more than three markers at a time. MALDI-immunohistochemistry (MALDI-IHC) overcomes this limitation by introducing photocleavable mass-tagged (PCMT) antibody immunolabeling, which uses peptide markers detectable via MS to enable multiplexed protein detection beyond traditional fluorescence-based antibody labeling, with simultaneous integration of up to 28 probes reported in literature thus far.^26–28 In contrast to flow cytometry–based cell sorting, which is only applicable to dissociated single cells, PCMT immunolabeling can be applied to both isolated single-cell analysis and spatial mapping. While MALDI-IHC has been successfully applied in tissue-scale MSI²⁷ and in a single-cell MSI approach to study lipid differences in patient-derived glioblastoma cells²⁹, its full potential for high-throughput, omics-based single-cell analysis across multiple brain cell types and cell states remains underexplored.

Here, we introduce a two-step, fluorescence-guided sequential single-cell MS (SSMS) approach that enables the direct targeting of isolated cells from the rodent hippocampus—a brain region known for its diverse cell populations and critical role in neurodegenerative diseases such as Alzheimer’s disease (AD).^30,31 This SSMS workflow integrates high-throughput, fluorescence-guided single-cell acquisition using the previously developed microMS³² workflow, a Python-based program that enables automated laser targeting of thousands of single cells within minutes, whereas traditional MALDI raster scanning would take hours to accomplish. MiicroMS, short for microscopy-guided MS, achieves this by converting fluorescence microscopy cell coordinates to MALDI system coordinates, using fiducial marker training to ensure accurate targeting of the laser to the cell locations. It also incorporates a step to remove locations of cell clusters, for true single-cell measurements. Utilizing this approach in SSMS, individual cells are first lipid-profiled through automated MALDI-MS (Figure 1A). Next, the MALDI matrix is removed, and the single cells are immunolabeled with PCMTs via ICC. The same cells are then analyzed in a second round of MS measurements. Each protein targeted by an antibody is associated with a specific mass tag, which is used to classify the cell into a specific cell type or cell state for downstream lipidomic analysis (Figure 1B–C). This automated SSMS workflow, developed in tandem with a data processing and statistical analysis pipeline, provides accurate and specific classification of cells, offering insights into cell-type-specific lipid heterogeneity in the functionally complex hippocampus, and can be extended to investigate other regions of the brain.

Figure 1.

Sequential single-cell MALDI-ICC mass spectrometry (SSMS) workflow. (A) Single cells isolated from the rat hippocampus are stained with Hoechst nuclear dye, deposited onto an ITO-coated glass slide, and imaged using optical microscopy. The cells are then lipid profiled via automated MALDI-MS. (B) After matrix removal; the same cells are immunocytochemically (ICC) labeled with photocleavable mass tags (PCMTs) and subjected to a second round of MALDI-MS analysis. (C) For each cell, the mass-tag spectrum is paired with its lipid spectrum through coordinate matching. Each cell is then assigned to a specific class based on the presence of a mass-tag corresponding to a specific antibody, defining cell type and cell state for downstream lipidomic analysis.

METHODS

General.

All chemicals were obtained from Sigma-Aldrich unless specified otherwise.

Animals.

Male Sprague-Dawley outbred rats (Rattus norvegicus, ~ 2.5-month-old) were housed on a 12 h light cycle and fed ad libitum. Animal euthanasia was performed in accordance with Illinois Institutional Animal Care and Use Committee (IACUC) protocol 23228, and in full compliance with both federal and ARRIVE guidelines for the humane care and treatment of animals.

Single-Cell Isolation.

Hippocampal tissues from three male rats were dissociated using a papain dissociation system (Worthington Biochemical, Lakewood, NJ) as previously described.³⁵ Briefly, the hippocampal region of the brain was sampled via a biopsy punch, enzymatically treated with the papain dissociation system, and then mechanically dissociated in modified Gey’s balanced salt solution (mGBSS). Cells on each slide were stained with Hoechst 33342 (0.1 μg/mL in mGBSS). Single cells from three animals were each dispersed at three different spots on indium tin oxide (ITO)-coated glass slides (Delta Technologies, Loveland, CO). After allowing cell adhesion to the slide surface, the extracellular mGBSS was replaced with a mixture of 33% glycerol and 67% mGBSS. The glycerol-containing medium was then removed by aspiration. Prepared samples were stored at 4°C for further analysis.

Brightfield and Fluorescence Microscopy.

Single cells were washed with 150 mM ammonium acetate preceding imaging to facilitate microscopy and MALDI matrix application. Brightfield and fluorescence images using DAPI channel (ex. 335–383 nm; em. 420–470 nm) were acquired using an Axio Imager M2 (Zeiss, Jena, Germany) at 10x magnification with 10% overlap. The images were then exported as big TIFF files using ZEN 3.1, blue edition (Zeiss, Jena, Germany).

Matrix Application.

Matrix was applied via sublimation using an HTX SubliMATE (HTX Technologies LLC, Chapel Hill, NC). 2,5- dihydroxybenzoic acid (DHB) was selected as the MALDI matrix following the Miralys Imaging Laboratory Workflow (AmberGen, Billerica, MA) due to its suitability for positive-ion mode lipid analysis and its consistent performance in preserving spatial localization of lipids. We adapted the application method by using the established DHB sublimation protocol included with the HTX SubliMATE system, instead of wet spraying, to achieve a homogeneous, fine-grained crystal layer while minimizing analyte delocalization. Briefly, 8 mg/mL of DHB was dissolved in 1.5 mL of HPLC grade acetone and added to three different regions on the matrix Wafer and allowed to dry. Ice was used to cool the top of the SubliMATE and allowed to thermally equilibrate under vacuum (~30 mTorr) for 10 minutes. DHB was then sublimed at 200 °C for 5 minutes. The ice was removed from top of the chamber and replaced with a prewarmed metal disc. The ceramic heater temperature was set to 0 °C for 15–20 minutes to allow the sample to equilibrate to room temperature before releasing the vacuum. A total of 0.7 ± 0.2 mg (~37.3 μg/cm²) of DHB matrix was applied on the slides. The samples were then placed inside a vacuum desiccator until ready for SSMS, which was performed within 12–24 hours after matrix application.

MALDI-QTOF Measurement with Single-Cell Automation.

microMS was used to perform automated single cell run on each animal replicate separately, as described previously.³² In this process, hundreds to thousands of single cells were selected using the Hoechst staining for cell identification. Cells were then filtered based on size, shape, and a distance between each targeted cells of 100 μm apart such that cells within this radius would not be analyzed simultaneously. Each run was evaluated by a test run of a few known spots on the slide to assess the targeting efficiency of the image co-registration. Single cell analysis was performed on a Bruker timsTOF fleX MALDI-2 MS (Bruker Daltonics, Billerica, MA). DHB ion adducts were used as reference peaks for lock mass calibration. Specifically, ions at m/z 137.0233 ([DHB + H - H₂O]⁺), m/z 155.0339 ([DHB + H]⁺), m/z 273.0392 ([2DHB + H - 2H₂O]⁺), and m/z 409.0554 ([3DHB - 3H₂O + H]⁺). Prior to each analysis, red phosphorus cluster ions were used for mass calibration.

Step I – Endogenous Lipid Profiling.

Mass spectra were acquired in standard MALDI mode, in positive polarity, over an m/z range of 200–1600, using a 50 μm laser spot size, sufficient to cover the overall dimensions of dissociated hippocampal cells including their processes, and a single smart beam with beam scanning enabled, at 45–50% laser intensity. Each acquisition consisted of 1000 shots at a laser repetition rate of 5000 Hz. Detailed instrument tuning parameters are provided in (SI Table S1).

Mass-Tag Immunolabeling.

PCMT ICC immunolabeling was performed using the MALDI HiPLEX-IHC Miralys Imaging Laboratory Workflow (AmberGen, Billerica, MA) with slight modifications as follows: single cells were washed twice with cold acetone (−80 °C) for 3 minutes per wash, dried with N₂ at room temperature, and then fixed in 4% PFA in 1xPBS for 5 minutes. The blocking buffer was modified to include 5% (w/v) BSA, 2% (v/v) normal rabbit serum, 2% (v/v) normal mouse serum (Jackson ImmunoResearch, West Grove, PA), 0.1% Tween^™ 20, and 0.05% octyl β-D-glucopyranoside (OBG) in 1xTBS. The list of antibodies and their respective concentrations is provided in (SI Table S2). Each region of interest (ROI) containing multiple targeted cells was outlined using a hydrophobic PAP pen and incubated with 67 μL of the mass-tag mixture, covering an area of ~2.9 ± 0.2 cm². For cell retainment assessment, Hoechst staining was reapplied, followed by photocleavage of the mass-tags and matrix application, as described above in the matrix application section and provided in (SI Figure S1).

Step II – Mass-Tag Immunolabeling Single-Cell Analysis.

Mass spectra were acquired in standard MALDI mode, in positive polarity, over an m/z range of 600–2000, using a 50 μm laser spot size and a single smart beam with beam scanning enabled, at 70–80% laser intensity. Each acquisition consisted of 1000 shots at a laser repetition rate of 10000 Hz. High-sensitivity detection was enabled to improve the detection of low sample amounts. The instrument tuning parameters are provided in (SI Table S3). Cell locations were determined by converting pixel coordinates from the microscopy image into physical coordinates of the MTP Slide Adaptor II using microMS.³² A distance filter of 100 μm was applied to ensure each acquisition sampled only one cell at a time.

Tandem MS Measurement.

Rat hippocampal tissue was isolated, freshly frozen, and stored at −20 °C. Lipid extraction was performed using the Bligh and Dyer method.³⁴ Briefly, the hippocampus from one hemisphere of the brain (18 mg) was removed from -20°C, immediately placed into a 0.5 mL tube containing beads from the Precellys Lysing Kit (Bertin Technologies, Rockville, MD) and processed for extraction. A chilled (1:2, v/v) mixture of chloroform:methanol was added to the tissue, which was homogenized using a Precellys tissue homogenizer at 1 cycle of 30 seconds at 1500 rpm. Additional chloroform and water were added to adjust the solvent ratio to (1:1:0.9, v/v/v) chloroform:methanol:water. The mixture was vortexed and centrifuged at 14,000 rpm for 10 minutes at 4 °C. The lower organic phase was collected into a new tube, washed with water (1:0.9, v/v, organic layer:water), and centrifuged again at 14,000 rpm for 5 minutes at 4 °C. The lipid extract was dried in a speed vacuum at room temperature and stored at -20°C until further analysis.

The liquid chromatography-tandem mass spectrometry (LC-MS/MS) was optimized based on a previously reported method.⁴¹ Briefly, the lipid extract was redissolved in 100 μL of (4:3:1, v/v/v) isopropanol:acetonitrile:water, quickly centrifuged, and the supernatant was transferred to a new tube. LC-MS/MS analysis was conducted using a Waters Acquity UHPLC system coupled to a Synapt G2-Si time-of-flight mass spectrometer (Waters Corporation, Milford, MA) equipped with an electrospray ionization (ESI) source. Chromatographic separation was performed with a CORTECS^™ UPLC C18 column (100 mm L × 2.1 mm inner diameter, 1.6 μm particle size) maintained at 50 °C. The binary solvent system consisted of mobile phase A (acetonitrile/water, 60:40, v/v) with 10 mM ammonium formate and 0.1% (v/v) formic acid, and mobile phase B (isopropanol/acetonitrile, 90:10, v/v) with 10 mM ammonium formate and 0.1% (v/v) formic acid. The gradient program was as follows: 60% mobile phase B at 0 minutes, 57% B at 2 minutes, 50% B at 2.1 minutes, 46% B at 12 minutes, 30% B at 12.1 minutes, 1% B at 18 minutes, returning to 60% B at 18.1 minutes and held until 20 minutes. A blank of LC-MS grade water along with three replicate chromatograms is provided in (SI Figure S2). The flow rate was maintained at 400 μL/min, with an injection volume of 5 μL and an autosampler temperature of 10°C.

Mass spectrometry was performed in positive mode using data-dependent acquisition (DDA). The MS survey range was set to 150–2000 Da, with an MS/MS switching ion intensity threshold of 6000 and a survey scanning rate of 0.2 scans per second in continuum mode. The MS/MS data range was set to 50–2000 Da, with a maximum of three ions selected for MS/MS per MS survey scan at a scan rate of 0.2 scans per second and an accumulated TIC threshold of 100,000. Peak detection was performed with a +1 charge state selection, capturing 30 ions per cycle. A tolerance window of ±0.5 Da was applied for peak identification, and a 3 Da extraction window was used for peak isolation. Trap collision energy was set with a collision energy ramp of 20–40 V for both low-mass (150 Da) and high-mass (2000 Da) ions

Lipid Annotation.

Peak detection, alignment, and lipid identification from the LC-MS data were performed using MS-DIAL (Version 5.5)³⁶, leveraging their in-silico LC-MS/MS lipidomics database. Lipids were identified based on MS2 spectral matching. All structurally (tandem MS) characterized lipids were exported and subsequently used to putatively identify MALDI MS peaks via custom Python scripts. Briefly, LCMS annotations were adjusted to their neutral mass by subtracting the adduct weight (e.g. H, Na, NH₄). Detected MALDI peaks were then systematically tested across this library of neutral weights by adding common MALDI adduct masses (H, Na, K) and checking for a ppm match less than 10. An overview of these annotations can be tablen in (SI Table S4).

Mass-Tag Cell Class Assignment.

Raw mass spectrometry data from the Bruker timsTOF fleX were converted to mzML using TIMSCONVERT⁴³ and processed in Python with pyOpenMS, focusing on MS1 spectra. A bin width of 0.01 Da was applied to each spectrum and normalized to the maximum aggregate intensity. Each cell was classified based on the a priori known m/z values of the mass tags, with each mass tag corresponding to a specific target marker. Several parameters were applied for cell classification, including the limit of detection (LOD), which was defined as two standard deviations above background noise. Background noise was determined from the baseline noise at the expected mass-tag location in cells lacking the target marker. Additionally, the presence of the isotopic peak distribution for each mass tag was used as another classification criterion. Spectra from cells containing both {target-marker} and {filter-marker} were excluded to isolate the target marker; however, both counts were included in the overall cell count analysis (SI Figure S3).

Python libraries NumPy, pandas, and Matplotlib were used for data handling and visualization of the average mass-tag profile for each sample, with an example shown in (SI Figure S4). Metadata was created to assign each lipid profile to its respective classified cell for lipidomic analysis.

Lipid Feature Extraction and Cell Classification.

Raw mass spectrometry data from a Bruker timsTOF fleX were directly read using functions from TIMSCONVERT⁴³, enabling direct extraction of MS1 data from Bruker .d files and conversion to m/z and intensity pairs via custom scripts in Python. Peak detection was performed using SciPy with a minimum threshold of 100. Selected m/z features were binned across the entire dataset via continuous binning with a 20 ppm bin width. The bins for each feature were recentered based on the weighted average of each peak within the bin and features detected in fewer than 0.1% of samples were removed. Cell classification metadata was integrated to lipid profile metadata to assign lipid features to specific cell types, enabling comparative lipid distribution analysis.

Statistical Analysis.

To identify lipid features distinguishing cell types and neuronal cell states, a nonparametric Wilcoxon rank-sum test was performed. For cell-type analysis, lipid feature intensities for each cell type were compared against all other cell types. For neuronal cell state analysis, the test was applied to compare lipid feature intensities between NPS and PS neuronal states. The resulting p-values were adjusted using the Benjamini-Hochberg method to control the false discovery rate. Lipid features with the lowest adjusted p-values were ranked, and the top-ranked lipids for each marker were selected for further analysis and representation in comparative plots.

RESULTS AND DISCUSSION

Cell-Targeting of SSMS.

Our SSMS approach is enabled by the identification of single cells via nuclei staining through fluorescence microscopy imaging, linking each cell to its corresponding lipid profile and mass-tag spectrum used to identify the antibody labeling the cell (Figure 2A–C). Cells isolated from the hippocampus were classified based on presence of glial fibrillary acidic protein (GFAP) for astrocytes, neuronal nuclei (NeuN) for mature neurons, and myelin basic protein (MBP) for oligodendrocytes. The expected isotopic distribution of each of these mass tags is provided in the (SI Figure S5). Mass spectrometry imaging (MSI) was performed to verify marker cell specificity (SI Figure S6–S7). We performed SSMS on 5,067 hippocampal cells from three animals, obtaining lipid profiles from all cells. Of these, 91 ± 4% retained their positions on the glass slide after matrix removal and mass-tag immunolabeling via ICC. After applying filtration criteria for cell-class assignment, 1,959 cells were included in the final MS analysis, representing approximately 39% of the total targeted cells. This includes cells for neuronal state analysis exhibiting cross-reactivity of neuronal markers: NeuN, synapsin-I (SYN-I), and neurofilament light (NFL), and compared to neuronal cells expressing only NeuN and NFL. The distribution of these marker-enriched cells, as visualized by UMAP of their lipid features, is shown in (SI Figure S8).

Figure 2.

Single cell classification using mass-tag markers and their lipidomic profiles. (A) Merged fluorescence and brightfield images of three marker-defined cells: astrocyte (top, GFAP+), neuron (middle, NeuN+), and oligodendrocyte (bottom, MBP+). (B) Lipid spectra for each cell, showing detected profiles. (C) Mass-tag spectra corresponding to each cell type, highlighting mass-tag-specific marker peaks from sequential analyses of the same cells displayed in (A).

Given the interconnected nature of brain cells, some cell fragments may adhere to others after single-cell extraction and isolation, potentially skewing lipid and immunolabeling results. To mitigate this, we excluded cells with antibody cross-reactivity, except for those used in neuronal state analysis. While co-expression of certain markers is expected in hippocampal progenitor cells, it is not the focus of our study. Given potential issues with cross reactivity from sampling, these cells were filtered from the dataset. Details on excluded cross-reactive cells, along with the complete dataset, are provided in the Associated Content and Supplementary Information (SI Figure S3).

Comparative Lipid Analysis of Neurons, Astrocytes, and Oligodendrocytes.

To explore lipidomic differences among classified cell types, we performed a comparative analysis of lipids enriched in neurons, astrocytes, and oligodendrocytes. These lipids were annotated through peak matching between MALDI and LC-MS/MS, unless otherwise noted.

Among the identified lipid features in neurons, we observed phosphatidylcholine (PC) species that align with previous studies including nano-LC MS single hippocampal neuron lipidomics³⁸ and MALDI-ICC analyses²⁴. For example, PC species, such as PC 32:0 and PC 34:1 (Figure 3A), as well as PC 36:2 and PC 40:6 (Figure 3B), were among the top differentially enriched lipids in neurons when compared to astrocytes, as reported by Neumann et al.²⁴ Similarly, several lipids in the hippocampal astrocytes, identified using our SSMS approach, correspond to the lipid species from the ICC dataset reported by Xie et al.²⁵, including PC O-32:1, PE P-38:4, and HexCer 32:1:O2 (SI Table S5). These alignments further support the ability of SSMS to capture biochemical heterogeneity with cell-type specificity in a similar fashion as traditional fluorescence-based MALDI-ICC.

Figure 3.

Comparative analysis of lipid profiles across cell types. (A) Subtracted average mass spectra comparing lipid profiles between neurons, astrocytes, and oligodendrocytes, highlighting key lipid ion differences. (B) Ranked dot plot of the top five lipid features distinguishing each cell type. Dot color represents the z-scored average intensity of each lipid, and dot size reflects the proportion of cells in that group expressing the lipid. N = 533 neurons, 404 astrocytes, and 366 oligodendrocytes, collected from three animals. Lipids annotated in black were confirmed by MS/MS, while those in red represent putative identifications from the LIPID MAPS^® based on accurate mass.

While previous studies primarily focused on neuronal and astrocytic lipidomics, our SSMS approach enables a more comprehensive cell-classified analysis. To demonstrate, we examined lipidomic differences between astrocytes and oligodendrocytes, as well as between neurons and oligodendrocytes. One key finding in our data was the differential distribution of hexosylceramides (HexCer), a class of lipids essential for myelin formation. These lipids are highly abundant in oligodendrocytes, where they contribute to neuronal support and axonal insulation. While previous studies have characterized the role of HexCer in oligodendrocytes⁵, our analysis identifies distinct HexCer species associated with specific cell types. For instance, statistical analysis revealed that HexCer 42:2;O2 was enriched in oligodendrocytes (Figure 3B), whereas mass spectral subtraction allowed us to observe the higher ion abundance of Hex2Cer 32:0;O2 in oligodendrocytes compared to astrocytes (Figure 3A). In contrast, HexCer 32:1;O2 was more abundant in astrocytes than in neurons (Figure 3A), highlighting cell-type-specific differences in sphingolipid metabolism.

In addition to HexCer, we also observed significant PC species—lysophosphatidylcholine (LPC), oxidized phosphatidylcholine (PCO), and PC—enriched in oligodendrocytes, reflecting their physiological role in myelin synthesis and membrane maintenance. Oligodendrocytes require a high supply of PC and PCO for building myelin sheaths, while LPC serves as a precursor in phospholipid metabolism. These findings further highlight the lipidomic specialization of oligodendrocytes in supporting neuronal function and maintaining hippocampal white matter integrity.

While the association of hydroxylated sphingomyelin (SM) species with specific brain cell types is not yet fully understood, our findings show a significant enrichment of SM 36:1;O2 in astrocytes. Given their role in lipid homeostasis^6,7,11 and support for myelination³⁴, this finding, in conjunction with previous studies^5,35 highlights astrocytes’ involvement in sphingomyelin metabolism, suggesting that astrocytes may process these lipids and facilitate their transfer to neurons and oligodendrocytes.

Furthermore, the hierarchical clustering of lipid profiles highlights lipidomic similarities between astrocytes and oligodendrocytes, reflecting their shared glial lineage and overlapping lipid metabolism. However, their distinct lipid enrichment patterns suggest functional lipidomic heterogeneity, as each cell type exhibits unique dominant lipid species (Figure 3B).

Lipid Differentiation Between Pre-synaptic and Non-Presynaptic Neuronal States.

To investigate significant lipid differences between neuronal states, we identified the top lipid species among two classified cell states. Cells positive for NeuN and NFL but lacking SYN-I expression were classified as non-presynaptic (NPS) neurons, while those expressing all three markers—NeuN, NFL, and SYN-I—were classified as presynaptic (PS) neurons. SYN-I, a synaptic vesicle-associated protein, is exclusively found in presynaptic neurons⁴⁰, allowing us to identify distinct lipid species between these two neuronal populations.

The differential lipid analysis shows distinct lipid compositions between PS and NPS states. Notably, four of the six most significant lipids in the PS state belong to the phosphatidylethanolamine (PE) class, whereas the top six significant lipids in the NPS state represent a more heterogeneous mix of lipid classes (Figure 4A). These findings reinforce the critical role of PEs in membrane curvature and vesicle fusion, both of which are essential for synaptic function.^37,38

Figure 4.

Differential lipid features between non-presynaptic (NPS) and presynaptic (PS) neurons. (A) Top six lipid species that significantly distinguish non-presynaptic (NPS, N = 216) from presynaptic (PS, N= 440) neuron states. (B) Bar plots showing the average lipid intensity of the top two differentially expressed lipid species per neuronal state. Statistical significance was determined using the Wilcoxon rank-sum test (*** p < 0.001). Error bars represent the standard error of the mean (SEM), calculated from individual cells per group. N = number of cells from three animals. Lipid annotations were confirmed by MS/MS.

Interestingly, we identified plasmalogen PE, specifically PE P-38:4, as one of the top two lipids distinguishing PS from NPS neuronal states (Figure 4B). Plasmalogen PEs have been reported to enhance membrane fusion and protect against oxidative stress, and their upregulation has been linked to AD, suggesting a potential biomarker role in AD.⁴³ Along with PE P-38:4, PE 40:4 was also among the top two significant lipids in PS neurons (Figure 4B), and both were also the top lipid features in astrocytes. Given the well-established role of astrocytes in neuronal lipid metabolism, our findings suggest that astrocyte-derived lipids may play an important role in supporting synaptic function in the PS state.

Additionally, we observed a preferential enrichment of sphingomyelin (SM) 36:2;O2 in NPS neurons (Figure 4B), which may reflect differences in membrane dynamics and fluidity. The presence of an additional double bond in the acyl chain of SM 36:2;O2 could contribute to a more fluid and flexible membrane structure compared to SM species with a single double bond, such as SM 36:1;O2, which shows higher relative ion abundance in astrocytes and may be associated with a more ordered and rigid membrane structure. However, further functional studies are needed to elucidate their precise role in synaptic physiology.

Beyond these biological findings, future studies integrating both positive and negative ion modes could provide a more comprehensive lipidomic profile. Although we demonstrated this method on a MALDI-QTOF-MS platform, the SSMS approach can readily be adapted to advanced MALDI-MS technologies. Since this workflow has been optimized on the timsTOF fleX MALDI-2 platform, several additional features could further enhance lipidomic investigations, such as the use of post-ionization laser⁴⁴ to detect low-abundance lipid species, MALDI-TIMS-MSI for isomeric separation, or MALDI-FT-ICR-MS to resolve isobaric species.²¹ Furthermore, with emerging technologies such as transmission-mode MALDI (t-MALDI)⁴⁵, it may be possible to integrate SSMS with t-MALDI, combining the microMS-guided laser targeting capabilities with the precise laser and imaging geometry of t-MALDI. This could eliminate the need for coordinate conversion and fiducial-based alignment, enabling fully automated, high-throughput single-cell analysis. Finally, future integration with PRescan Imaging for Small Molecule (PRISM)-MS⁴⁶ could expand SSMS to small-molecule applications.

CONCLUSIONS

In this work, we introduce SSMS as a two-step approach that integrates untargeted MS with antibody-targeted protein detection via PCMT probes. Using five immunomarkers, we classified mature neurons, astrocytes, and oligodendrocytes and identified neuronal subpopulations based on their PS and NPS states.

Our lipidomic analysis revealed distinct lipid species across these cell types and cell states, highlighting differences in HexCers, SMs, and various class of phospholipids including PEs and PCs. These findings demonstrate heterogeneity in lipid metabolism and functional specialization among hippocampal cell populations, emphasizing the importance of single-cell approaches for resolving cell-specific metabolic differences.

Due to cell isolation constraints, certain lipid classes may be underrepresented, highlighting the need for improved isolation procedures. Furthermore, while tissue dissociation into single cells inherently leads to loss of information on the native spatial organization of cells, future work could complement single-cell measurements with imaging of adjacent tissue sections to help correlate single-cell lipid profiles with their approximate spatial context. Despite this limitation, our findings emphasize the lipidomic diversity within glial cells and neurons, as well as lipidomic heterogeneity among neuronal subpopulations in the hippocampus. By identifying specific lipid species enriched in distinct cell types and cell states, this work provides a foundation for future studies investigating lipid-driven cellular functions, with potential relevance for biomarker identification and lipid-targeted therapeutic strategies.

Supplementary Material

Supporting Information

The Supporting Information is available free of charge on the ACS Publications website.

Miralys PCMT probes used for cell classification and timsTOF fleX MALDI-2 instrument parameters. MALDI-MSI ion images with overlaid Hoechst nuclear staining, an UpSet plot summarizing the distribution of targeted cells, and an example of cell-classified average mass spectrum. Cell retention fluorescence images, an LC-MS chromatogram of lipid extracts, lipid annotation from LC-MS/MS and MALDI-MS peak matching, and a comparison of lipids with literature data (PDF).

Accession Code: The codes used in this study are publicly available on GitHub (http://github.com/marisa-asadian)

Data supporting this study are publicly available from University of Illinois Data Bank: (https://doi.org/10.13012/B2IDB-9244584_V1)