Profiling lamina specific pyramidal neurons using postmortem human formalin fixed paraffin embedded frontal cortex tissue in combination with digital spatial profiling

Abstract

Background

Digital spatial profiling (DSP) is an innovative approach to perform RNA sequencing (RNA-seq), including in neuronal populations. DSP enables expression profiling linking RNA-seq data to spatially characterized samples utilizing tissue bound probes. We employ the GeoMx DSP system for spatial characterization of transcriptomic data from lamina specific pyramidal neurons and cortical ribbons containing admixed cell types using human postmortem brain tissue.

New method

We established a protocol using human postmortem formalin fixed paraffin embedded (FFPE) frontal cortex tissue from nondemented human control brains. Layer III (L3) and Layer V (L5) pyramidal neurons from Brodmann area 9 were identified with the neuronal marker Ca²⁺/calmodulin-dependent protein kinase II and selected for probe collection.

Results

This approach significantly reduced the amount of FFPE tissue needed for robust single population RNA-seq. We demonstrate ~20 identified L3 or L5 pyramidal neurons or one lamina-specific cortical ribbon from a single 5 μm thick section is sufficient to generate robust RNA-seq reads. Bioinformatic analysis of neurons and ribbons showed notable similarities and differences reflective of the single neuron and multiple admixed cell types within the former and latter, respectively.

Comparison with existing methods

Protocols exist for DSP of postmortem human FFPE brain tissue. However, this new approach enables profiling small groups of ~14–21 pyramidal neurons using the GeoMx DSP platform.

Conclusions

This optimized DSP assay provides high resolution RNA-seq data demonstrating utility and versatility of the GeoMx platform for individually characterized neurons and isolated cortical ribbons within postmortem FFPE human brain tissue for downstream analyses.

Graphical Abstract

1. Introduction

In the past decade, technical innovations have resulted in higher resolution methodologies to investigate genomic profiling in healthy and diseased tissues. RNA and protein expression profiling are necessary to understand the molecular underpinnings of cellular functions and mechanisms of activation. By performing profiling studies across a degenerative phenotype, researchers can identify novel targets for therapeutic intervention that are perturbed during aging and disease. This is especially important in the aging brain due to the complex cytoarchitecture of multiple cell types and innervation patterns that degenerate in neurodegenerative disease states, often coinciding with neuroinflammation and deterioration of brain circuits associated with memory and/or executive function (Tanaka et al., 2020, Liu et al., 2013b, Sullivan et al., 2018). Neuronal and glial subtypes are highly specialized with unique functional activation patterns based on spatial and temporal localization, underscoring the need for expression profiling of individual cells and/or spatially resolved cell populations (Gala et al., 2023, Yao et al., 2023, Engel et al., 2021). Several technologies exist that elucidate cellular profiles, but few can be applied to link spatial localization with gene expression in the brain. Further, these approaches are often limited due to restrictions associated with sample storage and/or fixation, particularly in postmortem human brain. Traditional and currently used RNA sequencing (RNA-seq) technologies (Johnson et al., 2020), including microarray analysis (Ginsberg and Che, 2004, Kumari et al., 2022, Chen et al., 2004), have greatly contributed to profiling of brain tissue especially in conjunction with newer tissue isolation techniques and next generation sequencing approaches (Alldred and Ginsberg, 2023, Ginsberg et al., 2010, Chen et al., 2004, Civita et al., 2019). However, a major drawback of these earlier methodologies is the lack of cell specificity, which have shifted cellular analysis towards next generation technologies (Alldred and Ginsberg, 2023) that allow for profiling of particular cell types (Mantione et al., 2014).

Single cell RNA sequencing (scRNA-seq) is a method which provides cell specific information and detects high resolution details such as single nucleotide polymorphisms (SNPs) and noncoding RNAs (Alldred and Ginsberg, 2023, Wang et al., 2009). Starting material for scRNA-seq is fresh tissue (Piwecka et al., 2023, Li and Wang, 2021) that must be processed and sorted for cells by varying dissociation methods (Haque et al., 2017, Mattei et al., 2020, Liu et al., 2021). Similarly, single nuclei RNA-seq (snRNA-seq) allows for the isolation of individual nuclei from previously frozen tissue (Kihara et al., 2022) while permitting simultaneous processing of multiple samples from brain banks and other sources, which eliminates errors from batch effects or processing differences. However, a major caveat is the genomic data collected is limited to nuclear expression, excluding RNA localized to synaptic or cytoplasmic regions, including mitochondrial gene expression (Sheng and Cai, 2012). Though scRNA-seq and snRNA-seq can provide profiles of each cell type, the large number of cells needed for sequencing resolves as an average genetic signature of each cell type in the originating tissue specimen and cannot be pinpointed to a small population of spatially characterized cells (i.e., dozens of cells) within a region.

Laser capture microdissection (LCM) can identify cells for higher resolution spatial information in sequencing from user defined cells microisolated from individual tissue sections. This method allows isolation of specific cell populations from fresh or fixed brain tissue but typically requires hundreds of cells, such as neurons or microglia, for sequencing (Alldred and Ginsberg, 2023, Alldred et al., 2021, Alldred et al., 2024, Liu et al., 2013a). An entire structure can be dissected for specific neuron types via LCM, but the quantity of cells needed limits spatial resolution to a structure or region of tissue. Other caveats of this technique are that it is labor intensive and the area of interest needs to be sufficiently large and/or cell type of interest must be abundant for the sample to reach sequencing quality control standards. Techniques such as patch-seq profiling of single neurons have been validated but are also labor intensive and have relatively low throughput (Cadwell and Tolias, 2024, Qiu et al., 2012, Lipovsek et al., 2021).

Rigorous, high throughput profiling of spatially characterized neurons is needed to elucidate spatial and cell specific gene expression that may be linked to disease or circuitry effects. For example, high throughput gene profiling of spatially identified neurons could provide mechanistic insight into neurodegenerative disorders including Alzheimer’s disease (AD) and Alzheimer’s disease related dementias (ADRD), among others (Alldred and Ginsberg, 2023) using animal/cellular models as well as postmortem human brain tissues. A drawback to performing RNA-seq in the brain is the difficulty of obtaining well-characterized human postmortem tissue, especially forebrain tissues associated with memory or executive function circuit regions (Leh et al., 2009, Takeuchi et al., 2012, Snow, 2016), including the hippocampus, neocortex, and hypothalamus, among others. The limited tissue required for digital spatial profiling (DSP) may potentially ameliorate this confound. DSP combines single cell or single population RNA-seq with spatial profiling, using immunohistochemical labeling to specifically demarcate cells and/or regions (e.g., lamina or nuclear structures) of interest (Wang et al., 2023, Zollinger et al., 2020, Walker et al., 2024).

DSP technologies were developed based on in situ hybridization protocols (Jensen, 2014). RNA probes of known sequence are bound to discrete areas on tissue sections mounted on glass slides, which can then be identified by a fluorescent label and subsequently collected from identified areas and sequenced (Hernandez et al., 2022, Zollinger et al., 2020). The GeoMx DSP platform (Bruker Spatial Biology, Billerica, MA) utilizes in situ labeling technology from a single 5 μm thick section of formalin fixed paraffin embedded (FFPE) tissue for probing (Zollinger et al., 2020, Merritt et al., 2020) in conjunction with histochemical profiling of individual cells. The procedure involves i) unmasking fixed RNA in tissue to enable access for system specific probes, ii) labelling targets (herein, pyramidal neurons) for probe extraction, iii) extracting probes from selected regions of interest, and iv) preparing libraries from probes for high throughput sequence analysis. Probes are cleaved from user drawn regions of interest (ROIs) and collected for amplification and sequencing, which allows for the spatial identification of each sequenced ROI. The current GeoMx platform is capable of profiling up to ~0.5 mm² with the ability to segment selections within the working area providing high resolution data for downstream applications.

The standard GeoMx protocol is validated for various types of tissues, although currently, most publications are limited to the cancer field (O’Rourke et al., 2024, Moffet et al., 2023, Bergholtz et al., 2021), as the target tissue provides abundant cell types of interest that can be readily isolated through ROIs. To date, studies in postmortem human brain using the GeoMx platform to profile specific neuron types via RNA-seq are lacking. We address this unmet need, through the generation of a method enabling rigorous and reproducible profiling of layer III (L3) and layer V (L5) pyramidal neurons within their respective laminae in human postmortem FFPE frontal cortex brain tissue for gene expression analysis and downstream bioinformatic inquiry.

2. Materials and Methods

2.1. FFPE tissue preparation and processing

Human postmortem brain tissues were obtained from the National Disease Research Interchange (NDRI, Philadelphia, PA). Tissues were fixed and stored at room temperature (RT) in 10% formalin from time of acquisition until dissection for the GeoMx protocol. Brodmann area 9 (BA9) frontal cortex was dissected in the laboratory from (n=8) brains (Table 1). Samples were processed with the Leica ASP 300S Tissue Processor (Leica, Buffalo Grove, IL) as described in the Leica ASP300 S Operating Manual (Leica, V1.2) and embedded in paraffin using the Shandon Embedding Center (Shandon Inc., Pittsburgh, PA).

Table 1.

Human postmortem tissue.

Case	Diagnosis	Sex	Age	Postmortem Interval	Race
421	CTR	F	67	16.8	W
429	CTR	M	70	19.9	W
434	CTR	M	51	22.4	W
440	CTR	M	51	13.1	W
435	T2DM	M	61	18.4	W
436	CTR	M	39	22.4	n/a
438	CTR	F	47	9.6	W
439	CTR	M	52	22.8	W

Age in years. Postmortem interval in hours. Control, CTR. Type II diabetes mellitus, T2DM. White, W. Not available, n/a.

2.2. Tissue sectioning

RNase-free conditions were maintained throughout the protocol using RNase Away (7005–1, ThermoFisher, Waltham, MA) for RNase deactivation of all surfaces and equipment. A 1% gelatin subbing solution was prepared by dissolving 6 g of gelatin and 0.2 g chromium potassium in 600 mL of distilled deionized water (ddH₂O) at ~55 °C followed by filtering (0.10 μm PES, 229724, Celltreat, Pepperell, MA). Glass slides (Superfrost Plus Microscope Slides, ThermoFisher) were coated with the gelatin solution for 30 sec at ~35–40 °C then dried vertically for 24–48 h at RT under low lint tissue paper (34133, Kimtech Kimwipes, Kimtech, Irving, TX). Paraffin embedded tissues were sectioned at 5 μm thickness on a microtome (Finesse Me, ThermoShandon, Cambridge, UK) and mounted using a water bath (Premiere XH1003 Tissue Floating Bath, Ted Pella, Inc., Redding, CA) with autoclaved ddH₂O at 35 °C onto gelatin coated glass slides. Each slide received one tissue section from a total of 2–4 brains, depending on the tissue area, in the “scan area” as defined in the GeoMx DSP Manual Slide Preparation User Manual (GM-DSP MSPUM; Bruker Spatial Biology, MAN-10150–04). Mounted slides were placed vertically in a slide rack and incubated in a hybridization oven (HS9360, Hybaid, Franklin, MA) for ~16–24 h at 37 °C. Immediately prior to tissue processing and probing, sections were baked at 60 °C for 30 min.

2.3. Slide preparation

The GM-DSP MSPUM reagent list was followed for each experiment apart from xylenes (214736, Sigma-Aldrich, St. Louis, MO), 10% neutral buffered formalin (NBF; SF100–4, Fisher Chemical, Pittsburgh, PA,), 20X SSC Buffer (AM9763, Ambion, Austin, TX), and formamide (A9342, Ambion). A hybridization tray sealed in an airtight plastic bag was used in place of a humidity tray and hybridization chamber.

A Ca²⁺/calmodulin-dependent protein kinase II alpha primary antibody (CaMKII-α; 6G9; 50049, Cell Signaling Technology, Danvers, MA; 1:50 dilution) was utilized to identify pyramidal neurons (Li et al., 2022, Nicole and Pacary, 2020). The GM-DSP MSPUM was followed as recommended for brain tissue, intended to validate immunostaining, no probing was completed on this tissue. Therefore, all solutions were prepared in ddH₂O instead of in DEPC treated water (DEPC H₂O) which is recommended for sterility when collecting probes. CaMKII-α (1:50 dilution) was incubated on slides overnight at RT followed by one wash in 2X saline sodium citrate (SSC) at RT. Tissues were incubated in secondary Donkey anti-Mouse IgG (H+L) Highly Crosslinked Secondary Antibody, Alexa Fluor Plus 647 (A28175, Invitrogen, 1:1000) for 2 h at RT.

The GM-DSP MSPUM was employed with the following adjustments made to maintain tissue integrity throughout slide preparation. All solutions were prepared fresh in DEPC H₂O (9613, Calbiochem, Darmstadt, Germany). A water bath was used in place of a steam cooker and RNA exposure was completed using proteinase K solution (100005393, 20 mg/mL stock, Invitrogen) diluted at 1:1,000 for 5 min. For DSP, 2 slides with 2 different pairs of samples (n=4 total) were prepared for RNA probing and followed by probe collection using the GeoMx Digital Spatial Profiler (Bruker Spatial Biology). Probing was completed using GeoMx Whole Transcriptome Atlas Human RNA Probes for NGS (hWTA kit; 121401102, Bruker Spatial Biology). Probes from 33 ROIs were selected from 3 of the 4 mounted samples.

A second probe collection from 2 additional slides each containing 1 section from 4 samples was completed with further adjustments to the GM-DSP MSPUM protocol as detailed herein, modifying the epitope retrieval and RNA exposure. Between deparaffinization and epitope retrieval, slides were placed directly into pre-warmed 1X Tris EDTA at 99 °C for 10 min. Slides were incubated for RNA exposure in proteinase K solution (1:2,000) at 37 °C for 5 min. These changes were made to maximize tissue quality and adherence for probe collection, to increase number of collected ROIs and reach sequencing quality control standards for the NovaSeq 6000 workflow. With these adjustments, probes from 94 ROIs were collected from 6 sections representing 3 of the 4 mounted samples (n=3 brains). None of the brain specimens probed were replicated in the final ROI collection between the 33 ROIs and 94 ROIs (n=6 total brains with ROIs collected).

2.4. ROI selection

The GM-DSP MSPUM refers to entire geometric regions from which probes are collected as ROIs and segmented regions within drawn ROIs as Areas of Interest (AOI). For this study, all collected areas are referred to as ROIs. Unsuccessfully stained tissue was defined as areas where structures or cells could not be positively identified. These areas were excluded from ROI selection (Supplementary Fig. 1). Varying intensity of staining across processed tissues dictated the type of ROIs collected. Robust and evenly stained tissue (Fig. 1A) allowed for proper selection of pyramidal neurons while weaker (Fig. 1B) and no staining (Fig. 1C) precluded ROI collection. A total of 3 categories of ROIs were defined: segmented neurons (Fig. 1D), outlined neurons (Fig. 1E), and ribbons (Fig. 1F). Segmented neurons were defined using the segmentation feature on the GeoMx and included 14–21 immunostained neurons per ROI. Segmentation adjustment in the GeoMx DSP Control Center (GEOMX-0478; Version 3.1.0.222) was based on staining intensity, limiting probe collection to positively identified CamKII-α-immunoreactive pyramidal neurons (Fig. 1D). To include areas where staining intensity was not sufficient to segment neurons using the GeoMx DSP Control Center, geometric regions were drawn tightly around neurons identified by the experimenter with areas between identified neurons narrowly connected to minimize inclusion of adjacent tissue (Fig. 1E). Ribbon ROIs were defined as elongated rectangular regions drawn using the freehand tool in the GeoMx ROI selection window, identifying cortical L3 or L5 (Fig. 1F). In the cortical ribbons, probes included sampling of admixed cell types, including pyramidal neurons. Probe collection and plate finalizing were executed as described in the GeoMx DSP Instrument User Manual (MAN-10152–04).

Figure 1.

Immunostaining of tissue with an antibody against CaMKII-α and drawn ROIs. A. Robust and even staining intensity of pyramidal neurons (white arrows). B. Staining sufficient for visual identification of neurons, but not for segmentation by GeoMx software. Pyramidal neurons are indicated by white arrows. C. Staining intensity below the level of detection. D. Top tier segmented neuron ROI. E. ROI of manually outlined neurons. F. Ribbon ROI including all cell types and tissue of selected layer. Scale bars: 100 μm.

2.5. Library preparation and sequencing

Upon completion of probe collection, the GeoMx DSP NGS Readout User Manual (MAN-10153–04) was followed for library preparation. Two different GeoMx Seq Code primer plates (“A” and “B”) were used for the first (33 ROIs) and second probe (94 ROIs) collections to allow for later pooling. Following amplification, library cleanup was completed according to each of the steps in the NGS readout manual using AMPure EXP beads (A63881, Beckman Coulter, Brea, CA). After preparation of each collection, library quality control was performed on a TapeStation 4150 (Agilent, Santa Clara, CA) with the High Sensitivity D1000 kit (Agilent) to determine the concentration and size of amplified probes. A range of 150–190 base pairs (bp) was defined to identify probes of ~160 bp length and overamplified probes >170 bp, according to the NGS Readout Manual. The library for the first collection was diluted according to manual specifications for cDNA measurement, which failed. Library quality control measurement was then modified to be performed without dilution, for first and second probe collection. Libraries were pooled with the first collection undiluted (0.2 nM) and the second collection at the 2 nM concentration as recommended by New York University Grossman School of Medicine Genome Technology Center (NYUGSOM GTC) for sequencing.

Pooled libraries were assayed on a NovaSeq 6000 system (Illumina, San Diego. CA) with the SP 100 cartridge kit and SP flow cell (20028401, Illumina). The same pool was split into 2 tubes and seeded onto the flow cell using the Illumina v1.5 lane XP lane separation kit (20043130) due to volume. The sequencing metrics were run as paired end 50 bp reads.

2.6. Data processing and bioinformatics

FastQ files for each ROI were processed through the GeoMx NGS Pipeline version 3.1.1.6 and converted to DCC files as described in the GeoMx DSP NGS Readout User Manual. DCC files were then uploaded to the GeoMx DSP Analysis Suite and bioinformatics were performed as follows. Segment tags were assigned to identify each ROI type (Fig. 1D-F). Ribbons in which the entire ROI surface area was highlighted (Fig. 1F) were included in data processing. Due to varying staining levels (Fig. 2A–C), a subset of the total collected ROIs was tagged and selected for data processing, with the criteria for inclusion as follows: i.) segmented neurons were included if pyramidal neurons were clearly highlighted by GeoMx software (Fig. 1D). ii). No ROIs of uneven staining/ intensity (Fig. 2A) or with high background intensity (Fig. 2B) were included when the software did not limit capture to pyramidal neurons (Fig. 2D–E). iii.) No ROIs of weakly stained tissue (Fig. 2C) were included if neuron capture was inaccurate (i.e., excluding areas of neurons or including surrounding tissue) (Fig. 2F).

Figure 2.

Representative images of ROIs excluded from analysis. A. Uneven staining/intensity. B. High background intensity. C. Uneven segmenting across ROIs. D. Inclusion of neurons and background in the segmenting process. E. Stained tissue at or below the limit of detection. F. Capture of neurons including surrounding tissue area for downstream analysis. Scale bars: 100 μm.

As described in the GeoMx DSP Data Analysis User Manual (MAN 10154–01), default values were used for segment quality control (QC), biological probe QC, scaling to surface area, and finally normalization to negative probes. Nuclei counts were not included in quality control due to inaccurate counts reported by the software; this inaccuracy was confirmed by visual investigation of stained tissue sections.

2.7. Statistical analysis

One-way ANOVA proceeded by Tukey’s multiple comparison test was used to compare the means of normalized ROI count groups. Statistical analysis was completed using GraphPad Prism version 10.4.0.

3. Results

3.1. CaMKII-α immunoreactivity is a robust morphological marker for pyramidal neurons with GeoMx

The GeoMx platform requires strong and clear immunoreactivity with low background to clearly select ROIs (Fig. 1D–F) and segment for cells of interest. The CaMKII-α monoclonal antibody produced robust staining to differentiate laminae in BA9, identifying L3 and L5 (Fig. 3A). Further, the staining protocol produced robust fluorescent signal in which individual pyramidal neurons could be clearly observed (Fig. 3B).

Figure 3.

A. Lamina and excitatory neuron localization within human postmortem FFPE tissue target layers L3 and L5 (white lines). Scale bar: 1 mm. B. Robust pyramidal neuron staining by morphology marker CaMKII-α in green along with staining in blue with nuclear marker SYTO13. Scale bar: 100 μm.

3.2. Diluted probe libraries do not provide adequate library concentration measurements

Quality control analysis was performed to ensure sufficient quantity and quality of library from each sample was collected and amplified prior to sequencing. Concentrations were measured (Fig. 4) within the 150–190 bp range. The first probe trial was diluted and measured for cDNA library, exhibiting an average size of 177 bp at 26.5 pmol/l, which represented 4.67% of the total sample measurement. The concentration of the library in the 150–190 bp range was 2.93 pg/μL. The corresponding diluted no template control (NTC) averaged 173 bp at 5.62 pmol/l, representing 0.33% of the total NTC (Supplementary Fig. 2A–B) and a concentration of 0.293 pg/μL. An undiluted aliquot of the first library suggested the diluted library was inaccurately measured, with an average 175 bp average size but 193 pmol/l concentration, which represented 25.95% of the total sample measurement. The concentration of the undiluted probe collection was 21.8 pg/μL. The corresponding undiluted NTC had an average size of 172 bp at 44.1 pmol/l representing 4.36% of the library (Supplementary Fig. 2C–D) and was 4.68 pg/μL concentration. The second library was measured undiluted, with a concentration of 2,730 pmol/l and an average size of 174 bp in the 150–190 bp range making up 58.08% of the library. The concentration within the size range was 307 pg/μL. The corresponding NTC was an average size of 177 bp at 66.2 pmol/l molarity representing 7.16% of the library (Supplementary Fig. 2E–F) and having a concentration of 7.45 pg/μL.

Figure 4.

Library concentrations. Concentration peaks of A. Diluted first collection library, B. Enlarged image of 150–190 bp peak, C. Undiluted first collection library, D. Enlarged image of 150–190 bp peak, E. Undiluted second collection library and F. Enlarged image of 150–190 bp peak.

3.3. Greater number of ROIs enables rigorous RNA-seq

Initial probe collection library of 33 ROIs did not meet sequencing concentration standards, therefore limited data was available. However, the subsequent 94 collected ROIs was of sufficient concentration for sequencing and produced robust raw reads for all collected ROIs. Even though all the ROIs from the second probe trial provided sequencing information, not all ROIs were representative of the target tissue due to uneven tissue quality and immunohistochemical staining. Sequencing probes from ROIs that did not meet inclusion criteria were not included in analysis (Table 2).

Table 2.

Sequencing quality control.

	Segmented Neurons	Outlined Neurons	Cortical Ribbons
Number of ROIs	9	19	54
Mean Surface Area	7,589	6,708	108,000
Mean Raw Reads	8,270,255	4,759,849	11,886,797
Mean Aligned Reads	7,457,603	4,313,095	10,956,616
% Alignment	91.0%	91.3%	92.7%
Mean Sequencing Saturation	95.3%	90.4%	94.7%
Mean Number of Normalized Counts Per ROI	43,479	31,083	97,769

A total of 9 ROIs were identified as representative samples of segmented neurons. The selected ROIs had an average area of 7,589 μm² yielding 8,270,255 average raw reads, and an average of 7,457,603 aligned reads with 91.0% alignment. The sequencing saturation for these ROIs was 95.3% with an average of 43,479 normalized counts per ROI (Table 2). In comparison, 21 ROIs were selected to represent outlined neurons, with a mean surface area of 6,820 μm² and an average of 5,057,385 raw reads, 4,581,838 average aligned reads with 91.3% alignment, 90.7% average sequencing saturation and 20,300 normalized counts per ROI (Table 2). Cortical ribbons had the highest number of included ROIs at 54 with a mean surface area of 108,000 μm². Ribbons averaged 11,886,797 raw reads, 10,956,616 average aligned reads at 92.7% alignment, 94.7% sequencing saturation, and 97,769 average normalized counts per ROI (Table 2).

3.4. Normalized counts differ between ROI types

Total normalized counts from each ROI type were compared (Fig. 5). Data analysis was completed using a one-way ANOVA (F = 85.44, P < 0.0001, R² = 0.6839), comparing the means of each group followed by Tukey’s multiple comparison test. Cortical ribbon normalized counts were significantly higher than those of segmented and outlined neurons (Tukey adj. P < 0.0001) for comparisons. Tukey post-hoc comparison showed no difference between counts of segmented and outlined neurons (Tukey adj. P = 0.2717).

Figure 5.

Comparison of normalized probe counts in segmented neurons, outlined neurons, and cortical ribbon ROIs.

4. Discussion

We developed a modified approach to optimize the GeoMx DSP System for use with postmortem human FFPE brain tissue affixed to glass slides to specifically profile cortical laminae and excitatory neuronal populations, using as little as 14 cells from a homogeneous neuronal population as well as cortical ribbons that include multiple admixed cell types. Studies utilizing the GeoMx system have been published for brain tissue but focus on protein and RNA of glial cells and neurons that were not isolated from surrounding tissue by segmentation or outlining, rather included in an ROI of admixed cell types (Noll et al., 2022, Walker et al., 2024, Walker et al., 2022, Westfall et al., 2022, Lee et al., 2022, Prokop et al., 2019, Valberg et al., 2023). Cortical neurons identified by neuronal nuclear protein or α-synuclein immunostaining have been identified and quantified through a DSP approach (Goralski et al., 2024). The present study uses DSP to profile postmortem human pyramidal neurons from two laminae (L3 and L5) with corticocortical and corticothalamic projections. We demonstrate the GeoMx DSP platform can be employed to elucidate gene expression from a relatively small number of spatially characterized pyramidal neurons in postmortem FFPE tissue, which is a common fixation and storage method that previously has been difficult to utilize in terms of generating sufficient high quality RNA for rigorous and reproducible RNA-seq (Zhao et al., 2019, Böning et al., 2024, Guerra et al., 2020). Future studies are being planned where downstream bioinformatic analysis can be performed to determine spatial and/or regional alterations in gene expression profiles within pyramidal neurons and cortical ribbons of admixed cell populations.

We detail assays optimizing the GeoMx system to: i) establish a morphology marker that produces robust and specific staining of pyramidal neurons, ii) maintain tissue integrity for probe collection, and iii) determine integrity of resulting sequencing data from individually characterized neurons and tissue ribbons. A challenge was variability in RNA quality and quantity levels in FFPE tissue samples from control subjects (Table 1). Many factors could contribute to variability, including postmortem interval from individual cases along with differences in fixation and storage conditions. We selected cases with no known neurodegenerative or neuropsychiatric disorders to initiate these optimization studies.

Paraffin embedding provides the advantage of relatively low cost and long-term efficient storage and preservation of postmortem brain tissues for long periods of time at RT (Sadeghipour and Babaheidarian, 2019) making this approach feasible for extended use compared to the handling and storage of unfixed fresh or frozen tissues. Uniform paraffin embedding allowed samples to be standardized for downstream analysis of tissue. The ability of the GeoMx system to work with FFPE tissue therefore provides an advantage over contemporary ‘omics approaches (e.g., scRNA-seq and bulk sequencing via LCM) by enabling staining and downstream capture workflows to be performed at RT. However, a caveat of utilizing postmortem brain FFPE tissue is that materials are often over fixed by prolonged exposure to standard fixatives such as neutral buffered formalin, impacting sample probing with antibodies or RNA probes for immunostaining and sequencing (Böning et al., 2024, Illumina, 2014), both critical steps of the GeoMx workflow.

Due to the inconsistencies in fixation of human postmortem tissue along with lot and manufacturer differences for antibody production (Luo et al., 2023), variability can exist in immunostaining. Even when selecting a robust morphological marker, such as the CaMKII-α antibody established herein for BA9 pyramidal neurons, epitope retrieval is requisite. Epitope retrieval and RNA exposure steps of the GeoMx protocol are harsh to the tissue section but are required to penetrate tissue to evenly immunostain cells and allow exposure of RNA species for probing. This complicates selection of ROIs as these steps affect tissue adherence and integrity, limiting ROI selection. Gelatin coating of slides was used to aid in tissue adherence. However, the GeoMx protocol produced uneven adherence of tissue sections, which complicates desired cell identification and microisolation for downstream analysis (Supplementary Fig. 3). Further, post epitope retrieval, variability in the staining intensity remained, often preventing appropriate segmentation of the desired cells/neurons by the GeoMx system, resulting in the need to generate multiple types of ROIs (Fig. 1D–F).

The initial probe collection assay resulted in limited tissue adhered to the slide with few areas of robust staining for neuron segmenting limiting ROI collection to 33 collected ROIs. As shown by library concentration measurements (Fig. 4A–B), library amplicon size was in the expected size range but lower in quantity than recommended in the NGS Readout Manual. Diluting the library returned a sequencing concentration below the quantitative range listed in the GeoMx protocol therefore the undiluted library was measured. Though the undiluted library measurement (Fig. 4C–D) was higher, this library still did not reach sequencing standards. The low concentration of the undiluted library likely reflects the low number of probes collected, rather than failure of the probes to amplify. Adjustments to the protocol, including heat induced and proteolytic epitope retrieval steps, were made improving tissue adherence and integrity. Following optimization, we were able to identify and collect 94 ROIs resulting in significantly higher concentration of the undiluted library (Fig. 4E–F), which met sequencing criteria. We recommend collecting as many ROIs as feasibly possible and measuring library quantity and quality with an undiluted sample when working with low surface area ROIs, including segmented or outlined neurons obtained from postmortem FFPE tissue or similar tissue sources.

Initial probe collection library did not meet sequencing standards, and unsurprisingly, sequencing results were not enabled. The subsequent probe collection, which met criteria, did produce robust sequencing results, which we could then stratify based on ROI type (Table 2). Importantly, ROIs were excluded based on their representation of the targeted pyramidal neurons. Exclusions were made to ensure that bioinformatic analysis was accurate, as non-representative probe collection due to poor staining likely indicates degraded RNA or limited somal area in the z-axis plane of the tissue section. After ROI exclusions, a difference in normalized counts was found between each of the three types of ROIs (Figs. 1, 5). Both segmented and outlined neurons had significantly less normalized counts than cortical ribbons, as expected due to the lower number of cells included in these ROIs and the smaller surface area. Neurons that did not stain sufficiently for segmenting by the GeoMx program had no difference in probe counts from those of segmented neurons. This suggests that manually outlined ROIs of pyramidal neurons are comparable to segmented neuron ROIs and may be pooled to build adequate sample sizes where they may not have been reached by segmenting alone.

Our modified methodology successfully characterized ~ 20 L3 or L5 pyramidal neurons (range 14–21 neurons) from spatially characterized human postmortem FFPE BA9 tissue using the GeoMx DSP system. Modifications of the GeoMx protocol enable preservation of the FFPE tissue throughout the protocol, while allowing probe accession resulting in genomic profiling of neurons and spatially characterized ribbons of tissue. Caveats of utilizing the GeoMx system include tissue processing variability and potential loss of RNA due to over fixation and improper segmentation. Further, profiling is limited to predefined sequences in the probe kit, preventing interrogation and potential discovery of novel sequences. However, the current GeoMx platform allows probes of ~19,000 genes using the hWTA kit, which was selected for the present study. Utilizing this probe kit allows for a wider selection of genes to be probed that may serve as internal controls for analysis. For example, a reasonable control gene for the present target tissue is myelin basic protein, which has been demonstrated independently to have higher expression in L5 compared to L3 (Tomassy et al., 2014). Additionally, the hWTA kit may be a platform for development and optimization of more specific probe kits as it may elucidate novel targets in brain tissues in the context of neurodegeneration, particularly in select neuronal populations, such as L3 and L5 pyramidal neurons, which are implicated in disease progression in a number of disorders, including AD dementia (Simic et al., 2014, Dharshini et al., 2021, Hof et al., 1990).

In conclusion, we successfully modified the GeoMx DSP technique to generate a genomic profile from as few as 14 spatially characterized neurons in FFPE fixed postmortem human brain tissue. We show this methodology allows for accurate and complete probe adherence to immunohistochemically labeled excitatory human neurons, which resulted in millions of reads that were linked to bona fide gene expression. The GeoMx platform is feasible for generation of accurate gene expression profiles from spatially characterized FFPE cortical tissue samples. We expect this revised GeoMx platform approach will expand our ability to examine human brain tissues under normative and disease conditions and determine dysregulation with spatial resolution in the human brain, which we posit will help accelerate discovery science in tissue samples that are abundant but underutilized.

Supplementary Material

1

Supplementary Figure 1. Unsuccessful staining of tissue section in which structures and cells cannot be identified. Scale bar: 1 mm.

2

Supplementary Figure 2. NTC Library concentrations. Concentration peaks of A. Diluted first collection NTC library, B. Enlarged image of 150-190 bp peak, C. Undiluted first collection NTC library, D. Enlarged image of 150-190 bp peak, E. Undiluted second collection NTC library, and F. Enlarged image of 150-190 bp peak.

3

Supplementary Figure 3. Tissue adhesion to glass slides prior to successful probe collection. Although postmortem FFPE BA9 tissue adhesion was not ideal, probe collection (A and B) was successful with the revised protocol resulting in the ROIs presented in Table 2.

Article Highlights.

• DSP was optimized for FFPE neurons and cortical ribbons from postmortem brains.

• 5 μm sections of human FFPE frontal cortex were used for GeoMx DSP and RNA-seq.

• Microisolation of ~20 L3 and L5 pyramidal neurons in frontal cortex enabled DSP.

• Robust staining and manual ROI selection permits DSP from FFPE pyramidal neurons.

Data Availability

Datasets supporting the conclusions of this article are available as follows. RNA-seq data analyzed within this study are available from GEO (www.ncbi.nlm.nih.gov/geo; GSE288628) or from the corresponding author upon request. ROI images for brain tissue are available from the www.synapse.org (syn64607994) portal or the corresponding author upon request.