ABCC9/SUR2 has a Complex Expression Pattern in Human Brain Gliovascular Unit Cells, Including Astrocytes

Abstract

The ABCC9 gene and its cognate protein SUR2 play important roles in neurovascular coupling and are implicated in hippocampal sclerosis of aging (HS-Aging). However, prior studies have not focused on human brain SUR2 expression or SUR2 in glial cells. Here we analyzed cell type-specific ABCC9/SUR2 expression patterns, and correlation with known genetic risk variants, using multiple data sets and a novel antiserum. Existing single-nucleus RNA sequencing data sets and new spatial transcriptomics data indicated that SUR2 transcripts were expressed primarily in human brain pericytes, astrocytes, smooth muscle cells, and endothelial cells. Evaluation of Sur2 expression in a sample of mice brains showed similar results except Sur2 was not detected in the mice astrocytes. In a SUR2-enriched subcluster of human astrocytes, the pattern of transcript expression suggested responsiveness to thyroid hormone signaling: SUR2-correlated gene products were enriched for thyroid hormone-sensitive transcripts and SLCO1C1, the astrocyte thyroid hormone importer, was the transcript with the strongest correlation with SUR2 expression. Cells in the SUR2 + astrocyte cluster tended to have been derived from individuals lacking severe Alzheimer’s disease pathology. An ABCC9 single nucleotide variant (rs1914361, also a HS-Aging risk allele) was associated with increased SUR2 expression in astrocytes, but not other cell types. SUR2 mRNA splicing differed between cell types; astrocytes preferentially expressed the SUR2B variant. A novel SUR2 antiserum immunolabeled blood vessel walls and some astrocyte-morphology cells in human brain. Overall, human brain SUR2 expression was enriched among different cell types of the gliovascular unit. A SUR2-enriched, possibly-homeostatic astrocyte subcluster suggested connections to thyroid hormone signaling.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s12031-026-02603-z.

Introduction

The ATP-binding Cassette, sub-family C member 9 (ABCC9) gene serves key cerebrovascular functions (Ando et al. 2022; Bowen et al. 2025; Hariharan et al. 2022; Isaacs et al. 2024; Mascarenhas et al. 2025; Rocha et al. 2020), but much remains unknown about its human brain cell-type specific gene expression. ABCC9 encodes sulfonylurea receptor 2 (SUR2) proteins. (We designate the gene “ABCC9”, the mRNA transcript “SUR2”, and the protein “SUR2”). SUR2 proteins are regulatory subunits of the ATP-sensitive potassium (K_ATP) channel (Nichols et al. 2013), facilitating sensitive ATP-dependent inhibition and ADP-dependent activation of an inwardly rectifying potassium channel (Minami et al. 2004; Nichols 2006). These special characteristics result in responsiveness to local metabolic conditions – the K_ATP channel opens at the plasma membrane when the intracellular ratio of ATP/ADP is low (Martin et al. 2023). Through this mechanism, SUR2 modulates various biologic processes, including neurovascular coupling (Ando et al. 2022; Bowen et al. 2025; Hariharan et al. 2022; Isaacs et al. 2024), stress responses (Taggart and Wray 1998; Yamada and Inagaki 2005), and ischemic preconditioning (Aggarwal et al. 2017; Hu et al. 2014).

Genetic variants in ABCC9 are also associated with human diseases in which multiple organ systems, particularly the brain, may be affected. For example, Cantú syndrome and ABCC9-related intellectual disability and myopathy syndrome (AIMS) are human conditions that, respectively, result from gain- and loss-of-function from ABCC9 mutations (Harakalova et al. 2012; Smeland et al. 2019; van Bon et al. 2012). Both of these rare monogenic syndromes are associated with neurological manifestations that include cerebrovascular pathology (Efthymiou et al. 2024a, 2024b; Nelson et al. 2015a, b, c). Multifactorial brain disorders linked to lower-penetrance ABCC9 gene variants include hippocampal sclerosis of aging (HS-Aging), sleep disorders, and depression (Nelson et al. 2015a, b, c). HS-Aging is a prevalent dementia-associated phenotype with pathologic features that include hippocampal cell loss, astrocytosis, and vasculopathy (Nelson et al. 2013; Neltner et al. 2014; Woodworth et al. 2025). The association between ABCC9 genetic variants and HS-Aging pathology was validated in several non-overlapping cohorts (Dugan et al. 2021; Katsumata et al. 2023; Nelson et al. 2014; Nelson et al. 2015a, b, c), but the finding was not replicated in one cohort from Finland (Hokkanen et al. 2020). A specific single nucleotide variant (SNV) in ABCC9, rs1914361, has been associated with differential risk for HS-Aging in persons of both European and African ancestry (Dugan et al. 2021; Katsumata et al. 2023).

With regard to the molecular biology of ABCC9/SUR2, limited insights have been gained into the regulation of ABCC9 transcription, mRNA processing, protein expression, or post-translational modifications (Shi et al. 2012), beyond the recognition that RNA splicing can produce alternative SUR2 mRNA transcripts (Chutkow et al. 1999; Shi et al. 2005; Ye et al. 2009). These include two common SUR2 splice variants, termed SUR2A and SUR2B (Fujita and Kurachi 2000), generated through differential inclusion of the last two ABCC9 exons, that encode the carboxy terminal region of the SUR2 protein (Chutkow et al. 1996; Davis-Taber et al. 2000; Inagaki et al. 1996; Isomoto et al. 1996).

Despite progress in other biological contexts, the regulation of ABCC9/SUR2 expression that occurs in the human brain remains to be characterized. A fundamental question is: which human brain cell types express the SUR2 mRNA and SUR2 proteins? Rodent Sur2 expression has been described as being highly enriched in brain pericytes, with key physiologic roles played by pericyte Sur2 in neurovascular coupling (Ando et al. 2022; Bowen et al. 2025; Hariharan et al. 2022; Isaacs et al. 2024; Ji et al. 2024; Longden and Isaacs 2025; Longden et al. 2023; Yang et al. 2022). Prior studies of ABCC9/SUR2 expression in mammalian (mostly not human) brains indicated ABCC9/SUR2 expression in brain vascular smooth muscle cells (SMCs) (Jansen-Olesen et al. 2005; Ploug et al. 2006, 2008), neurons (Ma et al. 2009; Wang et al. 2005; Zawar et al. 1999; Zoga et al. 2010), microglia (Ortega et al. 2012; Zhou et al. 2008), and/or other glial cells (Fogal et al. 2010; Zhou et al. 2012). Sur2 expression and K_ATP channel activity in endothelial cells have also been reported (Sancho et al. 2022).

We previously found evidence for co-regulation (i.e., decreased or increased transcript expression in register with each other) of ABCC9/SUR2 with another gene, Solute Carrier Organic Anion Transporter Family Member 1C1 (SLCO1C1) (Nelson et al. 2016), which is also referred to as Organic Anion Transporting Polypeptide 1C1 (OATP1C1) (Jansen et al. 2005). SLCO1C1 is an astrocyte-expressed protein that mediates transport of thyroid hormone from blood into the nervous system (Friesema et al. 2012). Astrocytes import thyroid hormone as part of their neurovascular unit (NVU) and gliovascular unit (GVU) functions, and astrocytes process thyroid hormone for sharing with other brain cells, resulting in regulation of neural development, neuroprotection, and vascular integrity (Brenowitz et al. 2018; Dezonne et al. 2015; Morte and Bernal 2014b; Noda 2018; Razvi et al. 2018; Soetedjo et al. 2024). The ABCC9 and SLCO1C1 genes are situated ~ 1 million base-pairs apart from each other on human chromosome 12 and specific germline gene variants were associated with altered expression (in the same “direction” of change) of both transcripts in human brain (Nelson et al. 2016). We also reported evidence for SUR2 being expressed in human astrocytes (Nelson et al. 2016) but had not further investigated astrocyte ABCC9/SUR2 expression prior to the current study.

Collectively, the existing scientific literature underscores the need for careful assessment of ABCC9/SUR2 gene expression in the human brain, stratified by specific cell types. In the present study, we deployed different methods and data sets, including single-nucleus RNA-seq (snRNA-seq) data analyses and immunohistochemistry, to query the cell type-specific expression of SUR2 transcripts and SUR2 protein. Studies included analyses of data derived from brain samples sourced from the University of Kentucky AD Research Center (UK-ADRC) community-based autopsy cohort. We found that in addition to other cell types of the NVU/GVU, an intriguing subpopulation of human astrocytes preferentially expresses SUR2 in human brain.

Methods

Single-nucleus RNA-seq and Genotype Data from the Seattle Alzheimer's Disease Brain Cell Atlas (SEA-AD) Project

For initial snRNA-seq data analyses, processed 10 × snRNA-seq data referent to human middle temporal gyrus (MTG) samples from the Seattle Alzheimer's Disease Brain Cell Atlas (SEA-AD) study (Gabitto et al. 2023; Hawrylycz et al. 2024) were obtained via the Registry of Open Data on Amazon Web Services (formatted as SEAAD_MTG_RNAseq_final-nuclei.2024–02–13.h5ad). Detailed descriptions of the SEA-AD cohort, sample processing, tissue dissection, and nuclei isolation are available at the Brain Knowledge Platform (https://brain-map.org/consortia/sea-ad/our-data). Briefly, the SEA-AD consortium includes the Allen Institute for Brain Science, the University of Washington, and the Kaiser Permanente Washington Research Institute. The study provides transcriptomic data from single nuclei isolated from postmortem brain tissue of 89 older adult donors (Gabitto et al. 2023; Hawrylycz et al. 2024) who were enrolled in the Adult Changes in Thought Study or the University of Washington ADRC cohort. The 10xV3.1 chip loading was performed as per 10 × guidelines (https://www.protocols.io/view/10xv3-1-genomics-sample-processing-dm6gpwd8jlzp/v3).

From the provided H5AD file, we extracted the Unique Molecular Identifiers count matrix and associated metadata for astrocytes based on the annotated cell types, and constructed Seurat objects using the “CreateSeuratObject” function in the Seurat R package (v5.3.1) (Hao et al. 2024). Because sex differences were observed in the initial subclustering analysis (Supplementa1 Fig. 1A), separate Seurat objects were generated for male and female nuclei and subsequently integrated using the “SelectIntegrationFeature”, “FindIntegrationAnchors”, and “IntegrateData” functions. Subclustering was then performed on the integrated object, effectively controlling for sex-related effects (Supplementa1 Fig. 1B). The same workflow was applied to the remaining cell populations.

Genotype imputation was performed using the Michigan Imputation Server 2 (v2.0.11) with Minimac 4(Das et al. 2016; Howie et al. 2012). The Haplotype Reference Consortium (HRC r1.1 2016) reference panel (GRCh37) with the European (EUR) population option was used for imputation, and Eagle v2.4 (Browning and Browning 2011; Delaneau et al. 2013; Durbin 2014; Koch et al. 1989; Loh et al. 2016) was used for pre-imputation haplotype phasing. We then extracted genotype data for rs1914361 (chr12: 22045853 on GRCh37, within a SUR2 intron), which showed high imputation accuracy (R² = 0.983).

Human Brain Samples from the UK-ADRC Autopsy Cohort

Some of the human brain samples that were used for single-nucleus analyses and antibody studies (immunoblots and immunohistochemistry) were sourced from the UK-ADRC autopsy cohort biobank. This is a community-based cohort that recruits from the Lexington, Kentucky USA region, and was described previously along with recruitment details (Nelson et al. 2019; Schmitt et al. 2012; Smith et al. 2017). Briefly, protocols were approved by the University of Kentucky Institutional Review Board, and all participants provided written informed consent. Participants and their study partners were administered the Clinical Dementia Rating (CDR) (Morris 1993) at each visit. Detailed protocols for the neuropathologic workup at the UK-ADRC were previously described (Abner et al. 2018; Nelson et al. 2023) and followed conventional diagnostic methodologies (Karanth et al. 2020; Nelson et al. 2010; Neltner et al. 2016). In terms of the data related to UK-ADRC participants, the demographic and neuropathologic features are shown in Refs (Sziraki et al. 2023a, b) for EasySci and (Lu et al. 2023) for LifeSci and in Supplemental Table 1 for western blots and immunohistochemistry.

Spatial Transcriptomics in Human Brain

For a separate readout of human brain cell type expression of ABCC9/SUR2, we performed spatial transcriptomics. Here we utilized the 10 × Genomics Xenium platform, following a published protocol (Saito et al. 2025), on a section of frontal cortex (Brodmann area 9) from a female research volunteer who died at age 93, after having been cognitively normal at last clinic visit. We used a custom Xenium probe panel targeting canonical cell-type markers for astrocytes, microglia, oligodendrocytes, neurons, and neurovascular unit cells (Lee et al. 2023). Xenium DAPI images were visualized in QuPath (v0.5.1) and Fiji/ImageJ (1.54f) with the BigDataViewer-Playground extension (Pietzsch et al. 2015; Schindelin et al. 2012). Seurat (v5.1) was used for initial object creation, excluding cells with zero detected transcripts (Hao et al. 2025). Using Python skimage (0.22.0) package and HALO software as previously described (Saito et al. 2025), images were annotated and pseudocolored for detected ABCC9/SUR2 transcripts, as well as astrocyte markers AQP4 and SOX9.

Single-nucleus RNA-seq using EasySci Platform to Compare Human SUR2 and Mice Sur2 Expression, and to Assess Specific (SUR2A and SUR2B) Splice Variants in Human Brain

We evaluated data obtained using the EasySci platform snRNA-seq method (Sziraki et al. 2023a, b) in order to accomplish three goals: 1. Validate (or refute) the findings from the SEA-AD data set in terms of human astrocyte SUR2 expression; 2. Compare cell-type specific SUR2/Sur2 expression results between human and mice brains using the same snRNA-seq methodology; and, 3. Evaluate cell type-specific SUR2 splice variants in human brains. To assess cell type-specific SUR2 expression we used a human nuclei data set for which the data generation methodology was described in detail previously (Sziraki et al. 2023a, b). Briefly, a human brain atlas data set was derived from twelve human brain hippocampus samples (Sziraki et al. 2023a, b). To validate the result, we analyzed a normal human brain atlas profiling twenty-nine post-mortem human brain samples across five regions brain (cerebellum, hippocampus, inferior parietal regions, motor cortex, and superior and medial temporal gyrus) and six individuals (three male and three female, all cognitively normal proximal to death) ranging from ages 70–94 years at death (Lu et al. 2023). For the comparison between human and mouse brains, we assessed the mouse brain atlas data comprised of 20 mice, with differing genotypes but all with C57BL/6 strain background, spanning different ages and genders as described in detail previously (Sziraki et al. 2023a, b). In both the human and mice data sets, the different vascular cell types were annotated based on conventional marker genes (Sziraki et al. 2023a, b).

Single-nucleus expression values for ABCC9/SUR2 were normalized using size factors, multiplied by 10,000, offset with a pseudo-count, and log-transformed. Cell-type–specific mean expression and standard error were then computed from the normalized single-nucleus data. To quantify ABCC9/SUR2 isoform usage across cell types, we first aggregated FASTQ files for each cell type within each sample and processed them using an established RNA-isoform quantification pipeline (Salmon (Patro et al. 2017)) with gencode.v19.transcripts.fa as the reference and GENCODE human Release 19 annotation (parameters: fldMean = 150, fldSD = 80). ABCC9-related (SUR2 transcript) isoforms were manually annotated and classified into SUR2A or SUR2B based on differential exon inclusion near the 3′ untranslated region. Isoform-specific expression differences were assessed using the Mann–Whitney U test.

Anti-SUR2 Antiserum: Development, Immunoblots, and Immunohistochemistry

A polyclonal anti-SUR2 rabbit antiserum was generated using 3 injections of polypeptide over 90 days at a commercial facility (Thermo/Pierce). The immunogen was a 385 amino acid long polypeptide corresponding to amino acids #597–981 of SUR2 proteins SUR2A (NP_005682.2) and SUR2B (NP_064693.2). See Supplemental Fig. 2 for more details on the antigen.

Plasmid transfection experiments with immunoblots were used to assess the specificity of the SUR2 antiserum. Plasmid ABCC8/SUR1 (Myc-DDK-tagged)-Human ATP-binding cassette (SUR1) was purchased from OriGene Technologies, Inc (Cat # RC215352, Rockville, MD 20850), plasmids pcDNA3.1-HsSUR2A and pcDNA3.1-HsSUR2B were previously reported (Gao et al. 2023a, b). Plasmids were prepared using E. coli DH5α competent cells. Prior to transfection, all plasmids’ sequences were verified by restriction digestion and whole plasmid sequencing (Eurofins Genomics, DNA Sequencing Lab, Louisville, KY 40299). Plasmids were purified using the QIAprep Spin Miniprep Kit and Qiagen Plasmid Maxi kit (Qiagen Science). To express the various plasmids for immunoblots, HeLa cells were grown in Dulbecco’s modified Eagle medium (DMEM) (Sigma life Science, D6546) supplemented with 10% fetal bovine serum (Gibco, Cat # 16000–044), 2 mM L-Glutamine, 100 units/ml penicillin, and 100 ug/ml streptomycin. The cultured HeLa cells were transfected with the SUR1, HsSUR2A and HsSUR2B plasmids using Lipofectamine 3000 transfection Kit (Thermo Fisher Scientific, Cat # L3000) according to the manufacturer’s instructions. Cells were seeded in a 6-well plate in DMEM supplemented with 10% fetal bovine serum (FBS) and transfected with 2.5 µg of plasmid vector using Lipofectamine 3000 following the manufacturer’s instructions. After 6 h transfection, the medium was removed and replaced with fresh 10% FBS/DMEM. After further incubation at 37 °C for 2 days, the cells were washed with ice cold PBS for 3 times and then lysed with RIPA buffer (EMD Millipore Crop, Billerica, MA) including Halt protein inhibitor cocktail (Thermo Fisher Scientific Cat # 1861279) and Halt phosphatase inhibitor cocktail (Thermo Fisher Scientific Cat # 78427) on ice for 2 h. The cell lysates were collected and centrifuged at 13,000 rpm, for 5 min at 4 °C. The supernatant was used for Western blot analyses on an equal protein basis.

Isolation of protein fractions (Low Salt, Triton X-100, sarkosyl [N-lauroylsarcosine] and Urea extractions, the latter two likely to solubilize the membrane-bound SUR2 protein) from human brain followed our published methodology (Gal et al. 2018). For immunoblots, proteins were separated by denaturing SDS gel electrophoresis using 8% SDS-PAGE gels (homemade); gels were run at 60 V constant. The proteins were then transferred to Immun-Blot PVDF membranes (BIO-RAD, Cat # 1620177) in transfer buffer containing 20% methanol. The membranes were blocked with 5% non-fat dry milk and the antibodies were applied in DPBS-T (8 g/L NaCl, 0.2 g/L KCl, 0.2 g/L KH₂PO_4, 1.15 g/L Na₂HPO₄, 0.05% [v/v] Tween-20). Primary antibodies/antisera used in immunoblots were the novel rabbit anti-SUR2 antiserum, diluted at 1:500; and, anti-β-Actin mouse antibody, 1:1000 (Santa Cruz Biotechnology Cat # sc47778). We also separately tested another anti-SUR2 antibody, clone N323A/31 (Thermo Fisher Cat # 75–298-FL650). The secondary antibodies were HRP-linked anti-rabbit (Jackson Immuno Research Laboratories, Inc. Cat # 111–035–144) and HRP-linked anti-mouse (Jackson Immuno Research Laboratories, Inc. Cat # 115–035-003). The blots were developed with the SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Fisher Cat # 34580), and imaged using a FlourChem R Imaging System (ProteinSimple Instruments).

For immunohistochemistry, slides were cut from formalin-fixed paraffin-embedded (FFPE) tissue blocks at 8 microns thickness. Cut slides were deparaffinized and antigen retrieval performed with pH 9 buffer for 6 min at 95ºC. Slides were then incubated in a 10% methanol and 3% hydrogen peroxide solution for 30 min. Following more washes, slides were incubated in a TBS buffer containing 3% milk, 4% goat serum, 0.2% Triton X-100, 1.9% L-lysine, for 1 h at room temperature. After that, 3% milk and 4% goat serum buffer was used for the remaining washes and antibody dilution. The SUR2 primary antiserum was incubated at 1:500 dilution at 4ºC. The next day, following additional washes, rabbit secondary antibody (Vector Labs Cat # BA-1000), diluted 1:500, was added and slides incubated at room temperature for 1 h followed by chromogen reaction using the VECTASTAIN ABC reagent (Vector labs) and 3,3’-diaminobenzidine (Thermo Fisher Scientific) according to the manufacturer’s instructions. For immunoabsorption immunohistochemical experiments, the primary and second antibodies were incubated in 1xTBS with 3% milk, 4% goat serum, while the blocking buffer was the same but contained an additional 0.2% Triton-X100 and 1.9% L-Lysine. The antiserum was incubated overnight for preadsorption prior to use at 4ºC at 1:1 ratio (undiluted antiserum:purified polypeptide at [2.5 mg/ml]). The same preadsorption method was used for immunoblotting.

Immunofluorescent double-labeling for SUR2 and GFAP was also performed on human brain tissue sections cut from FFPE blocks at 8 micron thickness. Slides were deparaffinized and antigen retrieval carried out in a pH 9 buffer at 95 °C for 6 min. For blocking nonspecific antibody binding, sections were incubated for 1 h at room temperature in 1 × TBS containing 3% milk, 4% goat serum, 0.2% Triton X-100, and 1.9% L-lysine. Primary antibodies—anti-SUR2 (rabbit, 1:500) and anti-GFAP (Santa Cruz Biotechnology clone 2E1 from mouse, Cat # sc-33673, 1:1000)—were diluted in buffer containing 3% milk and 4% goat serum and slides incubated with primary antibodies overnight at 4 °C. The following day, slides were washed twice in 1 × TBS for 5 min each. Secondary antibodies (anti-rabbit for SUR2 and anti-mouse for GFAP, both at 1:250: these were Alexa Fluor 488 goat anti-rabbit IgG second antibody, Thermo Fisher Scientific Cat # A11008; and, Alexa Fluor 594 goat anti-mouse IgG, Thermo Fisher Scientific Cat # A11032) were applied for 1 h at room temperature in the dark, followed by two additional 5-min washes in 1 × TBS. To reduce lipofuscin and other autofluorescence, sections were incubated in 1 × TrueBlack (Thermo Fisher Scientific Cat # 50–196–4406, 20 × stock; 50 µL stock diluted in 1 mL 70% ethanol) for 10 min. Slides were then re-rinsed and coverslipped using ProLong Gold Antifade Reagent (Invitrogen, Cat # P36935).

Statistical Analysis

An enrichment analysis was performed using a Fisher’s exact test to assess whether thyroid hormone–related genes reported in Ibanez et al. (Gil-Ibanez et al. 2015) were overrepresented among the genes most strongly correlated with SUR2 transcript levels (defined as Pearson correlation [r] > 0.15; while these were the strongest correlating transcripts, we note that r ~ 0.15 is still a relatively weak correlation).

Differential abundance of astrocyte clusters was assessed using the propeller method implemented in the speckle R package (version 1.12.0), with individual donors treated as biological replicates. Comparisons were performed between Braak NFT Stage 0 and Stages II–VI, Thal Aβ Phase 0 and Phases 1–5, and LATE-NC Stage 0 and Stages 1–3. Cell-type proportions were calculated for each. P values across clusters were adjusted for multiple testing using the false discovery rate (FDR) method.

To evaluate the association of the rs1914361 SNV with SUR2 expression in pericytes, SMCs, and astrocytes, we applied a hurdle (zero-inflated) regression model using the “zlm” function from the “MAST” R package (v1.33.0) (Chen et al. 2025). The model was fitted to a SingleCellAssay object constructed from the Unique Molecular Identifier count matrix using the FromMatrix function.

Results

Cell type-specific SUR2 expression patterns were initially analyzed from human brain (MTG) samples in the SEA-AD study. In this data set, the SUR2 transcripts were primarily detected in pericytes, vascular SMCs, and astrocytes (Fig. 1A). There was lower but above-background SUR2 expression detected in endothelial cells. By contrast, there were even more marginal levels of SUR2 detected in neurons, oligodendrocytes, and microglia. To reexamine the initial findings of SUR2 expression across human brain cell types, we analyzed an independent single-nucleus atlas generated from 12 human hippocampi profiled using the EasySci-RNA-seq platform (Sziraki et al. 2023a, b) (Fig. 1B). Data were analysed from the EasySci data set derived from 118,240 human brain cell nuclei, and ~ 1.5 million mouse (C57BL/6 strain) brain single nuclei; the acquisitions of these data were described in detail previously (Sziraki et al. 2023a, b). These data broadly confirmed the results of the SEA-AD data set, with the exception of slightly higher neuronal SUR2 expression and slightly lower SUR2 expression in vascular SMCs. When we evaluated for comparison sake the Sur2 expression levels in mice brains (C57BL/6 strain) using the same EasySci technique, some of the results were similar (Fig. 1C) – Sur2 was greatly enriched in mice pericytes as in human brain. However, in mice brains, Sur2 expression was relatively low, if not absent, in astrocytes.

Fig. 1.

Expression of SUR2 stratified by brain cell subtypes. Panel A shows results from the SEA-AD data set (Hawrylycz et al. 2024). The cell subtypes are determined based on Uniform Manifold Approximation and Projection (UMAP) clustering that were performed by the SEA-AD consortium aggregated from 89 different human middle temporal gyrus samples. Note that cells of the neurovascular unit expressed SUR2: pericytes, SMCs, astrocytes, and endothelial cells. Complementary analyses of EasySci snRNA-seq data sets (Sziraki et al. 2023a, b) were performed on human B and mice C brains. The human samples were hippocampal tissue from 12 humans (6 nondemented, 6 with Alzheimer’s disease); mice studies comprised pooled results from 20 mice (C57BL/6 base strain) spanning different ages and genders (Sziraki et al. 2023a, b). Note that as in the SEA-AD study set, the EasySci snRNA-seq data indicated that human astrocytes express SUR2. However, Sur2 was not detected in mice astrocytes

We followed up the analyses of existing data sets with an additional study of human brain ABCC9/SUR2 expression using new spatial transcriptomics. The spatial transcriptomic (Xenium platform) results in human frontal cortex (Fig. 2) were broadly similar to the findings using other methods, in that the predominant cell types expressing ABCC9/SUR2 were pericytes, followed by astrocytes, and next by cell types of the vascular unit (vascular smooth muscle cells and endothelial cells). Note that although shown to be expressed in astrocytes according to spatial transcriptomics, the ABCC9/SUR2 transcripts were only detected in a minority (~ 28%) of astrocytes.

Fig. 2.

Spatial transcriptomics (Xenium platform) in human frontal cortex show co-expression of ABCC9/SUR2 with astrocyte-expressed transcripts. A digitized photomicrograph shows cells with selected transcripts (ABCC9 along with astrocyte markers AQP4 and SOX9) pseudo-colored in red, blue, and yellow respectively. These results were quantified by cell types B, again demonstrating that ABCC9/SUR2 is expressed at highest levels (as percent of cells labeled) in pericytes, followed by astrocytes. In pericytes, 73% had detectable ABCC9/SUR2 transcripts, whereas only 28% of astrocytes were shown to express ABCC9/SUR2

We next sought clues as to whether a specific subset of astrocytes express ABCC9/SUR2. Using the integrated snRNA-seq data set of astrocytes from the SEA-AD study, we reran a UMAP analysis for astrocyte subclustering. This revealed seven subclusters, designated subcluster 0 through 6 (Fig. 3A). Although SUR2 expression was detected across all astrocyte subclusters, the expression of SUR2 was most enriched in the cells within subcluster 3 (Figs. 3B and C).

Fig. 3.

Expression of SUR2 in human astrocytes: SEA-AD data set analyses. Panel A shows the results of UMAP clustering. This analysis was performed on data from middle temporal gyrus samples in the SEA-AD data set (Hawrylycz et al. 2024) as in Fig. 1A, but only assessing the astrocytes. Subclusters identified across all the astrocytes are depicted A. Seven different astrocyte subclusters were identified, numbered #0–6. In panel B, the cells expressing moderate or high levels (i.e., more than median) of SUR2 are indicated with colored markers. Panel C shows the SUR2 expression in the different subclusters. SUR2 expression was highest in astrocyte subcluster #3. Other transcripts were used to help indicate the astrocyte phenotypes that were represented by the different UMAP-subdivided astrocyte subclusters. Panel D shows the expression pattern for SLC1A2, a transcript that was associated with protoplasmic astrocyte phenotype according to a prior study (Dai et al. 2023). This reference also indicated that CD44 is a marker for fibrous astrocytes, VIM for reactive astrocytes, and NRXN1 for homeostatic astrocytes (Dai et al. 2023). Based on those characterizations, and with the caveat that actual astrocyte phenotypes are more biologically complex than these categories would indicate, at least some of the seven astrocyte subclusters identified in the SEA-AD data set could be assigned to an astrocyte phenotype. As shown in tabular format in panel E, the astrocyte subcluster #3, that was enriched for SUR2 transcripts, did not represent a clear-cut example of either protoplasmic or fibrous astrocytic phenotype based on the abovementioned marker transcript expressions. The transcript most strongly correlated with SUR2 in astrocyte subcluster #3 was SLCO1C1. Further, the transcripts that correlated with SUR2 expression in subcluster #3 were enriched with thyroid hormone-responsive gene products as previously reported (Gil-Ibanez et al. 2015)

Given that the individual astrocyte subclusters expressed different levels of SUR2, we tested preliminarily whether astrocyte subclusters #0 to #6 were sub-classifiable according to stereotypical expression profiles, as reported by Dai et al. (2023): in this study, specific transcript markers were associated with protoplasmic (marked by relatively high SLC1A2 expression); fibrous (CD44); reactive (VIM); or, homeostatic (NRXN1) astrocyte classification. Not all of the potential astrocyte-subtyping markers were useful in this regard, e.g. VIM expression was too low in the SEA-AD astrocyte cells. However, some of the subclusters could be assigned to either protoplasmic/homeostatic (subclusters #0, #1, and #5) or fibrous (subcluster #2) astrocyte expression patterns (Figs. 3D,E). The SUR2-enriched subcluster #3 appeared to represent an intermediate phenotype with regard to the stereotypical type-defining transcript markers – compare the UMAP plot in Fig. 3B with those in Fig. 3D.

We next evaluated the astrocyte transcripts that were co-expressed with SUR2 in human brain. A list of transcripts that were correlated most strongly with SUR2 among astrocytes in the SEA-AD data set, stratified by UMAP-identified subclusters, is shown in Table 1. In the SUR2-enriched astrocyte subcluster #3, the transcript with the strongest correlation with SUR2 levels was SLCO1C1 (r = 0.205) (Fig. 3E, Table 1, and Supplemental Table 2). Note that SLCO1C1 was also among the top SUR2-correlated transcripts in astrocyte subclusters # 1, 2, and 6 (shown in bold font in Table 1), thus spanning a range of protoplasmic and fibrous astrocyte phenotypes. We performed an enrichment analysis to assess the biological implications of SUR2-correlated transcripts. First, we cross-checked the list of SUR2-correlated transcripts with ShinyGO 0.85.1(https://bioinformatics.sdstate.edu/go/) and found no basis of commonality related to pathways, exposures, etc., among this group of transcripts (data not shown). Since we had previously found evidence of a connection between SUR2 and thyroid hormone sensitive gene products (Nelson et al. 2016), we next assessed the subcluster-specific lists of SUR2-correlated transcripts (as presented in Supplemental Table 2) for enrichment in association with a prior study that evaluated the mouse brain transcriptome after thyroid hormone treatment (Gil-Ibanez et al. 2014). In this enrichment analysis, we found that there was over-representation for thyroid hormone-sensitive transcripts among the group of SUR2-correlated transcripts in subcluster #3 (Fig. 3E).

Table 1.

Top genes correlated with SUR2 in each astrocyte subcluster

Subcluster 0	Subcluster 1	Subcluster 2	Subcluster 3	Subcluster 4	Subcluster 5	Subcluster 6
GFAP	PLSCR4	FAM189A2	SLCO1C1	LMO3	PCDH9	AQP4
COLEC12	ASPH	AQP4	CADM1	ERBIN	NRXN1	F3
AC008250.1	ERBIN	TMTC2	AL589740.1	GPCPD1	NCKAP5	SLCO1C1
NEAT1	SLCO1C1	PLSCR4	TMTC2	SLC14A1	GABRB1	ATP1B2
SORBS1	COLEC12	CRISPLD1	PCDH9	CNTNAP3B	OBI1-AS1	MRVI1
ERBIN	SMAD1	NFIA	ASPH	ADCY2	ADGRV1	PCDH9
DTNA	STK3	SLCO1C1	PREX2	ACSS3	GPM6A	PMP2
TPST1	FAM189A2	OSMR	NRP1	MGST1	LMO3	NRP1
RFX4	BMPR1B	PRKG1	TENM2	DTNA	DNAH7	CADM1
MAOB	AQP4	CLU	PSD3	RHOBTB3	NRG3	PREX2

The proportion of astrocytes assigned to Cluster 3 was higher in participants with Braak NFT Stage 0 than in donors with Stages II–VI (FDR-adjusted P = 0.011; Fig. 4). This indicates that the SUR2 + astrocyte cluster tended to derive from brains lacking appreciable tau pathology. Suggestive findings in terms of astrocyte cluster assignments and Aβ and LATE-NC pathologies are shown in Supplemental Fig. 2.

Fig. 4.

Astrocyte subclusters in the SEA-AD data set, stratified by Alzheimer’s disease-type Braak neurofibrillary tangle (NFT) stages. A SUR2-enriched Cluster 3 astrocytes are delineated on the UMAP chart using a dashed outline (see Fig. 3). Panel B depicts the same astrocytes, false-colored according to Braak NFT stages. Astrocytes in the SUR2 + Cluster 3 tended to lack higher Braak NFT stages (P < 0.011). This tendency can also be appreciated in stacked bar chart formatted data (C). Arrows indicate Cluster 3. The color key on the right applies to both panels B and C

Research participants whose brains were included in the SEA-AD data set had their germline DNA (individual-specific SNVs) characterized, in addition to their snRNA-seq gene expression data. Using these data together, we next correlated brain ABCC9/SUR2 expression with the status of a germline SNV in the ABCC9 gene, rs1914361, which has been associated with altered risk for pathologically-confirmed HS-Aging and also with altered expression of SUR2 (Dugan et al. 2021; Katsumata et al. 2023, 2017; Nelson et al. 2014; Nelson et al. 2015a, b, c). The rs1914361 risk allele was shown previously to be associated with higher expression of SUR2 in some human tissues, and lower expression in other tissue types (Dugan et al. 2021). However, cell-type specific ABCC9/SUR2 expression patterns associated with rs1914361 have not previously been described. When data were cross-checked between germline SNV status and transcript expression in the SEA-AD data set, we found that cells with the rs1914361 HS-Aging risk allele (guanine) had higher expression of SUR2 in astrocytes (P = 1.5 × 10^–147) (Fig. 5) and SUR2 expression also trended higher in pericytes with the risk allele (P = 0.098), but SUR2 expression in those with the risk allele trended marginally lower in SMCs (P = 0.16) (Fig. 5). Analogous data for the other cell types (neurons, endothelial cells, oligodendrocytes, and microglia), stratifying by rs1914361 status, are shown in Supplemental Fig. 3.

Fig. 5.

Cell type-specific SUR2 expression stratified by rs1914361 G allele status in pericytes, vascular smooth muscle cells (SMCs), and astrocytes. The single nucleotide variant rs1914361 has previously been associated with HS-Aging and SUR2 expression (Dugan et al. 2021; Katsumata et al. 2023). The “G” allele is associated with increased risk for HS-Aging. Here we tested the association between rs1914361 G allele status and detected SUR2 transcript levels in human brain stratifying by cell types – pericytes A, SMCs B, and astrocytes C. Persons with the rs1914361 HS-Aging risk allele had higher expression of SUR2 in astrocytes (P = 1.5 × 10^–147 in additive mode of inheritance [MOI] and P = 2.5 × 10^–248 in dominant MOI). The rs1914361 risk allele was not associated with statistically significant differences of SUR2 expression in either pericytes or SMCs in additive MOI. For analogous results for neurons, microglia, and oligodendrocytes, see Supplemental Fig. 3. ADD = additive MOI; DOM = dominant MOI. Red dots represent median values

Single-nucleus RNA-seq to Test Cell-type Specific SUR2 Splice Variant Patterns

The next goal was to assess SUR2 transcript splicing variants in different cell types, particularly astrocytes. A key feature of the EasySci platform is its full gene body (complete transcript) coverage, achieved by implementing both oligo-dT and random primers during the reverse transcription step (Sziraki et al. 2023a, b). This enabled us to examine cell type-specific patterns of exon processing in human brain samples. Specifically, SUR2 transcripts can be spliced to produce different mRNA isoforms, two of which have been termed SUR2A and SUR2B. In our analyses of human brain tissues profiled by EasySci (Sziraki et al. 2023a, b), astrocytes expressed SUR2B levels significantly higher than SUR2A levels, P < 4 × 10⁻⁵, whereas pericytes expressed both SUR2A and SUR2B isoforms at statistically indistinguishable levels (Figs. 6A, B). To further confirm this result, we assessed another independent human brain atlas data set generated using EasySci (Lu et al. 2023) and obtained the same result: astrocytes preferentially expressed SUR2B, whereas pericytes expressed both SUR2A and SUR2B (Figs. 6C, D). When we attempted to further evaluate cell-type-specific patterns of SUR2 expression, the numbers of transcripts with isoform-specific features were insufficient for rigorous testing in SMCs, endothelial cells, or other brain cell types.

Fig. 6.

Expression of SUR2 splicing variants SUR2A and SUR2B in human brains as detected using snRNA-seq with exon-level resolution. The EasySci (Sziraki et al. 2023a, b) method enables assessment of exon-specific readouts to generate insights into cell type-specific mRNA splicing in human brain. In an EasySci data set generated by pooling data from 12 human hippocampi (Sziraki et al. 2023a, b), astrocytes expressed the SUR2B-specific exon at levels that were significantly higher than the SUR2A-specific exon, P < 4 × 10⁻.⁵, whereas pericytes expressed both SUR2A and SUR2B isoforms at statistically indistinguishable levels A, B. We further assessed another independent EasySci human brain atlas data set generated using different human brains (Lu et al. 2023) and obtained essentially the same result: astrocytes preferentially expressed SUR2B (p < 0.03), whereas pericytes expressed both SUR2A and SUR2B C, D

SUR2 Antiserum: Western Blots and Immunohistochemistry

A novel anti-SUR2 antiserum was developed in a rabbit immunized using a protein corresponding to amino acids #597–981 of SUR2 proteins (the protein sequence is presented in Supplemental Fig. 4). This sequence is entirely shared by SUR2A (NP_005682.2) and SUR2B (NP_064693.2) protein isoforms, both of which are 1549 amino acids long. The immunizing polypeptide sequence is 70% identical to a homologous sequence of the SUR1 protein (NP_001274103.1). To assess the specificity of the antibody, we ran immunoblots on extracts of HeLa cultured cells transfected with plasmids expressing SUR1, SUR2A, SUR2B, and a control plasmid lacking a cDNA insert. As expected, there was augmentation of high-molecular weight protein stained in anti-SUR2 immunoblots referent to the SUR2A and SUR2B transfections (Fig. 7A); the predicted molecular weight of SUR2A and SUR2B are ~ 174 kilodaltons. We next stained strong detergent- and urea-solubilized fractions from human brain and found similar high-molecular weight bands stained (Fig. 7A), including discrete bands at ~ 165 and 200 kDa, which may be unglycosylated and core glycosylated proteins, as well as diffuse higher MWs, which may reflect higher order glycosylation (Conti et al. 2002). However, we also note that there were some low-molecular weight bands stained on the immunoblot that indicated antiserum cross-reaction with other antigen(s).

Fig. 7.

An antiserum raised against human SUR2 reacts with transfected human SUR2 protein on immunoblots and also primarily labels vascular profiles in immunohistochemical (IHC) staining of human brain sections. The antigen corresponded to a fragment of SUR2 shared by both SUR2A and SUR2B isoforms; see Supplemental Fig. 3. A. The immunoblot on the left is stained with lanes representing plasmid-only, SUR1, SUR2A, and SUR2B, highlighting a high-molecular weight “smear” pattern (*) in both SUR2A and SUR2B-transfected cells. On the SUR-2 immunoblot on the right, human frontal cortex was extracted with sarkosyl detergent and with urea, to gain access to membrane-bound protein fractions, resulting in the staining of both high-MW and a lower-molecular weight (**) bands. B. Brightfield IHC staining shows primarily the walls of large and small blood vessels, but also labels some smaller cells with astrocyte morphology (arrows). Some presumed cross-reaction was seen on IHC with neuronal nuclei (Supplemental Fig. 4). The SUR2 antiserum was used for immunofluorescent staining of human brain sections, enabling the visualization of SUR2 (green; C), glial fibrillary acidic protein (GFAP; red; D), and both stained proteins together (E). The SUR2 antiserum stains the walls of blood vessels, and also co-labels the red-stained (GFAP +) astrocytes (arrows) and a portion of astrocytic end-foot processes that are concentrated around blood vessels (bv). Scale bars: 90 μm in B 80 μm in C-E

We next tested the anti-SUR2 antiserum using immunohistochemical methods that included brightfield (Fig. 7B) and immunofluorescence (Figs. 7C-E) microscopy. This anti-SUR2 antiserum stained blood vessels, large and small, including the mural portion of larger blood vessels and the delicate internal juxtaluminal layer (possibly pericytes and/or endothelial cells) of smaller blood vessels. In addition, other cells were stained including co-labeling of GFAP-immunoreactive astrocytes, as shown in Fig. 7C-E. The SUR2 + astrocyte immunostaining appeared to localize more in the cytoplasm than in astrocyte end-feet. To further assess the molecular specificity of the novel anti-SUR2 antiserum, we performed immunoadsorption experiments via immunoblots and immunohistochemistry. As expected, preincubation of the antibody with the polypeptide originally used as an immunogen abolished staining in western blots and immunohistochemistry (Supplemental Fig. 5). However, without pre-adsorption, there was in many cases faint immunolabeling in neuronal nuclei, that we judged to be likely cross-reaction, as shown in Supplemental Fig. 6. We also tested a commercially available anti-SUR2B monoclonal antibody, clone N323A/31 from Thermo Fisher. However, when assessing human brain-derived protein extracts on immunoblots, this antibody in our hands did not recognize a banding pattern consistent with recognition of SUR2 (Supplemental Fig. 7). Thus, further studies were not performed with this reagent. Overall, the immunohistochemical staining patterns for our novel SUR2 antiserum were consistent with expectations derived from snRNA-seq studies, because the antiserum preferably labeled blood vessels and cells with histologic appearance of astrocytes.

Discussion

ABCC9/SUR2 was expressed in multiple human brain cell subtypes involved in blood vessel function, particularly pericytes, vascular SMCs, and astrocytes. Endothelial cells showed a lower but detectable level of SUR2 expression. By contrast, SUR2 was expressed at still lower levels (if at all) in other human brain cell types including neurons, oligodendrocytes, and microglia. We also found evidence of complex cell type-specific gene expression regulation within the NVU/GVU: a HS-Aging-associated risk allele in ABCC9 was associated with differential changes in ABCC9/SUR2 expression in astrocytes, and distinct splice variants were preferentially enriched in different NVU/GVU cell types. The hypothesis that SUR2 is enriched in human brain pericytes, smooth muscle cells, and astrocytes, as indicated through single-nucleus expression profiling in two different data sets, was corroborated using immunohistochemical staining with a novel SUR2 antiserum that immunolabeled blood vessel walls and astrocyte-morphology cells.

To provide scientific context for the study findings, we first will review what is known about the normal functions of SUR2. Second, we describe the prior published literature related to SUR2 expression in the human brain. We then focus on the implication of astrocyte expression of SUR2 in human brains – including pitfalls in the methods that we used to detect and characterize the astrocyte SUR2 expression. Thyroid hormone-related pathways that are implicated in SUR2 + astrocytes are described. Next, we discuss the common human brain condition (HS-Aging) that is associated with ABCC9/SUR2 genetic variation, and why astrocyte SUR2 expression may play a role in that brain pathology. Finally, we consider the study limitations.

The biological and clinical implications of this study relate to the K_ATP channel of which SUR2 is a component. Numerous functions for K_ATP channels have been elucidated, varying by cell type, developmental stage, and other factors wherein SUR2 is involved in context-specific responses to different stimuli (Martin et al. 2023; Nichols 2006; Nichols et al. 2013). For example, K_ATP channels are involved in the paradigm of “ischemic preconditioning” (Rana et al. 2015; Sanada and Kitakaze 2004), which refers to homeostatic resistance to the adverse impact of loss of blood supply, oxygen, and/or glucose, in the aftermath of an ischemic challenge. Further, K_ATP channels mediate constitutive paracrine and metabolic signaling that modulate vascular tone and blood flow (Bowen et al. 2025; Shi et al. 2012). A common theme is that K_ATP channels compensate for stress and/or hypoxia by membrane hyperpolarization, reducing excitability and intracellular calcium levels (Nichols 2006). Although cardiovascular physiology has been a notable focus of K_ATP channel-related research (Nichols et al. 2013; Tinker et al. 2014), similar paradigms probably occur in human brain cells (Alkan 2009; Busija et al. 2008; Yuan et al. 2004).

Relatively little is known about how ABCC9 gene expression is orchestrated in the human nervous system. SUR2A is usually characterized as a skeletal and cardiac muscle-enriched transcript variant, and SUR2B as a vascular smooth muscle-enriched isoform (Chutkow et al. 1996; Davis-Taber et al. 2000; Isomoto et al. 1996; Ploug et al. 2010; Seino and Miki 2003; Shi et al. 2005), yet the actual splicing and expression patterns of SUR2 isoforms are not straightforward. There are other SUR2 transcripts for which the protein products’ functions are incompletely understood (Chutkow et al. 1999; Nelson et al. 2015a, b, c; Ye et al. 2009). Further, the human SUR2 3’UTR is variable in length in human brain (Nelson et al. 2015a, b, c); this variability can alter transcript stability and localization (Tushev et al. 2018). The results of the current study provided further clues: pericytes tended to express both SUR2A and SUR2B isoforms with a trend for higher SUR2A expression, whereas astrocytes were significantly enriched for SUR2B.

One of the limitations to the study of ABCC9/SUR2 in the human brain is the lack of technically optimal molecular probes that could characterize the subtypes of ABCC9-derived transcripts and polypeptides that are expressed. From the perspective of laboratory bench researchers that have tested multiple SUR2 probes and antibodies, we can attest to the imperfect specificity of these reagents in our hands (see (Nelson et al. 2014)). Thus, technically ideal SUR2A- and SUR2B-specific antibodies for human brain immunohistochemistry are not available. Even the novel antiserum that we introduced in the current study shows imperfect specificity via immunohistochemistry and on immunoblots.

In terms of SUR2 expression in individual brain cell types, prior studies have reported apparently contradictory results, with most previously published articles focusing on non-human species. For example, Zhou et al. evaluated Sur2 distribution in rat brain and found widespread expression (Zhou et al. 2012); neurons predominantly expressed Sur2A whereas Sur2B was more highly expressed in glial cells. In other rat studies, Sur2 protein was reported in dorsal root ganglia neurons (Zoga et al. 2010), and, Sur2 expression was reportedly enriched in hippocampal interneurons (Zawar et al. 1999). Other experiments in cultured primary rat neurons also found substantial Sur2 expression (Ma et al. 2009). A different study of rat brains showed that neuronal Sur2 expression increased following neurotoxin lesions in the prefrontal cortex (Wang et al. 2005). Other studies indicated that CNS vascular cells, e.g. endothelial and vascular smooth muscle cells, express SUR2 (Jansen-Olesen et al. 2005; Ploug et al. 2006, 2008). In addition to possible expression in neurons and vascular cells, ABCC9 has been implicated in astrocyte physiology. We previously presented data indicating that dysregulation of ABCC9 expression in human brain may relate to thyroid hormone signaling in astrocytes (Nelson et al. 2016). Finally, and appearing to disagree with findings in the abovementioned studies, recently published work has focused on the enrichment of Abcc9/Sur2 expression in mice brain pericytes (Ando et al. 2022; Hariharan et al. 2022; Isaacs et al. 2024). Our study supports the hypothesis that there is a relatively high level of expression of ABCC9 in multiple NVU/GVU cells, particularly pericytes, SMCs, and astrocytes, in human brains.

The detected expression of astrocyte SUR2/Sur2 was higher in human brain than mice brains in the present study. Thomzig et al. provided evidence for expression of the KATP pore-forming Kir6.1 protein in rat astrocytes (Thomzig et al. 2001), consistent with the presence of functional K_ATP channels in that species. The prior literature on snRNA-seq based characterization of astrocyte subtyping provides a rich collective source of reference (Gao et al. 2023a, b; Hennes et al. 2025; Lahaie et al. 2025). Some of the previously published snRNA-seq studies that included astrocytes focused on rodent species, tumors, and specific brain regions and developmental stages outside of the adult human cortex (Bocchi et al. 2025; Carroll et al. 2020; de Jager et al. 2025; Gonzalez-Velasco et al. 2025; Ikeda-Yorifuji et al. 2022; Kerr et al. 2025; Men et al. 2022; Pan et al. 2021; Qian et al. 2023; Scott et al. 2024; Tong et al. 2025; Xu et al. 2025; Zang et al. 2024); these studies did not refer specifically to ABCC9/SUR2, to the best of our knowledge. Additional recent snRNA-seq studies have focused on astrocyte subtypes in adult human brains (Dai et al. 2023; Galea et al. 2022; Sadick et al. 2022; Smith et al. 2022; Su et al. 2023; Yang et al. 2022).

We applied criteria derived from the study of Dai et al. (2023) to help assign general astrocyte classifications (e.g. protoplasmic vs fibrous phenotypes) based on specific astrocyte subcluster marking transcripts. According to the marker transcripts, the SUR2-enriched subcluster #3 appeared to occupy an intermediate phenotypic status between protoplasmic (SLC1A2-enriched) and fibrous (CD44-enriched) states. It is notable that we found that subcluster #3 astrocytes also tended to have been harvested from individuals with low severity of ADNC (lowest Braak NFT stages). While providing a useful and data-driven categorizing rubric, the definitional specificity of these markers could be debated and, biologically, there are surely more complex (anatomic site, developmental, and disease-specific) subtypes of astrocytes than are acknowledged by this classification scheme.

A key data source for the current study was SEA-AD (Gabitto et al. 2023; Hawrylycz et al. 2024), a publicly shared data set that was analyzed previously by others (Adeoye et al. 2024; Serrano-Pozo et al. 2024; Wei et al. 2025). In a study by Serrano-Pozo et al. (2024) that included analyses of SEA-AD and other human brain snRNA-seq data sets, nine astrocyte subclusters were identified of which one, designated “AstTinf”, most closely resembled the SUR2-enriched astrocyte subcluster #3 in the present study. This AstTinf subcluster was also enriched in transcripts encoding growth factors (particularly fibroblast growth factor-related), as well as interleukin signaling and extracellular matrix transcripts, and was one of several astrocyte subclusters that had a transcript profile that “deviate from the prototypical homeostatic-to-reactive transition path” (Serrano-Pozo et al. 2024).

Only a relatively small subset of prior snRNA-seq studies discussed astrocytes that are characterized by increased expression of genes induced following exposure to thyroid hormone. Yet astrocytes constitute the key cellular entry-points for blood-borne thyroid hormone in the CNS (Morte and Bernal 2014a) and numerous studies described the profound impact of thyroid hormone treatment on mammalian brain gene expression, including dementia-related genes (Brase et al. 2024; Contreras-Jurado and Pascual 2012) and across cell types (Das et al. 2018; Dezonne et al. 2015; Diez et al. 2021; Gil-Ibanez et al. 2015; Morte et al. 2010, 2018; Niedowicz et al. 2023; Zhang et al. 2024). Relying on the published work by Dr. Beatriz Morte and colleagues to define the thyroid hormone-induced brain transcripts (Gil-Ibanez et al. 2014, 2015; Morte et al. 2018), we here found intriguing correlations in human astrocytes between SUR2 expression and thyroid hormone sensitive astrocyte gene products. In addition, astrocyte SUR2 expression was positively correlated with that of the main thyroid hormone transporter gene, SLCO1C1. An intersection between ABCC9/SUR2 signaling and thyroid hormone pathways in astrocytes seems biologically credible given their common physiologic roles in tuning metabolic status based on energy needs and availability.

HS-Aging is a dementia-associated condition associated with ABCC9 genetic variation (Nelson et al. 2014; Nelson et al. 2015a, b, c). A pathologic phenotype present in ~ 20% of autopsied dementia cases beyond age 85 years, HS-Aging is characterized by cell loss and astrocytosis in the hippocampal formation, associated with clinical features that mimic Alzheimer disease (Brenowitz et al. 2014; Murray et al. 2014; Nelson et al. 2013; Zarow et al. 2012). The SNV designated rs1914361 resides within an intron of ABCC9 and has been shown to be a risk-modifying allele for HS-Aging pathology in both White and Black cohorts (Dugan et al. 2021; Katsumata et al. 2023). The rs1914361 risk allele (guanine, versus the non-risk associated adenine) previously was shown to be associated with higher SUR2 expression in brain (according to bulk tissue analysis) and lower SUR2 expression in large blood vessels (Dugan et al. 2021). Here, we analyzed data that went beyond tissue-level mRNA levels and found that the risk allele of rs1914361 was associated with a trend for higher SUR2 expression in astrocytes, but not in other cell types, in the same brain samples.

It remains to be determined if ABCC9/SUR2 mechanisms can be targeted for medical management of individuals at risk for HS-Aging. There are grounds for optimism, since SUR2 agonists and antagonists are widely prescribed medications (Gribble and Reimann 2002; Miura and Miki 2003; Nichols 2023; Nichols et al. 2013). Repurposing one of these established drugs, a NIH-sponsored clinical trial is ongoing that is termed the SMArT-HS study (Safety and Modulation of ABCC9 Pathways by Nicorandil for the Treatment of Hippocampal Sclerosis of Aging; ClinicalTrials.gov Trial# NCT04120766). This randomized, double-blinded, placebo-controlled trial investigates nicorandil, a SUR2 agonist used for the clinical indications of angina pectoralis and congestive heart failure (Ahmed 2019), as a potential medicine for individuals at risk for HS-Aging, with results of the trial to be reported in late 2026.

There were limitations and pitfalls intrinsic to our study design evaluating human brain cell type-specific gene expression, with multiple sources of variation including perimortem and postmortem gene expression changes, possible cell type-specific differences in survival and gene expression robustness, and potential technical bias in transcript and protein quantification. The different transcript expression patterns of astrocyte subclusters may be partly attributable to developmental phenotypes (“young or old”-type astrocytes) or other influences such as anatomic compartments sampled. These considerations also pertain to the human vs murine SUR2 expression pattern comparison; the SEA-AD and UK-ADRC participants included in the current study were generally old (average age of death > 80 years). The spatial transcriptomics result (Fig. 2) constitutes single-donor findings that served to corroborate the multi-cohort transcriptomic analyses. Studies of younger persons’ brains are required to more fully characterize the impact of neurodevelopmental factors on human brain ABCC9/SUR2 expression. We also acknowledge that the UMAP charts indicated that one could assign the astrocyte subclusters in different ways. In future studies, the generalizability of our snRNA-seq and immunohistochemical studies need to be tested in additional, preferably larger and more diverse, cohorts. Finally, there are intrinsic challenges to clinical research, e.g. in vivo manipulations are necessarily limited, so human studies are sometimes characterized as being “descriptive”. However, we also note that animal models, for all their benefits, can only capture a fraction of the complexity of human neurobiology (Alexandra Lopes and Guil-Guerrero 2025; Granzotto et al. 2024), so, the different research paradigms must coexist and complement one another.

In conclusion, the current study analyzed ABCC9/SUR2 expression in human brain using different methodologies, with emphases on SUR2 transcript and SUR2 protein within cellular elements of the NVU. The regulation of ABCC9/SUR2 in the NVU is complex – ABCC9 yields different mRNA splice variants and different responses to genetic variation in distinct NVU cells. These results (summarized in schematic form in Fig. 8) provide new insights relevant to the molecular biology underlying human cerebral blood flow regulation.

Fig. 8.

Cell-type specific expression of SUR2 in human brain: schematic overview. In addition to expected SUR2 expression in pericytes and vascular smooth muscle cells (SMCs), there was appreciable SUR2 expression detected in human astrocytes. Astrocyte SUR2 expression was sensitive to germline ABCC9 HS-Aging risk allele status and there was evidence of a link between ABCC9/SUR2 expression and thyroid hormone signaling. The SUR2 + astrocyte cluster tended to be present among donors who lacked Alzheimer’s disease neuropathologic change (ADNC) as operationalized by higher Braak neurofibrillary tangle stages

Supplementary Information

Below is the link to the electronic supplementary material.

Terms

ADP

Adenosine diphosphate

AIMS

ABCC9-Related intellectual disability and myopathy syndrome

ATP

Adenosine triphosphate

CDR

Clinical Dementia Rating

CNS

Central nervous system

FFPE

Formalin-fixed, paraffin-embedded

HS-Aging

Hippocampal sclerosis of aging

IHC

Immunohistochemistry

K_ATP channel

Adenosine-triphosphate sensitive potassium channel

MTG

Middle temporal gyrus

NVU

Neurovascular unit

SEA-AD

Seattle Alzheimer’s Disease Brain Cell Atlas

SMC

Smooth muscle cell

snRNA-seq

Single-nucleus RNA sequencing

SNV

Single nucleotide variant

TBS

Tris-buffered saline

UK-ADRC

University of Kentucky Alzheimer’s Disease Research Center

UMAP

Uniform Manifold Approximation and Projection

Genes, Transcripts, and Proteins

ABCC9

ATP-binding cassette, sub-family C member 9

AQP4

Aquaporin 4

CD44

Cluster of differentiation 44

GFAP

Glial fibrillary acidic protein

NRXN1

Neurexin 1

SLC1A2

Solute Carrier Family 1 Member 2

SLCO1C1

Solute carrier organic anion transporter family member 1C1

SUR1

Sulfonylurea receptor 1

SUR2

Sulfonylurea receptor 2

TDP-43

Tar DNA binding protein of 43 kDa

VIM

Vimentin

Author Contribution

YK did statistical analyses and helped conceptualize and write the paper JM and KS did experiments specifically related to Fig. 2 and helped write the paper AL and WZ did experiments specifically related to Fig. 5 and helped write the paper QQ and DF did experiments specifically related to Fig. 3 YZ and SF and SA and AW and TL and DN and GJ and CN and RS and ACN and CR did experiments for Fig. 6 and helped write the paper All authors reviewed the manuscript.

Funding

The study was supported by NIH grants P30 AG072946, R01 NS118584, RF1 AG082339, P01 AG078116, and R01 AG076932. The BRaIN laboratory is supported by the NIH grants for the UW Alzheimer's Disease Research Center (P50 AG005136 and P30 AG066509) and the Adult Changes in Thought Study (U01 AG006781 and U19 AG066567). This study is also supported by NIA grant U19 AG060909.

Data Availability

No datasets were generated or analysed during the current study.

Declarations

Ethics Approval

The University of Kentucky Institutional Review Board approved the use of autopsy material from the UK-ADRC (UK IRB # 44009).

Consent to Participate

Informed consent was obtained from research participants (and/or their caregivers) as stipulated in University of Kentucky IRB #44009.

Permission to Reproduce Material from other Sources

Not applicable.

Clinical Trial Registration

Not applicable.

Clinical trial number

Not applicable.

Conflict of interest

The authors declare no competing interests.