Cellular signatures of melanocortin pathway genes across the locus coeruleus

Abstract

Obesity and Alzheimer’s disease (AD) are epidemiologically associated. The locus coeruleus (LC)—the brain’s primary and most significant source of norepinephrine—is one of the earliest sites of neurodegeneration in AD. The LC participates in feeding behavior through connections with the hypothalamus. The cellular composition of the LC has been characterized at single-cell resolution. However, the constituent cellular signatures of genes related to energy homeostasis—such as the melanocortin pathway genes—in the LC are unclear. We performed single-nucleus RNA sequencing and spatial transcriptomics (Visium) in the human LC, and HiPlex RNAscope in the LC of mice. The melanocortin pathway gene MRAP2 was expressed in the majority of DBH neurons across the LC. Mrap2 was also co-expressed with AD-associated genes such as App, Psen1, Psen2, and Sorl1. More than 20% of Dbh neurons in the LC were positive for Mrap2, App, Psen1, and Psen2. Mrap2 is expressed in the central nervous system and modulates the trafficking and signaling of all five G-protein coupled receptors (GPCRs) of the melanocortin receptor family: Mc1r, Mc2r, Mc3r, Mc4r, and Mc5r. In mice, among the melanocortin receptors, Mc5r showed the highest co-expression with Mrap2, accounting for 17.9% of Mrap2-positive cells, followed by Mc2r with 10.9% of Mrap2-positive cells. Mc1r, Mc3r, and Mc4r showed very limited co-expression with Mrap2. Our study reveals that many Mrap2-positive cells do not express any melanocortin receptor genes, warranting future studies into metabolically relevant GPCRs downstream of MRAP2 in the LC. In summary, our study characterizes melanocortin molecular substrates in the human and mouse LC and highlights MRAP2 as a potential link between pathways of energy homeostasis and neurodegeneration.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40478-026-02287-x.

Introduction

Epidemiological and experimental studies have identified links between obesity and Alzheimer’s disease (AD), with obesity increasing the risk of AD [1–5]. Disruption of the locus coeruleus (LC) system is associated with neurodegenerative diseases such as AD [6]. The LC is a small brainstem nucleus located within the dorsal wall of the rostral pons, in the lateral floor of the fourth ventricle. It is the brain’s major source of norepinephrine (NE). LC NE neurons innervate both cortical and subcortical structures including the cerebellum, basal telencephalon, thalamus, amygdala, and hypothalamus [7]. Accordingly, LC NE neurons play a role in many physiological and behavioral processes including arousal, attention, memory, cognition, stress response, activity of the sympathetic nervous system, and energy homeostasis [8–14].

The LC is among the earliest brain regions to exhibit tau pathology in AD, with progressive loss of LC neurons evident as the disease advances [15–17]. Aberrant accumulation of hyperphosphorylated tau has been reported in LC neurons in addition to its presence in the hypothalamic tuberomammillary nucleus (TMN) [18]. While the most recognizable feature of AD is memory impairment, non-cognitive symptoms of AD such as sleep disorders, autonomic dysfunction, and endocrine alterations may be influenced by disruption of communication between the LC and hypothalamus [19, 20]. The hypothalamus is the “master regulator” of homeostatic functions, including energy homeostasis, sympathetic and parasympathetic activities, and sleep–wake cycles [21].

Both inputs to the LC from the hypothalamus and outputs from the LC to the hypothalamus have been reported [22]. Unidirectional projections from the TMN to the LC mediate the sleep–wake cycle [23–26]. Bidirectional neuronal projections to the LC arising from the ventrolateral preoptic nucleus (VLPO), suprachiasmatic nucleus (SCN), dorsomedial nucleus (DMH), and paraventricular nucleus (PVN) regulate arousal and the sleep–wake cycle [24, 27–29]. Additionally, there are afferent connections to the LC from the PVN of the hypothalamus involved in the coordination of the stress response [30–32]. Of note, there are bidirectional connections between the lateral hypothalamus (LH) and LC that regulate arousal and wakefulness, and may also play a role in the NE-mediated suppression of feeding via beta-adrenoceptors [27, 33–37].

LC NE activity, assessed via in vivo fiber photometry calcium imaging, optogenetics, and chemogenetics, is enhanced during food approach and progressively suppressed during food consumption, suggesting that LC responses to food are modulated by satiety state. Visual-evoked NE activity in LC neurons is attenuated in sated mice, suggesting that satiety modulates NE encoding of multiple behavioral states. Furthermore, food intake is reduced by brief or longer durations of NE activation in LC neurons and stimulation of LC neurons projecting to the LH suppresses feeding in mice [14]. While hypothalamic feeding circuitry is interconnected with the NE neurons of the LC, further investigation of the molecules that mediate this feeding response in LC NE neurons is necessary.

Previous studies on metabolically relevant G-protein coupled receptors (GPCRs), including hypocretin receptor 1 (OX1R), growth hormone secretagogue receptor (GHSR), and melanin concentrating hormone receptor 1 (MCHR1), revealed their expression patterns in the LC and function conveyed by LC NE neurons [38–46]. Furthermore, hypothalamic and brainstem adrenocorticotrophic hormone (ACTH)- and alpha-melanocyte-stimulating hormone (α-MSH)-containing fibers are present in the rat LC [47, 48]. ACTH is a ligand for melanocortin 1 receptor (MC1R) and melanocortin 2 receptor (MC2R) [49, 50]. α-MSH, as derived from proopiomelanocortin (POMC), is a ligand at MC1R, melanocortin 3 receptor (MC3R), melanocortin 4 receptor (MC4R), and melanocortin 5 receptor (MC5R) [51, 52]. The role of α-MSH as an anorexigenic peptide at MC4R has been studied in the hypothalamus [53]. Expression of Mc4r has been reported in the LC of rodents [54–58]. The melanocortin receptor genes and their ligands are part of the leptin-melanocortin pathway, which also includes the leptin receptor (LEPR) and the melanocortin 2 receptor accessory protein 2 (MRAP2) [59, 60]. In the hypothalamus, neurons of the leptin-melanocortin pathway participate in energy homeostasis by regulating food intake and energy expenditure [61]. The adipokine leptin has central effects mediated by the hypothalamic leptin-melanocortin pathway and its projections to various parts of the brain. Hyperleptinemia and obesity have been associated with AD risk [62]. Whether the effects of NE neurons on feeding are conveyed by the leptin-melanocortin pathway is unknown [63–65]. Further molecular characterization of the LC is required to identify how NE neurons contribute to the pathways of energy homeostasis [63–66].

Genetic, molecular, and clinical evidence established the roles of amyloid beta precursor protein (APP), presenilin 1 (PSEN1), presenilin 2 (PSEN2), sortilin related receptor 1 (SORL1), apolipoprotein E (APOE), and triggering receptor expressed on myeloid cells 2 (TREM2) in AD pathogenesis [67–72]. APP mediates proteolytic processing of amyloid beta (Aβ) peptides that may accumulate in amyloid plaques, a pathological hallmark of AD. Mutations in APP cause autosomal dominant early-onset AD, providing direct causal evidence for its role in disease [68–70, 73]. PSEN1 and PSEN2 encode presenilin-1 and presenilin-2, the catalytic components of the γ-secretase complex that cleaves APP. Pathogenic mutations in PSEN1 and PSEN2 alter Aβ production and cause familial early-onset AD [68, 70, 74]. SORL1 encodes a neuronal sorting receptor involved in APP trafficking. Reduced SORL1 expression or loss-of-function variants shift APP processing toward the amyloidogenic pathway, increasing Aβ generation. Genetic variants in SORL1 have been associated with both early- and late-onset AD [75–77]. APOE is the strongest genetic risk factor for late-onset AD. APOE influences Aβ aggregation and clearance, lipid transport, and neuroinflammation, making it a central modulator of AD susceptibility [70, 72, 78]. TREM2 encodes a microglial receptor involved in immune response and phagocytosis. Rare coding variants in TREM2 substantially increase the risk of late-onset AD, highlighting the importance of microglial dysfunction and neuroinflammation in disease progression [79–82]. Molecular characterization is warranted to examine co-expression of melanocortin and these AD-associated genes in the cellular populations of the LC.

Here, we characterize the cellular and molecular components of the melanocortin pathway within the LC in humans and mice. Using single-nucleus RNA sequencing (snRNA-seq) and spatial transcriptomics in human postmortem LC tissue, in conjunction with RNAscope HiPlex across the mouse LC, we achieved single-cell resolution of melanocortin gene expression. Our findings reveal substantial co-expression of melanocortin and AD-associated genes in LC cellular populations.

Materials and methods

Human postmortem samples

Postmortem human metencephalon tissue from deidentified neurotypical donors was obtained from the New York Brain Bank, Columbia University, New York. Briefly, sequential transverse slices of the hemi-brainstem of 0.3 cm thickness were obtained. The LC was identified via its distinct dark blue pigmentation. Slices were frozen in dry ice, transferred to a sealed bag, and stored at − 80 °C until further use. For snRNA-seq, we sourced the LC region from a 74-year-old female donor. For spatial transcriptomics, slices of the LC region were sourced from a 76-year-old female donor.

Single nuclei isolation for snRNA-seq

The nuclei isolation protocol was adapted from 10 × Genomics. To isolate the region of interest, slices were briefly transferred to a Petri dish on ice, and great care was taken to include all of the visible LC using a 2 mm diameter punch. The punched tissue was immediately transferred to a 1.5 ml microcentrifuge tube containing 500 µl of pre-chilled Lysis Buffer (10 mM Tris–HCl, Millipore-Sigma #T2194; 10 mM NaCl, Millipore-Sigma #59222C; 3 mM MgCl₂, Millipore-Sigma #M1028; 0.1% Nonidet P40 Substitute, Millipore-Sigma #74,385), homogenized by gently moving up and down a Pellet Pestle 15 times and centrifuged for 3 s at 300 g at room temperature. The supernatant was collected in a new 1.5 ml microcentrifuge tube and centrifuged for 5 min at 850 g at 4 °C. The supernatant was removed, and the pellet was washed twice in 1 ml of Nuclei Wash & Resuspension Buffer (0.04% BSA, Sigma-Millipore #SRE0036, in PBS, Gibco # 10010023). After the second wash, the nuclei suspension was passed through a 40 µm Flowmi Cell Strainer (Bel-Art #H13680-0040) and centrifuged for 5 min at 300 g at 4 °C. The nuclei pellet was resuspended in Nuclei Wash & Resuspension Buffer and immediately loaded onto the 10 × Chromium instrument (Single Cell Analysis Core at Columbia University).

Sequencing libraries were generated using 10 × Genomics Chromium Single-Cell 3’ Reagent Kit v3.

Human LC tissue processing for 10 × Genomics Visium spatial transcriptomics

Frozen samples were embedded in O.C.T. compound (Thermo Fisher Scientific, 23730571) and placed in dry ice to flash-freeze. A total of five 10-µm thick sections (2 consecutive sections from the rostral slice; 2 consecutive sections from the caudal slice) were prepared using a CryoStar NX70 cryostat (Epredia) at − 16 °C, in an RNase-free environment and mounted onto positively charged slides (Superfrost Plus, Fisherbrand #22-037-246). The sections were then stored at − 80 °C until use.

Spatial transcriptomics was performed using the 10 × Genomics Visium CytAssist platform (version 2) with human whole-transcriptome (WTA) probe chemistry (Visium V5 Slide—FFPE v2; Visium Human Transcriptome Probe Set v2.0, targeting 18,061 human genes), following the manufacturer’s Visium CytAssist FFPE v2 protocol. For each rostro–caudal level, the two consecutive 10 µm sections were placed on the same Visium slide: the rostral pair (Upper Rostral, Lower Rostral) on slide V52L19-072 (positions A1 and B1), and the caudal pair (Upper Caudal, Lower Caudal) on slide V43L24-365 (positions A1 and D1).

snRNA-seq analysis of human LC

Raw gene–barcode matrices (“filtered_feature_bc_matrix”) from two human LC sections (two Chromium reactions per section; LC001A–B, LC002A–B) were imported into R (v4.3.2) and processed with Seurat (v5.0.1) [83]. For each Chromium reaction, we generated a Seurat object, computed standard quality-control (QC) metrics: number of detected genes (nFeature_RNA), total UMI counts (nCount_RNA), and percentage of mitochondrial reads (percent.mt). The four Chromium reactions were then merged, yielding 2,347 nuclei before QC.

For initial processing, data were normalized, highly variable genes were selected, and expression values were scaled. Principal component analysis (PCA) was performed on variable genes, and the four Chromium reactions were integrated using Seurat’s reciprocal PCA (RPCA)–based layered integration workflow. A Uniform Manifold Approximation and Projection (UMAP) embedding was computed from the integrated reduction (40 dimensions), and a shared nearest-neighbor graph was built in the same space (FindNeighbors, 40 dimensions, k.param = 20). Unsupervised clustering was performed across a range of resolutions (0.1–1.5; FindClusters).

Broad cell types were annotated based on canonical marker expression. NE neurons were defined by dopamine beta-hydroxylase (DBH), tyrosine hydroxylase (TH) and solute carrier family 6 member 2 (SLC6A2) together with pan-neuronal markers stathmin 2 (STMN2) and synaptotagmin 1 (SYT1). Additional neuronal clusters lacked DBH/TH/SLC6A2, but they expressed neuronal markers. Oligodendrocytes were identified by myelin-associated glycoprotein (MAG), myelin basic protein (MBP), myelin oligodendrocyte glycoprotein (MOG), proteolipid protein 1 (PLP1) and related genes; oligodendrocyte precursor cells (OPCs) by platelet-derived growth factor receptor alpha (PDGFRA) and epidermal growth factor receptor (EGFR); astroependymal cells by glial fibrillary acidic protein (GFAP), gap junction protein alpha 1 (GJA1), aquaporin 4 (AQP4), forkhead box J1 (FOXJ1) and coiled-coil domain containing protein 153 (CCDC153); microglia by complement C1q B chain (C1QB) and allograft inflammatory factor 1 (AIF1). We also identified a low-quality cluster characterized by low gene counts and high mitochondrial percentages.

To remove low-quality nuclei, we implemented a data-driven, batch-aware QC strategy using custom R functions. QC batches were defined as sample × broad cell type (with small populations such as OPCs, astrocytes, and microglia grouped across sections). For each batch, kernel density estimates of log2-transformed nCount_RNA and nFeature_RNA were used to identify minima separating low-quality from high-quality nuclei, aided by Otsu-derived thresholds and, where applicable, Gaussian mixture modeling. Per-batch lower and upper thresholds for nCount_RNA and nFeature_RNA were then applied to the integrated object (Supplementary Table 1), and the low-quality cluster was explicitly removed. After QC, 2,071 of 2,347 nuclei (88.2%) were retained.

Cells passing QC were reprocessed using the same Seurat workflow: data were normalized, variable features were re-identified and scaled, sections were reintegrated using RPCA, and UMAP and graph-based clustering were recomputed. Final broad cell-type identities (NE neurons, other neurons and non-neuronal populations) were assigned based on UMAP visualizations and dot plots of canonical markers and were carried forward into downstream analyses.

snRNA-seq meta-analysis of human LC datasets

To place our LC dataset in a broader human context, we performed a meta-analysis integrating our snRNA-seq data with two published human pons datasets from Weber et al. [64] and Siletti et al. [66]. The Weber dataset comprised 20,191 neuronal and non-neuronal nuclei from the human LC region. For the Siletti dataset, six annotated pons sections were first screened for NE neurons by examining DBH expression across pontine nucleus (PN), afferent and efferent cranial nerve nuclei, pontine reticular formation (PnRF), parabrachial nuclei (PnRF–PB) and dorsal tegmental nucleus. A DBH-expressing neuronal cluster was detected in the PnRF–PB section, and only this section (24,155 neuronal and non-neuronal nuclei) was used for integration. These datasets were combined with the 2,071 QC-passed nuclei from our dataset, yielding a meta-dataset of 46,417 nuclei (Supplementary Table 2).

All meta-analysis steps were performed in R (v4.3.2) using Seurat (v5.0.1). Seurat objects from this study (Basak), Weber and Siletti were merged into a single object, with per-sample layers defined to enable multi-sample integration. Counts were normalized, highly variable genes were identified and data were scaled, followed by PCA.

Datasets were integrated using Seurat’s multi-dataset integration framework. We used RPCA-based integration to generate a shared low-dimensional embedding, from which a UMAP representation was computed (30 dimensions). A shared nearest-neighbor graph was then constructed on this RPCA-integrated space (FindNeighbors, 30 dimensions, k.param = 20), and unsupervised clustering was performed over a range of resolutions, with resolution 0.6 selected for final annotations.

Cell types in the integrated meta-dataset were assigned by examining dot plots and feature plots of canonical markers, using the same marker sets as for the Basak dataset. NE neurons were identified by DBH, TH and SLC6A2 expression; additional neuronal clusters expressed pan-neuronal markers (STMN2, SYT1) without DBH/TH/SLC6A2. Oligodendrocytes, OPCs, astrocytes, ependymal cells, microglia, and vascular cells were annotated based on established lineage markers. These integrated cell-type labels were used in subsequent meta-analyses.

Neuronal subtype analysis in the integrated snRNA-seq meta-dataset

To resolve neuronal diversity within the integrated LC meta-dataset, we first subset all nuclei annotated as neurons from the integrated object, yielding 33,375 neuronal nuclei across Basak, Weber, and Siletti. The RNA assay was retained, and per-sample layers were defined. We then applied Seurat’s standard preprocessing and integration workflow: counts were normalized, highly variable genes were identified, data were scaled and PCA was performed, followed by RPCA-based multi-sample integration to generate a neuron-specific integrated embedding. A UMAP was computed from this integrated space (30 dimensions), and a shared nearest-neighbor graph was constructed (FindNeighbors on 30 RPCA dimensions, k.param = 10). Graph-based clustering was performed over a range of resolutions, and a high-resolution setting (resolution 1.5) was selected for downstream neuronal subtype analyses, yielding 68 transcriptionally distinct neuronal clusters.

To functionally annotate neuronal subtypes, we focused on a curated list of G protein–coupled receptor and ligand genes derived from the IUPHAR Guide to Pharmacology (GtoP) database [84, 85]. Cluster markers were computed with Seurat’s FindAllMarkers function restricted to this curated gene set, and cluster-enriched genes were defined based on effect size and specificity (for example, avg_log2FC > 0.5 with higher expression in the index cluster and low background expression in other clusters). For each cluster, the top two discriminative genes from this curated list were selected and visualized using dot plots to summarize neuropeptide and receptor signatures (Supplementary Fig. 1 d). Cluster labels were then derived by concatenating these two marker genes, with numeric suffixes applied only when the same gene pair occurred in more than one cluster. This strategy yielded three DBH⁺ NE neuron subtypes: SLC6A2/DBH, DBH/SLC6A2, and SLC6A2/MET (MET proto-oncogene, receptor tyrosine kinase), which were used for downstream analyses.

Visium spatial transcriptomics analysis of human LC

For each Visium section, 10 × Genomics filtered feature–barcode matrices were imported into R (v4.3.2) using the Load10X_Spatial function in Seurat (v5.0.1). Standard spot-level QC metrics were computed, including total UMI counts (nCount_RNA), number of detected genes, and percentage of mitochondrial reads. To remove off-tissue or low-quality regions at the image periphery, we applied section-specific cropping based on Visium spot coordinates, visually matching areas with very low nCount_RNA to their positions in the coordinate space and excluding these spots by filtering on image row/column coordinates. Cropped objects from the four LC sections were then merged into a single Seurat object (12,495 spots in total).

Integration with published Weber LC Visium data and spatial cluster annotation

To place our spatial LC data in a broader context, we integrated our in-house Visium sections with published LC Visium data from Weber et al. [64] comprising 20,380 spots (Supplementary Table 3). Both datasets were merged into a single Seurat object (32,875 spots), with per-section layers defined. The combined LC Visium dataset was processed with SCTransform (default settings) and PCA, followed by Seurat’s RPCA-based multi-sample integration on SCT-normalized data to generate a joint Visium embedding. A UMAP was computed from this integrated space (30 dimensions), and a shared nearest-neighbor graph was constructed (FindNeighbors on 30 integrated dimensions, k.param = 10). Clustering was performed over resolutions 0.1–1.5, and resolution 0.3 was selected to obtain a manageable number of spatial clusters for biological interpretation.

Because Visium spots do not correspond to single cells and can capture mixtures of neurons, glia, and vascular cells, we interpreted “neuronal” clusters as spots enriched for neuronal transcripts rather than purely single-cell neuronal profiles. Potential contributions from neighboring astrocytes, microglia, and other non-neuronal cell types should therefore be kept in mind when interpreting these annotations. Cluster annotation for the integrated Visium dataset was guided by canonical marker expression, including noradrenergic (DBH, TH, SLC6A2), pan-neuronal (STMN2, SYT1) and glial/vascular lineage markers (for example, MAG, MOG, PLP1 for oligodendrocytes; AQP4 for astrocytes; FOXJ1, CCDC153, calcyphosine (CAPS) for ependymal cells; actin alpha 2 smooth muscle (ACTA2), transgelin (TAGLN) for mural/vascular smooth muscle cells; collagen type I alpha 1 chain (COL1A1), collagen type III alpha 1 chain (COL3A1), lumican (LUM), decorin (DCN) for vascular leptomeningeal cell (VLMC)-like cells; and hemoglobin subunit alpha 2 (HBA2), hemoglobin subunit beta (HBB) for erythrocytes), supplemented by a curated neurotransmitter/neuropeptide gene list. Based on these expression patterns, clusters were manually assigned to broad spatial cell classes: NE neurons, serotonergic neurons, tachykinin precursor 1 (TAC1)/CRH neurons, adenylate cyclase activating polypeptide 1 (ADCYAP1)/VGF nerve growth factor inducible (VGF) neurons, proenkephalin (PENK) neurons, parvalbumin (PVALB)/sodium voltage-gated channel beta subunit 4 (SCN4B) neurons, GABAergic neurons, glutamatergic neurons, oligodendrocytes, astrocytes, ependymal cells, vascular cells, and two unclassified clusters with no prominent marker signature (Supplementary Fig. 2c).

Animals

RNAscope in situ hybridization experiments were performed in 5-week-old male C57BL/6NTac mice. All mice were obtained from Taconic Biosciences. All animal care and experimental procedures followed the US National Institutes of Health Animal Care and Use guidelines and were approved by the Columbia University Animal Care and Use Committee (IACUC). Animals were sacrificed according to Columbia University regulations using the appropriate protocol (#AC-AABF6552).

Housing and diets

Mice were housed at 22–24 °C constant temperature with a regular 12-h light/12-h dark cycle (lights were turned off at 7 pm), with no more than 5 adult animals per cage and ad libitum access to Purina 5058 chow diet and water.

Mouse brain dissection and tissue preparation

Mice were sacrificed via cervical dislocation followed by decapitation. Brains were immediately removed, frozen in dry ice, subsequently embedded in O.C.T. compound (Thermo Fisher Scientific, 23,730,571) and placed in dry ice to flash freeze. All brains were immediately stored at − 80 °C in an airtight container until further use.

For in situ hybridization, fresh-frozen brains were cut into 20 µm-thick serial coronal sections (resulting in 54–57 sections for the LC) using a CryoStar NX70 cryostat (Epredia) at − 16 °C, collected on Superfrost Plus microscope slides (3 sections per slide), dried at 20 °C for 1.5 h and stored at − 80 °C in an airtight container until use.

The first section was collected approximately at Bregma − 5.1 mm to Bregma − 6.1 mm, spanning the LC from rostral to caudal.

RNAscope HiPlex in situ hybridization

Given that snRNA-seq and Visium have limited sensitivity, we moved forward with the highly sensitive.

RNAscope technology. It allows for the detection of single RNA molecules and can more reliably assess lowly.

expressed genes which may escape detection in snRNA-seq or Visium.

One series, consisting of 57 fresh-frozen brain sections with a thickness of 20 µm, covering the entire mouse LC from rostral to caudal, was pre-treated and stained following the RNAscope HiPlex 12 Reagents Kit (488, 550, 650) (ACD # 324,108) user manual protocol (Advanced Cell Diagnostics (ACD) Document #324,419-USM). Briefly, sections were taken from − 80 °C and put directly in a 4% paraformaldehyde solution for 60 min, rinsed in PBS, and dehydrated in increasing ethanol concentrations. Sections were dried for 10 min followed by 30 min of Protease IV (ACD #322,336) treatment at room temperature to permeabilize the cells and subsequent washed with PBS. The pre-warmed probes mixture was added, and the probes were hybridized to the targeted genes at 40 °C for 2 h in a HybEZ II oven (ACD). The following probes were used: Mc4r (ACD #319,181-T1), Dbh (ACD #407,851-T2), Lepr (ACD #471,171-T3), Mrap2 (ACD #1,067,291-T4), paired box 7 (Pax7, ACD #314,181-T5), Psen1 (ACD #451,011-T6), Psen2 (ACD #450,981-T7), Trem2 (ACD #1,082,891-T8), corticotropin releasing hormone (Crh, ACD #316,091-T9), App (ACD #578,001-T10), Sorl1 (ACD #1,082,901-T11), Apoe (ACD #590,131-T12). The sections were washed in 1X wash buffer (ACD #310,091), and amplification reagents AMP1-AMP3 were incubated in succession at 40 °C with wash buffer steps in between. In the first imaging cycle, fluorophores T1–T3 were applied and hybridized with the tissue at 40 °C for 15 min. The sections were then washed in wash buffer, counterstained with DAPI solution for 30 s, and coverslips were directly mounted with ProLong Gold Antifade Mountant (Fisher Scientific #P36930). Slides were kept dark at 4 °C and images were captured in the following days. Fluorophores were then cleaved with freshly prepared 10% cleaving solution (stock, ACD #324,399) twice to ensure the fluorophores were removed entirely. This process was repeated for 3 more rounds to image fluorophores T4–T12. Sections were counterstained with DAPI every other round. Target genes and channels for each round are listed (Table 1a).

Table 1.

Target genes for HiPlex RNAscope® Assay. (a) Channel setup for RNAScope HiPlex co-hybridization experiment. (b) Channel setup for RNAScope Multiplex Fluorescent v2 co-hybridization experiment

Gene	Channel	ISH Detection
Round 1
Mc4r	C1-T1	488-Green
Dbh	C2-T2	550-Orange
Lepr	C3-T3	650-Red
Round 2
Mrap2	C1-T4	488-Green
Pax7	C2-T5	550-Orange
Psen1	C3-T6	650-Red
Round 3
Psen2	C1-T7	488-Green
Trem2	C2-T8	550-Orange
Crh	C3-T9	650-Red
Round 4
App	C1-T10	488-Green
Sorl1	C2-T11	550-Orange
Apoe	C3-T12	650-Red

RNAscope multiplex fluorescent V2 assay

One series, consisting of 54 fresh-frozen brain sections with a thickness of 20 µm, covering the entire mouse LC from rostral to caudal, was pre-treated and stained according to the RNAscope Fluorescent Multiplex Reagents Kit (ACD #320,850) user manual protocol (ACD Document #320,293-USM). Briefly, sections were taken from − 80 °C and put directly in a pre-chilled 4% paraformaldehyde solution for 15 min at 4 °C and dehydrated in increasing concentrations of ethanol. Sections were dried for 5 min and then incubated with Protease IV at room temperature for 30 min. The pre-warmed probes mixture was added, and the probes were hybridized to the targeted genes at 40 °C for 2 h in a HybEZ II oven. The following probes were used: Mrap2 (ACD #1,067,291-C1), Mc3r (ACD #412,541-C2), Mc5r (ACD #478,381-C3), Mc1r (ACD #474,391-C2), and Mc2r (ACD #318,891-C3). Four consecutive rounds of amplifications (AMP1-AMP4) at 40 °C were followed by counterstaining with DAPI solution for 30 min and mounting of coverslips with ProLong Gold Antifade Mountant. Slides were kept dark at 4 °C and images were captured in the following days. Target genes and channels for each round are listed (Table 1b).

Digital image acquisition

Tiled z-stack images were acquired using the Zeiss LSM 710 laser scanning confocal microscope (Zeiss, Oberkochen, Germany). Images were obtained with 20X and 63X Zeiss Plan-Apochromat dry and oil objectives, respectively. All images were obtained in 1024 × 1024-pixel format. Zeiss ZEN software was used to gather the maximum intensity projection images.

For the sections processed using RNAscope HiPlex, 4 cycles of images were performed to image 12 RNA species, utilizing DAPI as reference in each round.

Image analysis

Images were analyzed using HALO (Indica Labs). For the sections processed using RNAscope HiPlex, the images from all rounds of staining were registered to each other to generate a 12 plex image using HALO software. In brief, a DAPI image from one of the rounds of imaging was used as the reference image. DAPI images from all other rounds of imaging were registered to this reference image, generating a transformation matrix of coordinate conversions that was applied to the remaining images from each imaging round to create one, unified coordinate system for all images from all rounds.

The HALO random forest classifier was trained for each section to identify the LC region by selecting two Dbh⁺ regions within the LC and two Dbh⁻ regions outside the LC (Supplementary Fig. 4). For the analysis, cell segmenting and signal thresholds were independently established on each section using the software for all genes of interest. The number of RNA molecules per cell for each gene was detected for every cell within our annotation layer of each section. All analysis matrixes were exported to Microsoft Excel.

Matrices from each section were further analyzed to identify cells expressing each gene of interest as well as the average number of molecules per cell. Positive and negative thresholds for each gene of interest were established by selecting a region outside the Dbh-LC annotation layer and within the Dbh-LC annotation layer from the left and right sides of the LC from sections 34, 35, and 36 and calculating the minimum average number of molecules per cell and the maximum average number of molecules per cell in the selected regions (Supplementary Fig. 5a–c). Thresholds were manually set based on evaluation of the average minimum value. The following thresholds for positive cells were established: Dbh⁺ ≥ 1, Lepr⁺ ≥ 2, Mc4r⁺ ≥ 2, Pax7⁺ ≥ 1, Psen1⁺ ≥ 3, Mrap2⁺ ≥ 2, Psen2⁺ ≥ 3, Trem2⁺ ≥ 3, Crh⁺ ≥ 4, App⁺ ≥ 7, Sorl1⁺ ≥ 2, and Apoe⁺ ≥ 4. The following thresholds for negative cells were established: Dbh⁻ = 0, Lepr⁻ ≤ 1, Mc4r⁻ ≤ 1, Pax7⁻ = 0, Psen1⁻ ≤ 2, Mrap2⁻ ≤ 1, Psen2⁻ ≤ 2, Trem2⁻ ≤ 2, Crh⁻ ≤ 3, App⁻ ≤ 6, Sorl1⁻ ≤ 1, and Apoe⁻ ≤ 3 (Supplementary Fig.  5 d).

Statistical analysis

Statistical analysis was performed using GraphPad Prism 10.0.0. Comparisons between two normally distributed groups were performed by simple two-tailed unpaired student’s t-test. All p (alpha) values < 0.05 were considered statistically significant.

Results

Single-nucleus transcriptomics and meta-analysis of the human LC

We first performed snRNA-seq on two anatomically defined human LC sections using the 10 × Genomics Chromium platform (Fig. 1a). After data-driven, batch-aware QC (Supplementary Table 1), 2,347 nuclei were retained for downstream analysis. UMAP visualization of this dataset revealed six major classes comprising 463 NE neurons, 897 other neurons, 493 oligodendrocytes, 56 OPCs, 131 astroependymal cells and 31 microglia (Fig. 1b).

Fig. 1.

Single-nucleus transcriptomics of the human LC a Schematic of the experimental workflow for snRNA-seq from the human LC region, including tissue dissection, single-nucleus suspension preparation, and processing with the 10 × Genomics Chromium platform. b UMAP representation of nuclei from this study colored by broad cell-type annotation, revealing NE neurons, other neurons, oligodendrocytes, oligodendrocyte precursor cells (OPCs), astroependymal cells and microglia. The accompanying table summarizes cell numbers per class. c UMAP of the integrated snRNA-seq meta-analysis combining Basak (this study; B), Weber (W) and Siletti (S) datasets, colored by cell type. The table lists cell numbers for each major class across studies, demonstrating robust cross-dataset integration. d Dot plot of canonical marker genes supporting the cell-type assignments in c (for example, DBH/SLC6A2 in NE neurons, MOG/PLP1 in oligodendrocytes). e Neuronal nuclei from c were subsetted and reintegrated. DBH⁺ NE neuron subtypes (SLC6A2/DBH, DBH/SLC6A2 and SLC6A2/MET) are highlighted on the UMAP, and their abundances per dataset are summarized in the table. f Heatmap showing average expression of key NE neuron markers and additional literature-curated and differentially expressed genes across the three DBH⁺ subtypes, illustrating shared core NE identity and subtype-specific transcriptional signatures. g Dot plot depicting expression of melanocortin pathway components and AD–related genes across NE neuron subtypes and major non-neuronal cell classes

To place this dataset in a broader human context, we next integrated our (Basak) LC nuclei with two recently published snRNA-seq datasets from the human pons: the Weber et al. dataset [64], centered on the LC region, and the Siletti et al. dataset [66] containing a parabrachial nuclei section that includes DBH⁺ NE neurons. The resulting meta-dataset comprised 46,417 nuclei (Supplementary Table 2). After Seurat RPCA-based integration, cells from all three studies were extensively intermingled in the UMAP space, indicating effective batch correction (Supplementary Fig. 1a). Within this integrated embedding, major cell classes again segregated into coherent clusters, and all three datasets contributed to each lineage, including NE neurons (346 Basak, 306 Weber and 113 Siletti), other neurons (1,012; 16,201; 15,397), oligodendrocytes (491; 1,842; 4,945), OPCs (56; 288; 1,168), astrocytes (122; 400; 1,510), ependymal cells (2; 380; 5), microglia (36; 277; 763) and vascular cells (6; 497; 254) (Fig. 1c). QC metrics—total UMI counts, number of detected genes and mitochondrial RNA fraction—showed largely overlapping distributions across Basak, Weber, and Siletti within each major class, with higher mitochondrial read fractions in neurons of Basak and NE neurons from Weber, and higher detected genes and UMI counts in neuronal clusters from Siletti (Supplementary Fig. 1b).

We then focused on neuronal diversity within this integrated LC meta-dataset. Subsetting all nuclei annotated as neurons yielded 33,375 neuronal nuclei, which were reintegrated and reclustered, resulting in 68 transcriptionally distinct neuronal subtypes (Supplementary Fig. 1c). To functionally annotate these subtypes, we restricted cluster-marker discovery to a curated list of GPCRs and ligands [84, 85]. Each cluster was labeled by a discriminative gene pair that captured its neuromodulatory identity (Supplementary Fig. 1 d).

Within the integrated neuron dataset, three DBH⁺ NE neuron subtypes were resolved and designated by their defining gene pairs: SLC6A2/DBH, DBH/SLC6A2 and SLC6A2/MET (Fig. 1e). These subtypes were present in all three datasets, although with differing abundances: for SLC6A2/DBH we observed 256 Basak, 285 Weber and 65 Siletti nuclei; for DBH/SLC6A2, 112, 177 and 5 nuclei; and for SLC6A2/MET, 3, 43 and 43 nuclei, respectively (Fig. 1e). A heatmap of average expression confirmed that all three subtypes shared a core noradrenergic transcriptional program with high levels of DBH, SLC6A2 and TH, but differed in a set of additional genes curated from the literature, including neuropeptide Y (NPY), GAL, glutathione peroxidase 3 (GPX3), delta like non-canonical Notch ligand 1 (DLK1) and cholinergic receptor nicotinic alpha 3 subunit (CHRNA3) [22, 86, 87], as well as from differential expression analyses (Fig. 1f). For instance, NPY expression was particularly enriched in the DBH/SLC6A2 subtype, whereas 5-hydroxytryptamine receptor 2 C (HTR2C) and syntrophin gamma 1 (SNTG1) were relatively higher in DBH/SLC6A2. LEPR, brain derived neurotrophic factor (BDNF), transmembrane channel like 3 (TMC3) and inversin (INVS) were higher in SLC6A2/MET. GAL, CART prepropeptide (CARTPT), GPX3, DLK1 and CHRNA3 showed subtype-biased expression across SLC6A2/DBH and DBH/SLC6A2. Together, these signatures indicate that human LC NE neurons comprise transcriptionally specialized populations beyond a shared catecholaminergic core.

Finally, we examined the distribution of melanocortin pathway components and AD-related genes across neuron subtypes and major non-neuronal classes (Fig. 1g, Supplementary Fig. 1e). LEPR expression was concentrated in the SLC6A2/MET subtype, with additional signal in other neuronal subtypes and in vascular cells. MC1R was mainly detected in the SLC6A2/MET subtype, whereas MC2R, MC3R, MC4R, and MC5R transcripts were generally rare and sparsely detected. MRAP2, a key melanocortin receptor accessory protein, was enriched in SLC6A2/MET and SLC6A2/DBH subtypes, as well as in additional neuronal subtypes and ependymal cells. AD-associated genes displayed broader patterns: APP and PSEN1 were widely expressed across neuronal and non-neuronal populations, with PSEN1 showing relative enrichment in the SLC6A2/MET subtype; PSEN2 was expressed predominantly in neuronal populations; APOE was strongly expressed in astrocytes with additional signal in microglia, vascular cells, and some neuronal subtypes, but was essentially absent from NE neurons; SORL1 was prominent in microglia with additional expression in astroependymal cells and neuronal subtypes including NE populations; and TREM2 expression was essentially restricted to microglia. Together, these snRNA-seq analyses define the cellular architecture of the human LC and its surrounding region and highlight transcriptionally specialized NE neuron subtypes with differential engagement of melanocortin and AD-related pathways.

Visium spatial transcriptomics of the human LC

The 10 × Genomics Visium Spatial Gene Expression platform was used to map the LC transcriptome in its morphological context. The anatomic right LC of a neurotypical donor was sampled in two blocks, one rostral and one caudal (Fig. 2a). Two adjacent 10 µm sections from the rostral block (Upper Rostral, Lower Rostral) and two adjacent 10 µm sections from the caudal block (Upper Caudal, Lower Caudal) were processed for Visium, providing coverage along the rostral–caudal axis of the LC (Fig. 2a). After image-based cropping of low-UMI peripheral regions, 12,495 spots were retained across the four sections.

Fig. 2.

Visium spatial transcriptomics of the human LC. a Schematic of workflow from separation of the hemi brainstem through sagittal cut to transverse sectioning of the metencephalon (LC shown in blue), followed by dissection of the LC (the boxed portion of the slices enlarged on the right side embedded in O.C.T. compound) and subsequent generation of consecutive sections for Visium spatial transcriptomics (10 micron sections, shown in orange). Photographs and sections were taken from the blocks’ rostral face. b Spatial organization of cell types in a representative Visium LC section (Lower Rostral), revealing distinct NE, other neuronal and glial populations. Each Visium spot captures transcripts from multiple cells; cell-type labels therefore indicate spots enriched for a given cell class rather than single-cell identities. c Spatial gene expression maps for selected noradrenergic markers, melanocortin pathway and AD–related genes in the same Visium section, illustrating their distribution and relative expression levels

To place these spatial data in a broader context, we integrated our LC Visium sections with published LC Visium data from Weber et al. [64], which contributed 22,853 spots from four additional donors, resulting in 35,348 spots in total (Supplementary Table 3). Integrated UMAP embedding of Basak and Weber spots showed contribution of both datasets to most clusters, indicating successful cross-study integration (Supplementary Fig. 2a). Clustering followed by marker-based annotation resolved spatial domains enriched for NE neurons, serotonergic neurons, TAC1/CRH neurons, ADCYAP1/VGF neurons, PENK neurons, PVALB/SCN4B neurons, GABAergic neurons, glutamatergic neurons, oligodendrocytes, astrocytes, ependymal cells, vascular cells and two unclassified spot populations (Supplementary Fig. 2b). Because each spot captures transcripts from multiple cells, these categories represent cell type–enriched spots rather than single-cell identities. A dot plot of canonical markers supported these annotations, and violin plots summarizing quality-control metrics (sum UMI counts, number of detected genes and mitochondrial RNA percentage) are provided in Supplementary Fig.  2c, d. Projection of these annotations back onto individual sections revealed a stereotyped LC architecture in both Basak and Weber donors, with a compact NE-enriched band surrounded by additional neuronal and glial territories (Fig. 2b, Supplementary Fig. 3a).

We next examined spatial gene-expression patterns for noradrenergic, melanocortin and AD-associated transcripts within the Visium LC sections. In the representative Lower Rostral section from our donor, DBH expression sharply delineated the LC NE neuron core (Fig. 2c). MRAP2 and PSEN2 were detectable in a subset of NE-enriched spots and in surrounding neuron-enriched domains, whereas APP and APOE showed a broad distribution across the tissue (Fig. 2c). PSEN1, SORL1 and TREM2 were expressed at lower levels, with signal scattered throughout the section (Fig. 2c). Examination of all sections across Basak and Weber donors revealed similar patterns for these genes (Supplementary Fig. 3b). Consistent with the noradrenergic identity of this domain, SLC6A2 closely paralleled DBH in outlining the LC band, while MC1R and MC4R were detected only in a small subset of LC spots, as well as in neuron-enriched regions adjacent to the LC. LEPR, MC2R, MC3R, and MC5R were expressed only very sparsely or were not detectably expressed (Supplementary Fig. 3b).

Together, these Visium analyses demonstrate that the human LC forms a conserved noradrenergic domain across individuals, defined by spatially restricted DBH and SLC6A2 expression, and that MRAP2 is detectable within this LC territory. AD-associated genes such as APP, PSEN1, PSEN2, APOE, SORL1, and TREM2 display broader spatial patterns, consistent with their expression in multiple neuronal and non-neuronal populations. These observations are largely in line with the cell type–resolved expression profiles obtained from our snRNA-seq analysis.

Melanocortin and AD-associated gene expression profiling across the entire mouse LC using RNAscope HiPlex

The human spatial transcriptomic data of the LC indicate heterogeneity of gene expression within the NE neurons (identified by DBH). We further assessed the heterogeneity with regards to genes of the melanocortin pathway and those associated with AD using RNAscope HiPlex. The complete mouse LC was included in 57 coronal sections (Fig. 3a). Each section was stained with twelve distinct probes, imaged using a confocal microscope, and then images were overlayed upon each other to generate a HiPlex image (Fig. 3b; Table 1). The images of the LC obtained were compared to reference images from the mouse Allen Brain Atlas using Dbh as the genetic marker for NE neurons within the region. Dbh⁺ cells were used to delineate the LC (Fig. 3c).

Fig. 3.

RNAScope HiPlex workflow and LC detection. a Fresh, frozen adult mouse brain tissue was obtained and sectioned, resulting in 57 sections of 20 µm each for the LC, extending from rostral to caudal, with 3 sections per slide. Brain Image adapted from Allen Brain Atlas. Scale Bar = 1860 µm. Image generated using Biorender (https://biorender.com/). b The RNAScope HiPlex assay was performed by hybridizing the gene-specific mRNA probes to our sample. Then one round of amplification and fluorophore application for 3 probes was completed, which was subsequently followed by imaging of all slides. The applied fluorophores were then cleaved and the process was repeated for a total of four rounds. The images from each round were overlayed and analysis was completed using Halo software. Image generated using Biorender (https://biorender.com/). c Sections from rostral (upper left corner) to caudal (bottom right corner). Cells were counterstained with DAPI (gray). Dbh⁺ cells were used as a marker for the LC. Six representative LC sections from the HiPlex experiment were matched to the six LC sections from the Allen Brain Atlas. Allen Brain (Right): Scale Bar = 1675 µm. Dbh⁺-LC (Left): Scale Bar = 200 µm

In Fig. 4, a representative overview of each transcript’s image is displayed from section 35, one of the sections which falls within the central region along the rostral-caudal axis. Here Dbh⁺ LC cells are highly abundant (Fig. 4a-l). All genes of interest present in our 12 channels were overlayed upon each other to generate a HiPlex image of each LC section (Fig. 4m). The random forest classifier was trained to identify the LC by selecting two Dbh⁺ regions within the LC and two Dbh⁻ regions outside the LC (Supplementary Fig. 4 a, b). An example of an input used to outline an LC region within a section is provided (Supplementary Fig. 4c). Across the sections, the number of Dbh-expressing cells increased from rostral to caudal, with higher numbers beginning at section 22, the maximum in sections 35–39, and remaining high up to section 57. None of the other transcripts examined in this experiment followed the Dbh gene distribution. While each gene presented with a distinct dynamic, within sections 25–28 and 34–39, all of them had one peak. However, there was no single section in which all genes peaked (Fig. 4n).

Fig. 4.

HiPlex RNAScope detection of AD-associated and leptin-melanocortin genes in the LC. a–m Cells were counterstained with DAPI (gray). Genes of interest were detected with fluorescent labeled probes and imaged in four rounds. Images displayed are from section 35 of the LC. Merged image shown in panel “m”. Magnification × 63. Scale Bar = 200 µm. n Dot plot graph showing expression levels of signature genes across each LC section. The size of the dot indicates the proportion of cells expressing each gene of interest. The spectrum of colors indicates the mean expression level. Section 1 is the most rostral section and section 57 is the most caudal section

We quantified the expression of each of the melanocortin pathway genes and each of the AD-related genes in Dbh⁺ or Dbh⁻ cells relative to the total number of Dbh⁺ or Dbh⁻ cells. We also quantified the average number of RNA foci per cell in the two categories, Dbh⁺ or Dbh⁻, within the LC (Fig. 5; Supplementary Fig. 5). Representative images for each gene of interest were selected from sections 31 to 35 (Table 2). Mrap2 is expressed in 71.8% of Dbh⁺ and 19.7% of Dbh⁻ cells. However, the number of Mrap2 molecules per cell are much greater in the Dbh⁺ compared to the Dbh⁻ cells (Fig. 5a). Less than 5% of both Dbh⁺ and Dbh⁻ cells express Lepr or Mc4r (Fig. 5b, c). AD-related genes are expressed at different levels in the Dbh⁺ and Dbh⁻cells of the LC (Fig. 5d–k; Table 3a, b).

Fig. 5.

HiPlex RNAScope co-expression of genes of interest in Dbh⁺and Dbh.⁻ LC cells. a–k Top panel: cells were counterstained with DAPI (gray). Genes of interest were detected with fluorescent labeled probes and imaged in four rounds. Magnification × 63. Scale Bar = 10 µm. Bottom panel: bar graphs represent quantification of double-positive cells expressed as a percentage with individual data points for each section value (Data are means +/− SD) and the average number of molecules per cell (Data are means +/− SEM). Two-tailed unpaired student’s t-test, ****p < 0.0001

Table 2.

Representative sections for selected genes of interest. Representative images displayed in Fig. 5 were selected from sections 31 and 35. The genes represented from each section are listed

Section number	Gene
31	Mrap2
	Mc4r
	App
	Psen1
	Psen2
	Apoe
	Trem2
35	Lepr
	Sorl1
	Pax7
	Crh

Table 3.

Selected values from Figs. 5 and 7

a		App	Psen1	Psen2	Apoe	Sorl1	Trem2	Pax7	Crh
	Dbh+	45.5%	55.7%	39.6%	33.6%	59.2%	18.3%	43.7%	4.8%
	Dbh−	25.9%	28.9%	10.0%	42.4%	26.0%	3.6%	12.5%	4.9%
b		App	Psen1	Psen2	Apoe	Sorl1	Trem2	Pax7	Crh
	Dbh+	20	7	6	23	10	10	10	10
	Dbh −	19	5	4	21	4	7	2	12
c		Familial AD genes	GWAS AD genes	All AD-related genes
	Mrap2+	21.4%	4.6%	0.03%
	Mrap2−	0.42%	0.35%	0.03%

(a) Percentage of total Dbh⁺ cells co-expressing Dbh and each selected gene of interest or percentage of total Dbh⁻ cells expressing each selected gene of interest. Percentage is mean value of all sections. (b) Foci per cell for each selected gene of interest in Dbh⁺ or Dbh⁻ cells. Value is the mean value of all sections. (c) Percentage of total Dbh⁺ cells co-expressing Dbh and familial AD genes (App, Psen1, Psen2), GWAS AD genes (Sorl1, Trem2, Apoe), or all AD genes (App, Psen1, Psen2, Sorl1, Trem2, Apoe). Percentage is mean value of all sections

Co-expression of melanocortin receptor family genes with Mrap2

Mrap2 interacts with all of the melanocortin receptor proteins, Mc1r-Mc5r [88]. One of this study’s goals is to identify which melanocortin receptors are co-expressed with Mrap2 in the LC. By HiPlex, Mc4r expression is very limited in the LC; necessarily the number of cells in which both Mc4r and Mrap2 transcripts could be detected. Less than 3% of Mrap2 positive cells co-express Mc4r (Table 4). We looked for other melanocortin receptors that might be co-expressed with Mrap2. In an RNAscope Multiplex experiment, limited co-expression of Mrap2 with Mc1r (< 1%) or Mc3r (< 5%), and higher co-expression of Mrap2 with Mc2r (10.9%) or Mc5r (17.9%) was detected within LC cells (Fig. 6a–d; Table 4). Overall, a higher percentage of cells co-expressed Mrap2 and Mc5r compared to the other melanocortin receptor genes.

Table 4.

Percentage of each melanocortin receptor Mc1r–Mc5r co-expressing Mrap2. Mean percentage of co-expression from all sections from each experiment

	Mrap2+
Mc1r	0.8%
Mc2r	10.9%
Mc3r	4.2%
Mc4r	2.3%
Mc5r	17.9%

Fig. 6.

RNAScope imaging of cells co-expressing Mrap2, Mc1r, Mc2r, Mc3r, and Mc5r in the LC. a Schematic diagram of RNAscope Multiplex experiment. Even and odd numbered slides were separated. Even slides were probed for Mrap2, Mc1r, and Mc2r and fluorescently labeled. Odd slides were probed for Mrap2, Mc3r, and Mc5r and fluorescently labeled. All slides were then imaged with the confocal microscope. b Cells were counterstained with DAPI (gray). Genes of interest were detected with fluorescent labeled probes and imaged. Magnification × 63. Scale Bar = 10 µm. c Bar graphs represent quantification of double-positive/Mrap2⁺ cells, expressed as a percentage with individual data points for each section value. Data are means +/− SD. d Dot plot graphs showing expression levels of Mrap2, Mc1r, and Mc2r (left; even slides) and Mrap2, Mc3r, and Mc5r (right; odd slides). The size of the dot suggests the proportion of Mrap2⁺ cells co-expressing each gene of interest. The spectrum of colors indicates the mean expression level. Section 1 is the most rostral section and section 27 is the most caudal section

Identification of LC cellular population co-expressing Mrap2 and AD-related genes

Finally, after identifying the abundant expression of Mrap2 in the LC, we analyzed the RNAscope HiPlex data further to see if there is a population of cells which co-expresses Mrap2 and the AD-related genes. We identified cells which co-express Mrap2 and familial AD-genes-App, Psen1, and Psen2- along with Dbh (Fig. 7a). 40.1% of App⁺, 49.3% of Psen1⁺, and 35.9% Psen2⁺ cells also expressed Mrap2 and Dbh; 21.4% of all cells in the LC co-expressed App, Psen1, Psen2, Mrap2, and Dbh (Fig. 7c, e). There were fewer cells co-expressing Mrap2, Dbh, and GWAS AD genes-Trem2, Sorl1, Apoe (Fig. 7b). 16.3% of Trem2⁺, 50.1% of Sorl1⁺, and 21.6% Apoe⁺ cells also expressed Mrap2 and Dbh and 4.6% of all cells in the LC co-expressed Trem2, Sorl1, Apoe, Mrap2, and Dbh (Fig. 7d, f; Table 3c). There were no cellular populations identified which co-expressed Mrap2, Dbh, and all AD-related genes (Fig. 7g; Table 3c).

Fig. 7.

RNAScope imaging of cells co-expressing Mrap2 and AD-related genes in the LC. a–b Cells were counterstained with DAPI (gray). Familial AD Genes-App, Psen1, Psen2-and Mrap2 were detected with fluorescent labeled probes and imaged in a. GWAS AD Genes-Trem2, Sorl1, Apoe-and Mrap2 were detected with fluorescent labeled probes and imaged in b. Merged images in final panels. Magnification × 63. Scale Bar = 50 µm. c–d Bar graphs representing quantification of triple-positive cells expressed as a percentage of Dbh⁺ cells with individual data points for each section value. Mrap2⁺ and Mrap2⁻ cell populations are represented. e–g Bar graphs represent quantification of cell populations co-expressing Familial AD genes in e, GWAS AD genes in f, and all AD related genes in g. Cell populations that were Mrap2⁺ or Mrap2⁻ were compared

Discussion

Numerous studies have described obesity as a risk factor for AD. The leptin-melanocortin pathway is a key mediator of food intake and energy expenditure and may be involved in the pathophysiology of this association [1, 62, 89–91]. Impaired leptin signaling may play a role in AD pathogenesis [19, 20, 92, 93]. We investigated whether the melanocortin genes are expressed in the LC—one of the brain regions known to be affected in the earliest stages of AD.

In the meta-analysis of snRNA-seq datasets from the human LC region, we identified three transcriptionally specialized DBH⁺ NE neuron subtypes and mapped melanocortin and AD-associated genes to this data. MRAP2 emerged as the most prominent melanocortin-related transcript in NE neurons, being enriched in SLC6A2/MET and SLC6A2/DBH subtypes. In contrast, melanocortin receptors themselves were expressed at low levels: MC1R was mainly detected in the SLC6A2/MET subtype, while MC2R, MC3R, MC4R, and MC5R transcripts were generally rare and sparsely detected. AD-associated genes showed broader, cell type–specific patterns that were consistent with their known biology. Together, these snRNA-seq data reveal a transcriptionally heterogeneous NE neuron population in the human LC with differential engagement of melanocortin and AD-related pathways.

Visium Spatial Transcriptomics identified distinct spatial arrangements of cellular populations in the human LC region. Our primary focus were the LC NE neurons, with the biosynthetic enzyme DBH serving as a marker for NE-producing neurons [94]. Integration of all Visium sections from our donor with published LC Visium data from Weber et al. [64] demonstrated a conserved noradrenergic LC domain and surrounding peptidergic, GABAergic, glutamatergic, and glial territories. Consistent with the snRNA-seq results, MRAP2 expression was detectable within the LC territory. By contrast, melanocortin receptors were at or near the detection limit of the Visium platform: MC1R and MC4R were observed only in a small subset of NE-enriched spots and adjacent neuron-enriched regions, whereas LEPR, MC2R, MC3R, and MC5R were very sparsely or not detectably expressed. As previously established, APP showed widespread expression across LC and non-LC tissue. PSEN1 and SORL1 were expressed at lower levels and were scattered throughout neuronal and glial regions, whereas PSEN2 was detectable in neuronal territories. TREM2 signal was sparse. In summary, our Visium analysis corroborates the presence of a conserved noradrenergic core in the human LC, confirms MRAP2 expression within this domain, and shows that AD-associated genes exhibit broader spatial patterns, consistent with their distribution across multiple neuronal and non-neuronal cell types observed in our snRNA-seq meta-analysis.

With the highly sensitive RNAscope technology, we were able to detect all probed genes of interest. There was considerable heterogeneity of all gene expression profiles analyzed within the LC. Each of the melanocortin pathway and AD-associated genes showed a distinct gene expression pattern within Dbh⁺ neurons across the mouse LC. These findings are in agreement with our Visium study in the human LC and confirm prior studies on the heterogeneous molecular signatures of the Dbh⁺ neurons within the LC [12, 86, 95–99].

Each gene had its own unique expression pattern within Dbh⁺ and Dbh⁻ cells of the LC. There was limited expression of Lepr within the mouse LC. This is in agreement with a previous snRNA-seq study of the LC that identified Lepr in addition to strong Galanin (Gal) expression in NE neurons [63]. The functional significance of this small population of Lepr-expressing cells is unknown. In other brain regions it has been shown that Gal can act as an important mediator of leptin action [100]. Among the melanocortin pathway genes, Mrap2-expressing cells were by far the most abundant within the LC. Mutations in MRAP2 have been associated with obesity [101–103]. MRAP2 interacts with all 5 melanocortin receptors, MC1R, MC2R, MC3R, MC4R, and MC5R, to regulate cell surface expression and signaling via response to α-MSH [88, 101, 103–107].

Of the melanocortin receptors co-expressed in Mrap2⁺ cells, Mc5r was highest, followed by Mc2r. Although MC2R and MC5R are classically associated with peripheral functions, there is evidence for the central effects of both receptors in stress, immune, and metabolic processes, which is relevant for our study of LC-NE neurons and AD pathology [12, 14, 51–55]. MC2R, the canonical ACTH receptor, is a key mediator of hypothalamic–pituitary–adrenal (HPA) axis signaling. Central ACTH–melanocortin signaling has been shown to modulate arousal, stress responsiveness, and noradrenergic activity, which are all core functions of LC NE neurons [12, 51, 56]. Chronic stress and dysregulated HPA axis activity are well-established contributors to LC degeneration and AD pathology, providing a mechanistic context for MC2R expression in these neurons [50, 52, 57, 58]. MC5R, which plays roles in both exocrine and immune regulation, has also been implicated in central anti-inflammatory and neuroprotective melanocortin signaling [12, 53, 55]. Neuroinflammation is a critical driver of LC vulnerability and early tau pathology in AD. Melanocortin receptors, including MC5R, have been shown to suppress pro-inflammatory signaling and oxidative stress in neural tissues [52, 53, 55, 59–63]. Within the context of our study, the identification of MC2R and MC5R expression in LC NE neurons expands the canonical view of melanocortin signaling beyond hypothalamic feeding circuits and supports a broader framework in which melanocortin pathways intersect with stress, immune regulation, and neurodegeneration.

Despite the reported findings, many Mrap2-expressing cells do not co-express melanocortin receptor genes (Supplementary Fig. 6). MRAP2 is considered a broad-spectrum GPCR modulator [108], and many GPCRs are expressed in the LC [63, 109]. This raises the question of which other GPCRs are co-expressed with MRAP2 in cells lacking melanocortin receptors. To address this, we systematically examined the broader GPCR landscape using a curated IUPHAR GPCR gene list [84, 85], generating dot plots across NE neuron and other neuronal subtypes and major non-neuronal classes in the integrated snRNA-seq meta-dataset, as well as corresponding spatial maps in the integrated Visium dataset (Supplementary Figs. 7 and 8). These analyses confirm that numerous additional GPCRs are expressed in the human LC/pons, including receptors involved in energy homeostasis that are known to interact functionally with MRAP2, such as orexin receptor (OXR1 [HCRTR1]) and MCHR1 [95–98]. The expression of these receptors has been also previously reported in the LC [39, 94, 99–102] and, together with our GPCR-wide profiling, highlights a broader set of candidate MRAP2 interaction partners beyond the melanocortin receptors.

Among the AD-associated genes, App and Apoe were highly expressed within the LC. Apoe expression was higher in non-neuronal cells. Consistent with our findings, Apoe is primarily expressed in astrocytes, but is also present in CNS neurons. Apoe is synthesized by astrocytes, however, neuronal expression and synthesis of Apoe increases with excitotoxic stress [110]. APP is expressed in neurons, astrocytes, microglia, oligodendrocytes, and endothelial cells [67, 73], we confirmed this distribution. Mutations in APP and presenilins contribute to increased amyloid-beta accumulation and AD pathogenesis [68–70, 73]. APP, PSEN1, and PSEN2 mutant genes are familial and strong contributors to early AD pathogenesis [71, 72]. Our study is agreement with prior work showing Psen1 expression in neurons, astrocytes, oligodendrocytes, microglia, and endothelial cells [67]. Consistent with the literature, we find PSEN2 is primarily expressed in neurons [111]. Cells co-expressing these familial AD genes were identified in the LC. Of note, APP and PSEN1 are not differentially regulated with respect to transcript level in LC neurons in the brains of individuals with mild/moderate AD relative to the brains of individuals with no cognitive impairment or mild cognitive impairment [112].

The expression of other AD-associated genes was also investigated. SORL1 is now considered to be a causal AD gene [77]. Sorl1 regulates the trafficking and processing of App and interacts with tau and Apoe as an Apoe receptor. Sorl1 is widely expressed in the brain, predominantly in hippocampal neurons, brainstem nuclei and Purkinje cells [75, 76]. Decreased SORL1 is associated with increased beta-amyloid accumulation. [75, 76]. Our study identified cellular populations expressing Sorl1 in both neuronal and non-neuronal populations of the LC. Other AD-related genes including App, Apoe, Psen1, and Psen2 were expressed in some of these cells. Trem2 is expressed in the LC, although at lower levels than the other AD-related genes. Whole exome sequencing and GWAS have identified Trem2 alleles that are associated with increased risk of developing AD [80, 81, 113]. Lack of Trem2 is believed to result in impaired recruitment of microglia leading to plaque formation, amyloid beta damage to neuronal cells, and neuroinflammation [82, 114, 115]. Trem2 is expressed throughout the brain including the hippocampus, brainstem, and cerebellum [79]. Literature predominantly suggests that Trem2 is expressed exclusively in microglia; however there have been reports of Trem2 expression in neurons, astrocytes, and oligodendrocytes [114–125].

Pax7 and Crh are expressed in the LC. There are Pax7⁺ and Pax7⁻cellular populations within the LC. The projections to the rest of the brain as well as function are thought to differ [96]. In our study, Pax7 expression was more abundant in specific regions of the LC. Crh expression within the LC was sparse, however, there was a dense population of Crh neurons detected adjacent to the LC. Crh expression has been detected in Barrington’s nucleus, which is adjacent to the LC [63, 126].

A high percentage of Mrap2 expressing cells within the LC are co-expressed with familial AD genes. This co-expression is a predicate for co-regulation of these receptors and has relevance for assessment of any functional links between the melanocortin and AD pathways to further the understanding of the conjunction of obesity and AD. LC NE neurons regulate neuroinflammation. NE has an anti-inflammatory effect on microglia through the suppression of inflammatory mediators, regulation of microglial motility, and reduction of oxidative stress [127–129]. In AD, the hypothalamus contains amyloid deposition and plaque formation [18, 20, 130, 131]. LC NE neurons project to the hypothalamus and damage to the LC is associated with atrophy and neuronal loss in the paraventricular and supraoptic nucleus of the hypothalamus, a process which may be mediated by neuroinflammation [10, 132–134]. Mrap2 is expressed throughout the hypothalamus and has been linked genetically to obesity [88, 135, 136]. Obesity is considered a chronic, low grade inflammatory state and it is a key risk factor for AD possibly through alterations in leptin signaling [62]. While mediators of leptin signaling have been implicated in AD, MRAP2 may also serve as a potential link between the LC and the hypothalamus in AD pathogenesis.

Identification of AD-associated genes and MRAP2 abundantly expressed within the LC is important as the LC had been implicated in both neurodegeneration and feeding dynamics [10, 14, 16, 17, 73]. There is significant cross species homology with regards to LC cellular organization, with some variation in neuronal abundance in some cellular subtypes of LC neurons between primates and rodents [64, 137]. Comparison of expression patterns using RNAscope in human samples would be valuable. The available brain tissue for our human studies consisted of female and male research subjects. However, due to the limited availability of female donors (n = 1 for snRNA-seq, n = 1 for Visium), sex specific analysis could not be performed. Anatomical and systems-level studies indicate that the core organization of the LC and its major afferent inputs, including hypothalamic projections, are conserved across sexes, with no evidence for qualitative sex differences in the presence or directionality of these pathways [26, 138, 139]. Thus, sex is as an important modulator of LC function that may influence the magnitude or behavioral consequences of noradrenergic signaling. While the underlying circuitry examined here is likely to be shared across sexes, lack of sex specific analysis is a limitation of this study.

Further, using HiPlex RNAscope and Multiplex RNAscope in mice (n = 1 for each assay) we generated complete, high-resolution anatomical maps, which has not been achieved previously. A limitation of this study is that it does not capture inter-animal variability or support statistical inference. Another consideration is that we do not make comparisons across age ranges in this study. Five-week-old mice are generally considered to be in mid adolescence (which is well beyond weaning). At this stage, the core anatomical organization of the melanocortin system including POMC and AgRP neuronal populations and their major projection patterns has already been established [140–142]. The melanocortin circuitry continues to mature, however, previous studies have demonstrated that expression of key melanocortin components such as Pomc, Agrp, and Mc4r is robust and readily detectable by 5-weeks of age [143–145].

Our study of human and mouse LC used spatially resolved transcriptomics, snRNA-seq, and RNAscope. Here we established the molecular characteristics of AD-related genes in the LC and identified a potential association with genes related to energy homeostasis via identification of cellular populations co-expressing Mrap2 and AD-associated genes. Currently there are gaps in the literature regarding the characterization of melanocortin pathway genes and AD-related genes in the LC of individuals with AD relative to individuals without AD. Further studies of the effect of MRAP2 loss- and gain-of-function on gene expression in the LC and on AD pathology are needed to understand the impact on AD disease progression [67].

Author contributions

C.A.D., A.B., and F.M.B.E. conceptualized the study. A.B., F.M.B.E., M.C.D.R., Z.D., V.O., H.J.G., R.R., G.H., J.C.M., H.B., Y.B., Q.S., B.C., C.A., D.F., B.C., H.X., H.R., P.A.W., C.R., M.W.S., J.Y.A., and A.F.T. generated data, conducted portions of the analysis, assisted with data interpretation, and participated in sample provision. A.B., F.M.B.E., and M.C.D.R. wrote the main manuscript text. A.B., R.L.L., L.O.Q., A.F.T., C.A.D. revised and helped prepare the manuscript. C.A.D. was tasked with funding acquisition. C.A.D. and A.F.T. supervised the project. All authors reviewed and approved the final manuscript.

Funding

This work was supported by New York Obesity Research Center, P30 DK026687-43 and DK52431-25. The snRNA-seq study was funded in part through the NIH/NCI Cancer Center Support Grant P30CA013696 and used the Genomics and High Throughput Screening Shared Resource. This work was supported by the National Center for Advancing Translational Sciences, National Institutes of Health, through Grant Number UL1TR001873.

Data availability

Data that support the findings of this study have been deposited to GEO.

Availability of data and materials

Raw snRNA-seq data (Cell Ranger output matrices) and processed Seurat objects (Basak snRNA-seq, meta-analysis all cells, and meta-analysis neurons) are deposited in NCBI GEO under accession GSE327442. Raw Visium spatial transcriptomics data (Space Ranger output matrices, tissue images, and spatial coordinates), along with the processed Seurat object from the integrated Basak and Weber et al. (2024) Visium dataset, are deposited under accession GSE327441.

Code availability

All code used to reproduce the snRNA-seq and Visium analyses and figures in this manuscript is available on GitHub: https://github.com/fmbetul/LC-melanocortin-analysis. Analyses were performed using R version 4.3.2 and Seurat version 5.0.1.

Declarations

Ethics approval and consent to participate

All autopsy materials were donated to Columbia by next of kin for both diagnostic and research purposes. Our IRB has designated our study “not human subject research.” All animal care and experimental procedures followed the US National Institutes of Health and Animal Care and Use guidelines and were approved by the Columbia University Animal Care and Use Committee (IACUC). Animals were sacrificed according to Columbia University regulations using the appropriate protocol (#AC-AABF6552).

Consent for publication

Not applicable.

Competing interests

Q.S., M.W.S., and J.Y.A. are full-time employees of Regeneron Pharmaceuticals Inc.