2024 Volume 71 Issue 2 Pages 101-118
Accumulating evidence suggests that cellular heterogeneity in organs and cell-cell and tissue-tissue interactions are crucial for maintaining physical homeostasis and disease progression. Endocrine organs also exhibit cellular heterogeneity and comprise multiple cell types. For instance, the pituitary gland comprises five types of pituitary hormone-producing cells as well as non-hormone-producing supporting cells, such as fibroblasts, endothelial cells, and folliculostellate cells. However, the functional roles of the interactions between hormone-producing and non-producing cells in the pituitary gland remain incompletely understood. Over the past decade, emerging technologies such as single-cell and spatial transcriptomics have provided excellent tools for studying cellular heterogeneity and their interactions; however, the application of these technologies in endocrine research remains limited. This review provides an overview of these technologies and discusses their strengths and limitations. Additionally, we also summarize the potential future applications of single-cell and spatial transcriptomics in the study of endocrine organs and their disorders.
The human body is composed of approximately thirty-to-forty trillion cells encompassing over two hundred distinct cell types, including neurocytes, myocytes, hepatocytes, and enterocytes [1]. These cells exhibit remarkable variability in their functions, morphology, and gene expression profiles, which arises from differences in their individual epigenomes and microenvironments. This diversity among cells substantially affects intercellular signaling, metabolism, and cell division, which actually plays important roles in various biological processes, such as development, regeneration, and aging. For instance, cellular diversity is a key factor in early embryonic development, during which the correct placement and timing of different cell types are vital for organ formation and function. In the process, cell-cell interactions among various cell types coordinate developmental processes [2]. As an example, during brain development, the localization of repressors plays a crucial role in delineating the neurogenic ectoderm, leading to the formation of the central nervous system [3]. As another example, in cancerous tissue, non-cancer cells, such as fibroblasts, immune cells, and vascular endothelial cells, coexist with cancer cells, with each cell type exhibiting distinctive gene expression profiles and functions. Such heterogeneity influences cancer progression and therapeutic responsiveness [4, 5].
Various types of cells coexist also in endocrine organs, such as the pituitary, thyroid, pancreas, and adrenal glands. These cellular heterogeneities play important roles in regulating the physical function, proper development, and disease pathophysiology of the gland. These functional mechanisms involved have been studied using various cell lines, engineered animals, and clinical samples. However, such cellular heterogeneity can substantially increase the amount of information to be handled, such as the interactions between multiple cell types and accompanying gene expression changes. Consequently, obtaining a comprehensive understanding of cellular heterogeneity becomes challenging.
Numerous technological advancements have been achieved in the last decade, including single-cell transcriptomics, spatial transcriptomics, and imaging mass cytometry. These technologies have revolutionized our ability to analyze cellular heterogeneity and identify previously unknown cell subtypes and the mechanisms of cellular interactions. However, the application of these technologies to endocrine organs remains limited.
In this review, we discuss the importance of cellular diversity and cell-to-cell interactions within endocrine organs in maintaining endocrine functions. Additionally, we highlight recent technical innovations with a focus on single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics. By reviewing the characteristics of these technologies, we discuss their potential impact on future endocrine research.
Cellular heterogeneity in the endocrine organs Pituitary glandThe pituitary gland is a small gland located at the base of the brain that plays a crucial role in the regulation of various physiological processes. This gland consists of two parts with different origins, the anterior and posterior lobes (Fig. 1a). These regions differ significantly in their function depending on the types of hormones they secrete. The anterior lobe synthesizes and releases five types of hormones, including adrenocorticotropic hormone (ACTH), which stimulates the adrenal cortex to produce cortisol; growth hormone (GH), which promotes the growth and development of tissue; thyroid-stimulating hormone (TSH), which stimulates the thyroid gland to produce thyroid hormones; luteinizing hormone (LH) and follicle-stimulating hormone (FSH), which regulate gonadal function; and prolactin (PRL), which is important for milk production in females. The posterior lobe serves primarily as a storage site for hormones synthesized by the hypothalamus. These hormones include oxytocin, which plays a role in social and reproductive behaviors; and arginine vasopressin (AVP) (recently referred to as vasotocin (VT) [6]), which regulates water balance in the body. These lobes have different origins: the anterior lobe differentiates from the oral ectoderm, which is derived from the non-neural ectoderm, whereas the posterior lobe differentiates from the hypothalamus, which is derived from the neural ectoderm. During embryonic development, the anterior lobe originates from the oral ectoderm and interaction with the adjacent hypothalamus is crucial for proper development of the gland. These cell-to-cell interactions determine the fate of part of the oral ectoderm to become the pituitary gland, and a precisely controlled spatiotemporal transcription network guides the differentiation to form the multicellular anterior lobe of the pituitary gland (Fig. 1b).
Cellular heterogeneity in the pituitary gland
a: The pituitary gland consists of the anterior and posterior lobes. Anterior lobe includes five types of hormone-producing cells (corticotroph, ACTH; somatotroph, GH; lactotroph, PRL; thyrotroph, TSH; and gonadotroph, LH/FSH) and non-hormone-producing supporting cells, such as fibroblasts, endothelial cells, and folliculostellate cells. b: During embryonic development, five types of hormone-producing cells are derived from the same progenitor cells. In the early stage, tissue interaction between oral ectoderm and ventral hypothalamus is important for development of the pituitary progenitor organ (Rathke’s pouch). Subsequently, timely and locally fine-tuned expression of transcription factors is essential for the specification of these hormone-producing cells. c: In adult pituitary, SOX2+ pituitary stem cells can produce all types of hormone-producing cells. The transcription factor networks involved in the turnover process of pituitary stem cells are consistent with the mechanisms present in pituitary development.
This review focuses on the anterior lobe. In addition to the five endocrine cell types, smaller populations of non-endocrine cell types, including folliculostellate cells, endothelial cells, and fibroblasts, are present in the anterior pituitary and provide structural and functional support to the hormone-secreting cells (Fig. 1a). Multicellular networks are important for controlling the function of adult pituitary glands. Folliculostellate cells are the most abundant type of non-endocrine cells in the anterior pituitary gland and play several roles in the pituitary gland, including acting as scavenger cells with phagocytic activity and as supportive cells. These cells form a network of interdigitating processes that surround hormone-producing cells and regulate hormone secretion and cell-to-cell communication. Folliculostellate cells also have the ability to secrete various cytokines and growth factors, which play important roles in pituitary development and hormone secretion [7, 8]. Endothelial cells and fibroblasts are also present in the anterior pituitary gland and contribute, respectively, to the formation and maintenance of the blood vessels and the extracellular matrix that surrounds the hormone-producing cells of the glands (Fig. 1a). This microenvironment also plays important roles in pituitary tumor development and function [9-11]. Additionally, pituitary stem cells play an important role in maintaining tissue homeostasis in the adult pituitary (Fig. 1a). Despite ongoing debate regarding the cellular characteristics of pituitary tissue stem cells, it is generally recognized that SOX2+ cells residing within the pituitary gland are tissue stem cells [12]. A SOX2-lineage tracing experiment showed that SOX2+ cells naturally give rise to multiple types of hormone-producing cells throughout life (Fig. 1c). The transcription factor networks involved in the turnover process of adult pituitary is consistent with the mechanisms during pituitary development (Fig. 1b and c). These SOX2+ cells proliferate in response to gonadectomy and adrenalectomy to generate gonadotropes and corticotrophs [13, 14]. SOX2+ cells exist within the marginal cell layer, a region between the anterior and posterior lobes, and within the parenchyma of the anterior lobe. Generally, the maintenance of tissue stem cells is believed to be dependent on a microenvironment called the stem cell niche. In the case of pituitary stem cells, it is thought that microenvironmental cues such as SHH, WNT, TGF-β, and Hippo signaling are crucial for stem cell maintenance; however, questions remain regarding what regulatory mechanisms modulate these signaling in these stem cells and commit the stem cells into their respective cell lineage.
Thus, the anterior pituitary gland contains a diverse population of cells, including stem cells and non-endocrine cells, which play important roles in the regulation of hormone secretion and the maintenance of gland structure. However, the precise mechanisms underlying these cellular communications have not been fully elucidated.
Adrenal glandAnother example of cellular heterogeneity in endocrine organs can be observed in the adrenal gland, which comprises two main parts: the cortex and medulla [15] (Fig. 2a). The adrenal cortex produces multiple steroid hormones, such as cortisol, aldosterone, and androgens, while the adrenal medulla produces catecholamines, such as epinephrine and norepinephrine. Different layers of the cortex produce different hormones: the zona glomerulosa produces aldosterone, the zona fasciculata produces cortisol, and the zona reticularis produces androgens (Fig. 2a). The precise balance of these hormones is essential for maintaining salt and water balance, blood pressure, and other physiological processes.
Cellular heterogeneity in the adrenal gland
a: The adrenal gland consists of the cortex and medulla. The cortex is further divided into the zona glomerulosa, zona fasciculata, and zona reticularis, which secrete aldosterone, cortisol, and androgens, respectively, while the medulla secretes catecholamines. b: The cortex of the adrenal gland shares a common origin with the kidneys, testes, and ovaries, which are all derived from the urogenital ridge. The medulla of the adrenal gland originates from the neural crest cells. c: During adrenal gland development, cortisol from the cortex enhances the expression of phenylethanolamine N-methyltransferase (PNMT) in the medulla, facilitating the conversion of norepinephrine to epinephrine. In the adult adrenal gland, steroid hormones from the cortex affect gene expression by binding to steroid receptors (SRs) in medullary cells. Conversely, catecholamines secreted from the medulla affect cortical cell function by binding to catecholamine receptors in the cortex. ACTH, adrenocorticotropic hormone; PNMT, phenylethanolamine N-methyltransferase; NE, norepinephrine; EP, epinephrine; SR, steroid receptor; CR, catecholamine receptor.
The adrenal cortex arises from the coelomic mesoderm of the urogenital ridge (Fig. 2b). During the fifth week of human fetal development, proliferating mesothelium-derived cells from the posterior abdominal wall give rise to the primitive adrenal cortex. A recent review article by Kastriti et al. comprehensively describes how mouse adrenal medulla develops from neural crest cells (NCCs) [16]. From embryonic day (E) 9.0 to E10.5, early NCCs migrate near the dorsal aorta and form the suprarenal ganglion. These progenitors give rise to chromaffin cells and sympathetic neuroblasts that supply nerves to the adrenal medulla. Although the cortex and medulla have different embryological origins and represent two discrete tissues, they are morphologically and functionally interconnected [17].
During the adrenal organogenesis, close interactions between the cortex and medulla (medullary-cortical interactions) are necessary for the proper differentiation and morphogenesis of the adrenal gland (Fig. 2c). Norepinephrine-producing chromaffin cells differentiate into epinephrine-producing chromaffin cells that are characterized by the expression of phenylethanolamine N-methyltransferase (PNMT). This cellular differentiation in the medulla requires glucocorticoids from the cortex [16, 18-20]. Absence or impairment of the adrenal cortex can cause developmental impairment in the adjacent medullary tissue [17]. When the melanocortin-2-receptor (MC2R), which is the ACTH receptor in the adrenal cortex, is knocked out in mice, the animals develop deficiencies in glucocorticoids and mineralocorticoids, as well as catecholamines [21]. Nr5a1 (encoding steroidogenic factor-1/SF-1, which is an important transcription factor during adrenal cortex development)-knockout mice lack adrenal cortical cells. In the mice, the number of chromaffin cells in the medulla is reduced by about 50% [22]. Oppositely, catecholamines and neuropeptides secreted from the adrenal medulla stimulate the production and release of steroid hormones (cortisol, aldosterone, and androstenedione) from the adrenal cortex [23]. Matsuo et al. used a human induced pluripotent stem cell (hiPSC)-derived adrenal cortex model to study the effects of dopamine produced from adrenal medulla on the steroidogenic differentiation. In the model, treatment with dopamine D1 receptor agonist upregulated the expression of various steroidogenic enzymes and increased the secretion of steroid hormones synergistically with ACTH. These results suggest the importance of dopamine D1 receptor signaling in cortical steroidogenic differentiation [24].
In addition to the medullary-cortical interactions during embryonic development, communication between the cortex and medulla ensures the regular functioning of the adult adrenal gland. In the normal adrenal gland, production and secretion of steroid and catecholamine hormones are regulated by a complex network of autocrine, paracrine, and endocrine interactions [23]. Neuroendocrine stimulation such as stress leads to the release of steroids from cortical steroid-producing cells. The released steroids subsequently work on adjacent chromaffin cells, where they bind to intracellular steroid receptors. Steroid-bound steroid receptors translocate into the nucleus, leading to the induction of specific gene expressions involved in catecholamine synthesis (Fig. 2c). Conversely, catecholamines released from the adrenal medulla regulate steroid biosynthesis in adrenocortical cells in a paracrine manner by binding to α- or β-adrenergic receptors (AR), resulting in the transcription of steroidogenic enzymes [25] (Fig. 2c). These synergistic mechanisms play coordinated roles in the regulation of stress response, metabolism, and blood pressure.
PancreasThe pancreas contains both exocrine (enzyme secreting) and endocrine (hormone secreting) cells. This organ primarily aids in the digestion of proteins, fats, and carbohydrates by secreting digestive enzymes from acinar cells and in the regulation of blood glucose levels by secreting insulin and glucagon from β- and α-cells, respectively (Fig. 3a). The pancreas is formed through the fusion of the ventral pancreatic bud and the dorsal pancreatic bud, both of which originate from the endoderm. The transcription factor PDX1 (pancreatic and duodenal homeobox 1) is crucial for early pancreatic specification, as it is expressed in pancreatic progenitor cells that give rise to both the exocrine and endocrine pancreas [26] (Fig. 3b). Acinar cells differentiate from multipotent progenitor cells under the regulation of transcription factors, such as PTF1A (pancreas transcription factor 1 subunit alpha) and MIST1 (Bhlha15, basic helix-loop-helix family member A15) [27]. As the pancreas develops, the transcription factors neurogenin 3 (NGN3) and NeuroD1 are crucial for endocrine progenitor specification. NGN3 is expressed in endocrine progenitor cells, which give rise to all endocrine cell types [28]. NeuroD1 is a transcription factor that acts subsequent to NGN3, promoting the differentiation and maturation of pancreatic endocrine cells.
Cellular heterogeneity in the pancreas
a: The pancreas consists of exocrine and endocrine cells. In exocrine glands, acinar cells secrete digestive enzymes into the pancreatic duct. Endocrine glands are referred to as islets of Langerhans, which consist of α-, β-, δ-, and PP (pancreatic polypeptide) cells. These cells secrete glucagon, insulin, somatostatin, and pancreatic polypeptide. b: The exocrine and endocrine cells of the pancreas originate from the same pancreatic progenitor cells that differentiate into exocrine progenitor cells characterized by GATA4/6 and endocrine progenitor cells characterized by NGN3. Exocrine precursor cells further differentiate into acinar and duct cells, whereas endocrine precursor cells differentiate into their respective endocrine cells. c: Within the islets of Langerhans, the α-, β-, and δ-cells regulate each other’s functions via paracrine mechanisms.
The exocrine pancreas comprises of acinar and ductal cells. Acinar cells secrete digestive enzymes (e.g., amylase, lipase, and proteases) into the pancreatic ducts. Ductal cells secrete bicarbonate to neutralize stomach acids and ensure optimal enzymatic function (Fig. 3a). The endocrine pancreas comprises several different cell types, including α-cells, β-cells, δ-cells, and PP cells, which produce glucagon, insulin, somatostatin, and pancreatic polypeptide, respectively. In the pancreas, paracrine interactions between these endocrine cell types within the islets of Langerhans play crucial roles in regulating hormone secretion and maintaining glucose homeostasis (Fig. 3c). Somatostatin released from δ-cells inhibits insulin secretion from β-cells by binding to somatostatin receptors on the β-cell surface [29]. Insulin released from β-cells inhibits glucagon secretion by α-cells [30]. Insulin has also been reported to stimulate somatostatin secretion from δ-cells under certain conditions, such as high glucose levels. Thus, the precise ratio of these different cell types is critical for maintaining glucose homeostasis and deviations from this ratio can lead to conditions such as diabetes. However, the underlying regulatory mechanisms are not yet fully understood, and nor are the processes through which the ratio of these endocrine cells is retained.
As discussed above, endocrine organs consist of various cell types, and cell-cell interactions play vital roles in regulating their functions and development. Single-cell-level analysis techniques are indispensable for comprehensive understanding of this cellular diversity. Accompanying recent advancements in next-generation sequencing (NGS) methods, techniques such as scRNA-seq and spatial transcriptomics have been developed and are increasingly being employed in cellular diversity analyses. The journal Nature Methods selected each of these as the Method of the Year in 2019 and 2020, respectively [31, 32]. In addition, because biological insights need to be extracted from the vast amount of data generated by NGS analyses, recent computational advancements have contributed substantially. The computational challenges associated with handling vast amounts of data include noise reduction, dimensionality reduction, clustering, differential expression analysis, and trajectory inference. For instance, analysis tools such as Seurat and Scanpy have been widely used to preprocess, visualize, and cluster single-cell data [33, 34]. Additionally, pseudotime analysis tools such as Monocle and Slingshot have been used to infer the developmental trajectories of cells [35, 36]. Machine learning strategies, particularly those based on artificial neural networks such as autoencoders are increasingly used to interpret complex patterns and predict potential cell-cell interactions from single-cell transcriptomic data [37, 38]. These specific tools and strategies exemplify computational efforts in the field to fully exploit the potential of single-cell transcriptomics data. Furthermore, several computational approaches aim to integrate the results of scRNA-seq and spatial transcriptomics to obtain single-cell-level gene expression data that retain spatial information. In this section, we focus on scRNA-seq and spatial transcriptomics and review their technical features and computational approaches to integrate their results.
Single-cell RNA sequencingWhen studying the cellular heterogeneity in tissues, scRNA-seq allows researchers to obtain comprehensive gene expression data in individual cells within a heterogeneous population. This technique can reveal previously unknown cell subtypes, analyze rare cell types, and provide insights into cellular interactions.
The first scRNA-seq method was developed by Tang et al. in 2009 [39]. This method, known as Smart-seq, can capture mRNAs from individual cells and amplify them using reverse transcription and polymerase chain reaction (PCR). The amplified complementary DNAs (cDNAs) were then sequenced using NGS. Although Smart-seq enabled researchers to study the transcriptomes of individual cells at high resolution, its low throughput and high cost limited the use of the method in biological studies. To overcome these limitations, several microfluidic-based scRNA-seq methods have been developed, including Drop-seq [40], 10x Genomics [41], and inDrops [42]. These methods use microfluidic devices to capture individual cells and barcode their mRNAs, which are then amplified simultaneously and sequenced by NGS. These methods greatly increased the throughputs and enabled the profiling of tens of thousands of cells in a single experiment (Fig. 4).
Single-cell RNA sequence and spatial transcriptome (in situ capture)
In single-cell RNA sequence (scRNA-seq), tissues are dissociated into single cells, which are then partitioned and lysed, followed by barcoding of intracellular mRNA to distinguish between individual cells. In the in situ capture spatial transcriptome methods, tissue sections are mounted onto RNA-binding probes that include barcode sequences that represent spatial information. This tissue is then permeabilized to barcode the mRNA therein. In subsequent procedures, which are similar for both techniques, a cDNA library is created from the barcoded mRNAs. This library is then sequenced, and through data analysis, comprehensive gene expression information can be obtained for each individual cell or according to spatial information.
To date, scRNA-seq has been used in numerous studies that analyze cellular heterogeneity in various biological systems. As one example, scRNA-seq of the developing human brain revealed cellular heterogeneity in brain tissue, its molecular profiles, and possible inter-cellular communications [43, 44]. As another example, Enge et al. analyzed human pancreatic cells from eight donors spanning six decades of life using scRNA-seq, and found that old pancreatic endocrine cells exhibited increased transcriptional instability and potential functional impairment. Furthermore, they analyzed somatic mutations from scRNA-seq data and identified a mutational signature present during the aging of pancreatic endocrine cells [45]. scRNA-seq has also been utilized for a wide range of cellular analyses, including hematopoietic stem cells [46, 47], lymphocytes in the thymus [48, 49], and mesenchymal stem cells [50].
Overall, the development of scRNA-seq has revolutionized the study of cellular heterogeneity, gene expression networks, and cellular interactions, which enabled the discovery of novel cell types and biological processes. As scRNA-seq technologies continue to evolve and improve, they are likely to have an even greater impact on biology and biomedicine.
Spatial transcriptomicsOne limitation of scRNA-seq is its inability to keep spatial information during the library construction for NGS, which hinder the analysis of cellular interactions. To address this limitation, spatial transcriptomics has been developed, which can map the transcripts of individual cells within their tissue context. This technique enables researchers to analyze the gene expression of each cell in its native environment with positional information, providing insights into the interactions between different cell types [51]. To date, several methods of spatial transcriptomics have been developed, which are roughly divided into two groups depending on the method of transcript detection: High-Plex RNA imaging (HPRI)- and NGS-based methods (Fig. 5).
Overview of spatial transcriptomics
Spatial transcriptomics is mainly grouped into High-Plex RNA imaging (HPRI)-based and next generation sequencer (NGS)-based methods. In situ sequencing (ISS) is a pioneering method in HPRI-based spatial transcriptomics. In ISS, mRNA in the tissue on the slide is reverse transcribed to synthesize complementary DNA (cDNA). The cDNA is then hybridized with padlock probes designed for each target gene and circularized through ligation. These padlock probes contain an ID sequence corresponding to each gene, which is amplified using rolling cycle amplification. Bridge probes and fluorescently labeled detection probes are annealed at this site and fluorescence is detected through imaging. By changing the combination of fluorescence that detects the target gene in each cycle and repeating this cycle, it is possible to detect several hundreds or thousands of genes simultaneously. The in situ capture method is a representative NGS-based method. In this method, tissue slices are mounted directly onto RNA-binding probes containing barcode sequences that provide spatial information, thereby barcoding the mRNA in the tissue. A cDNA library is synthesized from mRNA. The arrangement of the RNA-binding probes differed in each method. In 10x Visium, the probes are printed on a glass slide. Each spatial spot is 55 μm, and the distance between spots is 100 μm. In Slide-seq, resolution is improved by laying out 10 μm beads on the slide. In high-definition spatial transcriptomics (HDST), barcoding is performed using 2 μm beads on a silicon wafer.
HPRI-based methods can directly determine the location and identity of RNA molecules in intact tissue sections. In situ sequencing (ISS) was established in 2013 as a pioneering HPRI-based method [52]. This technique consists of the following procedures: tissue sectioning, hybridization of padlock probes, ligation of both ends of the probes, rolling cycle amplification, hybridization of detection probes, imaging, repeated rounds of de-hybridization and hybridization, and data analysis (Fig. 5). To date, several modifications have been made to improve the detection efficiency and resolution of this method [53, 54]. By repeating imaging cycles, several thousands of genes can be detected in the tissue. Furthermore, HPRI-based methods can be combined with other methods. When coupled with immunostaining, they enable the acquisition of data on protein expressions and epigenomic modifications along with transcriptomic profiles, thus enriching multi-omics analysis. Moreover, integration with clustered regularly interspaced short palindromic repeats (CRISPR) screening [55] expands the capability of HPRI-based methods, allowing for the collection of both transcriptomic and phenotypic data including cellular morphology, cell proliferation rates, and the subcellular protein localization assessed with immunostaining. This simultaneous evaluation of transcriptomic and phenotypic parameters at the single-cell level facilitates high-throughput screening of phenotypic alterations induced by CRISPR/Cas9-mediated gene-targeting.
Chen et al. performed ISS of the brain in a mouse model of Alzheimer’s disease (AD) [56], and combined ISS and β-amyloid (Aβ) immunostaining to elucidate the spatiotemporal changes in gene expression caused by Aβ. In another study, hiPSC-derived somite organoids (axioloids) were analyzed using ISS [57]. Gene expression profiles during somite development were shown with spatial information by mapping fifty-eight genes, including all HOX family members and other marker genes important for somite development. The observed patterns were consistent with those predicted from other vertebrates.
NGS-based methods represent the other category of spatial transcriptomics method. NGS-based methods include laser-captured microdissection (LCM) [58], in situ capture, Digital Spatial Profiler (DSP) [59], and APEX-sequencing [60]. Among these, the in situ capture method is the most representative and is now widely used in many biological studies. The seminal study by Ståhl et al. introduced the in situ capture method that combines tissue imaging and NGS to provide high-resolution spatial gene expression data [61]. This method comprises the following steps: mounting freshly frozen samples onto specialized glass slides containing an array of spatially barcoded oligonucleotide spots, tissue permeabilization, cDNA synthesis on the slide, library preparation, sequencing, and data analysis (Figs. 4 and 5). The constructed cDNA library contains spatial barcode sequences that facilitate the identification of the location of the detected mRNA in the imaged tissue. This technique forms the basis for the development of other spatial transcriptomics platforms; however, it has a limited resolution of 100 μm. Rodriques et al. developed Slide-seq, another spatial transcriptomics technique that utilizes a spatially barcoded bead array [62]. In this method, 10 μm of beads are laid on a glass slide, and their position on the slide can be determined based on their barcode sequence. Subsequently, frozen tissue is attached to the beads and the RNA is bound to the barcode on the beads. After bead collection and library construction, the cDNA and barcodes are sequenced by NGS to determine the position on the glass slide where the mRNA was present. This approach provides single-cell resolution (Fig. 5). Vickovic et al. developed an enhanced version of the spatial transcriptomics method, high-definition spatial transcriptomics (HDST) of which resolution was 2 μm with spatially barcoded bead array on silicon wafer [63] (Fig. 5). Chen et al. developed spatial-enhanced resolution omics sequencing (Stereo-seq), which utilizes DNA nanoballs (DNB). The DNB-patterned arrays used in this method have approximately 220 nm diameter spots and center-to-center distances of 500 or 715 nm [64].
NGS-based spatial transcriptomics is now widely used to analyze many biological events, including development and disorders. Asp et al. analyzed the spatial transcriptome of developing human fetal heart [65]. This study provides a comprehensive spatiotemporal transcriptome of the cell types present in the embryonic heart during different developmental stages. Moncada et al. investigated the tissue architecture of pancreatic ductal adenocarcinoma via spatial transcriptomics [66], providing insights into tumor cell-microenvironment interactions and changes in cellular states. Additionally, spatial transcriptomes have been used to analyze tumor heterogeneity in various cancers, such as melanoma [67], prostate cancer [68], and breast cancer [63].
HPRI- and NGS-based spatial transcriptomics have both advantages and disadvantages, and it is necessary to select an appropriate method according to the size of the tissue of interest, cellular composition, and the purpose of the analysis. In general, HPRI-based spatial transcriptomics methods have a superior resolution, enabling mRNA detection and identification at the subcellular level. However, the imaging resolution does have its limits, and only a few to a dozen mRNA molecules can be observed per cell, compromising its quantitative capabilities. NGS-based spatial transcriptomic methods provide superior quantitation because the number of detected mRNA molecules can theoretically be increased without limit by increasing the number of reads. However, it has spatial resolution limitations; a single spatial spot can contain multiple cells. Currently, among commercially available NGS-based spatial transcriptomics, Visium from 10x genomics, is the most commonly used platform and has a single spot of 55 μm. In the following section, we discuss ways to overcome the disadvantages of NGS-based spatial transcriptomics.
Data repositories and analytical tools for single-cell RNA sequence and spatial transcriptomicsRecent advancements in these scRNA-seq and spatial transcriptomics technologies have generated an immense amount of data from all over the world. This increase underscores the need for accessible data repositories that organize and facilitate the sharing of this information. These repositories are invaluable for the integration and analysis of extensive datasets, thereby contributing to the attainment of more universal insights. Several data repositories have been established, which are currently acquiring and utilizing data. Gene Expression Omnibus (GEO), managed by NCBI, is a publicly available cost-free data repository that supports the storage and sharing of scRNA-seq and spatial transcriptomics data (https://www.ncbi.nlm.nih.gov/geo/). The Broad Institute launched the Single Cell Portal, which is dedicated explicitly to scRNA-seq (https://singlecell.broadinstitute.org/single_cell). ArrayExpress is hosted by the European Bioinformatics Institute (EBI), providing a robust platform for data deposit for reuse by researchers (https://www.ebi.ac.uk/biostudies/arrayexpress). The Human Cell Atlas Data Coordination Platform (HCA DCP) is a part of the Human Cell Atlas project. It is designed for storing comprehensive data related to human cells from scRNA-seq studies (https://data.humancellatlas.org/). These data repositories play a pivotal role in sharing scRNA-seq and spatial transcriptomics data and in facilitating advanced research in various biological fields. To date, hundreds or thousands of items of sequencing data for each endocrine gland have been registered to these repositories.
To accurately manage and understand the large amount of available data, robust and sophisticated analytical tools are required. Several analytical tools have been developed to facilitate efficient data processing and to improve result reliability. 10x Genomics offers a subset of tools for preprocessing, visualization, and data analysis. Cell Ranger and Space Ranger are used for preprocessing the scRNA-seq and spatial transcriptomics data, respectively. Additionally, Loupe Browser, with its user-friendly graphical user interface (GUI), allows for data analysis and visualization without the necessity for coding skills (https://www.10xgenomics.com/support/software/loupe-browser). However, integration and compatibility are only available with other 10x Genomics products, potentially limiting its use with data from different platforms. Seurat, developed by the Satija Lab (https://satijalab.org/seurat/) [33], stands out as one of the most widely used analytical tools. This tool enables data analysis, visualization, and integration of different batches of data. It supports a variety of scRNA-seq and spatial transcriptomics technologies, including 10x Genomics, Drop-seq, and Smart-seq2, and spatial transcriptomics, such as Visium, Slide-seq, and MERFISH. Seurat, built on the R programming language, reached its fourth version (September 25, 2023) [69], with ongoing updates, and the fifth version was launched in October 2023 [70]. Scanpy is another extensively used tool, constructed using the Python programming language (https://scanpy.readthedocs.io/en/stable/) [34]. It offers comprehensive capacities for preprocessing, visualization, and data exploration, and is compatible with both scRNA-seq and spatial transcriptomics. For visualizing scRNA-seq data, SPRING serves as a GUI web tool that employs k-nearest-neighbor graphs to provide robust and reproducible data representations (https://kleintools.hms.arvard.edu/tools/spring.html) [71]. Lastly, the Automated Single-cell Analysis Portal (ASAP), a web-based platform (https://asap.epfl.ch/) [72], enables the automated analysis of scRNA-seq data. ASAP is designed to perform intricate scRNA-seq data analysis, including data quality control, normalization, dimensionality reduction, clustering, differential expression analysis, and data visualization without coding skills.
Integration analysis of single-cell and spatial transcriptomicsSpatial transcriptomics is a promising tool for studying complex biological processes, including tissue-tissue and cell-cell interactions. Although NGS-based spatial transcriptomics can be used to detect tens of thousands of genes, it has limited resolution as each spatial spot covers up to dozens of cells. The lower resolution makes it difficult to precisely elucidate cellular interaction networks. To overcome this limitation, various integration methods for single-cell and spatial transcriptomics have been developed to obtain single-cell-level transcriptome data with spatial information [73]. The first step in integration is to identify the cell subtypes present in a certain tissue along with their transcriptional profiles. Because of its depth and single-cell precision, scRNA-seq is currently the optimal platform for defining cell subtypes in tissue. Subsequently, the integration of single-cell and spatial transcriptomics is performed. There are primarily two methods for integrating transcriptomic data: deconvolution and mapping methods (Fig. 6). The deconvolution method determines the proportions of cellular subtypes for a certain spatial spot, while the mapping method assigns the most likely dominant cell subtype to a spatial spot.
Overview of integration of single-cell RNA sequence and spatial transcriptomics
Integration analyses can be categorized into deconvolution and mapping methods. Deconvolution methods identify the cell types and their proportions at each spatial spot. Mapping methods aim to identify the most likely dominant cell type within each spot and map it accordingly.
Several methods are available for deconvolving cell types from the mRNA mixture of a particular spatial spot. These methods can be grouped into four main categories: enrichment scoring-based, regression model-based, probabilistic model-based, and deep learning-based methods [74]. Enrichment scoring-based methods generally determine the presence probability of each cell type in a spot based on the enrichment score of a gene set, such as cell type-specific marker genes identified from scRNA-seq data. Seurat [33], Giotto [75], and MIA [76] use this approach. The regression model-based approach assumes that a spot profile is a linear combination of cell type-specific expression profiles and cell type proportions. SPOTlight predicts the cell subtype population of each captured spatial spot using non-negative least squares regression [77]. SpatialDWLS utilizes a dampened weighted least squares regression [78]. Danaher et al. proposed SpatialDecon, which uses log-normal regression, and showed that log-normal regression improved the deconvolution accuracy compared with least squares regression [79]. CARD is built on a non-negative matrix factorization model [80]. Probabilistic model-based methods fit a probability distribution using statistical models. Robust cell type decomposition (RCTD) assumes that the spatial gene expression follows a Poisson distribution [81]. The сell2location is based on a Bayesian model that resolves cell types in spatial transcriptomics data [82]. The algorithm Stereoscope uses negative binominal distribution [83]. Additionally, a reference-free approach has been developed. STdeconvolve recovers cell-type transcriptional profiles along with their proportions within each spatial spot without relying on external single-cell transcriptomic references [84]. The deep learning-based methods deconvolve spatial spots by borrowing information from scRNA-seq data. DSTG is an artificial intelligence model that deconvolves spatial transcriptomic data using graph-based convolutional networks [85].
The mapping method aims to assign scRNA-seq-based cell type to each cell from HPRI data (Fig. 6). One common approach is that of clustering, where cells are projected into a shared space of lower dimensions after detection of clusters. Several methods have been developed and employ various data integration approaches to account for the batch effects between the two datasets. For the clustering, Seurat Integration uses canonical correlation analysis (CCA) [86], Harmony uses Principal component analysis (PCA) [87], and LIGER uses non-negative matrix factorization (NMF) [88]. SpaGE uses a domain adaptation method named PRECISE [89] to correct for the batch effect [90]. Instead of a clustering approach, pciSeq projects cell types based on a probabilistic model [91]. Tangram is a deep-learning-based method which can map any type of scRNA-seq data as “puzzle pieces” to align in space to match “the shape” of spatial data [92]. gimVI is a deep generative model that can be used to impute undetected transcripts [93].
Thus, many related methods have been developed to integrate scRNA-seq and spatial transcriptomics. Li et al. reported the results of benchmarking sixteen integration methods using forty-five paired and thirty-two simulated datasets, and found that that cell2location, SpatialDWLS, and RCTD were the top-performing methods for cell type deconvolution of spatial spots and that Tangram, gimVI, and SpaGE outperformed other mapping methods [94]. Another group benchmarked eighteen methods and concluded that CARD, cell2location, and Tangram were the best methods in terms of performance [95].
Applications of advanced analysis integrating single-cell and spatial transcriptomicsThe integrated analysis of single-cell and spatial transcriptomics is already being employed across many biological studies that investigate organ development and tumor microenvironment. In this section, we will review several studies utilizing these methodologies.
In organogenesis, cellular heterogeneity within a tissue and cell-to-cell communication play important roles in cell fate determination. Unraveling these intricate relationships requires sophisticated investigation methods. Therefore, the integrated analysis of scRNA-seq and spatial transcriptomics is a powerful tool that enables a comprehensive understanding of these complex biological processes. In one study, a comprehensive atlas of cell types in mouse brains was created by combining single-cell and spatial transcriptomics (Slide-seq), and RCTD, a computational method described above, was used to map single-cell clusters onto Slide-seq spatial beads. The integrated datasets revealed the cell type composition of neuroanatomical structures and enabled the comprehensive characterization of neuropeptide and neurotransmitter signaling, region-specific gene expression patterns, and heritability enrichment of neurological phenotypes [96]. In another study, spinal cord development was studied using integrated data processed with Stereoscope. Human prenatal samples were used to create a comprehensive developmental cell atlas of the spinal cord, which revealed the spatiotemporal regulation of neural progenitor cell fate commitment and positioning [97]. One research group investigated brain regeneration mechanisms using SpaTially Enhanced Resolution Omics sequencing (Stereo-seq) to capture spatially resolved single-cell transcriptomes. They focused on the axolotl, an animal known for its regenerative abilities, and analyzed its telencephalon sections during development and regeneration. Distinct spatial distributions, molecular features, and functions of the annotated cell types were observed. This study identified an injury-induced ependymoglial cell cluster at the wound site as a progenitor cell population potentially responsible for replenishing lost neurons through a cell-state transition process, which is similar to neurogenesis during development [98]. Mantri et al. investigated chicken heart development by integrating scRNA-seq and spatial transcriptomics. Integration was performed using an anchor-based method to predict cell type annotation for spatially resolved transcriptomics by Seurat package [99]. In another study, human heart development was examined by combining snRNA-seq and imaging mass cytometry. Heart tissue from control subjects and patients with congenital heart disease (CHD), including hypoplastic left heart syndrome and tetralogy of Fallot was analyzed, which revealed CHD-specific cell states in cardiomyocytes, such as enhanced insulin resistance and FOXO signaling, as well as a perivascular microenvironment consistent with an immunodeficient state in CHD [100]. Furthermore, differentiation processes in various organs, such as the kidney and pancreas, have been analyzed using these integration methods [101, 102].
These analyses also provide a highly valuable method for examining microenvironments, such as the infiltration of immune cells into tissue and intracellular interactions among multicellular entities. Recent advances in these methodologies have shed light on cell diversity within tissue and intricate cellular networks, particularly in cancerous tissue. Cancer cells within tumor tissue are not uniform and exhibit cellular heterogeneity at the genetic and epigenetic levels, which contributes to tumor progression, metastasis, and resistance to treatment. Moreover, tumor tissue is not solely composed of cancer cells; it also contains a significant number of non-cancerous cells, such as immune cells and fibroblasts. For instance, cancer-associated fibroblasts (CAFs) interact with cancer cells, thereby affecting tumor growth and responsiveness to treatment. Therefore, integrated analysis of scRNA-seq and spatial transcriptomics provides a highly beneficial approach for understanding cellular diversity within tumor tissue and the intracellular interactions. In one study, the glioblastoma microenvironment was analyzed by integrating scRNA-seq and spatial transcriptomics using SPOTLight. Integrative modelling revealed that a subset of myeloid cells that release IL-10 contributes to the immunosuppressive tumor microenvironment by causing T cell exhaustion [103]. Ji et al. performed an integrated analysis of the cutaneous squamous cell carcinoma microenvironment and identified tumor-specific keratinocytes localized in the fibrovascular niches. These cells communicate intercellularly with the tumor cells and have potential immunosuppressive features [104]. Thus, integrated methods have been used to investigate the cellular heterogeneity and microenvironment of various types of cancers, such as pancreatic cancer [66, 105], prostate cancer [106], and esophageal squamous cell carcinoma [107].
scRNA-seq and spatial transcriptomics have been extensively used in various fields. However, despite their promising potential, their application in endocrine organs has been relatively limited. We will review several studies that utilize scRNA-seq to understand cellular heterogeneity in endocrine organs.
Chen et al. reported the result of scRNA-seq on mouse pituitary posterior and intermediate lobes. They identified major cell types present in the lobes using scRNA-seq and illustrated their spatial organization by employing single-molecule in situ hybridization. Their results revealed that the macrophages are closely located to the pituicytes, suggesting possible intercellular communications between them [108]. Cheung et al. analyzed over ten thousand pituitary cells from seven-week-old male mice using scRNA-seq to investigate stem cell proliferation and differentiation in the pituitary glands [109]. They confirmed pituitary endocrine and stem cells as well as other support cell types present in the pituitary and identified novel markers of pituitary cell populations. Ruf-Zamojski et al. reported the results of a single-nucleus multiome analysis (transcriptome and chromatin accessibility) and the DNA methylation status of over seventy thousand single nuclei in the adult mouse pituitary [110]. They identified transcriptional and chromatin accessibility programs for each major cell type, highlighting the importance of chromatin accessibility in shaping cell-defined transcriptional programs. Zhang et al. reported the results of single-nucleus RNA-seq and ATAC-seq resources from postmortem pituitaries [111]. They focused on human pituitary stem cells and identified uncommitted stem cells, committing progenitor cells, and sex differences. These findings improve our understanding of human stem cell lineages and reveal the diverse mechanisms regulating key pituitary stem cell genes and cell type identity. Deng et al. studied the association between the pituitary and the kidneys. Their ligand-receptor analysis on scRNA-seq data of pituitary and kidney revealed ligand-receptor pair expressions, such as GH-GH receptor, pleiotrophin-syndecan-2/4, and delta-like non-canonical notch ligand (DLK) 1-Notch3, suggesting that the pituitary might directly regulate kidney function [112].
The hypothalamus contains a variety of neuronal types distributed across different regions, as well as numerous cell types such as oligodendrocytes, astrocytes, and microglia. The crosstalk among these cells plays a crucial role in functional regulation. Steuernagel et al. developed a comprehensive single-cell transcriptomic atlas of the mouse hypothalamus named HypoMap [113]. Moffitt et al. combined the HPRI-method with scRNA-seq to develop a spatially resolved cell atlas of the mouse hypothalamic preoptic region. They identified specific neuronal populations that are activated during social behaviors in male and female mice [114]. Mickelsen et al. also performed scRNA-seq on the hypothalamus. They mainly focused on mammillary bodies and revealed anatomical subpopulations with different projections to the thalamus [115].
Human adrenal cortex development was studied using scRNA-seq, which demonstrated the longitudinal transcriptional changes of diverse cell types included in the developing adrenal gland. In the study, ligand-receptor analysis revealed the potential bidirectional interactions between mesenchyme and adrenal cortex, such as DLK1-NOTCH2 (cortex to mesenchyme) and RSPO3-LGR4 (mesenchyme to cortex) [116]. These techniques have been used in several studies to examine pancreases. In one study, scRNA-seq analysis of human islets revealed that specific subpopulations of islet endocrine cells (α cells co-expressing ARX and MAFB and β cells co-expressing MAFA and MAFB) displayed increased functionality, including enhanced glucose sensing and hormone secretion, indicating that combinatorial transcription factor expression is indicative of mature functional islet cells [117]. Zheng et al. compared islet β cells from old and young mice using scRNA-seq and identified forty-seven significantly altered genes related to cellular aging [118]. These age-related genes might be associated with cellular senescence and functional impairments in islet β cells. Vivoli et al. studied the mechanisms underlying β cell proliferation induced by fatty acids using scRNA-seq [119]. Thus, scRNA-seq is actually useful for capturing cellular diversity and cell fate changes and studying cellular interactions also in endocrine cells.
In addition to the physiological conditions, scRNA-seq and spatial transcriptomics have been applied to pathological conditions including tumor tissue to investigate tumor heterogeneity and intra-tumoral communication. A comparison of silent and active pituitary corticotroph tumors with scRNA-seq revealed that silent corticotroph tumors exhibit characteristics of epithelial-to-mesenchymal transition, coupled with a decrease in transcripts regulating proopiomelanocortin (POMC) processing and secretion. These findings suggest a potential common transcriptional reprogramming mechanism that simultaneously impairs POMC processing and promotes tumor invasion [120]. The application of in situ capture spatial transcriptomics to pituitary corticotroph tumors confirmed the cellular diversity within these tumors. Their findings suggested that these diverse cells may have varying ACTH-producing capabilities [121]. Yu et al. investigated the role of telomerase reverse transcriptase (TERT) in a BRAF V600E mutant-induced thyroid cancer model using spatial transcriptomics. They showed that TERT activation promotes dedifferentiation of BRAF-caused thyroid cancer. Spatial transcriptomics revealed cellular heterogeneity in the tumor, and cell dedifferentiation was positively correlated with ribosomal biogenesis, suggesting that TERT upregulated ribosomal RNA expression and protein synthesis [122]. Spatial transcriptomics combined with metabolomics on adrenocortical aldosterone-producing adenoma (APA) revealed intra-tumoral transcriptional heterogeneity. A comparison of KCNJ5-wild type and KCNJ5-mutant APAs demonstrated the differences in metabolites that cause oxidative stress. This study also identified APA-like populations adjacent to APA, suggesting the existence of tumor precursor states [123].
In this review, we focused on cellular heterogeneity and cooperative intercellular interactions in endocrine organs and on the use of scRNA-seq and spatial transcriptomics as emerging technologies to analyze their functions. Data-driven approaches have recently become mainstream in biological research and have led to numerous breakthroughs. NGS-based techniques, such as scRNA-seq and spatial transcriptomics, are typical examples. While this review primarily discussed scRNA-seq and spatial transcriptomics, advancements have also been made in multiomics analyses, such as chromatin accessibility via histone modifications, DNA methylation, and single-cell protein expression analysis. Furthermore, the development of long-read sequencers, such as Single Molecule Real-Time (SMRT) sequencing (Pacific Biosciences) and nanopore sequencing (Oxford Nanopore Technologies), has further broadened the possibilities of NGS analyses [124-126]. Combining these long-read sequencers with scRNA-seq and spatial transcriptomics techniques may enable us to obtain isoform-specific transcriptome information at the single-cell level or retained spatial information, expanding the depth and resolution of our understanding of diverse biological systems [127].
Although cellular diversity and its networks play crucial roles in endocrine organs, the mechanisms underlying these roles are not yet fully understood. Moreover, the pathogenesis of many well-known diseases, such as Graves’ disease, Hashimoto’s disease, and lymphocytic hypophysitis, remain poorly understood. The advancements in the technologies reviewed in this study offer great opportunities for dissecting the complex cellular landscapes and cellular interactions within endocrine organs.
We would like to thank the members of the Yamamoto Lab. for discussions on the manuscript.
This work was supported by a JES Grant for Promising Investigator (to R. Matsumoto), grants in aid for JSPS Fellows (21J01689 to R. Matsumoto), an iPS Academia Japan Grant (to R. Matsumoto), and a Fellowship Program of Development of Young Researchers from Center for iPS Cell Research and Application (to T. Yamamoto).
Authors have nothing to disclose.
