Spatial transcriptomics in human skin research

Spatial transcriptomics is a revolutionary technique that enables researchers to characterise tissue architecture and localisation of gene expression. A plethora of technologies that map gene expression are currently being developed, aiming to facilitate spatially resolved, high‐dimensional assessment of gene transcription in the context of human skin research. Knowing which gene is expressed by which cell and in which location within skin, facilitates understanding of skin function and dysfunction in both health and disease. In this review, we summarise the available spatial transcriptomic methods and we describe their application to a broad spectrum of dermatological diseases.


| INTRODUC TI ON
Tissues and organs of multicellular organisms are spatially organised by the compartmentalisation of cell types and subpopulations that are coordinated in internal gene regulatory networks, and by signalling from their surrounding tissue environment. In order to perform a subset of biologic processes, these cells differ in terms of localisation, morphology and gene expression patterns. 1 The development of single-cell methods has enabled studies that can retain cellular properties and heterogeneity at a single-cell resolution, which has led to a more detailed understanding of physiologic and disease conditions. 2 One of the most popular methods, single-cell RNA-sequencing (scRNA-seq), has made it possible to profile the whole transcriptome of each individual cell. 3,4 However, it results in the loss of information on spatial relationships among the cell populations since it entails dismantling of the original tissue into single-cell suspensions through mechanical and enzymatic dissociation steps, which could perturb or alter the cell's gene expression. 5 Recently, various techniques have been developed aiming for transcriptome mapping while preserving the spatial location of the expressed transcripts within intact tissues. 6 These methods, collectively known as spatial transcriptomics (ST), have been voted as the method of the year 2020 by Nature Methods and expect to provide remarkable novel insights in the near future. 7 Spatial information can be particularly useful in characterising conditions such as tumour microenvironment, 8 certain immune responses 9 or cell-cell interaction during development and (patho) physiological conditions. 10 Skin is one of the most stratified and multi-functional organs in humans, characterised with high-cell heterogeneity in homeostasis, wound healing and cancer. 11 For instance, fibroblasts in the dermis have distinct functions according to their location, whether they are close to a hair follicle, in the papillary layer or near the hypodermis. 12 In parallel, keratinocytes in the basal layer have different roles when compared to keratinocytes in the superficial layer. 13 Even a minor distribution in the skin structure can alter the cutaneous cell organisation and lead to disease. This is why histology has been attributed as the main tool in diagnosing skin diseases. 14 Therefore, the acquisition of location specific information of gene expression has a very high impact in cutaneous research.

| AVAIL AB LE TECHNOLOG IE S
Spatial transcriptomics can be categorised into three major methods: in situ hybridisation, in which the tissue is labelled by predesigned probes; in situ sequencing, in which transcripts are amplified in the tissue, labelled by fluorescent nucleotides and detected by imaging; and next-generation sequencing-based methods, which label all acquired transcripts with spatial barcodes that can map their appropriate locations within the tissue. 6,15 The integration of high-throughput next-generation sequencing in situ has revolutionised spatial gene capturing since they enable the entire transcriptome to be sequenced and analysed. 15 The first attempts in profiling the transcriptome spatially were conducted with laser capture microdissection (LCM) or single-molecule fluorescent in situ hybridisation (smFISH) techniques in the late 1990s. 16 smFISH visualises transcripts of specific genes with non-radioactive fluorescent or colorimetric dyes. 17 In the early 2000s, a combinational barcoding method was developed in which gene-specific probes are used, and transcripts of up to five genes can be quantified in each tissue. 18 19 More recently, the advancement of scRNA-seq and its application together with LCM, enabled the generation of a more detailed spatial transcriptomic profiling. This technology is referred to as (Geo-seq). 20 Despite the high robustness of LCM, it is a very labour-intensive method that limits the sample throughput.
There is a plethora of next-generation sequencing methods, including Visium by 10× Genomics, Slide-seq and Nanostring GeoMx, which capture transcripts in tissue sections in a specific manner 21 ( Figure 1). In the commercialised approach created by 10× Genomics called Visium, tissue sections are placed onto slides that contain reverse transcription (RT) primers and are x and y coordinated to undergo fixation, staining, imaging and permeabilisation. During permeabilisation, mRNA diffuses into the slides and hybridises into the RT primers of its location. Then the transcriptome is extracted for NGS library preparation. 22 In Slide-seq technology, the workflow is conceptually similar to Visium, but instead of having RT primers onto slides, barcoded beads in solution are randomly packed into a glass coverslip. The position of each barcode is acquired by in situ indexing. 23 Nanostring has launched the GeoMx commercialised system, which can capture both proteins and RNA but in different tissue sections. In this method, the user selects regions of interest that are excited with ultraviolet light triggering the release of proteintargeting antibodies or RNA-targeting probes barcoded by specific tags. The tags are then collected and analysed by the NanoString nCounter instrument. 24 These technologies provide a resolution larger than a single cell within a tissue (3-30 cells in a single spot) but can go through the conventional downstream processes similar to scRNA-seq. 25 MERFISH is another available spatial transcriptomics technology which combines in situ hybridisation and spatial barcoding of the RNA transcripts, establishing single-cell resolution. This workflow entails hybridising the sample with encoding probes which contain incorporated barcodes, that are imprinted into the RNA of the tissue and are detected by imaging. Nevertheless, this method can only detect hundreds to a few thousand genes. 26

| APPLI C ATI ON S IN S KIN RE S E ARCH
Spatial transcriptomics can be a powerful tool in mapping cell-tocell interactions and cell diversity within a tissue. Several studies have been conducted in the dermatologic field aiming to decipher wound healing processes, skin cancer and inflammatory diseases (Table 1) (representative visualisation of Nanostring data are shown in Figure 2). ing neovascularisation, as well as stimulating abnormal fibrosis. 32 Spatial transcriptomics was also utilised to assess diabetic foot ulceration healing, which revealed a significant localisation of healingassociated fibroblast subtype towards the wound bed of the ulcer in comparison to unwounded skin. Furthermore, mapping of the immune landscape of healing and non-healing ulcers has revealed an abundance of M1 macrophages in healers and M2 macrophages in non-healers, suggesting an interplay between healing fibroblasts and acute inflammatory responses in efficient wound healing resolution. 33

| Inflammatory diseases
Some forms of leprosy are defined by the formation of granulomas, due to the recruitment of lymphocytes and macrophages that have been infected by Mycobacterium leprae. 37    cytokines such as IL17F, IL21, IL22, TNF, IL4 were found in psoriasis and atopic dermatitis. These cytokine positive spots were associated with sets of immune-related genes associated with pathways such as oxidative stress, neutrophil migration (IL17A, IL17F, IL26), apoptosis (IFNG) and type 2 immunity (IL13, IL4). 30 Numerous studies have been conducted to decipher multiple inflammatory pathways leading to psoriasis. Ding et al. utilised ST to identify pathways underlying neutrophilic cell death and neutrophil extracellular traps (NETs), which consist of enzyme and peptides aggregating dead neutrophils. 28 In their study, they hypothesised that an important driver gene of psoriasis SHP2, might cause NET formation, which aggravates the disease. They had identified two neutrophil marker genes that PI3 and S100A8, and two markers for NET formation, CASP4 and PPIF, to be expressed in psoriatic skin tissue.
The association of SHP2 and NETs was confirmed by the decrease in the expression of NET-forming proteins in SHP2 knockout mice. 28 The role of SH2 in the promotion of psoriasis was also studied by

| Skin cancer
Spatial transcriptomics has also been a valuable tool in dissecting information from several skin cancers. With respect to melanoma, a study mapped three individual profiles from a specific expression of patterns in melanomas from short-and long-term survivor patients. Pathways involving epithelial-mesenchymal transition and immune-mediated regression of metastases were found, which can be exploited in feature therapeutic purposes. 40 The melanoma microenvironment was also studied in BRAF V600E -driven melanomas in zebrafish. 8 Distinct cell populations were identified within zebrafish melanomas and the neighbouring tissues, further assessed by scRNA-seq. Also, upregulation of cilia genes was found, which are conserved in human samples, suggesting a potential feature and pathway in human melanomas. 8 Melanoma is classified as one of the most metastatic neoplasms, which most frequently occurs in the lungs, the liver and the brain. A multi-omics study investigating the spatial and temporal composition of human brain metastatic melanomas revealed a restricted expression of type-I interferon responses in specific areas and clusters of lymphoid aggregates which are dominated by plasma cells. ATAC-seq and scRNA-seq analysis indicated larger fractions of lymphoid aggregates as well as dysfunctional TOX + CD8 + T cells occurring during distinct expression of immune checkpoints. 39 The cellular composition of squamous cell carcinoma (SSC) has been explored using ST by Ji et al. in order to assess the regions within the tumours in which cells and pathways found by scRNAseq reside. 35 They discovered a tumour-specific (TSK) population surrounded by a fibrovascular niche in the tumour-leading edges.
The expression of interferon signalling was enriched in the non-TSK-leading edge, as well as cancer-associated fibroblast-associated transcripts located in the vicinity of the TSK cells. Ligand receptor mapping revealed that TSK cells participated in autocrine and paracrine interactions mostly with cancer-associated fibroblasts (CAFs), endothelial cells and macrophages that may contribute to tumour progression, immunosuppression and heterogeneity. 35 In basal cell carcinoma (BCC), the tumour-stroma interface was genetically separated, with a total of 86 genes found to be differentially expressed between the infiltrative and nodular tumour regions. 34 A high percentage of the genes was observed in a fibroblast cluster which was reported to drive BCC infiltration. 34 The spatial genomic landscape of tissues in clinical trials and drug development and testing can be particularly useful in identifying specific areas responsive to the drug and deciphering pathways that are associated with each response. Head and neck squamous cell carcinomas (HNSCC) are classified as the most common malignancies in the pharynx, larynx and oral cavity. 42 A clinical trial has used spatial genomic profiling to investigate the response of PD-L1 + and FoxP3 + T-cell antitumor cellular density to PD-L1 immune checkpoint inhibitor durvalumab. 36 HNSCCs were subjected to a single dose of durvalumab or supplemented with metformin which has been shown to promote antineoplastic effects in tumour microenvironments. Transcriptomic analysis has shown a significant increase of FoxP3, as well as the increase of CTLA-4 gene post-therapy, which is responsible for T-cell exhaustion but has a mixed role in immune response, suggesting a novel pathway leading to PD-L1 response. 36 The induction of immune checkpoint blockade drugs in multiple cancer patients can lead to multiple immune-related adverse events.
One of these events is immune-related dermatitis, in which the spatial context of T cell activation was investigated using formalin-fixed paraffin embedded (FFPE) skin samples from patients with dermatitis and psoriasis. 29 Spatial transcriptomic data have shown a higher expression of inhibitory receptors in dermatitis cases, as well as a portion of tissue-resident memory T cells aggregated to the inflamed part of the skin. 29

| FUTURE PER S PEC TIVE S AND TECHNOLOG IE S
Methods for analysing the spatial profile of tissues are constantly evolving with the release of new high-resolution technologies.
Currently, these methods are at the forefront of their development and lack the sufficient molecular depth for diagnostic purposes.
However, the development of spatial single-cell resolution, larger tissue area, detection sensitivity and higher number of profiled genes made spatial transcriptomics a great option in translational research and diagnostics, since most pathology methods rely only on cell and tissue structure by means of H&E staining. 43 With the use of ST, genes and pathways responsible for a disease or dysfunction in a tissue area will provide an in-depth, high-throughput analysis resulting in a more accurate and quicker diagnosis. 43 Recently, two technologies called Stereo-seq 44 and Seq-scope 45  This method was applied in mice embryonic tissues and was proven to provide single-cell resolution. 47  in extracted nuclei from the cells after the tissue has been stained. 56 All these cutting-edge technologic advances can improve our ability to study tissue complexity and become the foundation for discovering novel biomarkers or establishing new or stratified therapies.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors have no conflict of interest to declare.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing not applicable to this article as no datasets were generated or analysed during the current study.