Method for high‐plex analysis of immune cells in human skin using the GeoMx system

Specific T cell populations in the skin have been demonstrated as important disease drivers in several dermatoses. Due to the unique skin architecture, these cells are not grouped together in structures but dispersedly spread out throughout the epidermis. Following tissue disruption and isolation, only about 10% of skin T cells are recovered and any in vitro expansion may alter their bona fide phenotype. The Nanostring GeoMx system was developed to address cellular phenotype and protein expression in a tissue spatial context. To do so, regions of interest (ROI) must exceed a certain area threshold (usually 100 μm in diameter) to generate a sufficient signal‐to‐noise ratio. Here, we present an approach that allows for the pooling of numerous smaller ROIs within the skin, enabling T cell and melanocyte phenotyping. Skin samples from healthy individuals and vitiligo patients were analysed using the GeoMx system and several immune profiling panels. A sufficient signal‐to‐noise ratio was achieved by pooling smaller ROIs and analysing them as a single group. While this prevents spatial analysis, this method allows for detailed analysis of cells as a population in the context of their physiological environment, making it possible to investigate in situ phenotype of rare cells in different tissue compartments.


| INTRODUCTION
The human skin is comprised of two distinct anatomical layers: the epidermis and the underlying dermis.Together they form an interface between our body and the external environment.As such, the skin homoeostasis provides an immune barrier providing tolerance, microbial control and tumour surveillance. 1Resident and patrolling immune cells serve as a first line of defence against pathogenic microorganisms. 2,3One of the key cell types responsible for both homoeostasis regulation as well as threat elimination are the skin T cells.During encounter with a harmful pathogen, a large number of circulating T cells may enter the skin in order to quickly dispose of the microorganism and subsequently undergo apoptosis as a result of fine balance between pro-and anti-inflammatory responses. 4However, a specific subset of T cells may instead establish a long-term local population, primed to respond more rapidly towards secondary insults from the pathogen. 5These cells are known as tissue resident memory T cells (Trm) and have been shown to safeguard the body from both external threats such as viruses, as well as cancer cells of own origin. 5,6Unfortunately, despite these important protective functions that these cells have, it has been shown that improperly activated Trm have pathogenic potential themselves and play a role in the maintenance of 'disease memory' in certain inflammatory skin conditions.An enrichment of CD49a+CD8+ Trm cells has been observed in the skin of vitiligo patients. 7These cells demonstrated a higher cytotoxic capacity and produced IFNγ considered to be a hallmark of this dermatosis. 7,8urthermore, Trms lacking expression of CD49a were enriched in psoriasis patient's skin and were able to produce IL-17 cytokine, a key psoriasis cytokine, demonstrating the functional plasticity that these cells possess.Thus, a detailed investigation of the Trm gene and protein expression is essential to understand their involvement and contribution in inflammatory skin diseases.
Current protocols used for complex analyses of tissuederived T cells often rely on mechanical and enzymatic treatment of tissue biopsies. 9,10It has been demonstrated that isolation methods introduce distortion into T cell number estimation 11 and that in vitro culturing of certain T cells may lead to their activation. 12Therefore, advanced new technologies are used for the analysis of complex gene and protein patterns in situ. 13One challenge is to employ high-resolution analysis of small structures in tissue.Human epidermal Trm cells never exceed 1%-2% of total cells in epidermis 14 and are located throughout the epidermal-dermal border, which complicates the analysis of this subset.
Here, we present a method of using the GeoMx technical platform that allows for in situ analysis of up to 72 proteins (Table 1) on T cells and melanocytes in epidermis, and T cells in dermis, in healthy and vitiligo skin (Figure 1).The method relies on sample pooling and bioinformatics post-processing in order to obtain enough sampling material and successfully pass negative and positive quality control (QC).The diameter of recommended area for proteomic analysis with the GeoMx instrument is 200 μm with a potential range of 50-400 μm in diameter.While the instrument allows for the selection of circular regions of interest (ROI) as small as 10 μm in diameter, T cells and melanocytes scattered through the tissues typically measure up to 10-20 μm in diameter.Therefore, 50-μmwide ROI would primarily map the signal from cells surrounding the cell of interest (Figure 1A).Here, pooling of individual ROIs as a group and the processing of this sample emerges as a viable solution (Figure 1B,C).Such grouping comes at the cost of losing 'true' spatial data, as the obtained counts can no longer be linked to the individually selected ROIs, and does not allow for a single-cell resolution.However, it still gives the opportunity to assess the in situ expression of protein targets on a population level.In our experimental set-up, we pooled together T cells, melanocytes, T cell and melanocyte colocalizations in epidermis, as well as dermal T cells, and demonstrate that the method is able to markedly distinguish them from surrounding microenvironment, enabling their further phenotyping at their site of residence.

| MATERIALS AND METHODS
The general workflow of the method is represented in Figure 1B,C.

| Patient material and FFPE preparation and sectioning
Four-millimetre full skin thickness biopsies were collected from six vitiligo patients or healthy volunteers.From each vitiligo patient, skin was sampled from non-lesional, perilesional and lesional skin (for a total of 18 vitiligo samples from six patients), and one sample was obtained from each healthy volunteer (for a total of six) (Table 2).All patient samples were collected in accord with the Declaration of Helsinki principles after approval of Stockholm region ethical committee, and signed informed consent has been provided (2015/933-23).Biopsies were submerged in 1 mL of Dulbecco's phosphate-buffered saline (DPBS) and fixed in 4% formaldehyde no longer than 2 h after collection.The tissue was dehydrated with the 'Microwave Hybrid Tissue Processor' Logos from Milestone, embedded in formalinfixed paraffin-embedded (FFPE) block and sectioned using Thermos Scientific Micro HM 355S.Three 5-μmthick transverse tissue sections per sample were mounted on a Fisherbrand™ Superfrost™ Plus Microscope Slides and baked in a dry oven (JULABO WS 100, JULABO GmbH, Seelbach, Germany) for 50 minutes at 60°C.To ensure maximal possibility of encountering rarer cell types, up to three sections from the same paraffin block were mounted on the same slide (Figure S1).The slides can be stored up to 1 month before proceeding with antigen retrieval and incubation with staining and profiling antibodies.
Once the staining protocol was completed, four slides (one slide per healthy and one each for vitiligo nonlesional, perilesional and lesional patient-matched skin biopsies) were loaded onto the GeoMx instrument, and we proceeded with ROI selection.

| ROI selection strategy for epidermis and dermis
To minimize signal dilution and obtain cell-specific protein expression, we devised а strategy aimed at obtaining data from individual cells in epidermis and cell clusters in dermis (Figure 2).ROIs can be further subdivided into areas of Illumination (AOI) using pre-made or custom F I G U R E 1 Experimental design and general workflow.(A) In order to obtain enough sampling material and pass signal-to-noise ratio (SNR) quality control, a circular ROI approximately 50 μm in diameter needs to be profiled with the GeoMx instrument.However, since T cells are rare in the skin and have small diameter (approx.10 μm), this would include a lot of the surrounding cell types and environment and dilute the T-cell-specific expression.Therefore, we decided to pool T cells, melanocytes, T cell and melanocyte colocalizations and keratinocyte in the epidermis.In the dermis we pooled T cell clusters and surrounding T cell lacking stroma.(B) General Workflow -a total of six biopsies from healthy volunteers, and six donors matched biopsies from lesional, perilesional and non-lesional skin each of vitiligo patients were obtained.Following FFPE block preparation, three 5-μmthick sections of each sample were mounted on slides, stained and proceeded to ROI selection.(C) Post-sample selection samples were pooled, and hybridization with HYB probes was performed.After nCounter readout, RCC output file is redacted for the GeoMx instrument to verify the experiment.Afterwards, raw data can be exported and analysed.
T A B L E 2 Cell-type availability in patient samples.masks in the GeoMX software.Here, due to their small size, each epidermal cell or colocalization was designated as individual ROI and analysed in entirety, without subdivision.Epidermal ROIs were defined as 'Circular ROI' with a diameter of 10 μm ensuring that the cell of interest (T cell, melanocyte and their colocalization) was covered in the ROI (Figure 2A).Of note, not all cell types were present within all samples (Table 2).Ten random epidermal areas 10 μm in diameter each that showed PanCK expression but did not express CD3 or MelanA were selected to represent the keratinocyte compartment.To obtain the lowest threshold for reliable analysis, at least five cells per condition and cell type were collected.All cells of interest were identified by the expression of cell-specific markers and nuclear staining Syto83 and sequentially annotated.The clustering pattern of dermal T cells (Figure 2B) made individual T cell selection with the circular ROI impossible.Instead, dermal T cell population was collected using 'Geometrical ROI' selection, minimizing acellular areas between closely located groups.To obtain background protein expression, three 660 μm by 785 μm areas lacking T cell staining was selected from the dermal interstitium.Once ROIs for all slides were defined, and the slides were marked as 'finalized' in the GeoMx software, we initiated sample collection from ROIs.Note that if the number of ROIs exceeds the capacity of a 96-well plate, several plates may be used in a single run.In our experiments, between two and three plates were used per sample collection run depending on abundance of target.When exchanging plates from the same sample collection run, full plates were kept at 4°C for <24 hours before next step of manual pooling.After the run is completed, the instrument will generate two files: a Library preparation text file containing information about the area of each ROI, used in the hybridization step, and cartridge definition file (CDF) used to initiate the experiment on the nCounter instrument and later validate the readout raw data.

| Post-sample collection pooling and hybridization
After ROI sample collection was completed, each sample group on the 96-well plate was manually labelled, according to the corresponding ROI group (Figure S2A).To ensure equal working volumes, each collection plate was dehydrated for 30 minutes at 65°C and rehydrated in 7 μL of nuclease-free water.We then proceeded to manually pool each ROI sample group on a separate collection plate (Figure S2B).In order to keep working volumes within a sensible range and ensure that postpooling hybridization does not take more than 24 h, no more than 15 samples from the original collection plates were pooled.Samples were again dehydrated at 65°C, until no visible liquid phase was left in the wells and rehydrated in 7 μL of nuclease-free water.During this step, a salt pellet may form in the wells.This does not affect the final results, and the pellet completely dissolves upon rehydration.We repeated the dehydration and rehydration steps until a single well corresponding to our pooled ROI cell type was present on a 96-well plate.As a result of this ROI pooling, the subsequent analysis addresses the cell types as a group and not at a single-cell level.We then proceeded with hybridization and nCounter instrument run as per GeoMx-nCounter Readout User Manual MAN-10089-07 p.17 -p.26.The nCounter system generates a Recap Compressed Structured Scan Data file.

| CDF and RCC file data structure and editing
CDF and RCC files were opened and edited in Microsoft Windows NotePad Microsoft Office 365 Excel, and final archiving of files was performed using Windows 10 builtin archiving tool (all Microsoft Corporation, Redmond, Washington, United States) (Figure 3A).
In order to proceed to data analysis, at least one slide from the scanned and collected slides needs to be associated with the experiment in the software.A critical step to finalize the experiment is a sample check which the GeoMx performs to validate whether the number of ROIs from the nCounter RCC readout file corresponds to the number of ROIs in the GeoMx CDF file.The machine will accept the file if the number of ROIs from the RCC is equal to or higher than the one in the CDF but will refuse to proceed with any analysis if it is lower.Due to the post-collection pooling of samples, nCounter RCC files will contain information about fewer ROIs than the ones selected in the GeoMx instrument.
To fulfil this criterion, generation of a 'zero counts' RCC file containing no counts for the target proteins needs to be generated and uploaded.To create the RCC file, experiment-specific data need to be obtained from the CDF file generated by the GeoMx instrument at the end of plate collection (Figure 3B).Both CDF and RCC files can be opened by conventional text editors such as WordPad or Notepad, or in Microsoft Excel using the 'Text to Columns' function to separate the columns from the comma delimited file structure of the RCC (Figure S3).The information needed to be extracted from the CDF file is located in the CartridgeID, the Lane ID, and the Gene RLF file sections.
The general structure of RCC file is shown in Figure 3 and Figure S3A.To generate a 'zero counts' RCC file, we copied and edited a file already generated from the nCounter MAX/FLEX RCC during the run of the cartridge containing the pooled samples in Microsoft Excel.We used the 'Text to Columns' function to conveniently visualize the file contents (Figure S3A).In the <Sample Attributes> ID section, after P (indicating 'plate') input the barcode number obtained from the LaneID section of the CDF.Date should correspond to the same date the cartridge was scanned.GeneRLF relates to the RLF (Reporter Library File) used to scan the cartridge and should be always inputted as DSP_V1.0.The <Lane_Attributes> section of the file corresponds to the lanes of the cartridge that were scanned.Input numbers from one to 12 post the comma to simulate the scanned lanes.FoVCount and FoVCounted should be kept the same number as they correspond to the number of fields of view the instrument was instructed to scan, and the actual scanned number and discrepancy there may generate an error and QC flag.Change the Car-tridgeID value, with the value obtained from the CDF file.In <Code_Summary> subsection change, all values in the Accession column to NA and all values in Count column to 0. Values in the 'Name' columns should go from DSP_0001 to DSP_0972.Save the file as *.csv and post-save change the file extension to.RCC.Open the file using WordPad and through the Find and Replace function, change all and to blank space (no value).Generate a total of 12 files corresponding to the 12 lanes of the cartridge, sequentially changing the ID value in <Lane_Attributes> from 1 to 12.After all 12 files have been generated, archive them to a ZIP file using the Windows built in tool following the name convention described in Figure S3B.Proceed to data import on the GeoMx instrument.

| Data import and zero count prune
Once data have been imported from the scanned RCC files with pooled samples, together with the 'zero count' RCC to fulfil the scan to sample match criteria, analysis queue and QC as per GeoMx instrument manual can be performed.On the QC screen, scanned slide and initial heatmap of the entire dataset including pooled ROIs and the 'zero count' files are shown (Figure 4A).Due to the pooling procedure, sample wells no longer correspond to the actual image ROIs on the scanned slide, but to the designated cell type.
In order to remove 'zero count' data and continue further with the analysis of the scanned pooled data, the user may either one by one deselect the samples from the representative wells or select all 'zero count' data from the heatmap and remove it (Figure 4B).Once the initial data is properly curated, the user may proceed with relevant QC step and sample annotation.The samples containing data from the pooled cells will correspond to the cartridge lanes scanned by the nCounter machine.

| Normalization and statistical analysis and visualization of GeoMX data
The geometric mean of epidermal protein expression of each profiling protein was normalized to the geometrical mean of housekeeper protein set (Histone H3, S6, GAPDH) included in the core panel.Since dermal T cells displayed a much more grouped three-dimensional localization and determining nuclear count was challenging, we opted to use the signal-to-noise ratio itself as a normalization parameter.
The GeoMX output data were analysed, and figures and graphs were produced by GraphPad Prism (version 9; GraphPad Software, La Jolla, California, United States).
Mann-Whitney U test was performed when assessing the abundance of specific protein in populations of melanocytes, T cells and keratinocytes.Unsupervised k-mean F I G U R E 3 CDF and RCC file data structure and editing.(A) CDF file structure -after pooling, hybridization and nCounter readout data need to be obtained from the CDF file and edited in the RCC files; otherwise, the GeoMx instrument would refuse to validate the experiment, as the number of ROIs in the nCounter files will be lower due to pooling of samples.Relevant information from CDF file is in lanes labelled 'CartridgeID', 'ID', 'GeneRLF'.(B) RCC file structure -file should be edited with the information from the CDF file in corresponding lanes.Afterwards, all 12 generated RCC files should be archived using Window's built-in archiving tool.
clustering was performed in R, and data were visualized by ComplexHeatmap and ggplot2.Each experiment consisted of four slides -one healthy and one matched nonlesional, perilesional, and lesional vitiligo patient sample.

| Pooling method identifies enrichment of key cell-specific expression markers and successfully discriminates between epidermal cell types
In order to confirm that our pooling strategy generates sensible data and is able to correctly identify the cells of interest, we decided to measure several cell type-specific markers in melanocytes and T cells expression and compare it to the background of keratinocytes and rabbit isotype antibodies in the epidermis (Figure 5).Keratinocytes are by far the most abundant epidermal cell type, 15 so they may be the primary cause of signal dilution in this layer.Each cell type was analysed separately.We observed enriched expression of CD3 in the non-competitive CD3 antibody-labelled T cells in all conditions, as compared to keratinocytes (Figure 5A).CD3 expression in cells with a keratinocyte morphology was close to that of the rabbit IgG, indicating that the observed enrichment is indeed due to the correct identification of T cells and not a product of background.
Next, we wanted to evaluate how well the method discriminates between melanocytes and keratinocytes (Figure 5B), as well as between melanocytes and T cells (Figure 5C), we examined the expression of MART1.7][18] We observed that MART-1 expression was significantly higher in melanocytes when compared to keratinocytes (Figure 5B) and non-specific IgG.
Finally, we decided to test whether the method successfully distinguishes the T cells and melanocytes F I G U R E 5 Pooling method successfully discriminates between epidermal cell types by key expression markers.In order to validate that our pooling method generates sensible cell-type corresponding data, we examined the housekeeper protein (HK) normalized expression of several key cell markers in our pooled samples (Mann-Whitney U test).Representative images (same as Figure 1

) of epidermal ROI selection (right). (A) CD3 expression successfully distinguished between T cells and keratinocytes in all sample types. (B) MART-1 expression distinguished melanocytes from T cells. (C) MART-1 expression distinguished melanocytes from T cells.
themselves.To this end, we investigated the expression of MART1 in melanocytes and T cells and observed statistically significant enrichment of this protein in melanocytes (Figure 5C).The expression in T cell and melanocyte colocalization was observed as intermediate between T cells and melanocytes (Figure S4A) and in the different sample groups we observed an average 10-fold enrichment of the CD3 and MART1 expression when compared to keratinocytes, and between 5-and 30-fold increase of MART1 in melanocytes compared to T cells (Figure S4B).

| Dermal T cells are distinguished from the surrounding interstitial stroma
The dermal layer of the skin is located underneath the epidermis and has a markedly different cellular and intercellular environment. 1Unlike epidermis, this layer is vascularized and T cell presence here is higher and distributed in clusters rather than single cells, which aided our previously described ROI selection (Figure 2B).
When addressing the dermal T cell compartment, we observed statistically significant higher CD3 expression in the T cell as compared to the interstitial tissue in all sampled vitiligo areas and healthy skin (Figure 6A).This confirmed that the proposed pooling method is able to distinguish specific cell populations from the surrounding environment.
Furthermore, we needed to address the discrepancy in area between the T cell ROI (variable based on the number of clusters in skin) and the interstitial ROI (fixed 660 × 785 μm).Since the two areas vary, we would expect that the expression of intercellular matrix components to be higher in the fixed areas.To investigate this, we measured the expression of fibronectin between the two area types.0][21] We observed a significantly higher expression of fibronectin in our T cell barren dermal ROIs compared to the T cell-annotated clusters (Figure 6B).Since fibronectin is a component of the dermal interstitium, our observed fibronectin expression in T cell AOIs likely originates from the surrounding stroma.When addressing the fold change of CD3 and fibronectin in our groups, we observed an average 10-fold higher expression of the former in T cells, and 2-10-fold enrichment of the latter in the healthy and vitiligo biopsies (Figure S5).This further confirms that the observed expression of profiling panel CD3 in the T cells was specific for this cell-type ROI, and our method was able to distinguish between the T cells and their surrounding interstitium.

| Pooling method allows for high-plex analysis of proteins on rare cell populations and colocalizations
After we validated our method on the epidermal and dermal compartments, we decided to investigate the differential expression of vitiligo-associated proteins previously described in the literature (Figure 7). 7,22,23Our selected markers of interest formed three clusters based on protein expression (Figure 7A).Cluster 1 showed increased expression of S100B protein, cluster 2 was predominantly enriched in expression of programmed cell death protein 1 (PD-1), while cluster 3 displayed a more heterogenous protein expression and enrichment of beta-2-microglobuling.S100B is a melanocyte expressed protein with clinical relevance protein in vitiligo, as its serum levels were demonstrated to positively correlate with disease activity. 22The role of PD-1 has been explored both in mouse models and a potential therapeutic target in humans. 23,24It has been noted that patients undergoing cancer therapy blocking PD-1 signalling, a type of checkpoint inhibitor, can develop therapy-induced vitiligo. 25e therefore focused our analysis on these two clinically relevant proteins.When we evaluated their expression in our different epidermal cell groups, statistically significant changes were observed in the colocalizations of the T cells and melanocytes (Figure 7C).The expression of the melanocyte damage protein S100B was increased in colocalizations from lesional compared to healthy skin.PD-1 expression, however, was decreased specifically in the colocalizations in the perilesional group compared with lesional vitiligo.Overall, these changes demonstrate that there is a differential expression of disease-associated proteins in the colocalizations of T cells and melanocytes in the vitiligo skin.

| DISCUSSION
The skin is the largest organ by measured surface area in the human body, and within it resides a vast network of adaptive and innate immune cells. 1 Due to its unique microanatomical composition -a keratinocyte-rich and -dense epidermis, and stromal-rich connective tissue dermal layer, the isolation and characterization of the immune cells in the different tissue compartments remain a technical challenge.0][11][12] While new and improved methods are being developed for the more rapid and sparing recovery, this still disrupts their connection with surrounding microenvironment and other cells. 10,26herefore, a method which enables the in situ profiling of protein expression, without separating the cells from the tissue, may prove to be a particularly valuable instrument in the experimental immunological repertoire.
Several novel methods for in situ gene and protein expression have recently revolutionized the field of spatial visualization. 13Here, we demonstrate a method of pooling individually annotated skin cells and cell colocalizations using the GeoMX spatial transcriptomics platform to overcome the systems requirement for minimum sampling material.While this approach forgoes the exact spatial location of each cell, it allows for analysis of in situ protein expression on a population level and characterization of cell-cell interactions between different areas and conditions.The method requires F I G U R E 6 Dermal T cells are distinguished from surrounding interstitial stroma.In order to validate that our method allows for proper distinction between T cells and dermal stroma despite differences in ROI size, we analysed the signal-to-noise ratio normalized (SNR) expression of two markers (Mann-Whitney U test).Representative images of dermal ROI selection (right).(A) CD3 correctly identifies T cell ROIs as significantly enriched for the protein's abundance.(B) Fibronectin is specifically enriched in dermal ROIs lacking T cells.

F I G U R E 7
Pooling method allows for high-plex analysis of proteins on rare cell populations and colocalizations.After we validated our method provides cell-specific data, we investigated the expression of several disease-related proteins in vitiligo.(A) Unsupervised k-mean clustering revealed three clusters based on expression of either melanocyte damage or immune regulation proteins.(B) Protein enrichment and log-fold change between different skin sample sites.(C) Mann-Whitney U test analysis showed that there is a statistically significant S100B overexpression in T cell and melanocyte colocalizations in lesional compared to healthy skin (0.0381), and decreased expression of PD1 in perilesional compared to lesional skin (0.0476).optimization on several steps, namely sample slide mounting, ROI selection and pooling, and changes in the data structure of the GeoMX primary output file to facilitate further analysis.Due to the dispersed localization of T cells within the human skin, we mounted several consecutive sections of the same FFPE-embedded biopsies on the same slide to increase the chance to observe these cell types.ROI selection strategy was performed based on the specific compartmental architecture in epidermis and dermis.We selected 10 μm circular ROIs to annotate T cells, melanocytes, and T cell and melanocyte colocalizations in the epidermis.For the dermal T cell population, we opted for a geometric ROI finely delineating their clustered position and compared it to T cell barren interstitium.We needed to introduce changes in the CDF and RCC data structure to be able to import and analyse the pooled ROI samples.These changes affect only the way the GeoMX instrument perform equipment and procedural verification and introduce no changes in the actual expression data.Post-import analysis of data easily identifies the introduced changes in the files and can be excluded from further analysis.
In the epidermis, we verified that our method correctly identifies the annotated cell types by cross-referencing their fluorescent marker expression with well-described cell-specific proteins.We observed statistically significant increase in expression of CD3 in T cells and MART-1 in melanocytes compared to the surrounding CD3 and MART-1 lacking keratinocyte environment and also between the two cell types.In dermis, we demonstrated that despite discrepancies of area between T cell clusters and defined interstitium ROI, our method successfully shows statistically higher expression of CD3 in T cells.However, it should be noted that since fibronectin is a component of the environment itself, it can be detected in the T cell AOIs as they are not on a single-cell resolution.Therefore, it is advisable not to use the method in experiments investigating the expression of proteins found in abundance in the surrounding environment compared to target cells.Such contamination may show falsely elevated protein of interest levels.Ultimately our method aided in the identification of several differentially expressed proteins between the different areas in vitiligo and healthy skin, including clinically relevant proteins S100B, a melanocyte intracellular protein suggested as a marker following vitiligo patients, 22 and PD-1 in which inhibition can lead to development of vitiligo in cancer patients undergoing checkpoint immune therapy. 25Taken the large number of proteins that can be simultaneously measured by GeoMx, this method has the potential to provide tissue-wide multiplex analysis of small tissue samples.
Overall, we proposed and validated a method optimization of the GeoMX system which allows for high-plex protein expression analysis of rare and dispersedly distributed cell types.It is important to underline that the method relies on pooling individual cells together and does not allow for single-cell analysis.Furthermore, due to the relatively limited number of morphology markers (3), distinguishing between different T cell subsets such as T reg and T RM is limited, so the potential effect of skin T cell heterogeneity must be considered in the study design.Likewise, previously described melanocyte heterogeneity in human skin is hard to assess with this method. 27Instead, the method still enables spatial protein expression studies in valuable samples in small quantity.It can also serve as a valuable source of supplementary data in large-scale skin atlas projects that rely on tissue disruption or spatial transcriptomics.Whilst promising, mass spectrometry-based spatial proteomics is cumbersome and is based on larger areas captured by laser microdissection and thereby not assessing single-cell types. 28Other antibody-based methods, such as Hyperion, are promising but while preserving single-cell resolution only allows identification of up to 50 proteins in its current set-up.It would be interesting to compare these methods side by side in clinical material. 29While in our study we focused on cells from healthy and vitiligo skin, we believe the method can be applied in multiple other tissues and samples that face the same experimental challenges.

F I G U R E 2
ROI selection strategy.(A)In order to obtain cell-specific expression we annotated each cell type of interest in the epidermis (T cells, melanocytes, colocalizations and keratinocytes) with a 10-μm diameter ROI.Samples were pooled afterwards corresponding to annotation.(B) Due to distinctly different cluster-like localization of T cells in the dermis, we opted for geometric ROI and finely marked all CD3 expressing areas.To obtain background expression, we manually selected three T cell absent 660 × 785 μm areas of the dermal interstitium.

F I G U R E 4
Data import and zero count prune.(A) Following edited RCC files into the GeoMx instrument, the software will display the added 'zero counts' in the expression heatmap flanking the experimental protein data.(B) Deselection of 'zero count' ROIs can be performed either individually from the sample selection panel or as a group from the heatmap panel.
GeoMx protein profiling panel and controls.
T A B L E 1