Single-Cell RNA Sequencing Reveals Tissue Compartment-Specific Plasticity of Mycosis Fungoides Tumor Cells

Mycosis fungoides (MF) is the most common primary cutaneous T-cell lymphoma. While initially restricted to the skin, malignant cells can appear in blood, bone marrow and secondary lymphoid organs in later disease stages. However, only little is known about phenotypic and functional properties of malignant T cells in relationship to tissue environments over the course of disease progression. We thus profiled the tumor micromilieu in skin, blood and lymph node in a patient with advanced MF using single-cell RNA sequencing combined with V-D-J T-cell receptor sequencing. In skin, we identified clonally expanded T-cells with characteristic features of tissue-resident memory T-cells (TRM, CD69+CD27-NR4A1+RGS1+AHR+). In blood and lymph node, the malignant clones displayed a transcriptional program reminiscent of a more central memory-like phenotype (KLF2+TCF7+S1PR1+SELL+CCR7+), while retaining tissue-homing receptors (CLA, CCR10). The skin tumor microenvironment contained potentially tumor-permissive myeloid cells producing regulatory (IDO1) and Th2-associated mediators (CCL13, CCL17, CCL22). Given their expression of PVR, TNFRSF14 and CD80/CD86, they might be under direct control by TIGIT+CTLA4+CSF2+TNFSF14+ tumor cells. In sum, this study highlights the adaptive phenotypic and functional plasticity of MF tumor cell clones. Thus, the TRM-like phenotype enables long-term skin residence of MF cells. Their switch to a TCM-like phenotype with persistent skin homing molecule expression in the circulation might explain the multi-focal nature of MF.


INTRODUCTION
Primary cutaneous T-cell lymphomas (CTCL) comprise a heterogeneous group of peripheral non-Hodgkin's lymphomas (1,2), with mycosis fungoides (MF) as their most frequent clinical entity. Usually, MF shows an indolent course with stable or only slowly progressing lesions confined to the skin, resulting in an overall 5-year survival rate of 70-80% (3). In line with this biological behavior, malignant T-cells show many features consistent with non-migratory, skin-resident memory cells (T RM ) (4). Yet, in some patients, initial patches and plaques develop into tumors, eventually leading to systemic disease, with potential involvement of lymph nodes, blood, bone marrow and internal organs. Increasing expression of exhaustion markers by infiltrating CD4 + and CD8 + T-cells (5), and a shift from a Th1 towards a more Th2-like immune microenvironment are currently assumed to interfere with effective anti-tumor immune responses during progression of disease (6). However, exact mechanisms remain to be elucidated, and only little is known about the actual processes facilitating dissemination of tumor cells to extracutaneous sites. Using high-throughput TCR-b and TCR-g sequencing, Kirsch et al. demonstrated seeding of a clonal population of malignant T RM to distant skin sites and the peripheral blood, and found the malignant clone to descend from a mature T-cell according to the number of rearranged TCR-g genes (7,8). Recently, this hypothesis has been challenged by data from copy number aberration and TCR clonotype analyses using whole exome and whole transcriptome sequencing, that suggest the presence of oligoclonal malignant T-cells in MF lesions (9,10). Single-cell RNA sequencing (scRNA-seq) combined with single-cell V-D-J sequencing now bears the potential to shed new light on these incongruities. By simultaneous determination of clonality through sequencing of the TCR-a and b chains and analysis of differentially expressed genes, malignant and tumor infiltrating cells can be easily discriminated, and their interaction can be readily analyzed (11). Here we profiled skin, lymph node and peripheral blood from a patient with stage IVB MF at single cell resolution. We were able to consistently detect the malignant clone and assess its properties and dynamic interplay with the microenvironment specific to each of these body compartments, demonstrating considerable tumor cell plasticity spanning from T RM characteristics in skin to more T CMlike properties in blood and lymph node.

Patient Recruitment
The study was conducted under a protocol approved by the Ethics Committee of the Medical University of Vienna, Austria (EK 1360/2018). After written informed consent, a 65 years old Caucasian woman with stage IVB (T3 N3 M1 B1) Mycosis fungoides (MF) was included. At time of sampling, the patient did not receive disease-specific treatment. Routine laboratory investigations revealed an LDH of 322U/L (normal range <250U/L) and a CRP value of 11.47mg/dL (normal range <0.5mg/dL).

Sample Acquisition and Cell Sorting
A 6mm skin punch biopsy from an MF lesion was obtained from the right flank. The biopsy was minced with scalpels and digested in collagenase IV (0,5U/ml, Worthington) for 30min at 37°C to obtain a single cell suspension. Lymph node cells were obtained from a small fraction of a diagnostic biopsy from a pathologically enlarged axillary lymph node. PBMC were prepared by Ficoll ® Paque density gradient centrifugation (GE Healthcare). All cell suspensions were frozen at -80°C until further processing. Cell suspensions were thawed and stained with CD7 FITC, CD4 PE, CD3 APC, CD45 ECD and 7-AAD. In the case of skin and lymph node, viable CD45 + CD3 + CD4 + T-helper cells, other CD45 + cells, and CD45-negative cells were sorted on a FACS Aria III (BD Biosciences) at a ratio of 1:2:1. In the case of PBMC, viable CD3 + CD4 + helper T-cells and remaining CD45 + cells were sorted in a ratio of 1:2. Immediately after sorting, cells were subjected to scRNA-seq processing (10X Genomics, Pleasanton, CA) according to the manufacturer's instructions.

Droplet-Based, Single-Cell RNA Sequencing
T-cell receptor (TCR) sequencing and 5' gene expression sequencing was performed using the Chromium Controller and Single Cell 5' Library & Gel Bead Kit (10x Genomics) according to the manufacturer's protocol. Sequencing was performed using the Illumina NovaSeq 4000 platform and the 150bp paired-end configuration.

Data Analysis
RNA sequencing files were preprocessed with the Cell Ranger software from 10x Genomics (version 3.0.2, RRID: SCR_017344). After demultiplexing with the Cell Ranger command 'mkfastq', reads from each transcriptome library were aligned to the human reference genome assembly 'refdata-cellranger-GRCh38-3.0.0' using the 'cellranger count' pipeline. The pipeline generated both a raw UMI count matrix, including counts for all the droplets, and a filtered UMI count matrix, including only those droplets which are likely to contain at least one cell.
Reads from TCR receptor libraries were assembled using the 'cellranger vdj' pipeline in the reference-assisted mode using the vdj_GRCh38_alts_ensembl-3.1.0 reference version. The pipeline also annotated the assembled contigs according to the reference and grouped cells into clonotypes.
Filtered gene expression matrix, TCR amino acid sequences and clonotypes were then used for secondary analysis with R version 3.6.3 (2020-02-29).

Secondary Analysis
Seurat package (version 3.1.4, RRID: SCR_016341) was applied to perform quality control and integrate all samples. The filtering criteria applied to each cell were the number of genes expressed (between 200 and 4000) and mitochondrial gene percentage (less than 12%) in order to discriminate for multiplets and dead cells leaking mRNA but include cell types with naturally higher mitochondrial content (such as macrophages). All cells not satisfying these criteria were discarded. Following QC, data from TCR receptor sequencing and transcriptome sequencing were merged by adding clonotype frequency and CDR3 amino acid sequences to the metadata column of the Seurat objects. All samples were aligned with the standard integration pipeline, as recommended by the Seurat developers (12,13). Briefly, gene expression counts were log-normalized and 2,000 variable features were selected individually for each sample, and used to find integration anchors, and for principal component analysis. Based on explained variance by each principal component (elbow plot), we selected the first 22 principal components as input for dimension reduction and clustering using the Louvain algorithm at a resolution of 0.6. Clusters were visualized in twodimensional space by Uniform Manifold Approximation and Projection (UMAP). The corresponding cell types of clusters were annotated by finding cluster markers with the "FindAllMarkers" command and running the SingleR package (1.0.5) (14). Differential gene expression (logFC>|0.25|, adjusted p-value<0.05) was calculated using the FindMarkers command and the Wilcoxon Rank Sum Test. P-values were adjusted for multiple comparisons with Bonferroni correction. We used scran package to find droplets containing more than one cell (15). The applied approach simulates thousands of doublets by adding together two randomly chosen single cell profiles. For the doublet score calculation cell clustering including the set randomly generated doublets was performed. Then for each cell of the original dataset the number of simulated doublets in their neighborhood was recoded and used as input for score calculation. We used 200 nearest neighbors for each cell. Doublet score was log10 of the ratio between simulated doublet cells and total number of neighbors taken into consideration for each cell.
Calculation of cell cycle scores was performed as implemented in the Seurat package, where gene expression of cell cycle marker genes was combined to a score. The score consisted of 43 genes primarily expressed in G1/S and 55 primarily expressed in G2/M, described in more detail by Tirosh I et al. (16).

Monocle Analysis
Trajectories and Pseudotime were calculated using Monocle 2 version 2.13 (RRID: SCR_016339) (17,18). Briefly, malignant cells (i.e. TRB1+ or TRB2+ cells) of clusters TC-1 and TC-3 of the filtered 'Seurat' object were converted to a CellDataSet and size factors and dispersions were estimated. Low quality cells and genes expressed in fewer than 10 cells were removed and unsupervised clustering was performed after "tSNE" dimension reduction (to 15 dimensions) and by specifying "tissue" as model formula for batch effect removal.
As model formula input for the first step of trajectory construction, differentially expressed genes between the tissues skin, blood and lymph node were used. In the next step dimensions were reduced with DDRTree and cells were ordered along the constructed trajectory and colored according to pseudotime, tissue of origin or cell cycle phase. For each branching point of the trajectory, we did branched expression analysis modelling (BEAM) (19) and genes with q-value <0.00001 were displayed in a heat map, after eliminating ribosomal and mitochondrial genes.

Copy Number Variation Analysis
Copy number variation analysis was performed using inferCNV of the Trinity CTAT Project (https://github.com/broadinstitute/ inferCNV). The software compares gene expression values across genomic positions between case and control cells to identify genomic regions with consistent higher or lower signal intensities. We used a window of 201 for moving average smoothing and minimum of 5 cells per gene to include the corresponding gene into the analysis. CNVs were predicted with a Hidden Markov Model. We used a 6-state CNV model, attempting to predict either complete loss, loss of one copy, no change, addition of one copy, addition of two copies or addition of more than two copies.

Resource Availability -Data and Code Availability
The 10X Genomics datasets generated during this study are available via Gene Expression Omnibus GSE165623. The published article includes detailed descriptions on how publicly available coding pipelines were used during this study. The exact code is available from the corresponding author upon request. Further information and requests for resources and reagents should be directed to the Lead Contact, Patrick M. Brunner (patrick.brunner@meduniwien.ac.at).

Patient Characteristics
A 65-year-old female patient suffering from MF had been in care at our department for a total of 14 years. She was originally referred in 2006 with stage IIB (T3 N0 M0 B0) disease with patches, plaques and tumors on her head and lower extremities. The Modified Severity Weighted Assessment Tool (mSWAT) showed a severity score of 8.2, indicating generally limited skin involvement. Until 2012, treatments consisted of topical corticosteroids, involved site radiation therapy (ISRT), systemic retinoid treatment (bexarotene), total skin electron beam (TSEB) radiation therapy, and gemcitabine chemotherapy, which led to temporary remission of disease. After seven years of loss to follow up, she returned to our clinic in 2019, presenting with generalized patches, plaques and tumors, and an mSWAT of 108. Skin histopathology revealed a dense subepidermal infiltrate of small to medium sized lymphocytes positive for CD2, CD3, CD4, PD-L1 and TCRbF1, partial positivity for CD5, and negative stainings for CD7, CD30 and PD-1. Ki67 staining showed a proliferation index of approximately 20% ( Figure 1A). In peripheral blood, the absolute leukocyte count was 7400/µl (normal range 4000-10,000), with 885 T-cells (CD3) per microliter (data not shown). 13% of blood T-cells (ie 1.6% of all leukocytes) showed an aberrant phenotype of high forward scatter (FSC-A), elevated CD4 and decreased CD7 levels ( Figure  1B), with negativity for CD5 and CD30 (data not shown). Contrast-enhanced computed tomography (CT) of the chest, abdomen and pelvis revealed pathologically enlarged axillary, para-aortic, iliac and inguinal lymph nodes, as well as hepatic and pulmonary lesions suspicious of CTCL manifestations ( Figure 1C). Routine histologic examination of a right axillary lymph node showed paracortical infiltration of conspicuous atypical small-to-medium sized lymphocytes, positive for CD2, CD3, CD4 and TCRbF1, with partial loss of CD5 and CD7 ( Figure 1D), consistent with lymph node involvement of MF (Dutch grade N3). These cells were negative for PD-1, with a proliferation index of up to 50% (data not shown). There was no histological evidence for large cell transformation (LCT) neither in the skin nor the lymph node. In sum, the patient was diagnosed with stage IVB (T3 N3 M1 B1) disease.

Skin, Blood and Lymph Node Profiling Using Single-Cell RNA Sequencing
To better understand tumor cell characteristics across tissues in this MF patient, we performed single-cell RNA sequencing  (scRNA-seq) of cells sorted from involved skin, blood and lymph node, as outlined in Figure 2A. Sequencing data were filtered for low-quality cells and normalized, yielding 1,215 lymph node, 3,301 blood and 4,512 skin cells ( Table S1). Clustering of these three compartments followed by visualization using uniform manifold approximation and projection for dimension reduction (UMAP) (20) revealed 19 distinct cell clusters ( Figures 2B, C). We assigned cell identities on the basis of canonical markers ( Figure 2D) and identified the top 10 upregulated genes (according to average log fold change and smallest adjusted p-value) for each cluster in comparison to all other clusters ( Figure S1 and Table S2). We found PTPRC/ CD45-negative cells consistent with endothelial cells ( Two T-Cell Clones Are Consistently Expanded in Skin, Blood and Lymph Node, Likely Derived From One Single Tumor Cell When displaying cells separately for lymph node, blood and skin, we found most T-cell clusters to be present in all of these three body compartments ( Figure 3A). By using 5' scRNA-seq that allows simultaneous V-D-J sequencing of the T-cell receptor (TCR) alpha-and beta-chains (TRA and TRB), we overlaid TCRpositive cells onto UMAP plots (colored cells in Figure 3B). Generally, TRA and/or TRB sequences were detected in 2,952 cells which showed a polyclonal pattern in 45% of TCR-bearing cells ( Figure 3B, labelled in blue, and Table S3). Similar to protein expression measured by immunohistochemistry ( Figures 1A, D), the putative malignant cells expressed CD3D and CD4 but lacked CD5 and CD7 ( Figures S2A, B). Further analysis of the TCR sequences revealed two clones to be strongly expanded at 46% and 9% of all TCR + cells, which were confined within the CD4 + T-cell clusters TC-1, TC-3 and TC-4 (top expanded clones marked in red and green, Figure 3B). The two TRB chains of these two clones differed by only one amino acid, namely CASSQDRALENTIYF ("TRB1") and CASSQDRTLENTIYF ("TRB2"), due to a single nucleotide change (guanine to adenine). These TRB chains were mainly paired with the TRA chain CAVDHARLMF, and in <5% of clonal cells also with CALSKKPGRKAYLRT (Table S3). TC-1 in skin contained 79% TRB1 + and 20% TRB2 + cells, with comparable distributions in blood (TRB1: 82%, TRB2: 17%) and lymph node (TRB1: 73%, TRB2: 26% of all cells) within this cluster ( Figure  3C). The much smaller cluster TC-4 was sufficiently present only in skin and blood, with roughly the same distribution of TRB1 + and TRB2 + cells. In cluster TC-3, only skin showed largely clonal TRB1 and TRB2 percentages, while blood cells were mostly polyclonal ( Figures 3B, C).
The expression of two distinct TRB within the clonally expanded T-cell population either represents spontaneous mutation from one single clone, or the presence of two independent clones (22). When comparing gene expression of cells harboring TRB1 versus TRB2, we did not detect any significant differences (at an adjusted p<0.05, data not shown). Furthermore, we also used inferCNV to identify copy number variations between these two populations (23). Also on this level we did not find relevant differences between TRB1 and TRB2 clonotypes, but identified similar gains on chromosomes 5 and 7 in both clones ( Figure 3D and Table S4), the latter being consistent with observations in previous CTCL whole exome sequencing studies (24). Taken together, these findings suggest that both expanded clones have a common T-cell ancestor. Based on these observations we conclude that clusters TC-1, TC-4 and in part TC-3 represent the malignant CTCL clone, while clusters TC-2 and TC-5 harbor benign helper and cytotoxic T-cells, respectively. To further define characteristics of these cells, we calculated differentially expressed genes between polyclonal, presumably benign CD4 + helper T-cells (TC-2) and clusters TC-1, TC-3 and TC-4 (Table S5). We observed substantial transcriptomic differences between these three clusters ( Figures 3E, F, and Table S6). Cluster TC-3 contained mostly proliferating cells, as evidenced by the upregulation of MKI67 ( Figure 2D and Table S5). Despite consistent CD3D expression in cluster TC-4, most cells co-expressed markers typically found in myeloid cells such as CD14, CD68, and CD1C (Table S5), and overall RNA content was approximately doubled in TC-4 when compared to other, non-proliferating cells ( Figures S2C, D), suggesting either the presence of technical doublets, or the engulfment of T-cells by professional phagocytes (25), a question that cannot be definitively answered with the current dataset.

Malignant Cells Display Markers of Dermal
Tissue Resident Memory T-Cells, and Simultaneously Express Th2, Th17 and Th22-Associated Cytokines in a Tumor-Permissive Microenvironment Malignant cells in MF are thought to be closely related to tissueresident memory T-cells (T RM ), given many phenotypic similarities (26). In our patient, clonally expanded cells in cluster TC-1 were CD4 + CD69 + ITGAE/CD103 -( Figure 4A) suggesting their close relation to dermal, but not epidermal  T RM (27). In addition, these cells expressed the skin homing molecule CCR10, but only weakly CCR4 ( Figure 4A). Another chemokine receptor, CXCR3, was specifically expressed in cells of cluster TC-1 ( Figure 4A), but was hardly present in benign helper T-cells of cluster TC-2 (data not shown). In cells from TC-1, representing the largest cluster of malignant cells, we found increased expression of various cytokines in comparison to benign helper T-cells from cluster TC-2, including type 22 (IL22), type 2 (IL4, IL13), and type 17 (IL26) cytokines, as well as IL21 and IL32 ( Figures 4B, C, Table S5). In addition, they were rich in LTA, TNF, CSF2, and GNLY ( Figure 4C and Table  S5), creating a highly inflammatory environment that could not be attributed to a single classic T helper cell subset (28). We also found increases in the T-cell exhaustion-associated markers TIGIT and the CTCL-associated markers TOX (29) and MIR155 (30) ( Figure 4D). IL22, IL32 and GNLY, as well as TOX and TIGIT were generally increased in skin, blood and lymph nodes, when compared to benign helper T-cells or cytotoxic T-cells ( Figures 4C, D), consistent with a proinflammatory malignant phenotype spanning several body compartments (Table S7). By contrast, Th2 cytokines (IL4 and IL13) and associated mediators (IL21, TNFSF14/LIGHT) (31, 32), but also IFNG and MIR155 were mostly upregulated in malignant cell from skin ( Figures 4C, D, Figures S3A, B and Table S8). The Th17-associated cytokine IL26, colonystimulating factor-2 (CSF2), the co-stimulatory molecule CD40LG and tumor necrosis factor-alpha (TNF) were upregulated both in lymph nodes and skin, but less so in blood ( Figures 4C, D). Expression of the transcription factors AIRE (previously associated with negative selection of self-recognizing T-cells) (33) and TCF7 (associated with central memory T-cells) (34) discriminated the malignant cells in peripheral blood from those in skin ( Figure 4D). CD27, previously described to be absent from terminally differentiated memory T-cells (35), was universally downregulated in lymphoma cells ( Figure 4A). Taken these data together, we found that malignant clones harbored a multitude of pro-inflammatory mediators, in stark contrast to benign T-cells. Next, we assessed factors involved in the control of cell activation, namely inhibitory receptors such as CTLA4, PDCD1 (PD-1), HAVCR2 (TIM-3), LAG3, and CD101 (IGSF2) (36). In line with their strong pro-inflammatory phenotype, clonally expanded cells showed absence of these markers, with only a partial expression of CTLA4 ( Figure 4E). By contrast, benign helper T-cells (TC-2) expressed PDCD1 and partly LAG3 and CTLA4, while cytotoxic T-cells (TC-5) highly expressed LAG3 ( Figure 4E), consistent with their less inflammatory phenotype. These data indicate that the lack of inhibitory molecules might be involved in the aberrant inflammatory cytokine pattern of clonally expanded cells, suggesting a cell-intrinsic contribution of hyperactivation in these cells.
Besides the malignant clone, benign inflammatory cells surrounding CTCL cells have also been assumed to play a role in control versus progression of the disease (6). Activated CD8 + cytotoxic T-cells have been postulated as major anti-CTCL cells in early lesions when the disease is still confined to the skin, but lose this capacity when the disease progresses (6). CD8A + cells in cluster TC-5 were rich in the killer molecules GZMA ( Figure  4C), GZMK, GZMH, and GZMB ( Figure S1), and were main producers of IFNG in skin and lymph nodes ( Figure 4D and Table S2), the type-1 lead cytokine associated with anti-CTCL properties (37). However, CD8A + cells also ubiquitously showed highest levels of CCL5 (RANTES) ( Figure 4F and Table S2), a chemokine previously shown to maintain CD4 + T RM cells after infection or sensitization (38), but that also attracts monocytes which were shown to promote the survival of MF cells in a mouse model (39). NK/NKT-cells were also rich in GZMA, GZMB, CCL5 and IFNG ( Figure S1 and Table S2), but did not show significant differences between skin and lymph nodes ( Table S9). Benign helper T-cells (TC-2) displayed significantly increased levels of IL2RA and TNFRSF18 (GITR) in skin compared to lymph node tissue ( Figure S3C and Table S10), consistent with a more regulatory phenotype. Compared to blood, they showed increases in the checkpoint molecule CTLA4, most likely due to increased levels of FOXP3 + regulatory T-cells within this cluster ( Figure S3D and Table S10). Conversely, levels of the cytotoxic molecule TNFSF10 (TNF-related apoptosis-inducing ligand TRAIL) were decreased compared to blood cells ( Figure S3D and Table S10), potentially marking decreased anti-tumor activity (40). Benign skin versus blood helper T-cells also showed increases in ICOS (CD278) ( Figure S3D and Table  S10), a costimulatory molecule present in activated T-cells especially associated with type 2 inflammation (41), corroborating a more Th2 versus Th1-skewed skin phenotype, consistent with CTCL progression (6,42). These data suggest that both, helper and cytotoxic T-cells surrounding CTCL cells in skin, have several regulatory features and may thus help in sustaining a pro-tumorigenic environment.

Myeloid cells, including dendritic cells, play a central role in Tcell biology, including in vitro survival of malignant T-cells (39).
We found a large population of ITGAX/CD11c dendritic cells (DC-1) in skin, strongly positive for the Th2-associated markers amphiregulin (AREG), and the chemokines CXCL2, CXCL3 and CXCL8 ( Figure S1 and Table S2). While previous reports suggest an abundance of immature dendritic cells in advanced CTCL lesions implicated in tumor progression (39), DC-1 highly expressed the maturation marker CD83 ( Figure 5A). Nevertheless, they were the main producers of the regulatory cytokine IL10 (Figure 5B), and positive for VEGFA ( Figure 5C), a major angiogenetic growth factor associated with advanced CTCL (43).
Dendritic cells from cluster DC-2 harbored the maturation marker LAMP3 ( Figure 5D) and were present only at relatively small frequencies in comparison to all other myeloid cells ( Figure 2B), but were also found in lymph node tissue, consistent with previous reports (44). They were characterized by peak expression of Th2-associated chemokines (45) CCL17 and CCL22 (Figures 5E, F). DC-2 also expressed IL15 both in  skin and lymph node tissue ( Figure 5G), a cytokine previously shown to promote CTCL (46) and implicated in the maintenance of T RM (47), as well as the receptors for tumor cell derived TIGIT, TNFSF14 (LIGHT) and CTLA4, namely PVR, TNFRSF14 and CD80/CD86, respectively ( Figures 5H-K). They also expressed IDO1 ( Figure 5L), coding for the enzyme indoleamine-pyrrole 2,3-dioxygenase that has immunosuppressive and T-cell modulatory functions, potentially involved in CTCL pathogenesis (48). Skin macrophages (MФ) were also positive for Th2promoting chemokines such as CCL18, CCL13 and CCL17 ( Figure S1, Figure 5E and Table S2). Blood monocytes in clusters MC-1, MC-2 and MC-3 were found only at traces in lymph node and skin tissues, and generally lacked chemokine expression or activation markers such as CD83 ( Figure 5A). Taken together, myeloid cells were largely skewed towards more type 2 inflammation and regulatory mediators, consistent with a more pro-CTCL tumor microenvironment (6).

Trajectory Inference Reveals Transcriptomic Heterogeneity Within the Malignant Clone, Reflecting Differences in Inflammatory and Migratory Properties
To gain further insight into the relationship of cellular fates in skin, blood and lymph node (19), we applied trajectory inference analysis using the Monocle 2 algorithm on malignant T-cells ( Figure 6A). In the resulting trajectory, cells from peripheral blood and skin were found at opposing ends, while lymph node cells were mostly scattered along the manifold ( Figure 6B), suggesting transcriptomic properties related both to skin and blood cells. The influence of cell cycle genes in proliferating cells was not the major determinant for pseudotime alignment since cycling cells were found on all branch ends, and were spread throughout the trajectory ( Figure 6C). To discern genes responsible for the construction of the manifold, branched expression analysis modeling (BEAM) was performed ( Figure 6D). Genes with a q-value<10 -5 were considered for further analyses. For the main branching ("cell fate 1" versus "cell fate 2"), 500 genes fulfilled this criterion (Table S11). There was also a small side branch within "cell fate 1" containing a few lymph node cells ( Figure 6B), but it did not elicit significantly regulated genes (all q-value>0.1) and was therefore not further analyzed (data not shown). In blood, the transcription factor KLF2 as well as the target downstream gene S1PR1 were upregulated ( Figures 6D-F), consistent with the loss of tissue retention (49). Concordantly, the transcription factor TCF7, which is usually downregulated on T RM and is important for T CM differentiation (49), was expressed on malignant cells in the blood, together with the T CM surface markers and lymph node homing receptors SELL and CCR7 ( Figure 6D).
Importantly, malignant cells of the peripheral blood expressed the skin-homing receptor CLA as shown by flow cytometry (Figure S4), consistent with corresponding gene upregulation of both SELPLG and the fucosyltransferase FUT7 (50) (Figure 6D), indicating their retained capability to home to the skin (51,52).

DISCUSSION
Reliable tracing of individual T-cell populations in humans can be a major challenge. scRNA-seq with simultaneous V-D-J sequencing of the TCR now allows for the investigation of specific T-cell clones and their transcriptomic behavior throughout various body compartments. In our patient, malignant cells of the skin showed all the characteristics of benign T RM cells as described before, with expression of the skin homing molecule CCR10 as well as CD69 accompanied by downregulation of the transcription factor KLF2 (57), and ample production of cytokines (58). In blood, by contrast, these MF cells showed a loss of the tissue retention signature as evidenced by the upregulation of KLF2 and, consequently, upregulation of S1PR1 and downregulation of CD69, indicative of a shift towards a more T CM -like phenotype, consistent with increased TCF7 expression (57). In accordance, lymph node homing receptors (SELL and CCR7) were upregulated. Until recently, T RM were believed to be a sessile, non-recirculating, terminally differentiated population restricted to their non-lymphoid tissue of residence, such as the skin (59,60). This dogma has recently been challenged by observations in mice showing equilibration of skin resident memory T-cells upon parabiosis over 12-16 weeks (38). Importantly, the malignant clone in our patient displayed exactly the same shift in phenotype as observed in T RM in a murine model, demonstrating that CD8 + T RM from skin can rejoin the circulation after antigenic stimulation or activation (61). They could then either maintain their phenotype and home back to their tissue of origin, or even differentiate into T EM and T CM cells (61). In a xenograft mouse model, T RM cells had the capability to migrate out of the skin, but showed preferential homing back to a human skin equivalent (62). In line, blood CD4 + CD69 -CD103 + cells were transcriptionally and clonally related to skin CD4 + CD69 + CD103 + T RM in this model system (62). In our MF patient, malignant cells in blood seemed to also retain their skin-homing capabilities, indicated by maintained CLA and CCR10 expression. The systemic dissemination of tumor cells in MF could therefore reflect the physiological migratory behavior of tissue resident memory Tcells, and might help to explain the clinical observation that MF lesions can spread to virtually any skin site, which, intuitively, needs to happen via the bloodstream. Importantly, it is not yet understood which factors regulate the egress or the reentry of T RM from non-lymphoid tissues. Upregulation of certain promigratory mediators (LAIR2, CD99, FXYD5) by a subset of skin lymphoma cells might contribute to the altered migratory behavior of these cells, but this hypothesis needs corroboration in functional studies.
T RM cells are poised to rapidly react after antigenic rechallenge, a process that is assumed to be under stringent control by inhibitory receptors (36), that can also be present on CTCL cells (5). In our patient, these receptors were largely absent, except for CTLA4. Secreted CTLA4 can act on dendritic cells inducing the expression of IDO and rendering them tolerant (63). Additionally, the resulting tryptophan deprivation and generation of toxic metabolites have been shown to preferentially induce apoptosis in Th1 cells and less so in Th2 cells, which would favor a pro-tumorigenic environment (6). In our patient, a distinct population of LAMP3 + dendritic cells expressed the CTLA4 receptors CD80 and CD86, as well as IDO1. They also expressed TNFRSF14 and PVR, the receptors for TNFSF14 (LIGHT) and TIGIT, respectively, which are both synthesized by the tumor cells, and might additionally enhance local Th2 immune skewing (31,64). The main population of skin

Rindler et al.
Plasticity of Cutaneous Lymphoma Cells myeloid cells (DC-1) were CD83 + co-expressing VEGFA and small amounts of IL10, features that are implicated in the sustainment of a pro-tumorigenic milieu (65,66). This study is limited by data derived from only one patient, at one single (advanced) stage of disease. Also, due to scarcity of sample, we only had few lymph node cells available. Given the substantial heterogeneity not only between CTCL patients, but also within the malignant clone (11,67,68), our findings need corroboration in additional patients and subtypes of CTCL. Nevertheless, the general possibility to trace a defined clone throughout three human body compartments sheds new light on the transcriptomic plasticity within these cells, which might be directly linked to their migratory behavior. Future studies will need to clarify the role of cell intrinsic mechanisms versus the impact of the tissue microenvironment in this regard.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ncbi. nlm.nih.gov/geo/, GSE165623.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Ethics Committee of the Medical University of Vienna, Austria. The patients/participants provided their written informed consent to participate in this study.   for T-cell clusters, according to highest log fold change ordered by smallest adjusted p-value using Wilcoxon Rank Sum Test with Bonferroni correction, as compared to the rest of the dataset, calculated separately for skin, blood and lymph node (LN) samples.

AUTHOR CONTRIBUTIONS
Supplementary Table 8 | Differential gene expression analysis within cell types of scRNA-seq analyses as defined by logFC>|0.25| and adjusted p-value<0.05, using Wilcoxon Rank Sum Test and Bonferroni correction. Comparison of lymph node (LN) vs skin, and skin vs blood (PBMC) samples within cluster TC-1.
Supplementary Table 9 | Differential gene expression analysis within cell types of scRNA-seq analyses as defined by logFC>|0.25| and adjusted p-value<0.05, using Wilcoxon Rank Sum Test and Bonferroni correction. Comparison of lymph node (LN) and skin samples within the NK cell cluster.