Clonal expansion of intra‐epithelial T cells in breast cancer revealed by spatial transcriptomics

The spatial distribution of tumor‐infiltrating lymphocytes (TIL) predicts breast cancer outcome and response to systemic therapy, highlighting the importance of an intact tissue structure for characterizing tumors. Here, we present ST‐FFPE, a spatial transcriptomics method for the analysis of formalin‐fixed paraffin‐embedded samples, which opens the possibility of interrogating archival tissue. The method involves extraction, exome capture and sequencing of RNA from different tumor compartments microdissected by laser‐capture, and can be used to study the cellular composition of tumor microenvironment. Focusing on triple‐negative breast cancer (TNBC), we characterized T cells, B cells, dendritic cells, fibroblasts and endothelial cells in both stromal and intra‐epithelial compartments. We found a highly variable spatial distribution of immune cell subsets among tumors. This analysis revealed that the immune repertoires of intra‐epithelial T and B cells were consistently less diverse and more clonal than those of stromal T and B cells. T‐cell receptor (TCR) sequencing confirmed a reduced diversity and higher clonality of intra‐epithelial T cells relative to the corresponding stromal T cells. Analysis of the top 10 dominant clonotypes in the two compartments showed a majority of shared but also some unique clonotypes both in stromal and intra‐epithelial T cells. Hyperexpanded clonotypes were more abundant among intra‐epithelial than stromal T cells. These findings validate the ST‐FFPE method and suggest an accumulation of antigen‐specific T cells within tumor core. Because ST‐FFPE is applicable for analysis of previously collected tissue samples, it could be useful for rapid assessment of intratumoral cellular heterogeneity in multiple disease and treatment settings.

T cells, B cells, dendritic cells, fibroblasts and endothelial cells in both stromal and intraepithelial compartments. We found a highly variable spatial distribution of immune cell subsets among tumors. This analysis revealed that the immune repertoires of intraepithelial T and B cells were consistently less diverse and more clonal than those of stromal T and B cells. T-cell receptor (TCR) sequencing confirmed a reduced diversity and higher clonality of intra-epithelial T cells relative to the corresponding stromal T cells.
Analysis of the top 10 dominant clonotypes in the two compartments showed a majority of shared but also some unique clonotypes both in stromal and intra-epithelial T cells.
Hyperexpanded clonotypes were more abundant among intra-epithelial than stromal T cells. These findings validate the ST-FFPE method and suggest an accumulation of antigen-specific T cells within tumor core. Because ST-FFPE is applicable for analysis of previously collected tissue samples, it could be useful for rapid assessment of intratumoral cellular heterogeneity in multiple disease and treatment settings.

Whats new?
The abundance and spatial distribution of tumor-infiltrating lymphocytes (TIL) in breast cancer tumors correlates with outcome. Here, the authors demonstrated the feasibility of a spatial transcriptomics method to analyze formalin-fixed paraffin-embedded tissue samples, or ST-FFPE, in triple-negative breast cancer. Using this method, they observed the cellular composition of the tumor microenvironment and found that intra-epithelial TIL showed reduced diversity and increased clonality compared with stromal TIL. In addition to studying TIL populations, the method could have clinical usefulness for predicting response to immunotherapy.

| INTRODUCTION
Solid tumors are composed of cancer cells and non-cancer cells such as immune cells, fibroblasts and endothelial cells, 1,2 whose proportions and state vary from tumor to tumor. The abundance of tumorinfiltrating lymphocytes (TIL) correlates with a better outcome in various cancers. Importantly, the prognostic and predictive value of TIL depends on their spatial localization. 3 In breast cancer, TIL that are located in the stroma between cancer cells and do not directly contact cancer cells, namely stromal TIL (sTIL), have a prognostic value. 4 Assessment of the abundance of sTIL by hematoxylin and eosin (H&E) staining is now implemented in routine clinical practice. 5 In contrast, in other cancers such as melanoma, it is the presence of intra-epithelial CD8 + T cells, that is, lymphocytes located in tumor core (cTIL) and directly interacting with cancer cells with no intervening stroma, that has prognostic value 6 and is predictive of response to PD-1 blockade, 7 highlighting the importance of spatial analysis of tumor immune microenvironment.
Stromal and intra-epithelial TIL can be highly heterogenous in terms of composition, phenotype and antigen specificity. Notably, a substantial proportion of TIL are bystander T cells, which are specific to diverse epitopes unrelated to tumor antigens such as viral antigens. [8][9][10] Hence, it is important to assess the tumor antigen-specificity of T cells, since successful cancer immunotherapy requires their reactivation and clonal expansion of tumor-reactive T cells. 11,12 Because antigen specificity comes from T-cell receptors (TCR), TCR repertoires are actively characterized in order to identify tumor-reactive T cells.
For instance, the increased prevalence of TIL sharing the same TCR correlates with a better response to cancer immunotherapy, [13][14][15] suggesting that these TIL include T cells reactive to tumor antigens.
Laser-capture microdissection (LCM) is a suitable tool for studying spatially distinct regions of interest such as tumor core and adjacent stroma. [16][17][18] Here, we developed ST-FFPE, a robust spatial transcriptomics method that combines LCM with RNA extraction, exomecapture and sequencing, and can be used for the analysis of formalinfixed paraffin-embedded (FFPE) tumor samples. We demonstrate the relevance of this approach by investigating the spatial distribution of clonally expanded T cells localized within tumors from TNBC patients.
Our method can elucidate intratumoral immune cell dynamics with potential implications for cancer immunotherapy.

| Laser capture microdissection of FFPE tissue
Polyethylene terephthalate (PET) slides were pretreated by coating with 300 μl poly-lysine (Sigma-Aldrich, St-Louis, United States) and exposure to UV for 30 min. Consecutive sections from FFPE blocks were cut at 8 μm, floated onto a 45 C water bath and mounted on PET slides. After drying in a slide holder for 30 min at room temperature, tissue sections were dewaxed in xylene, rehydrated with decreasing concentrations of ethanol, stained with hematoxylin, dehydrated with increasing concentrations of ethanol, and cleared in xylene (Detailed timing and concentrations in Table S1 and Section 3). Eight protocols were tested and the optimal protocol was b: briefly, deparaffinization was performed in fresh xylene for 1 min twice followed by 100% ethanol for 1 min, 95% for ethanol 1 min and 75% ethanol for 1 min. The slides were transferred into diethyl pyrocarbonate (DEPC) water for 2 min before staining with hematoxylin for 2 min. Subsequently, slides were rinsed in DEPC water until they became clear before undergoing dehydration in 75% ethanol for 1 min, 95% ethanol for 1 min, 100% ethanol for 2 min and xylene for 1 min.
A minimal surface of 2.5 mm 2 of tumor core (TC), fibroblasts (Fib) and sTIL was microdissected using the Laser Microdissection Systems LMD7000 with the laser parameters (laser power of 39 mW, a wavelength of 349 nm, pulse frequency of 664 Hz and pulse energy of 120 μJ) and collected separately in RNase-free tube. For each microdissected block, a consecutive 4 μm thick tissue section was stained with H&E to help recognize tissue's morphology under the LMD7000 microscope ( Figure 1).   We first optimized the tissue preparation process before LCM on three TNBC blocks. We aimed to shorten the H&E staining procedure in order to reduce further degradation of nucleic acids, 19 while conserving maximal morphological contrast between cell subtypes. Thus, we tested eight different protocols (a-h) previously described for LCM procedures [20][21][22] with three different dyes (hematoxylin, eosin and cresyl violet) and varying durations of incubation in xylene and ethanol ( Figure S1). The optimal staining was obtained with protocol b (see methods) that allowed the best morphological distinction between the three cell subsets (sTIL, TC and Fib; Figure S2) while having the shortest exposure to xylene that partially degrades RNA. 19 Second, we aimed to optimize RNA extraction from FFPE samples in view of subsequent RNA-sequencing (RNA-seq), for which we had to choose the most appropriate protocol. In highly degraded and small amounts of RNA, RNA-seq via poly(A) selection does not perform well because poly(A) tail can be lost. 23 Exome-capture protocol has been previously shown to perform better than ribosomal RNA depletion, 23 with minimal differences when comparing FFPE and matched FF samples. 24 Extracting RNA from 2 mm 2 microdissected area of TC using manufacturer's protocol (eg, elution volume of 30 μl) resulted in an undetectable yield ( Figure S3A). Therefore, we introduced several modifications to this protocol, by reducing the eluted volume to 15 μl, adding mechanical digestion by stirring (400 rpm) and concentrating the volume using a vacuum concentrator centrifuge to reach a final volume of 4 μl.

| RNA extraction, quantification and fragment size analysis
This optimized protocol rendered RNA detectable (yield of 18-40 ng).

Next, we tested four different durations (15 min, 1 h, 3 h and overnight)
of enzymatic digestion with proteinase K. RNA yield tended to increase with prolonged digestion, whereas RIN was not affected and DV 200 initially reached a maximum at 3 h but declined after overnight digestion ( Figure S3B-D). Thus, 3 h was set as the optimal duration of tissue digestion. These modifications led to the final protocol ( Figure S3E).
We used quantitative RT-PCR (qRT-PCR) to assess the expression of genes to identify those that are significantly upregulated in TC vs sTIL and sTIL vs TC. Ten genes were evaluated: five genes upregulated in sTIL (CD28, CCR7, CD79B, FCMR and PAX5) and five genes upregulated in TC (ELF3, MAL2, MUC1, TFAP2A and GPR37) ( Table S7). The qRT-PCR results confirmed our RNA-sequencing data ( Figure S6A,B).

| Spatial distribution of immune cell subsets in distinct tumor compartments revealed through deconvolution with MCP-counter
We estimated the abundance of immune cell subpopulations in both TC and sTIL compartments by deconvoluting RNAseq data with Microenvironment Cell Population (MCP)-counter. 28 The spatial distribution of immune cells was variable across the seven cases of TNBC. For instance, CD8 + T-cell abundance estimate was high in microdissected TC from case 73 (full gray arrow, Figure 3A) and low in case 69 (full black arrow, Figure Figure 3C) where they were rather present at tumor margins. High spatial variability was also observed in the monocytic lineage. For instance, the monocyte abundance estimate was high in microdissected TC from case 65 (empty gray arrow, Figure 3A) and low in case 88 (empty black arrow, Figure 3A). Multiplexed IHC confirmed the accumulation of intraepithelial CD68 + cells in this case, whereas they were mainly in the stromal compartment surrounding TC in case 88 ( Figure 3C). Overall, the spatial distribution of immune cell subsets by transcriptomics was concordant with multiplexed IHC.

| Spatial transcriptomics revealed clonal expansion of intra-epithelial T cells as compared with matched stromal T cells
We took advantage of the high potential of the spatial transcriptomics method to address biologically relevant questions and investigate the immune repertoires of intra-epithelial T and B cells, that is, located in TC (cTIL), and matched stromal T and B cells (sTIL). Interestingly, the immune repertoires of cTIL were systematically less diverse than sTIL in the seven TNBC cases (P = .02, paired Wilcoxon test; Figure 4A), while their clonality was higher in six out of seven cases (P = .03125, paired wilcoxon test, Figure S7). In order to further study the diversity and clonality of T-cell receptor (TCR) repertoire of intra-epithelial and matched stromal T cells, we performed TCR-sequencing 29 on microdissected TC (that contains cTIL) and sTIL from seven fresh frozen TNBC samples (Table S8). Diversity was reduced in cTIL as compared with matched sTIL in six out of seven cases (two-tailed paired T test P = .008; Figure 4B). Conversely, clonality of cTIL repertoires was consistently higher (two-tailed paired T test P = .021; Figure 4C), and tumor core samples harbored a higher fraction of hyperexpanded clonotypes compared with matched stromal samples (P = .005; Figure 4D). Next, we evaluated the overlap between cTIL and sTIL TCR-repertoires. In six out of seven patients, the cumulative frequency of shared clonotypes (ie, their relative contribution in each repertoires) was higher in cTIL as compared with sTIL ( Figure 4E), consistent with the higher clonality of cTIL TCR repertoire. We next   by the concern of tumor heterogeneity, as well as undersampling. As the tumor regions were randomly selected, we can assume that even though the clonality of the tumor may vary from one region to the other, it is more clonal than the adjacent stromal part. Our spatial transcriptomics approach will also be useful to address other hypotheses regarding the tumor microenvironment, for instance whether close vicinity of (subpopulations of) dendritic cells with T cells is beneficial for patients, and whether T cells essential for antitumor response are located within cTIL, sTIL and/or elsewhere. This latter question is of major importance for a better selection of patients susceptible to respond to immunotherapy. Additionally, there is increasing focus on models suggesting that different subtypes of T cells need to associate with different antigen presenting cells and stromal components in order to optimally support anti-tumor immune responses. Our approach will be helpful in addressing these questions and optimize future therapies.

AUTHOR CONTRIBUTIONS
The work reported in the paper has been performed by the authors,