Cell type and region-specific transcriptional changes in the endometrium of women with RIF identify potential treatment targets.

Recruitment of a well-characterized patient cohort with regard to demographic features and menstrual cycle phase [(LH) + 7 ± 2 d] was used to reduce commonly present biases in previous studies ( Table 1 ). An overview of the study design is shown in Fig. 1 . Two ROIs were selected in each endometrial tissue section: i) LE coupled with underlying stroma (SLS) and ii) GE with surrounding FS. The transcriptional profile of individual compartments and cellular components of the endometrium were compared between FC and RIF groups. Specifically, epithelial cells [cytokeratin positive (PanCK + )] in the LE and functionalis GE, stromal cells (CD56 − /CD45 − /PanCK − ) in the SLS and FS, and CD45 + leukocytes in the SLS, and CD56 + leukocytes in the FS. Participant demographics Age Mean (SD) 35 (5.0) BMI Mean (SD) 26.7 (4.6) Smoker n (%) 0 (0) 0 (0) Alcohol n (%) 2 (25) 5 (62.5) Live birth n (%) 8 (100) 1 (12.5) Miscarriage med (range) 0 (0) 0 (0 to 1) Failed embryo transfers med (range) 0 (0) 3 (3 to 9) * t test. † Chi-square test. Study overview. Endometrial biopsies were collected from RIF patients and FC. Antibodies were used to distinguish specific cell types of interest: epithelia (pan-cytokeratin), leukocytes (CD45 and CD56), and stroma (nuclear staining only). ROIs were selected comprising the LE, GE, SLS, FS. NanoString GeoMx spatial transcriptomics was employed for region- and cell-type-specific gene expression profiling. Transcriptomic data underwent cellular deconvolution, differential expression analysis, GSEA, and in silico drug screening. Following quality control of the in situ transcriptome profiles, principal component analysis (PCA) confirmed the well-known transcriptional differences between cell types, e.g., stroma and epithelial ( Fig. 2 A and B ). Cellular deconvolution of these gene expression profiles with existing endometrial single-cell transcriptomics data from the Reproductive Cell Atlas ( https://www.reproductivecellatlas.org/endometrium_reference.html ) effectively distinguished the known identity of these cell types ( Fig. 2 C ), and thus, externally validated the transcriptomic profiles generated. Cell abundances were similar between RIF and FC groups. Spatial transcriptomics profiling of endometrial biopsies. PCA of upper-quartile normalized RNA expression data showing ( A ) PC1 vs. PC2 and ( B ) PC1 vs. PC3, colored by cell type and shape corresponding to disease state. Tissue-specific differences can be observed with PANCK differences mainly accounted for in PC1 and other tissues separating alongside PC2. No clear structure observed between patient types via PCA. ( C ) Spatial deconvolution cell-type prediction using reproductive atlas single-cell sequencing data. Note that the different segment tags present the highest beta value for their counterpart cellular type (highest beta values-red), e.g., PanCK-epithelial, CD45/CD56 immune and stromal. ( D ) Heatmap of top four DEGs between cell types. Annotations with disease, cell type (segment tag), and tissue are shown. Very clear differential expression patterns were observed between segment tags. We compared gene expression across the spatially resolved regions to further confirm the specificity of our ROI selection, which allowed identification of DEGs including both known and novel region-specific markers ( Dataset S1 ). Gene expression of the guiding AOI protein markers PTPRC (CD45), NCAM1 (CD56), and KRT8 (PanCK) was enriched in the expected cell types ( SI Appendix , Fig. S1 ). As anticipated, CD45 + segments had significantly higher expression of the MHC class II cell surface receptors HLA-DRA [fold change (FC) 3.51, false discovery rate (FDR) 3.3E-05], HLA-DPB1 (FC 3.89, FDR 1.29E-05) and significantly lower expression of the natural killer (NK) cell markers NKG7 (FC 1.86, FDR 0.00942) and IL2RB (FC 1 .96, FDR 0.00475) compared to the CD56 + segments. Similarly, LGR5 (FC 7.69, FDR 1.93E-15), AGR3 (FC 4.3, FDR 2.07E-07), and MUC1 (FC 2.05, FDR 0.0036) expression was significantly up-regulated in the LE compared to the GE, whereas WNT5A (FC 2.25, FDR 0.00574) [differentially expressed in endometrial cancer ( 42 )] and PROK1 (FC 3.45, FDR 0.00169) [implemented in decidualization and implantation ( 43 )] distinguished between the SLS and FS ( Fig. 2 D and Dataset S1 ). Supporting these findings, GO enrichment identified “positive regulation of T cell activation” (FDR 9.32E-09) and “MHC protein complex assembly” (FDR 6.66E-05) between SLS CD45 + and FS CD56 + segments ( Dataset S2 ). The GO Terms enriched between SLS and FS included “extracellular matrix organization” (FDR 1.11E-10), “regulation of the canonical Wnt signaling pathway” (FDR 0.00345) and “response to transforming growth factor beta” (FDR 0.00882). Changes in gene expression between FC and RIF groups were identified by comparing their transcriptional profiles across each endometrial region and cell type ( Fig. 3 and Dataset S3 ). Differential expression profiles in each region between RIF and control patients. Volcano plots displaying differential gene expression between cell types from patients with RIF when compared to healthy controls. Top DEGs are labeled and significantly DEGs are colored red. ( A ) LE. ( B ) GE. ( C ) SLS. ( D ) FS. ( E ) Subluminal stromal CD45 + . ( F ) Functionalis stromal CD56 + . 685 DEGs were identified between the FC and RIF cohorts in the LE ( Fig. 3 A ). Notably, ULBP1 was up-regulated in those with RIF, a ligand for the NKG2D NK cell activating receptor ( 44 ). XRCC2 , a gene known to regulate RNA repair and chromatin stability, was also up-regulated ( 45 ). Transcription factors JUN , JUND, and FOS were down-regulated in those with RIF. FOS and JUN regulate expression of many target genes that function in concert to properly control uterine epithelial cell proliferation, stromal differentiation, angiogenesis, and local immune response to render the endometrium receptive and allow embryo implantation ( 46 ). Progesterone up-regulates poFUT1 expression via c-FOS/c-JUN, which increases trophoblast cell proliferation and adhesion potential ( 47 ). JUND and FOS are up-regulated in endometriosis but down-regulated in the LE of women with RIF ( 48 ). 293 DEGs were detected between the FC and RIF cohorts in the GE ( Fig. 3 B ). Down-regulated DEGs notably included ADAM 15 , a gene known to play a role in wound healing ( 49 ). CCDC13 , which is required for primary cilia formation and promotes the localization of the ciliopathy protein BBS4 to both centriolar satellites and cilia, was also down-regulated ( 50 ). PTBP1 , identified as a down-regulated DEG, had been previously associated with RIF in combination with HOXA11-AS and impaired endometrial decidualization ( 51 ). 419 SLS DEGs were identified between the FC and RIF cohorts ( Fig. 3 C ). ULBP1 was found to be up-regulated in SLS and GE, suggesting a universal importance for this gene in RIF. TM4SF4 was down-regulated in SLS of women with RIF. TM4SF4 has previously been reported to be decreased in the WOI in those with polycystic ovarian syndrome (PCOS) and is associated with cell adhesion ( 52 ). PGRMC2 (progesterone receptor membrane component 2), a gene required for the maintenance of uterine histoarchitecture and normal female reproductive lifespan ( 53 ), was also down-regulated in those with RIF. Endometrial PGRMC2 expression fluctuates during the human menstrual cycle and is abundantly expressed in stromal cells during decidualization ( 54 ). PGRMC2 knockdown significantly compromised the ability of the decidualized human endometrial stromal cells to support trophoblast expansion in an outgrowth model ( 54 ). WNT4 , a critical regulator of embryo implantation and decidualization, was down-regulated in those with RIF ( 55 ). A WNT4 agonist has been shown to improve impaired decidualization in RIF ( 56 ). 264 FS DEGs were identified between the FC and RIF cohorts ( Fig. 3 D ), including down-regulated genes previously recognized in other regions/cell types in this study, e.g., PTBP1 (GE) , ADAM15 (GE) , PGRMC2 (SLS), and up-regulated gene ULBP1 (SLS, LE). IL24 expression in the FS was down-regulated in the RIF population and is postulated to be required for decidual stromal cells to promote differentiation of CD56 bright CD16 − NK cells, which elicit high levels of inhibitory receptors, immunotolerance, and angiogenic cytokines by secreting IL-24 during decidualization ( 57 ). IL-24 may influence the invasion of trophoblasts and is involved in normal pregnancy ( 58 ). We analyzed the SLS CD45 + population as well as the FS CD56 + population. The SLS CD56 + cell numbers were insufficient for analysis. We propose that future studies, focusing on the different immune cell populations, can explore the immune-cell-type-specific transcriptional differences in RIF. In the SLS CD45 + leukocytes, 1,125 DEGs were detected between the FC and RIF cohorts ( Fig. 3 E ). HSD3B2 was seen to be up-regulated, which has previously been reported as increased in infertile women with endometriosis versus fertile women ( 59 ). HSD3B2 upregulation may be related to the abnormal biological effect of estrogen in the endometrium affecting the development of human embryos ( 59 ). Interestingly, ULBP1 was also up-regulated in these cells, mirroring the GE and FS in the RIF endometrium. APH1A , a gene known to increase Notch and WNT signaling cascades ( 60 ), and PGRMC2 were down-regulated in SLS CD45 + cells in those with RIF. PTBP1 and ADAM15 were again identified to be down-regulated in RIF, among the 1,049 DEGs between the FC and RIF cohorts in the FS CD56 + population ( Fig. 3 E ). Similarly, FS CD56 + ULBP1 expression was found to be up-regulated in those with RIF. CD27 , an important marker distinguishing between NK cell subsets, was down-regulated in the RIF FS CD56 + cells. CD27 + uNK cells maintain the immune-tolerant microenvironment by inhibiting the proliferation of inflammatory Th17 cells ( 61 ). To explore connections between RIF-associated DEGs and reproduction, we leveraged existing knowledge from mouse genetic models. We consulted our DEGs within the Mammalian Phenotype Browser and found 52 genes associated with abnormal embryo implantation (MP:0001727) or reduced female fertility (MP:0001923) phenotypes dysregulated in at least one cell type in our study ( Dataset S3 ). DEGs associated with aberrant embryo implantation phenotypes were observed mostly in immune cells (SLS CD45 + , FS CD56 + ). These included TROPBP and ANXA7 , genes involved in NK cell activation and prostaglandin E2 regulation, respectively ( 62 ). The transcription factor WT1 was observed to be down-regulated in SLS in RIF patients, which has previously been reported to regulate endometrial receptivity in PCOS patients ( 63 ). These results emphasize that genes previously linked to female fertility are also dysregulated in a region-specific manner in RIF ( 63 ). We further contextualized the observed differential gene expression in RIF with GSEA using both GO terms and KEGG pathways, which identified multiple dysregulated pathways ( Datasets S4 and S5 and Fig. 4 A and B ). The top enriched GO Term across several cell types was “sensory perception of smell” (FS FDR 1.73E-34), which included many up-regulated G protein–coupled receptors. These findings are consistent with previous reports of olfactory receptors expressed in the endometrium regulating receptivity ( 64 ). Dysregulated pathways in specific regions included the “WNT signaling pathway” (FDR 3.27E-05, SLS), which was altered in the FS and SLS. The “response to estradiol” (FDR 0.019) and “ovulation cycle” (FDR 0.000456) pathways were dysregulated in the SLS ( Fig. 4 A ). To investigate potentially altered cellular communication between cell types in RIF, we filtered known ligand–receptor interactions with differential expression. The luminal area circos plot (including the LE, SLS, and CD45 + leukocytes) highlights the immune cells as main receivers ( SI Appendix , Fig. S2 ), whereas the glandular area circos plot (including the GE, the FS, and the CD56 + leukocytes) shows the immune, epithelial, and stromal cells to be comparative as receivers ( SI Appendix , Fig. S2 ). This analysis highlighted interactions such as ULBP1 (a ligand for the NKG2D NK cell activating receptor ( 44 )) universally sending from each of the six regions, in the luminal circos [component of the dystrophin–glycoprotein complex known to be important for cell adhesion ( 65 )] and HCST_KLRK1 in the glandular circos [involved in the regulation of decidualized NK cells ( 66 )] receiving from six senders. Dysregulated pathways in regions from RIF patients. Dot plots of GSEA results between RIF patients and FC with ( A ) GO terms and ( B ) KEGG pathways are shown. Positive score (red) and negative score (blue) indicates positively or negatively enriched pathways, respectively. We identified drug candidates for future functional validation by screening the LINCS L1000 library of drug induced transcriptional profiles for compounds capable of mimicking or reversing the observed transcriptional changes in RIF ( Table 2 and Dataset S6 ). Highly ranked compounds with reverse gene expression signatures to RIF vs. FC across multiple endometrial regions included the serotonin receptor antagonist GR-55562, the selective estrogen receptor (ER) modulator raloxifene (which is a partial agonist for ERα and pure antagonist for ERβ) and the β1-adrenergic receptor antagonist, bisoprolol. GSEA of the molecular targets of the prioritized compounds showed significant enrichment of the drug targets in Wnt signaling, FoxO1 signaling, basal transcription factors, and glutamatergic synapse signaling across several endometrial regions, thus highlighting potential therapeutic pathways for future experimental investigation ( Dataset S7 ). Drug candidates capable of mimicking or reversing the observed transcriptional changes in RIF An IHC experiment was conducted to assess whether transcriptional differences can be readily detected with protein level markers. Four protein targets (ADAM15, PTBP1, PGRMC2, and ULBP1) were selected based on having high mRNA level differential expression and pragmatically by antibody availability. Immunostaining was present for all four tested antibodies in the endometrial tissue, confirming translation ( SI Appendix , Fig. S3 ).

RIF is a devastating multietiological condition without robust treatment options. It is apparent that discrepancies exist when differing methods and homogenized samples are utilized to identify DEGs between those with and without RIF ( 67 ). In this study, we employed spatial transcriptomics to elucidate the endometrial transcriptional profile of RIF at a cell-type- and region-specific resolution. We identified transcriptional differences in all endometrial subregions examined in women with idiopathic RIF relative to FC ( Fig. 5 ). Our findings endorse careful consideration of functionality of different cellular neighborhoods and cell types in the endometrium, which should be examined separately. Ignoring individual endometrial regions and their composite cell populations will overlook important aberrations in pathologies and thus forego the identification of potential treatment targets. Overview of region- and cell-type-specific gene expression changes in the RIF endometrium. Dysregulated signaling pathways and physiological processes are highlighted in red. Arrows relate to up or downregulation of gene expression in the RIF endometrium versus the fertile controls. Abbreviations: functionalis glands (FG), glandular epithelium (GE), subluminal stroma (SLS), and luminal epithelium (LE). Our findings show agreement with some published work identifying DEGs in the WOI in those with and without RIF, while also defining previously unrecognized differences. Sets of hub genes that have been previously proposed as potential biomarkers for the prediction of RIF were not seen to be differentially regulated in the specific cellular neighborhoods examined ( 26 , 27 , 68 – 72 ). In contrast, some DEGs were highlighted by others previously. Koot et al. ( 25 ) produced a list of 303 genes that they suggested could accurately predict RIF in their cohort with 100% positive predictive value ( 25 ), and of these, 39 were found to be differentially expressed in our study, but in various specific endometrial regions. For example, 17 were in SLS CD45 + leukocytes, 16 in FS CD56 + leukocytes, two in the LE, two in the SLS, and two in the FS population. WWC1 was the only overlapping DEG between our work and the Albayrak et al. study. WWC1 showed increased expression by digital droplet PCR ( 73 ) and this was also seen to be up-regulated in the SLS of our RIF cohort. WWC1 upregulation is thought to impair decidualization ( 73 ). One out of the three identified hub genes identified by Lai et al., PSMD10 ( 74 ), was seen to be up-regulated in the SLS CD45 + leukocytes; others were not differentially expressed in any of the subcompartments. Previously, increased HOXA11-AS was seen to modulate PKM1/2 alternative splicing by competitive binding to PTBP1 , thus leading to attenuated decidualization in RIF patients ( 51 ). We found that PTBP1 was consistently down-regulated in RIF across all endometrial regions we interrogated. Out of the proposed 10 hub genes that were significantly differentially expressed in RIF patients in the Zeng et al. study ( 75 ), only HOXB4 was found to be up-regulated in the SLS CD45 + population in our RIF cohort. In a bulk RNA study, Wang et al. showed that UBE2I , a hub DEG, was up-regulated in the RIF endometrium ( 76 ). In contrast, it was down-regulated in the FS CD56 + leukocytes of the current RIF cohort. Wang et al. found NR3C1 to be a main target gene for drugs annotated in drugbank ( 76 ) and we found this gene to be down-regulated in our SLS CD45 + leukocytes. We found RPS6KA1 to be down-regulated in FS CD56 + leukocytes of the RIF group, while Liaqat Ali Khan et al. demonstrated its downregulation in RIF women ( 77 ). Interestingly, Liaqat Ali Khan et al. reported GPN3 to be down-regulated in the RIF endometrium, while we observed an upregulation in SLS CD45 + cells of the RIF cohort ( 77 ). Interestingly, Liaqat Ali Khan et al. presented several other DEG that conflict with our findings and other studies ( 77 ). An overall observation from the current study was that RIF endometrial subregions demonstrated more down-regulated genes in comparison with the FC, which is in agreement with Koot et al., who also reported 81% of DEGs to demonstrate a decreased expression ( 25 ). The discrepancy between identified DEGs across studies is inherent in the utilization of different methodologies and potentially different cohorts. While bulk RNA-seq provides a comprehensive view of gene expression changes across the entire homogenized tissue, spatial transcriptomics offers a targeted approach focusing on localized gene expression patterns within distinct regions of interest, for example, the LE, the first maternal cellular contact for the implanting blastocyst. Understanding of the specific and conducive functional changes during the implantation process that occur in the LE is essential to understand how pregnancy establishment occurs. We have identified a particular set of 57 DEGs that were universally dysregulated in all the different subregions and cell types considered in the RIF endometrium (36 up-regulated and 21 down-regulated), providing avenues for research, exploring their function and potential as biomarkers or therapeutic targets in RIF. Currently, there are no medications available for RIF. Many immunological therapies (e.g., intrauterine infusion of peripheral blood mononuclear cells/platelet rich plasma, administration of granulocyte colony-stimulation factor, intravenous immunoglobulins, intravenous intralipid, and corticosteroids) have been trialed and continue to be studied, but a consensus is yet to be reached on an effective treatment option ( 78 – 80 ). Our in silico drug screening has identified several potential therapeutic agents to treat RIF. Interestingly, none of the top ranked drugs are postulated to target all of the six different regions/cell types studied ( Table 2 ). Of these, Raloxifene (shown to affect four of the six regions/cell types studied) has a specific pure antagonistic effect on ERβ, while being a partial agonist of ERα and G-protein-coupled ER ( 81 ). Our previous work demonstrated that RIF endometrial glandular and perivascular cells have reduced ERβ ( 40 ), and ERβ has opposing action to ERα in the endometrium ( 82 ). Therefore, Raloxifene may rectify RIF associated abnormalities via potentiating ERα mediated estrogenic action, such as inducing PR expression, thus enhancing progestogenic action on the endometrium. Further elucidating the role of estrogen, working through all three types of ER in the endometrium of RIF women, will facilitate the development of targeted therapies. We propose future studies to elucidate the potential for reversion of the RIF endometrial transcriptional profile using the therapeutic agents identified herein, revert the aberrations we identified in the RIF endometrium by studying the endometrium before and after exposure to these proposed therapies. The existing large body of transcriptomic data have not sufficed to advance clinical management of RIF due to the aforementioned shortcomings and those highlighted previously ( 67 ). In this context, our findings provide precedent for future transcriptomic studies of the RIF endometrium to surmount current limitations by granting robust validation of the identified DEGs, thus, allowing comparison and generalizability ( Fig. 5 ). Although previous studies have demonstrated the utility of spatial transcriptomics in understanding endometrial physiology, this study highlights the strength of the spatially resolved endometrial-cell-type-specific transcriptomic profiles in addressing pathological conundrums such as RIF. The region- and cell-type-specific endometrial differences identified in women with RIF in our deeply phenotyped and LH timed endometrial biopsies advances the existing knowledge base, delineating causative biological processes relevant to RIF. These aberrations can consequently be targeted in future for developing novel personalized treatment strategies for this poorly understood, highly prevalent, and distressing condition.

Collection and use of all samples were approved by the Liverpool Adult Ethics committee (REC references; 05/Q1505/115 and 05/Q1505/014). Informed written consent was obtained from all participants. Human endometrial pipelle biopsies were collected in the WOI (urinary luteinizing hormone (LH) + 7 ± 2 d) from 16 women (8 RIF and 8 FC). The FC group were volunteer women with at least one previous live birth and no fertility treatment. All women included had regular menstrual cycles, no known endometrial pathology (with a normal gynecological ultrasound scan), or use of hormonal medications for three months prior to sampling at Liverpool Women’s Hospital NHS Foundation Trust. Women were included in the RIF cohort after ≥3 embryo transfers and no positive pregnancy test. Demographics were summarized using means and medians for continuous data and counts and percentages for categorical data. Statistical analysis was undertaken using the t test and Chi-square test. All tests were undertaken at the 5% significance level. Formalin-fixed paraffin-embedded (FFPE) 4 µm thick endometrial biopsies were sectioned, mounted on slides (Thermo Scientific Superfrost Plus), and sent to the Technology Assessment Program (NanoString Technologies, Seattle) for GeoMx Digital Spatial Profiling (DSP) analysis. Briefly, NanoString GeoMx DSP uses a pool of ∼20,000 transcript-specific RNA probes linked to a unique barcode via a UV cleavable linker ( 28 ). The probe pool was hybridized to target mRNAs on tissue slides overnight and slides were then stained using fluorophore-labeled antibodies [epithelial cells (cytokeratin positive (PanCK + )), stromal cells (CD56 − /CD45 − /PanCK − ) and immune cell populations (CD45 + and CD56 + )], before imaging on the GeoMx profiler platform. For each sample two regions of interest (ROIs) were selected, encompassing i) luminal epithelium (LE) and subluminal stroma (SLS), and ii) glandular epithelium (GE) and the surrounding functionalis stroma (FS). UV light was applied to release and collect the DSP barcodes from selected ROIs. This was followed by library preparation and RNA sequencing as described previously ( 29 ). Illumina NovaSeq 6000 sequencing was performed, S4 kit, pair end reads and 27 × 27 read length was used. FASTQ sequencing files were converted into digital count conversion files using NanoString’s GeoMx next generation sequencing pipeline. All downstream spatial transcriptomics analyses was performed in R ( 30 ). GeomxTools (v3.5.0) was used to perform quality control and filter the lower quality samples using the approach described in GeoMx Workflows ( 31 ). Further quality control included filtering out segments where <10% genes were detected and genes with <10% detection across areas of interest (AOI). Downstream analysis was performed on the resulting 88 AOIs and 12,195 genes. Normalization was performed using the upper-quartile method, and all normalized counts were transformed to log2 space for downstream analysis. DEGs between cell types and disease state were identified using a linear mixed effect model [package lme4 (v 1.1-35.2) ( 32 )] treating Patient ID as a random effect. Benjamini–Hochberg adjustment of the P -values was used to correct for multiple testing. DEGs were defined as those with an absolute fold change ≥1.5 and an adjusted P -value < 0.05. Mouse phenotype annotations were obtained from the mouse genome informatics (MGI) database and used to further contextualize DEG results. Database access in https://www.informatics.jax.org/downloads/reports/HMD_HumanPhenotype.rpt on 21/02/2024 and gene annotations for abnormal embryo implantation (MP:0001727) and reduced female fertility (MP:0001923) were retrieved. Altered pathways and gene ontology (GO) biological processes were identified with Gene Set Enrichment Analysis (GSEA) using either GO biological process annotations or KEGG pathways ( 33 , 34 ), implemented in clusterProfiler (v 4.10.1) ( 35 ) using a minimum and maximum gene set size of 10 and 500, respectively. Analysis was performed using all genes passing the gene detection threshold, ranked by fold-change. Benjamini–Hochberg adjustment of the P -values was used to correct for multiple testing. An in silico screen of drugs that could mimic or reverse the observed differential expression was performed using signatureSearch (v1.16.0) ( 36 ) with the LINCS L1000 library of drug transcriptomic responses (Experiment Hub accession–EH3226) ( 37 ) filtered to retain only the high quality TouchStone compounds. The differential expression responses between RIF and FC for each segment were used as input and the correlation-based similarity measure (“Cor”) was used to rank candidate drugs. Endometrial single-cell sequencing data from the Reproductive Cell Atlas ( 38 ) were downloaded from https://cellgeni.cog.sanger.ac.uk/vento/reproductivecellatlas/endometrium_all.h5ad and converted to a Seurat (v 5.0.3) ( 39 ) data object with SeuratDisk (v 0.0.0.9021). The provided broad cell-type annotations were used to create a custom profile matrix with SpatialDecon (v 1.12.3) ( 40 ) and subsequently used to estimate the proportion of cell types in each segment of the GeoMX spatial transcriptomics data with default settings. The OmniPath database of curated consensus ligand–receptor interactions was filtered to retain those interactions, where for each combination of cell types either the ligand or the receptor were differentially expressed ( 41 ). The absolute sum of the fold change for both the ligand and the receptor was used to rank the interactions and select those for visualization with circos plots. FFPE blocks were sectioned at 4 µm (n = 24) and mounted on Superfrost microscope slides (ThermoFisher Scientific). Immunohistochemistry (IHC) was performed using the Bond RXm Automated Stainer with the Bond polymer refine detection systems in green, red, brown, and blue ( SI Appendix , Table S1 ) according to the manufacturer’s instructions (Leica Biosystems). Following deparaffinization and heat induced epitope retrieval, multiple cycles of sequential staining using Bond Refine detection systems were conducted. Each antibody step was followed by heat-mediated antibody stripping using high pH or low pH BOND Epitope Retrieval Solution at 97 °C for 20 min and then repeated with the next round of staining with antibody, Leica detection system, and chromogen. Sections were counterstained with hematoxylin following chromogen application.

Appendix 01 (PDF) Dataset S01 (XLSX) Dataset S02 (XLSX) Dataset S03 (XLSX) Dataset S04 (XLSX) Dataset S05 (XLSX) Dataset S06 (XLSX) Dataset S07 (XLSX)