Patient material was obtained via the biobank of the Institute of Pathology and Neuropathology of the University Hospital Tübingen. Respective tissue sections were considered irrelevant for diagnostic purposes by the responsible pathologist. All patient material included in this study was retrieved from the central biobank of the Comprehensive Cancer Centre Tübingen. All patients gave their written informed consent for biobanking and the use of biomaterials and clinical data for scientific assessment, as approved by the ethics committee of the University Hospital Tübingen (ethics approval No. 508-2016BO1) in accordance with the Declaration of Helsinki. To test our panel on diverse tumor types showing distinct architectures and representing major cancer types, we decided to include CRC, PCa, and different types of liver cancer, namely CCC and HCC. Patient characteristics are listed in Supplementary Table S2 . Tissue identity was validated by a board-certified pathologist. Antibody validation was performed on healthy tonsil tissue, received from the Department of General, Visceral, and Transplant Surgery at the University Hospital Tübingen. Fresh or frozen tissue was embedded in Tissue-Tek^® O.C.T.™ Compound (Sakura Finetek USA) in cryomolds (25375-500, Polysciences, Inc.). Slow and controlled freezing was performed in an aluminum dish containing viscous ethanol, altogether cooled in liquid nitrogen. Embedded samples were stored at -80°C and equilibrated to the cutting temperature in the cryostat chamber for at least 20 mins before sectioning. Sections of 4-5 µm were cut with a cryostat (Leica CM 1950, Leica Biosystems) and placed onto SuperFrost^® plus slides (R. Langenbrinck GmbH). Slides were stored until the experimental procedure at -80°C.

Frozen slides were incubated with 4% paraformaldehyde solution (J19943.K2, Thermo Scientific) for 10 mins at room temperature, washed three times with MACSima™ Running Buffer (130-121-565, Miltenyi Biotec), and immediately mounted onto an appropriate MACSwell™ Imaging Frame (One: 130-124-673, Two: 130-124-675, or Four: 130-124-676, Miltenyi Biotec). Sections were covered with MACSima™ Running Buffer until initial 4′,6-diamidino-2-phenylindole (DAPI) staining. Primary DAPI staining was performed right before the experiment by removing the MACSima™ Running Buffer and adding a 1:10 DAPI (130-111-570, Miltenyi Biotec) dilution in Running Buffer. After 10 mins incubation at room temperature, the samples were washed three times with Running Buffer and covered with the final sample volume. Staining and washing volumes depend on the respective working volumes for the Imaging Frames (One:1000 µl, Two:500 µl, Four: 250). MACSwell™ Imaging Frames were sealed with MACSwell™ Sealing Foil (130-126-866, Miltenyi Biotec) to prevent evaporation and protect the samples from contaminants.

Primary fluorochrome-labeled antibodies conjugated to fluorescein isothiocyanate (FITC) or phycoerythrin (PE) were used for MICS. All antibody clones, dilutions, and fluorochrome recommendations are documented as our immune-oncology panel in Table 1 . Additionally, we recommend 34 alternative fluorochromes which we tested on diverse tissues and which could be used for precise individual panel design, adding up to a total of 155 validated antibody conjugates ( Table 1 ). Moreover, we provide antibodies that did not fulfill our validation criteria ( Supplementary Table S1 ). Antibodies were prepared in a MACSwell™ Deepwell Plate (130-126-865, Miltenyi Biotec) in the appropriate dilution with MACSima™ Running Buffer for the respective MACSwell™ imaging frame working volumes. Before pipetting the antibodies into the Deepwell Plate, we recommend centrifuging the tubes for 20 sec at 1000 x g to sediment possible precipitates. DAPI was added to the antibody-containing Deepwell Plate in a 1:50 dilution in every eighth antibody cycle. The antibody-containing Deepwell Plates were sealed with MACSwell™ Sealing Foil (130-126-866, Miltenyi Biotec) to prevent evaporation.

The MACSima™ Imaging System provides a fully automated workflow for iterative cycles of immunofluorescence sample staining, multi-field imaging, and fluorochrome removal (photobleaching or enzymatic digest). The sample slides mounted with a MACSwell ™ Imaging Frame as well as the antibody-containing MACSwell™ Deepwell Plate were placed on the xy-stage which moves reagents and samples to the correct needle and microscope positions. Antibody incubation was performed for 10 mins (standard setting), followed by excitation of the fluorochromes by specific laser/filter combinations and detection with a monochromatic scientific CMOS camera. A detailed description of the MACSima™ hardware (liquid handling system, microscope, stage) was published previously (40). We used hardware version 1.5.0 as well as software version 0.13.2 for image acquisition.

Before the iterative cycles were initiated, ROIs were defined and the focus was adjusted. ROIs were defined based on the DAPI signal shown in the overview scan and are based on customized numbers of overlapping fields of view (FoV). Additionally, we performed an H&E staining of a sequential tissue section and used this H&E stain as a reference which ensured optimal region definition. The focus was primarily set using the hardware autofocus option and adjusted manually for each ROI based on the DAPI signal. The MACSima™ captures various raw images of each FoV and each cycle. These images are utilized for subsequent image processing performed in MACS iQ View. One imaging cycle includes the following: DAPI image (used for FoV stitching), as well as bleach and staining images for each channel captured at three different exposure times (used for background subtraction and optimal exposure selection). After the MICS experiment, slides were H&E stained according to pathological standards (41) and imaged with a slide scanner (Panoramic Midi II, 3D Histech).

Individual raw FoV images were processed with the automated pipeline on the MACS iQ View software (Version 1.2.2). Detailed information on image processing was described before (40). In brief, different exposure times are combined in a high dynamic range image, normalizing based on multiple exposures when overexposure is detected. Exposure selection was set for “automatic”, validated and, if necessary, adjusted to optimal exposure times. Flatfield and distortion corrections are applied for the calibration data of the individual instrument. Next, neighboring FoV are automatically stitched together based on the DAPI staining of overlapping margins. Finally, the pre-stain bleach images of each cycle, acquired before the subsequent antibody incubation and showing potential residual staining of the previous cycle, are subtracted from the subsequent staining image, resulting in the final processed and stitched images. The images were cropped to DAPI based on experimental values of 200 pixels for top/bottom and 100 pixels for right/left correction.

Image datasets for each ROI, including all staining and autofluorescence images in TIFF format, were integrated into the MACS iQ View software (Version 1.2.2). Advanced cell segmentation was performed using the MACS iQ View segmentation pipeline. In detail, segmentation is conducted based on nuclear and cytoplasmic markers. We used the “Advanced Morphology for Tissue” nuclear detection option based on the first DAPI staining. Min/Max Diameter, Detection Sensitivity, Separation Force, and Smoothing Filter Sigma were adjusted based on the DAPI staining intensity and the respective tissue, see Supplementary Table S3 . Cytoplasmic signal was allocated to single cells by using the “Constrained Donut” option based on automatically chosen constraint channels. Detection Sensitivity and Donut Width were specified for each ROI individually based on signal intensities ( Supplementary Table S3 ). We provide recommended parameters for efficient cell segmentation for densely packed and hard-to-segment tissues, like tonsils, and adapted settings for various cancerous tissues based on tumor morphology. Depending on staining efficiency, exposure time, and tissue composition the recommended parameters need to be adjusted. Therefore, we give guidelines for the optimal usage of segmentation parameters. Segmentation was cross-validated via the visual control in MACS iQ View by displaying nuclear and cytoplasmic markers overlayed by the segmentation mask.

Antibody cross-validation was performed for each antibody, dilution, and experiment. Having the ability to stain several markers for a specific cell type, we compared markers of a single experiment (in-run comparison) and between different experiments. Additionally, we included a reference tissue (tonsil) on each slide for the in-run staining control. As a third reference, we used public databases like the Human Protein Atlas (proteinatlas.org (42)) or classical IHC staining to compare the MACSima™ staining and validate the antibodies.

Marker expression was allocated to the segmented cells and was used for expert-based gating of cell populations and cell type annotation (exemplary gating strategy: Supplementary Figure S2A ). Additional markers that were not used for gating were displayed as Color Maps to ascertain the cell type annotation ( Supplementary Figure S2B ). Gates were directly side-by-side compared with the staining image in the MACS iQ View image control window. Expert-based (sub-)gating of cell populations was adjusted for ROI and cell type-specific cell expression patterns, depicted as Scatter Plots and Histograms in MACS iQ View. Using the “Merge” operator, gated cell populations from different sub-gates were integrated into a combined cell type gate.

Neighborhood analysis of annotated cell types (i.e. PD1^low/high T cells) was conducted with the MACS iQ View distance analysis tool, applied for an annotated cell type of interest. Based on the distance histogram, we defined a range of 5 µm (≙ 47,14 pixels) or 25 µm (≙ 235,85 pixels) around the annotated cell type of interest. All cells within the distance range were merged (“Merge” operator, option “AND”) with all pre-defined annotated cell types. Cells present in both an annotated cell type and the range around the cell type of interest resulted in a new intersection population.

An exemplary MACS iQ View workflow script containing advanced segmentation settings, expert-based gating of cell populations, cell type annotation, and distance analysis is deposited at Zenodo: 10.5281/zenodo.10057717.

After advanced cell segmentation using the MACS iQ View, data was exported as a CSV file and analyzed with R (version 4.1.0). To limit the effect of potentially spurious outliers, we applied winsorization to the raw expression data, capping values at the 1st and 99th percentiles. We computed percentiles separately for every marker. Next, to reveal patterns in the data that are not apparent in the original, antibody-dependent scale, we applied the inverse hyperbolic sine transformation (arcsinh transformation) to the winsorized data. Replicable code for the analyses is available on Zenodo: 10.5281/zenodo.10057717.

Graphs were produced using GraphPad Prism (Version 9.5.1) and images were exported from MACS iQ View (Version 1.2.2). We created the faceted violin plots using the {ggplot2}-library (43).

Here, we describe the establishment, validation, and application of a comprehensive immunophenotyping panel for MICS, tailored to address immuno-oncology research questions. MICS multiplexed tissue imaging is based on iterative cycles of 1) fluorochrome-labelled antibody staining, 2) epifluorescence imaging, and 3) fluorochrome removal ( Figure 1A , top). After image processing, staining accuracy was assessed through cross-validation by co-staining of reference tissues as well as comparing staining patterns with publicly accessible databases and traditional IHC staining ( Figure 1A , middle). Following antibody cross-validation, we performed advanced segmentation analysis within the MACS iQ View software and used the ascertained marker expression profiles for expert cell annotation, which serves as a basis for 1) defining cellular composition, heterogeneity, and spatial distribution, 2) performing deep spatial cell phenotyping, 3) analyzing cellular neighborhoods, and 4) utilizing data for external bioinformatic analysis pipelines ( Figure 1A , bottom). In detail, image processing, image quality control, antibody cross-validation, advanced segmentation, marker-based cell annotation, and data analysis of all markers were performed within the MACS iQ View software. For optimal advanced cell segmentation and applicability for various tissues, we provide recommended parameters and guidelines ( Supplementary Table S3 ). Utilizing the advantages of the MACS iQ View software, the expert-based gating analysis (exemplary gating strategy in Supplementary Figure S2A ) was side-by-side controlled with the staining image and scatter plot gating was complemented by depicting additional marker expressions as Color Maps ( Supplementary Figure S2B ). For external downstream analyses, segmentation-based expression data as well as annotated cell subtypes were further analyzed in R, thereby combining the strengths of MACS iQ View and R in our analysis workflow. In the current study, we present our immunophenotyping panel comprising 121 antibodies addressing 118 antigen markers ( Figure 1B ). Detailed information on clones, vendors, tested fluorochromes, and suggested dilution is provided in Table 1 . In addition to the core panel used in this study, we tested alternative fluorochrome conjugates across various tissues, resulting in a total of 155 validated antibody conjugates for multiplexed imaging ( Table 1 ). We provide optional fluorochrome recommendations to facilitate further customization and optimization of our immunophenotyping panel. Essentially, we emphasized having more than two markers to distinguish distinct cell types, functioning as an internal staining control and strengthening the analytical power. The immunophenotyping panel is suitable for cell phenotyping of at least 16 different cell populations (lymphocytes: B cells, plasma cells (PCs), natural killer (NK) cells, T cells; myelocytes: Monocytes, myeloid dendritic cells (mDCs), plasmacytoid dendritic cells (pDCs), granulocytes, mast cells, myeloid-derived suppressor cells (MDSCs), macrophages (MΦ); endothelial cells, epithelial cells, fibroblasts, platelets, erythrocytes), as well as for definition of the cellular cytoskeleton and extracellular matrix (ECM) components. To further dissect the cellular landscape, we included 40 markers for ultra-deep sub-phenotyping, resulting in highly defined and diversified cell populations and cellular states including markers for characterizing cellular stress and senescence. Moreover, we integrated markers for possible therapeutic intervention, especially for immunotherapeutic targeting with monoclonal antibodies (mAbs), chimeric antigen receptor (CAR) T cells, or ICB. Representative staining of all markers on tonsil tissue is provided in Supplementary Figure S1 . Since panel design is laborious and costly, we additionally provide a list of tested antibodies that did not fulfill our validation criteria ( Supplementary Table S1 ). Taken together, we provide a validated immunophenotyping panel plus an analytic workflow for multiplexed tissue imaging using MICS.

To validate our immunophenotyping panel and test the feasibility of the approach, we chose palatine tonsil tissue as a reference. Palatine tonsils are secondary lymphoid organs playing an important role in developing self-tolerance and establishing adaptive immunity (44). A plethora of cells of the innate as well as adaptive immune system in different activation or differentiation states can be found within tonsils, complemented by stromal and vasculature components (45). Therefore, tonsil tissue has been broadly used for antibody panel validation (26, 46, 47). Applying our panel to human tonsil sections enabled a highly detailed mapping of tonsillar immune cells ( Figure 2A , middle and Figure 2B ) and stromal components ( Figure 2A , right). Despite performing 89-92 MICS cycles, including staining, imaging, and fluorochrome removal, which resulted in an experimental duration of more than seven days, the tissue integrity was preserved and used for final post-experimental H&E staining ( Figure 2A ). Advanced segmentation resulted in a total of 8074, 7379, and 6213 cells for tonsils 1, 2, and 3, respectively. We were able to classify 13 main cell types, namely T cells, B cells, PCs, NK cells, mDCs, pDCs, granulocytes, mast cells, MΦ, fibroblasts, blood, and lymphatic vessels, as well as squamous epithelial cells ( Figure 2C ). We further sub-characterized the T cells into CD4⁺ helper T cells (T_h) and CD8⁺ cytotoxic T cells (T_c) ( Figure 2C ). We could identify B cell subclusters in the germinal center (GC) organization, categorizing them into CD21⁺ germinal center B cells, CD22⁺ mantle zone B cells, and CD11b⁺ B cells ( Figure 2D ). Additional B cell activation states (CD25⁺, CD44⁺, CD69⁺) and memory phenotypes (CD40⁺) could be deciphered. In line with previous findings (48–52), we defined different architectural zones of the tonsil based on multiparametric staining (mantle zone: CD27⁺, CD39⁺; GC: CD9⁺, CD10⁺, CD171+; non-GC: BCL-2⁺, CD44⁺) ( Supplementary Figure S2C ). Expert marker-based gating using our analysis workflow allows definite cell annotation and quantification of immune cell subsets ( Figure 2E ), consistent with literature (53). Taken together, we validated an antibody panel of 121 markers using palatine tonsil tissue which serves as a platform to be adapted for a plethora of research questions.

Next, we validated our antibody panel on samples of three different cancer entities, particularly CRC, PCa, and CCC ( Figure 3 ). The entities were chosen to provide a spectrum of histologically different tumors to demonstrate the broad applicability of the antibody panel. The diverse morphology as well as tissue integrity was preserved over 92 MICS cycles (H&E staining in Supplementary Figure S3A ). Samples were cross-validated and analyzed using the presented analysis workflow ( Figure 1A ). Cytokeratin staining depicted the histological features of glandular CRC and PCa and ductal intrahepatic CCC, all originating from epithelial cells ( Figure 3 , second row). By using the markers for ECM and structural components, we differentiated the respective tumoral stroma differences of the three entities, consistent with literature (54–56). While we could define a reduced collagenous stroma for juxtaposed glands in CRC (cribriform, back-to-back aspect), ( Figure 3A , second row), the PCa and intrahepatic CCC samples were dominated by a dense stroma consisting of collagen III and collagen IV and fibroblastic markers ( Figures 3B, C , second row). Quantification of 36 different cancer-associated markers showed the heterogeneity of the different tumor types ( Supplementary Figure S3B ). We spatially mapped the expression of possible immunotherapeutic targets like EpCAM, uniformly expressed in all samples, and HER2, demonstrating a higher degree of intratumoral heterogeneity, particularly in the CRC sample. On the contrary, CD155 and CD243, e.g., were only detected in intrahepatic CCC, whereas PD-L1 was highly expressed in the CCC and CRC samples ( Supplementary Figures S3C–E ). Our data clearly demonstrates the feasibility of the applied technology to individually characterize tumor samples in terms of quantitative expression and spatial heterogeneity of druggable immune targets. This detailed tumor phenotyping can serve as a prerequisite for complex precision immunotherapy approaches.

In line with our findings in palatine tonsil tissue, we were able to identify different immune cell populations as well as tertiary lymphoid structures (TLS) within the tumor tissues ( Figure 3 , third row). Segmentation resulted in 7963 cells for the CRC, 5029 cells for the PCa, and 7101 cells for the CCC sample. Cells were annotated marker-based and we quantified the cellular components, including ten different immune cell types (T cells, B cells, PCs, NK cells, mDCs, pDCs, granulocytes, mast cells, MΦ, MDSCs), fibroblasts, vessels, lymphatic vessels, epithelial, and tumor cells. While percentages of total infiltrating immune cells (CRC: 20,24%, PCa: 26,45%, CCC: 23,91%) were similar, cell types and particularly spatial distribution differed dramatically, see Figures 3A–C . In both the CRC and the CCC samples, TLSs were identifiable ( Figures 3A, C ). In the CRC sample, the TLS presents as a peritumoral primary follicle-like structure with clearly distinguishable B and T cell zones and spotted mDCs. In contrast, the TLS in the CCC sample is located within a dense collagenous stroma and exhibits a more mature secondary follicle-like structure. GC reaction was defined by CD21⁺ GC B cells and surrounding CD34⁺/CD54⁺ high endothelial venules (HEV) ( Supplementary Figures S3F, G , HEV highlighted by white arrows). In contrast, B cells were found to be absent in the PCa sample. Taken together, this set of data clearly validates the feasibility of our immunophenotyping panel to comprehensively dissect tissues from various cancer entities at a single-cell level to study functional cellular composition and spatial architecture. Additionally, it underscores the dramatic phenotypical differences between cancer entities and the eminent need for comprehensive spatial analyses to study and understand cancer biology and immunology.

To further reveal the potential for spatial ultra-deep functional phenotyping with our established and validated panel, we demonstrate a comprehensive step-by-step analysis of an HCC sample. We chose three different tumor regions of interest (ROIs) from one patient sample and subsequently analyzed and compared the different tumor regions ( Figure 4 ), performed ultra-deep phenotyping of the T cell fraction ( Figures 5 , 6 ), and elucidated the T cell neighborhood ( Figure 7 ). Selected tumor regions included peritumoral, immune-rich stroma ( Figure 4C ), tumor margin ( Figure 4D ), and tumor core ( Figure 4E ). We further introduced subarea classifications within the different ROIs to describe spatial relationships: Intratumoral stroma (ITS), defined as areas without malignant cells, and intratumoral malignant cell clusters (ITM), defined as areas with densely packed malignant cells. Again, tissue integrity of the specimen after 92 MICS cycles was verified by H&E staining ( Figure 4A ). ROIs were selected based on initial DAPI staining ( Figure 4B ). Advanced cell segmentation resulted in a total of 13006 cells for the stroma, 7552 cells for the tumor margin, and 4917 cells for the tumor core ROI. Tumor stroma was composed of numerous blood and lymphatic vessels, a biliary duct (white arrow, top right corner), and a high percentage of immune cells, particularly T cells, PCs, mDCs, and MΦ, organized in clusters without distinguishable B and T cell zone as seen in canonical TLSs ( Figure 4C ). Approaching the tumor site, we identified a collagen- and fibronectin-rich stroma layer with a spotted T cell infiltrate ( Figure 4D ). Interestingly, we found a densely packed immune infiltrate of predominantly myeloid cells at the tumor border region ( Figure 4D ). These myeloid cell populations, dominated by mDCs and MΦ, form a barrier-like structure surrounding the malignant cells ( Figure 4D ). We also identified peritumoral T cell clusters further characterized below. mDCs, MΦ, and T cells are clustered not only at the tumor margin but also in the collagen-rich ECM ITS areas within the tumor core ( Figure 4E ). In contrast, most immune cells are strongly excluded from ITM areas, with only a few T cells and MDSCs being able to infiltrate ( Figures 4D, E ). Expression profiles of 36 tumor-specific markers were compared for tumor margin and border. Violin plots as well as staining showed uniform expression of HCC-related markers like AFP or Glypican 3, as well as pro-oncogenic CD155 and CD36, which are associated with disease progression and poor prognosis in HCC (57, 58) ( Supplementary Figure S3B and Supplementary Figures S4A, B ). Quantification of the different cell populations within the three ROIs is provided in Figure 4F . In sum, this set of data demonstrates spatially restricted intratumoral heterogeneity in cellular and ECM composition. We show compartmentation and spatial exclusion of certain immune cell populations, elucidating their functional properties as shown below.

To further demonstrate the potential of the 121 antibody immunophenotyping panel, we next analyzed the T cell compartment within the different HCC tumor ROIs. T cells can be roughly categorized into CD4⁺ T_h and CD8⁺ T_c. Among CD4⁺ T_h, FoxP3⁺/CD25⁺ regulatory T cells (T_reg) play a crucial role in orchestrating the immune system, mediating immunosuppressive function within the TIME. We first identified CD45⁺/CD3⁺ T cells in two different HCC tumor regions, namely margin and core. Next, cells were categorized as T_c, T_h, or T_reg based on the expression of CD8, CD4 or CD4, FoxP3, and CD25 ( Figure 5A ). Additionally, T cells can be phenotypically and functionally described based on their differentiation state. Along this line, we identified six T cell differentiation states based on deep sub-phenotyping using the expression levels of CR45RA, CD45RO, CD27, CD95, and CD11a ( Figure 5B ): naïve T_N cells, stem cell-like memory T_SCM cells, central memory T_CM cells, effector memory T_EM, effector memory cells re-expressing CD45RA T_EMRA, and terminally differentiated effector T_TE cells (see gating strategy in Supplementary Figures S5A, B ). We saw a shift towards more CD4⁺ T_h and less CD8⁺ T_c in the tumor core compared to the tumor margin. T_EM cells, both T_h cells and T_c cells, are the dominant T cell differentiation state in the tumor margin and peritumoral stroma. In contrast, T_CM cells and particularly CD8⁺ T_CM cells are predominantly found in the core ( Figure 5A ). Looking at the spatial distribution, T_C cells, mainly T_CM, were found to be the only T cell subset capable of infiltrating into ITM areas ( Figures 5C, D , left). The overall fewer CD4⁺ T_h cells (19,1% CD4⁺, 80,9% CD8⁺ of all CD3⁺ T cells) were generally located more distant to tumor cell clusters compared to the CD8⁺ T_c cells. Utilizing MICS data for external bioinformatic analyses, we were able to cross-validate T cell phenotyping based on marker expression in an unsupervised way. To demonstrate coherence between methods, we provide expression profiles for decisive markers on pre-gated T cell subtypes in Figure 5E , confirming subset affiliation.

Beyond a phenotypical description of different T cell differentiation states, functional characterization plays an important role in understanding and predicting responses to therapy. ICB has revolutionized the therapy of certain cancer entities, however reliable biomarkers for response prediction remain sparse (59–61). Particularly the functional state, spatial distribution, and cellular neighborhood of T cells are crucial for understanding mechanisms of ICB and other immune modulatory therapies. Dysfunctional, exhausted T cells are characterized by the expression of inhibitory immune checkpoint receptors, including PD1, LAG3, and TIM3. In contrast, activated T cells can be identified by the expression of CD69, CD25, and HLA-DR. We used the strength of MICS and analyzed 16 markers associated with T cell function in tumor margin and core. We identified different activation states of T cells – early (CD69), late (CD25), and very late (HLA-DR) ( Figure 6A ). The late and very late activated T cells were clustered in regions proximal to the ITM and within ITS areas ( Figure 6A , green arrows), while T cells within the tumor core, particularly in ITM areas, showed mostly no activation. Along with activation states, we saw a gradient in the expression of exhaustion markers. In Figure 6B , PD1 expression levels are depicted based on quartiles of normalized PD1 expression (0-25% = low, 25-75% = intermediate, 75-100% = high PD1 expression). Highest PD1 expression levels were present on cells infiltrated into ITM areas and on the previously described ITM-proximal immune cell clusters ( Figure 6B , blue arrows). Using MICS, and compared to other technologies, we were able to stain additional immune checkpoints and deep sub-phenotype fractions of T cells which were i) PD1 single positive ii) double positive for PD1 and LAG3 or TIM3, or iii) triple positive T cells expressing PD1, LAG3, and TIM3 ( Figure 6C ). Highest T cell exhaustion states were found in ITM-proximal immune cell clusters ( Figure 6C , pink arrows), while ITM-infiltrating T cells predominantly expressed only PD1. Combining the T cell subset definition ( Figure 5 ) with the T cell activation and exhaustion states, we were able to thoroughly profile a multitude of T cell subsets for their functional marker and potential therapeutic target expression ( Figure 6D ). In summary, not being limited by markers and having validated antibodies for functional immunophenotyping allows for a stringent and precise T cell classification, which is fundamental for patient-specific therapeutical decisions.

The major advantage of spatial biology applications towards sc/snRNASeq or other technologies is spatial resolution. Physiological or disease-relevant cellular function and interaction are often limited to restricted areas within a tissue. To further highlight the power of MICS in combination with our established and validated panel, we performed neighborhood analysis focusing on PD1^high and PD1^low T cells, two clinically relevant T cell subsets. In detail, we performed a distance analysis in MACS iQ View of our anchor cells, PD1^high and PD1^low T cells, and characterized proximal cell populations. To investigate the relevance of distance to anchor cells in the context of cell-cell interactions, we applied different radii for our analysis. We found that in a 5 µm range a very distinct neighborhood can be described, whereas at broader ranges the informative value faded ( Supplementary Figure S5C ). We therefore focused our downstream analyses on the 5 µm neighborhood ( Figures 7A–C ). Intriguingly, the immediate cellular neighbors around PD1^high T cells in ITS areas were predominantly myeloid cells and among them primarily mDCs, M2-like MΦ, and MDSCs ( Figure 7E ). We saw a tendency towards a higher percentage of mDCs and M1-like MΦ in the margin and MDSCs in the core area ( Figure 7E ). As described above, neighborhoods consisting of PD1^high T cells and myeloid cells were sited in spatially restricted clusters in close proximity, but strictly excluded from ITM areas ( Figures 7A–C ). In contrast, PD1^high T cells infiltrated into ITM areas were almost exclusively surrounded by malignant or other T cells ( Figures 7A–C ). Looking at PD1^low T cells, we did not only find a different spatial distribution within the ROIs but also different cellular neighborhoods ( Figure 7F ). PD1^low T cells are more often located in perivasculature and fibroblast-rich niches. Furthermore, we found a spatial association of PD1^low T cells with plasma cells ( Figure 7E ). To further understand the functional interaction of neighboring cell populations and to cross-validate findings, we analyzed cell populations in the proximity of PD1^high and PD1^low T cells for the expression of 12 immune checkpoint or immunomodulatory molecules ( Figure 7G and Supplementary Figures S5D, E ). Providing intrinsic validation of data, we found relatively higher expression of PD-L1 on cells in PD1^high vs. PD1^low T cells neighborhoods, particularly expressed by M2-like MΦ, pDCs, plasma cells, endothelial cells, lymphatics vessels, and tumor cells. A similar tendency was found for CD276 on M2-like MΦ, fibroblasts, endothelial, and malignant cells, CD73 on fibroblasts and endothelial cells, CD155 on M1-like and M2-like MΦ, fibroblasts, endothelial and tumor cells, IDO on M2-like MΦ, pDCs, plasma cells, endothelial cells, and lymphatics as well as VISTA on plasma and endothelial cells ( Figure 7G ). Particularly on vasculature, we found a clear spatial correlation between PD1^high T cells and protein expression of IFN-γ regulated genes like PD-L1, CD155, CD276, or IDO, as described individually before (62–65). In contrast, there is a tendency towards higher expression of CD112 on M1-like MΦ and CD155 on pDCs in PD1^low T cell neighborhoods ( Figure 7G ). Cells in the neighborhood of PD1^low and PD1^high cells in the tumor margin ( Figures 7A, B ) showed comparable patterns except for higher expression levels of PD-L1 of pDCs, M1-like MΦ, MDSCs, fibroblasts, and tumor cells in PD1^low neighborhoods ( Supplementary Figure S5E ). Together, this set of data demonstrates the capacity of MICS-based spatial biology to functionally dissect tumor tissues at single-cell spatial distribution. We identified multilayered, spatially restricted, and functionally distinct cellular neighborhoods. Comparable to T cells within ITM areas, we found less-activated and less-exhausted cells T cells in perivascular and stromal niches facing inhibitory signals via CD39, CD112, and IDO. We identified areas of dense immune infiltration proximal but strictly excluded from ITM areas. These “battle-grounds” show the highest T cell activation but also exhaustion and are dominated by myeloid cells with a tendency towards, but not restricted to suppressive phenotypes, expressing PD-L1, CD39, IDO, and VISTA. Further, we found a limited number of CD8⁺, mainly PD1^high T cells, infiltrating into ITM areas where they are surrounded by PD-L1, CD73, CD155, and CD276 expressing tumor cells. These findings clearly demonstrate the spatially restricted functional heterogeneity of distinct cell populations within the very same tumor sample and highlight the eminent need for comprehensive spatially resolved context analyses to derive clinically relevant conclusions.