‡ Both authors share equal contribution. * Both authors share last authorship. Corresponding authors

The determinants of metastasis in mismatch repair deficiency with high levels of microsatellite instability (MSI-H) in colorectal cancer (CRC) are poorly understood. Here, we hypothesized that distinct immune and stromal microenvironments in primary tumors may discriminate between non-metastatic MSI-H CRC and metastatic MSI-H CRC.

We profiled 46,727 single cells using high-plex imaging mass cytometry and analyzed both differential cell type abundance, and spatial distribution of stromal and immune cells in primary CRC tumors with or without metastatic capacity. We validated our findings in a second independent cohort using immunohistochemistry.

cohort.

High-plex imaging mass cytometry and hierarchical clustering based on microenvironmental markers separated primary MSI-H CRC tumors with and without metastatic capacity. Primary tumors with metastatic capacity displayed a high stromal content and low influx of CD8 + T cells, which expressed significantly lower levels of markers reflecting proliferation (Ki67) and antigen-experience (CD45RO) compared to CD8 + T cells in non-metastatic tumors. CD8 + T cells showed intra-epithelial localization in non-metastatic tumors, but stromal localization in metastatic tumors, which was validated in a second We conclude that localization of phenotypically distinct CD8 +

metastasis formation in MSI-H CRC. 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69

A deficient DNA mismatch repair system (dMMR) is the underlying cause of a specific subtype of colorectal cancer (CRC). A direct consequence of dMMR is a high level of microsatellite instability (MSIH). Both can be measured in clinical practice as part of the pathological classification of CRC1. In addition, dMMR causes a high mutational load, the generation of tumor-specific immunogenic neo-antigens, and infiltration by T cells. High numbers of tumor-infiltrating T cells are associated with a good prognosis and decreased risk of metastasis2,3. Indeed, patients with primary MSI-H CRC have a reduced risk of developing distant metastases4. Nevertheless, in a small proportion of patients with MSI-H CRC, distant metastasis formation does occur. These patients account for 3-5% of all cases of metastatic CRC. Importantly, MSI-H status in metastatic CRC is associated with poor prognosis and reduced benefit from chemotherapy5-7.

Genetic, histopathological, immunological, and molecular characterization has demonstrated the heterogeneous nature of CRC8-13. The association of (some of) these variables with disease progression and therapy response in CRC has promoted precision treatment strategies14,15. For instance, MSI-H tumors do not benefit from adjuvant chemotherapy with 5-fluorouacil (5-FU) monotherapy following surgical removal of the primary tumor16. Consequently, dMMR tests are now routinely applied in the clinic as a decision tool to select stage II/III CRC patients for adjuvant treatment. Recently, we demonstrated that MMR status in patient-derived CRC epithelial organoid cultures does not correlate with intrinsic tumor cell sensitivity to 5-FU17. As organoid cultures lack cells from the tumor microenvironment (TME), we proposed that differences in the TME between MSI-H and microsatellite stable (MSS) tumors may determine benefit from chemotherapy17. Indeed, the composition of the TME is now recognized as an important determinant of treatment response in cancer in general18. 70 71 72 73 74 75 76 77 78 79 80 81 82 infiltrating lymphocytes, which are kept in check by expression of immune checkpoints11,19-22. Different immune cells (e.g., macrophages, NK-, T- and B cells) in the TME can have pro- or anti-tumorigenic effects. Particularly, reduced infiltration of CD8 + cytotoxic T cells into the primary tumor is associated with disease progression and metastasis in CRC (and other cancer types)2,3. This observation has formed the basis for developing the 8Immunoscore9 as a tool to predict CRC progression2,3. In addition, a high content of stromal fibroblasts characterizes the metastasis-prone consensus molecular subtype 4 (CMS4) in primary CRC, which predominantly consists of MSS tumors8,23,24. To date, prognostic tumor- or TMEcentric biomarkers that predict metastasis formation in the MSI-H CRC subtype are lacking and, given the therapy resistance in a substantial proportion of patients with advanced metastatic MSI-H CRC25, these are urgently needed to allow for early treatment.

Here, we investigated whether distinct stromal features such as the number, location, and activation status of immune cell subsets and fibroblasts could distinguish between primary MSI-H CRC non-metastatic tumors and tumors with metastatic capacity. To this end, we applied high-plex imaging mass cytometry (IMC), which enables single-cell spatial analysis of multiple distinct cell types in patientderived tissue samples. We show that metastatic capacity in MSI-H CRC is associated with limited (intraepithelial) infiltration by CD8 + T cells and localization of inactive CD8 + T cells within the primary tumor stroma. These findings suggest that a stroma-rich immunosuppressive TME may drive metastasis in MSIH CRC and may help to devise strategies aimed at early identification and treatment of metastasis-prone MSI-H CRC. 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 (MSI-H) and metastatic (MSI-H or MSS) primary colorectal tumors. We employed tissue microarrays of patients with non-metastatic MSI-H (n=3), metastatic MSI-H (n=3) and metastatic MSS (n=3) CRC and processed three distinct TMA cores per patient (Supplementary Table 1). All cores (n=27 Regions of Interest (ROI9s)) were used for multiplex imaging mass cytometry analysis (IMC), using a 35-plex antibody panel (Supplementary Table 326). The panel included markers identifying distinct cell types (i.e. immune cells and fibroblasts), transcription factors, activation/exhaustion markers, and signaling markers (Fig. 1a, Supplementary Fig. 1). We used machine learning tools Ilastik and CellProfiler for pixel classification, and tissue region and single cell segmentation26-29, and excluded the necrotic areas in ROI9s, which showed non-specific signals with poor signal-noise ratio (Supplementary Fig. 2a, Supplementary Table 4). Next, we generated single cell data of all segmented stromal (immune and fibroblast) cells (Supplementary

Analysis of segmented epithelial and stroma compartments showed that the stroma content (percentage of stroma in the tissue) was significantly higher in MSS/MSI-H metastatic tumors compared to MSI-H non-metastatic tumors (Fig. 1b). Principal component analysis showed separate clustering of metastatic and non-metastatic tumors based on differential marker expression (Fig. 1c). Hierarchical clustering revealed a separate cluster containing most non-metastatic tumors (Fig. 1d). Both analyses demonstrate that, in comparison to metastatic tumors (MSS and MSI-H), non-metastatic MSI-H tumors display high expression of markers for proliferation (Ki67) and cell death (cleaved caspase 3), and of markers identifying immune cells (CD45, Granzyme B), including T cells (CD3, CD8, CD45RO, ICOS, CTLA4) (Fig. 1c, d and Supplementary Fig. 3a). Moreover, MSI-H non-metastatic tumors show low expression of the stromal marker aSMA, and markers associated with inflammation (phosphorylated STAT3 (PSTAT3), NFkB) (Fig. 1d).

Metastatic MSI-H tumors contain low numbers of cytotoxic T cells To gain further insight into the differences in TME composition between metastatic and non-metastatic tumors, we performed unsupervised clustering by Phenograph30, and manually annotated clusters of cell types based on marker expression (Supplementary Fig. 3b). We identified 20 clusters and 12 distinct cell types (Fig. 2a). Uniform Manifold Approximation and Projection (UMAP) analysis on the expression of lineage markers per cell showed a clear separation of the identified cell types (Fig. 2b). Hierarchical clustering analyses of cell type abundance revealed strong predominant ordering according to disease state and showed two major clusters (Fig. 2c). The first cluster mainly consisted of MSI non-metastatic tumors and displayed elevated proportions of T cell populations (cytotoxic, helper and regulatory), while the second cluster mainly contained MSI-H/MSS metastatic tumors and displayed higher abundance of 8Fibroblast9, 8Fibroblast & Monocyte9, 8Neutrophil & NK-cell9 populations (Fig. 2c). Indeed, in line with the higher stroma content observed in metastatic tumors (Fig. 1b), MSS metastatic tumors showed significantly elevated numbers of 8Fibroblast9 and 8Fibroblast and Monocyte9 populations compared to MSI-H non-metastatic tumors (Fig. 2d). MSI-H non-metastatic tumors contained a significantly higher number of 8Cytotoxic T cell9, 8Regulatory T cell9, and 8Macrophage & Monocyte9 when compared to MSI-H metastatic tumors (Fig. 2d). 8B cells9, 8T helper cells9, and other cell types were present in equal numbers in MSI-H non-metastatic and MSI-H metastatic tumors (Fig. 2d). 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155

Metastatic MSI-H tumors lack proliferating antigen-experienced CD8+ T cells Next, we investigated the relative proportions of specific T cell sub-populations (CD4+, CD8 + and FOXP3+) in the tumors. This analysis showed that the CD8 + T cell population was more abundant in MSI-H nonmetastatic tumors compared to MSS metastatic tumors (Fig. 3a). Vice versa, the proportion of CD4+ T cells was higher in MSS metastatic tumors compared to MSI-H non-metastatic tumors. The proportion of regulatory T cells (Tregs; FOXP3+) was similar in all tumors (Fig. 2d, Fig. 3a).

Analysis of immune checkpoint expression showed that CTLA-4 was significantly higher in helper T cells in MSI-H non-metastatic tumors compared to MSI-H/MSS metastatic tumors (Fig. 3b). Other proteins expressed by T helper cells (CD45RO, Granzyme B, IFNy, IL10, IL17a, Ki67, LAG3, PD1) were not significantly different between the three groups. Tregs showed no changes in expression of immune checkpoints and displayed significantly increased proliferation (Ki67) in non-metastatic MSI-H CRC compared to the metastatic tumors (Supplementary Fig. 4). In cytotoxic T cells, expression of the effector molecule Granzyme B was elevated in MSI-H non-metastatic tumors compared to MSS metastatic tumors. Moreover, CD45RO and Ki67 expression in cytotoxic T cells were increased in MSI-H non-metastatic tumors compared to both MSI-H and MSS metastatic tumors (Fig. 3c). Stromal, but not intra-epithelial, localization of cytotoxic T cells in metastatic MSI-H tumors We observed that the immune cells in MSI-H and MSS metastatic tumors appeared to be excluded from the tumor compartment compared to non-metastatic tumors. To quantify this, we determined the distance towards the epithelial tumor cells of all distinct cell populations in the TME (fibroblasts, B cells, T cells, monocytes, neutrophils, NK cells). To this end, we constructed a spatial graph in which individual cell-centroids are represented as dots and segmented tumor regions are converted into polygons (Fig. 4a, Supplementary Fig. 4). The minimal distance between each given single cell and the tumor outline 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 was then determined for each cell type. The distance between the tumor compartment and 8Fibroblasts9, 8B cells9, 8Fibroblast & Monocyte9, and 8Neutrophils & NK cell9 were similar between groups. However, T cell populations, including cytotoxic T cells, localized very close to the tumor cells in MSI-H nonmetastatic tumors (median 2 µm distance) but significantly further away from the tumor cells in MSIH/MSS metastatic tumors (median distance 21 and 22 µm; p=2e-16) (Fig. 4b, c, Supplementary Fig. 6). Similarly, macrophage-containing cell populations were localized near the tumor cells in MSI-H nonmetastatic tumors, but further away from the tumor cells in metastatic tumors (Fig. 4b).

To assess cytotoxic T cell numbers in relation to their distance from the tumor compartment, we divided the cytotoxic T cell population in three distinct distance-based bins, <intra-epithelial= (≤ 0 µm), <proximal= (1 – 25 µm), and <distant= (> 25 µm) (Fig. 4d). The accumulation of T cells in the 8intraepithelial9 and 8proximal9 bins was significantly higher in MSI-H non-metastatic tumors when compared to MSI-H/MSS metastatic tumors, while 8distant9 T cells were similar among the groups (Fig. 4e). The levels of the 8intra-epithelial9 tumor infiltrating cytotoxic T cells clearly differentiated non-metastatic from metastatic tumors, and this difference appeared more robust than the difference in total T cell numbers (Fig. 3a). Moreover, intra-epithelial T cells expressed significantly higher levels of Granzyme B when compared to T cells located further away from the tumor cells in MSI-H non-metastatic tumors. This was not observed in MSI-H or MSS metastatic tumors (Fig. 4f).

To validate the relevance of the robust difference in cytotoxic T cell localization in the TMA samples, we validated our findings in a second independent cohort of patients with MSI-H metastatic disease (n=10) or MSI-H non-metastatic disease (n=1 1), using large tumor sections (Fig. 5a). We performed immunohistochemistry staining for the detection of epithelial tumor cells (E-cadherin), stromal fibroblasts, (αSMA) and cytotoxic T cells (CD8a) ( Fig. 5b). Cytotoxic T cells in non-metastatic MSIH tumors localized significantly (p=0.00012) closer to tumor cells than those in metastatic MSI-H tumors (Fig. 5c, d). Stromal content was significantly (p=0.029) highe r in MSI-H metastatic tumors than in non– metastatic tumors (Fig. 5e).

Together, our findings support a model in which metastatic primary MSI-H tumors characterized by extensive formation of fibroblast-rich stroma in which cytotoxic T cells are residing and remain inactive, while in non-metastatic MSI-H tumors the cytotoxic T cells localize intra-epithelially, close to tumor cells, proliferate, and have experienced tumor antigen (Fig. 5f).

In this study, we profiled 46,727 single cells using multiplex imaging mass cytometry and analyzed both differential cell type abundance, and spatial distribution of stromal and immune cells in metastatic and non-metastatic primary colorectal tumors. We demonstrate that MSI-H tumors with metastatic capacity are characterized by extensive formation of fibroblast-rich stroma containing inactive cytotoxic T cells which fail to invade the tumor cell compartment. By contrast, non-metastatic MSI-H tumors were characterized by infiltrating and actively proliferating antigen-experienced cytotoxic T cells (CD8 +CD45RO+Ki67+), which produced high levels of Granzyme B, indicating cytotoxic activity. Moreover, these tumors displayed a higher number of proliferating Tregs, and helper T cells with elevated expression of immune checkpoint molecules (CTLA-4). These results show that both anti-tumor and immune-suppressive responses occur in the microenvironment of non-metastatic MSI-H tumors, with a predominantly cytotoxic T cell response.

T cell infiltration and activation is associated with a better prognosis in CRC2,3,31, and our findings confirm the prognostic value of cytotoxic T cell infiltration for metastatic MSI-H CRC. The numbers of cytotoxic T cells detected might be less reliable due to location-dependent bias and sample variation, in particular when analyzing relatively small tumor regions. Here, we show that intra-tumoral localization of 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224

T cells is a more robust method. Our results are in line with a correlation between intra-tumoral localization of cytotoxic T cells with a memory phenotype, and survival demonstrated in breast cancer32, and confirm the value of spatial analysis of the tumor microenvironment33. Tumors containing high numbers of fibroblasts often have low levels of lymphocytes34,35. A recent study demonstrated that specific subsets of cancer-associated fibroblasts with each unique molecular programs and organization patterns, associate with T cell exclusion in human lung cancers36. Increased stroma is generally associated with worse overall survival and recurrence in CRC patients after surgery23,24,37. Stromal localization of cytotoxic T cells has thus far not been associated with the risk of metastasis formation in CRC patients. Our study suggests that tumor stroma may contribute to metastasis by trapping cytotoxic T cells and preventing T cell expansion and activation.

Immune Checkpoint Blockade (ICB) therapy has led to durable responses and significantly improved overall survival, and now replaces first-line chemotherapy in metastatic MSI-H CRC patients3840. Furthermore, neoadjuvant PD-1 and CTLA-4 ICB therapy in primary colon cancers (stage I-III) showed that MSI-H tumors responded to ICB (n=20), and the presence of CD8 +PD1+ T cells predicted ICB response41. Here, our spatial analysis revealed that primary MSI-H tumors with metastatic capacity were largely devoid of intra-epithelial cytotoxic T cells. The success of ICB in relation to T cell intra-epithelial versus stromal localization in the tumor remains to be investigated in the future.

In conclusion, our study reveals that primary MSI-H tumors that gave rise to metastases are characterized by low numbers of inactive cytotoxic T cells localized in the stroma, in both early (I-III) and late tumor stages (IV). We propose that combined histopathological evaluation of stroma content, cytotoxic T cell activation status and spatial localization, may have value in predicting metastasis formation in early-stage MSI-H CRC. Such prognostic value should now first be determined in larger patient cohorts. The next step would be to develop a robust metastasis-prediction test for MSI-H CRC 225 226 which may have significant clinical utility in terms of patient selection for intensified follow-up and/or 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 We generated a collection of primary tumor tissues derived from metastatic MSI-H-CRC (n=3; CAIRO3 (Supplementary Table 1). Patients did not receive systemic therapy before resection. The collection and processing of human tissue specimens used for validation of our findings (Supplementary Table 2) was performed in accordance with the Declaration of Helsinki and approved by the Medical Research Ethics Committee of Utrecht, the Netherlands (METC 20-102) and Dutch Nationwide Pathology Databank (PALGA 2020-53). This tissue is classified as <residual material= and the collection and processing of this biological material is in accordance with the <no objection= procedure defining the release of anonymized residual material without broad consent under strict conditions and approval under aforementioned Research Ethics Committee. One patient (out of 21) received capecitabine + oxaliplatin treatment before resection.

For a comprehensive list of antibodies, compounds, and kits, see Supplementary Table 326. Tumor sections of 4 µm thickness were deparaffinized and rehydrated. Endogenous peroxidase activity was blocked with 1.5% hydrogen peroxide for 10 min. Heat-induced antigen retrieval was carried out using citrate buffer pH6.0 for 20 min, followed by cooling of tissue sections for 20 min. Sections were incubated overnight at 4°C with primary antibodies (Supplementary Table 3) diluted in PBS with 0.1% Sodium azide and 3% BSA. The next day, sections were washed three times with 0.05% Tween-PBS solution for 5 min, followed by a 1 hour incubation with a HRP-conjugated secondary antibody. 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270

Subsequently, sections were washed three times with 1× PBS for 5 min and developed with 3,39Diaminobenzidine (DAB) chromogen for 10 min at room temperature. Sections counterstained with hematoxylin, followed by dehydration and mounting.

Stained tissue sections were scanned using NanoZoomerXR whole slide scanner (Hamamatsu) at 40X magnification, with a resolution of 0.25 μm/pixel. Quantification of the staining was performed using QuPath software43. Targets of interest were quantified using QuPath9s Trained Pixel classification command that allows an automated recognition of background, tissue (haematoxylin) and DAB staining areas. The percentage of target in regions of interest (e.g. αSMA for stroma content) was then calculated by using the following formula: DAB area / (DAB + tissue – background area) * 100%.

Imaging mass cytometry was performed exactly as described in the MATISSE protocol26,29. Segmentation of single cells was performed on IMC data in a two-step process as described before26,29, with the following deviations: First, a pixel classification workflow was used to identify nuclei and membranes of stromal cells in all images using IMC data. The resulting probability maps were used to perform single-cell segmentation. Importantly, individual tumor cells were neglected in these steps. The following channels were used for nucleus and membrane pixel identification: Histone H3, DNA-lr193 and CD3, CD4 CD8a, CD14, CD16, CD20, CD45, CD45RO, CD68 .

Tissue type training for identification of tumor, stroma, and background regions was performed in Ilastik with a pixel classification workflow using the following IMC channels: Histone H3, DNA-Ir193, E-Cadher in, Pan Keratin, and Beta-catenin. The following features were selected: Gaussian Smoothing, Laplacian of

Gaussian, Gaussian Gradient Magnitude, Differenced of Gaussians, Structure Tensor Eigenvalues, Hessian of Gaussian Eigenvalues (σ3.5, 5, 10).

Single cell data was generated in R (v4.0) by extracting pixel intensities from unscaled 32-bit images for all channels for all cells represented in the segmentation maps, as described before26,29. Necrotic tissue regions were manually excluded from analysis based on IMC and toluidine blue images. High-intensity noise artefacts in IMC images were corrected by excluding single-cell events overlapping with these regions. The artefacts were identified with a pixel classification workflow in Ilastik using all acquired channels. The following feature was selected: Gaussian Smoothing (σ0.3, 1.0). Single-cell events with a mean probability above 0.1 for predicated noise were excluded. Samples from patient 2 were excluded from analysis due to bad tissue quality resulting in poor signal-noise ratio in staining.

Intensities were log1p scaled. Next, signal levels for individual image regions were normalized using the median of scaled intensities of the following channels: AKT (pan), CD20, CD45, CD45RO, Cl_Casp3, CTLA4, ERK (pan), FOXP3, IFNγ, LAG3, NFkB, pAKT, pSTAT3, Tbet.

Single-cell events were clustered using Rphenograph (k=20) on normalized and rescaled (R scale function) intensity values using the following lineage markers: aSMA, CD3, CD4, CD8a, CD14, CD16, CD20, CD45RO, CD68, FOXP3, IL17a. Resulting cluster s were manually assigned to cell types based on marker expression patterns. analysis.

Unless otherwise specified, expression values were calculated as median expression of all cells per image. Heatmaps were generated using the pheatmap (RRID: SCR_016418) package in R. For IMC data, segmented tumor regions were converted to polygons in R using the sf package44. Buffer regions around tumor regions were generated using the st_buffer function. Distance to tumor regions are centroid based. In frequency distribution graphs distance is clipped to 75μm.

For IHC data, tumor and stroma regions per image are segmented using a pixel classification workflow in QuPath on CD8a-stained sections. Segmen tation was optimized on a single-sample level and verified using a serial E-Cadherin stained section. CD8a cytotoxic T cells were identified using a pos itive cell detection workflow in QuPath. Single-cell events were converted to centroids, and closest distance to tumor borders was calculated. Events with distances above 75μm were discarded from further For cellular frequency analysis we used one-way ANOVA and Tukey HSD post-hoc test. For expression analysis, we used Kruskal-Wallis and Dunn post-hoc test with Bonferroni correction.

All data will be made publicly available in Zenodo repository (https://doi.org/10.5281/zenodo.7692926) All code will be made publicly available in Github (https://github.com/VercoulenLab). 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

Taieb, J. et al. Deficient mismatch repair/microsatellite unstable colorectal cancer: Diagnosis, prognosis and treatment. European Journal of Cancer 175, 136-157 (2022). https://doi.org:https://doi.org/10.1016/j.ejca.2022.07.020 Galon, J. et al. Type, density, and location of immune cells within human colorectal tumors predict clinical outcome. Science 313, 1960-1964 (2006). https://doi.org:10.1126/science.1129139 Bruni, D., Angell, H. K. & Galon, J. The immune contexture and Immunoscore in cancer prognosis and therapeutic efficacy. Nat Rev Cancer 20, 662-680 (2020). https://doi.org:10.1038/s41568020-0285-7 Popat, S., Hubner, R. & Houlston, R. S. Systematic review of microsatellite instability and colorectal cancer prognosis. J Clin Oncol 23, 609-618 (2005). https://doi.org:10.1200/JCO.2005.01.086 Kim, C. G. et al. Effects of microsatellite instability on recurrence patterns and outcomes in colorectal cancers. Br J Cancer 115, 25-33 (2016). https://doi.org:10.1038/bjc.2016.161 Tran, B. et al. Impact of BRAF mutation and microsatellite instability on the pattern of metastatic spread and prognosis in metastatic colorectal cancer. Cancer 117, 4623-4632 (2011). https://doi.org:10.1002/cncr.26086 Wensink, G. E. et al. Survival of patients with deficient mismatch repair metastatic colorectal cancer in the pre-immunotherapy era. Br J Cancer 124, 399-406 (2021). https://doi.org:10.1038/s41416-020-01076-0 Guinney, J. et al. The consensus molecular subtypes of colorectal cancer. Nat Med 21, 1350-1356 (2015). https://doi.org:10.1038/nm.3967 Zhang, B. et al. Proteogenomic characterization of human colon and rectal cancer. Nature 513, 382-387 (2014). https://doi.org:10.1038/nature13438 Vasaikar, S. et al. Proteogenomic Analysis of Human Colon Cancer Reveals New Therapeutic Opportunities. Cell 177, 1035-1049 e1019 (2019). https://doi.org:10.1016/j.cell.2019.03.030 Lee, H. O. et al. Lineage-dependent gene expression programs influence the immune landscape of colorectal cancer. Nat Genet 52, 594-603 (2020). https://doi.org:10.1038/s41588-020-0636-z Joanito, I. et al. Single-cell and bulk transcriptome sequencing identifies two epithelial tumor cell states and refines the consensus molecular classification of colorectal cancer. Nat Genet 54, 963975 (2022). https://doi.org:10.1038/s41588-022-01100-4 Pages, F. et al. International validation of the consensus Immunoscore for the classification of colon cancer: a prognostic and accuracy study. Lancet 391, 2128-2139 (2018). https://doi.org:10.1016/S0140-6736(18)30789-X Punt, C. J., Koopman, M. & Vermeulen, L. From tumour heterogeneity to advances in precision treatment of colorectal cancer. Nat Rev Clin Oncol 14, 235-246 (2017). https://doi.org:10.1038/nrclinonc.2016.171 Biller, L. H. & Schrag, D. Diagnosis and Treatment of Metastatic Colorectal Cancer: A Review. JAMA 325, 669-685 (2021). https://doi.org:10.1001/jama.2021.0106 Sargent, D. J. et al. Defective mismatch repair as a predictive marker for lack of efficacy of fluorouracil-based adjuvant therapy in colon cancer. J Clin Oncol 28, 3219-3226 (2010). https://doi.org:10.1200/jco.2009.27.1825 Kucukkose, E. et al. Mismatch Repair Status in Patient-Derived Colorectal Cancer Organoids Does Not Affect Intrinsic Tumor Cell Sensitivity to Systemic Therapy. Cancers (Basel) 13 (2021). https://doi.org:10.3390/cancers13215434 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

Wu, T. & Dai, Y. Tumor microenvironment and ther apeutic response. Cancer Lett 387, 61-68 (2017). https://doi.org:10.1016/j.canlet.2016.01.043 Llosa, N. J. et al. The vigorous immune microenvironment of microsatellite instable colon cancer is balanced by multiple counter-inhibitory checkpoints. Cancer Discov 5, 43-51 (2015). https://doi.org:10.1158/2159-8290.CD-14-0863 Khaliq, A. M. et al. Refining colorectal cancer classification and clinical stratification through a single-cell atlas. Genome Biol 23, 113 (2022). https://doi.org:10.1186/s13059-022-02677-z Pelka, K. et al. Spatially organized multicellular immune hubs in human colorectal cancer. Cell 184, 4734-4752 e4720 (2021). https://doi.org:10.1016/j.cell.2021.08.003 Lin, A., Zhang, J. & Luo, P. Crosstalk Between the MSI Status and Tumor Microenvironment in Colorectal Cancer. Front Immunol 11, 2039 (2020). https://doi.org:10.3389/fimmu.2020.02039 Isella, C. et al. Stromal contribution to the colorectal cancer transcriptome. Nat Genet 47, 312319 (2015). https://doi.org:10.1038/ng.3224 Calon, A. et al. Stromal gene expression defines poor-prognosis subtypes in colorectal cancer. Nat Genet 47, 320-329 (2015). https://doi.org:10.1038/ng.3225 Andre, T. et al. Pembrolizumab in Microsatellite-Instability-High Advanced Colorectal Cancer. N Engl J Med 383, 2207-2218 (2020). https://doi.org:10.1056/NEJMoa2017699 Baars, M. J. D. et al. MATISSE: a method for improved single cell segmentation in imaging mass cytometry. BMC Biol 19, 99 (2021). https://doi.org:10.1186/s12915-021-01043-y Berg, S. et al. ilastik: interactive machine learning for (bio)image analysis. Nat Methods 16, 12261232 (2019). https://doi.org:10.1038/s41592-019-0582-9 Carpenter, A. E. et al. CellProfiler: image analysis software for identifying and quantifying cell phenotypes. Genome Biol 7, R100 (2006). https://doi.org:10.1186/gb-2006-7-10-r100 Krijgsman, D. et al. MATISSE: An analysis protocol for combining imaging mass cytometry with fluorescence microscopy to generate single-cell data. STAR Protoc 3, 101034 (2022). https://doi.org:10.1016/j.xpro.2021.101034 Levine, J. H. et al. Data-Driven Phenotypic Dissection of AML Reveals Progenitor-like Cells that Correlate with Prognosis. Cell 162, 184-197 (2015). https://doi.org:10.1016/j.cell.2015.05.047 Bindea, G., Mlecnik, B. & Galon, J. Immune sunrise: from the immunome to the cancer immune landscape. Oncoimmunology 11, 2019896 (2022). https://doi.org:10.1080/2162402x.2021.2019896 Egelston, C. A. et al. Resident memory CD8+ T cells within cancer islands mediate survival in breast cancer patients. JCI Insight 4 (2019). https://doi.org:10.1172/jci.insight.130000 van Dam, S., Baars, M. J. D. & Vercoulen, Y. Multiplex Tissue Imaging: Spatial Revelations in the Tumor Microenvironment. Cancers (Basel) 14 (2022). https://doi.org:10.3390/cancers14133170 Jiang, P. et al. Signatures of T cell dysfunction and exclusion predict cancer immunotherapy response. Nat Med 24, 1550-1558 (2018). https://doi.org:10.1038/s41591-018-0136-1 Chakravarthy, A., Khan, L., Bensler, N. P., Bose, P. & De Carvalho, D. D. TGF-beta-associated extracellular matrix genes link cancer-associated fibroblasts to immune evasion and immunotherapy failure. Nat Commun 9, 4692 (2018). https://doi.org:10.1038/s41467-01806654-8 Grout, J. A. et al. Spatial positioning and matrix programs of cancer-associated fibroblasts promote T cell exclusion in human lung tumors. Cancer Discov (2022). https://doi.org:10.1158/2159-8290.CD-21-1714 Calon, A. et al. Dependency of colorectal cancer on a TGF-beta-driven program in stromal cells for metastasis initiation. Cancer Cell 22, 571-584 (2012). https://doi.org:10.1016/j.ccr.2012.08.013 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423

Overman, M. J. et al. Durable Clinical Benefit With Nivolumab Plus Ipilimumab in DNA Mismatch Repair-Deficient/Microsatellite Instability-High Metastatic Colorectal Cancer. J Clin Oncol 36, 773-779 (2018). https://doi.org:10.1200/JCO.2017.76.9901 Diaz, L. A., Jr. et al. Pembrolizumab versus chemotherapy for microsatellite instability-high or mismatch repair-deficient metastatic colorectal cancer (KEYNOTE-177): final analysis of a randomised, open-label, phase 3 study. Lancet Oncol 23, 659-670 (2022). https://doi.org:10.1016/S1470-2045(22)00197-8 Cervantes, A. et al. Metastatic colorectal cancer: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up(†). Ann Oncol (2022). https://doi.org:10.1016/j.annonc.2022.10.003 Chalabi, M. et al. Neoadjuvant immunotherapy leads to pathological responses in MMRproficient and MMR-deficient early-stage colon cancers. Nat Med 26, 566-576 (2020). https://doi.org:10.1038/s41591-020-0805-8 Simkens, L. H. et al. Maintenance treatment with capecitabine and bevacizumab in metastatic colorectal cancer (CAIRO3): a phase 3 randomised controlled trial of the Dutch Colorectal Cancer Group. Lancet 385, 1843-1852 (2015). https://doi.org:10.1016/s0140-6736(14)62004-3 Bankhead, P. et al. QuPath: Open source software for digital pathology image analysis. Sci Rep 7, 16878 (2017). https://doi.org:10.1038/s41598-017-17204-5 Pebesma, E. Simple Features for R: Standardized Support for Spatial Vector Data. The R Journal 10, 439-446 (2018). https://doi.org:https://doi.org/10.32614/RJ-2018-00 9 427 428 429 430 431 432 433 434 435 436 437 438 439

Fig. 1: TME markers distinguish non-metastatic from metastatic colorectal tumors. (a) Illustration of the data acquisition workflow used for Imaging mass cytometry (IMC): (1) One 4 µm-thick section of a TMA composed of 2 mm2 CRC cores (2) was stained with a cocktail of 35 antibodies labeled with unique metal isotopes. (3) Samples were ablated with a high-energy laser in a rasterized pattern. The resulting plumes were ionized and analyzed by a mass cytometer, which returned the metal isotope composition per pixel. (4) Each antibody resulted in a single image per sample, and together constructed a multi-channel image stack. (b) Stromal content was quantified by analyzing α-Smooth muscle actin (aSMA) expression. Boxplot represents the percentage of aSMA+ stroma (fibroblasts) in tissue per group, individual patients are color-coded. (c) Principal component analysis using the median intensity of each marker per image region as input variables. Each data point is a given ROI, color-coded by tumor state. Black lines depict the contribution of individual markers to the first two principal components (PC1 and PC2) of the analysis. (d) Hierarchical clustering analysis of ROIs according to median marker intensity of segmented stromal/immune cells. Tumor state and patient ID are indicated using a color code on the right side.

Fig. 2: Cell-type annotation reveals differential microenvironment composition between metastatic and non-metastatic primary CRC tumors. (a) Hierarchical cluster analysis heatmap of lineage marker 443 444 445 446 expression. The heatmap shows the expression levels of 11 markers used for defining cell clusters. (b) Uniform Manifold Approximation and Projection (UMAP) of single cells (N=46,727) of all ROIs combined using lineage marker intensities, color-coded by annotated cell lineages. (c) Heatmap showing hierarchical clustering on cell-type abundance for all cell lineages and individual ROIs. (d) Boxplot showing the median of number of cells per mm2 stratified for annotated cell-types, across disease groups 448 449 450 451 452 453 454 stacked bar plot represent the proportions of cytotoxic-, helper-, regulatory- or other T cell of all T cells. (b) UMAP of all stromal and immune cells, highlighting the helper T cells as red dots. Boxplots represent the median expression per ROI of given markers for helper T cells. (c) UMAP of all stromal and immune cells, cytotoxic T cells are indicated as red dots. Boxplots represent the median expression of given markers per ROI for cytotoxic T cells. Spatial representation of representative ROIs per disease group with segmented tumor regions displayed as grey polygons and the identified cells are shown as dots, colored by cell-type. (b) Tumor distance plots demonstrating the relative frequency distribution of given cell populations. (c) Boxplot represents the median epithelial distance of cytotoxic T cells per group. (d) Spatial representation of cytotoxic T cell localization relative to the tumor, color-coded by distance bin: 8intra9 (blue), 8proximal9 (1 – 25 µm; green) and 8distant9 (>25 µm; red). Scale bar indicates 100 µm. (e) Boxplots represent the number of cytotoxic T cells per mm2 per distance bin and group (median per ROI). (f) Boxplots represent the median Granzyme B expression in cytotoxic T cells per distance bin and group (median per ROI).

Fig. 5: Cytotoxic T cells reside in high stromal content in metastatic MSI-H tumors. (a) Illustration of the data acquisition workflow used for the IHC validation cohort: (1) FFPE sections of resected primary CRC with (n=10) or without (n=11) metastatic disease were (2) stained for E-cadherin, aSMA and CD8a using IHC. (3) Tumor segmentation was performed, and masks were generated using Qupath43. Individual cytotoxic T cells were identified. (b) Representative IHC images of E-cadherin, aSMA, and CD8a. Scale bar is 100 µm. (c) Tumor distance plots demonstrating the frequency distribution of cytotoxic T cells per tumor and group. (d) Boxplot representing the median epithelial distance of cytotoxic T cells per ROI, shown per group. (e) Boxplot representing the percentage of stroma per ROI and group. (f) Schematic model of CD8+ T cell phenotype and localization in the CRC microenvironment of metastatic and nonmetastatic MSI-H tumors.

Supplementary Fig. 1: Representative images demonstrating all markers acquired in the multiplex imaging mass cytometry. All 35 antibody-stainings used for IMC are shown. Scale bar sizes are indicated.

Supplementary Fig. 2: Negative controls, exclusion of necrotic regions, and single-cell and tumor segmentation in IMC data. (a) Negative controls for p-STAT3, p-AKT, p-ERK, IL1- and IFNy stainings in IMC. Scale bar is 50 µm. (b) Representative ROI demonstrating necrotic area, left: IMC staining data, necrotic region outlined in yellow. Right: Slide-scan image of same section, necrotic region outlined in black (see also Supplementary Table 4). Scale bar is 100 μm. (c) Representative image showing the predicted outlines for single-cell (magenta) and tumor (green) segmentation.

Supplementary Fig. 3: Cell-type annotation by clustering single-cells on lineage marker expression. (a) Representative IMC images showing pan-keratin (red), CD3 (blue), CD8 (white) and CD4 (yellow) per group. Scale bar is 100 µm. (b) Violin plot representing scaled expression of lineage markers per generated Rphenograph cluster.

Supplementary Fig. 4: Increased proliferating regulatory T cells in non-metastatic MSI-H tumors compared to metastatic tumors. UMAP of all stromal and immune cells, highlighting the regulatory T cells as red dots. Boxplots represent the median expression per ROI of given markers for regulatory T

Supplementary Fig. 5: Spatial immune landscape of CRC microenvironment for all samples. Spatial representation of segmented tumor regions shown as grey polygons, and single-cell events shown as dots, colored by the 12 identified cell populations. All ROIs are included.

Supplementary Fig. 6: Relative tumor distance of cytotoxic T cells in individual IMC ROIs. Frequency distribution of cytotoxic T cell tumor distance for all CyTOF IMC ROIs, per group.

Supplementary Table 1: Patient characteristics used for CyTOF IMC.

Supplementary Table 2: Patient characteristics used for IHC.

Supplementary Table 3: Antibody list IMC and IHC.

Supplementary Table 4: Single-cell event count and curation per image region. 509 510 512 513 514 515 517 518 520 521 522 523

Study design: EK, MJDB, OK, YV. Experimental work: EK, MJDB, MA, EF. Resources: SS, GB, JMLR, MK. Data interpretation: EK, MJDB, IHMBR, JL, OK, YV. Study supervision: OK, YV. Writing manuscript: EK,