This study was conducted in compliance with ethical standards and approved by the University of Florida Institutional Review Board (IRB) under protocol IRB202001363. All subjects were informed on the purpose of the study and gave written informed consent before participation. Informed consent was obtained from all participants or their legal guardians prior to the collection of any tissue sample. The study was carried out in accordance with the protocol and relevant guidelines and regulations, and the experimental protocol was approved by the University of Florida IRB (IRB202001363). No animal models were used during the conduct of this study.

The procedure of microexplant generation was previously reported (24). In brief, the surgically resected specimens were immediately submerged in ice-cold, calcium and magnesium-free PBS (1X) containing 10% FBS and transported to the laboratory within two hours of the surgery. External membranes and observable adipose tissues were removed using sterile forceps and scissors. The samples were washed three times with ice-cold PBS (1X). Holding on to one end of the sample with forceps, the sample was cut into smaller pieces using a surgical scalpel. The pieces were further cut with surgical scissors to sizes no larger than 1 mm. Without enzyme digestion, these microexplants were then transferred to a 15-mL tube containing approximately 10mL of ice-cold PBS and further chilled on ice for at least 5 minutes. The supernatant was discarded once a relatively clear collection of microexplants had settled at the bottom. This procedure was repeated three times to remove single cells and decellularized extracellular matrix (ECM) fragments. The collected microexplants were placed in 3D perfusion culture using pre-wetted wide-bore micropipette tips.

Peripheral blood mononuclear cells (PBMCs) were isolated from the patient peripheral blood (~ 30–50 mL) using a Ficoll gradient (Ficoll-Paque Plus; Amersham Biosciences). The cells were labeled with carboxyfluorescein succinimidyl ester - CFSE (Cell Division Tracker Kit, 423801; BioLegend) and counted with a hemocytometer prior to the co-culture experiment.

Perfusion culture was conducted as previously described (24). In brief, the microtumor samples were suspended in media-equilibrated liquid-like solid (LLS) microgels. A small sample of the microtumor-LLS mixture was placed on an inverted light microscope to count the microtumors per unit volume. Each well of a perfusion plate was loaded with 200µL of LLS containing PBMCs uniformly suspended at 10⁶ cells/mL and approximately 30–40 CRC microtumors (Figures 1A–D). After loading, the plate was centrifuged at 100xg for 5 minutes at room temperature with a moderate acceleration and deceleration setting (Thermo Fisher Scientific, Waltham, MA, USA, Sorvall ST 40R). Culture media was carefully added to each quadrant of the plate to avoid disturbing the centrifuged microgels and samples. A 100cc Jackson-Pratt drain (Care Express, Cary, IL, USA, SU130-1305) was connected to establish a pressure gradient and initiate media perfusion as previously described (24). The culture conditions were maintained at 37°C with 5% CO₂. The perfusion plates were maintained every 24 hours by adding pre-warmed media to the media reservoirs and collecting perfused media from the waste collection compartments.

Patient-derived microtumors and autologous PBMC were co-cultured long-term in the presence of pembrolizumab (anti-PD-1). We established three (3) control conditions: co-culture of tumor microexplants and CFSE labelled PBMCs, incorporating a human IgG4 isotype control (ISO ctrl); microtumors or PBMC alone with ISO ctrl or with anti-PD-1. Experimental conditions are clearly defined in the Results and the figures. The analysis was centered on differentiating evidence of immune response from MSI CRC as compared to MSS CRC. The IFNγ production between the responding versus non-responding cases was measured (Figures 1E–G). Note: pembrolizumab (anti-PD-1) was kindly provided by Merck & Co., Inc., Rahway, NJ, USA. The concentration of both pembrolizumab and the isotype control (IgG4) was 20 µg/mL for all experiments. Clinically, pembrolizumab achieves steady-state trough serum concentrations of approximately 50 µg/mL in patients receiving the standard 300 mg every two weeks (Q2W) dosing regimen. In ex vivo co-culture models, the absence of clearance mechanisms and the defined volume of culture medium enable consistent antibody exposure to target cells. A concentration of 20 µg/mL offers a conservative and effective dose that falls within the biologically active range and has been widely used in previous preclinical studies to block PD-1 signaling and elicit T-cell activation (29, 30).

The immunofluorescence (IF) staining protocol was as previously completed in (24, 25). CRC microtumor samples were fixed in 4.0% formaldehyde in 1X PBS overnight at 4°C, washed twice, and incubated in PBS (1X) for 1 hour at room temperature. The samples were then permeabilized in 0.5% Triton X-100 (Sigma-Aldrich, St. Louis, MO, USA, X100-100ML) for 2 hours, washed twice, and blocked with 3% bovine serum albumin in PBS for 3 hours at room temperature. After blocking, samples were washed 3 times with PBS and incubated overnight with conjugated antibodies at 4°C. The antibodies used in this assay included E-cadherin (BD Biosciences, San Jose, CA, USA, 560062), Alexa Fluor^® 488 Anti-CD45 antibody (Abcam, Cambridge, UK, ab197730), and Alexa Fluor™ 568 Phalloidin (Invitrogen, Waltham, MA, USA, A12380). After overnight incubation with the antibodies, the samples were washed 3 times with PBS and counterstained with Hoechst 33342 (Invitrogen, Waltham, MA, USA, H3570) for 15 minutes before imaging.

For live-cell viability assay, a live/dead kit (Thermo Fisher, Waltham, MA, USA, R37601) containing Calcein AM and BOBO-3 iodide was used, following the manufacturer’s protocol. Briefly, Calcein AM was gently mixed with BOBO-3 iodide. The solution was added to an equal volume of media-containing samples and incubated at 37°C with 5% CO₂ for 30 minutes. Hoechst 33342 (Invitrogen, Waltham, MA, USA, H3570) was added to visualize cell nuclei. Subsequently, the samples underwent at least two washes with PBS (1X) before imaging. Samples were submerged in pre-warmed culture media and placed in a custom-built stage incubator for live-cell imaging.

Terminal deoxynucleotidyl transferase (TdT) dUTP Nick-End Labeling (TUNEL) assay (Click-iT Plus TUNEL assay, Invitrogen, Waltham, MA, USA, C10619) was applied to detect apoptosis in cancer cells and correlated with infiltrating CFSE labeled PBMCs. The assay was conducted following manufacturer protocol. All CRC microtumors were then imaged using a Nikon A1R HD25 confocal microscope equipped with a high-definition Galvano scanner.

Formalin-fixed paraffin-embedded (FFPE) tissues were sectioned at 5µm thickness and stained on the Leica Bond autostainer using optimized conditions for CD8 (CST#85336), CD4 (CST#48274), PD-1 (CST#86163), or PD-L1 (Abcam, Cambridge, UK, ab228415) antibodies. Serial sections were subjected to hematoxylin and eosin (H&E) staining.

FFPE tissues were sectioned at 5µm thickness, deparaffinized with xylene, and rehydrated with successive washes in reagent alcohol and water. Antigen retrieval was performed using Tris/EDTA pH9 solution (Dako) prior to blocking with 5% donkey serum. Slides were stained overnight at 4°C using a 25-plex cocktail of antibodies (dsDNA, beta-tubulin, CD298, CD4, CD11C, FOXP3, PD-1, PD-L1, GZMB, CD56, CD31, Ki67, CD14, CD68, CD8, CD3, CD45RO, VIM, alpha-SMA, pan-cytokeratin [pan-CK], CD20, PDPN, HLA-DR, HLA-ABC, and CD45). Following washings with TBS-T, slides were fixed for 5 minutes in 2% glutaraldehyde (Electron Microscopy Sciences), dehydrated with decreasing dilutions of ethanol, and dried prior to acquisition on MIBIscope. Ten (10) 800x800µm regions of interest (ROIs) per sample were selected based on the presence of immune infiltrates in serial H&E stains. The spatial locations of ROIs are clearly defined in the figures.

Following image quality control (QC), ROI regions were quantified as tumor based on pan-CK positivity or stroma based on VIM or alpha-SMA positivity. Watershed segmentation using a combination of nuclear, cytoplasmic, and membrane markers was used for immune population quantification. Cell classification was performed using a machine learning framework with pathologist-defined intensity thresholds for immune lineage markers. Marker intensities were quantified as ion counts per cell and cell densities were quantified as cells per mm². The nearest neighbor analysis quantified the average distance from tumor cells to immune lineages. Statistical analyses were performed with GraphPad Prism8 using a one-way ANOVA with Tukey’s multiple comparison test. A cell density heat map was generated by normalizing population density with a z-score computation. Each column represents a single ROI, and each row reflects a different cell population. Red indicates higher values, and blue indicates lower values, as shown in the color scale.

A multi-parameter bead-based ELISA MSD IFNγ Small Spot assay (Meso Scale Diagnostics, MD, Cat#K151TTK-2) was employed to measure the IFNγ level from the daily perfused effluent media from the individual well of the perfusion platform. The assay was conducted following manufacturer protocol.

Whole exome sequencing (WES) reads were aligned to human reference genome GRCh37 with Sentieon v201808.06 (31) including post-processing steps (deduplication, indel realignment and base quality recalibration) to generate the bam files for downstream analysis. Somatic mutations in the tumor samples were called with GATK v4.1.4.0 Mutect 2 (32) using the matched tumor-infiltrated lymphocyte (TIL) samples as paired-normal controls. A panel of 522 normal samples was also used. Mutations with total read depth < 15 or mutant allele read depth < 4 were excluded, and the resulting mutations were further filtered with dbSNP b151 (33) and COSMIC v90 (34) databases. Specifically, mutations listed in dbSNP while not present in COSMIC were excluded. VEP v98 (35) was used to annotate the mutations. All protein-altering mutations (including splicing site mutations) were used for calculating mutation load. The length of regions used to calculate the normalized mutation count per megabase value was obtained by intersecting the WES bait regions and refGene exon regions padded with +/- 2 bp for splicing sites. The R package maftools (36) was used to generate the plots of mutations on the POLE protein.

GraphPad Prism software was used for statistical analysis. For Figure 2, we used a two-way ANOVA followed by Tukey’s HSD posthoc test for IFNγ level between the treatment groups at each time point. For Figure 4, we used one-way ANOVA with multiple comparison tests for multiple-group comparisons. Statistical analysis was also indicated in the figure captions.

The CRC cases were selected for the ex vivo model based on the following criteria: being treatment-naïve, having a sufficient blood draw (at least 35–50 mL) for PBMC isolation, and having a sufficient sample size (≥ 1 cm³ or 1g) without observable fibrotic tissue (Figures 1A, B). Tissue specimens were collected from properly consented colorectal cancer patients enrolled in our research program, Microtissues for Oncology and Drug Discovery (MODEL or MOD) (Figure 2A, Supplementary Table S1).

After surgical resection, specimens were promptly collected and delivered to the pathology department for analysis. Subsequently, a portion of the tissue was forwarded to our laboratory for experiment. The specimens were mechanically dissociated into smaller tissue fragments (≤ 1 mm) and co-cultured with CFSE-labeled PBMCs (10⁶ cells/mL) as shown in the workflow in Figures 1A–D and 2B. In previous work, we demonstrated the capacity of T cells to navigate through the interstitial space of a Liquid-Like Solid (LLS) material, then infiltrate, and effectively eliminate target tumors in a 3D static co-culture (28). Building upon these previous findings (28), our objective was to construct an ex vivo model involving the co-culture of patient-derived tumors and immune cells to assess the anti-tumor response to Immune Checkpoint Inhibitor (ICI) therapy. We also employed a custom-built 3D platform to support the ex vivo co-culture of cells and tissues through the mechanism of media perfusion transport (24). In the culture chamber, the LLS provides a stable 3D support structure for immune cells and tumor microexplants, referred to as microtumors interchangeably in this study (28). The interstitial space between microgel particles facilitates cell motility and allows the perfusion of media (24, 28), (Figure 2C). The platform ensures a constant supply of culture media and the removal of waste metabolites through a continuous flow (24).

We observed a heterogeneous microtumor morphology with an intact tumor boundary, confirming the presence of viable cell populations and indicating signs of proliferation and self-organization over time, as typically shown in Figures 1F, 2D. To evaluate immune cell infiltration, the co-cultured samples were collected and thoroughly washed with PBS (1X) to remove surrounding non-infiltrating PBMC cells. Consequently, the detectable CFSE+ PBMC population in the microtumors exclusively represents infiltrated immune cells after long-term co-culture (Supplementary Figure S1). In summary, we demonstrated that microtumors and autologous PBMCs remained viable over an extended co-culture in our perfusion platform, making them suitable for drug screening.

We co-cultured patient-derived CRC microtumors with autologous CFSE-labeled PBMCs and treated them with pembrolizumab (defined as microtumors + PBMCs + anti-PD-1) at 20 µg/mL under perfusion. We established three control conditions to assess anti-tumor response: (1) co-culture of tumor microexplants and CFSE labelled PBMCs, incorporating a human IgG4 isotype control (defined as microtumors+ PBMCs+ ISO ctrl) at 20 µg/mL. This critical control ensured that any observed anti-tumor response would be specific to the PD-1 blockade. In addition, (2) microtumors alone and (3) PBMCs alone were used to evaluate the baseline anti-tumor activities from pre-existing Tumor-Infiltrating Lymphocyte (TILs) populations and the patient’s immune system respectively.

The levels of IFNγ in the daily perfused media were analyzed. IFNγ serves as a key indicator of immune-mediated anti-tumor function resulting from anti-PD-1 treatment. A significant elevation in IFNγ levels was observed in CRC microtumors-PBMC co-culture following anti-PD-1 treatment in MSI CRC MOD1 (Figure 2E). The IFNγ level remained consistently higher than in all three control conditions throughout the study, reaching its peak at 72 hours. This suggests a potential recovery period for these biological samples post-resection, tissue process, and ex vivo culture. Furthermore, the IFNγ level in MSS CRC MOD2 coculture, which was expected to be a non-responder to anti-PD-1 therapy, remained consistently low across all treatment conditions (Figure 2E).

Subsequently, we investigated immune infiltration by detecting CFSE+ cells and identified late-stage apoptosis using TUNEL assay. In the MSI CRC responder group, we observed a correlation between the spatial localization of a high number of infiltrating CFSE+ PBMCs (green) and TUNEL-labeled cancer cell death (white), indicating immune-mediated killing in microtumors-PBMCs co-cultured treated with anti-PD-1 (Figure 2F). This correlation was not observed in the three control groups. Consistently, we did not observe significant CFSE+ or TUNEL signal in the non-responder MSS CRC case. To verify the specificity of TUNEL labeling and CFSE+ signals, we co-stained for nuclei and the actin cytoskeletal network, ensuring the identification of cells and confirming intact tissue and cellular morphology respectively (Figure 2F).

Given recent clinical case studies reporting partial or complete responses in POLE-mutated MSS CRC, we aimed to investigate evidence of ex vivo response within our model. We received a case from a 37-year-old Caucasian male diagnosed with treatment-naïve, moderately differentiated MSS CRC harboring POLE mutations, who underwent surgery for tumor removal. The tumor specimen was collected, promptly sent to pathology, and then delivered directly to our laboratory for experiment. In this report, the case is labeled as MOD3, and its clinical data are presented in Figure 3A and Supplementary Table S1. NextGen sequencing data retrieved from the UF Shands Clinical and Translational Science Institute (CTSI) RedCap database (PID 11877, project title: MOD2EL v2) revealed multiple POLE mutations for this case. We also conducted whole exome sequencing (WES) and confirmed the presence of six missense mutations on the POLE full-length sequence as illustrated in Figure 3B. Notably, the amino acid mutation p.Ala456Pro (p.A456P; c.1366G>C) identified on exon 14, is situated in the exonuclease domain and has been suggested as a recurrent somatic aberration and identified as a pathogenic POLE hotspot mutation for endometrial cancer (37, 38). In addition, mutations in the POLE exonuclease domain are often found in MSS tumors, leading to hypermutation and associated response to immunotherapy (12). The remaining five mutations, distributed across other domains of the POLE gene, are of unknown significance.

In subsequent studies, we further characterized and compared the mutation profiles of additional CRC cases, including three MSS CRC MOD 4, 5 and 6. In addition, we performed an ex vivo experiment on an MSI CRC case namely MOD7, which also harbors POLE mutations, for comparison with MSS CRC MOD3. WES analysis of MOD7 unveiled the presence of two missense mutations and one splice site mutation within the POLE gene. The pathogenic significance of these mutations has not been reported. We presented the clinical data for these CRC cases in Figure 3A and Supplementary Table S1, along with their somatic mutation profiles in key CRC-associated genes (39) in Figure 3C. Although MOD3 and MOD7 share multi-hit POLE mutations, they demonstrate distinct driver gene profiles. Notably, MSS CRC MOD3 presents a significantly higher number of these mutations compared to other MSS CRC cases and even to MSI MOD7. MSS MOD4 and 5 exhibited the most similar driver gene mutations (TP53 and APC) and copy number alteration (CNA) profiles, with these two genes displaying the most widespread CNAs across the genome. Additionally, typical CNAs observed in CRC, such as gains in chromosome 13 (chr 13) and losses in chromosome 18 (chr 18), are evident in these samples. On the other hand, MOD6 exhibits a unique driver gene mutation profile, featuring alterations in PIK3CA, PTEN, and KRAS, along with specific CNAs, including losses in chr 1p, 17, and 18.

Furthermore, MSI CRC MOD7 with POLE mutations, demonstrates hypermutation with a high mutation load of 51 mutations per megabase (mutations/Mb) (Figure 3D). MOD7 exhibits a distinctive driver mutation profile, featuring numerous CNAs (gains in chr 1q, 2, 13), and a higher frequency of indel mutations (12 mutations/Mb), which is commonly observed in MSI CRCs. This mutation profile sets it apart from all other MSS cases in this study. Remarkably, MSS CRC MOD3 exhibits an ultra-mutated phenotype, revealing an even higher TMB with an estimated total mutation count of 11,329 and a mutation load of 236 mutations/Mb (Figure 3D). This is approximately more than four times that of MOD7 and approximately 70 times more than MOD4, 5, and 6 cases.

We proceeded to investigate the evidence of response to anti-PD-1 therapy for MSS CRC MOD3 and MSI CRC MOD7 using our ex vivo model, as described above. The workflow is illustrated in Figure 1, where patient whole blood and tumor tissue specimens were collected for the experiment following selection for pathology examination. IF assays at specific time points revealed microtumors with robust cell viability and intact boundaries, indicative of a well-organized and thriving culture (Figure 3E).

We established four conditions for the studies: (1) co-culture of MOD3 and MOD7 microtumors and patient-matched CFSE-labeled PBMCs with pembrolizumab (20 µg/mL) through continuous, unidirectional vertical perfusion. As a control (2) we also treated the co-culture with human IgG4 isotype control (ISO). In addition, (3) we treated microtumors alone with pembrolizumab (defined as: microtumors + anti-PD-1) to investigate potential anti-tumor responses stemming from TILs populations and (4) microtumors alone with ISO (defined as: microtumors + ISO ctrl) as an additional control. In line with the previous co-culture experiments, we evaluated daily perfused media for IFNγ levels, serving as an indicator of an immune-mediated anti-tumor response. Although not statistically significant, a higher trend in IFNγ production was observed in the microtumor-alone condition following anti-PD-1 treatment, suggesting a potential anti-tumor response from the TILs population (Figure 3F). Further investigation is needed to confirm this finding and clarify whether local immune reinvigoration contributes to effective recruitment of PBMCs and overall anti-tumor response. Figure 3G presents confocal IF images of a representative MOD3 microtumor cultured for 96 hours, revealing abundant CD45+ TILs in the microtumors. On the other hand, the co-culture of microtumors with PBMCs at the same anti-PD-1 dose resulted in a minimal response, possibly due to the presence of myeloid-derived suppressor cells within the PBMCs (40, 41), potentially dampening the anti-tumor effect (42, 43).

We closely examined the tumor microenvironment (TME) in the selected CRC cases to explore intra- and inter-tumor heterogeneity influencing the response to anti-PD-1 treatment. We employed multiplexed ion beam imaging (MIBI) platform to capture ten regions of interest (800 x 800µm) within formalin-fixed paraffin-embedded (FFPE) specimen sections (Figure 4A), ensuring a precise representation of the TME.

IHC staining of the entire tissue displayed a heterogeneous distribution of CD8+ and CD4+ T cells across all samples, with notably significant expression observed in MSI CRC MOD1 (Figure 4B). In addition, TILs are notably present in MSS CRC MOD3 compared to MOD2 and even MSI CRC MOD7. We applied a comprehensive 25-marker panel to conduct a thorough profiling of various immune subsets present within the complex landscape of the TME in these cases. The analyses were done blinded to patient outcome and CRC phenotype. Macroscopic region analysis initially revealed that MOD7 exhibits the highest stromal distribution, whereas MOD3 has the greatest tumor area, as quantified in Figure 4C and visually depicted in Figure 4A.

Cell segmentation and phenotyping revealed cellular heterogeneity, as evidenced by variations in cell population densities (Figure 4D), which shows higher cancer cell populations for MSS MOD3 and significantly high TILs in MSI MOD1. Further analysis showed significantly higher CD8+ T cell densities in MSI MOD1 compared to other samples (Figure 4E). Consistent with other studies, there is more PD-1 expression on CD8+ T cells for both MSI cases, MOD1 and MOD7. Remarkably, although the count of CD8+ T cells is not markedly elevated in MSS MOD3, spatial analysis of T cell distribution reveals that, on average, these immune cells are localized closer to tumor cells in both MOD1 and MOD3 cases compared to the rest of the samples (Figure 4F). However, these findings are limited by sample size and the number of ex vivo cases. Further investigations are necessary to draw a better definitive conclusion. This close cancer-immune proximity could contribute to the observed anti-tumor response to anti-PD-1 treatment in both cases. Intra- and inter-tumor heterogeneity in cell populations and phenotypes across all cases are effectively summarized and depicted in the cell density heatmap (Figure 4G). Notably, there is a significant elevation in the T cell population for MOD1, contrasting with MOD7 despite both sharing the same MSI CRC phenotype.