Multiomics Reveals the Immunologic Features and the Immune Checkpoint Blockade Potential of Colorectal Medullary Carcinoma

Medullary colorectal carcinoma (MeC) is one of the rare pathologic subtypes of colorectal cancer with an incidence of approximately 2.2% to 2.8% (1). The World Health Organization describes MeC as vesicular nuclei with prominent nucleoli and frequent mitotic activity, well-circumscribed neoplasms with solid growth patterns and pushing borders, and prominent lymphocytic infiltration within and around the tumor (2). In histologic features, MeC lacks glandular differentiation with CDX2 deficiency and often confuses with poorly differentiated adenocarcinoma and undifferentiated carcinoma (3, 4). In clinical features, MeC mostly localizes to the right colon in older women. In addition, it tends to have a more favorable clinical outcome than conventional poorly differentiated adenocarcinomas with a lower probability of lymph node metastasis or distant metastasis (5–7).

MeC exhibits distinct pathology characterized by numerous lymphocytes, indicating a robust inflammatory and immune response (8). Genomically, more than 80% of MeC cases show deficient mismatch repair (dMMR), with higher mutation rates in ARID1A and BRAF^V600E (1, 6). Key pathways like IFN-γ and immune checkpoints are enriched, enhancing antigen recognition and recruitment of killer immune cells such as CD8⁺ T cells and NK cells (9, 10). MeC demonstrates higher CD8⁺ T-cell density compared with other colorectal carcinoma histotypes, even with high microsatellite instability (MSI-H)/dMMR (11). Despite its strong immune response, further immune-related studies are needed with high-throughput sequencing.

Immune checkpoint blockade (ICB) prolongs survival and improves quality of life in patients with tumors, but less than 10% are effective in colorectal cancer (12, 13). Currently, only dMMR/MSI-H and POLD1/POLE gene mutations are used to screen immunotherapy-advantaged populations from patients with colorectal cancer (14). Different pathologic types of colorectal cancer exhibit distinct immune microenvironment compositions, which more likely influence the efficacy of ICB therapy. However, whether MeC, as a pathologic subtype of colorectal cancer associated with rich lymphocyte infiltration and a high immune response, can benefit from ICB therapy is still unknown.

To investigate the landscape and immunologic characteristics of MeC, we conducted an integrated genomic and transcriptomic analysis, combined with immune cell staining and T-cell antigen receptor sequencing (TCR-seq) of 47 MeC tissue samples from Harbin Medical University Cancer Hospital (HMUCH). We supplemented this with data from 24 patients with MeC in The Cancer Genome Atlas Colon Adenocarcinoma (TCGA-COAD) database and retrospectively analyzed the ICB therapy process in 47 patients with MeC for clinical validation. This study aims to comprehensively elucidate the immunologic characteristics of MeC and provide new insights into ICB therapy for this subset of patients.

Pathologic tissue sections of 47 patients with MeC who underwent primary resection at HMUCH between January 1, 2012, and December 12, 2015, were collected in this study (Supplementary Fig. S1A). The remaining non-MeC (nMeC) tissues were used in the immune multiomic analyses as controls (Supplementary Tables S1–S6). Moreover, we screened patients with nMeC with available clinical results, particularly for microsatellite instability (MSI) status and the tumor mutation burden (TMB). Propensity score matching was used to reduce treatment selection bias between the MeC and nMeC groups, accounting for variables such as sex, age, tumor–node–metastasis stage, and tumor location (Supplementary Table S2). Public data from the TCGA-COAD database were utilized to validate our conclusions, of which 24 cases were diagnosed as MeC, and detailed interpretation of pathologic subtypes is shown in Supplementary Table S7 (https://portal.gdc.cancer.gov/). An additional 47 patients with MeC who received ICB therapy were enrolled as a clinical cohort (Supplementary Table S8). Each participant in the study provided their informed consent in writing, which was approved by the HMUCH Clinical Research Ethics Committee (No. KY2022-21).

Histologic types of all cases were evaluated descriptively. Assessment of all cases was carried out by two pathologists with experience in gastrointestinal pathology. The definition of colorectal cancer subtype refers to the 2019 World Health Organization classification criteria. Among the objects of our study, MeC is described as sheets of malignant cells with vesicular nuclei, prominent nucleoli, abundant eosinophilic cytoplasm, and obvious infiltration by intraepithelial lymphocytes. Other specific descriptions of the subtypes are shown in Supplementary Table S9.

Formalin-fixed, paraffin-embedded (FFPE) tissue samples diagnosed with medullary carcinoma were analyzed by pathologists for exon sequencing. Beijing Genetron Health Technology Co., Ltd. performed and analyzed the DNA sequencing based on the NovaSeq high-throughput sequencing platform. According to the manufacturer’s instructions, genomic DNA was extracted from FFPE samples and underwent a process of quantification, fragment cutting, and cfDNA library construction. We chose a high-throughput sequencing panel to identify variants in 825 genes and the partial intron region, including major tumor‐related genes (Onco Panscan; Genetron Health; Supplementary Table S10). The test results include point mutation, short fragment insertion-deletion mutation, copy-number variation, and other variants, which can provide the analysis results of the TMB and MSI. After obtaining the raw data by sequencing, a screening process for information analysis was performed based on the HG19 human reference genome. The reference genome was GRCh37/hg19, and the mutation naming principle was based on Human Genome Variation Society (HGVS) rules. In addition, the “maftools” (v2.18.0) package of R software was used for evaluating the mutational landscape and characteristics in HMUCH and TCGA-COAD cohorts, and the function “oncoplot” was used for visualizing them. MSI and TMB status are shown using the “ggplot2” (v3.4.4) package in R.

Archived diagnostic FFPE histologic specimens were cut into 10 consecutive 10-μm sections for RNA extraction. Directly before and after these 10-μm-thick sections, a 4-μm section was cut for hematoxylin and eosin staining. Pathologists annotated tumor regions on these two slides to ensure that all 10-μm slides contained the tumor. RNA was isolated according to the manufacturer’s instructions. Total RNA was extracted from the FFPE tumor samples using the FFPE isolation kit (Qiagen). The concentration of RNA was determined using the Qubit 4 Fluorometer (Thermo Fisher Scientific). Imaging, binding density, positive control linearity, limit of detection, and visual inspection were used to assess the overall performance of nCounter analysis as quality control flags. The 20 housekeeping genes in the IO 360 panel facilitated sample-to-sample normalization. The data were analyzed using nCounter Advanced Analysis module v.2.0 software (NanoString Technologies).

Differentially expressed genes (DEG) in the HMUCH cohort were performed using nCounter Advanced Analysis module V.2.0 software (NanoString Technologies), which were defined by a log₂ fold difference of >1 or <1 and an adjusted P value of <0.05 (Supplementary Table S11). DEG in the TCGA-COAD cohort were performed using the DESeq2 package (v1.32.1; ref. 15). After reading the DEG with the R package “clusterProfiler” (v4.3.2), gene set enrichment analysis (GSEA) was conducted using hallmark gene sets from the MSigDB database (16). A significance threshold of P < 0.05 was applied for GSEA results. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses of the DEG were carried out using the enrichment plot function and the “clusterProfiler” R package, respectively (17, 18). Single-sample GSEA was performed using the “GSVA” (v1.50.0) package of R software with default parameters to assess immune infiltration (19). The output of the function includes the single-sample GSEA enrichment score for each sample. The “ESTIMATE” (v1.0.13) R package was used to calculate the “StromalScore” and “ImmuneScore,” and the “ESTIMATEScore” and “TumorPurity” was generated from the combined “StromalScore” and “ImmuneScore.” ESTIMATE is convenient for describing immune infiltration of tumor tissue through expression profiles (20). The relative infiltration level of immune cells for each sample was estimated using the CIBERSORT algorithm based on the gene expression matrix of 22 immune cells (21). Consensus molecular subtypes (CMS) groups were determined using the Single Sample Predictor (SSP) classifiers published in the CMSclassifier R package (22). Other statistical analysis and visualization, such as the volcano plot and heatmaps, were constructed using “ggplot2” and “ComplexHeatmap” (v2.18.0) packages to display gene expression.

FFPE tissue sections of 4-μm thickness were dewaxed and rehydrated, followed by inactivation of endogenous peroxidase and antigen recovery. IHC stains for ARID1A (1:500, Abcam, cat. No. ab182560, RRID: AB_3096240), CD8 (1:500, Abcam, cat. No. ab101500, RRID: AB_10710024), CD163 (1:500, Abcam, cat. No. ab182422, RRID: AB_2753196), FOXP3 (1:500, Abcam, Cat. #ab20034, RRID: AB_445284), and IL-17A (1:500, Abcam, Cat. #ab79056, RRID: AB_1603584) on all tumor tissues were processed as previously described (23, 24). Antibodies were diluted in PBS buffer. Tissue sections were incubated with the primary antibody overnight at 4°C, followed by incubation with the secondary antibody at room temperature for 30 minutes. Two pathologists performed the biomarker quality control and immune cell density assessment by analyzing 0.52 μm/pixel resolution digital images of the stained tissue sections that were obtained under 20× magnification. Each stained tissue section was randomly selected as the locations of five centers of the tumor regions and five invasive margin regions. An open-source platform for biological image analysis (Fiji/ImageJ; RRID: SCR_002285) was used to estimate the densities of immune cells in the center of the tumor regions or invasive margin regions (cells/μm²).

Multiplexed immunofluorescence (mIF) staining was performed on 4-μm-thick FFPE tissue microarray sections. The slides were deparaffinized and rehydrated with graded ethanol, and the process of antigen retrieval and nonspecific binding blocking is the same as in the previous IHC staining. For the mIF of CD8 (pink), CD163 (red), FOXP3 (light blue), IL-17A (green), pan-cytokeratin (CK; brown), and 4′,6-diamidino-2-phenylindole (DAPI) staining (for nuclear staining) in the tissue microarray, we used a validated multiple fluorescence staining kit from Servicebio, according to the manufacturer’s recommendation. The sample scanning, spectral unmixing, and quantification of signals were conducted with the PANNORAMIC MIDI Pathology Imaging System, using CaseViewer V.2.3 (RRID: SCR_017654).

DNA samples were used for TCR-seq. A NanoDrop spectrophotometer was used to detect the concentration and purity of DNA, agarose gel electrophoresis was used to analyze the degree of DNA degradation and whether there was protein contamination, and Qubit 4.0 accurately quantified the DNA concentration. The remaining amount of gDNA (500-1000 ng) could be used for the construction of multiplex amplicon libraries. Primers were designed for the V and J regions of complementarity‐determining region 3 (CDR3) of the TCR, and the CDR3 region of the TCR was amplified by multiplex PCR technology, and linker sequences were added to construct a library containing target sequences in the CDR3 region. High-depth sequencing was performed on Illumina NovaSeq 6000 high-throughput sequencing platforms. Once the library was constructed, Qubit 4.0 was used for quantification, followed by Qsep100 to detect the insert size of the library, and PE150 sequencing was performed using an Illumina NovaSeq 6000 high-throughput sequencing platform after the insert size met expectations. After the raw data were obtained from sequencing, the information analysis process of VDJ was performed based on the process software MiXCR (https://github.com/milaboratory/mixcr/; RRID: SCR_018725; ref. 25). The R package “immunarch” (v 0.6.5) was used to calculate and visualize the T-cell repertoires, which includes the distribution of CDR3 length, number of clones, number of clonotypes, top clone proportion, and other parameters to evaluate the clonability of immune infiltration (26).

Overall survival (OS) and progression-free survival (PFS) were analyzed using Kaplan–Meier analysis by the R package “survival” (v3.5-7). OS was defined as the time between the first completed ICB treatment and the date of death from any cause. Patients who were lost to follow-up or still alive at the time of the last follow-up without an event (progression or death) were censored at the respective timepoint. PFS was defined as the time between the first completed ICB treatment and death or the occurrence of tumor progression. Patients without progression of any kind or death until the end of follow-up were censored at the end of follow-up. The patient cohort receiving ICB therapy included individuals treated with anti-PD1, anti-PDL1, and anti-CTLA4 therapies, either as monotherapies or in combination with chemotherapy and targeted therapies. Other statistical analyses were performed using R v4.3.1 or GraphPad Prism v9.5.0 (RRID: SCR_002798). An unpaired two-sided Mann–Whitney U-test or unpaired two-tailed student t test was used for analysis when comparison was made between the pathologic groups where appropriate, and the χ² analysis was used to determine the statistical significance of the differences between the groups in relation to categorical variables. Results are expressed as mean ± SD. Significant differences are indicated: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

All human subjects provided informed consent. This study adhered to the principles of the Declaration of Helsinki and was approved by the Clinical Research Ethics Committee of HMUCH (No. KY2022-21).

The datasets supporting the conclusions of this article are included within the article and its supplementary files. The accession numbers for the raw sequencing in this article are Genome Sequence Archive for Human in BIG Data Center (Beijing Institute of Genomics, Chinese Academy of Sciences): HRA008019 https://ngdc.cncb.ac.cn/gsa-human/browse/HRA008019. The publicly available TCGA datasets were directly downloaded from TCGA Data Portal at https://tcga-data.nci.nih.gov/tcga/. Raw data are available from the corresponding author upon request.

We explored the somatic mutation characteristics of 21 MeC samples (Supplementary Table S12). The HMUCH somatic mutation landscape depicted that EGFR, NF1, APC, KRAS, and TP53 were the top five genes with the highest genetic mutations, and the mutation rate was above 65% (Fig. 1A). Three cases had copy-number variations, and all of them showed ASCL2 amplification, the gene associated with low differentiation (Fig. 1B; ref. 27). The TCGA-COAD waterfall plots illustrated similar mutation features in about 20 MeC cases (Supplementary Fig. S1B). In addition, the landscape from HMUCH revealed that 48% of MeC had ARID1A mutations. We also performed IHC staining on 41 MeC cases to determine the expression of ARID1A (Fig. 1C; Supplementary Table S1). ARID1A deficiency was observed in 39.02% (16/41; Fig. 1D) of patients with MeC, which was higher than that in nMeC (4.44%, 54/1216; Supplementary Fig. S1C). We evaluated oncogenic signaling pathways as defined by the TCGA PanCancer Atlas Project (28). Figure 1E and Supplementary Figure S1D systematically describe the high mutation rate in the cell proliferation signaling pathway in MeC. In addition, MeC had a significantly higher incidence of MSI-H/dMMR status, TMB levels, and CMS1 subtype in the HMUCH cohort, with similar features observed in the TCGA-COAD cohort (Fig. 1F and G).

The NanoString nCounter analysis system was used to detect 26 samples. A total of 46 overlapping DEG related to MeC were identified from the HMUCH cohort, including 26 upregulated genes and 20 downregulated genes. GSEA algorithm with the MSigDB Hallmark gene sets (Supplementary Table S13) exhibited the top 15 enrichment results of gene sets ranked by the absolute value of the normalized enrichment scale in Supplementary Fig. S2A. In particular, inflammatory pathways were significantly upregulated in MeC, such as gene sets “IFN-α” and “IFN-γ” (Fig. 2A). Ridge plots presented the expression levels of the key genes included in each pathway. We also analyzed the enrichment of oncogenic pathways between MeC and nMeC. Snapshot of enrichment results showed downregulation in metabolism (glycolysis and cholesterol homeostasis), cell function (epithelial–mesenchymal transition and angiogenesis), and signaling pathways (Wnt/β-catenin and TGF-β pathways; Fig. 2B–D). It was suggested that the function of tumor cell proliferation, metastasis process, and malignant evolution in MeC was poor, which is consistent with the previously proposed better prognosis (28–32). Subsequently, we divided the cohort into four groups according to their mismatch repair (MMR) status: MeC_dMMR, nMeC_dMMR, MeC_proficient mismatch repair (pMMR), and nMeC_pMMR. We analyzed the pathway enrichment difference of these four groups compared with other groups using the same gene sets. The dot plot showed the unity between the MeC_dMMR and MeC_pMMR groups in MeC (Supplementary Fig. S2B).

To further elucidate the functions of these DEG, enrichment analyses, including GO and KEGG pathway analyses (Supplementary Tables S14 and S15), were conducted. The analyses revealed that the DEG were mainly enriched in the activation and regulation of the immune microenvironment, such as T-cell differentiation, antigen processing and presentation, and Th17 cell differentiation (Supplementary Fig. S2C and S2D). As shown in the volcano plot of DEG, genes associated with CD8⁺ T-cell activation were elevated (Supplementary Fig. S3A). To further characterize the immune profile, we thoroughly examined the differences in immune-related gene expression patterns between MeC and nMeC in the HMUCH cohort (33–35). Gene expression levels related to immune processes were generally upregulated in MeC, including genes involved in antigen presentation (CD74) and co-stimulation (CD86, CD28, and CXCR3). Immune checkpoints, such as CTLA4, PDCD1, LAG3, TIGIT, CD274, and HAVCR2, were significantly upregulated in MeC. We supplemented the heatmap to visualize the relative abundance of infiltrating immune cell populations and found that MeC was positively correlated with “activated CD8 T cells” and “type Th1 cells” (Fig. 3A). We estimated tumor purity and immune infiltration using algorithmic analysis, which revealed significantly higher ESTIMATEScores, ImmuneScores, and TumorPurity in MeC, with no significant difference in StromalScore (Fig. 3B). Subsequently, we assessed the level of immune infiltration in four groups: MeC-dMMR, nMeC-dMMR, MeC-pMMR, and nMeC-pMMR. The relative fractions of 22 immune cells in each sample were identified, and six immune cell subtypes showed significant disparity among them, including memory B cells, CD8 T cells, follicular helper T cells, regulatory T cells, resting NK cells, and M1 macrophages (Fig. 3C; Supplementary Table S16). MeC exhibited abundant immune infiltration regardless of MMR status, even surpassing the traditional dMMR-type nMeC. Additionally, immune cell infiltration showed a similar trend in the MeC_pMMR and MeC_dMMR groups. These results were validated in the TCGA-COAD cohort (Fig. 3D; Supplementary Fig. S3B and S3C).

We performed IHC staining of CD8, FOXP3, CD163, and IL-17A to verify the tumor-infiltrating lymphocytes in MeC. Representative images are shown in Fig. 4A and Supplementary Fig. S4A. Counting results demonstrated abundant infiltration of immune cells in MeC, especially CD8⁺, FOXP3+, and CD163⁺ cells (Fig. 4B). In further analysis, it was verified that there was no significant difference between pMMR and dMMR types in MeC (Fig. 4C). mIF was composed randomly on 94 cases to further characterize regional immune infiltrates with multiple immunophenotypic markers, and six samples were diagnosed as MeC (Supplementary Table S3). Representative pictures show merged fluorescent images of CD8 (pink), CD163 (red), FOXP3 (light blue), IL-17A (green), pan-CK (brown), and DAPI staining of DNA (dark blue) in MeC and nMeC cells (Fig. 4D). Analysis also showed abundant infiltration in MeC (Fig. 4E; Supplementary Fig. S4B). To evaluate the infiltration of immune cells within the epithelium, we utilized immunostaining on pan-CK, a specific epithelial cell marker to indicate tumor cells. CK-positive cells are considered to be located within the epithelium (Fig. 4F; Supplementary Fig. S4C). The results showed that the infiltration of CD8⁺ T cells within the epithelium was significantly increased, whereas CD8⁺ T cells in the stroma showed no significant difference (Fig. 4G).

We performed hematoxylin and eosin staining and pathologic diagnosis of the 1,257 colorectal cancer samples from HMUCH. Among them, adenocarcinoma (74.22%), mucinous adenocarcinoma (8.11%), serrated adenocarcinoma (7.16%), MeC (3.26%), low-grade tubular adenocarcinoma (2.23%), and signet ring cell carcinoma (1.03%) are the top six pathologic types in incidence (Fig. 5A and B). MeC showed the highest proportion of dMMR status among all these pathologic subtypes (Fig. 5C). The abundance of various immune cells, such as CD8⁺, CD163⁺, and FOXP3+ cells, was significantly increased in MeC, especially CD8⁺ T cells (Fig. 5D). We also identified the pathologic subtypes of all 460 samples in the TCGA-COAD cohort and diagnosed four subtypes with high incidence of adenocarcinoma (68.51%), mucinous adenocarcinoma (14.48%), serrated adenocarcinoma (8.74%), and MeC (5.52%; Fig. 5E and F). MSI status and immune-infiltration analysis results were consistent with the HMUCH cohort (Fig. 5G and H). We also analyzed the TMB, indel neoantigen, and single-nucleotide variant neoantigen levels and four classic immune reaction markers, namely, antigen presentation, co-stimulation, effect module, and immune checkpoint of various pathologic types and found that MeC showed the highest level under these evaluation indicators (Supplementary Fig. S5A–S5E).

We investigated the TCR repertoire characteristics of both MeC and nMeC tumors and extracted TCR α-chain sequences from bulk DNA sequencing data using MiXCR (https://github.com/milaboratory/mixcr/; ref. 25). The median length of the CDR3 region, which plays a key role in the antigen recognition process (36, 37), was 14 amino acids in both MeC and nMeC, whereas MeC showed slightly longer CDR3 regions when compared with nMeC, as shown in the bar chart (Fig. 6A). We further analyzed the top 10 V–J combinations in these samples, and the top 10 clonotypes constituted approximately 30% of all reads. Representative results are shown in the donut plot (Fig. 6B). MeC samples tended to show higher top 10 proportions. We explored the indicators related to clonability, including the number of clones, number of clonotypes, top clonal proportion, and some evaluation index. MeC tended to show the lower number of clones and number of clonotypes and a higher top clonal proportion (Fig. 6C–F). The results of the Hill number, which rigorously combined the species richness and relative abundances (38), suggested that the diversity of MeC is lower and the clonality is higher (Fig. 6G). The analysis results of the sample divided into four groups based on the MMR and pathologic type are shown in Supplementary Fig. S6. In conclusion, infiltrated T cells in MeC had higher clonality.

We finally included a retrospective clinical cohort of 47 individuals who had received ICB monotherapy or combination therapy (Supplementary Table S8). The results demonstrated that ICB had shown superior therapeutic effects in MeC, and 30 achieved a partial response or complete response, and 10 achieved stable disease after treatment (Fig. 7A). Among these, POLE gene mutational status was collected from 21 of the 47 patients with MeC, with mutations detected in 9.5% (2/21) of cases, similar to the 10% (2/20) mutation rate of patients with MeC in TCGA (Supplementary Fig. S7A and S7B). We further assessed the overall response rate (ORR) and disease control rate (DCR) in the MSI-H and microsatellite-stable (MSS) groups, finding that outcomes were similar between the two groups. Specifically, the MSS group achieved an ORR of 66.7% and a DCR of 83.3%, whereas the MSI-H group reported an ORR of 60.9% and a DCR of 87.0% (Fig. 7B and C). The median PFS was 14.33 months, whereas the median OS was 38.33 months (Fig. 7D and E). Moreover, we did not observe a significant difference in PFS and OS between patients with MeC with MSI or MSS tumors (Supplementary Fig. S7C and S7D). Representative CT images of three patients with pMMR/MSS MeC are shown in Fig. 7F–H. Both liver and lung metastases had shrunk significantly after treatment. These results demonstrate a significant response to ICB therapy in patients with MeC, even those with MSS.

This study is a systematic investigation that employs multiomic approaches to elucidate the immunologic characteristics of medullary carcinoma. The graphic abstract is presented in Fig. 8. First, we presented an atlas of high-frequency gene mutations (TOP25) in MeC. TES revealed a high rate of ARID1A gene mutation and ASCL2 gene amplification, a high proportion of MSI-H/dMMR, and TMB-high in MeC. We identified the protein of ARID1A deletion in MeC with a rate of 39.02%, which was higher than that in nMeC (4.44%). ASCL2 is associated with low tumor differentiation, and copy-number variation occurs in 14% of MeC samples, which can partially explain features of MeC differentiation and the disorder of polarity (27). TMB and MMR status have been found to affect the response to ICB, and CMS1 subtype is considered to be prone to immune infiltration and activation (39). Our cohort had a remarkable higher proportion of TMB-high (TMB ≥100 muts/mb), dMMR/MSI-H status, and CMS1 subtype compared with nMeC. Analysis of oncogenic pathways showed massive alterations in cell proliferation, suggesting the dilated growth pattern of MeC (28).

We also analyzed differences in the gene expression profiling between MeC and nMeC, and oncogenic pathways showed downregulation in glycolysis, angiogenesis, and Wnt/β-catenin pathways. Gene expression profile analysis showed that the antitumor immune response of MeC was highly active. IFN-α and IFN-γ response pathways were enriched in MeC, and the expression of genes, such as CXCL9, CXCL10, GBP4, STAT1, and GZMA, was significantly upregulated (P < 0.05). GO and KEGG enrichment analysis also confirmed the upregulation of T-cell–activated regulation and Th1/Th2 cell differentiation. The expression features of antigen presentation, co-stimulation, effect module, and immune checkpoint genes were also activated in MeC. All these results revealed the key mechanisms for the superior prognosis of medullary carcinoma over poorly differentiated adenocarcinoma (5–7).

The composition and density of immune cells determine the potency of the immune response (40–42). Using the ESTIMATE method to infer the proportion of tumor intermediates and immune cells, we demonstrated that the TumorPurity was lower, but the ImmuneScore, ESTIMATEScore, and StromalScore were higher in the MeC, suggesting that its intercellular substance infiltrated a large number of immune cells (20). CIBERSORT analysis further showed that immune infiltration was more abundant in MeC, especially CD8⁺ T cells. The distribution levels of CD8⁺, CD163⁺, and FOXP3+ cells were evaluated in tissue samples, and the conclusions were consistent with the gene expression profiling. The spatial distribution and proximity among cells are also essential features with important impact on the functional status of the tumor immune microenvironment (43–45). mIF provided an opportunity to assess the spatial relationship of different cellular components. Using CK expression as a criterion to distinguish the location of immune infiltration, we observed significant intraepithelial infiltration of CD8⁺ T cells in MeC, suggesting intense immunogenicity. Specific TCR recognize tumor-specific antigens and promote tumor-killing function (46). The data from TCR-seq showed that some TCR in MeC were greatly amplified, and their clonal expansion degree was significantly higher than that in nMeC. We also paid attention to other pathologic subtypes and evaluated the immune superiority of MeC among them. Cohorts from HMUCH and TCGA both confirmed that the proportion of dMMR status and the density of immune infiltration in MeC ranked first among all pathologic types.

Notably, we found that the activated immune response of MeC was superior to that of dMMR/MSI-H nMeC, even independent of the MSI status. In other words, the immune responses of both MSI and MSS MeC are highly active and similar. In previous descriptions, MeC has a large proportion of dMMR/MSI-H states, and the high level of immune response inevitably leads to suspicion of factors arising from its MMR/MSI status. Specially, we paid attention to this confusion in our analysis and divided the samples into subgroups based on MSI status. The transcriptional, immune infiltration, and T-cell clonality landscapes all suggested that a high immune response in MeC was consistent between the pMMR/MSS and dMMR/MSI-H subgroups. Our findings suggest the presence of a third immune state in colorectal cancer, distinct from both MSI adenocarcinoma and MSS adenocarcinoma, representing an exciting development in our understanding of tumor immunology.

To verify the potential of MeC applying ICB, we collected 47 patients diagnosed with MeC who had received ICB. We subsequently conducted a systematic analysis of their treatment outcomes. Most of the metastases reduced well. The ORR was up to 63.8%. Survival analysis showed a favorable prognosis for MeC with ICB. Previous literature has reported that MeC has a good prognosis, and there have been case reports of patients with medullary carcinoma of the small intestine who have achieved good efficacy after immunotherapy (5, 47). The clinical retrospective cohort indicates that MeC may be a potential population of immune responders (12, 48). Therefore, to clarify the pathologic diagnosis of medullary carcinoma is an important supplement to distinguish the suitability of ICB treatment by relying solely on MSI status.

As a multiomic analysis of pathologic tissue samples, our study has several inherent limitations. First, inadequate tissues for single-cell platform integrative analyses precluded us from deriving data for immune-infiltrating cell subtypes and distinguishing the biases caused by bulk tissue. Second, the limited number of new patients available for prospective clinical cohort studies hindered robust clinical validation, making it challenging to exclude the potential influence of other treatments. In addition, our current study did not fully explain the reason why MeC is extremely immunogenic. Despite these challenges, the comprehensive observation of mutation atlas, gene expression, pathway enrichment, and immune cell infiltration all strongly revealed that MeC may be the promising population for ICB responders, and our study stressed that MeC even could directly suggest a better immune activity independent of MMR/MSI status. The clinical cohort further confirmed the potential of ICB in patients with MeC. As for further exploration of the mechanism of the high immune response, the analysis of antigenic peptides has the potential to explain it from the perspective of the antigen–antigen presentation process (49, 50). We will improve the identification of tumor antigens with immunopeptidomics for different pathologic subtypes of colorectal cancer in the future.

No disclosures were reported.

C. Liu: Conceptualization, writing–original draft. H. Zou: Writing–original draft. Y. Ruan: Visualization, methodology. L. Fang: Methodology. B. Wang: Validation. L. Cui: Methodology. T. Wu: Visualization, methodology. Z. Chen: Methodology. T. Dang: Methodology. Y. Lan: Methodology. W. Zhao: Methodology. C. Zhang: Visualization. H. Meng: Visualization. Y. Zhang: Conceptualization.

Funding for this work was provided by grants from the National Natural Science Foundation of China (Nos. U22A20330, 82373372, 82102858, and 82072985), the Key Project of Research and Development Plan in Heilongjiang Province (Nos. 2022ZX06C01 and JD2023SJ40), the Natural Science Funding of Heilongjiang (No. YQ2022H017), and Haiyan Foundation of Harbin Medical University Cancer Hospital (No. JJJQ 2024-02).