Selection criteria for the public scRNA-seq dataset

To compare the immunological function of immune cells at single-cell resolution, we analyzed scRNA-seq data of colorectum-resident immune cells. The GSE125527 dataset from the Gene Expression Omnibus (GEO) and the E-MTAB-8107 dataset from the European Bioinformatics Institute (EMBL-EBI) database were used [21, 22]. To minimize the influence of technical factors, we selected datasets whose scRNA-seq libraries were generated by 10x Genomics, aligned to the human reference genome hg38, and processed to create a UMI count matrix through 10x Genomics Cell Ranger software. The GSE125527 dataset supplies counts on CD45+ immune cells from intestinal biopsies, while the E-MTAB-8107 dataset provides transcripts on CRC neoplasm core tissue, including immune cells and nonimmune cells. Furthermore, seven UC patients from GSE125527 received medical therapy, 4 of whom received 5-aminosalicylates, 1 of which received azathioprine, 1 of which received vedolizumab, and 1 of which received tofacitinib. Conversely, patients with CRC in the E-MTAB-8107 cohort had undergone surgery without prior treatment.

Processing of the scRNA-seq dataset

We clustered cells by performing Louvain’s algorithm and confirmed that GSE125527 was composed of immune cells (S1A and S1B Fig) whereas E-MTAB-8107 consisted of seven immune cell populations numbered 1, 2, 5, 6, 7, 11, and 12 among the thirteen clusters (S1C and S1D Fig). Thus, we selected and included only immune cell clusters. After the data were collected, we used the ‘Seurat’ (version 4.0.4) package in the R program (4.1.1) for quality control, dataset integration, normalization, and clustering [23]. We filtered out the cells with a mitochondrial RNA percentage of more than 10%. The data were normalized using NormalizaData, and highly variable features were recognized through the FindVariableFeatures function. Then, sets of anchors between each subject and integration were completed using FindIntegrationAnchors and IntegrateData. Samples with fewer than 500 cells were removed, and we finally obtained a total of 19 patients, 7 healthy donors, 7 UCs, and 4 CRC patients.

Integration and cell type identification

During the process of integrating the entire dataset, the component of each group with the most number of cells was used as a reference. We then normalized the combined data using SCTransform. The first 50 PCs, which were determined by elbow plots, were utilized for scaling and PCA reduction. Subsequently, we visualized uniform manifold approximation and projection (UMAP) based on the first 50 dimensions of the data reduction method called ‘harmony’ [24]. When setting the optimal number of clusters, the number of clusters containing 100 or more cells for each state was selected, and nine clusters were suitable. Initial clusters were classified as T cells, B cells, plasma cells, and myeloid cells depending on cell type probability calculated by package named ‘SingleR’ and mean expressions of canonical marker genes [21, 22, 25–28]; T-cell (CD3D, CD3E, CD3G, CD7), CD4 T-cell (CD4, TCF7, FOXP3, IL2RA), CD8 T-cell (CD8A, CD8B), NK cells (NCAM1, TBX21), MAIT cell (SLC4A10, TRAV1-2, KLRB1, IL7R, DPP4), mast cell (ENPP3, M+S4A2, TPSAB1, CPA3), B-cell (CD19, MS4A1, CD79A, CD79B), plasma cell (SDC1, SDC1, CD27), macrophage (CD14, FCGR3A) and myeloid cell (CD68, LYZ, AIF1).

Differentially correlated gene and functional enrichment analysis

To identify genes differentially correlated with NCR3+ NK cells compared to NCR3- NK cells in the UC group and with HAVCR2+ NK cells compared to HAVCR2- NK cells in the CRC group, we converted cell-level gene expression into pseudobulk gene expression. We employed averaged pseudobulked data and applied the Pearson correlation coefficient using the ddcorAll function inherited from the DCGA package [29]. The correlated gene pairs with p-values greater than 0.05 were filtered out. The genes correlated with NCR3+ NK cells from the UC group or with HAVCR2+ NK cells from the CRC group were identified based on the criteria of gene pairs that exhibited a positive correlation in NCR3/HAVCR2+ NK cells but not in NCR3/HAVCR2- NK cells.

To explore the functional characteristics of the identified correlated gene sets, we initially conducted a coexpression module analysis using the CEMiTool [30]. Subsequently, we carried out an overrepresentation analysis of the genes in each module utilizing the Gene Annotation & Analysis Resource Metascape [31]. We set a cutoff p-value of 0.05, and the immune-related terms were visualized.

Gene module scores

We evaluated the degree of cytotoxicity and exhaustion by calculating predefined gene expression values for individual cell types. The cell scores were computed using the AddModuleScore function in Seurat with default settings. We used 12 cytotoxicity-associated genes (PRF1, IFNG, GNLY, NKG7, GZMB, GZMA, GZMH, KLRK1, KLRB1, KLRD1, CTSW, and CST7) and 6 well-defined exhaustion-associated markers (LAG3, TIGIT, PDCD1, CTLA4, HAVCR2, and TOX) to calculate cytotoxicity and exhaustion scores [32].

Cell‒cell communication

Cell‒cell interactions were quantitively inferred based on known ligand–receptor pairs in identified cell types using the ‘CellChat’ (version 1.1.3) package [33]. We loaded transform-based normalized gene expression values as input and ran CellChat on each group. We used a ligand‒receptor interaction database for humans and followed the official workflow. We identified overexpressed ligand–receptor pairs and projected gene expression data onto the protein–protein interaction network using identifyOverExpressedGenes, identifyOverExpressedInteractions and projectData. The communication probability between cell types was computed by ComputeCommunProb. Finally, the respective estimated cell‒cell interactions for each group were merged. Communication was visualized through built-in graphic functions and the ‘ggplot2’ package [34].

Gene set enrichment/variation analysis (GSVA)

To explore the pathway enrichment score in NK cells, we conducted GSVA using the R package scGSVA (v0.0.14) [35]. An annotation database was built with biological functions in the Gene Ontology (GO), and the normalized enrichment score (NES) for each cell was calculated. A heatmap was generated from the scGSVA data using the heatmap function.

Human specimen collection

Colon and rectum specimens from UC patients and healthy donors were procured through the Korea Biobank Network. UC tissues were acquired via diagnostic endoscopic biopsies, encompassing 5 patients from the colon and 6 patients from the rectum. Additionally, 10 normal colon tissues and 10 rectal tissues were collected during colonoscopies for health screenings. The demographic details of the subjects are presented in S1 Table.

Ethics approval and consent to participate

This procurement was conducted with the approval of the Ethics Committee at Gachon University Gil Medical Center (GBIRB2022-138). Informed consent was duly obtained from all participants. The data were collected for research purposes after September 29, 2022, and only accessed personal information provided by the Korea Biobank Network.

Immunostaining

Paraffinized tumor tissue blocks were processed into sections and deparaffinized, followed by incubation with 3% hydrogen peroxide to block endogenous peroxidases. The tissue sections were blocked with 1% bovine serum albumin and then incubated with polyclonal antibodies against BAG-6 (Abcam, Cat.No. ab137076), CLEC10A (OriGene, Cat.No. TA810180), KLRD1 (AbboMax, Cat.No. 605–080) and NCR3 (LSBio, Cat.No. LS-C334983-20) overnight at 4°C. The tissues were then stained with goat anti-rabbit IgG H&L (Alexa Fluor® 488 or Alexa Fluor® 594). Confocal images were analyzed using a Carl Zeiss CLSM21 laser scanning microscope and the ZenBlue program (magnification: ×200).

Survival analysis of COAD dataset from The Cancer Genome Atlas (TCGA)

To validate our findings in CRC, we obtained transcriptional profile data from TCGA under the COAD project. Patients were stratified based on HAVCR2 expression levels using the surv_cutpoint function from the survminer R package. Kaplan-Meier survival analysis and log-rank test were conducted to compare the groups.

A scRNA-seq profile of immune cells from healthy controls, UC patients, and CRC patients

We hypothesized that the immune system is remodeled by status in healthy controls, UC patients, and CRC patients. Thus, we extracted immune cells by identifying the populations. In total, 34,209 immune cells were included and integrated; these cells consisted of 12,193 cells from seven healthy donors, 16,909 cells from seven UC patients, and 5,107 cells from four CRC patients. The integrated data exhibited no obvious batch effects, indicating that sample sequencing techniques had less influence on the data (S2A Fig). These immune cells were segregated into nine populations and identified based on the expression of canonical marker genes (S2B and S2C Fig). Our initial rough clustering identified the cells as T cells, B cells, myeloid cells or plasma cells (Fig 1A). In addition, the cells were uniformly distributed across health conditions (Fig 1B). We subsequently investigated the cellular composition fraction (Fig 1C). There were no notable differences in cellular compartments between healthy donors and UC patients. However, compared with those in the other groups, CRC patients in the other groups had greater proportions of myeloid cells and a lower ratio of plasma cells, although there were some variations, to some extent, depending on the individual. Additionally, the lymphocyte-to-myeloid cell ratio, which is a negative indicator of inflammation and poor prognosis, exhibited a steep decrease in CRC patients (Fig 1D). These findings imply that CRC substantially alters the cellular composition of treated UC patients, whereas the cellular composition of treated UC patients is similar to that of healthy individuals.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Overview of the scRNA-seq profiles of immune cells from healthy controls, UC, and CRC patients.

(A) UMAP plot of annotated cell types. Myeloid, T, B, and plasma cells were identified. (B) UMAP visualization split by disease status. (C) Stacked bar plot of immune cell proportions. The proportions for each group are shown in the left panel, and those for each individual are shown in the right panel. (D) Lymphocyte-to-myeloid ratio according to disease status.

https://doi.org/10.1371/journal.pone.0315981.g001

The compositional alteration of immune cell subtypes depending on disease status

To determine the fine spectrum of immune cell heterogeneity and states, we reclustered immune cells and identified distinct populations. T lineage cells were identified as naive T, CD8T, Treg, Tfh, Th1, MAIT, mast and NK cells (Fig 2A and S3A Fig), and the B cells were separated into naive B cells, memory B cells, and plasma cells (Fig 2B and S3B Fig). For myeloid cells, DCs and macrophages were revealed (Fig 2C and S3C Fig). Next, we investigated the compositions of T lineages in healthy donors, UC patients, and CRC patients. Compared with healthy donors and UC patients, CRC patients had the largest proportion of regulatory T cells (Tregs) but the lowest proportion of follicular helper T (Tfh) and type 1 helper T (Th1) cells (Fig 2D–2F). Among the B-cell subtypes, UC patients had immunological profiles comparable to those of healthy donors, whereas CRC patients had a greater proportion of naive B cells (Fig 2G). Among myeloid cells, DCs were more common in healthy individuals and UC patients than in macrophages (Fig 2H and 2I). In contrast, although macrophage fractions are considerably increased in certain CRC patients, they exhibit heterogeneous fluctuations according to individual differences. These results suggest that the CRC environment is primarily remodeled by T cells and exhibits heterogeneous perturbations in B and myeloid cells, whereas the UC environment after treatment demonstrates a compositional predisposition parallel to that of normal conditions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. High-resolution cell compartment analysis.

(A) UMAP and proportion bar plot of subclustered T cells. The T cells in Fig 1A were reclustered and identified into hierarchical subclusters of T cells and NK cells. The subclusters identified included naive T cells, CD8T cells, regulatory T (Treg) cells, follicular helper T (Tfh) cells, type 1 helper T (Th1) cells, Mucosal-associated invariant T (MAIT) cells, and mast and natural killer (NK) cells. The stacked bar plot shows the proportions of cells in the T-cell lineage and NK cells. (B) UMAP and proportion bar plot of subclustered B cells. As B cells and plasma cells are positioned on the same differentiation lineage, B cells and plasma cells were grouped simultaneously. Subclustering identified naive B, memory B, and plasma cells. The stacked bar plot shows the proportions of B-cell subtypes. (C) UMAP and proportion bar plot of myeloid cells. The myeloid cell group shown in Fig 1A was reclustered and identified as macrophages and dendritic cells (DCs). The stacked bar plot shows the proportions of myeloid cell subtypes. (D-I) The proportions of immune cells in healthy, UC and CRC patients. The boxplot represents the ratio of cellular components for (D) Tregs, (E) Tfh cells, (F) Th1 cells, (G) naive B cells, (H) DCs and (I) macrophages under each condition. The horizontal line indicates the mean ratio, and the Wilcoxon test was performed to compare two groups. p ≤ 0.1 (•), p ≤ 0.05 (*), p ≤ 0.01 (**), p ≤ 0.001 (***), p ≤ 0.0001 (****), not significant (ns).

https://doi.org/10.1371/journal.pone.0315981.g002

Comparison of effector cell characteristics between UC patients and CRC patients

While CD8+ T cells and NK cells can kill infected or abnormal cells directly and defend against pathogens, long-lasting secretion of cytotoxic agents can impair tissue cells and lead to chronic inflammation [36]. Therefore, we next explored the attributes of cytotoxic cells, CD8+ T cells and NK cells. We calculated cytotoxicity and exhaustion scores for 12 cytotoxicity-associated genes (PRF1, IFNG, GNLY, NKG7, GZMB, GZMA, GZMH, KLRK1, KLRB1, KLRD1, CTSW, and CST7) and 6 exhaustion-associated genes (LAG3, TIGIT, PDCD1, CTLA4, HAVCR2, and TOX) (Fig 3A). There was a notable alteration in effector function in both UC and CRC patients. In the UC group, NK cells exhibited lower cytotoxicity than the healthy group, although the CD8+ T cells showed no changes in cytotoxicity or exhaustion. In the CRC group, all cytotoxic cells exhibited elevated toxicity, but they also demonstrated a greater level of exhaustion (Fig 3B). CRC patients exhibited increased expression of most cytotoxic genes in both CD8+ T and NK cells (Fig 3C). The CRC group presented heightened levels of granzymes, including GZMB and GZMA, in CD8+ T cells and KLRB1 and KLRD1 in NK cells. Regarding exhaustion, all the CRC patients exhibited increased expression of many inhibitory receptors on CD8+ T cells, especially PDCD1 and HAVCR2 on NK cells (Fig 3D). To consider the potential for functional changes to be induced by signals exchanged with other cells, we subsequently examined the strength of the cell‒cell interactions between NK and CD8+ T cells (Fig 3E). CD8+ T cells and NK cells were distinctly segregated by the intensity of the received signal rather than the sending signal. CD8+ T cells displayed a sequentially lower intensity of received signals in UC and CRC patients. In contrast, the number of NK cells increased in UC and CRC patients. This observation could imply that NK cells contribute to the reconstruction of the immune microenvironment in both UC and CRC through cytotoxicity and communication with other immune cells.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3.

The altered traits of effector cells, CD8+ T cells, and NK cells. (A) Box plots of the cytotoxicity scores of NK and CD8+ T cells. The x-axis indicates three groups, healthy, UC, and CRC, and the horizontal line in the box shows the average score. To explore the significance of the differences, the Wilcoxon test was performed; p ≤ 0.05 (*), p ≤ 0.01 (**), p ≤ 0.001 (***), p ≤ 0.0001 (****), and not significant (ns). (B) Box plots of exhaustion scores in NK and CD8+ T cells. Like in Fig 3A, the scores were compared. (C) Heatmap of cytotoxicity-associated genes across participants. The genes used for cytotoxicity scoring are listed, and scaled expression levels of the genes are shown. (D) Heatmap of exhaustion-associated genes spanning participants. The genes used for exhaustion scoring are listed, and scaled expression levels of the genes are shown. (E) Inferred interaction strength of CD8+ T cells and NK cells as a sender and a receiver. The X-axis indicates the outgoing interaction strength of CD8+ T cells and NK cells, which served as signal senders for each condition. The Y-axis indicates the incoming interaction strength of CD8+ T cells and NK cells, which serve as signal receivers for each condition.

https://doi.org/10.1371/journal.pone.0315981.g003

UC-specific immune cell interaction network

To inspect the UC- and CRC-intensified pathways, we investigated ligand and receptor interactions using CellChat. A BAG signal was exclusively detected in UC patients (S4 Fig). Interestingly, the BAG signaling network was composed of the ligand BAG6 and the receptor NCR3 (Fig 4A). The signal was sent from the DCs and transferred to the NK cells. In addition, the expression level of BAG6 in DCs was slightly increased in UC patients (Fig 4B). To examine whether the BAG signal is specifically related to UC status, we conducted immunostaining on tissues obtained from healthy donors and UC patients (Fig 4C). Notably, both BAG6 in DCs and NCR3 in NK cells were strongly detected in UC tissues. In contrast, they were rarely found in normal tissues. The level of NCR3 detected in NK cells from UC patients was approximately 7 times greater than that in NK cells from healthy individuals, consistent with the inferred signaling network. To determine the immunological attributes of NCR3+ NK cells in comparison to NCR3- NK cells, we identified a coexpression module using genes that are differentially correlated with NCR3+ NK cells from CRC patients, and we investigated the biological roles of these genes (Fig 4D). The NCR3+ NK cells from UC patients exhibited upregulated expression of genes associated with lymphocyte activation, the TNF signaling pathway, and NK cell cytotoxicity. Additionally, only under UC conditions did NCR3+ NK cells express a greater amount of IFNG, which is known to promote the accumulation and activation of NK cells, than NCR3- NK cells (Fig 4E) [37]. GSVA revealed that NK cell activation and cytokine production exhibited distinctly high activity depending on the presence of NCR3 under UC conditions (Fig 4F). We thus inspected the cytotoxicity of NK cells based on NCR3 expression by further classifying them into cytokine-mediated and granule-mediated cytotoxic effects (Fig 4G) [38]. Consequently, in the normal state, NK cells exhibit no difference in effector function depending on NCR3 expression. However, in UC patients, we observed significant differences in the cytotoxicity of NK cells, both in cytokine-mediated and granule-mediated cytotoxicity; the cytotoxicity was greater in the NCR3+ NK group than in the NCR3- NK group. Next, we investigated the expression level of NCR3 in two GEO datasets, GSE92415 and GSE107593 (Fig 4H and 4I). GSE92415 includes samples from 21 healthy individuals and 87 UC patients, while GSE107593 contains 24 non-inflamed and 24 inflamed tissues obtained from UC patients. Compared to the healthy group and non-inflamed colon tissues, both the UC group and inflamed colon tissues exhibited higher levels of NCR3. Collectively, This finding suggested that the NCR3 receptor of NK cells, which receives signals from DCs through BAG signaling, mediates the secretion of cytotoxic molecules and contributes to the persistence of colitis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. UC-specific cell‒cell communication network among tissue-resident immune cells.

(A) Circle plot of cell‒cell interactions in the BAG pathway in UC patients. For each immune cell type, the color of a dot is assigned, and the color of the arrow is the same as the color of the cell type that sends the signal. (B) Gene expression of the ligand BAG6 and the receptor NCR3. The violin plots for BAG6 are normalized transcripts in DCs (left), and those for NCR3 are normalized in NK cells (right). The crossbar of the plot indicates the average expression level of each transcript. (C) Comparison of BAG6 and NCR3 expression in normal and UC patient tissues through immunostaining. BAG6 expression was explored in DC (CLEC10A) cells, and NCR3 was explored in NK (KLRD1) cells. To explore the significance of the difference, the Wilcoxon test was performed; p ≤ 0.05 (*), p ≤ 0.01 (**), p ≤ 0.001 (***), p ≤ 0.0001 (****), and not significant (ns). (D) Functional analysis of differentially coexpressed genes in NCR3+ NK cells compared to NCR3- NK cells in the UC group. Immune-related pathways associated with P<0.05 are shown. (E) Violin plot of IFNG expression in healthy donors and UC patients separated by NCR3 expression in NK cells. The statistical significance was determined using the Wilcoxon test; p ≤ 0.05 (*), p ≤ 0.01 (**), p ≤ 0.001 (***), p ≤ 0.0001 (****), and not significant (ns). (F) GSVA of NCR3+ NK cells from UC patients. Heatmaps depicting NK cell-related pathway activities according to the presence of NCR3 expression in the healthy and UC groups. The color bar indicates the NES, with higher activity represented in orange and lower activity in green. (G) Boxplot of cytotoxicity in NK cells separated according to the expression of NCR3. The groups were separated by the presence (+) or absence (-) of NCR3 expression in healthy controls and UC patients, and the color is presented according to the disease condition. To calculate the score, we utilized the expression levels of cytokine-encoding genes (IFNG, TNF, CSF2, IL10, and IL15) and granule-encoding genes (PRF1, GZMB, GZMA, GZMH, GZMM, GZMK, NKG7, and GNLY). In each panel of the boxplot, the overall cytotoxicity of NK cells was measured using total cytokine- and granule-encoding genes (left), cytokine-mediated toxicity (middle), and granule-mediated toxicity (right). The horizontal line in the box indicates the average score, and Wilcoxon signed-rank test was performed; p ≤ 0.05 (*), p ≤ 0.01 (**), p ≤ 0.001 (***), p ≤ 0.0001 (****), not significant (ns). (H-I) Expression level of NCR3 in GEO datasets, GSE92415 (H) and GSE107593 (I).

https://doi.org/10.1371/journal.pone.0315981.g004

The GALETIN pathway in CRC patients

We next detected an increase in the GALECTIN signaling pathway within the CRC environment. Among the GALECTIN family members, LGALS9 (GALECTIN-9) plays a primary role in this signaling pathway (Fig 5A). CD45 and CD44 were observed as receptors, and DCs sent signals to overall immune cells. However, the LGALS9-HAVCR2 pair was also newly discovered in CRC macrophages and was an additional signal sender. In particular, only NK cells communicated with myeloid cells via LGALS9-HAVCR2 signaling under CRC conditions (Fig 5B and S2 Table). LGALS9 expression was elevated in DCs and macrophages from the CRC group, and the level of HAVCR2 in NK cells was increased (Fig 5C). From the coexpression module analysis, two modules were extracted from the HAVCR2+ NK cells of CRC patients, and their biological functions were inspected (Fig 5D and S5A Fig). The functions associated with the regulation of immune system processes were primarily recognized. Among the enriched terms, we detected the PD-1 checkpoint pathway, NK cell-mediated cytotoxicity, and induction of tolerance. The results of GSVA were consistent with the results of the coexpression module analysis; in particular, HAVCR2+ NK cells from CRC patients exhibited strong inactivity of NK cell-mediated cytotoxicity (Fig 5E, and S5B Fig). We also examined the molecular cytotoxicity score of NK cells from healthy individuals and CRC patients stratified by HAVCR2 expression. Although there was no difference in the granular cytotoxicity score, the cytotoxicity score mediated by cytokine secretion was decreased in the HAVCR2+ NK cells of CRC patients. Subsequently, we investigated the association between the expression level of HAVCR2 and cytokine production in the GSE50760 dataset contating 18 healthy individuals and 18 CRC patients. In the healthy control group, an increase in HAVCR2 expression levels was linearly associated with higher cytokine production, whereas no significant correlation was observed in the CRC group (Fig 5G). In addition, high levels of HAVCR2 demonstrated worse clinical outcomes in the TCGA-COAD cohort (Fig 5H). Finally, we explored the association of NCR3 and HAVCR2 expression with disease status (Fig 5I). Interestingly, the NCR3+ UC group exhibited the second-highest expression level of HAVCR2, following the HAVCR2+ CRC group. In addition, the HAVCR2+ CRC group showed the second-highest level of NCR3, following the NCR3+ UC group. Considering these findings, we suggest that the LGALS9/HAVCR2 crosstalk found only in CRC interferes with the killing function of NK cells through cytokine secretion rather than via granule formation and suggest that HAVCR2 is an important immunological target for improving cytokine levels in CRC. Furthermore, we observed a potential correlation between NCR3-expressing NK cells in UC patients and HAVCR2-expressing NK cells in CRC patients, indicating a possible shared mechanism or pathway in these distinct conditions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. CRC-specific intercellular signaling pathways.

(A) The GALECTIN pathway according to disease condition. Each immune cell type is assigned the color of a circle, and the color of the arrow is the same as the color of the cell type sending the signal. (B) Circle plot of the interaction between the LGALS9-HAVCR2 pair and the GALECTIN pathway in healthy donors, UC patients, and CRC patients. (C) Gene expression levels of a ligand and a receptor in the GALECTIN pathway. The violin plot shows the expression levels of the ligand LGALS9 in DCs and macrophages and the receptor HAVCR2 in NK cells. The crossbar indicates the average expression level of transcripts in each status. (D) Immunological functions of differentially coexpressed genes in HAVCR2+ NK cells compared to HAVCR2- NK cells in the CRC group. To determine the significance of the differences, the Wilcoxon test was performed; p ≤ 0.05 (*), p ≤ 0.01 (**), p ≤ 0.001 (***), p ≤ 0.0001 (****), and not significant (ns). (E) Violin plot of NES for negative regulation of natural killer cell-mediated cytotoxicity. The groups were separated based on disease condition and the expression of HAVCR2. (F) Boxplot of cytotoxicity in NK cells separated according to HAVCR2 expression. As shown in Fig 4G, groups were separated by the presence (+) or absence (-) of HAVCR2 expression in healthy individuals and CRC patients. (G) Correlation between HAVCR2 and cytokine expression levels. The correlation coefficients and p-values are displayed in the upper left. (H) Kaplan-Meier survival plot for 3 years. The 371 CRC patients from TCGA-COAD are classified based on the expression levels of HAVCR2, and survival plot is described in days. (I) The expression level of NCR3 and HAVCR2 in NK cells of UC and CRC patients. The groups are categorized based on the expression levels of NCR3 and HAVCR2 with disease.

https://doi.org/10.1371/journal.pone.0315981.g005