Immunohistochemistry and digital pathology assessment

Tumor specimens were assessed for the presence of CD68⁺ and CD8⁺ T lymphocytes, by immunohistochemistry. Formalin-fixed, paraffin-embedded (FFPE) tissue sections were evaluated using antigen retrieval, primary antibody incubation, and 3,3′-diaminobenzidine staining (EnVision FLEX System). High-resolution images were digitally scanned (Aperio Scanscope XT) and the whole tissues area was segmented using MiaQuant,23 24 to quantify the cell marker-pixel densities in the tumor core (TC), adjacent mucosa (AM), invasive margin (IM) and distinct 400 µm-thick region of interest, internal (intra) and peripheral (peri) from the IM. FC were identified as CD68⁺ elements with an area greater than 100 μm². Immunofluorescence (IF) staining was also used to detect CD68⁺ Bodipy^493/503+ cells, CD8⁺ T infiltrate, PD-1, Foxp3, TGF-β and Ki67 markers. Images were captured by an Olympus BX63 microscope (online supplemental tables S2 and S3).

Cell cultures

In vitro-derived FC were generated by culturing sorted CD14⁺ cells from PBMCs of HDs and patients with CRC with macrophage colony-stimulating factor in absence or presence of oxLDL for 48 hours. T cell proliferation study was performed using CFSE staining prior TCR stimulation with anti-human CD3/CD28 mAb-conjugated beads and then cocultured at ratio 1:1 for 5 days with in vitro generated FC or monocyte-derived early macrophages as control, in presence or absence of anti-TGF-β1,2,3 and anti-PD-L1 mAbs (online supplemental table S3). Lymphocytes cultured alone were used as internal reference.

Flow cytometry analyses

Flow cytometry assays were used to analyze single-cell suspensions derived from surgical specimens (n=21 CRC) and in vitro cell cultures. Surface antigens and intracellular lipid uptake were stained with fluorochrome-conjugated antibodies and reagents (online supplemental tables S3 and S4). Cytokine quantification was performed by Cytometric Bead Array. Samples were analyzed using a CytoFLEX S flow cytometry (Beckman Coulter). Data analyses were conducted using the Kaluza Software (Beckman Coulter).

Transcriptomic analyses

RNA was isolated from FFPE intestinal tissue sections of patients with CRC who underwent surgery and from cultured CD14⁺ cells in the absence or presence of oxLDL for 48 hours. Library preparation was performed using Illumina Stranded Total RNA Prep according to the manufacturer’s instructions. Sequencing and bioinformatics analyses (online supplemental table S5) were performed and annotated in the Gene Expression Omnibus (GEO) Super Series: GSE227206 and GSE273106.

Lipidomic analyses

Total cholesterol, high-density lipoprotein (HDL), low-density lipoprotein (LDL), and triglycerides levels were analyzed using a Cobas Roche Automated Clinical Chemistry Analyzer (Roche Diagnostics), following standard clinical procedures. Plasma esterified fatty acids were analyzed as methyl esters. Quantification was performed using gas chromatograph equipped with a flame ionization detector, using the chromatographic peak area according to the internal standard method.

Macrophages accumulate lipid-droplets in primary CRC tissue

First, we investigated the presence of macrophages with lipid droplets in patients with stage I–III CRC. In tumor lesions, a subset of CD68⁺ cells displayed a specific morphology of large polygonal cells with optically empty vacuoles-rich cytoplasm and centrally placed faint-colored nuclei. The cytoplasmic vacuoles stained strongly positive for neutral lipids, and weakly positive for acidic mucins (figure 1A).

Open in a separate window

Figure 1

Spatial distribution of cancer-associated foam cells (FC) in primary colorectal cancer (CRC). Representative images for H&E, CD68-, oil red O-, and Alcian blue PAS-stained sections in stage I–III CRC; (B) correlation plot of Bodipy^493/503 and cell size in tumor associated CD68⁺ cells and relative confocal images (n=150 cells). (C) Representative images of FC and non-FC distribution across different tumor localization. Boxplots of (D(i)) FC (green) and non-FC (red) distribution across tumor core (TC), invasive margin (IM) and adjacent mucosa (AM), quantified by MiaQuant digital pathology, and (D(ii)) the IM/TC ratio of FC and non-FC. Values are expressed as logarithm of marker densities; (E) correlation between FC tissue accumulation and patient’s body mass index (BMI); (F) boxplot of plasma lipid profile along FC^high and FC^low CRC obtained by K-means clustering methods. P values are determined by two-tailed Mann Whitney U-test.

The accumulation of Bodipy⁺ lipid droplets positively correlate with enhanced size of CD68⁺ cells (p<0.0001), resembling macrophages turned into FC. Contrarily, CD68⁺ macrophages displaying no lipid droplet content, which were reproducibly small in size, were defined as non-FC (figure 1B). Thus, a specific algorithm for digital pathology was developed to quantify FC and non-FC based on the morphology and size of CD68⁺ elements, in the whole FFPE tissue section from the retrospective cohort, discriminating tumor core (TC), invasive margin (IM), and adjacent mucosa (AM) localization (figure 1C). Results from this analysis showed that FC accumulated at IM, displaying a “barrier-like” distribution. In contrast, non-FC were more ubiquitous, involving also intratumoral localizations and AM, and more homogeneously detected in the majority of patients (figure 1D(i)). Specifically, FC clustering was 54.2-fold higher at the IM than TC, while non-FC marker density showed an increase of 6.7-fold at the IM compared with TC (p<0.0001) (figure 1D(ii)).

Although clearly detectable along the tumor IM, the FC infiltrate was highly heterogenous among different patients (IQR: 0.006–0.035), allowing us to define a median cut-off value (0.02), and dichotomize the patients into the FC^high (43%) versus FC^low (57%) groups. While no association was observed between FC abundance and the main clinical and pathological variables, including CEA and CA19.9 levels, pT versus pN stages, and grading (online supplemental figure S1), a significant positive correlation was observed between FC accumulation in tissue and BMI (figure 1E). Conversely, BMI did not show any correlation with non-FC (online supplemental figure S1) and did not display a significant relationship with cancer-associated variables (online supplemental table S6). On this basis, we then investigated the association between FC infiltrate and plasma lipid profiles. FC^high CRC exhibited a clustering with higher triglycerides and lower HDL in plasma compared with FC^low CRC, while cholesterol and LDL did not show statistically significant differences between the two groups (online supplemental figure S2). Moreover, a thorough characterization of lipid species revealed elevated levels of saturated fatty acid, including palmitic acid (C16:0; p=0.04) in FC^high CRC. This group also displayed an enrichment of monounsaturated fatty acid (p<0.0002), specifically of oleic acid (C18:1; p<0.0003). In contrast, polyunsaturated fatty acids (PUFAs) were significantly lower in FC^high tumors (p<0.0076), with no significant differences in omega-6 (n-6) PUFA arachidonic acid (C20:4) while there a remarkable decrease of eicosapentaenoic acid (C20:5; p<0.001), docosapentanoic acid (C22:5; p<0.0006), and docosahexaenoic acid (C22:6; p<0.02), both belonging to omega-3 (n-3) PUFA, in FC^high versus FC^low CRC (figure 1F, online supplemental figure S2).

FC accumulation at the tumor edge prevents CD8⁺ intratumoral infiltration

Next, CD8⁺ T-cells were also quantified by digital pathology and tumors divided into CD8^high (>0.025 median cut-off) and CD8^low (<0.025) with respect to their spatial distribution at IM. FC^high CRC showed a significantly lower density of CD8⁺ T cells, compared with FC^low cases (p=0.04) (figure 2A(i)), and reciprocally, FC accumulation at IM was significantly higher in CD8^low CRC (p=0.01) (figure 2A(ii)).

Open in a separate window

Figure 2

CD8 spatial distribution with respect to foam cells (FC) presence. (A) Violin plot displaying (A(i)) CD8 and (A(ii)) FC accumulation with respect to FC and CD8 quantification, respectively; (B) representative image of the regions of interest segmented in the whole slide colorectal cancer (CRC) tissue section within 1.2 mm intra- and peri-tumoral to the invasive margin (IM); (C(i)) curves representing spatial distribution of CD8 T infiltrate in FC^high and FC^low; (C(ii)) CD8 T infiltrate in the different intracellular localizations was expressed in % with respect of total CD8 at IM, accounted as 100%. Statistic by two-tailed Mann Whitney U-test (and two-way analysis of variance.

The quantification of CD8 gradient from peri-tumoral (peri) to intra-tumoral (intra) localization by applying a 400 µm width intervals (figure 2B) showed that in FC^high tumors, CD8⁺ T cells progressively declined from the lesion edge towards the internal localization, with the marker density decreasing to 0.017 (±0.004; p=0.01), 0.014 (±0.003; p=0.001), and 0.013 (±0.003: p=0.0003) at 0.4, 0.8, and 1.2 mm, respectively. Contrarily, the CD8 quantification was not significantly modulated with regards to the IM in FC^low cancer (figure 2C(i)). Globally, only 35% of the CD8 infiltrate present in the IM could reach the TC (1.2 mm) in FC^high with respect to the 80% of the FC^low setting (figure 2C(ii)). Summarizing, this detailed special analysis showed that the disparity in CD8 intratumor infiltrate detected in our CRC case set is strictly influenced by FC abundance at the IM, indicating a potential role of these cells in restraining the influx of T-cell effectors deep into the TME.

FC-wall assumes a clinical significance in stage I–III CRC

The impact of the FC versus CD8⁺ T cells interplay on disease-free survival (DFS) was evaluated using Kaplan-Meier analyses. First, to elucidate the reciprocal impact of FC and CD8⁺ T cells on DFS, patients with a high versus low FC/CD8 ratio (cut-off 2.6) were compared. Patients with high FC/CD8 ratio exhibited a significant poor DFS (log-rank p=0.002 (online supplemental figure S3A). Subsequently, Kaplan-Meier curves were applied separately to assess this impact in poorly and highly immunogenic tumors. CRC with high CD8⁺ cells in IM (cut-off 0.025) showed a 3-year DFS rate of 63.6% versus 20.8% in CD8^low CRC (log-rank p=0.009) (figure 3A). However, within the CD8^low patients, the presence of the FC^high phenotype (cut-off 0.02) was associated with a significantly lower 3-year DFS rate (8.6%) compared with FC^low cases, which showed a 28.7% 3-year DFS (log-rank p=0.001) (figure 3A). In contrast, no prognostic difference was observed between low and high non-FC density at IM (cut-off 0.107) in CD8^low CRC (log-rank p=0.19 (online supplemental figure S3B)), nor between FC^high and FC^low density in CD8^high patients (log-rank p=0.17 (online supplemental figure S3C)). This likely suggests that in the CD8^high group, the density of FC was not sufficient to counteract the antitumor activity of effector CD8⁺ T-cells, which is associated with improved survival.

Open in a separate window

Figure 3

Foam cells (FC) quantification identifies a subset of poorly immunogenic colorectal cancer with worse disease outcome. (A) Kaplan-Meier curves and number of patients at risk for disease-free survival (DFS) with CD8^high, CD8^low, FC^low CD8^low, and FC^high CD8^low. Log-rank p values from Mantel-Cox test are indicated; (B) stacked graph representing % of patients with recurrent (R) and non-recurrent (NR) disease according to the type of immune infiltrate at the invasive margin. χ² test for statistic; (C) multivariate analysis and forest plot of clinical and immune infiltrate profile associated with DFS. P values are determined by Cox proportional regression model.

Clustering patients by the presence (n=33) or absence (n=32) of recurrence at 3-year follow-up showed a notable enrichment of FC^high CD8^low and general reduction of CD8^high cases within the recurrent population compared with disease-free patients (figure 3B).

Univariate Cox regression analysis indicated that FC^high CD8^low phenotype was one of the strongest negative prognostic factors associated with DFS (HR 2.47; p=0.02), together with CEA (HR 1.14; p=0.002), Ca19.9 (HR 1.57; p=0.004), neutrophil-to-lymphocyte ratio (HR 1.13; p=0.025), extramural vascular invasion (HR 2.55; p=0.01), and perineural invasion (HR 2.12; p=0.038). In contrast, CD8^high phenotype showed a potent protective effect (HR 0.29; p=0.01) (online supplemental table S7), in line with the validated role of the Immunoscore.4 On multivariate Cox analysis, the only independent factors associated with DFS were pT stage (HR 3.5; p=0.04), FC^high CD8^low (HR 2.9; p=0.01), and the CD8^high cells (HR 0.33; p=0.04). Harrell’s C-index was 0.66 (figure 3C).

These findings confirm the clinical relevance of the immunosuppressive features of FC, and these cells significantly impact on disease progression when protective CD8-mediated immunosurveillance is lacking.

Tissue FC associated with distinct intratumoral immune cell contexture

Multiparametric flow cytometry analysis performed on fresh single cell suspension from tumor lesions of our CRC case set (n=21) depicted an overall enrichment of CD14⁺ (4.9±4.3 vs 0.98±1; p=0.0001) and CD4⁺ T cells (39.2±15.1 vs 14.3±7.7; p<0.0001) paired with reduced CD8⁺ (17.5±7.3 vs 28.5±16.8; p=0.01), compared with patient-matched unaffected mucosa (figure 4A(i)).

Open in a separate window

Figure 4

Stage I–III colorectal cancer immunoprofile in relation to the presence of FC^high and FC^low infiltrate. Frequency of (A(i)) CD14⁺, CD8⁺, and CD4⁺ T cells in tumor (T) and distant unaffected mucosa (M), and (A(ii)) representative flow cytometry plots showing physical parameters (forward and side scatter-area, FSC-A and SSC-A) of non-foam cells (FC) and FC CD14⁺ with boxplots relative to their distribution in T and M; (B) histograms and dotplots of the % of positive cells for the indicated markers among FC and non-FC in tumor sample; p values are determined by Wilcoxon signed-rank test (n=21); (C) regression plots between infiltration FC and activated subsets (PD1⁺CD8⁺) or regulatory T cells (Tregs) (CD4⁺CD25^hiCD127⁻) in tumors. P value and r coefficient are determined by Spearman correlation; (D) representative images of CD68-immunostained FC^high and FC^low CRC and confocal microscopy of CD68, Ki67, CD8 (red, green, and blue, respectively) and Foxp3, PD-1, CD8 (pink, yellow, and blue, respectively) matrix panels.

As expected, the whole immune context of the tumor lesion was remarkably different from that of the corresponding normal colon mucosa (online supplemental figure S4), with immunosuppressive and regulatory populations being enriched along with reduced antitumor immune effectors. Indeed, CRC displayed enhanced infiltrate of the monocytic MDSC (CD14⁺ HLA-DR⁻) (17.7±19.1 vs 9±10.9; p=0.004) and reduced frequency of CX3CR1⁺ CD14⁺ cells (39.5±25 vs 53.4±33; p=0.02), which are associated with resident macrophages in steady-state colon.25 Within T-subsets, activated PD-1⁺ (49.7±21 vs 29±22.4; p=0.003) and regulatory HLA-DR⁺ (40.3±21.7 vs 17.7±14.5; p=0.004) CD8⁺ T-cells were expanded in tumors, while the resident memory CX3CR1⁺ CD8⁺ T-cells26 significantly declined (4.9±4.1 vs 8.4±8; p=0.02). CRC specimens were also characterized by increased frequency of regulatory CD4⁺CD25hiCD127⁻ T cells (Tregs) (20.7±8 vs 9±6.1; p=0.0004) and CCR4⁺ effector Treg (21.2±18.3 vs 15±14.8; p=0.004), along with a contraction of Th1 CD4⁺CXCR3⁺ (9.2±10.4 vs 25.4±15.2; p<0.0001) T cell subsets (online supplemental figures S4 and S5). We also detected FC within the tumor-infiltrating CD14⁺ myeloid population based on the larger morphology and the higher structural complexity with respect to the non-FC counterpart. Remarkably, FC were significantly more abundant in tumor specimens compared with matched distant mucosa (49.29±13.22 and 32.67±21.10, respectively; p=0.002), while non-FC were significantly accumulated in mucosa (68.09±20.67) compared with tumor (49.16±13.00; p=0.001) (figure 4A(ii)).

The results obtained from multiparametric flow cytometry revealed that FC were characterized by the expression of distinct metabolic and functional markers associated with the induction of tumor immune tolerance. The modulation of these markers differed significantly compared with their non-FC counterparts (figure 4B). Specifically, Bodipy^493/503 staining, indicating cellular lipid content, was expressed in 70% of FC, compared with only in 11.1% of non-FC (p=0.001). Mirroring their active lipid metabolism, FC exhibited modulated expression of lipid transporter proteins, including upregulation of TREM2 (44.5%; p=0.008) and ABCA1 (38.6%; p=0.008), and downmodulation of CD36 (44.3%; p=0.02), in comparison with their non-FC counterpart (20.6 %, 4.9%, and 58.4%, respectively) (figure 4B). Additionally, FC showed significantly higher expression of the phosphatidylserine (PS) receptor Tim4 (40.9%; p=0.03), the chemokine receptor CX3CR1 (52.5%; p<0.0001) and the inhibitory immune checkpoint PD-L1 (63.5%; p<0,0001) compared with non-FC (19.6%, 24.5%, and 16%, respectively) (figure 4B), defining features of long-lived resident macrophages27 with a potential role in impairing antitumor T cell response and inducing regulatory subsets.28 Along with the increased expression of the myeloid lineage markers CD68 and CD163, FC notably exhibited higher expression of CD206 (80.7%; p=0.06) and PD-L1 (compared with the non-FC, as an additional immunosuppressive pathway to blunt antitumor immunity (online supplemental figure S6). Notably, the abundance of FC was found to negatively correlate with the presence of activated CD8⁺PD-1⁺ effector subsets (p=0.01), while it was positively associated with Tregs (p=0.02) (figure 4C). In contrast, no correlation between non-FC and effector or regulatory T cell subsets was observed (online supplemental figure S7). In summary, the immunological hallmarks of FC-enriched CRC indicate a tolerogenic and pro-tumor environment, characterized by the exclusion of activated CD8⁺ T cells and the predominance of Tregs. This observation was also supported by confocal analyses, revealing lack of the PD-1 activation marker in CD8⁺ T cells, decreased total CD8⁺ T cell infiltration, and increased presence of FoxP3⁺ Tregs in FC^high tumors. Furthermore, CD8⁺ T cells in FC^high CRC did not colocalize with the proliferative marker Ki67 (figure 4D), thus revealing poor proliferative potential.

Transcriptional profile of FC-enriched CRC is compatible with a multimodal dampening of tumor immunosurveillance

TGF- β is an acknowledged player of the immunosuppressive activity mediated by FC in chronic inflammatory setting such as atherosclerosis,29 and a key immune escape mediator in TME.30 Hence, our subsequent objective was to assess whether TGF- β could be linked to the restrained T-cell immunity associated with FC. IM CRC regions with an average area of 0.05 mm² were analyzed by IF. The FC^high area, marked by a substantial accumulation of Bodipy⁺ and CD68⁺ FC, was compared with matched FC^low area, where the average FC accumulation was 80% vs. 32% (figure 5A). The FC^high area exhibited a twofold higher TGF-β level compared with the FC^low area (13% vs 5.7%; p=0.007) (figure 5A). Furthermore, TGF-β was found to be independently associated with FC compared with non-FC elements across different observed areas, reaching statistical significance particularly in FC^high area (p=0.02) (figure 5B).

Open in a separate window

Figure 5

Foam cells (FC) alter T-cell functional program at local levels of colorectal cancer (CRC). (A) Representative images of (A) CD68 (red) and Bodipy^493/503 (green), and (B) transforming growth factor-beta (TGF-β) (white) staining of FC^high and FC^low regions of CRC tissue and relative boxplots of marker expression; DAPI (blue) was used for the nuclei; (C) heatmap of the significant enrichment annotations of coherent DE genes between FC^high versus FC^low CRC of our study and those of micro-dissected FC-enriched CRC from Luca et al study; (D) immune-associated signature score in FC^high versus FC^low CRC. (E) %positive terminal effector cells (TEMRA) in FC^high and FC^low CRC. P values are calculated by Wilcoxon and Mann-Whitney test for dependent and independent observations, respectively; ns, not significant.

Our next step was to functionally characterize FC immune suppressive features by assessing the transcriptional profile of FC^high and FC^low CRC, selected from the digital pathology quantification (online supplemental table S5A).

Analyses from bulk RNAseq data revealed 2900 differentially expressed (DE) genes (FDR<0.05) from FC^high and FC^low CRC tissue specimens (onlinesupplemental figure S8A table S5B). Specific FC-associated transcriptomic programs were identified by crossing the expression levels with cell-state specific genes listed in micro-dissected FC-enriched and deprived tumors.31 881 genes with coherent differential expression were identified (onlinesupplemental figure S8B table S5B). Specifically, 676 genes were upregulated, whereas 205 were downregulated in FC^high CRC in both studies (FDR<0.05). Unsupervised enrichment analyses based on upregulated and downregulated coherently modulated genes (online supplemental table S5C) revealed a scenario resembling a state of immune hyporeactivity in FC^high CRC. Selected terms from the Reactome Gene Set and GO Biological Processes (figure 5C, online supplemental table S5D), showed upregulation of negative type I interferon production and innate immune response as detected by the genes RNF125, HAVCR2, NLRC3, PTPN2 and CD96, as well as, enhanced cell proliferation as depicted by the upregulation of “EGFR signaling pathway” and downregulation of “degradation of β-catenin.”

Additionally, FC^high CRC showed significantly negative enrichment of a core set of genes implicated in class I-II MHC mediated antigen presentation and processing, such as AP2B1, ARF1, DYNC1I2, KIF5B, DYNLL1, SEC24C, BLMH, TPP2). Moreover, the term related to “downstream TCR signaling” was also significantly negative enriched (PSMA4, AP2B1, ARF1, KIF3B, SEC24C, URB4, TUBA1B), indicating that the functions and cross-talk between adaptive and innate immunity are inhibited in FC-enriched CRC.

Alongside the altered antitumor immunity, FC^high CRC exhibited changes in lipid metabolism, indicated by the significant negative enrichment of the term “cholesterol biosynthesis process,” including downregulation of ACLY, DHCR24, HMGCR, IDI1, EBP, NSDHL, and PLPP6 genes. Furthermore, this process was accompanied by mitochondrial dysfunction, as evidenced by the significant downmodulation of a core set of gene involved in mitochondrion organization and biosynthesis (TIMM8A, HSPA4, NDUFA8, VDAC2, COA1, ACACA, ACAT1, and HMGCR).

Next, we examined immune-related gene signatures in FC^high and FC^low CRC. The interferon gamma (IFNγ) activating signature (IFNG_score_21050467),32 comprising genes associated with effective immunosurveillance and favorable prognosis in patients with cancer was found to be reduced in FC^high CRC (figure 5D, online supplemental table S5E). Conversely, there was an increase in signature associated with T-cell exhaustion (inhibitory)32 and Tregs26 in FC^high CRC. Additionally, compared with FC^low tumors, transcriptional pathways indicative of memory CD8 (CX3CR1⁺ CD8 Tmem) and various subsets of activated colonic macrophages (MMP9⁺ inflammatory macrophage) were downregulated in FC^high samples26 (figure 5D, online supplemental table S5E). To validate the findings of the dysfunctional state of antitumor immunosurveillance in the presence of FC, tumor cell suspensions were assessed for abundance of differentiated effector cells. FC^high CRC were characterized by lower proportion of terminally differentiated effector memory CD8⁺ T cells (TEMRA) compared with FC^low CRC (figure 5E). This suggests that weak T cell receptor engagement and limited immune response to tumor antigens may occur within the TME in the presence of lipid-engulfed macrophages.

Overall, these findings suggest that FC-enriched CRC are characterized by immunosuppressive milieu that, through the impaired CD8 T-cell proliferation and activity, accrual of regulatory CD4, and loss of homeostatic function in resident myeloid cells, leads to the complete subversion of local immunosurveillance.

Graphical abstract was created with “BioRender.com”.