3.1. gPTGS2 Quantification in 100 CRC Lysates and its Relation to Tissue PTGS2
gPTGS2 was detectable by WB in 96/100 CRC (Figure 1
a,c) (median = 156.86 pg, mean = 293.3 pg, range 0.00–1515.64 pg of protein, in 30 µg of tissue lysate, according to the hu PTGS2 standard) and in 11/100 of matched normal mucosa (median = 0.00 pg, mean = 0.003 pg, range 0.00–79.8 pg of protein, in 30 µg of tissue lysate). Compared to other studies (see Supplementary Table S1
), this is a high detection rate. The replicate of WB analysis on 60 CRC (Figure 1
b) showed a high correlation (Pearson’s correlation r = 0.907, p = 0.0000000000000000000000217, ensuring sufficient reproducibility.
PTGS2 was also evaluated by IHC in 100 matched CRC paraffin embedded tissues, using the same primary antibody. Tumor-associated and stroma-associated PTGS2 were scored independently. The correlation coefficient of tumor PTGS2 compared with stromal PTGS2 was 0.334 (Spearman’s rank, p
< 0.001). Thus, the contemporary presence of high or low PTGS2 levels in the tumor and stromal populations of the same sample was apparently infrequent in our cohort, suggesting the existence of distinct mechanisms of PTGS2 induction in the different cell populations of the same tumor. In tissue lysates, both tumor and stromal cells contributed to total gPTGS2 levels, showing a directly proportional correspondence with IHC data (Figure 2
PTGS2-positive cells of the stromal component almost invariably localized in the luminal area of the tumor, with a strong intensity of staining. These cells frequently lined the limit between living tissue and necrotic areas or surrounded crypts of the outer epithelial border (Figure 2
b), suggesting a protective function. In CRCs with medium–high PTGS2 epithelial staining, an irregular distribution of positive areas was observed (Figure 2
3.2. Identification of gPTGS2 Positive Cells in the Stromal Component
As in our CRC cohort, PTGS2-positive stromal populations with a luminal distribution were previously observed in colon adenomas: Chapple and Bamba independently attributed PTGS2 positivity to macrophages, according to cell morphology or CD68 expression [16
]. Tumor-infiltrating macrophages have been classified as M1 (antitumor) or M2 (protumor) according to the co-expression of CD68, iNOS or MRC1/CD206, CD163, Arg1, and other markers in in vitro models. In human pathology, this subdivision is an oversimplification, and these markers can be expressed or downregulated in macrophages with high plasticity, according to different microenvironmental stimuli [18
]. In Apc (Min/+) mice, the inhibition of PTGS2 reduces the M2 component [19
]; thus, the expression of PTGS2 in CRC macrophages could be associated to the induction of a prevalent M2 phenotype. On the other hand, PGE2 is able to induce M1 differentiation in other mouse models [20
], suggesting a positive influence of PTGS2 on the M1 component. We first tested the correspondence of PTGS2 expression with CD68 and CD163 on serial sections, assuming that CD68 positivity would indicate the total macrophage population, while CD163 would indicate the M2 component. The comparison of positive areas indicated a possible coexistence of PTGS2 and CD68 staining in some samples, while the correspondence of PTGS2 and CD163 was less evident (Figure 3
The quantification of cells, expressing these antigens in overlapping areas of equal extension, corroborated this observation (Figure 3
b). The Pearson correlation coefficient was 0.422, (95% CI 0.229–0.582, p = 0.0000586) for CD68/PTGS2 and 0.316 (95% CI 0.110–0.496, p = 0.00324) for CD163/PTGS2.
To obtain a more specific quantification of the involvement of M1 and M2 macrophages in PTGS2 production in CRC, we also tested a multiplex IHC approach. Using consecutive destaining, stripping, and reprobing of the same tissue slices, we tested the CD68–iNOS–PTGS2 and the Arg1–MRC1–CD163–PTGS2 series (Figure 4
Multiplex analysis showed a complex picture with high variability among samples and different fields of the same sample. iNOS was strongly expressed by several non-macrophage cells, being the majority of iNOS positive signals in CRC tissues (Figure 4
a). Few Arg1dim
-positive cells were detected, with minimal overlap with other M2 markers (Figure 4
b and Supplementary Figure S1
). A strong overlay was observed only between MRC1 and CD163 (Figure 4
b), which was confirmed by co-localization analysis (Supplementary Figure S2
). PTGS2 showed a limited co-localization with macrophages: the mean Pearson’s coefficient for CD68-PTGS2 was 0.063, while the mean Manders’ overlap coefficient, evaluating the extent of PTGS2 positivity in the CD68-positive area, was 0.237 (Figure 4
c and Supplementary Figure S2
). We also attempted a simplified representation of the co-localization of M1 and M2 single markers, assuming their expression only in macrophages (Figure 4
c, right). This analysis suggested a mixed contribution of both M1 and M2 polarized cells in PTGS2 expression, with a possible prevalence of iNOS+ cells.
Thus, the expression of PTGS2 at the luminal surface of CRC, in non-tumor cells, did not apparently associate to a M1->M2 switch of macrophages. Moreover, macrophages did not appear as the main PTGS2-positive cell population in stromal areas.
Adegboyega et al. described subepithelial myofibroblasts as another source of PTGS2 in colorectal adenomas, mainly in luminal mucosal areas with damaged surface [21
]. Sonoshita et al. observed the same localization of PTGS2 in the ApcΔ716
mouse model and in human adenomas, finding a strong co-localization of PTGS2 and vimentin, the marker of mesenchymal cells [22
]. To verify the involvement of mesenchymal cells, we performed fluorescent double staining to co-localize PTGS2 and vimentin in multiple fields of representative CRC samples (Figure 5
Indeed, almost all PTGS2-positive cells corresponded to vimentin-positive cells, although stromal cells with the bright PTGS2 signal usually showed a lower signal for vimentin. Eliminating dim vimentin-positive signals by a color threshold, co-localization analysis showed a mean Pearson co-localization coefficient of 0.472 (Figure 5
b; range 0.367–0.600, see Supplementary Figure S3a
). Manders’ overlap coefficients, evaluating the overlap of PTGS2 with vimentin and vice versa, showed that a high proportion of vimentin-positive PTGS2-negative mesenchymal cells was also present (Supplementary Figure S3b
). The mean of Manders’ overlap coefficients evaluating the extent of PTGS2 signal in bright vimentin-positive areas was 0.301 (Figure 5
c). The possible involvement of cancer-associated fibroblasts (CAF) as a prevalent non-tumor source of PTGS2 in CRC led us to analyze in vitro the possible inducers of PTGS2 in primary colorectal CAF.
3.3. In Vitro, IL1β is a Powerful Inducers of PTGS2 in CAF and Correlates to gPTGS2 Levels in Tissue Lysates
CAF primary cells [15
] were stimulated for 24 h with IL1β, a known inducer of PTGS2 expression by NFkB and AP1 activation [23
]. Other factors linked to inflammatory/trophic CRC progression (IL8/CXCL8, GROβ/CXCL2, as agonists of G-protein coupled receptors; PGE2 produced by PTGS2 activation; EGF as the prototype EGFR tyrosine-kinase agonist) were also tested as potential PTGS2 inducers [1
]. The analysis for gPTGS2 expression by Western blot showed a powerful induction of gPTGS2 by IL1β, and weak responses elicited by PGE2 and EGF (Figure 6
IL1β involvement in the induction of gPTGS2 in our cohort was verified by the quantification of its levels in 60 tissue lysates (Figure 6
b). Pearson correlation coefficient of gPTGS2 versus IL1β levels was 0.593 (p = 0.00000609), indicating a strong link between IL1β and gPGTS2 levels in more than half of the samples. Western blot analysis of CRC cell lines stimulated by IL1β revealed a possible explanation for the existence of samples with a low gPTGS2/IL1β correlation. CRC cell lines showed high or low basal levels of gPTGS2 with a limited response to IL1β stimulation compared to CAF (Figure 6
c), even at 10× increased doses of IL1β (Supplementary Figure S4
). Thus, the prevalence of stroma, or tumor-derived gPTGS2, could mirror the strong or weak response to the presence of IL1β in CRC tissues.
3.4. Effects of Stromal PTGS2 on Patients Prognosis
In our cohort, stromal PTGS2 affected patients OS according to low, medium, or high IHC positivity (Figure 7
a). In particular, intermediate levels of stromal PTGS2 (n = 22) were apparently associated to a better prognosis, while high PTGS2 (n = 6) showed a negative outcome compared to low/null PTGS2-expressing CRC.
The exclusion of stage IV tumors from Kaplan–Meier analysis did not modify the prognostic potential of stromal PTGS2 expression (Supplementary Figure S5
), reducing a potential bias on OS evaluation linked to the surgical or therapeutic treatment of this group of patients. Neither gPTGS2 total levels quantified in tissue lysates nor tumor-associated PTGS2 scored by IHC influenced patients OS in our cohort (Figure 7
Our observations suggested a possible subgrouping of CRC on the basis of stromal, or tumor-derived PTGS2, the former modulated by IL1β, with a prognostic significance, the latter relatively independent. We asked if we could identify these two categories of tumors on the basis of gPTGS2 and IL1β levels in tissue lysates and if they could influence patients prognosis. When dichotomized by the median value, PTGS2 showed no relation with the main pathological data (Table 1
), but it maintained a significant relation with IL1β (the database with pathological data, PTGS2, and IL1β quantification is available in Supplementary data S1
We used gPTGS2 and IL1β dichotomization by the median of data and defined as stroma-derived gPTGS2 the cases with low + low or high + high gPTGS2 and IL1β levels (= directly proportional IL1β and gPTGS2 levels). On the contrary, tumor-derived gPTGS2 was defined in cases with low + high or high + low gPTGS2 and IL1β levels (= independence of IL1β and gPTGS2 expression). Dichotomized gPTGS2 or IL1β levels, individually assessed by Kaplan–Meier analysis, did not influence patients’ OS, showing almost superimposable curves (Figure 7
d,e). On the contrary, an immediate divergence of curves was observed when CRC with putative stroma-derived PTGS2 were compared to CRC with putative tumor-derived PTGS2 (Figure 7
f). This categorization suggested a different involvement of stroma-expressed versus tumor-expressed PTGS2 in patient outcomes, although it did not reach statistical significance.