185 stage I-IV patients, who underwent surgery due to colorectal cancer between 2009–2017, were selected retrospectively and their slides and formaline fixed paraffin embedded (FFPE) blocks were retrieved from the archives of the Department of Pathology and Experimental Cancer Research, Semmelweis University, Budapest, Hungary. Patients who received neoadjuvant treatment, died within 30 days of surgery, had other synchronous or metachronous primary colorectal cancer, or another malignancy requiring systemic treatment in their history, were excluded from the study. Patient history and laboratory results were collected using the medical database of Semmelweis University (MedSolution, T-Systems, Budapest, Hungary). Preoperative serum CA19-9 and CEA levels were measured routinely using Abbott Architect CA 19-9 XR immunoassay (Chicago, IL, United States of America) and Abbott Architect CEA immunoassay (Chicago, IL, United States of America). Pathological characterization of surgically resected primary tumors was performed according to the UICC TNM classification, 8^th edition.

TSR was evaluated on HE-stained slides in accordance with recommendations [8]: stromal content was estimated visually per 10% increments at ×10 magnification field of the invasive front of the tumor (containing the stromal hotspot). If stromal content exceeded 50% of the examined area, patients were classified as TSR-high, if not, they were categorized as TSR-low (Figure 1).

KM grade was estimated according to the criteria established by Klintrup et al [6]. Briefly, inflammatory reaction was graded as KM-low when there was no or only mild increase in inflammatory cells at the invasive front, and a KM-high score was given when there was a band- or cup-like infiltrate of inflammatory cells at the margin with destruction of cancer cells (Figure 1). The GMS is a combination of KM score and TSR. Briefly, when there was extensive inflammatory reaction (KM-high), a score of GMS0 was given. In case of stroma-low and KM-low tumors, a score of GMS1, and in case of stroma-high and KM-low tumors a score of GMS2 was given (Supplementary Table S1), as described previously [12, 13]. All parameters were graded by two independent observers (AJ and TSM) blinded to clinicopathological data and disease outcome.

All blood samples were obtained from patients within 30 days before surgery. Cutoff values and scoring of SIR markers are included in Supplementary Table S1.

Tissue microarray (TMA) blocks containing 6 × 9 cores (core diameter: 2 mm) selected from surgically derived FFPE blocks of 167 patients were created using TMA Master1000 (3DHistech, Budapest, Hungary). At least two representative cores were selected per case. Non-neoplastic kidney samples were used as normal tissues and stain controls in each block.

Immunohistochemistry was performed on 4 um thick sections of TMAs. For mismatch repair status (MMR) assessment, anti-MLH1, anti-PMS2, anti-MSH2, and anti-MSH6 primary stains were used and graded as recommended [31]. CMS classification was carried out on proficient MMR (pMMR) samples using anti-cytokeratin (CK), anti-CDX2, anti-FRMD6, and anti-ZEB1 immunhistochemistry as described by Ten Hoorn et al [32]. For further details of immunohistochemistry, please see Supplementary Table S2. Cytoplasmic CK and FRMD6, nuclear CDX2, as well as membrane and cytoplasmic HTR2B were graded as low, moderate or high. In case of ZEB1, the presence of nuclear staining was scored as either present or absent. All CMS stains were graded both in accordance with Ten Hoorn et al’s recommendations, and, apart from ZEB1, using H-score as well [33].

All immunohistochemical reactions were assessed by two observers (AJ and TSM).

Cases with deficient MMR-status (dMMR) were classified as CMS1. For proficient MMR tumors, CMS classification was performed using an online, TMA-based and validated, robust and reliable random forest classifier ¹ as described previously [32, 33]. Briefly, four immunostains (ZEB1, HTRB2, FRMD6 and CDX2) were selected based on distinct gene expression profile differences in CMS2/3 and CMS4 tumors. In pMMR tumors, the stain intensity and content of these four stains in tumor epithelium correlates with their CMSs [32]. Typically, low FRMD6 and HTR2B staining intensities, lack of nuclear ZEB1 expression and strong CDX2 stain correlate with epithelial subtypes (CMS2/3); while strong positive FRMD6 and HTR2B, loss of CDX2 and positive nuclear ZEB1 reaction is expected in mesenchymal CRCs (CMS4) (Figure 2) [33]. In case the probability of a CMS was estimated higher than 0.6, we automatically labeled the case in concordance with the software. Where the probability of estimated CMS was between 0.5 and. 0.6, the case was automatically excluded from our analysis. In total, 12 cases were excluded due to uncertain subtyping.

To reflect the biological behaviour of certain cancer subtypes classified by TSR and CA19-9, a novel scoring system was established by combining CA19-9 and TSR into stroma-tumor marker (STM) score. In case of TSR-low and CA 19-9 low cases, STM 0 score was given. If either CA19-9 or TSR was classified as high, but the other marker as low, an STM 1 score was given. When both markers were classified as “high,” STM 2 score was given (Supplementary Table S1).

Shapiro-Wilks test of normality was performed on the continuous variables. In order to determine the relationship between clinicopathological features and categorical variables, Chi-squared test was performed. Mann-Whitney U-test or Kruskal-Wallis H-test was performed to examine the correlation between clinicopathological features and continuous variables. Spearman’s rank order correlation coefficent was calculated to investigate the correlation between certain systemic inflammation-related and tumor markers. The relationship between TME, systemic and tumor markers and survival was carried out using Kaplan-Meier log-rank survival analysis. To assess how certain variables affect the overall survival (OS), uni- and multivariate Cox regression analysis was performed. All variables that reached p < 0.1 in the univariate analysis were included for the multivariate analysis. The complete statistical analysis was performed using SPSS version 28.0.1.0 (IBM, Armonk, NY, United States).

Altogether, 185 patients were included in our cohort, 155 patients had available CEA, and 135 patients had available CA19-9 results. CMS classification was carried out in 155 patients. Further information can be found in Tables 1–3 and Supplementary Table S1.

Patients with TSR-high tumors were significantly associated with higher pT (p = 0.043), pN (p < 0.001) and M (p < 0.001) descriptors (Table 1). Also, lymphatic and perineural invasion was significantly higher amongst TSR-high patients (p < 0.001 and p = 0.002) (Table 1). TSR correlated with age, CEA and CA19-9, using Chi-squared test (p = 0.022, p = 0.029, p = 0.035). Similarly, KM-low correlated with advanced pT, pN, stage and M-status (p = 0.029, p = 0.024, p = 0.017 and p = 0.009) and also, a tendency towards lymphatic invasion (p = 0.093) was found (Table 1). As expected based on TSR and KM grading results, GMS was also associated with more advanced pT, pN and M descriptors and stage (p = 0.015, p = 0.001, p < 0.001 and p < 0.001) and lymphatic and perineural invasion (p < 0.001, p = 0.010), and there was a tendency towards vascular invasion (p = 0.065) (Table 2). KM, TSR or GMS were not associated with any SIR markers (Supplementary Tables S3, S4).

Elevation of serum CRP was associated with increasing stage (p = 0.002), pT (p < 0.001), distant metastasis (p = 0.007), higher grade (p = 0.027), vascular invasion (p = 0.026), lymphatic invasion (p = 0.032) and there was a trend towards perineural invasion (p = 0.063) (Supplementary Table S3). There was a significant correlation between ANC and higher pT (p = 0.011) and a trend towards advanced pN (p = 0.058) (Supplementary Table S3). ALC did not correlate with any of the examined features, but with younger age (p = 0.007). APC was significantly elevated in males (p = 0.003), associated with right-sidedness (p = 0.013), CMS1 (p < 0.001), showed significant association with lymphatic invasion (p = 0.045) and there was a tendency towards distant metastasis (p = 0.068) and vascular invasion (p = 0.077) (Supplementary Table S3, Figure 3).

NLR did not show any correlation with tumor descriptors, but PLR showed significant association with higher age (p = 0.046), right sidedness (p = 0.022) and CMS1 (p = 0.005) using nonparametric Kruskal-Wallis H-test (Supplementary Table S3). The mGPS showed significant association with higher grade (p = 0.042) and a tendency towards elevated pT (p = 0.057) (Supplementary Table S4). NLR and PLR did not show any correlation with clinicopathological descriptors, whereas NPS was significantly associated with male gender (p = 0.003), higher pT and pN stages (p = 0.043 and p = 0.032) and was inversely associated with the frequency of epithelial phenotype using Chi-squared test (p = 0.034) (Supplementary Table S4).

CEA was significantly lower in left sided tumors (p = 0.033). Elevated CEA levels were associated with stage (p < 0.001), pT (p = 0.044) and distant metastasis (p < 0.001) and also showed a tendency towards higher pN (p = 0.062) stage, besides, using Chi-squared test, TSR-high tumors were associated with higher CEA levels (p = 0.029) (Supplementary Table S3, Table 1 and Figure 4).

CA19-9 was also associated with stage (p < 0.001), pT (p = 0.002), distant metastasis (p < 0.001), lymphatic invasion (p = 0.015), and GMS (p = 0.027). There was a tendency towards vascular (p = 0.065) and perineural (p = 0.081) invasion, and with Chi-squared test there was significant association between TSR-high status and elevated CA19-9 (p = 0.035) (Supplementary Table S3, Table 1 and Figure 4).

Low CK expression was associated with higher age, and TSR-low (p = 0.012, p = 0.003), and inversely associated with presence of lymphatic invasion (p = 0.020) (Supplementary Table S5). Low FRMD6 expression was positively associated with TSR-low (p = 0.041), elevated serum CEA (p = 0.008) and albumin concentration (p = 0.026). Loss of CDX2 expression positively correlated with lymphatic and perineural invasion (p = 0.026 and p = 0.037), with NLR-high (p = 0.023) and showed tendency towards higher pT (p = 0.087) and increased serum albumin concentration (p = 0.081). The expression of ZEB1 was observed in only 14% of CMS classified cases. Interestingly, ZEB1 positive cases had lower CRP-levels (p = 0.039) and showed tendency towards lower mGPS (p = 0.074). No significant associations were revealed regarding HTR2B expression, only a tendency towards higher pN (p = 0.092) and M (p = 0.075) and higher PLR (p = 0.083) (Supplementary Table S5).

CMS1 was significantly associated with right colonic localization (p = 0.006) and higher histological grades (p < 0.001). CMS4 was associated with higher stage (p = 0.006), lymphatic and perineural invasion (p < 0.001 and p = 0.006, respectively) pN (p = 0.001) and M (p = 0.022) descriptors, and there was a tendency towards high TSR just failing to be significant (p = 0.054) as well (Tables 1, 3; Figure 3). We did not find significant correlation between the examined tumor markers (CEA and CA19-9) and CMS (p = 0.439 and p = 0.215) (Figure 3; Table 3).

With Kaplan-Meier survival analysis we found that some microenvironmental and systemic markers of CRC were associated with OS (Supplementary Table S6). Patients with high stromal content (p < 0.001), high GMS (p = 0.003), high ANC (p = 0.007), low albumin (p = 0.027), elevated CRP (p = 0.006), elevated CEA (p < 0.001) and CA19-9 (p < 0.001) as well as higher mGPS (p = 0.002) and mesenchymal subtype (p = 0.049) had shorter overall survival (Supplementary Table S6, Figure 5).

As for local relapse free survival, CEA was the only variable that stratified survival significantly (p = 0.009), and also a tendency for high PLR (p = 0.087) was observed (Supplementary Table S6). For distant metastasis free survival, TSR (p = 0.017) and serum albumin (p = 0.031) were associated with survival, while there was a tendency towards poor survival with GMS (p = 0.057), CEA (p = 0.066) and CRP (p = 0.092). Figure 6.

In the univariate Cox regression analysis TSR, GMS, mGPS, ANC, CRP, Albumin, CEA and CA19-9 were significantly associated with OS; CMS presented a tendency (with a p = 0.055, just failing to be significant) towards increased risk of death in CMS4 patients and NLR and NPS also showed tendency for poorer OS (Supplementary Table S7).

In the multivariate analysis, TSR (p = 0.029), NPS (p = 0.033), mGPS (p = 0.003), Albumin (p = 0.003), CRP (p = 0.018), CA19-9 (p = 0.013), and STM-score (p ≤ 0.001) were significant predictors of OS (independently of sex, grade, stage and vascular invasion) (Supplementary Table S7).

Scoring systems comprising of semi-quantitative aspects of certain TME or systemic inflammation-based markers are no novelty in oncology. A combination of strong prognostic factors have the ability to further identify a subset of patients with particularly poor outcome.

In our research the strongest independent TME-based marker was the TSR (Supplementary Table S7). Also, CA19-9, a tumor marker, often, though not routinely used in colorectal cancer follow up, came out as a predictor of overall survival in our analysis (Supplementary Table S7). Incorporating these two, STM was created. The assessment of STM is described in Methods previously.

Cases classified as STM2 were associated with younger age (p = 0.014), higher pN (p = 0.033) and M (p < 0.001), as well as higher TNM stage (p < 0.001), and presence of lymphatic (p < 0.001) and perineural invasion (p < 0.001), and also with elevated CEA-levels (p = 0.002) (Table 4). The mesenchymal subtype of CMS was also more prevalent in STM1 and STM2 groups (p = 0.048) (Table 4). There was a tendency towards higher pT (p = 0.071), also, preoperative serum CRP and CEA levels correlated with STM1 and STM2 (p = 0.017 and p < 0.001) (Table 4 and Supplementary Table S8).

The STM score significantly stratified 5-year overall survival (86% versus 54% versus 42%) with Kaplan-Meier analysis (Supplementary Table S6). Also in stage I-III patients there was significant difference between the distant metastasis free survival of STM0, 1 and 2 patients (p = 0.005) (Supplementary Table S6). In the univariate Cox-regression analysis STM was significantly associated with OS (HR: 7.4 (3-18), p < 0.001), and in the multivariate Cox-regression analysis STM was found to be an independent prognosticator of OS independently of sex, grade, stage and vascular invasion (p < 0.001, HR: 4.3 (1.5-12) (Supplementary Table S7).