The reporting of this study follows the guidelines outlined in the Reporting Recommendations for Tumor MARKer prognostic studies (REMARK). The study adhered to the Declaration of Helsinki and received approval from The Regional Committees on Health Research Ethics for Southern Denmark (S-20190164), with dispensation from obtaining informed consent from the study patients. No patients were excluded based on registration in the Danish Registry of Tissue Utilization.

This population-based study included 493 patients who underwent curative surgical resection of UICC stage II colon cancer between 2014 and 2016 in the Region of Southern Denmark. The patients were sourced from a screened population across four hospitals identified using the Danish Colorectal Cancer Group database and the Danish Pathology System. None of the patients included received neoadjuvant chemotherapy, and they had no history of colon cancer or any malignant disease (except non-melanoma skin cancer) within the 10 years leading up to the diagnosis of colon cancer. Patients who received postoperative adjuvant chemotherapy were included in the study, comprising 69 individuals, corresponding to 14% of the cohort. Further information on patient selection is described in detail elsewhere [28].

A retrospective histopathological characterization of the tumors was performed, coupled with a comprehensive review of medical records encompassing surgical details, follow-up information, and survival data. Details of the baseline clinicopathological characteristics have previously been presented [28].

All archived formalin-fixed, paraffin-embedded tissue blocks and slides originally utilized for routine diagnostic purposes were obtained from the four pathology departments in the Region of Southern Denmark. The number of tumor-containing tissue blocks per patient varied from 2 to 48, with a mean of 7. Histologic sections of 4-um thickness were cut from the tumor block with the highest degree of H&E-based tumor budding, and consecutive sections were stained with H&E and pan-cytokeratin, respectively.

Immunohistochemical staining was performed automatically on a DAKO Autostainer Link 48 platform (DAKO, Glostrup, Denmark) as described elsewhere [23]. In short, the primary antibody used was mouse monoclonal anti-cytokeratin (clone AE1AE3, code M3515, DAKO, Glostrup, Denmark) diluted at 1:250. Following deparaffination and rehydration, antigen retrieval was performed using Envision Target Retrieval Solution (DAKO, Glostrup, Denmark) at pH 9 and 97 °C for 20 min. Slides were treated with EnVision FLEX Peroxidase-Blocking Reagent (DAKO) for 5 min to inhibit endogenous peroxidase, followed by a 30-min incubation with primary antibody at room temperature. Amplification was achieved using Envision Flex + Mouse (Linker) (DAKO, Glostrup, Denmark) for 20 min. Detection of bound antibodies was carried out using Envision FLEX/HRP (DAKO, Glostrup, Denmark) and visualized with Envision FLEX DAB (DAKO, Glostrup, Denmark) and Chromogen. Hematoxylin served as the counterstain.

Tumor budding is defined as single tumor cells or clusters of up to four cells budding of the primary tumor [5]. The evaluation of tumor budding was done following the ITBCC guidelines, and all diagnostic H&E slides were reviewed at low power to identify the tumor block with the highest degree of budding at the invasive front [5]. Based on two consecutive sections from this tissue block, tumor budding was assessed using both H&E and IHC, following the same method: Ten individual fields were scanned at medium power, and tumor buds were counted in the hotspot area normalized to the field size of 0.785 mm2 using a Leica HC microscope. The tumors were categorized based on the proposed categories by ITBCC, including a separate category for Bd0 tumors. Consequently, a four-tiered scoring system, as suggested by Zlobec et al. [25], was implemented, classifying them into Bd0 (zero) 0 buds, Bd1 (low) 1–4 buds, Bd2 (intermediate) 5–9 buds, and Bd3 (high) ≥ 10 buds. We enumerated up to 100 buds and assigned a count of 100 to tumors exceeding this number. The pan-cytokeratin-stained tumor budding cells were required to show cytoplasmatic positivity and a clearly defined hematoxylin-stained nucleus to distinguish the cells from apoptotic bodies and cellular debris.

Caution was exercised when assessing tumor budding in regions exhibiting significant inflammation in order to differentiate true buds from mechanically fragmented glands and not erroneously count these so-called pseudo buds as genuine tumor buds.

The assessment of tumor budding was conducted by one observer, MPK, while SKF contributed to the interobserver evaluation. The observers scored the tumors independently of each other and were blinded to former bud count, as well as clinical and histopathological information. The intra- and interobserver reproducibility was assessed on 50 randomly selected tumor slides from both T3 and T4 tumors.

Summary statistics included mean and standard deviation (normal-distributed variables) or median and interquartile range (non-normal-distributed variables). Categorical variables are presented as numbers and percentages. Analyses of associations between tumor budding categories and clinicopathological characteristics used the chi-squared test or Fisher’s exact test, where appropriate. The Wilcoxon rank-sum test or Wilcoxon signed-rank test was employed for independent or matched continuous variables, respectively.

Weighted kappa statistics were used to determine the intra- and interobserver agreement between the tumor budding categories. A comparison of the tumor budding categorization assessed by H&E or IHC was performed using descriptive statistics and visualized using a scatter plot and a Bland–Altman plot.

For the prognostic evaluation, a receiver operating characteristic (ROC) curve analysis with either recurrence or death as an endpoint was performed to determine a clinically relevant cut-off score for IHC-evaluated tumor budding.

Time to recurrence (TTR) was defined as the time from surgery to the date of local or distant recurrence of colon cancer or the date of death from colon cancer. Recurrence-free survival (RFS) was defined as the time from surgery to the date of local or distant recurrence or death from any cause, whichever occurred first. Overall survival (OS) was defined as the time from surgery to death from any cause or end of follow-up. If no events occurred, all records were censored either at the point of loss to follow-up (n = 2) or at the end of the study period (May 15th, 2023). Events of metachronous cancer in the follow-up period were not considered a censoring event in the analyses [29].

Kaplan–Meier curves and log-rank tests were used to test for differences in survival times by the tumor budding groups. Uni- and multivariable Cox regression models were used to estimate hazard ratios (HR) and 95% confidence intervals (CIs). Bd0 was used as the reference group.

The multivariable analysis was adjusted for potential confounders identified by a previously published causal-directed acyclic graph (DAG) [28] and included the T category, mismatch repair (MMR) status, and histologic type. Multivariable analyses were conducted on complete cases (n = 492) due to minimal missing data (MMR status not assessed in one tumor).

Scaled Schoenfeld residuals checked the proportional hazard assumption for each regression analysis and did not violate it.

All analyses were carried out using Stata software (version 18.0 BE). All data were recorded in a Research Electronic Data Capture (REDCap®) database with an automatically generated entry check via the Open Patient Data Explorative Network (OPEN) organization. P-values of < 0.05 were considered to be statistically significant.