Characterizing the phenotypic landscape of tumor buds
Aim
Tumor buds are tumor cells that are detaching themselves from the main tumor body and are hypothesized to be the starting point of a metastatic progression of the disease. Tumor buds are therefore a powerful biomarker in colorectal cancer. However, Currently, no phenotypic marker for tumor buds exists. Additionally, the interactions between tumor buds and their tumor microenvironment need to be investigated more thoroughly. To understand the phenotypes of tumor buds and their interacting cells, sequential Immunofluorescence (seqIF) is used to visualize up to 32 different proteins. An analysis pipeline is built to detect tumor buds and other cell types. Subsequently, unsupervised methods are used to uncover distinct subpopulations as well as enriched cell-cell interactions.
Methods
To characterize tumor buds and their transition states, a 31-plex sequential immunofluorescence (seqIF) panel was applied across two cohorts: a set of five colon cancer whole-slide images used to establish the marker panel and validate the pipeline, and a cohort of 159 locally advanced rectal cancer patients, represented as pretreatment biopsy cores on tissue microarrays. A custom image analysis pipeline was built to process the images: nuclei were segmented with StarDist, cells were phenotyped through a semi-automated gating strategy, and epithelial tissue was segmented with a random forest classifier to detect tumor buds (clusters of fewer than five malignant cells) alongside larger tumor clusters. From these, we derived quantitative morpho-molecular features — a collective epithelial downregulation (CED) score averaging CDX2, E-cadherin and membrane-bound Catenin beta-1, an interaction score capturing tumor–stroma contact and finger-like projections, and tumor cluster size. These features were correlated with molecular expression and, in the rectal cohort, linked to patient survival.
Results
Across both cohorts, epithelial markers were progressively lost as tumor clusters became smaller, with tumor buds showing the strongest downregulation. In rectal pretreatment biopsies, tumor buds displayed a marked reduction of epithelial and cell-status markers compared to larger, non-budding clusters (all p < 0.0001):
| Marker | Effect size (Cohen's d) |
|---|---|
| Catenin beta-1 (membrane-bound) | 2.50 |
| CDX2 | 2.34 |
| E-cadherin | 2.15 |
| Ep-CAM | 1.78 |
| Ki-67 | 1.62 |
| Lamin-B1 | 1.15 |
| Caspase-3 | 0.80 |
Treated as a continuous feature, cluster size tracked this molecular transition more finely than a simple budding/non-budding split, with CED score correlating negatively with cluster size in both cohorts (Spearman ρ ≈ −0.32 to −0.35) and levelling off around a cluster size of 128 cells. The same epithelial downregulation was seen in finger-like projections and cells in close contact with the stroma, supporting the idea that tumor buds, poorly differentiated clusters and infiltrative projections lie along a single morphological continuum.
On their own, no single marker or morphological feature including CDX2 expression, cluster size, or tumor bud counts separated patients by outcome. However, the within-patient correlation between CDX2 expression and cluster size was strongly prognostic: patients whose smaller clusters actively lost CDX2 had significantly worse survival (log-rank p < 0.0001 for both overall and disease-free survival), and this held in multivariate Cox models (OS: HR 2.8, p = 0.004; DFS: HR 2.0, p = 0.022).
In short, the loss of CDX2 in relation to tumor cluster size acts as a morphology-anchored biomarker of infiltrative growth, measurable directly in rectal cancer pretreatment biopsies and associated with worse disease-free survival.
Members
Mauro Gwerder
Inti Zlobec
Martin Weigert
Hannah Williams
Cristina Graham Martínez
Funding source
Publications
Gwerder, M., Saygili Demir, C., Williams, H. L., Lugli, A., Graham Martinez, C., Kowal, J., Khan, A., Kirchner, P., Koessler, T., Berger, M. D., Weigert, M. & Zlobec, I. CDX2 expression dynamics in tumor clusters: a morpho-molecular biomarker in rectal cancer pretreatment biopsies revealed by sequential immunofluorescence. Scientific Reports 16, 10129 (2026). https://doi.org/10.1038/s41598-026-40005-8
Collaboration