Cancer cell invasion takes place at the cancer-host interface and is a prerequisite for distant metastasis. The relationships between current biological and clinical concepts such as cell migration modes, tumour budding and epithelial-mesenchymal transition (EMT) remains unclear in several aspects, especially for the 'real' situation in human cancer. We developed a novel method that provides exact three-dimensional (3D) information on both microscopic morphology and gene expression, over a virtually unlimited spatial range, by reconstruction from serial immunostained tissue slices. Quantitative 3D assessment of tumour budding at the cancer-host interface in human pancreatic, colorectal, lung and breast adenocarcinoma suggests collective cell migration as the mechanism of cancer cell invasion, while single cancer cell migration seems to be virtually absent. Budding tumour cells display a shift towards spindle-like as well as a rounded morphology. This is associated with decreased E-cadherin staining intensity and a shift from membranous to cytoplasmic staining, as well as increased nuclear ZEB1 expression.

In cancer biology, the architectural concept "form follows function" is reflected by cell morphology, migration, and epithelial-mesenchymal transition protein pattern. In vivo, features of epithelial-mesenchymal transition have been associated with tumor budding, which correlates significantly with patient outcome. Hereby, the majority of tumor buds are not truly detached but still connected to a major tumor mass. For detailed insights into the different tumor bud types and the process of tumor budding, we quantified tumor cells according to histomorphological and immunohistological epithelial-mesenchymal transition characteristics. Three-dimensional reconstruction from adenocarcinomas (pancreatic, colorectal, lung, and ductal breast cancers) was performed as published. Tumor cell morphology and epithelial-mesenchymal transition characteristics (represented by zinc finger E-box-binding homeobox 1 and E-Cadherin) were analyzed qualitatively and quantitatively in a three-dimensional context. Tumor buds were classified into main tumor mass, connected tumor bud, and isolated tumor bud. Cell morphology and epithelial-mesenchymal transition marker expression were assessed for each tumor cell. Epithelial-mesenchymal transition characteristics between isolated tumor bud and connected tumor bud demonstrated no significant differences or trends. Tumor cell count correlated significantly with epithelial-mesenchymal transition and histomorphological characteristics. Regression curve analysis revealed initially a loss of membranous E-Cadherin, followed by expression of cytoplasmic E-Cadherin and subsequent expression of nuclear zinc finger E-box-binding homeobox 1. Morphologic changes followed later in this sequence. Our data demonstrate that connected and isolated tumor buds are equal concerning immunohistochemical epithelial-mesenchymal transition characteristics and histomorphology. Our data also give an insight in the process of tumor budding. While there is a notion that the epithelial-mesenchymal transition zinc finger E-box-binding homeobox 1-E-Cadherin cascade is initiated by zinc finger E-box-binding homeobox 1, our results are contrary and outline other possible pathways influencing the regulation of E-Cadherin.

Motivation Mass spectrometry imaging (MSI) characterizes the molecular composition of tissues at spatial resolution, and has a strong potential for distinguishing tissue types, or disease states. This can be achieved by supervised classification, which takes as input MSI spectra, and assigns class labels to subtissue locations. Unfortunately, developing such classifiers is hindered by the limited availability of training sets with subtissue labels as the ground truth. Subtissue labeling is prohibitively expensive, and only rough annotations of the entire tissues are typically available. Classifiers trained on data with approximate labels have sub-optimal performance. Results To alleviate this challenge, we contribute a semi-supervised approach mi-CNN. mi-CNN implements multiple instance learning with a convolutional neural network (CNN). The multiple instance aspect enables weak supervision from tissue-level annotations when classifying subtissue locations. The convolutional architecture of the CNN captures contextual dependencies between the spectral features. Evaluations on simulated and experimental datasets demonstrated that mi-CNN improved the subtissue classification as compared to traditional classifiers. We propose mi-CNN as an important step toward accurate subtissue classification in MSI, enabling rapid distinction between tissue types and disease states. Availability and implementation The data and code are available at https://github.com/Vitek-Lab/mi-CNN_MSI.