MALDI image-based molecular classification of difficult-to-diagnose rare gastrointestinal tumors in dialog with pathological and clinical aspects

By using analytical methods such as MALDI imaging on formalin-fixed tumor tissues, we can examine cancer proteomes sensitively and rapidly from a diagnostic point of view. To reflect the heterogeneity, we analyzed a collection of gastrointestinal cancer samples, including rare gastric and ampullary cancers, according to WHO classification and MALDI imaging TOF classification algorithms. Redetermination of the cancer tissues by mass spectra demonstrated intra-tumoral heterogeneity as well as relatedness to tumor stroma and overlapping mass spectra of tumor subtypes - independent of patient. Interestingly, greater parallels were found with the detailed histo-morphological tumor classification than with other clinical parameters at the time of surgery. This also shows the good usability of this tool as a substitute for classical histological analyses with the addition of the representation of paracrine (cellular) and tissue group-related correlations. Tumor detection is successfully demonstrated on the tumor sample itself, metastasis or artificially generated samples with little surrounding material such as organoids. For example, out of this analysis mass peaks 1441, 1775 and 2886 m/z were exclusively related to gastric tumor areas of the stomach. Transferability and comparability from one sample to another was calculated by leave-one-out classification. Mass spectra of unknown tissue matched the reference tumor spectra yielded positive predictive values ranging from 52% -77%, and showed the highest migration from tumor specific pixels to tumor-altered material such as stroma or inflammation. Gastric cancer organoids with different histological subtypes had divergent molecular signatures to that extent, that could be traced back to the signatures of the original tumor out of the gastric intestinal cancer collection. Overall, the classification of organoids and donor tissues only was 54% successful on average and need to be improved by sample size and training method before relying on the representative restriction of organoids. In addition, subclasses that are particularly difficult to diagnose, such as cancer mixed types, showed relationships to subclasses of the pre-trained MALDI imaging classes similar to the pathological assessment. Using ampullary carcinoma as an example, we were able to identify all subgroups and reveal new features of those mixed ones. In summary, we can demonstrate the potential of a proteome-based classification system for the development of grading of gastrointestinal carcinomas. In addition to histological detection, MALDI TOF analysis reveals many other aspects of how tumor tissue can be characterized. Rare tumor types in particular can benefit from this robust and fast detection of similarities or differences to existing tumor subtypes. The use of AI analysis will bring further significant developments in this area.