Endoscopic polypectomy specimens (n = 16) used in this study were obtained from the Department of Pathology of the Leiden University Medical Center. Tissues were formalin-fixed and paraffin-embedded (FFPE) following routine diagnostic protocols. All samples were anonymized, according to the national ethical guidelines (“Code for Proper Secondary Use of Human Tissue,” Dutch Federation of Medical Scientific Societies).

FFPE tissues were sectioned at 6 μm thickness. Deparaffinization, rehydration and sialic acid linkage-specific derivatization were performed as described previously (8). Tissue enzymatic N-glycan release was performed applying 10 layers of PNGase F (0.1 μg/μL in Tris buffer) and CHCA matrix was used (5 mg/mL CHCA in 50:49.9/0.1 (% v/v/v) ACN:mQ:TFA).

All specimens were mixed and measured in randomized fashion. Spatial glycomics was performed on a rapifleX MALDI-TOF/TOF-MS instrument in positive ion reflectron mode (Bruker Daltonics). Spectra were recorded using 1,000 laser shots per pixel with 50 × 50 μm² pixel size (laser setting “M5 small”) over an m/z range of 900–3,300. After the analysis, excess MALDI-matrix was removed, tissues were hematoxylin and eosin (H&E) stained following routine histopathological procedures and then scanned using a digital slide scanner (IntelliSite Pathology Ultra-Fast Scanner, Philips, Eindhoven, Netherlands). In the flexImaging software, image scans and spatial glycomics data were co-registered and regions of interest (ROIs) were selected based on the histopathological analysis of the H&E slides by expert pathologists (AFS, HM). The spatial glycomics data, co-registered H&E images and ROIs from the whole cohort were imported into SCiLS lab software (2016b, version 4.01.8781, Bruker Daltonics).

Data preprocessing were performed using the same methods as in Boyaval et al. (8). Spatial segmentation using uniform manifold approximation and projection for dimension reduction (UMAP) analysis was performed on peak picked, TIC normalized single-pixel spectra, exported from SCiLS Lab for each individual tissue. UMAP was performed using the Seurat v3 package in R using default parameters. Each pixel were translated to RGB colors corresponding the color of the UMAP cluster, cluster corresponding to non-cellular areas were colored as white in the main figure. Molecular histology images were reconstructed in MATLAB. The correlation-based distance measures were performed using Pearson methods in R for each single tissue between the different morphologies by using the average intensity of the ROIs. In the Pearson correlation distance, similar object close to one other have a score close to 0, the bigger the distance the higher the score.

To test that spatial glycomics was capable of differentiating invasive adenocarcinoma from (non-invasive) dysplastic areas in colorectal polyps, eight cases of T1 colorectal cancer (malignant polyp) were analyzed (Table 1). In seven cases, the histological annotations of the invasive areas directly showed the expected molecular profiles. In Figure 1A, a representative case with invasive adenocarcinoma, high-grade, low-grade dysplasia and normal epithelium is depicted (Tissue-section 5, Table 1). Each of these morphological entities corresponded to a specific molecular signature as shown in the molecular histology reconstruct-image in Figure 1B.

In one case (Tissue 4, Supplementary Figures S1A,B and Table 1), the molecular signature of the invasive part did not fully match the histological annotation but highlighted the presence of mucinous pools with invasive cells as a novel separate cluster in the submucosa.

Then a total of eight cases with suspected PI were analyzed (Table 2). In seven cases, the morphological appearance of the suspected PI areas was similar to the one of the surface low-grade dysplasia epithelium (Supplementary Figures S1C,D). Rather than showing up as a distinct molecular phenotype, as would be expected for true invasive adenocarcinoma, the molecular histology of the suspected PI areas correlated highly with the histologically confirmed low-grade dysplasia areas in the same tissue (Pearson correlation distance = 0.05; Table 2 and Supplementary Table S1).

In another case, in Figures 1C,E (Tissue 9, Table 2), PI of normal and high grade dysplastic epithelium correlated thoroughly with the corresponding molecular histology in Figures 1D,E (Pearson correlation distance = 0.01 & 0.06).

Here, the application of MALDI-MSI-based spatial glycomics has been used to address the differential diagnosis of PI from true invasive cancer.

Dimensionality reduction on the high-dimensional spatial glycomics data provided molecular histology. This molecular histology showed comparability to conventional histomorphology in recognizing normal and different grades of neoplastic epithelium, but it displayed enhanced discriminatory potential between PI and true invasion.

In this report we showed that in all analyzed cases the glycomic molecular phenotype of PI was identical to that of the corresponding surface epithelium (either high-grade dysplasia, low-grade dysplasia or normal epithelium). We also highlight that the same technology is able to detect true invasion as separate molecular phenotype even with a high correlation score with the high-grade dysplasia (Pearson correlation distance = 0.26; Supplementary Table S1). Therefore, when a novel molecular phenotype is found in the deepest submucosal part, it might be used as an argument in favor of true invasion. These findings are in line with our previous studies that showed strong changes in glycomic signatures during the progression of stage I and stage II colorectal cancer (8). Additionally, others have demonstrated that employing the same technique has the potential to enhance tumor classification accuracy in pancreatic, ovarian, and hepatocellular carcinoma (9–11).

The fact that FFPE material is suitable for the spatial glycomics platform, and that the total diagnostic time spent is comparable to most immunohistochemistry protocols (approximately 2 days), makes this technique appropriate for running in parallel with the routine diagnostic workflow. The molecular histology reconstruct-image resulting from the spatial glycomics platform facilitates the direct comparison with H&E and other histopathological stainings by trained pathologists. However, while the samples, sample preparation, and required reagents for these analyses are easily obtainable and manageable within most pathology departments, it’s important to note that an imaging mass spectrometer and the expertise to operate the instrument and extract the data are necessary. The initial investment in the instrument and the requirement for specialized skills during operation should be carefully considered. For the data analysis part, once the pipeline and code are developed, it becomes user-friendly and applicable on any computer. This underscores the dual aspects of both accessibility and the requisite expertise associated with deploying this spatial glycomics platform in routine diagnostics. Nevertheless, this technique holds the potential to serve as a diagnostic tool in challenging scenarios where distinguishing between T1 colon cancer and PI is a subject of debate.

Clinical pathologists often pay more attention to morphological changes instead of the possible explanations behind them. Recently, an article proposed the concept of ecological pathology, stating that it is essential to apply the ecological (−evolutionary) principles and approaches to study the etiology, pathogenesis, pathological changes and outcomes of human diseases (12). For example, in this paper, tumor vascular mimicry is proposed to be an ecological adaptation. Similarly, the now discussed pseudoinvasion is a benign condition in which cancer is mixed in the colon by misplacement of dynamic lands in the submucosa, which might also be thought as an ecological adaptation due to the similar habitat.

This proof-of-concept study, while limited in cases, emphasizes the necessity for a more extensive cohort, particularly incorporating additional adenomas with high-grade dysplasia in the pseudoinvasion component, to validate these findings. Increasing the sample size holds the potential to significantly enhance statistical robustness, pattern generalizability, stability, and detection reliability. Such a comprehensive dataset is crucial for establishing a strong foundation, ensuring precision across cases, and minimizing the impact of variability in practically applying glycomic patterns to clinical diagnoses. While ongoing optimization of MSI analysis techniques remains valuable, and several studies have been working on it (13–16), the true strength of the separation criteria lies in the extensive representation of diverse cases through a larger sample size.

However, the presented results highlight the potential of spatial glycomics-based molecular histology as novel tool for routine diagnostics. We hope that this technical report will be a first steppingstone in this area with MSI based technology combined with dimensionality reduction algorithm.