Tumour sample collectionThe study cohort consists of human tumour tissues resected from patients with CRC (n = 14). The cohort has a balanced distribution of the anatomical origin of the tumour, stages and morphology of the tumour, as detailed in the sample information table (Table 1: Sample information). The samples were collected and used according to the approval and regulations of the Ethics Committee of the University Medical Center Goettingen (Ethics approval: #5/10/17). All patients provided written informed consent for using pathology specimens for research purposes. The study was performed in accordance with the Declaration of Helsinki.Table 1 Sample informationLabel-free imaging2PLSM images were acquired using an upright TriM Scope II multiphoton microscope (Miltenyi Biotec, Bielefeld, Germany) equipped with a tunable femtosecond laser (Cronus 2 P, Light Conversion, Vilnius, Lithuania). The laser was tuned to 870 nm at 15% of its maximal output power (1.0 W) for SHG and 2PEF excitation. Images were acquired using an Olympus XLPLAN 25x (NA 1.05) water immersion objective. The backscattered emitted light was split by 495 nm and 560 nm long pass dichroic mirrors (Semrock) and detected through the objective lens at photomultiplier tube (PMT) and GaAsP detectors (Hamamatsu). A PMT in the transmission position collected the forward scattered emitted light gathered by a 1.4 NA condenser lens under the stage. SHG and 2PEF signals were collected using the filter settings 434 ± 20 nm and 525 ± 50 nm, respectively (BrightLine HD filters, Semrock, AHF Analysentechnik Tübingen, Germany). Collagen fibre signals were collected as forward scattered SHG (F-SHG). The 2PEF signal was collected from backscattered photons. 2D mosaics of entire sections were acquired with individual image sizes of 393 × 393 µm with 1024 × 1024 pixels, a frequency of 600 Hz with 2-fold averaging, and 10% overlap within each tile of the mosaic.MALDI IMS sample preparationAll tissue samples were fixed in 4% paraformaldehyde and embedded in paraffin. 5 µm thick sections were cut from the paraffin blocks using a vibratome (VT1000 S; Leica Biosystems) and mounted on conductive glass slides coated in indium tin oxide (Bruker Daltonik GmbH, Bremen, Germany). The sections on the slides were covered with a coverslip before imaging by 2PLSM. After 2PLSM imaging, the coverslip was removed, and the samples were prepared for MALDI MSI as detailed by Wu et al.24. Sections were preheated to 80 °C for 15 min. Deparaffinisation was performed by successive immersion in xylene, 100% isopropanol, and successive hydration steps of 100%, 96%, 70%, and 50% ethanol for 5 minutes each. Heat-induced antigen retrieval was performed in MilliQ-water in a steamer for 20 min. After drying the slides for 10 min, tryptic digestion was performed using an automated spraying device (HTX TM-Sprayer, HTX Technologies LLC, ERC GmbH, Riemerling, Germany). 16 layers of tryptic solution (20 µg Promega® Sequencing Grade Modified Porcine Trypsin in 800 µL digestion buffer-20 mM ammonium bicarbonate with 0.01% glycerol, flow rate = 0.015 ml/min velocity = 750 mm/min, track spacing: 2 mm, concentration = 0.025 mg/ml, pattern CC) were sprayed onto each section at 30 °C. Tissue sections were then incubated for 2 h at 50 °C in a humidity chamber saturated with potassium sulphate solution. After incubation, the HTX TM Sprayer applied 4 layers of the matrix solution (7 g/L a-cyano-4-hydroxycinnamic acid in 70% acetonitrile and 1% trifluoroacetic acid, flow rate = 0.012 ml/min velocity = 1200 mm/min, track spacing: 3 mm, concentration = 7 mg/ml, pattern HH) at 75 °C.MALDI MSIMALDI imaging was performed on the rapifleX® MALDI Tissuetyper® (Bruker Daltonik GmbH, Bremen, Germany) in reflector mode with a detection range of 800–3200 m/z, 500 laser shots per spot, a 1.25 GS/s sampling rate and raster width of 50 µm. FlexImaging 5.1 and flexControl 3.0 software (Bruker Daltonik GmbH) were used in coordination. External calibration was performed using a peptide calibration standard (Bruker Daltonik GmbH).Protein identificationIdentification of proteins based on m/z value was conducted on adjacent tissue sections utilising a bottom-up approach involving nano-liquid chromatography electrospray ionization tandem mass-spectrometry, as outlined in previous work22. Firstly, the samples were preheated for 15 min at 80 °C, followed by deparaffinization, antigen retrieval, and tryptic digestion. The samples were then incubated for 2 h at 50 °C in a humidity chamber filled with a potassium sulphate solution. Peptides were separately extracted from each tissue after 2 h into 40 µL of 0.1% trifluoroacetic acid (TFA) and then incubated for 15 min at room temperature. As per the manufacturer’s guidelines, the digests were filtered through a ZipTip®C18, and the filtrates were concentrated using a vacuum evaporator (Eppendorf® Concentrator 5301, Eppendorf AG, Germany) and re-dissolved in 0.1% TFA. 2 µL of the peptide mix was loaded to an Acclaim PepMap™ 100 C18 trap column (100 µm × 2 cm, PN 164564, Thermo Fisher Scientific, USA) and pre-treated with 10 mM sodium hypofluorite (flow rate 20 µL/h) before being separated on an Acclaim PepMap™ RSLC C18 column (75 µm × 50 cm, PN 164942, Thermo Fisher Scientific, USA) using a 2–35% acetonitrile gradient in 0.1% formic acid (flow rate 400 nL/min, pressure range 10–800 bar) over 90 min at 60 °C. The tandem mass spectrometer (Impact II, ESI QTOF MS, Bruker Daltonik GmbH, Bremen, Germany) detected the ionized peptides through a full-mass scan (150–2200 m/z) at a 50,000 FWHM resolution. The autoMS/MS Insant Expertise feature selected peaks for fragmentation via collision-induced dissociation. For peptide identification, the peak lists were searched in the human Swiss-Prot database using PEAKS-studio-proteomics software (Bioinformatics Solutions, Version 11.6) using PEAKS-DB and PEAKS-de-novo sequencing. For MALDI MSI and LC–MS/MS m/z value comparisons, identification required more than one peptide with mass differences <0.15 Da. Peptides that had the highest score (-10lolgP, protein confidence score) and the smallest mass differences between the MALDI MSI and LC-MS/MS datasets were assumed as identified. Deisotoping was performed.HistologyFollowing MALDI -MSI, the matrix was removed from the tissue sections with 70% ethanol, and the sections were then washed in distilled water for 7 min. They were stained with haematoxylin (Himedia) for 10 min and washed with warm water twice for 10 min each. This was followed by washing with distilled water. The slides were then stained with eosin (Himedia) for 60 s and washed twice with distilled water. Serial dehydration was performed by immersing the slides in 80%, 96%, 100% ethanol, and xylol, followed by mounting them with a coverslip. The slides were then scanned with a whole-section scanner (NanoZoomer, Hamamatsu). These haematoxylin and eosin (H&E) images were used to annotate tumour regions in Qupath SW23 by a pathologist before transferring them into SCiLS Lab software version 2024a Pro (12.00.15110).Image processingThe F-SHG signal from 2PLSM images was extracted and denoised using the global Otsu adaptive thresholding. Texture analysis was performed on the SHG signal as explained before14 to quantify the coherence of collagen fibre. The local coherence heatmap is bound between values 0 and 1, where 1 indicates highly aligned structures, and 0 indicates chaotic and disorganised fibres. This coherence map was overlaid on the original 2PLSM image in FIJI by merging images to stack, followed by taking a maximum intensity projection of the merged stack. The coherence regions were classified as chaotic (local coherence values from 0 to 0.5) and organised (local coherence values from 0.5 to 1). The percentage area occupied by chaotic and organised regions was quantified.The nuclei were segmented from the H&E images using the “2D_versatile_he”pre-trained model from StarDist24. During post nuclei segmentation, the nuclei distribution was quantified by identifying the number of nuclei per unit area. A heatmap of nuclei distribution was formed, bound between 0 and 1, where 1 indicates densely packed nuclei, and 0 indicates nuclei scattered largely in space. This nuclei distribution map was then overlaid on the original H&E image in FIJI by merging images to stack, followed by taking a maximum intensity projection of the merged stack. The nuclei distribution regions were classified as low nuclei distribution (values from 0 to 0.5) and high nuclei distribution (values from 0.5 to 1). The percentage area occupied by low and high nuclei distribution regions was quantified. The nuclei segmentation and texture analysis source codes are available at dedicated GitHub repositories25,26.STRING analysisProtein-protein associations were derived from the STRING database (https://string-db.org/). The list of UniProt IDs of peptides identified by MALDI MSI was added in the ‘Multiple proteins’ search for the organism Homo sapiens and the confidence interval score of 0.4 to generate a full string network indicating both functional and physical edges. The thicker edge indicates the strength of the data support. Functional enrichment analysis was performed by applying the following settings: Maximum FDR < 0.01, strength >0.01 and minimum count in network.MALDI MSI data processing for statisticsMALDI MSI raw data were imported into the SCiLS Lab software version 2024a Pro (Bruker Daltonik GmbH) using settings to preserve the total ion count without baseline removal and converted into the SCiLS base data .sbd and .slx file. An attribute table was made, which consists of annotated regions of LSCC, RSCC, high collagen coherence, low collagen coherence, high nuclei distribution and low nuclei distribution. These attributes were used to divide the dataset into respective spatial groups to compare the differences in the spectra in particular regions of the tissues. Peak finding and alignment were conducted across a dataset using a standard segmentation pipeline (SciLS Lab software) in maximal interval processing mode with TIC normalisation, medium noise reduction and no smoothing. Annotated regions were transferred from QuPath as sef.file into SCilS Lab.Statistics evaluationDiscriminative MALDI MSI m/z values from annotated regions of each group (LSCC, RSCC, high collagen coherence, low collagen coherence, high nuclei distribution and low nuclei distribution) were identified using supervised receiver operating characteristic (ROC) analysis. The area under the curve (AUC) varies between 0 and 1, where values close to either 0 or 1 indicate discriminatory peptides and values close to 0.5 indicate that the peptides have a similar distribution in the groups. The presence of multiple peptides of the same protein with a similar ROC value (all peptides with AUC ≥ 0.6 ≤ 0.4) was considered a good indicator of the presence of protein. The peptides with an AUC ≥ 0.6 ≤ 0.4 were selected as candidate markers for principal component analysis (PCA) and segmentation using bisecting k-means clustering analysis on the respective groups.A two-way ANOVA was performed to compare the percentage area occupied by chaotic regions against organised regions and similarly to compare the percentage area occupied by the high against low nuclei distribution regions in the LSCC v/s RSCC. Statistical analysis was performed with Graph Pad Prism 9. A p-value of 0.05 was considered as a margin for statistical significance. Data is presented as mean ± SD, * indicates p ≤ 0.05, ** indicates p ≤ 0.01, *** indicates p ≤ 0.001, **** indicates p ≤ 0.0001.Figures were created using the SCiLS Lab software (Bruker, Bremen, Germany, 2024a Pro), biorender, and Inkscape, and graphs were created using GraphPad Prism 9.
