Project five 5

Raman data were processed using the Project FIVE 5.2 software (WITec GmbH, Ulm, Germany). Cosmic rays were removed, and a baseline correction was employed on all spectra. Spectra were cropped to the range from 300 cm⁻¹ to 3045 cm⁻¹. True Component Analysis (TCA) was employed to identify major spectral components in the images. To extract single spectra, masks were generated based on TCA heat maps. To reduce the dimensionality of the spectral data, principal component analysis (PCA) was performed using the Unscrambler × 14.0 software (Camo Software, Oslo, Norway) [21 (link)]. For PCA analysis, the spectral fingerprint region between 400–1800 cm⁻¹ was investigated. PCA results are presented as score plots and loading plots. In brief, PCA is an unsupervised algorithm elaborating spectral differences and similarities on a vector-based approach. The identified vectors are also called principal components (PCs), where PC-1 explains the most relevant difference in the data set, PC-2 the second most relevant difference, etc. PC-1 often only describes differences based on spectral intensities or background signals. PCs were selected based on a clustering of the data sets of interest and the biological relevance of the most influencing peaks in the assigned loadings plot.

Raman images were processed and analyzed by TCA using the Project FIVE 5.2 software (WITec GmbH).^[23] Briefly, TCA is a non‐negative matrix factorization‐based MVA tool that elaborates spectral components that predominantly occur in the data set and allows the identification and localization of these components by false color intensity distribution heatmaps. GVI per pixel were determined in ImageJ to semiquantify the distribution of the spectral components in both conditions. Furthermore, TCA allowed the preselection of ROIs of similar spectral information, which were extracted for further in‐depth analysis of the molecular composition by PCA using Unscrambler X (Camo). Moreover, MCR analysis was applied to define the local spectral composition within the islet periphery and islet core of the pseudoislets. NID1 core, control core, NID1 periphery and control periphery were defined by the islet area and compared to determine the protective effect of NID1 on the pseudoislets.

Images were acquired at 1 µm/pixel resolution, resulting in a matrix of 62,500 pixels for a scan area of 250 by 250 µm, or 90,000 pixels for 300 by 300 µm (for combination with immunofluorescence staining). This led to a total of more than 62,000,000 data points (90,000 columns x 692 rows). Raw spectral data was preprocessed by cosmic ray removal, background subtraction and baseline correction using Project FIVE 5.2 (Witec, Ulm, Germany). All methods for data preprocessing haven been described before^{45 (link)}. We analyzed the “whole spectrum” with wavenumbers from 300 to 3000 cm⁻¹ and biological “fingerprint spectrum” between 400 and 1800 cm⁻¹, as defined previously^{18 (link)}. Technical outliers, identified as extreme values of the spectra, were filtered out. A limitation in use of formalin-fixed, paraffin-embedded (FFPE) tissue sections is artifacts caused by paraffin residues leading to sharp, distinct peaks, which may lead to unwanted clustering of pixels. Hence, areas contaminated with paraffin were detected as distinct subcluster (Supplementary Fig. 2b). For reasons of reproducibility we decided to exclude the paraffin peaks occurring at wavenumbers of 1064, 1132, 1294 and 1441 cm⁻¹, which have been described in literature before^{46 (link)}. For cryosections, no paraffin removal was applied.

Raman microspectroscopy (WITec alpha 300 ​R, Ulm, Germany) was performed at the interface of the fibrotic capsule as described previously [23 ]. A 63× objective (W Plan-Apochromat 63×/1.0 M27, Carl Zeiss AG) was used to image deparaffinized and hydrated tissue sections, sequential to sections subjected to IF staining. Spectral preprocessing and analysis were conducted in Project Five 5.2 (WITec GmbH). True component analysis (TCA) was applied to generate false-color coded intensity distribution heatmaps based on reference spectra of αSMA, Col I, Col III and nuclei generated in a previous study [23 ]. To obtain reference spectra of immune cells, Raman imaging was correlated with CD68/CCR7 or CD68/CD204 positive immunofluorescence images and spectra from co-localized pixel were extracted. Triplicate images were selected for each sample. Similar to ECM signatures, the retrieved spectra were used as reference spectra to identify macrophage polarization states via TCA. MGV was also performed to quantify the signal intensities of the specific proteins. The cell numbers were counted manually by using ImageJ V 1.52p.

All acquired Raman spectra were preprocessed in WITec ProjectFIVE 5.2 and Matlab following the same pipeline on a per pixel basis: cosmic ray removal (filter size: 4; dynamic factor: 4.1), setting minimum value in Rayleigh region (-150 - 50 cm^-1) to zero (detector dark current to zero), normalization setting the main water peak average value (3220−3420 cm⁻¹) equal to one, matrix/medium background subtraction using a matrix blank (i.e., 1 % agarose (w/v) in PBS), and rolling circle baseline correction (shape size: 300) to remove any other non-specific background signal artifacts. Finally, all spectra were cropped from 400 - 3100 cm⁻¹while also ignoring the biological "silent region" from subsequent unmixing (1800 - 2700 cm⁻¹).