Sample preparation. For this study, fresh frozen (FF) rat brain tissues were cut using a cryostat at -20°C. All sections were obtained at the same time and stored at -80°C until their use.
Rat brain sagittal 12 µm sections were prepared, to finally reach 22 batch of 4 consecutive sections. Tissues were fixed on ITO slides and respectively intended to: back-up, lipid in negative and positive mode imaging, protein imaging and peptide imaging in positive mode.
10 others consecutive rat brain sagittal sections of 12 µm were mounted on poly-lysine coated slide for lipid analysis carried out by SpiderMass technology. Three consecutive another 20 µm sections were fixed on poly-lysine coated slide for spatial proteomic analysis.
Finally, 3 different rat brain sagittal 12 µm section were fixed onto ITO coated slide as a validation cohort for the lipid predictive model.
For the analysis of horizontal rat brain tissues, 4 consecutives sections were prepared for multi-omics MSI analysis as describe bellow, followed by another consecutive sections for spatial proteomic analysis. This schema was repeated on 4 different rat brains.
Lipid MALDI MS imaging. Tissues were dried in a desiccator before a matrix deposition. Norharman was used as MALDI matrix for positive and negative lipid imaging. The matrix was deposited at 7 mg/mL in CHCl3: MeOH (2:1, v/v). The HTX parameters for norharman spray were : spray at 30°C with 10 psi pressure, a pattern CC, a flow rate of 0.1 mL/min, a velocity of 1200 mm/min, for 12 passages with 2 mm track spacing. Lipid images were performed on the MALDI-TOF Rapiflex Tissuetyper mass spectrometer. The spectra were acquired within the m/z 200-1200 range in positive ion mode and the m/z 400-1500 range in negative ion mode. All data were performed in the delayed extraction reflectron mode with an average of 300 laser shots per pixel for a spatial resolution of 50 µm. The laser energy was set around 60 % and the voltages of the ion source were 20 kV and 11 kV for the lens.
Other images were performed with DHB matrix in positive ion mode. The matrix was deposited at 10 mg/mL in MeOH: TFA 0.1% (7:3, v/v). The HTX parameters for DHB spray were : spray at 75°C, tray at 55°C, with 10 psi pressure, a pattern CC, a flow rate of 0.1 mL/min, a velocity of 1200 mm/min, for 8 passages with 2 mm track spacing. Lipid images were performed on the MALDI-TOF Rapiflex Tissuetyper mass spectrometer. The spectra were acquired within the m/z 200-1200 range in positive ion mode. All data were performed in the delayed extraction reflectron mode with an average of 300 laser shots per pixel for a spatial resolution of 50 µm. The laser energy was set around 85 % and the voltages of the ion source were 20 kV and 11 kV for the lens.
Protein MALDI MS imaging. Tissues were vacuum dried before being subjected to delipidation using sequential baths of EtOH: H2O (70:30, v/v) for 30 s, EtOH 100% for 30 s, Carnoy solution (EtOH/Chloroform/Acetic acid, 3:6:1, v/v/v) for 2 min, EtOH 100% for 30 s, TFA 0.1%/H2O for 30 s and EtOH 100% for 30 s. After drying the sections, SA-Aniline (SA-ANI) MALDI matrix was deposited on tissue. SA-Aniline was prepared by dissolving sinapinic acid matrix at 10 mg/mL in ACN/TFA 0.1% (50:50, v/v and adding 24.3 µL of aniline. The HTX parameters included a temperature of spray at 75°C with 10 psi pressure, a pattern CC, a flow rate of 0.1 mL/min, a velocity of 1100 mm/min, a temperature of tray at 55°C, for 8 passages with 2 mm track spacing. The slides were analyzed on the MALDI-TOF Rapiflex Tissuetyper mass spectrometer. MS spectra were acquired in the positive linear delayed extraction mode, on the m/z 2400-30,000 range with an average of 700 laser shots per pixel and at a spatial resolution of 50 µm. The laser energy was set around 90 %. The voltages of the ion source were 20 kV and 9 kV for the lens.
Peptide MALDI MS imaging. For peptide imaging, the slides were dried and delipidated using a similar protocol as for protein MS Imaging. The tissue sections were then submitted to trypsin digestion. The tryptic digestion was performed by applying trypsin (40 µg/mL in NH4HCO3 50 mM). The HTX parameters included a temperature of spray at 65°C with 10 psi pressure, a pattern CC, a flow rate of 0.1 mL/min, a velocity of 1100 mm/min, for 12 passages with 2 mm track spacing. Once the trypsin was deposited the slides were incubated overnight at 56°C in a humidified box containing MeOH/H2O. The slides were then dried under vacuum over the next day. An HCCA-aniline matrix was deposited by the HTX M5-Sprayer. Briefly, 43.2 µL of aniline were added to 5 mL of a solution of 10 mg/mL HCCA dissolved in ACN/TFA 0.1% (7:3, v/v). Slides were analyzed on a MALDI-TOF Rapiflex. Spectras were obtained in the positive delayed extraction reflector mode analysis, with a mass range of 700-3200 m/z, and averaged from 500 laser shots per pixel for a spatial resolution of 50 µm. The laser energy was set around 40 %. The voltages of the ion source were 20 kV and 11 kV for the lens.
Multi-Omics-MSI segmentation. The raw MALDI-MSI data for lipids in both ionization modes, peptide and protein data were initially converted into the imzML format using SCiLS lab software. Subsequently, the imzML converter, version 1.3.3, was employed to import these datasets into MATLAB R2019a. It's worth noting that MSI data is characterized by high dimensionality, often reaching sizes of up to 100 GB per image. This magnitude makes it infeasible to analyze such data. To address this issue and prevent data loss using peak list generation, data compression was implemented as a preprocessing step before segmentation. Several data reduction (compression) algorithms were explored, including t-SNE (t-distributed stochastic neighbor embedding), NNMF (non-negative matrix factorization) and SVD (Singular Value Decomposition). For the segmentation process, the k-means++ algorithm was utilized, implemented as the 'kmeans' function in the MATLAB Statistics Toolbox. K-means++ offers an improved initialization of centroids, enhancing the quality of clustering. The cosine distance metric was employed to calculate the cosine angle between two spectra for quantifying the similarity. For visualization, each cluster's pixels are uniformly assigned a specific color, facilitating the creation of a segmentation map. This map delineates the cluster or region of interest to which each pixel (spectrum) belongs. To estimate the right numbers of clusters, the Silhouette criterion was used. After predefining the number of clusters, the silhouette plot method was used to assess the stability of the clusters. The silhouette plot displays a measure of the proximity of each point in a cluster. This measure has a range (-1, 1). A value close to 1 indicates that the cluster is distant from neighboring clusters (the spectra are very compact within the cluster to which it belongs and distant from other clusters). A value of 0 indicates that the sample is very close to the decision boundary between two neighboring clusters (overlapping clusters). Negative values indicate that these samples may have been assigned to the wrong cluster. Silhouette plot was calculated using the function silhouette in Matlab. Subsequently, each centroid within these clusters is thoughtfully exported in CSV format, ready for further in-depth analysis and exploration.
Differential analysis between clusters. The centroids generated from the image segmentation were imported into Python using the panda’s library. All centroid data was structured into a data frame. A custom script was developed to automate the execution of a statistical test. This script iterates over all m/z variables, identifying ions that exhibited statistical significance between the regions of interest (ROIs). To enhance data quality, a peak picking algorithm was employed. Specifically, the find_peaks_cwt function from the Scipy library was utilized to effectively remove instrument noise. A non-parametric statistical test, the Kruskal-Wallis test with Bonferroni correction, was conducted. Only features deemed statistically significant, with a p-value equal to or less than 0.05, were retained. A manual step is added to isolate and retain only the mono-isotopic peaks. The seaborn library was utilized to generate corresponding box plots.
Prediction model based on lipid MALDI imaging. The previously developed pipeline (ref glioblastoma 28 March) served as the foundation for constructing the optimal model adapted to the dataset. The SGD (Stochastic Gradient Descent)) model with isotonic loss function for calibrated probability scores, achieved the highest accuracy and a high F1 score. SGD combines multiple weak classifiers with slightly better than random performance to create a strong classifier. Sample weights are adjusted in each iteration to prioritize misclassified samples, and a series of boosting iterations are employed to train and combine the weak classifiers.
Lipid annotation by SpiderMass technology. The basic design of the instrument setup has been described in detail elsewhere20. In addition, here, the laser system used was an Opolette 2940 laser (OPOTEK Inc., Carlsbad, California, USA). The infrared laser microprobe was turned at 2.94 µm to excite the most vibrational band of water (O-H). The laser beam was injected into a 1 m reinforced jacketed fiber of 450 µm inner core diameter equipped at its extremity with a handheld including a focusing lens with 4 cm focal distance to get a 500 µm spot on the tissue. To aspirate and analyze the ablated material, a TygonⓇ tubing (Akron, OH, USA) is directly connected to Q-TOF mass spectrometer (Xevo, Waters, UK) through a REIMS interface. Each rat brain cerebellum clusters, observed by MSI, were analyzed by SpiderMass with four independent biological repetitions. Briefly, the laser was directly placed above the region of interest at the 4 cm focal distance. The laser energy was fixed to 4 mJ/pulse. On each spot, three acquisitions of 10 repetitive laser shots (10 Hz) were performed which resulted in 3 individual MS spectra. The data were acquired in both negative and positive polarities, in the sensitivity mode over a m/z 100-2000 range. The previously identified discriminative ions were selected for MS/MS analysis with 0.1 m/z isolation window. MS/MS was performed using collision induced dissociation (CID) with argon as collision gas and a collision energy of 25 eV.
Each rat brain cerebellum clusters, observed by MSI, were analyzed by SpiderMass with four independent biological repetitions. Briefly, the laser was directly placed above the region of interest at the 4 cm focal distance. The laser energy was fixed to 4 mJ/pulse. On each spot, three acquisitions of 10 repetitive laser shots (10 Hz) were performed which resulted in 3 individual MS spectra. The data were acquired in both negative and positive polarities, in the sensitivity mode over a m/z 100-2000 range.
The previously identified discriminative ions were selected for MS/MS analysis with 0.1 m/z isolation window. MS/MS was performed using collision induced dissociation (CID) with argon as collision gas and a collision energy of 25 eV.
Spatially resolved proteomics extraction. The different clusters identified by the segmentation process were submitted to spatially resolved proteomics. Each cluster was analyzed in triplicate from the same tissue section as describe bellow. A localized digestion was carried out by deposing a trypsin solution (40 µg/mL in NH4HCO3 50mM), on a region of 800 µm2 of tissue (4 x 4 droplets of 200 µm in diameter), using CHIP-1000. The deposition method comprises approximately 1205 cycles per digestion spot, i.e., 3 hours of deposition, with a drop volume of 150 pL. Finally, each spot was digested with 0.112 µg of trypsin. Following the micro-digestion, each spot was extracted by liquid micro-junction using the TriVersa Nanomate device, with LESA (Liquid Extraction and Surface Analysis) parameters. The tryptic peptides were extracted by performing 2 consecutive extraction cycles for three different solvents mixtures (TFA 0.1%; ACN/0.1% TFA (8:2, v/v); and MeOH/0.1% TFA (7:3, v/v)) for a total of 6 extractions. For each cycle, 2 µL of solvent was drawn into the tip of the pipette, of which 0.8 pL was brought into contact with the surface. 15 back and forth movements were performed to extract the peptides before collecting the solution in a recovery tube. All extracts were pulled in one tube and 50 µL of ACN were finally added before drying the samples in a SpeedVac. The samples were then stored at -20°C prior to nLC-MS/MS analysis.
nLC-MS/MS bottom-up analysis. All sample analysis were performed on a timsTOF fleX mass spectrometer online coupled to an Evosep One nano-flow liquid chromatography system. Peptides were separated using an 8 cm x 150 µm C18 column with 1.5 µm beads and the 60 samples per day method from Evosep One. The mobile phases comprised 0.1% FA in water as solution A and 0.1% FA in ACN as solution B. To perform DIA analysis in PASEF mode, one MS1 scan was followed by 10 dia-PASEF scans from m/z 100 to 1700. The ion mobility range was set to 1.42 and 0.65 V.s/cm-2. The accumulation and ramp times were specified as 100 ms. As a result, each MS1 scan and each MS2/dia-PASEF scan last 100 ms plus additional transfer time, and a dia-PASEF method with 22 dia-PASEF scans has a cycle time of 1.06s. The mass spectrometer was operated in high sensitivity mode, with a collision energy ramped linearly as a function of the ion mobility from 59 eV at 1/K0=1.6Vs.cm-2 to 20 eV at 1/K0=0.6Vs.cm-2. The ion mobility was calibrated with three Agilent ESI Tuning Mix ions (m/z, 1/K0: 622.02, 0.98 V.cm−2, 922.01, 1.19 V.cm−2, 1221.99, and 1.38 V.cm−2).
Proteomic data analysis. DIA-NN version 1.8.1 was used to search DIA raw files and dia-PASEF files. A Rattus library was generated with the software parameters set as following: complete proteome of Rattus norvegicus from UniProt database (Release January 2024, 92958 entries), Trypsin protease with 2 missed cleavages and a maximum number of variable modification at 3, methionine oxidation as variable, peptide length range from 7 to 30, precursor charge range from 1 to 4, precursor m/z range comprised between 100 and 1700, fragment ion m/z range between 200 and 1700, 0.1% precursor FDR, protein inference set on ‘genes’, neural network classifier on single-pass mode, quantification strategy set on robust LC (high accuracy), RT-dependent cross-run normalization, and library generation fixed on the ‘IDs, RT & IM profiling’ ruban. Samples were interrogated according the resulting Rattus library with the same options. Statistical analyses were carried out using Perseus software v2.0.5.0. ANOVA tests were performed with p-value ≤ 0.01 to be statistically significant and generate heat maps of differentially expresses proteins across sample. STRING and Gene Ontology (GO) analysis were performed using ClueGO with GO term database, on Cytoscape v3.10.2.
Glioblastoma samples. A retrospective cohort of 50 FFPE glioblastoma tissues was obtained from the Pathology department of Lille Hospital, France. A prospective cohort of 31 fresh frozen glioblastoma tissues were also included in this study. 31 patients with newly diagnosed glioblastoma were prospectively enrolled between September 2014 and November 2018 at Lille University Hospital, France (NCT02473484). All patients gave written informed consent before enrollment. According to the French Public Health Code and in application of the General Data Protection Regulations, all patients had been informed at the time of care that their standard clinical and biological data could be used for research purposes regarding the retrospective analysis of FFPE samples, and none had expressed his opposition. Regarding the prospective collection of samples, each patient's informed consent for the collection and publication of clinical and biological data was obtained at the time of hospitalization prior to surgical intervention.