Lipidsearch

All spectra were analysed using Xcalibur™ (Thermo Fisher Scientific), with drug analytes being identified by their protonated molecular ion peak (within 1 ppm) and retention time compared to certified reference materials. LipidSearch™ (Thermo Fisher Scientific) was used to make lipid peak assignments in the single cells. Only peaks above a 1E4 intensity threshold and within 5 ppm of the calculated m/z value were selected. The data was corrected using blank samples extracted from identical PBS solution as the surrounding cells using the pressure injector with the same settings as used for cell extraction (diluted with 50 : 50 MeOH/EtOH – blank correction of 10× signal to noise ratio).
To compare different cell populations, MetaboAnalyst 5.0 (MKS Umetrics) was used to conduct partial least squared discriminant analysis (PLS-DA) analysis. Prior to analysis, the data was pareto scaled. The PLS-DA analysis provided a list of lipid features and their corresponding variable importance in projection (VIP). A VIP score is a measure of a variable's importance in the PLS-DA model. It gives the contribution each lipid feature makes to the model, therefore the higher the VIP score, the more it contributes to the model.

Metabolite assignments, isotopologue distributions, and correction for expected natural abundances of deuterium, ¹³C, and ¹⁵N isotopes were performed using MAVEN (Princeton, NJ, USA) [57 (link)], against an in house library of deuterated lipid standards (SPLASH LIPIDOMIX Mass Spec Standard, Avanti Lipids) and in house libraries of 3,000 unlabeled (MSMLS, IROATech, Bolton, MA, USA; IroaTech; product A2574 by ApexBio; standard compounds for central carbon and nitrogen pathways from SIGMA Aldrich, St Louis, MO, USA) and labeled standards (see below for the latter). Untargeted lipidomics analyses were performed with the software LipidSearch (Thermo Fisher, Bremen, Germany). Results from LipidSearch were exported as a library and additional discovery mode analyses were performed with standard workflows using Compound Discoverer 2.1 SP1 (Thermo Fisher Scientific, San Jose, CA). From these analyses, metabolite IDs or unique chemical formulae were determined from high-resolution accurate intact mass, isotopic patterns, identification of eventual adducts (e.g., Na⁺ or K⁺, etc.) and MS2 fragmentation spectra against the KEGG pathway, HMDB, ChEBI, and ChEMBL databases.

To select potential lipid biomarkers that contributed to the separation between the 0 d and 4 d groups grown under nitrogen starvation stress, three criteria, specifically, the high variable importance in the projection (VIP) and jack-knifed confidence interval (CIJF_JK) excluding zero and an absolute value of p(corr) higher than 0.4, were used29 (link)30 . The lipid metabolites were identified using a commercial MS/MS database, Lipid Search (Thermo Fisher Scientific, USA). Changes in PC and PE were expressed as the relative abundance (%), which was the ratio of the peak intensity of the test group (4 d) to that of the control group (0 d) with the same dry weight, and the relative abundances (%) of the control was expressed as 100%43 (link)44 45 .

Thermo Scientific™ LipidSearch™ software version 4.2 was used for lipid identification and quantification. An in-house script written in R (version 4.0.2) was used for data QC analysis, normalization, and plotting. The data was filtered to exclude any peak concentration estimates with a signal to noise ratio (SNR parameter) of less than 2.0 or a peak quality score (PQ parameter) of less than 0.8. If this exclusion resulted in the removal of two observation within a biological triplicate, the remaining observation was also excluded. The individual concentrations were then gathered together by lipid identity (summing together the concentration of multiple mass spectrometry adducts where these adducts originated from the same molecular source and averaging together biological replicates) and grouped within the broader categories of AcCa, TG, DG, PC, PE, PG, CL, PI, PS. The result was nine groups containing multiple lipid concentrations corresponding to specific lipid identities, which were then compared between wild type and KO samples using a (paired, non-parametric) Wilcoxon signed-rank test at an overall significance level of 5% (using the Bonferroni correction to account for the large number of tests performed). As the Bonferroni correction is fairly conservative, significant differences are reported at both pre-correction (*) and post-correction (***) significance levels.