LipidCreator workbench to probe the lipidomic landscape

Abstract

Mass spectrometry (MS)-based targeted lipidomics enables the robust quantification of selected lipids under various biological conditions but comprehensive software tools to support such analyses are lacking. Here we present LipidCreator, a software that fully supports targeted lipidomics assay development. LipidCreator offers a comprehensive framework to compute MS/MS fragment masses for over 60 lipid classes. LipidCreator provides all functionalities needed to define fragments, manage stable isotope labeling, optimize collision energy and generate in silico spectral libraries. We validate LipidCreator assays computationally and analytically and prove that it is capable to generate large targeted experiments to analyze blood and to dissect lipid-signaling pathways such as in human platelets.

Fig. 1. The LipidCreator workbench and its integration into Skyline.

a Main steps include the decision of which lipids to query (Lipid Query), the calculation of the lipid precursor and corresponding fragment masses (Calculation) and the assay design (Targeted Assay). b These main steps consist of the selection of species or lipidomes to target (1–2), the inclusion of isotope-coded internal standards for validation (3), the calculation of precursor and fragment masses for the assay (4), library generation (5) and the in silico optimization of CE for individual fragments (6). The steps (1–6) are performed in LipidCreator. After the submission of the transition list to Skyline and the execution of the targeted lipidomics experiment, lipids can be validated by spectral library matching (7) and/or their coeluting internal standards, which are further used for quantification (8). Due to the integration features of Skyline, different downstream quality control systems, such as Panorama, are available (9). c Graphical user interface of LipidCreator integrated into Skyline.

Fig. 2. LipidCreator system levels and flow of information.

LipidCreator uses an internal knowledge base that stores information about lipids and their fragments. This information is assembled based on the user’s selection of lipid classes and additional parameters from the two main levels, the precursor and fragment level. Then, LipidCreator combines the user’s selection with information about lipid classes, precursors, backbones, fatty acyl chains, lipid fragments and information on isotope labels and forwards it to the transition layer for targeted assay generation. The downstream filter layer then applies collision-energy models to the generated transitions, if either automatic CE optimization or manual CE mode are enabled. The final export layer generates the final transition lists and spectral libraries and either stores them locally or transmits them directly to Skyline.

Fig. 3. Relative fragment intensity prediction for collision-energy optimization.

The workflow is connecting the assay development with LipidCreator (1), lipid standards measurement (2) and transition extraction from the measured data to create feature tables (3) for the training of fragment-specific nonlinear regression models for collision-energy-dependent, relative intensity prediction using nonlinear regression models (4). The subsequent QC and visualization step (5) supports model fit quality inspection and selection of the parameterized model parameters. LipidCreator uses these model parameters to calculate the optimal collision energy for a lipid based on the selected fragments (6) for further assay refinement and integration with the MS acquisition workflow.

Fig. 4. Lipid distribution and lipidome coverage of LipidCreator in different model organisms.

a–j Numbers of lipid species per lipid class within different organisms. A list with all unsupported lipid classes within this experiment is available in Supplementary Note 1 and in Supplementary Data 1. The latter table further contains detailed descriptions of all lipid name abbreviations and the lipid classes supported by LipidCreator. Source data are provided as a Source Data file.

Fig. 5. Probability and false match analysis proves LipidCreator output as correct.

For probability calculations, the yeast lipidome, a set of target lipids and a set of decoy lipids (in total: 188 negative/273 positive lipid ions) were chosen. a, b The minimum number of lipid ions that can be unambiguously identified using 0 fragments (i.e., precursor mass), 1 arbitrary fragment etc. was calculated in negative and positive ion mode, respectively. The MS and MS/MS mass tolerance was set to ±0.5 Da (representing low resolution instrumentation) or ±2.5 ppm (representing high resolution instrumentation). Sphingolipids (SL) and glycerophospholipids (PL) were investigated in both polarities while, glycerolipids (GL) and sterol lipids (ST) were investigated in positive ion mode only. c, d Cumulative probability to unambiguously identify any lipid ion when using 0 fragments, 1 arbitrary fragment etc. using 0.5 Da (low) or 2.5 ppm (high) tolerance in positive and negative ion mode, respectively. To verify the calculations, LC/ESI MS/MS experiments were conducted in a yeast lipidome background. The monitored lipid set contained yeast lipids, 21 target lipids, and 21 decoy lipids. Here, (un)identified target lipids are referred to as true positives or false negatives, whereas (un)identified decoys are denoted as false positives or true negatives, respectively. The identification is based on the upper number of used fragments, where 0 fragments means the identification is solely based on precursor mass. The accuracy ratio (%) is plotted above each bar. e Numbers of identified and unidentified lipids when considering only positive precursor ions (note that the majority of lipids do not contain more than one fragment in positive mode). Our knowledge base contains only one or two positive fragments for some lipids, therefore the percentage remains unchanged when choosing more fragments. f Numbers of identified and unidentified lipids in negative mode. g Total number of identified and unidentified lipids when taking both polarities into account. When considering three fragments, all target lipids could be identified and only one decoy was positively identified. Measurements were carried out on a pooled sample of five individual extraction experiments and were conducted in technical triplicates (n = 3 independent experimental replicates). Source data are provided as a Source Data file.

Fig. 6. Quantitative comparisons in the human plasma lipidome confirms LipidCreator output.

Lipid molecular species from major classes were quantified from the plasma samples of 21 healthy individuals (n = 21), and the NIST SRM 1950 reference material, by using transitions generated by LipidCreator. a Concentration differences of selected lipid species between healthy individuals (black dots) and average NIST 1950 plasma samples analyzed in this study (blue line). Each sample was measured with four independent technical replicates. 276 lipid species passed QC filters with a linear response R² > 0.8 and CV < 20%. Shaded areas represent standard deviation around the mean of the individuals and the mean of the NIST 1950 plasma samples. b The plasma lipid species concentrations across 22 lipid classes are displayed as blue bars. The sum of the concentrations of individual lipid species of the lipid classes are indicated as vertical thick red bars (right). Source data are provided as a Source Data file.

Fig. 7. Lipid regulation during human platelet activation.

a Network visualization of the lipid-lipid correlations. Nodes are lipid species. Node size represents the degree of connectivity, and node color represents the analyzed lipid class (see inset). Edges are correlations with r ≥ 0.85. b–e Color-coded nodes in the network show the lipid fold change with respect to resting platelets during activation by 0.01 U mL⁻¹ thrombin (0.01 Thr), 1 U mL⁻¹ thrombin (1 Thr), 1 µg mL⁻¹ collagen-related peptide (1 CRP) or 5 µg mL⁻¹ CRP (5 CRP); red indicates a fold change greater than or equal to 2, and olive green indicates a fold change less than or equal to 0.5. Data are combined from five independent biological experiments (n = 5), and mean values are shown. f Arachidonic acid (AA)-based mediator production and release upon stimulation. Bar graphs display the determined mediators in platelet cells. Bar graphs with diamand shape display the secreted mediators. The absolute quantities are reported in pmol mg⁻¹ protein. Error bars are presented as the standard deviation of the mean (n = 3 independent experimental replicates). 12-lipoxygenase (12-LOX). 15-lipoxygenase (15-LOX). cyclooxogenase (COX). phospholipase A2 (PLA₂). Unst: unstimulated. not detectable (N.D.). Source data are provided as a Source Data file.