Trapped ion mobility spectrometry (TIMS) and parallel accumulation - serial fragmentation (PASEF) enable in-depth lipidomics from minimal sample amounts

Lipids form a highly diverse group of biomolecules fulfilling central biological functions, ranging from structural components to intercellular signaling. Yet, a comprehensive characterization of the lipidome from limited starting material, for example in tissue biopsies, remains very challenging. Here, we develop a high-sensitivity lipidomics workflow based on nanoflow liquid chromatography and trapped ion mobility spectrometry. Taking advantage of the PASEF principle (Meier et al., PMID: 26538118), we fragmented on average nine precursors in each 100 ms TIMS scans, while maintaining the full mobility resolution of co-eluting isomers. The very high acquisition speed of about 100 Hz allowed us to obtain MS/MS spectra of the vast majority of detected isotope patterns for automated lipid identification. Analyzing 1 uL of human plasma, PASEF almost doubled the number of identified lipids over standard TIMS-MS/MS and allowed us to reduce the analysis time by a factor of three without loss of coverage. Our single-extraction workflow surpasses the plasma lipid coverage of extensive multi-step protocols in common lipid classes and achieves attomole sensitivity. Building on the high precision and accuracy of TIMS collisional cross section measurements (median CV 0.2%), we compiled 1,327 lipid CCS values from human plasma, mouse liver and human cancer cells. Our study establishes PASEF in lipid analysis and paves the way for sensitive, ion mobility-enhanced lipidomics in four dimensions.

very cell contains large amounts of lipids in various concentrations and chemical compositions 1,2 . Aberrant lipid homeostasis is a hallmark of many diseases, including cancer 3,4 and metabolic disorders 5 .
Therefore, analyzing lipidomes on a large scale promises novel insight into basic biology, as well as the onset and progression of disease [6][7][8] .
Lipid extracts from biological sources can be analyzed either directly via high-resolution mass spectrometry 9,10 or via online liquid chromatography (LC-MS) 8 . Lipids can be identified based on accurate mass and the MS 2 or MS 3 fragmentation pattern, which is facilitated by recent software developments and ever growing reference databases 11-14 . Established LC-MS lipidomics workflows separate lipids at flow rates in the higher micro-or milliliter per minute range, which ensures high sample throughput and robustness, but also compromises sensitivity.
As the available sample amount becomes a limiting factor, for example with small tissue sections from biobanks or small cell subpopulations, it is increasingly attractive to employ nanoflow chromatography [15][16][17] .
MS technology has greatly improved and state-of-the-art high-resolution Orbitrap or time-of-flight (TOF) instruments transmit ions very efficiently and achieve low-to sub-ppm mass accuracy 18,19 . The high acquisition speed of TOF analyzers makes them compatible with very fast separation techniques such as ion mobility spectrometry (IMS) 20,21 . Nested in-between LC and MS, IMS provides an additional dimension of separation based on the ions' shape and size (collisional cross section, CCS). This is particular interesting for lipidomics, as it provides an opportunity to separate otherwise unresolved isomers [22][23][24][25] . Furthermore, the chemical structure of lipids is closely linked to the CCS, which allows predictions by machine learning and could facilitate lipid identification [26][27][28][29] .  36 . In proteomics, PASEF increases MS/MS scan rates more than ten-fold, importantly, without the loss of sensitivity that is otherwise inherent to faster acquisition rates 37 .
Here, we explore whether the PASEF

Development of the nanoflow PASEF lipidomics workflow
We aimed to develop a rapid workflow that enables global lipid analysis in a straightforward manner (Fig. 1). Our lipid extraction protocol is applicable to common biological sample types, such as body fluids, tissue, as well as cell lines (Fig. 1a) and requires only a few manual liquid handling steps that could easily be automated in the future. We found that our extraction protocol scales well from small sample volumes (1 uL blood plasma) to relatively large cell counts (0.5 million HeLa cells) and can be performed in less than 1 hour.
We loaded the lipid extracts directly onto a C18 column and eluted them within 30 min, for a total of about one hour total analysis time per sample when using both positive and negative ionization modes (Fig. 1b).
Chromatographic peak widths were in the range of 3 to 6 s full width at half maximum (FWHM), at least two orders of magnitude slower than TIMS ion mobility analysis (100 ms) and the acquisition speed of highresolution TOF mass spectra (~100 µs  (Fig. 1b). We make use of this information in the postprocessing ( Fig. 1c)  In a 30 min analysis of plasma, we found that on average nine precursors were fragmented per PASEF scan (Fig. 2c)  We evaluated the performance of our PASEF method with lipid extracts from human plasma, mouse liver and HeLa cells ( Fig. 2d and Suppl. Fig. 1). In all three samples types,

Comprehensive and accurate lipid quantification
Having ascertained that PASEF achieves a very high MS/MS coverage of lipidomics samples, we investigated our automated data analysis pipeline in more detail (Fig. 3a).
Starting from the thousands of 4D features  Indeed, we observed that 408 out of the total 437 identified lipids in human plasma were quantified in four out of four replicates (Fig.   3b), resulting in a data completeness of 97%.
The median coefficient of variation (CV) was 9.2%, and 91% of all quantified lipids had a CV below 20% ( Fig. 3c and Suppl.   Fig. 4). Considering only identified lipids, we found a median CV of ion mobilities of 0.30% in repeated injections of the same sample (Fig. 5b) and, reassuringly, we achieved this very high precision also across commonly identified lipids from the three different biological samples ( Fig. 5c and Suppl.  comparison revealed a very high correlation (r = 0.993) and 95% of all values were within ±2% deviation centered at zero (Fig. 5d and Suppl.    Fig. 6a and Suppl. Fig. 4).
The investigation of the correlation of lipids mass and ion mobility has been a long term interest in ion mobility spectrometry-based lipidomics 26,27 . TIMS and PASEF provide a very efficient way to extend the scope of such studies to complex biological samples. To illustrate, we zoomed into narrow elution windows of the LC gradient (Figs. 6b and c).
In addition to coarse separation of lipid classes, there is a fine-structure within each of them. For example, extension of the acyl chain of phosphocholines (PCs) increases the CCS incrementally and results in clusters of lipids with the same acyl chain composition (Fig. 6b). Within each cluster, the lipids are differentiated by their degree of unsaturation as the addition of a double bond decreases the CCS value almost linearly. Similar trends can be derived for triacylglyerides (Fig. 6c) and other lipid classes (Suppl. Fig. 4).   Lipid CCS values were predicted in MetaboScape based on a support vector machine learning approach by Zhou et al. 27 .
Mass spectrometric metadata such as the PASEF frame MS/MS information were extracted from the .tdf files using the SQLite database viewer (SQLite Manager v3.10.1).