DDA‐imaging with structural identification of lipid molecules on an Orbitrap Velos Pro mass spectrometer

Abstract Matrix‐assisted laser desorption/ionization‐mass spectrometry imaging (MALDI‐MSI) is a useful technique for visualizing the spatial distribution of lipid molecules in tissues. Nevertheless, the use of MSI to investigate local lipid metabolic hallmarks has until recently been hampered by a lack of adequate technology that supports confident lipid identification. This limitation was recently mitigated by the development of DDA‐imaging technology where high‐resolution MSI is combined with parallel acquisition of lipid tandem MS2 spectra on a hybrid ion trap‐Orbitrap Elite mass spectrometer featuring a resolving power of 240,000 and a scan time of 1 s. Here, we report the key tenets related to successful transfer of the DDA‐imaging technology onto an Orbitrap Velos Pro instrument featuring a resolving power of 120,000 and a scan time of 2 s. Through meticulous performance assessments and method optimization, we tuned the DDA‐imaging method to be able to confidently identify 73 molecular lipid species in mouse brain sections and demonstrate that the performance of the technology is comparable with DDA‐imaging on the Orbitrap Elite. Altogether, our work shows that DDA‐imaging on the Orbitrap Velos Pro instrument can serve as a robust workhorse for lipid imaging in routine applications.


| INTRODUCTION
Matrix-assisted laser desorption mass spectrometry imaging (MALDI-MSI) is a useful analytical technique for imaging the spatial distribution of a wide range of biomolecules, including lipids. 1,2 Hundreds of putative lipid molecular ions in tissue sections can be visualized routinely and easily with a spatial resolution of 40 μm 3 and with extra effort these can be mapped to a spatial resolution of 1.4 μm. 4 MALDI-MSI technology has so far been used for lipid imaging in diverse biological systems, including individual cell culture cells, 5 tissues from animals, 6 microbes, 7 plants, 8 parasites, 9 and insects. 10 Nevertheless, a key challenge in lipid imaging is deciphering whether detected molecular ions are genuine lipid molecules or false-positive annotations related to other molecules with various adduct ions (e.g., H + , Na + , 37 K + , and 41 K + ), isotopologues, in-source fragmentation as well as chemical noise.
One way to mitigate false-positive lipid annotations is to use Fourier transform (FT) ion cyclotron resonance or Orbitrap mass spectrometry. 11,12 High-resolution FTMS 1 analysis affords accurate mapping of putative lipid molecular ions based solely on accurate m/z values and identification at the lipid species-level with notation of the total number of carbon atoms, double bonds and hydroxyl groups in all hydrocarbon chains. However, the high-resolution FTMS 1 analysis alone can still yield a high degree of false-positive lipid annotations, 13 representing "lipid molecules" that are unlikely based on known lipid metabolic pathways and that cannot be confirmed by lipid analysis of homogenized tissue extracts.
Another way to reduce false-positive lipid annotations is to combine the high-resolution FTMS 1 -based imaging with data-dependent (DDA) acquisition of lipid tandem MS 2 spectra at a single pixel-level. 14 In doing so, it becomes possible to identify lipids at the molecular species-level with notation of the number of carbon atoms, double bonds, and hydroxyl groups in individual hydrocarbon chains, and importantly, confirm whether a particular FTMS 1 -based image stem from a genuine lipid molecule or not. This DDA-imaging technology was recently established on a hybrid ion trap (IT)-Orbitrap Elite mass spectrometer, which due to its hardware design allows parallel recording of FTMS 1 and ITMS 2 spectra with a scan time around 1 s per pixel. 14 Analysis of the FTMS 1 and ITMS 2 data by the lipidomic software ALEX 12315 allowed high-confidence identification of 113 molecular lipid species in tissue sections of rat cerebellum. Notably, transferring the DDA-imaging technology to other instruments comes with an overhead of assessing and benchmarking the performance in reference to fixed, hardwarerelated specifications (e.g., number of detectors, resolving power) as well as tunable acquisition method-specific parameters (e.g., DDA settings).
Here, we outline the implementation and validation of DDAimaging technology on a hybrid IT-Orbitrap Velos Pro, which inherently features an inferior resolving power and scan time as compared with the Orbitrap Elite. Nevertheless, through systematic assessment of hardware-specific specifications, optimization of acquisition method-specific parameters, and benchmarking, we show that the performance of DDA-imaging is comparable with that of the Orbitrap Elite. Overall, our method validation shows that DDA-imaging on the Orbitrap Velos Pro can serve as a robust workhorse for lipid imaging in routine applications.

| Mouse brain section preparation
Animal experiments were conducted in accordance with German law (in congruence with 86/609/EEC) for the use of laboratory animals and approved by the local animal welfare committee at the Johannes Gutenberg University Mainz. Male C57BL/6 wild-type mice (8 weeks old) were euthanized by an overdose of ketamine by intraperitoneal injection. Subsequently, the mice were perfused intracardially with cold 155 mM ammonium acetate, and the brains were quickly removed, snap-frozen in liquid nitrogen and stored at À80 C until sectioned, as previously described. 16 For MALDI imaging experiments, frozen brain tissue was cut into 15 μm thick sections in a cryostat (CM 3050 S; Leica Microsystems, Nussloch, Germany), placed on a glass slide, and stored at À80 C until further processing.

| Matrix deposition for MALDI-MSI
Tissue sections were put in a desiccator for 20 min to minimize condensation of atmospheric water on their surfaces. For MSI in positive and negative ion mode, norharmane (7 mg/ml) in chloroform:methanol (7:3, v/v) was applied with an iMatrixSpray (Tardo, Switzerland). Spray conditions were as follows: height, 60 mm; line distance, 1 mm; speed, 180 mm/s; density, 1 μl/cm 2 ; cycles, 15; delay, 0 s; pressure, 1.6 bar. The sample stage of the MALDI source and the high-pressure ion funnel was maintained at 7.5-7.8 Torr. The low-pressure ion funnel was maintained at 1.6-1.7 Torr. Applied radio frequency voltages to the high-and low-pressure ion funnels were set to 590 kHz, 210 V 0-peak and 890 kHz, 80 V 0-peak , respectively. Ejection of MALDI-generated ions into the ion funnels was accomplished by an electric field gradient of $100 V/cm between the sample holder and the first ion funnel electrode.

| Lipid nomenclature
Lipids reported at the "species-level" are denoted by their lipid class abbreviations, followed by the total number of carbon atoms, double bonds and hydroxyl groups in all hydrocarbon chains. For example, "SM 36:1;2" denotes a SM lipid with a total of 36 carbon atoms, 1 double bond, and 2 hydroxyl groups in its two hydrocarbon chains. 18 Lipids reported at the "molecular species-level" are denoted by their lipid class abbreviations, followed by the number of carbon atoms, double bonds, and hydroxyl groups in each hydrocarbon chain.

| Lipid identification
Identification of lipid molecules was done using the ALEX 123 , as previously described. 14 Briefly, lipid precursor ions detected by FTMS 1 were identified using a m/z tolerance of ±0.0045 Da, corrected for potential 13 C isotope interference, required to have a relative detection frequency greater than 0.5 (equivalent to being detected in 50% of all DDA-imaging data files of identical polarity) and annotated at the "species-level." Lipid fragment ions detected by ITMS 2 were identified using a m/z tolerance of ±0.2 Da, required to have a relative detection frequency greater than 0.50 (equivalent to being detected in 50% of all DDA-imaging data files of identical polarity) and annotated as "molecular lipid species-specific fragments" (MLF) or "lipid class-specific fragments" (LCF). 19

| Imaging of tissue sections by optical microscopy
Optical microscopy was performed on a Nikon Eclipse Ti microscope (Nikon, Japan). Tissue sections were stained with hematoxylin and eosin (H&E), and images were acquired at 10Â magnification.  Figure 2A). Importantly, this performance is comparable with that of the Orbitrap Elite instrument, which exhibits almost identical calibration drifts, that is, 0.9 ppm and 0.8 ppm in positive and negative ion mode, respectively (Figure 2A).
To reduce these shorter term calibration drifts, we investigated the possibility of applying online FTMS lock mass calibration using the omnipresent dopant norharmane as a calibrant. This reduced the calibration F I G U R E 1 Key parameters that can influence the performance of DDA-imaging technology. The parameters can be grouped into those related to the hardware of the mass spectrometers, which cannot be modified, and those that are related to data acquisition, which to some extent can be modified. F I G U R E 3 Benchmarking lipid identifications obtained by DDA-imaging on the Orbitrap Velos Pro and the Orbitrap Elite. Confident lipid identifications were obtained from five positive and five negative DDA-imaging experiments of mouse brain carried out using the Orbitrap Velos Pro and three positive and three negative DDA-imaging experiments of rat brain using the Orbitrap Elite. Detected lipid molecules were identified using the ALEX 123 software. Venn diagram displays the lipidome overlap between the two DDA-imaging platforms. The 21 unique lipid identifications obtained using the Orbitrap Velos Pro were validated by high-resolution FTMS 2 analysis using an Orbitrap Fusion 20 (see Figure S1).  intensity (data not shown). We also found that it was best to use lipid standards, and not non-lipid calibration mixtures, for optimization of collision energy settings for DDA-ITMS 2 analysis (data not shown).
Another beneficial feature was to use a DDA-ITMS 2 inclusion list to emulate sequential parallel reaction monitoring of all precursor m/z values across an m/z range 400-2000 and within an ion isolation window of ±0.5 Da. 20 The benefit of this amendment is that that ITMS 2 scans will always be acquired at the same precursor m/z value (e.g., m/z 760.58) and prompt only one unique "scan header" in proprietary. RAW mass spectral data files instead of multiple practically identical "scan headers" (e.g., m/z 760.57, 760.58, and 760.59). Our approach thereby makes it is easier to manually inspect ITMS 2 spectra with the QualBrowser software and, more importantly, identical ITMS 2 scan events will be averaged together improving the signal-tonoise of detected fragment ions. Nevertheless, DDA-imaging on the Orbitrap Velos Pro was able to recapitulate 51 lipid identification previously identified using the Orbitrap Elite. Furthermore, the Orbitrap Velos Pro-based platform shortlisted 21 unique lipid identifications, which could be verified by manual inspection of high-resolution FTMS 2 spectra of mouse brain tissue recorded using an Orbitrap Fusion ( Figure S1). 20 Notably, majority of these lipid annotations are borderline identifications in the Orbitrap Elite dataset (Figure S1), rejected originally due to strict constraints.

| Benchmarking the DDA-imaging technology
These results, together with the assessments of instrument hardware, sensitivity and FTMS 1 calibration stability, demonstrate that the performance of DDA-imaging on the Orbitrap Velos Pro is inferior to that on the Orbitrap Elite instrument. Nevertheless, it appears to be relatively easy to transfer the DDA-imaging technology from one system to another.

| Reproducible DDA-imaging of brain lipids
Next, we assessed the performance of the DDA-imaging technology in terms of its ability to reproducibly generate images of lipid molecules across multiple tissue sections as required for making tissue F I G U R E 6 Reproducibility of DDA-imaging of SM species across multiple brain sections. (A) Ion images of protonated, sodiated and potassiated SM species across four distinct brain sections. (B) FTMS 1 intensity profile of SM species with different adduct ions across the four brain sections. atlases from consecutive sections. To this end, we first identified lipid molecules displaying the most pronounced differences in distribution across the mouse brain. Our aim was to utilize such molecular ions as markers to assess image reproducibility across consecutive tissue sections. To identify such lipid markers, we performed principal component analysis of representative FTMS 1 ion images in positive and negative ion mode (Figures 4A and S2A). This analysis shortlisted 22 distinct lipid molecules that display major differences in topology ( Figure S2B). These biomarkers are primarily glycerophospholipids, including phosphatidylcholine (PC) and phosphatidylethanolamine (PE) species, and sphingolipids, including sphingomyelin (SM), hexosylceramide (HexCer), and sulfatide (SHexCer) species as well as a putative ganglioside. Importantly, we found that images obtained by DDAimaging in both polarities showed a close correlation to the underlying mouse brain anatomy assessed by conventional H&E staining and optical microscopy ( Figure 4B).

| Reproducible DDA-imaging of glycerophospholipids
To examine more closely the reproducibility of the lipid imaging, we first inspected the topology of the shortlisted glycerophospholipid markers PC 36:1 and PC 40:6 across four horizontal brain sections.
Notably, ALEX 123 automatically identified these as the molecular lipid ( Figure 5B). This demonstrated that the DDA-imaging routine across multiple tissue sections produces highly reproducible profile of glycerophospholipid ionization, which underpin the utility of DDA-method as a robust tool for cross-sectional lipid imaging.

| Reproducible DDA-imaging of sphingolipids
We also examined the reproducibility of DDA-imaging of sphingolipids, given that these molecules generally yield less intense intensities as compared to more abundant glycerophospholipids. Among the sphingolipids with major differences in localization we identified four distinct SM molecules: SM 34:1;2, SM 36:1;2, SM 38:1;2, and SM 42:1;2. Notably, the parallel ITMS 2 analysis of these lipids only yielded detection of the confirmatory lipid class-specific phosphocholine fragment ion with m/z 184.1 and no long-chain base-specific fragments that allows inferring the structure of the ceramide-backbone (e.g., m/z 264.3). 19 Nevertheless, the topology of these SM species again demonstrated a good reproducibility across the four brain sections and for several adduct ions ( Figure 6A). Notably, we found that the most abundant SM 36:1;2 localizes to areas of gray matter throughout the brain whereas the three other SM species are more abundant in the choroid plexus. When assessing the FTMS 1 intensity distribution across all identified SM molecules we again observed a highly reproducible profile of ionization for protonated, sodiated as well as potassiated precursor ions ( Figure 6B).

| DDA-imaging of gangliosides
Our principal component analysis-based search for topology markers also yielded a prominent molecular ion with m/z 1544.8615 ( Figures 7A   and S2B). By manual inspection of the DDA-imaging data we elucidated that this ion is a deprotonated, singly charged ganglioside with the species-level annotation GM1 36:1;2 (detected with mass accuracy error of À3 ppm). This identification was further supported by the parallel ITMS 2 analysis of m/z 1544.9 showing multiple characteristic neutral loss fragments corresponding to the oligosaccharide-based head group structure ( Figure 7B). Unfortunately, no long-chain base-or FAspecific fragment ions were detected to provide insights into the molecular structure of the ceramide-backbone. Nevertheless, it is reasonable to predict that the molecular species is primarily GM1 18:1;2/18:0, since numerous reports have previously shown that majority of brain sphingolipids have a backbone composed of a C18 sphingosine with an amide-linked C18 acyl chain. 21 Finally, the ion image shows that GM1 36:1;2 is most abundant in the cerebral cortex, the hippocampus and the midbrain region of the mouse brain ( Figure 7A).

| CONCLUSIONS
Here, we outlined the successful transfer and implementation of DDAimaging technology on an Orbitrap Velos Pro machine and benchmarked its performance for lipidome imaging across multiple consecutive tissue sections. Overall, our report provides a guideline and insights into various technical issues that warrant special attention when implementing the technology on another type of mass spectrometer. More specifically, we found that DDA-imaging on an Orbitrap Velos Pro is comparable with that of the Orbitrap Elite in terms of easeof-operation, lipid image reproducibility and ability to shortlist differentially localized lipid molecules such as GM1, which was not identified in the original study by Ellis et al. 14 Moreover, we show that DDA-imaging on the Orbitrap Velos Pro was able to identify half of the confidently annotated molecular lipid species reported previously as well as another 21 lipid molecules that could be verified by high-resolution shotgun lipidomic analysis of mouse brain extracts and that were previously considered borderline identifications in the Orbitrap Elite dataset. Our work also indicated several drawbacks of using an Orbitrap Velos Pro, which includes an inherently lower resolution, slower scan rate, lower sensitivity and dynamic FTMS detection range, which concur with an overall lower lipidome coverage. We note that other technical parameters could in principle have affected the comparison of the two platforms.
These include the animal species used for making tissue sections (rat cerebellum was used for benchmarking the Orbitrap Elite-based platform), the thickness of tissue sections (herein we used 15 μm thick sections whereas the rat cerebellar tissue was 10 μm thick) and whether tissues are obtained following whole-body perfusion to remove blood contamination (as was done herein). To the best of our knowledge, these parameters are, however, of minor relevance for showcasing the overall performance metrics of DDA-imaging on the Orbitrap Velos Pro machine. In summary, based on our systematic performance assessment we conclude that Orbitrap Velos Pro serves as a robust tool for routine lipid imaging, covering a limited number of tissue sections, as well for more challenging and time-consuming endeavors such as making tissue atlases, which can entail analysis of hundreds of consecutive tissue sections.