We reasoned that novel lipid structures may arise from yet unknown combinations of already known modules and both known and novel combinations may be recognized in the same AIF LC-MS/MS analysis (Fig. 1). For presentation clarity, let us assume that anticipated lipid molecules may consist of two structural modules: a polar head group (H) and a fatty acid (F). To screen for new molecules, we will first design plausible structures by combining all fatty acid moieties f_j with all head groups h_i from the H and F lists, respectively. For each compound f_jh_i we compute m/z of its precursor ion [f_jh_i]⁺ and characteristic polar head group fragment [h_i]⁺ which would be most likely detectable in tandem mass spectra. AIF LC-MS/MS of a total extract detects all ionisable precursors (in MS spectra) and all fragments (in AIF MS/MS spectra). Chromatographic peaks of anticipated precursors (XIC [f_jh_i]⁺) and fragments (XIC [h_i]⁺) are then aligned by the in-house developed Arcadiate software. They should neatly overlap if [h_i]⁺ is produced from [f_jh_i]⁺ (see aligned XIC trace in Fig. 1A), while no or partial match indicates that this pair of ions is unrelated. The same scheme also applies for more complex fragmentation patterns since peaks alignment specificity increases if XICs of several fragments are required to match. A dataset produced by a single AIF LC-MS/MS experiment suffices for in silico screening for an unrestricted number of combinations of modules. If the structure of candidate molecules remains ambiguous, AIF LC-MS/MS should be followed by the targeted acquisition of full high resolution MS/MS spectra of their precursors, chemical derivatization¹¹ and, ultimately, by chemical synthesis.

We applied AIF LC-MS/MS to profile endogenous endocannabinoid-related compounds (ERC) in a total extract of rat kidney (Fig. 1A). ERC consist of two structural modules: a fatty acid or fatty alcohol linked to a polar head group (which determines its class), such as ethanolamine, glycerol, dopamine or amino acids (such as Ser or Gly). Bona fide endocannabinoids are physiological ligands of type 1 and type 2 cannabinoid receptors (CB₁R and CB₂R) and implicated in neurological, metabolic and cardiovascular diseases (reviewed in)¹². ERC are widespread across phyla¹³,¹⁴ and have a broad spectrum of activities¹⁵,¹⁶. They are present in fluids and tissues at the nM or picogram per mg range¹⁷,¹⁸, respectively, and therefore common LC-MS/MS analyses only target a few key molecules. ERC are structurally diverse, however no method afforded a comprehensive overview of their composition: although in total more than 80 ERC have been identified this list is certainly not exhaustive¹⁶,¹⁹,²⁰.

We first benchmarked our method by comparing the number of N-acylethanolamines (NAE) identified by AIF and, independently, by the two targeted LC-MS/MS methods¹⁷,²¹ – multiple reaction monitoring (MRM) on a triple quadrupole mass spectrometer and by parallel reaction monitoring (PRM) on a hybrid high resolution tandem mass spectrometer Q Exactive (Fig. 1B). Both MRM and PRM rely upon an inclusion list of targeted precursors compiled prior the analysis. Contrary, AIF profiling is unbiased since a full dataset of MS and MS/MS spectra is acquired first and then XIC traces for any desired combination of precursor and fragment masses are retrieved and aligned. For PRM and MRM analyses we compiled a list of m/z of 19 candidate NAE molecules that included species with common C16-C20 fatty acids with 0 to 6 double bonds; we also added two shorter saturated C12 and C14 fatty acids and three polyunsaturated C22 fatty acids with 4 to 6 double bonds (Supplementary Table S1). This list comprised all currently known NAE species and a few plausible candidates comprising common fatty acid moieties. Considering even larger number of candidate molecules could compromise the sensitivity of both reference methods, while having no impact on the sensitivity of AIF. In all three experiments NAE were identified by detecting the ethanolamine head group fragment (m/z 62.060) produced by MS/MS of 19 expected precursor ions (Supplementary Table S1). We note that, although MRM and PRM relied on the same mass transitions, they were performed independently on different instruments and were complementary. MRM was expected to have higher sensitivity, yet lower specificity in selecting precursor and fragment m/z, while PRM was performed at much higher mass resolution and therefore should be detecting precursors and fragments with higher specificity albeit lower sensitivity.

In a rat kidney extract AIF LC-MS/MS identified 15 NAE molecules, including 12 out of 13 molecules found by both MRM and PRM (Fig. 1C). In total, it recognized the same number of NAE molecules (15) as the method of MRM. It failed to detect NAE 22:6 because of background interference; this, however, was unrelated to the high degree of unsaturation of its fatty acid moiety. Hence the identification of endogenous NAE by the untargeted AIF and both targeted (MRM, PRM) LC-MS/MS methods was consistent.

We further hypothesized that ERC might feature a broader selection of polar head groups and fatty acids/alcohols, yet sharing the same modular organization. In principle, amino- and hydroxyl groups of any endogenous small molecule are plausible targets for N-acylation, O-esterification or -etherification, respectively. We therefore selected 52 polar head groups including amino acids, nucleobases and biogenic amines, each of which was conjugated in silico to 39 fatty acid/fatty alcohol moieties (Supplementary Dataset 1), altogether representing over 2000 putative molecules. A single AIF LC-MS/MS run of rat kidney extract recognized 132 ERC of 18 individual classes, out of which 74 molecules were novel (Supplementary Table S2). Furthermore, we identified 3 molecules of a novel class of ERC: N-acylaspartates (NAAsp) having 16:0, 18:2 and 20:4 fatty acid moieties attached to the amino group of aspartic acid (Fig. 2). Considering an extended list of uncommon fatty acids (e.g. having odd number of carbon atoms or a hydroxyl group) produced no further hits.

To further validate the identification these three NAAsp were synthesized. We found that their retention times and ratio of abundances of precursor and fragment peaks were the same as of endogenous molecules (Fig. 2A,B).

How abundant were NAAsp molecules compared to other endogenous ERC? We quantified NAAsp and four major ERC classes (1-acylglycerols, 2-acylglycerols, NAE and N-acylglycines)²² in the same extract. NAAsp was 2-fold less abundant than the least abundant endocannabinoid class of NAE (including anandamide) and 100-fold less abundant than 1-acylglycerols (Supplementary Table S3), which underscores the superior sensitivity of AIF LC-MS/MS.

We recently reported that anandamide (NAE 20:4) inhibits Hedgehog (Hh) signalling by binding the 7-pass transmembrane protein Smoothened¹⁵. To ask whether NAAsp may also interfere with Hh, we compared NAE 20:4 with NAAsp for their capacity to block both Sonic Hedgehog (Shh) and SAG-induced Smoothened activation (Fig. 2C). NAAsp 20:4 and NAAsp 18:2 were as active as anandamide in both assays. In contrast to anandamide, NAAsp do not bind CB₂R, only weakly bind CB₁R, and are resistant to major anandamide-degrading enzyme FAAH²³ (Fig. 2D). This makes NAAsp promising candidates for pharmacological targeting of the Hedgehog signalling pathway.