**Chasing the Major Sphingolipids on Earth: Automated Annotation of Plant Glycosyl Inositol Phospho Ceramides by Glycolipidomics**

**Lisa Panzenboeck 1, Nina Troppmair 1, Sara Schlachter 1, Gunda Koellensperger 1,2,3, Jürgen Hartler <sup>4</sup> and Evelyn Rampler 1,2,3,\***


Received: 20 August 2020; Accepted: 16 September 2020; Published: 19 September 2020

**Abstract:** Glycosyl inositol phospho ceramides (GIPCs) are the major sphingolipids on earth, as they account for a considerable fraction of the total lipids in plants and fungi, which in turn represent a large portion of the biomass on earth. Despite their obvious importance, GIPC analysis remains challenging due to the lack of commercial standards and automated annotation software. In this work, we introduce a novel GIPC glycolipidomics workflow based on reversed-phase ultra-high pressure liquid chromatography coupled to high-resolution mass spectrometry. For the first time, automated GIPC assignment was performed using the open-source software Lipid Data Analyzer (LDA), based on platform-independent decision rules. Four different plant samples (salad, spinach, raspberry, and strawberry) were analyzed and the results revealed 64 GIPCs based on accurate mass, characteristic MS2 fragments and matching retention times. Relative quantification using lactosyl ceramide for internal standardization revealed GIPC t18:1/h24:0 as the most abundant species in all plants. Depending on the plant sample, GIPCs contained mainly amine, N-acetylamine or hydroxyl residues. Most GIPCs revealed a Hex-HexA-IPC core and contained a ceramide part with a trihydroxylated t18:0 or a t18:1 long chain base and hydroxylated fatty acid chains ranging from 16 to 26 carbon atoms in length (h16:0–h26:0). Interestingly, four GIPCs containing t18:2 were observed in the raspberry sample, which was not reported so far. The presented workflow supports the characterization of different plant samples by automatic GIPC assignment, potentially leading to the identification of new GIPCs. For the first time, automated high-throughput profiling of these complex glycolipids is possible by liquid chromatography-high-resolution tandem mass spectrometry and subsequent automated glycolipid annotation based on decision rules.

**Keywords:** glycolipidomics; GIPC; glycosyl inositol phospho ceramides; Lipid Data Analyzer; lipidomics; sphingolipids; ultra-high pressure liquid chromatography; high-resolution mass spectrometry; LC-MS; automated annotation

#### **1. Introduction**

The sphingolipidome of plants contains glycosyl inositol phospho ceramides (GIPCs), glycosylceramides and ceramides, whereas sphingomyelin, globosides, sulfatides or gangliosides are absent. GIPCs were characterized as the major sphingolipid on earth due to their high abundance

in plants and fungi, which comprise a large portion of the biomass of the biosphere [1]. GIPCs were first described more than 60 years ago as "phytoglycolipids" [2]. The total plant lipid content can consist of up to 40% GIPCs [3]. The structure of these plant sphingolipids has three major subunits: (1) a polar inositol containing part, (2) the sphingoid backbone with a long-chain base (amino-alcohol) linked by an amide bond to a (3) fatty acyl chain moiety [2,4]. The terms d, t and q refer to the hydroxylation state of the whole ceramide or long-chain base (LCB) moiety, ranging from two (d) to four (q) hydroxy groups. The term h denotes a hydroxylation of the fatty acyl group (i.e., the ceramide moiety q40:1 can correspond to a t18:1 LCB connected to a h22:0 fatty acyl). Di- and trihydroxylation of LCBs with t18:0, t18:1(8Z and 8E) (the main sphingoid base in some species), and d18:0, d18:1(8Z and 8E), d18:2 (4E/8Z and 4E/8E) and fatty acid components varying in chain-length, saturation and hydroxylation state (h16:0–h26:1, 20:0 to 28:0) have been reported in plant GIPCs [5,6]. Different GIPC core structures were determined from higher plants ranging from simple high-abundant A-series species with Hex-HexA-IPC and HexN(Ac)-HexA-IPC (Hex = hexose, HexA = hexuronic acid, IPC = inositol phospho ceramide, HexN = hexosamine, and HexNAc = N-acetyl hexosamine) to low abundant F-series species containing several arabinoses and hexoses [3,7]. Despite the fact that GIPCs are an integral part of the plant plasma membrane, there is still little knowledge concerning its molecular organization and the way this organization is involved in signaling processes necessary for cellular adaptation [1]. To understand the interplay of GIPCs with different enzymes and their detailed function in the plasma membrane in plants, comprehensive structural information provided by observation tools such as NMR or MS are necessary.

Even though GIPCs were discovered 60 years ago, their analysis remains challenging due to the lack of available standards, automated annotation software and reference databases. For example, CHEBI [8] does not provide any GIPCs and the comprehensive LIPID MAPS Structure Database (LMSD) contains only one GIPC (A-NH2-t18:1/h24:0) [9]. As GIPCs consist of a sugar head group linked to a lipid subunit causing amphiphilic properties, they are neither well covered by common glycomics nor lipidomics workflows. Consequently, specialized glycolipidomics analysis strategies are required, e.g., applying a mixture of 2-propanol (IPA), hexane and water [10]. The combination of liquid chromatography and mass spectrometry (LC-MS) has been used due to its unprecedented potential to annotate GIPCs by *m*/*z*, retention time and fragmentation pattern [7,11]. Unambiguous GIPC identification requires both retention time evaluation and detection of structural subunits by tandem mass spectrometry (MS2), due to the absence of commercial standards. Most GIPC LC-MS-based analysis workflows were performed almost a decade ago by electrospray ionization followed by analysis with low resolution mass spectrometers (QQQ, QTRAP) [7,11]. Meanwhile, high-resolution mass spectrometers (such as TOF, orbitrap, FTICR) have been established with up to 1,000,000 resolution enabling GIPC analysis by accurate mass [12]. Additionally, ultra-high pressure liquid chromatography (up to 1500 bar) with sub 2-μm particles provides high chromatographic resolution and excellent sensitivity. Up to now, GIPC analysis has been performed by tedious manual annotation and curation [1,7,12,13] and expert knowledge was necessary to interpret glycosphingolipid tandem mass spectrometry fragmentation patterns [14–16]. The instrumentation advancements of the recent years paved the way for automated high-throughput GIPC analysis. In this work, a variety of plants, i.e., iceberg lettuce (*Lactuca sativa var. capitata nidus tenerimma*), deep frozen spinach (*Spinacia oleracea*), raspberries (*Rubus idaeus*), and strawberries (*Fragaria*) were analyzed by the combination of reversed-phase (RP) ultra-high pressure liquid chromatography (UHPLC) and high-resolution mass spectrometry (HRMS). For the first time, automated GIPC annotation will be performed using the Lipid Data Analyzer (LDA) and platform-independent decision rules [17].

#### **2. Results**

Here we describe a novel workflow by RP-HRMS/MS using the open-source program LDA [17] for automated GIPC assignment. Method development considerations and guidelines for the automated structural analysis of GIPCs are provided. Finally, we test the developed glycolipidomics workflow for different plant samples, leading to a reference database of GIPCs, including fragmentation and retention time information.

#### *2.1. Method Development for Automated GIPC Assignment*

GIPCs were extracted by a mixture of IPA, n-hexane and water [18]. So far, most LC-MS-based GIPC chromatographic separations relied on the use of tetrahydrofuran (THF) containing solvents [7,11–13]. However, the usage of THF in the eluent system has some drawbacks: (1) it is aprotic and cannot donate a proton; thus, for ionization, pairing with a protic solvent (usually water) is necessary; (2) it can attack tubing (especially PEEK tubing); (3) it tends to polymerize (usually in APCI mode); and (4) it is highly flammable. In order to avoid the use of THF, we developed a novel GIPC method based on RP-HRMS/MS, facilitating a 30 min isopropanol gradient (detailed information can be found in the Materials and Methods Section 4.3). GIPC detection was performed using both negative and positive electrospray ionization and high-resolution Orbitrap MS (see Materials and Methods Section 4.4). Importantly, GIPC analysis requires relatively high RF voltages (S-lens RF level of 45) to ensure efficient transport of medium size glycolipids in the mass spectrometer. Figure 1 shows the extracted ion chromatogram of GIPCs in salad samples analyzed by RP-HRMS, based on data-dependent MS2 (ddMS2) in positive and negative ion modes. The GIPCs displayed in Figure 1 belong to the A-series (Hex(R1)-HexA-IPC) with R1 being a hydroxyl group and the ceramide portion consisting of a hydroxylated saturated fatty acyl chain attached to a t18:1 long chain base.

**Figure 1.** Extracted ion chromatogram of glycosyl inositol phospho ceramides (GIPCs) in salad samples analyzed by RP-HRMS/MS analysis using ddMS2 in positive (red) and negative (blue) ion modes. Assigned GIPCs belong to the Hex-HexA-IPC series with a t18:1 long-chain base (LCB) and varying chain length of the hydroxylated saturated fatty acids. Retention times coincided in positive and negative ion modes. Increasing carbon numbers result in belated elution.

As no commercial standards are available, GIPC assignment has to be conducted with caution. In such a situation, the use of the equivalent carbon number model (ECN) is required [19,20]. The ECN model originates from state of the art lipidomics workflows and is based on elution orders observed in RP columns: (1) longer fatty acid chains will increase the retention time (see Figure 1) and (2) more double bonds will decrease the retention time [21] (see Table S1). To increase the level of confidence in GIPC annotation, we accepted only GIPCs that: (1) were detectable by accurate mass (±5 ppm) in MS1 at the same retention time in both positive and negative ion modes (Figure 1); (2) showed MS2 spectra with characteristic fragments for the ceramide and sugar part in at least one ion mode and; (3) fulfilled the ECN model.

#### *2.2. Structural Elucidation and GIPC Annotation Based on MS2 Information*

In this work, we introduce the first automated GIPC annotation workflow based on structural information provided by acquired MS2 spectra. Structural analysis and automated GIPC annotation was performed based on a set of in-house developed decision rules for the freely available software LDA [17,22]. As no standards were available, blank extractions (no GIPC annotations found) and GIPC annotations in salad [13] and spinach [12] reported in the literature were used to validate GIPC assignments (Figure 1, Table A1 and Table S2). Various LCBs (d18:0, d18:1, d18:2, t18:0, and t18:1) and fatty acids (FAs) (16–26) with or without hydroxylation have been reported [5,13]. Moreover, R1 in Figure 2A can either be a hydroxyl (OH), an amine (NH2) or an N-acetylamine (NAc) group, increasing the number of putative GIPCs even within a single series.

**Figure 2.** *Cont*.

**Figure 2.** Overview of the GIPC fragmentation for the example of GIPC A-OH-t18:1/h24:0 in salad: (**A**) The fragment assignment of GIPC A-OH-t18:1/h24:0 (adapted from [23]). The W fragment is shown in a light blue color. Please note that a full structural characterization is not possible by RP-HRMS/MS, (**B**) The product ion spectrum in negative ion mode at *m*/*z* 1260.7237, showing characteristic fragments *m*/*z* 241 and 259, 355, 373 and 417. The sugar head group was confirmed by the [C3PO3] − fragment (*m*/*z* 597, R1 = OH). [Z0PO3] <sup>−</sup> and [Y1-H]<sup>−</sup> fragments prove the ceramide moiety. (**C**) The positive ion mode ddMS2 spectrum of the [M + H]<sup>+</sup> precursor, exhibiting the [W]<sup>+</sup>, [W-H2O]<sup>+</sup> and [W-2H2O]<sup>+</sup> fragments at *m*/*z* 298, 280 and 262, which are characteristic for the t18:1 LCB.

The final decision rule set was based on well-defined fragments (fragment rules) and their intensity relationships (intensity rules) (Folder S1). The characteristic fragments [IP]<sup>−</sup> (*m*/*z* 259) and [IP-H2O]<sup>−</sup> (*m*/*z* 241) are mandatory in negative ion mode (e.g.: Figure 2B). However, these fragments are not specific, since they are produced by other phosphoinositol-containing lipids too. Thus, for a confident identification, negative or positive ion mode fragments indicating the sugar or ceramide part have to be detected.

In the majority of cases (see level 2 annotations, Table A1 and Table S2), MS2 spectra with GIPC fragmentation patterns were detected in both negative and positive mode. Depending on the fragmentation pattern and the level of confidence [24] of the structural elucidation, GIPCs are assigned as either: (1) series-R1-hydroxylation stage-carbon number (LCB + FA)-number of double bonds (LCB + FA) if the exact ceramide composition is not known or (2) series-R1-LCB/FA. Figure 2B displays an exemplary ddMS2 spectrum of A-OH-q42:1 with *m*/*z* 1260.7237, in salad recorded in negative ion mode. The positive ion mode fragmentation pattern of the [M + H]<sup>+</sup> precursor (*m*/*z* 1262.7389, Figure 2C) revealed further structural details, based on the identification of [W]<sup>+</sup>, [W-H2O]<sup>+</sup> and [W-2H2O]<sup>+</sup> fragments, indicating an A-OH-t18:1/h24:0 GIPC. Additional GIPC confirmation is possible by Z0 fragments ([Z0] <sup>+</sup>, [Z0-H2O]<sup>+</sup>) of the [M+ H]<sup>+</sup> precursor and by the sodium adduct [M+ Na]<sup>+</sup> (Figure A1), where sugar fragments are readily observable. GIPCs were annotated based on single ionization information only if (1) in negative ion mode in addition to the apparent [IP]−/[IP-H2O]−/[H2PO4] − fragments at *m*/*z* 259, 241 and 97, other characteristic fragments were detectable e.g., [C3PO3] − (*m*/*z* 596 − R1 = NH2, *m*/*z* 597 − R1 = OH, *m*/*z* 638 − R1 = NAc), [C3PO3-C1-CO2] − (*m*/*z* 373) or [C3PO3-C1-CO2-H2O]<sup>−</sup> (*m*/*z* 355) or (2) in positive ion mode the [IP]<sup>+</sup> (*m*/*z* 261)/[IP + Na]<sup>+</sup> (*m*/*z* 283) and fragments indicating the ceramide moiety (e.g., Z0) were identified by LDA. The detailed fragment information used for GIPC annotation can be found in Table S2.

GIPC annotation can be hampered by the presence of isobaric masses for qX:Y NH2 and t(X − 2):(Y − 1) NAc (where X refers to the carbon number (LCB + FA) and Y refers to the number of double bonds (LCB + FA), respectively). This may result in false positive GIPC identifications, because these classes share the same characteristic fragments *m*/*z* 241, 259, 355, 373 and 417. The correct structural elucidation is possible if additional fragments such as [C3PO3] <sup>−</sup> (R1 = OH − *m*/*z* 597, R1 = NH2 − *m*/*z* 596, R1 = NAc − *m*/*z* 638) in negative ion mode or if LCBs in positive ion mode can be identified based on [W]<sup>+</sup>, [W-H2O]<sup>+</sup> and [W-2H2O]<sup>+</sup> fragments. In the ddMS2 spectra of the [M + H]+-precursor, trihydroxylated LCBs are characterized by the presence of three W fragments ([W]<sup>+</sup>, [W-H2O]<sup>+</sup> and [W-2H2O]<sup>+</sup>), such as t18:0 (*m*/*z* 300, 282 and 264) and t18:1 (*m*/*z* 298, 280, 262), while dihydroxylated species miss the [W-2H2O]<sup>+</sup> fragment, e.g., d18:0 (*m*/*z* 284, 266), d18:1 (*m*/*z* 282, 264) and d18:2 (*m*/*z* 280, 262). As such, both LCB hydroxylation levels can be clearly distinguished.

#### *2.3. Analysis of Di*ff*erent Plant GIPCs by UHPLC-HRMS Suggesting t18:2 LCB*

The novel RP-HRMS/MS and GIPC annotation workflow was used to analyze different plant samples, namely salad (*Lactuca sativa var. capitata nidus tenerimma*), deep frozen spinach (*Spinacia oleracea*), raspberries (*Rubus idaeus*) and strawberries (*Fragaria*). As glycosphingolipid analysis is not negatively impacted by alkaline hydrolysis [10], alkaline hydrolysis was performed to simplify lipid profiles by removing the phospholipid background in the unknown plant samples (strawberry and raspberry, detailed information can be found in the Materials and Methods Section 4.2.2). Figure A2 shows the RP-HRMS/MS GIPC profile for the five most abundant GIPCs determined in spinach, strawberry and raspberry samples. For the sake of clarity, the five most abundant GIPCs in salad (A-NAc-t18:1/h24:0, A-NH2-t18:1/h24:0, A-OH-t18:1 h22:0 and h24:0, A-OH-t18:0/h24:0) are not displayed in Figure A2. Irrespective of the plant sample, the species group A-R1-t18:1/h24:0 was always the most abundant one. While in spinach R1 was always N-acetylamine (A-NAc-t18:1 h22:0 to h26:0) for the five dominating GIPCs, in strawberries the major GIPCs contained a hydroxyl group as R1 (A-OH-t18:1 h23:0 to h26:0 and A-OH-t18:0/h24:0). In contrast to that, raspberries had an amine group as R1 for four out of five shown GIPCs (A-NH2-t18:0/h24:0, A-NH2-t18:1 h22:0 and h24:0, A-NH2-t18:2/h24:0 and A-OH-t18:1/h24:0), emphasizing the structural diversity of GIPCs in different plants.

By analyzing different GIPCs, the NAc, NH2 and OH-species from the A series could be detected (Figure 3A–C) with high confidence by (1) accurate determination of mass, (2) matching retention times of ion modes, (3) characteristic fragments and (4) the ECN model. We recommend checking isotopic patterns to avoid false positive hits. For a comprehensive overview of the annotated GIPCs see Table A1.

Due to the absence of commercially available GIPC standards, relative quantification of the individual species was performed using C16 lactosyl(ß) ceramide (d18:1/16:0) as the internal standard. This compound is similar in structure (sugar and ceramide moiety) and retention time (14 min). Even though lactosyl ceramide (d18:1/16:0) may be present in plants, we could not detect it in our samples, thus, making it suitable as the internal standard in our workflow. Normalization by the internal standard (area ratio) and dry weight was performed for MS1-based relative quantification by Skyline [25] (Figure 3A–C). Estimated concentrations in the nmol to μmol range per gram dry weight were observed, which is consistent with the literature [12,18].

In summary, 64 GIPCs in salad (19), spinach (8), strawberry (10) and raspberry (27) were annotated (Table A1). Ranking of the GIPC annotations was performed according to the guidelines of the metabolomics society [24,26], leading to 48 level 2 (matching accurate masses and MS2 in negative and positive mode) GIPCs, 13 level 3 (MS2 in one ion mode with matching accurate masses in both ion modes) GIPCs and 3 level 3\*\* (matching accurate masses in both ion modes, MS2 in one ion mode but lacking information on IP fragments in positive ion mode or lacking sugar information in negative ion mode) GIPCs. The annotations found in spinach and salad are in accordance with literature [12,13]. To the best of our knowledge, this is the first report on GIPCs in strawberries and raspberries.

**Figure 3.** The normalized ratio per gram dry weight for annotated GIPCs in salad (light-green), spinach (green), strawberries (rose) and raspberries (dark-red), by using different substituents for the functional group R1: (**A**) NAc, (**B**) NH2, and (**C**) OH (more detailed information can be found in Table A1).

Within the annotated GIPCs with structural information on LCB and fatty acyl composition, t18:1 followed by t18:0 and t18:2 were the most prominent LCBs in the analyzed plants (Table A1). For GIPCs containing N-acetylamine residues t18:1 was the most abundant LCB with regard to normalized ratios per gram dry weight (Figure 3A). The same holds true for the amine or hydroxyl group containing GIPCs, with additional high abundance of t18:0 LCBs (Figure 3B,C). While in spinach solely t18:1 LCBs were detected, salad, strawberries and raspberries show more variation in terms of LCB composition with presence of both t18:1 and t18:0 (Table A1).

Interestingly, besides the expected t18:0 and t18:1 LCBs (R1 = NAc, NH2, OH), we additionally annotated four t18:2 (R1 = NAc, NH2, OH) species in raspberries (Figure 3A–C). These annotations are verified by coinciding retention times in positive ion mode (Figure A3A), detection of characteristic fragments in MS2 spectra (Figure A3B,C) and conformity with the ECN model (Table S1). However, we could not find any report in the literature of t18:2 species, which can be explained as up to now no automated GIPC annotation was possible and t18:2 GIPC species were only detected in raspberries. As no standards are available, it is difficult to prove the presence of this species and further investigation is needed. A confirmed t18:2 LCB would indicate a much higher diversity in sphingolipids than anticipated in the past. Another hint for the complex nature of GIPCs in raspberries is the additional annotation of GIPCs with di- and trihydroxylated variants compared to all other analyzed plants.

Concerning LCB and fatty acyl combinations, t18:1/h22:0 and t18:1/h24:0 showed equal annotation numbers for N-acetylamine or amine containing GIPCs (Figure 3A,B). Independent of the NH2, OH or NAc functional group, only two odd chain fatty acids (h23:0, h25:0) were detected and no fatty acids with a length from 17 to 21 carbon atoms were found. For the hydroxyl group containing GIPC variants, the combinations t18:1/h22:0, t18:1/h23:0, t18:0/h24:0, t18:1/h24:0 and t18:1/h25:0 were found in equal annotation numbers. Overall, plant GIPCs with a combination of t18:1 LCB and a h24:0 fatty acyl moiety were the most abundant ones in terms of normalized ratios per gram dry weight (Figure 3A–C, Table A1).

#### **3. Discussion**

GIPCs are the major sphingolipids on earth [1]. Hence, it is important to understand their function and distribution in plants and fungi. However, GIPC analysis remains extremely challenging, as tailored extraction strategies for this glycolipid class are necessary. GIPC analysis is in its infancy due to the lack of standards and databases. In this work, we present the first automated high-throughput GIPC annotation workflow which is based on RP-HRMS/MS. By using a novel 30 min gradient based on isopropanol with a reversed-phase column, packed with sub 2-μm particles, fast GIPC analysis was possible at the same time avoiding standard eluent use of tetrahydrofuran. Four different plant samples were analyzed. For salad and spinach, literature information has been available [12,13], while for raspberry and strawberry, GIPC profiles were completely uncharacterized. Using strict filtering by (1) accurate mass determination (±5 ppm) with matching retention times for both ion modes in MS1, (2) MS2 spectra with characteristic fragments and (3) expected retention time series, we produced a database of 64 GIPCs (Table A1). As no GIPC standards are available, only GIPC annotation hits with level 2 and 3 confidence [24] were possible. The most prominent MS2 fragments for GIPCs are [IP] fragments in both ion modes ([H]−: *m*/*z* 241, 259; [H]+: *m*/*z* 261; [Na]+: *m*/*z* 283). However, additional sugar or ceramide fragments are essential for correct GIPC annotation. The high MS2 mass range coverage (*m*/*z* 65 to 2500) provided by the Orbitrap was beneficial to determine GIPC low mass fragments such as *m*/*z* 79 [PO3] <sup>−</sup> or 97 [H2PO4] <sup>−</sup>, besides high mass precursors such as 1261 [M − H]<sup>−</sup> (Figure 2).

Relative quantification with the internal standard lactosyl ceramide revealed GIPC t18:1/h24:0 as the most abundant species, independent of the plant sample. Depending on the plant sample, GIPCs contained mainly amine, N-acetyl or hydroxyl residues. Most GIPCs showed a Hex-HexA-IPC core with a trihydroxylated t18:0 or t18:1 long-chain base ceramide part a and hydroxylated fatty acid chains ranging from h16:0 to h26:0. Interestingly, in raspberry, four GIPCs contained t18:2, which was not reported so far. This finding would suggest the existence of more complex sphingolipid species in nature than previously anticipated. Further analysis by orthogonal methods such as NMR, GC-MS or IMS and available GIPC standards would be necessary to confirm the presence of the t18:2 GIPC group. Different analytical strategies could also resolve potential isomeric species and provide comprehensive details on the sugar moiety present in GIPCs. Nevertheless, this example shows the power of this workflow to detect promising novel GIPC candidates in an automated fashion. In order to support LC-MS-based GIPC analysis in general, we provide the mass lists for GIPCs in positive and negative ion modes (Tables S3 and S4), as well as the fragmentation rules (Folder S1) for setting up the automated GIPC analysis by Lipid Data Analyzer. Even though we confirmed the GIPCs exclusively from the A-series, the presented strategy is also suitable to determine less or more complex GIPC series, such as 0, B, C, D, E and F. However, extended analytical workflows (e.g., multi-stage fragmentation/MSn) and additional software method development might be necessary. Precursor mass lists for positive ([M + H]+) and negative ([M <sup>−</sup> H]−) ion modes comprising series 0–F, LCBs d18:0, d18:1, d18:2, t18:0 and t18:1 and fatty acyls h15:0–h26:0, h15:1–h26:1 and n20:0–n28:0 (n = non-hydroxylated), as reported in the literature [5,13], can be found in Tables S5 and S6. In general, we believe that LC-HRMS/MSn combined with automated annotation based on decision rules will pave the way for more complex glycolipidomics profiling.

#### **4. Materials and Methods**

#### *4.1. Material*

The plant material used was derived from salad (*Lactuca sativa var. capitata nidus tenerimma*), deep frozen spinach (*Spinacia oleracea*), raspberries (*Rubus idaeus*) and strawberries (*Fragaria*). (A more detailed description of plant samples can be found in Table A2.)

All chemicals were of LC-MS grade. Acetonitrile (ACN), methanol (MeOH), IPA and water were bought from Honeywell (Offenbach, Germany) and n-hexane was bought from VWR (Vienna, Austria). Butylated hydroxytoluene (BHT) was purchased from Sigma-Aldrich (Vienna, Austria), ammonium formate (AF) from Sigma-Aldrich (Vienna, Austria) and formic acid from VWR (Vienna, Austria). C16 Lactosyl(ß) Ceramide (d18:1/16:0) (D-lactosyl-ß-1,1' N-palmitoyl-D-erythro-sphingosine) was purchased from Avanti Polar Lipids, Inc. (Alabaster, Albama, USA), was used as internal standard (IS) and dissolved in an appropriate amount of IPA to achieve a concentration of 100 μM.

#### *4.2. Sample Preparation*

Salad was manually cut into small pieces before being weighed into falcon tubes (50 mL, VWR, Vienna, Austria) using a CPA225D balance (Sartorius, Vienna, Austria). Raspberries and strawberries (whole fruits) were homogenized with a hand blender (Tefal/SEB, Ecully, France). Raspberries, strawberries and deep-frozen homogenized spinach were directly weighed into 10 mL glass vials (more details can be found in Table A2). In order to prevent potential oxidation of lipids, 3 mL of an approximately 0.01% BHT solution in IPA were added and samples were mixed. Subsequently 30 μL IS were spiked into all samples except for one replicate (to test for potential IS presence in plants). Salad samples were homogenized using an ultra-turax (miccra d-1, Heitersheim, Germany) which was cleaned with 70% IPA and dried between the samples. In order to inhibit lipase activity, all samples were incubated at 75 ◦C for 30 min under constant shaking [27]. The warm salad samples were subsequently transferred into glass vials. The following sections provide a detailed overview of the extraction strategies that were applied.

#### 4.2.1. One-Phase Extraction

The extraction of GIPCs from salad and spinach was performed as previously reported [18] using a mixture of IPA, n-hexane and water. Amounts of 3.47 mL IPA, 0.6 mL n-hexane and 1.93 mL water were added to the salad and spinach samples. In order to ensure sufficient accessibility of the plant material, samples were vortexed and manually shaken prior to incubation at 60 ◦C for 15 min under constant shaking.

#### 4.2.2. One-Phase Extraction Combined with Alkaline Hydrolysis

To avoid the occurrence of glycerophospholipids, which might reduce GIPC ionization efficiency and lead to potential false identifications, alkaline hydrolysis was applied for the raspberry and strawberry samples, using an adapted workflow [28]. After incubating the plant material with the BHT solution for 30 min at 75 ◦C under constant shaking, 3.47 mL IPA and 0.6 mL n-hexane were added. Samples were vortexed and put on a shaker for 15 min at 60 ◦C. As soon as the samples had reached room temperature 707 μL 1 M KOH in MeOH was added and the solution was vortexed. After shaking the samples for 2 h at 37 ◦C, they were left at room temperature. Subsequently 100% formic acid was added until a pH of ~6–7 was reached and 1.93 mL water was added before repeating the incubation step.

#### 4.2.3. Centrifugation, Drying and Reconstitution

Irrespective of the extraction strategy, the warm samples were centrifuged at 1000 rpm for 10 min at 4 ◦C and the supernatant was transferred into a separate glass vial. The solvent was evaporated to dryness overnight in a Genevac EZ-2 Series Personal Evaporator (SP Scientific, Ipswich, UK) and the dried residue was reconstituted in 2 mL IPA:H2O (65:35) [13]. Samples were vortexed prior and after ultrasonication at 30 ◦C for 15 min. Subsequently, 500 μL of this solution was filtered directly into HPLC vials through a ClariStep filter (Sartorius, Vienna, Austria). Pools were prepared separately for each plant by pipetting 50 μL of each biological replicate into a separate HPLC vial. A quality control pool was prepared by combining 30 μL of the pooled samples.

#### *4.3. Reversed-Phase Chromatography*

Liquid chromatography was performed using a C18 Acquity UHPLC HSS T3 reversed phase column (2.1 × 150 mm, 100 Å, 1.8 μm, Waters, Vienna, Austria) equipped with a VanGuard Pre-column (2.1 × 5 mm, 100 Å, 1.8 μm, Waters, Vienna, Austria) at a column temperature of 40 ◦C. The flow rate was 0.25 mL/min and the backpressure was 460 bar at the starting conditions. Gradient elution with a total runtime of 30 min was performed using the solvent A: ACN:H2O (3:2, *v*/*v*) and the solvent B: IPA:ACN (9:1, *v*/*v*), both of which contained 0.1% formic acid and 10 mM ammonium formate.

The gradient can be described as follows: 0–2 min 30% B, 2–3 min ramp to 55% B, 3–17 min ramp to 67% B, 17–22 min ramp to 100% B, 22–26 min 100% B, followed by an equilibration step from 26 to 30 min using 30% B. A Vanquish Duo UHPLC system (Thermo Fisher Scientific, Germering, Germany) was used and injections were performed with an autosampler. An injection volume of 10 μL was chosen and the injector needle was flushed with 75% IPA and 1% formic acid in between the injections.

#### *4.4. High-Resolution Mass Spectrometry*

The LC system was coupled to a Q Exactive HF (Thermo Fisher Scientific, Bremen, Germany) high resolution mass spectrometer, applying a HESI ion source with an S-lens RF level of 45. Measurements were carried out in positive and negative modes using different parameters. The following settings were applied in positive mode: spray voltage: 3.5 kV, capillary temperature 220 ◦C, sheath gas flow rate: 30, and auxiliary flow rate: 5. In negative mode parameters were adapted as follows: spray voltage: 2.8 kV, capillary temperature 250 ◦C, sheath gas flow rate: 35 (a.u.), and auxiliary flow rate: 10 (a.u.). The top 10 data-dependent MS2 spectra were obtained at a scan range of 500 to 3000 *m*/*z* with HCD using normalized collision energies of 35 (+35 in positive ion mode, −35 in negative mode), an MS1 resolution of 15,000 or 30,000 with an AGC target of 1e6 and MS2 resolution of 15,000 with an AGC target of 1e5. MS2 spectra were acquired based on an inclusion list ("do not pick others" option) containing the GIPC series 0–F (*m*/*z* values were calculated using enviPat Web 2.4 [29]). A more

comprehensive picture of the GIPC composition of the analyzed plant material was obtained using several rounds of automatically generated exclusions lists for the sample pools [30].

#### *4.5. Data Analysis*

The GIPC assignment was performed using LDA (version 2.8.0) [17]; corresponding settings (Table A3), mass lists (Tables S3 and S4) and decision rule sets for series A (Folder S1) can be found in the Appendix A and Supplementary Materials. The correct GIPC annotation was ensured by a manual inspection of the results. MS1-based relative quantification of annotated GIPCs was performed with Skyline [25]. Total areas were divided by the corresponding calculated dry weights and areas of the IS, resulting in normalized ratios per g dry weight, of which the average was taken based on the number of replicates (3 for salad and spinach, 4 for strawberries and raspberries). More information can be found in Appendix B.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2218-1989/10/9/375/s1, Table S1. Application of the ECN model to ensure correct GIPC annotation, Table S2. List of annotated GIPCs including all detected MS2 fragments, Table S3. LDA mass list used for automated annotation of the series A GIPCs in positive mode, Table S4. LDA mass list used for automated annotation of the series A GIPCs in negative mode, Table S5. List of [M <sup>+</sup> H]<sup>+</sup> precursors comprising the GIPC series 0–F, Table S6. List of [M <sup>−</sup> H]<sup>−</sup> precursors comprising the GIPC series 0–F, Folder S1. Fragmentation rules for GIPC analysis by LDA.

**Author Contributions:** Conceptualization, E.R.; methodology, L.P. and E.R.; software, L.P., S.S., N.T. and J.H.; validation, L.P., E.R. and J.H.; formal analysis, L.P., S.S., N.T. and E.R.; investigation, L.P. and E.R.; resources, E.R. and G.K.; data curation, L.P. and E.R.; writing—original draft preparation, E.R. and L.P.; writing—review and editing, E.R., L.P., J.H. and G.K.; visualization, L.P.; supervision, E.R.; project administration, E.R. and G.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** This work was supported by the University of Vienna, the Faculty of Chemistry, the Vienna Metabolomics Center (VIME; http://metabolomics.univie.ac.at/), the research platform Chemistry Meets Microbiology and the Mass Spectrometry Centre of the University of Vienna. The authors thank (1) the Department of Food Chemistry and Toxicology (University of Vienna) for sharing their sample preparation equipment, (2) Sophia Mundigler for the preparation of a literature database for glycolipid fragmentation (3) Martin Schaier for graphical abstract support, as well as (4) all members of the Koellensperger lab (University of Vienna) for continuous support.

**Conflicts of Interest:** The authors declare no conflict of interest.


**Table A1.** Overview of GIPCs annotated in salad, spinach, strawberries and raspberries. The precursor ion *m*/*z* and retention times are listed as provided by the LDA display results function. In cases where only matching retention times but no *m*/*z* were explicitly shown by the LDA, because only one MS2 spectrum was annotated (level 3, level 3\*\*), corresponding values (marked with an asterisk\*) were manually assigned at the peak maximum using Thermo Scientific FreeStyle. In the Level column the levels of identification are listed. For all annotated GIPCs, accurate mass and retention times were observed. Their level of identification depends on the ddMS2 spectra. Level 2 annotation is based on the ddMS2 spectra in both ion modes. For level 3, ddMS2 spectra with characteristic fragments could only be detected in one ion mode. Putative hits, which cannot be annotated with such high confidence, because they were only observed with a sugar fragment in positive mode, but did not show the [IP]+/[IP+Na]+ fragment, are listed as level 3\*\* at the end of the table and were not included in Figure 3.




**Figure A1.** The ddMS2 spectrum of the [M + Na]<sup>+</sup> of GIPC A-OH-t18:1/h24:0 (*m*/*z* 1284.7193, Rt 17.30 min), measured in positive ion mode, showing the characteristic [IP + Na]<sup>+</sup> and additional sugar fragments.

**Figure A2.** Comparison of the RP-HRMS/MS GIPC profiles in spinach (green), strawberry (rose) and raspberry (dark-red), showing the five most abundant GIPCs found in each plant sample measured in positive ion mode (detailed information can be found in Table A1).

**Figure A3.** *Cont*.

**Figure A3.** (**A**) Extracted ion chromatograms of GIPCs in positive ion mode, having a t18:2 LCB annotation on levels 2 and 3 in raspberries (A-NAc-t18:2/h24:0, A-NH2-t18:2 h22:0 and h24:0, as well as A-OH-t18:2/h24:0). The ddMS2 spectra of GIPCs A-NH2-t18:2/h24:0 at the retenion time of 16.1 min (**B**) A-OH-t18:2/h24:0 at the retention time of 15.2 min and (**C**) in negative and positive mode, showing characteristic fragments.

**Table A2.** The description of the plant samples, including plant species, origin, number of biological replicates, and the average fresh- and dry weights [g]. The extraction of GIPCs from strawberries and raspberries was performed one day after the collection was performed (28 June 2020).



**Table A3.** The exemplary LDA parameters and settings used for automated GIPC annotation in negative ion mode.


**Table A3.** *Cont*.

#### **Appendix B**

Automated GIPC annotation was performed using LDA (version 2.8.0) [17] with the settings provided in Table A3. The mass-to-charge ratios included in the mass lists (see Tables S3 and S4) were calculated separately for negative and positive ion modes, with enviPat Web 2.4 [29] and decision rules (see Folder S1) were created based on fragments reported in the literature [12,13]. Please note that the raw data acquired in negative ion mode has to be analyzed using the mass list of Table S4 and the fragmentation rules ending with '-H.frag', while for positive mode the mass list of Table S3 and corresponding fragmentation rules ('H.frag' and 'Na.frag') should be used. Further information on working with the LDA can be found in [31].

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Plasma Lipid Profiling of Three Types of Drug-Induced Liver Injury in Japanese Patients: A Preliminary Study**

**Kosuke Saito 1, Tatehiro Kagawa 2, Keiji Tsuji 3, Yuji Kumagai 4, Ken Sato 5, Shotaro Sakisaka 6, Naoya Sakamoto 7, Mitsuhiko Aiso 8, Shunji Hirose 2, Nami Mori 3, Rieko Tanaka 4, Toshio Uraoka 5, Kazuhide Takata 6, Koji Ogawa 7, Kazuhiko Mori 9, Motonobu Sato 10, Takayoshi Nishiya 11, Kazuhiko Takamatsu 10, Noriaki Arakawa 1, Takashi Izumi 12, Yasuo Ohno 12, Yoshiro Saito 1,\* and Hajime Takikawa 8,13**


Received: 16 July 2020; Accepted: 28 August 2020; Published: 31 August 2020

**Abstract:** Drug-induced liver injury (DILI) is a major adverse event caused by drug treatment, which can be categorized into three types: hepatocellular, mixed, and cholestatic. Although nearly every class of drugs can cause DILI, an overall understanding of lipid profiles in DILI patients is lacking. We used lipidomics to analyze the plasma lipid profiles of patients to understand their hepatic pathophysiology and identify DILI biomarkers. We identified 463 lipids and compared their levels between the acute and recovery phases of the three types of DILI patients. Mixed and cholestatic types demonstrated specific plasma lipid alterations between the phases, but the hepatocellular type did not. Moreover, as specific indicators of mixed-type DILI, levels of several ceramides increased in the acute phase, while those of arachidonic acid-containing ether-linked phosphoglycerolipids decreased. In contrast, as specific indicators of cholestatic-type DILI, levels of palmitic acid-containing saturated or monounsaturated phosphatidylcholines increased in the acute phase, while those of arachidonic acidor docosahexaenoic acid-containing ether-linked phosphoglycerolipids and phosphatidylinositols

decreased. We also identified lipids with a relatively high capacity to discriminate the acute phase from the recovery phase and healthy subjects. These findings may help with understanding the pathophysiology of different DILI types and identify candidate biomarkers.

**Keywords:** lipidomics; drug-induced liver injury; biomarker; plasma lipid profiles

#### **1. Introduction**

Drug-induced liver injury (DILI) is a major adverse event caused by drug treatment and is the most frequent cause of acute liver failure in the U.S. [1,2]. Depending on the histological location of the tissue damage, DILI is categorized as hepatocellular, cholestatic, or mixed type, which is usually based on changes in blood levels of alanine transaminase (ALT) and alkaline phosphatase (ALP). The causal relationship between DILI and suspected drugs has been digitized by the CIOMS/RUCAM and DDW-J2004 scoring scales (in Japan), which are used in clinical practice [3–5]. The mechanisms of DILI are diverse and include direct toxicity by the administered drug or its metabolites and immune reactions against the drug or its metabolites [6,7]. The most studied drug causing DILI is acetaminophen, which is metabolized to a toxic and electrophilic intermediate by cytochrome P450 isoenzymes (such as CYP2E1 and CYP3A4); this intermediate interacts with intracellular proteins resulting in hepatocyte damage [8]. Although specific mechanisms of drugs with relatively high incidence of DILI have also been studied [6,7], nearly every class of drug can cause DILI. However, biomarkers and characteristics of DILI that are important to understand its pathophysiology are limited.

Lipids, such as phosphoglycerolipids, sphingolipids, and neutral lipids, are components of cellular membranes that also play important roles in multiple biological processes, including apoptosis, inflammation, proliferation, and differentiation [9–12]. The liver is a central organ in regulating lipid levels, and therefore, aberrations in lipid homeostasis are associated with hepatic injury and disease. In addition, a recent study demonstrated that the composition of plasma lipids correlates well with that of hepatic lipids [13]. Thus, plasma lipid profiles could be useful tools to understand the biological processes in the liver. To analyze plasma lipid profiles, lipidomics based on mass spectrometry has been established [14–17]. Plasma lipidomics has already been used to study hepatocellular carcinoma [18,19], liver phospholipidosis [20], nonalcoholic fatty liver disease [21], and other hepatic diseases and toxicities. For example, plasma lipidomics of hepatocellular carcinoma demonstrated decreased levels of lysophosphatidylcholine (LPC) in plasma, suggesting the hepatic activation of autotoxin and its involvement in hepatocarcinogenesis [22]. Moreover, plasma lipidomics of liver phospholipidosis demonstrated increased levels of d18:1/24:0 glucosylceramide (GluCer), which was proposed as a biomarker for the disease [20]. Therefore, the characterization of overall plasma lipid profiles could lead to a better understanding of hepatic pathophysiology and identify new DILI biomarkers.

In this study, we aimed to analyze the differences in lipid profiles among three DILI types (hepatocellular, mixed, and cholestatic) during acute and recovery phases in human patients. We present novel lipidomic data for the three injury types, which could be used to screen for DILI biomarkers and/or develop future novel therapies by understanding lipid homeostasis in DILI.

#### **2. Results**

#### *2.1. DILI Patients Recruited in the Present Study*

We recruited 54 DILI patients, comprising 33 hepatocellular, 13 mixed, and 8 cholestatic types (Table 1). Of these patients, 11 males and 22 females were diagnosed with hepatocellular type, 9 males and 4 females were diagnosed with mixed type, and 4 males and 4 females were diagnosed with cholestatic type. Their median ages were 56, 60, and 69 for the hepatocellular, mixed, and cholestatic types, respectively. The median CIOMS/RUCAM scores were eight, nine, and eight for the hepatocellular, mixed, and cholestatic types, respectively. In addition, the median DDW-J2004 scores were eight for each of the respective DILI patient types. The causal relationship between suspected drug and liver damage was definite in all patients using the CIOMS/RUCAM scale and all patients using the DDW-J 2004 score, except for one case.


AST; aspartate transaminase, ALT; alanine transaminase, ALP; alkaline phosphatase, T. Bil; total bilirubin, ad.; administrated. The reference ranges of the liver blood test were <30 for AST, <30 for ALT, 100–325 for ALP, and 0.2–1.2 for T. Bil. The threshold numbers of patients in "Suspected drugs" and "ATC level 2 of suspected drugs" were judged by the sum of all types of drug-induced liver injury (DILI).

The suspected drug with the highest frequency of culpability was loxoprofen, which was responsible for four cases out of the 54 patients (two, one, and one case in the hepatocellular-, mixed-, and cholestatic-type patients, respectively). In addition, when the prescribed drugs were categorized according to the World Health Organization (WHO) Anatomical Therapeutic Chemical (ATC) codes, the highest number of cases was found in antibacterial agents for systemic use (J01, eight cases), followed by antineoplastic agents (L01, six cases), anti-inflammatory and antirheumatic products (M01, six cases), and psycholeptics (N05, six cases). The causes of DILI were widely diverse among cases, which hindered the analysis of drug-specific or drug category-specific effects.

#### *2.2. Global Plasma Lipid Profiling in the Three DILI Types*

Global plasma lipidomic profiling using our lipidomics platform detected 463 lipids spanning 31 lipid classes (Table S1 and summarized in Table 2). Note that in our assay platform, unconjugated bile acids were detectable but not quantitative because their liquid chromatography (LC) retention time is close to the void fraction where ionization is unstable due to the presence of unretained salts. The exemplar LC/MS traces are shown in Figure S1. To distinguish stereoisomers, each quantified lipid was assigned a specific metabolite ID. The fatty acid side chains in the lipids were confirmed

using mass spectrometry (MS), the confirmed fatty acid fragments were indicated after a semicolon in the name. The combination of fatty acid side chains was combined using a slash. If two different sets of fragments were confirmed, we provided both of them, separated by a comma. The identified lipids comprised 184 phospholipids, 80 sphingolipids, 180 neutral lipids, and 19 others, including coenzyme Q10 (CoQ10), free fatty acids (FAs), and acylcarnitines (Cars). The major phospholipid class, phosphatidylcholines (PCs), contained 56 lipids. In addition, the major sphingolipid class, sphingomyelins (SMs), comprised 37 lipids, and the major neutral lipid class, triacylglycerols (TGs), comprised 138 lipids. The identified lipid levels were compared between the acute and recovery phases in each DILI type. Lipids with both high effect sizes (g > 0.8) and statistically significant differences (*p* < 0.05) were defined as altered.


**Table 2.** Identified lipid classes and numbers of individual lipids.

Although 112 lipids were significantly different between the phases, no lipid was defined as altered in the hepatocellular type (Figure 1a). In contrast, 9 and 20 lipids were defined as altered in the mixed and cholestatic types, respectively (Figure 1b,c). Thus, we focused on the mixed and cholestatic types for further analysis.

**Figure 1.** Volcano plot of lipid alterations in three types of DILI. Statistical probability (*p* value) and effect size (g) were determined by a comparison of lipid levels between acute phase and recovery phase of the DILI patients. Volcano plots show -log *p* value versus g value for (**a**) hepatocellular type, (**b**) mixed type, and (**c**) cholestatic type. Each dot represents an individual lipid.

#### *2.3. Discrimination Ability for Mixed-Type DILI between Acute Phase and Recovery Phase or Healthy Volunteers*

In the mixed-type DILI patients, three lipids, ceramide (Cer)(d34:1; d18:1/16:0), Cer(d36:1; d18:1/18:0), and oxidized ganglioside GM3 (GM3+O)(d34:1), were increased in the acute phase compared with the recovery phase, while six lipids, LPC(18:2), ether-linked LPC LPC(16:1e), ether-linked PC PC(38:6e; 18:2e/20:4, 16:1e/22:5), ether-linked phosphatidylethanolamine (PE) PE(36:4e; 16:0e/20:4), PE(38:4e; 18:0e/20:4), and PE(38:6e; 18:2e/20:4), were decreased in plasma (Table 3). The increased lipid of the highest effect size in the mixed-type patients was Cer(d34:1; d18:1/16:0), and the corresponding decreased lipid was PE(38:4e; 18:0e/20:4). All PCes and PEes contained the same FA (20:4: arachidonic acid).

Once we had characterized the specific lipids that were altered in mixed-type DILI, we next evaluated their discrimination ability between the acute phase and recovery phase by receiver operating characteristics (ROC) analysis. As shown in Table 3, four lipids, PE(38:4e; 18:0e/20:4), Cer(d34:1; d18:1/16:0), Cer(d36:1; d18:1/18:0), and GM3(d34:1)+O, had area under the curve (AUC) values over 0.8. The lipid with the highest AUC was Cer(d34:1; d18:1/16:0), with a value of 0.87.

We further compared the lipids levels of acute phase mixed-type DILI patients with the lipid levels of healthy subjects. Although the median ages of the three DILI patient types were approximately 60 years, we recruited the healthy subjects in four groups according to sex and age (HM1; middle-age male, HM2; old-age male, HF1; middle-age female, HF2; old-age female, where middle age was approximately 45 years and old age was approximately 60 years) (Table S2). The different lipids between mixed or cholestasis type DILI and all healthy subjects were listed in Table S3 (mixed) and Table S4 (cholestasis). As shown in Table 3, 6 lipids, LPC(18:2), LPC(16:1e), PE(38:6e; 18:2e/20:4), Cer(d34:1; d18:1/16:0), Cer(d36:1; d18:1/18:0), and GM3(d34:1)+O, were significantly different when comparing the acute phase DILI patients with all groups of healthy subjects. Cer(d34:1; d18:1/16:0), Cer(d36:1; d18:1/18:0), and GM3(d34:1)+O also had AUC values > 0.8 by ROC analysis versus all groups of healthy subjects. The representative individual plots of lipid levels discriminating the acute phase of mixed-type DILI from the recovery phase or the healthy volunteer groups are shown in Figure 2. Furthermore, we also calculated the ratio of altered specific lipids and evaluated their discriminating ability to acute phase mixed-type DILI patients from other groups, but no ratio of altered specific lipids further improved the discriminating ability. In addition, the absolute correlation coefficient of the altered specific lipids in mixed-type DILI with clinical parameters (AST, ALT, ALP, and total bilirubin) were all less than 0.6 (Table S5).

**Figure 2.** Representative individual plots of lipids levels discriminating the acute phase of mixed-type DILI from the recovery phase and the four healthy volunteer groups. Each dot represents an individual sample. Statistical significance is indicated as follows: \*\* *p* < 0.01, \*\*\* *p* < 0.001. Acute; acute phase DILI patients, Recovered; recovery phase DILI patients, HM1; healthy male subject group 1 (approximately 45 years old), HM2; healthy male subject group 2 (approximately 60 years old), HF1; healthy female subject group 1 (approximately 45 years old), HF2; healthy female subject group 2 (approximately 60 years old).


**Table 3.** Specific lipids altered in mixed-type DILI.

The values fulfilled the threshold values (*<sup>p</sup>* < 0.05, effect size > 0.8, ROC–AUC > 0.8) are indicated by bold fonts. M1; middle-age male, M2; old-age male, F1; middle-age female, old-age female, where middle age was approximately 45 years and old age was approximately 60 years, ROC: receiver operating characteristics, AUC: area under the curve.

#### *2.4. Discrimination Ability for Cholestatic-Type DILI between Acute Phase and Recovery Phase or Healthy Volunteers*

In the cholestatic-type DILI patients, 4 lipids, PC(30:0; 14:0/16:0), PC(31:0; 15:0/16:0), PC(32:1; 16:0/16:1), and PC (33:1; 15:0/18:1, 16:0/17:1), were increased in the acute phase then the recovery phase, while 16 lipids, PC(36:5e; 16:1e/20:4), PC(38:6e; 18:2e/20:4, 16:1e/22:5), PE(36:4e; 16:0e/20:4), PE(38:4e; 18:0e/20:4), PE(40:6e; 18:0e/22:6), PE(40:7e; 18:1e/22:6; M160), PE(40:7e; 18:1e/22:6; M161), phosphatidylinositol (PI)(38:3), PI(38:4; 18:0/20:4), PI(40:4), triglycosylceramide (CerG3)(d40:1), CerG3(d42:1), CerG3(d42:2), SM(d40:1; d18:1/22:0), TG(44:0; 14:0/14:0/16:0, 12:0/16:0/16:0), and CoQ10, were decreased (Table 4). The increased lipid of highest effect size in the cholestatic-type patients was PC(33:1; 15:0/18:1, 16:0/17:1) and the corresponding decreased lipid was PE(40:7e; 18:1e/22:6). Three FA(20:4)-containing ether-linked phospholipids, PC(38:6e; 18:2e/20:4, 16:1e/22:5), PE(36:4e; 16:0e/20:4), and PE(38:4e; 18:0e/20:4), were common with the mixed-type cases, but FA(22:6), corresponding to docosahexaenoic acid, was contained in three PEes, PE(40:6e; 18:0e/22:6), PE(40:7e; 18:1e/22:6; M160), and PE(40:7e; 18:1e/22:6; M161), which are specific for the cholestatic-type cases. In addition, all increased PCs in the acute phase of cholestatic-type patients contained the same FA(16:0: palmitic acid).

We also evaluated the discrimination ability of specific lipids that were altered in cholestatic-type DILI between the acute and recovery phases by ROC analysis. As shown in Table 4, 12 lipids, PC(31:0; 15:0/16:0), PC (33:1; 15:0/18:1, 16:0/17:1), PE(36:4e; 16:0e/20:4), PE(38:4e; 18:0e/20:4), PE(40:6e; 18:0e/22:6), PE(40:7e; 18:1e/22:6; M160), PI(38:3), PI(38:4; 18:0/20:4), PI(40:4), CerG3(d40:1), CerG3(d42:1), and CoQ10, had AUC values over 0.8. The lipid with the highest AUC was PE(40:7e; 18:1e/22:6; M160), with a value of 0.91.

We further compared the lipids levels of acute phase cholestatic-type DILI patients with the lipid levels of healthy subjects (grouped as indicated in Section 2.3). As shown in Table 4, eight lipids, PC(30:0; 14:0/16:0), PC(31:0; 15:0/16:0), PC(32:1; 16:0/16:1), PC(33:1; 15:0/18:1, 16:0/17:1), PC(36:5e; 16:1e/20:4), PI(38:3), PI(38:4; 18:0/20:4), and SM(d40:1; d18:1/22:0), were significantly different when comparing the acute phase DILI patients with all the compared groups of healthy subjects. PC(30:0; 14:0/16:0), PC(31:0; 15:0/16:0), PC(32:1; 16:0/16:1), PC (33:1; 15:0/18:1, 16:0/17:1), PI(38:3), PI(38:4; 18:0/20:4), and SM(d40:1; d18:1/22:0) also had AUC values >0.8 using ROC analysis versus all the groups of healthy subjects. The representative individual plots of lipid levels discriminating cholestatic-type DILI in the acute phase from the recovery phase or the healthy volunteer groups are shown in Figure 3. Furthermore, we also calculated the ratio of altered specific lipids and evaluated their discriminating ability to acute phase cholestatic-type DILI patients from other groups, but no ratio of altered specific lipids further improved the discriminating ability. In addition, the correlation coefficient of the altered specific lipids in mixed-type DILI with clinical parameters (AST, ALT, ALP, and total bilirubin) demonstrated over 0.6 (with *p* < 0.05) for three out of four palmitic acid-containing saturated or monounsaturated PCs, PC(31:0; 15:0/16:0), PC(32:1; 16:0/16:1), and PC (33:1; 15:0/18:1, 16:0/17:1) (Table S5). The absolute correlation coefficient of all other specific lipids was less than 0.6.


 **4.** Specific lipids altered in cholestatic-type DILI.

**Table**

old-age female, where middle age was

approximately

 45 years and old age was

approximately

 60 years, ROC: receiver operating

characteristics,

 AUC: area under the curve.

**Figure 3.** Representative individual plots of lipids levels discriminating the acute phase of cholestatic-type DILI from the recovery phase and the four healthy volunteer groups. Each dot represents an individual sample. Statistical significance is indicated as follows: \* *p* < 0.05, \*\* *p* < 0.01, \*\*\* *p* < 0.001. Acute; acute phase DILI patients, Recovered; recovery phase DILI patients, HM1; healthy male subject group 1 (approximately 45 years old), HM2; healthy male subject group 2 (approximately 60 years old), HF1; healthy female subject group 1 (approximately 45 years old), HF2; healthy female subject group 2 (approximately 60 years old).

#### **3. Discussion**

In this study, we used plasma lipid profiling to characterize the pathophysiology of three different types of DILI in human patients and made five broad observations. First, the mixed and cholestatic types of DILI demonstrated specific plasma lipid alterations between acute and recovered phases, but the hepatocellular type did not. Second, as specific features of mixed-type DILI, when compared with levels in the recovery phase, several ceramides were increased in the acute phase, while arachidonic acid-containing ether-linked phosphoglycerolipids were decreased. Third, as specific features of cholestatic-type DILI, when compared with levels in the recovery phase, palmitic acid-containing saturated or monounsaturated PCs increased in the acute phase, while arachidonic acid- or docosahexaenoic acid-containing ether-linked phosphoglycerolipids and PIs decreased. Fourth, of the specific lipids altered in mixed-type DILI, the levels of Cer(d34:1; d18:1/16:0), Cer(d36:1; d18:1/18:0), and GM3(d34:1)+O demonstrated relatively high discrimination ability for the acute phase over the recovery phase in all groups of healthy subjects. Finally, of the specific lipids altered in cholestatic-type DILI, the levels of PC(31:0; 15:0/16:0), PC(33:1; 15:0/18:1, 16:0/17:1), PI(38:3), and PI(38:4; 18:0/20:4) demonstrated relatively high discrimination ability for the acute phase over the recovery phase in all groups of healthy subjects.

Although the number of subjects in the cholestatic-type DILI group was limited, the number of specific lipids altered was larger in this group than in the other two groups. This result suggests that the alteration in hepatic lipid homeostasis in cholestatic-type DILI shares a common mechanism among diverse suspected drugs. One representative plasma lipid that was increased in cholestatic-type DILI was palmitic acid (16:0)-containing saturated or monounsaturated PCs. Palmitic acid has been reported as the major fatty acid in biliary PCs [23]. In addition, the partner FAs of palmitic acid in the specifically altered PCs in cholestatic-type DILI, FA(16:1), and FA(17:1) are preferentially secreted into bile [24]. Thus, these increased levels of palmitic acid-containing saturated or monounsaturated PCs in the plasma were probably due to the reduced bile secretion of palmitic acid-containing saturated or monounsaturated PCs by cholestasis. This is also supported because the total bilirubin levels, which were also elevated by biliary structure, were highly correlated with the lipid levels in the DILI patients in this study.

Along with the palmitic acid-containing saturated or monounsaturated PCs, PIs, such as PI(38:3) and PI(38:4; 18:0/20:4), were also specifically increased in the cholestatic-type DILI patients. To date, the role of increased plasma PIs in cholestatic-type DILI remains unclear. However, supplementation with PIs decreases mRNA levels of the inflammatory cytokines/chemokines, tumor necrosis factor-alpha (TNF-α), and monocyte chemoattractant protein-1 (MCP-1), which are upregulated in steatosis [25]. In addition, blood and liver PIs were shown to increase with hepatic steatosis [26,27]. Therefore, one plausible reason for the increased PIs in plasma is to counteract the hepatic inflammation that can occur with lipid dysregulation.

Unlike other lipid classes, arachidonic acid-containing ether-linked phosphoglycerolipids were commonly altered in mixed and cholestatic-type DILI. Decreased levels of serum ether-linked phosphoglycerolipids have been reported in patients with nonalcoholic steatohepatitis and nonalcoholic fatty liver disease when compared to the levels in healthy controls [21]. In addition, plasma and liver ether-linked phosphoglycerolipid levels were decreased in a valproic acid-induced rat model of hepatic steatosis [28]. Thus, the decreased levels of ether-linked phosphoglycerolipids that we observed in the plasma of mixed and cholestatic-type DILI patients could be caused by mechanisms that are like those in steatosis and steatohepatitis, and they could reflect reduced levels in the liver. Arachidonic acid is well-known to be metabolized to inflammatory eicosanoids, such as prostaglandin E2; thus, decreased levels of arachidonic acid-containing ether-linked phosphoglycerolipids in the plasma and the liver in the reference would implicate the inflammatory incidences in the liver of mixed and cholestatic-type DILI patients as well as patients with steatosis and steatohepatitis. Alternatively, ether-linked phosphoglycerolipids have been characterized as peroxisome-synthesized lipids and are a key component of peroxisome [29]. In fact, decreased levels of hepatic glyceronephosphate O-acyltransferase, which is a key peroxisomal enzyme for the synthesis of ether-linked phosphoglycerolipids, have been observed in a rat model of hepatic steatosis [28]. In addition, rescuing ether-linked phosphoglycerolipid levels by alkyl glycerol treatment could prevent impaired peroxisomal metabolism and hepatic steatosis [30,31]. Taken together, the decreased levels of plasma ether-linked phosphoglycerolipids that we observed may be caused by peroxisomal dysfunction in mixed and cholestatic types of DILI, and the rescue of ether-linked phosphoglycerolipid levels could be utilized for the therapeutic treatment of these DILI types.

Besides arachidonic acid-containing phosphoglycerolipids, increased plasma Cer was a characteristic feature of mixed-type DILI. To date, whether the increase in Cers plays a pivotal role in mixed-type DILI is unclear. However, Cers possess cell-signaling properties that are relevant to inflammation and apoptosis [32,33], and they may be involved in cystic fibrosis in the lung [34,35]. Thus, it is reasonable to speculate that increased Cer levels in mixed-type DILI patients contribute to hepatic inflammation and trigger subsequent pathological fibrosis.

In the present study, we also evaluated the differences in plasma lipids and their ability to discriminate between acute state DILI and healthy subjects divided into four age/sex groups. We identified 3 and 4 lipids in mixed and cholestatic types of DILI, respectively, as lipids with high

discrimination ability. Although their scores did not exceed those of ALT and ALP (data not shown), these lipids could be utilized as biomarkers for DILI patients with ALT and ALP levels that are not diagnostic of liver disease. For example, ALT is elevated in patients with muscle injury and ALP is elevated in bone diseases. These lipids may also be helpful to discriminate DILI types and determine therapeutic approaches. Further analysis is needed to corroborate these speculations.

There are several limitations in the present study. First, it was performed with a few subjects of mixed and cholestatic types. Although we collected samples in both the acute and recovery phases from the same patients, the number of analyzed patients was limited, thus restricting the statistical power of our analysis. Second, due to the sparse number of events and limited ability to follow up patients, we recruited DILI patients from seven core hospitals. Although we used the same sampling protocol, hospital-to-hospital variation in sample preparation may have produced slightly different results in plasma lipid levels. Third, postprandial effect has been reported to have a global impact on lipidomics, although the impact is less than that of inter-individual variations [36,37]. Thus, this impact should be taken into consideration even though it is less than the impact caused by inter-individual variations. However, it is difficult to control the food intake of DILI patients, especially during the acute phase. Therefore, we believe that the state of fasting can be disregarded for this preliminary study. Fourth, although we recruited self-reported healthy subjects who had taken no medication for at least 1 week as controls, they may have been unaware of their disease status. Fifth, we did not control the alcohol and food intake of the patients, and the time of blood draw was not standardized, both of which might have affected plasma lipid levels. Sixth, since we used one internal standard (PC[1 2:0/12:0]) for all the classes of lipids, ionization efficiency should be different among the classes. Thus, the fold changes can effectively be calculated/estimated even using the same IS for all the lipids, but the comparison between lipid classes is not valid then. Last, it is difficult to consider the effects of other disease states and external factors, such as sexes and ages. In fact, several lipids, such as PEes and CerG3s, have high discriminant ability between acute phase and recovered phase DILI, while those lipids could not discriminate acute phase DILI and some groups of healthy subjects, which may be attributed to differences in sexes and ages. In addition, as was reported in the literature, many diseases, including liver-related diseases, which are possibly base diseases and complications, alter the plasma lipid levels [18–22]. Multivariate analysis including these potentially affecting factors should be performed using more patients' samples. Therefore, to address these limitations, a future, large-scale study with updated protocols should be performed.

In conclusion, we characterized the plasma lipid profiles of three types of DILI patients using a lipidomics approach. By comparing samples in acute and recovery phases, we revealed that mixed and cholestatic types of DILI produce specific alterations in plasma lipid profiles. In addition, by comparing these data to those of healthy subjects, we found several candidate markers of mixed and cholestatic DILI that discriminate the acute phase from the recovery phase and healthy state. Our study provides insights into the alterations in plasma lipidomic profiles, which reflect alterations in lipid homeostasis in the livers of DILI patients. These findings may help to understand the pathophysiology of different types of DILI.

#### **4. Materials and Methods**

#### *4.1. Subjects and Sample Collection*

DILI patients were recruited at the Teikyo University Hospital, Tokai University Hospital, Hiroshima Atomic-bomb Survivors Hospital, Kitasato University Hospital, Gunma University Hospital, Fukuoka University Hospital, and Hokkaido University Hospital. The inclusion criteria for DILI in the acute phase were ALT ≥150 U/L and/or ALP ≥ 2× upper limit of normal, as described previously [38,39]. In addition, each DILI patient was scored using the CIOMS/RUCAM [3] and DDW-J2004 [4,5] scales, and the highest probability cases in these scores were included in this study. The CIOMS/RUCAM scale involves a scoring system that categorizes the cases into "definite or highly probable" (score > 8),

"probable" (score 6–8), "possible" (score 3–5), "unlikely" (score 1–2), and "excluded" (score ≤ 0). The DDW-J2004 scale involves a scoring system that categorizes the cases into "highly probable" (score > 5), "possible" (score 3–4), and "unlikely" (score ≤ 2). The DILI type and entry into the recovery phase were also diagnosed by DILI experts at each hospital. All healthy subjects were non-smoking, self-reported healthy volunteers who had taken no medications for at least 1 week before the study.

Blood samples were collected by venipuncture into 7 mL EDTA-2Na-containing vacuum blood collection tubes (VENOJECT II, TERUMO, Tokyo, Japan). The blood samples were immediately centrifuged (2500× *g*, 10 min, 4 ◦C); the resulting plasma was dispensed into screw-capped polypropylene tubes and stored in a deep freezer (−80 ◦C) before use. The plasma was typically frozen within 2 h from blood draw, although this occasionally extended to 4 h.

This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the National Institute of Health Science (256, and 260 for Kihara Memorial Foundation), Teikyo University Hospital (15-127-2), Tokai University Hospital (15R-117), Hiroshima Atomic-bomb Survivors Hospital (H27-399-2), Kitasato University Hospital (B13-182), Gunma University Hospital (1487), Fukuoka University Hospital (18-8-04), Hokkaido University Hospital (016-0345), Daiichi Sankyo Co., Ltd. (15-0504-00), and Astellas Pharma Inc. (150028-01, 150047-01). Written informed consent was obtained from all participants.

#### *4.2. Lipidomics*

Lipid extraction was performed using the Microlab NIMBUS workstation (Hamilton, Binaduz, GR, Switzerland). The plasma samples were mixed with nine volumes of methanol/isopropanol (1/1) containing an internal standard (PC[12:0/12:0]), which is not detectable endogenously, at 2 μM. The mixed samples were filtered through a FastRemover Protein Removal Plate (GL Science, Tokyo, Japan) using an MPE2 automated liquid handling unit (Hamilton). The resulting lipid-containing filtrate was directly subjected to lipidomics. To obtain the lipidomics data, we performed reversed-phase LC (RPLC; Ultimate 3000, Thermo Fisher Scientific, Waltham, MA, USA) and MS (Orbitrap Fusion, Thermo Fisher Scientific), as described previously [40,41]. Compound Discoverer 2.1 (Thermo Fisher Scientific) was used with the raw data for peak extraction, annotation, identification, and lipid quantification, as described previously with a prior version of the software [40,41]. For isomers (same class, carbon length, and number of double bonds) showing different retention times in RPLC, each lipid was assigned a metabolite ID to distinguish it. Lipids with two different fatty acid combinations (e.g., 38:6e; 18:2e/20:4, 16:1e/22:5) indicate that the quantified lipid is a mixture of two different lipids that could not be separated. The quantified raw data were normalized to the internal standard. Since the lipidomics analysis was combined across two batches, the median value of each lipid in all samples was set to one in each batch to consolidate data from two batches after normalization. The processed data for the lipid levels are presented in Table S1.

#### *4.3. Statistical Analysis*

Significant differences in lipid levels were assessed by paired t-tests and Welch's t-test, and the effect size, which is calculated by Hedge's *g*, was considered. In this study, due to the limitation of sample size, a lipid level was considered specifically altered if its *p* value was <0.05 and its absolute effect size was >0.8. The discrimination ability was assessed by AUC score in ROC analysis using GraphPad Prism 6 (GraphPad Software, San Diego, CA, USA). The correlation coefficient was calculated as Pearson's correlation coefficient.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2218-1989/10/9/355/s1, Table S1: Lipidomics data set used in the present study. Table S2: Details of healthy subjects in the present study. Table S3: Lipids different between mixed-type DILI and healthy subjects. Table S4: Lipids different between cholestatic-type DILI and healthy subjects. Table S5: Correlation coefficient of altered specific lipids to clinical parameters in DILI patients. Figure S1: The exemplar LC/MS traces of plasma lipid profiles. Text S1: The details on the lipidomic analyses and associated data processing.

*Metabolites* **2020**, *10*, 355

**Author Contributions:** Conceptualization, Y.O., Y.S., and H.T.; subject recruitment, T.K., K.T. (Keiji Tsuji), Y.K., K.S. (Ken Sato), S.S., N.S., M.A., S.H., N.M., R.T., T.U., K.T. (Kazuhide Takata), K.O., and H.T.; blood sample preparation, T.K., K.T. (Keiji Tsuji), Y.K., K.S. (Ken Sato), S.S., N.S., M.A., S.H., N.M., R.T., T.U., K.T. (Kazuhide Takata), K.O., N.A., and H.T.; lipidomics analysis K.S. (Kosuke Saito); writing—original draft preparation, K.S. (Kosuke Saito); writing—review and editing, T.K., K.T. (Keiji Tsuji), Y.K., K.S. (Ken Sato), S.S., N.S., M.A., S.H., N.M., R.T., T.U., K.T. (Kazuhide Takata), K.O., K.M., M.S., T.N., K.T. (Kazuhiko Takamatsu), T.I., Y.O., Y.S., and H.T.; funding acquisition, K.S. (Kosuke Saito), T.K., K.T. (Keiji Tsuji), Y.K., K.S. (Ken Sato), S.S., N.S., N.A., Y.O., Y.S., and H.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Japan Agency for Medical Research and Development, grant number (19mk0101045h0005, 19mk0101045s0105, 19mk0101045s0205, 19mk0101045s0305, 19mk0101045s1205, 19mk0101045s1405, 19mk0101045s1505, 19mk0101045s1605, 19mk0101045j0105, 19mk0101045j0305, 19mk0101045j0405).

**Acknowledgments:** We thank C. Sudo (National Institute of Health Sciences) for administrative assistance; M. Kojima, R. Iiji, R. Kaneko (National Institute of Health Sciences) for analytical assistance; K. Kubota, T. Hirata (Daiichi Sankyo RD Novare Co., Ltd.), K. Hashimoto (Daiichi Sankyo Co., Ltd.), Y. Hashimoto (Astellas Pharma Inc.), for technical advice.

**Conflicts of Interest:** Tatehiro Kagawa has received research funding from AbbVie GK, Sumitomo Dainippon Pharma, Eisai, and Daiichi Sankyo and honoraria from AbbVie GK, Sumitomo Dainippon Pharma, Bayer Yakuhin, Eisai, Daiichi Sankyo, and Gilead Sciences. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results. All the other authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
