Proteomic and Biochemical Studies of Lysine Malonylation Suggest Its Malonic Aciduria-associated Regulatory Role in Mitochondrial Function and Fatty Acid Oxidation^*

Materials

Chemicals were purchased as analytical grade from Sigma-Aldrich, Inc. (St. Louis, MO). Modified sequencing grade trypsin was purchased from Promega Corp. (Madison, WI). Pan-anti-malonyllysine antibody and pan-anti-malonyllysine-agarose beads were from PTM Biolabs, Inc. (Chicago, IL). MS grade water and acetonitrile were from Thermo Fisher Scientific (Waltham, MA). C₁₈ ZipTips were purchased from Millipore Corp. (Billerica, MA). SILAC DMEM (CCFDA003-132J01) was purchased from UCSF Cell Culture Facility (San Francisco, CA). XerumFree reagent (XF205) was purchased from MayFlower Bioscience (St. Louis, MO). Dialyzed serum (Gibco 26400) was purchased from Life Technologies, Thermo Fisher Scientific.

Preparation of Mouse Liver Lysate

Four two-month-old male Sirt5 KO mice (25, 26) were anesthetized with an isoflurane overdose, and the blood in the liver was removed by perfusion with ice-cold PBS for 5 min. Liver was homogenized in a glass Dounce homogenizer in SDS lysis buffer (20 mm Tris-HCl, pH 6.8, 1% SDS, 5% β-mercaptoethanol, 10% glycerol, 25 mm nicotinamide). The lysates from four livers were pooled together, and the sample was clarified by centrifugation at 16,000 × g. The protein in the supernatant was precipitated with 10% (v/v) trichloroacetic acid. Then the precipitated proteins were in solution-digested with trypsin as described previously (27).

Preparation of SILAC Samples

Human dermal fibroblast cells lines MCD+/+ (control cells) and MCD−/− cells were obtained from Gaslini Biobank, Italy. The cells were grown in SILAC DMEM supplemented with l-glutamine (584 mg/liter), 10% (v/v) dialyzed serum, and 2% (v/v) serum-free reagent. Regular l-lysine (¹²C₆,¹⁴N₂-labeled) and l-arginine (¹²C₆,¹⁴N₂-labeled) were added to the “light” medium (final concentration, 100 mg/liter) used for culturing MCD−/− cells. “Heavy” isotopic l-lysine (¹³C₆,¹⁵N₂-labeled) and light l-arginine (¹²C₆,¹⁴N₂-labeled) were added to the heavy medium (final concentration, 100 mg/liter) used for culturing MCD+/+ control cells. Both cell lines were grown in parallel until MCD+/+ cells were sufficiently labeled by the isotopic lysine.

Both MCD+/+ and MCD−/− cells were lysed in SDS buffer (20 mm Tris-HCl, pH 6.8, 1% SDS, 5% β-mercaptoethanol, 10% glycerol, 25 mm nicotinamide). Twelve milligrams of each cell lysate were mixed and precipitated overnight by 10% TCA for tryptic digestion.

HPLC Fractionation

The tryptic peptides were fractionated by using a reversed phase column (Luna C₁₈, 10 × 250 mm, 5-μm particle, 100-Å pore size, Phenomenex Inc., Torrance, CA) in a Discovery VP preparative HPLC system (Shimadzu Corp., Kyoto, Japan). The peptides were fractionated into 75 fractions using a gradient from 2 to 90% buffer B (10 mm ammonium formate in 90% acetonitrile and 10% water, pH 7.8) in buffer A (10 mm ammonium formate in water, pH 7.8) at a flow rate of 4 ml/min in 60 min. The 75 fractions were finally combined equally into five final fractions for mouse liver samples and 10 final fractions for MCD SILAC samples, respectively. Each fraction was condensed by using a SpeedVac (ThermoSavant SPD111V). The peptide solution was used for immunoaffinity enrichment.

Affinity Enrichment of the Peptides Containing Kmal

The peptides containing Kmal were enriched using a procedure described previously (27). The tryptic peptides from each fraction were resolubilized in 100 mm NH₄HCO₃, pH 8.0. Samples were centrifuged at 20,000 × g for 10 min to remove insoluble particles. The peptides were incubated with 15 μl of agarose beads conjugated with anti-malonyllysine antibody at room temperature for 4 h with gentle rotation. The beads were washed three times with NETN buffer (50 mm Tris-HCl, pH 8.0, 100 mm NaCl, 1 mm EDTA, 0.5% Nonidet P-40), twice with ETN buffer (50 mm Tris-HCl, pH 8.0, 100 mm NaCl, 1 mm EDTA), and once with water. Enriched Kmal peptides were eluted from the beads by washing three times with 0.1% trifluoroacetic acid. The eluted Kmal peptides were dried in a SpeedVac.

Nano-HPLC-MS/MS Analysis

The enriched Kmal samples were first desalted using OMIX C₁₈ tips (Agilent Technologies Inc., Santa Clara, CA) and then dissolved in solvent A (0.1% formic acid in water). Samples were injected onto a manually packed reversed phase C₁₈ column (100 mm × 75 μm, 3-μm particle size, Dr. Maisch GmbH, Ammerbuch, Germany) connected to an Easy-nLC 1000 HPLC system (Thermo Fisher Scientific Inc., Waltham, MA). Peptides were eluted from 5 to 90% solvent B (0.1% formic acid and 1% water in acetonitrile) in solvent A with a 1-h gradient at a flow rate of 200 nl/min. The analytes were directly ionized and sprayed into a Q Exactive mass spectrometer (Thermo Fisher Scientific Inc.) by Nanospray Flex^TM Ion Sources. Full MS scans were acquired in the Orbitrap mass analyzer over the range m/z 300–1400 with a mass resolution of 70,000 at m/z 200. The 15 most intense peaks of the precursor ions were fragmented in the high energy collision dissociation collision cell with normalized collision energy of 27, and tandem mass spectra were acquired with a mass resolution of 17,500 at m/z 200. Lock mass at m/z 445.120024 was enabled for internal calibration of the full MS spectrum. Ions with either a single charge or more than four charges were excluded from MS/MS fragmentation, and the dynamic exclusion duration was set to 25 s.

Data Processing and Analysis

MaxQuant software (version 1.3.0.5) was used for identifying and quantifying protein and malonylated peptides. Peak list generation and precursor mass recalibration of the raw MS data were carried out by MaxQuant software. Trypsin was specified as the cleavage enzyme, and the maximum number of missed cleavage was set at 3. Methionine oxidation, protein N-terminal acetylation, lysine acetylation (Kac), Kmal (specified for neutral loss of CO₂ in MS/MS fragmentation), and Ksucc were specified as variable modifications, and cysteine alkylation by iodoacetamide was specified as a fixed modification for all database searching. Database searching was performed against the UniProt mouse (50,807 sequences; release date, May 2013) or human (88,817 sequences; release date, February 2014) reference protein sequence database concatenated with a reversed decoy database with initial precursor mass tolerance of 7 ppm. Mass tolerance for fragment ions was set at 20 ppm. False discovery rate thresholds for protein, peptide, and modification site were fixed at 0.01. The identified peptides with MaxQuant Andromeda score below 50 and localization probability below 0.75 were removed prior to bioinformatics analysis.

Malonyl-CoA Measurement

The cells were treated with 15 μm orlistat or vehicle only for 24 h at 80% confluence. The media were quickly removed, and the dishes were placed on top of dry ice. One milliliter of extraction solvent (80% methanol in water) was immediately added, and the dishes were transferred to the −80 °C freezer. The dishes were left for 15 min, and then cells were scraped into extraction solvent on dry ice. The whole solution was centrifuged at 20,000 × g at 4 °C for 10 min. Cell extracts were prepared from 3 wells to make biological triplicates. The supernatant from the tissue extract was transferred to a new tube for LC-MS/MS analysis. All samples were dried in a vacuum concentrator (SpeedVac, Thermo Fisher Scientific Inc.).

An Ultimate 3000 UHPLC (Thermo Fisher Scientific Inc.) was coupled to Q Exactive mass spectrometer (Thermo Fisher Scientific Inc.) for metabolite separation and detection. For acyl-CoA analysis, a reversed phase LC method was used. A Luna C₁₈ column (100 × 2.0-mm inner diameter, 3 μm, Phenomenex Inc.) was used with mobile phase A (water with 5 mm ammonium acetate, pH 6.8) and mobile phase B (methanol) at a flow rate of 0.2 ml/min. The linear gradient was as follows: 0 min, 2% B; 1.5 min, 2% B; 3 min, 15% B; 5.5 min, 95% B; 14.5 min, 95% B; 15 min, 2% B; 20 min, 2% B. The column was at room temperature.

The Q Exactive mass spectrometer was equipped with an heated ESI probe, and the relevant parameters were as follows: heater temperature, 120 °C; sheath gas, 30; auxiliary gas, 10; sweep gas, 3; spray voltage, 3.6 kV for positive mode. The capillary temperature was set at 320 °C, and S-lens was 55. A full scan range was set at 300–1000 (m/z). The resolution was set at 70,000 (at m/z 200). The maximum injection time was 200 ms. Automated gain control was targeted at 3 × 10⁶ ions. For CoA analysis, cell extract was dissolved into 30 μl of water with 50 mm ammonium acetate, pH 6.8. Samples were centrifuged at 20,000 × g at 4 °C for 3 min, and the supernatant was transferred to LC vials. The injection volume was 8 μl for CoA analysis.

Raw data collected from LC-MS/MS were processed on Thermo Scientific software Sieve 2.0. Peak alignment and detection were performed according to the manufacturer's protocols. For a targeted metabolomics analysis, a frame seed including Acyl-CoA metabolites that was previously validated was used for targeted metabolite analysis with data collected in positive mode with the m/z width set at 8 ppm. Statistical significance was calculated based on Student's t test (unpaired, two-tailed).

Motif Analysis for Lysine Malonylation Substrates

The standalone version of iceLogo (version 1.2) software was used to analyze the preference of flanking Kmal site sequence from mouse liver or human MCD cells (28). The embedded Swiss-Prot “Mus musculus” or “Homo sapiens” was used as the negative set. Six flanking amino acid residues on each side of a lysine malonylated site were selected as the positive set.

Functional Enrichment Analysis

Functional enrichment analysis of lysine malonylated proteins was carried out using DAVID (Functional Annotation Bioinformatics Microarray Analysis) Bioinformatics Resources version 6.7 with the total mouse or human genome information as the background (29). All identified lysine malonylated proteins were subjected to database analyses using gene ontology (GO) (30) and Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolic pathways (31). The GO Fat database from DAVID was selected in this analysis. The family-wide false discovery rate was corrected by the Benjamini-Hochberg method using an adjusted p value cutoff of 0.05.

Protein-Protein Interaction Network Analysis

Protein-protein interaction networks of the lysine malonylome were analyzed using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database (version 9.1; confidence score, 0.7) visualized by Cytoscape software (version 3.1.0) with the MCODE App toolkit (32). The confidence score is the approximate probability that a predicted link exists between two enzymes in the same metabolic map in the KEGG database. Confidence limits are as follows: low confidence, 0.2 (or better); medium confidence, 0.5; high confidence, 0.75; the highest confidence, 0.95.

Protein Complex Enrichment Analysis

Manually curated core complexes indexed by the CORUM (the comprehensive resource of mammalian protein complexes) database were used for the analysis of lysine malonylated substrates. Mouse or human complexes indexed in the database were used for enrichment analysis of mouse liver or MCD human cells by Fisher's exact test. Complexes with an adjusted p value <0.01 were considered as significant.

Kmal Stoichiometry Calculation

Absolute stoichiometry calculation of malonylated site in SILAC samples was based on the previously reported algorithm (33) with slight modification (34). The calculation was based on the MS quantification data (SILAC ratio) of the Kmal peptides (x), the corresponding protein (z), and the corresponding unmodified peptide (y) with the assumption that only one type of PTM occurs at a given site. The SILAC ratios of unmodified peptides (y) and proteins (z) were calculated from the global protein expression analysis using the whole cell lysate mixture of SILAC labeled MCD+/+ and MCD−/− cells without antibody affinity enrichment. The calculation assumed that only one type of PTM occurred at the given site of interest. The unmodified peptide was defined as the longest completely digested part of the peptide sequence derived from the malonylated peptide, which contains no other PTM. The absolute stoichiometry was calculated based on the SILAC ratios of x, y, and z using the same formula as reported previously (33).

Mitochondrial Respiratory Flux Analysis

Measurements of cellular oxygen consumption were performed using an extracellular flux analyzer (Seahorse BioScience, Billerica, MA). Fao hepatoma cells were incubated for 24 h in culture medium (DMEM supplemented with 2 mm HEPES, 2% penicillin/streptomycin, and 10% FBS) containing 50 mm malonate. Next, cells were plated at 20,000 cells/well in Seahorse 96-well culture plates followed by overnight incubation in malonate-free medium. Human fibroblasts were maintained and plated in DMEM supplemented with 2 mm HEPES, 2% penicillin/streptomycin, and 10% FBS at 30,000 cells/well. Seahorse mitochondrial function analysis was performed using the digitonin cell permeabilization protocol (35). Prior to measurements of respiration, culture medium was replaced with MAS buffer (pH 7.4, 220 mm mannitol, 70 mm sucrose, 10 mm KH₂PO₄, 5 mm MgCl₂, 2 mm HEPES, 1 mm EGTA, 0.6% fatty-acid free BSA). Oxygen consumption rate (OCR) was analyzed following a single injection of either pyruvate/malate/ADP/digitonin, succinate/rotenone/ADP/digitonin, or octanoylcarnitine/malate/ADP/digitonin dissolved in MAS buffer without BSA at pH 7.4. Final digitonin concentration was 30 μg/ml for Fao hepatoma cells and 100 μg/ml for fibroblasts. Final substrate concentrations were as follows: pyruvate, 5 mm; malate, 2.5 mm; succinate, 10 mm; octanoylcarnitine, 100 μm; ADP, 1 mm. After injection of substrate, oligomycin was injected at 1.5 μm final concentration followed by injection of antimycin (2.5 μm) and rotenone (1.25 μm).

Very Long-chain Acyl-CoA Dehydrogenase (VLCAD) Activity Analysis

VLCAD activity was analyzed by monitoring the specific conversion of palmitoyl-CoA (C16:0-CoA) into palmitenoyl-CoA (C16:1-CoA) in cell lysates (36). Cell lysates (0.1 mg/ml) were incubated in 0.125 mm Tris, pH 8.0, with 0.4 mm ferrocenium and 0.25 mm palmitoyl-CoA for 10 min at 37 °C, and the reaction was stopped by addition of 10 μl of 2 n HCl followed by neutralization with 10 μl of 2 m KOH, 0.6 m MES. Samples were deproteinated with acetonitrile followed by separation of substrate and products by reversed phase C₁₈ HPLC and UV detection.

Long-chain 3-Hydroxyacyl-CoA Dehydrogenase (LCHAD) Activity Analysis

LCHAD activity was analyzed by incubating cell lysates (0.1 mg/ml) with 3-ketopalmitoyl-CoA (0.26 mm; synthesized in house) and 0.4 mm NADH in 100 mm MES, 200 mm potassium phosphate buffer with 0.1% Triton X-100, pH 6.2, for 5 min at 37 °C using a procedure described previously (37). To control for the conversion of 3-ketopalmitoyl-CoA by short-chain 3-hydroxyacyl-CoA dehydrogenase, samples were incubated with and without N-ethylmaleimide because N-ethylmaleimide inhibits only LCHAD and not short-chain 3-hydroxyacyl-CoA dehydrogenase. After incubation, reactions were stopped with 10 μl of 2 n HCL followed by neutralization with 10 μl of 2 m KOH, 0.6 m MES. Samples were deproteinated with acetonitrile followed by separation of substrate and products by reversed phase C₁₈ HPLC and UV detection.

Immunocytochemistry

MCD+/+ and MCD−/− cells were grown on coverslips and treated with 15 μm orlistat for 48 h. MitoTracker Red was added to the culture medium at 0.1 μm final concentration and incubated for 30 min. The cells were washed with PBS twice, fixed with 4% (v/v) paraformaldehyde, and permeabilized with 0.2% (v/v) Triton X-100. The cells were blocked with 2% bovine serum albumin for 2 h and incubated with the corresponding primary antibodies at 1.5 μg/ml final concentration overnight. The cells were washed with PBS twice, incubated with secondary antibody Alexa Flour 488 (Invitrogen) for 2 h, and washed with PBS twice. Hoechst (BD Biosciences, San Jose, CA) was added at 2 μg/ml final concentration and incubated for 15 min. The coverslips are washed with PBS twice and mounted. The imaging was performed using a Leica SP2 DMIRE2 confocal microscope with HCX PL APO Ibd.BL 63× 1.4 oil objective.

Kmal Is Affected by SIRT5 and MCD

Our previous studies showed that SIRT5 can catalyze removal of malonyl groups from malonylated lysine residues both in vitro and in vivo (15). In addition, exogenous malonate can boost lysine malonylation, possibly by increasing intracellular concentrations of malonyl-CoA catalyzed by a short-chain acyl-CoA synthase (15). Consistent with this result, Sirt5 KO mice showed increased Kmal and Ksucc levels compared with their wild-type counterparts, but not Kac (Fig. 2A).

We previously showed, by Western blotting analysis, that Kmal levels are higher in MCD−/− cells than MCD+/+ cells (24). This result, in combination with our earlier observation that malonate can enhance Kmal (15), supports a hypothesis that MCD−/− induces an increase of malonyl-CoA concentration that in turn boosts Kmal. If this is true, a reduction of lipid biosynthesis by reduced activity of fatty-acid synthase may also increase malonyl-CoA and Kmal levels. To test this, we treated both control MCD+/+ and MCD−/− cells with orlistat, an inhibitor of fatty-acid synthase (38). Consistent with our hypothesis, we observed an increase of Kmal levels in response to orlistat in MCD+/+ cells (Fig. 2B). In addition, orlistat further increased Kmal levels in MCD−/− cells compared with MCD+/+ cells, whereas Kac and Ksucc levels remained largely unchanged (Fig. 2C).

To test whether the enhanced Kmal levels are correlated with higher amounts of malonyl-CoA, we measured intracellular malonyl-CoA levels in MCD+/+ and MCD−/− cells using an HPLC-MS-based metabolomics method. Our data showed that orlistat significantly increased intracellular malonyl-CoA levels in both cell lines (Fig. 2D), suggesting that increased lysine malonylation induced via orlistat treatment might be due to enhanced concentration of malonyl-CoA.

Taken together, three different strategies for enhancing malonyl-CoA levels lead to increased levels of lysine malonylation. This result is consistent with our previous work showing that increasing crotonyl-CoA, succinyl-CoA, and glutaryl-CoA levels all result in increases of their respective lysine acylations (11, 13, 39).

Proteomic Identification of Kmal Peptides

Identification of protein substrates is critical to studying the biology of a PTM pathway as was demonstrated in characterization of the lysine acetylation pathway (40–43). To identify Kmal substrate proteins and their modification sites, we used a proteomics approach involving affinity enrichment and subsequent HPLC-MS/MS analysis (Fig. 3). Two experimental models were used: Sirt5 KO mice and MCD-deficient fibroblasts from malonic aciduria patients. Analysis of Kmal substrates in mouse liver allows us to identify Kmal substrates in an organ important for cellular metabolism (Fig. 3A). The liver also has the highest lysine malonylation levels among the mouse tissues that we screened (supplemental Fig. S1). Quantification of Kmal substrates in MCD-deficient cells versus wild-type controls can reveal key Kmal substrates whose modification status is changed in response to malonic aciduria and whose increased malonylation may play a pathogenic role in this disorder (Fig. 3B).

Protein extracts from liver tissues of Sirt5 KO mice were prepared, tryptically digested, and resolved into five fractions by high pH reversed phase HPLC. Kmal peptides were enriched using pan-anti-malonyllysine antibody. The enriched Kmal peptides were analyzed by HPLC-MS/MS (Fig. 3A). The acquired raw MS data were analyzed by MaxQuant software with a false discovery rate of 0.01 at protein and peptide levels for the identification of Kmal peptides. To ensure high confidence of the identifications, we removed Kmal peptides with Andromeda scores between 40 and 50 and localization probability below 0.75 prior to bioinformatics analysis (supplemental Table S1A). The Andromeda score is used for ranking the confidence of peptide identifications for the MS/MS spectrum by the Andromeda search engine integrated in MaxQuant software. A higher score indicates a more confident peptide identification. This analysis led to identification of 4016 Kmal sites in 1395 proteins in Sirt5 KO mouse liver (Fig. 3C, top). A significant portion of the 427 malonylated peptides (9.6% of the total) with Andromeda scores between 40 and 50 may represent true positives, and these proteins were listed as Kmal candidates (supplemental Table S1B). Annotated speactra of lysine malonylated peptides in mouse liver is included n the supplements.

In a parallel experiment, we identified and quantified Kmal peptides in human dermal fibroblasts isolated from normal individuals (MCD+/+; labeled with heavy lysine isotope) and from malonic aciduria patients that are deficient in MCD (MCD−/−; labeled with light lysine isotope). Equal amounts of protein lysates from both MCD+/+ and MCD−/− cells were combined in a 1:1 ratio and processed using the same procedure as described above for analysis of Kmal peptides. The study identified 4943 Kmal sites, with Andromeda scores >50, on 1831 proteins in human fibroblasts (MCD+/+ and MCD−/− combined; Fig. 3C and supplemental Table S1C). We considered 732 Kmal sites with Andromeda scores between 40 and 50 as the true positive Kmal candidate peptides, and these are listed in supplemental Table S1D. Annotated speactra of lysine malonylated peptides in human cells is included n the supplements.

Among the Kmal substrates, we identified 21 histone marks in mouse liver and 19 histone marks in human fibroblasts (supplemental Table S2). Interestingly, most of them were not located at N-terminal tails of histones (Fig. 3D). Similarly, 35 non-redundant histone lysine succinylation sites have been reported in mouse liver that mostly localize to C-terminal globular domains as well (10). These results suggest that both Kmal and Ksucc histone marks are likely to have a regulatory function different from the widely studied histone acetylation marks (supplemental Table S2).

We previously reported that Ala and Gly were over-represented in the flanking sequences of Ksucc sites, whereas Arg was largely depleted at both −1 and +1 positions (10). Similarly, we evaluated the flanking sequences of Kmal sites to identify whether there was a structural preference for the location of this modification on the peptides. Motif analyses of Kmal sites in mouse liver (supplemental Fig. S2A, left) and human fibroblasts (supplemental Fig. S2B, left) showed significant similarity. Aliphatic amino acids including Ala, Val, Ile, and Gly were over-represented at the flanking sequence of Kmal sites similar to the situation with Ksucc sites, whereas Ser, Pro, and Leu were under-represented. In contrast to the similarity of Kmal and Ksucc flanking sequences, positively charged residues, such as Lys and Arg, predominate in Kac motifs in mouse liver (44).

Quantification of Changes in Kmal Modification Levels from MCD-deficient Cells Versus Wild Type

Using a SILAC-based quantitative proteomics approach, we quantified the difference in Kmal substrate levels between MCD+/+ and MCD−/− cells based on the levels of Kmal peptides and those of protein expression. In parallel, we also quantified changes of protein expression using whole cell lysates derived from a mixture of SILAC labeled MCD+/+ and MCD−/− cells. The changes of Kmal peptides were normalized to the change of their corresponding proteins' levels in MCD cells. Normalized changes of Kmal peptides were used for the subsequent analysis.

Among 4943 Kmal sites on 1822 proteins identified in MCD human fibroblasts, 3181 Kmal sites on 1257 proteins could be quantified (supplemental Table S1, C and D). Among the 1762 unquantified Kmal sites, 1452 are present only in MCD−/− cells (light only); these are the Kmal peptides that have no detected signal in heavy labeled peptide from MCD+/+ cells but significant intensity for the corresponding light labeled peptide from MCD−/− cells by MaxQuant analysis (supplemental Table S1C). The median MCD+/+:MCD−/− ratio of the quantifiable Kmal sites was 0.8284 (Fig. 4A). These results clearly suggest that MCD deficiency has an impact on elevating Kmal levels in MCD−/− cells. 461 Kmal sites on 339 proteins increased in abundance by 2-fold or more (normalized log₂ ratio (MCD+/+:MCD−/−) ≤ −1), whereas 1452 Kmal sites on 822 substrate proteins were present in light-only MCD−/− cells (Fig. 4B and supplemental Table S1C) Forty-eight Kmal sites on 38 Kmal proteins showed more than a 10-fold increase in MCD−/− cells (supplemental Table S1C). We considered these Kmal substrates to represent the core group of MCD−/− stimulated Kmal substrates. KEGG pathway analysis indicated that these substrates are associated with tricarboxylic acid cycle, oxidative phosphorylation, amino acid degradation (valine, leucine, isoleucine, and lysine), fatty acid metabolism, and propanoate metabolism pathways (supplemental Table S5E).

To calculate the stoichiometry of Kmal in MCD+/+ and MCD−/− cells, we modified a reported algorithm (33) as we described previously (10, 34). The calculation was based on the successful quantification of a Kmal site, its corresponding protein, and the unmodified peptide form in the SILAC experiment (for details, see Ref. 34). To achieve a more accurate calculation, we removed those Kmal sites that were previously reported to be acetylated and succinylated (47, 48) to minimize errors caused by the two modifications at the same residues. This analysis enabled us to calculate the stoichiometry of 325 Kmal sites on 222 proteins in MCD−/− cells with calculated stoichiometries ranging from 0.07 to 50.0% and in MCD+/+ cells with a range from 0.01 to 48.6%, respectively (Fig. 4C and supplemental Table S3). The two highest Kmal stoichiometry sites were Lys³⁷⁶ of adenylyl cyclase-associated protein 1 (50.0% in MCD−/− cells and 48.6% in MCD+/+ cells) and K41 of phosphoglycerate kinase (49.4% in MCD−/− cells and 48.2% in MCD+/+ cells). VLCAD catalyzes the first step of mitochondrial fatty acid oxidation. Nine Kmal sites were identified in VLCAD among which five sites were detected in MCD−/− cells only, whereas the other four were up-regulated in MCD−/− cells, suggesting a dramatic increase of Kmal on this protein. The dynamic increase of two Kmal sites in VLCAD was 390- and 137-fold, respectively, in MCD−/− cells. Among 324 sites whose stoichiometries were determined, 179 sites (55%) have more than a 2-fold increase of Kmal stoichiometry in MCD−/− cells (supplemental Table S3). For example, malonylation at Lys²⁹⁵ of mitochondrial 10-formyltetrahydrofolate dehydrogenase, which is responsible for formate oxidation, is increased from 0.7% in MCD+/+ cells to 31% in MCD−/− cells. Malonylation at Lys¹²⁶ of prohibitin-2, a mediator of transcriptional repression by nuclear hormone receptors, increased from 12.8 to 42.3% in MCD−/− cells.

Overlap among Kmal, Ksucc, and Kac Sites

To understand the similarities and differences among Kmal, Ksucc, and Kac sites, we compared our lysine malonylome data with previously published data (10, 41, 43, 47). We found that, of all the identified Kmal sites in mouse liver, 640 (16%) sites (Fig. 5A, right) and 595 (42%) proteins (Fig. 5A, left) overlapped with Kac sites in mouse embryonic fibroblasts (43). Five hundred and ten (36.5%) Kmal sites (Fig. 5A, right) and 262 (6.5%) proteins (Fig. 5A, left) overlapped with Ksucc sites in Sirt5 KO mouse liver (10). When we pooled the Ksucc sites reported in Sirt5 KO mouse liver and mouse embryonic fibroblasts and carried out the same analysis, 706 (17.6%) sites (supplemental Fig. S2C, right) and 406 (29%) proteins (supplemental Fig. S2C, left) overlapped with Kmal sites identified in Sirt5 KO mouse liver. Interestingly, we found that a significant portion of the malonylated proteins (46.2%) and sites (71.1%) identified in our mouse liver data do not overlap with the previously reported Ksucc and Kac data.

In a parallel experiment, we carried out a similar analysis for the human malonylome. In this experiment, we obtained the Kac and Ksucc data from previous publications (41, 47). Among the Kmal sites identified in human fibroblasts (combination of MCD+/+ and MCD−/−), 776 Kmal sites (Fig. 5B, right) and Kmal 827 proteins (Fig. 5B, left) overlapped with the human Kac proteome (41), and 671 sites (Fig. 5B, right) and 550 proteins (Fig. 5B, left) overlapped with the human Ksucc proteome (47). Similar to the mouse malonylome data, a significant portion of the human malonylated proteins (46.7%) and sites (75.7%) did not overlap with previously reported Ksucc and Kac data. Overall, the spectrum of lysine sites and protein targets subject to malonylation shows substantial non-overlap with Kac and Ksucc, suggesting that this modification likely plays roles in modulating biological processes distinct from other lysine PTMs.

Cellular Localization of Lysine Malonylomes

SIRT5, a regulatory enzyme of Ksucc, Kglu, and Kmal, localizes predominantly to mitochondria but is also present in the cytosol and nucleus (10, 49). Previously, we reported that 17.8% of Ksucc substrates (351) are localized in the mitochondria in mouse liver (supplemental Fig. S2F, left) (10). Among the Ksucc substrates identified in human cervical cancer cells (HeLa) (47), 17% of Ksucc substrates exclusively localize to mitochondria (supplemental Fig. S2F, right).

To understand the cellular localization of Kmal substrates in mouse liver, we performed the same analysis for the Kmal data set generated from mouse liver. Here, we compared our Kmal data set with the mitochondrial genes annotated in the GO database (50). Of all the identified Kmal substrates, 316 (58%) of them are present in the mitochondria, and 274 (50%) of them are exclusively mitochondrial proteins (Fig. 5C, left). Therefore, a comparable fraction of Kmal and Ksucc proteins from mouse liver localizes to mitochondria.

In parallel, we carried out a similar analysis for Kmal proteins derived from human fibroblasts. Our result shows a striking difference of subcellular localization among Kmal substrates. Among the 1024 Kmal substrates identified in human fibroblasts, 338 (33%) of them localize to mitochondria of which 265 (26%) of them are exclusively mitochondrial (Fig. 5C, right). The number of mitochondrial Kmal substrates from either mouse liver or human fibroblasts is comparable. However, in human fibroblasts, we identified a significantly higher number of nuclear and cytosolic substrates with 262 (30%) and 342 (39%) proteins, respectively (Fig. 5C, right). The cellular enzymes that catalyze lysine malonylation in mammalian cells are still unknown.

Additionally, we compared the Kmal proteins and sites in mitochondria with our previously reported Ksucc data (10). We found that 198 mitochondrial Kmal proteins (13.9% of all Kmal proteins) (supplemental Fig. S2D, left) and 432 mitochondrial Kmal sites (10.7% all Kmal sites) (supplemental Fig. S2D, right) overlapped with mitochondrial Ksucc in mouse liver, whereas 37% of mitochondrial Kmal proteins and 62% of Kmal sites did not overlap. In human fibroblasts, 59% of mitochondrial Kmal proteins (199) (supplemental Fig. S2E, left) and 31% of mitochondrial Kmal sites (344) (supplemental Fig. S2E, right) overlapped with mitochondrial Ksucc data (47).

We also performed immunostaining of MCD+/+ and MCD−/− human fibroblasts with anti-malonyllysine, anti-acetyllysine, and anti-succinyllysine antibodies along with Hoechst nuclear stain and MitoTracker Red (Fig. 6, A and B). Our staining results suggest that the strongest signals for Kac and Ksucc are confined in the nucleus in both MCD+/+ and MCD−/− human fibroblasts (Fig. 6, A and B, second and third rows). However, Kmal signals are distributed among the cytosol and nucleus in MCD+/+ cells (Fig. 6A, top row). Interestingly, most of the Kmal signal overlaps with MitoTracker Red in MCD−/− cells (Fig. 6B, top row), suggesting that Kmal levels increase specifically in the mitochondria of MCD−/− cells.

Functional Annotation of Lysine Malonylomes

To understand the biological functions of Kmal proteins, we performed enrichment analysis by using the GO database (30) and KEGG (51) for Kmal substrates identified in mouse liver and human fibroblasts. The GO biological process analysis of mouse liver Kmal substrates (supplemental Table S4) showed enrichment in oxidation/reduction (adj p = 5.45 × 10⁻⁵¹), protein translation (adj p = 5.71 × 10⁻⁴⁹), cofactor metabolism (adj p = 4.18 × 10⁻²⁶), and fatty acid metabolism (adj p = 5.09 × 10⁻¹²) (supplemental Fig. S3A, left). The GO analysis of human fibroblast malonylome (supplemental Fig. S3B, left and supplemental Table S5) showed enrichment in protein expression processes, such as translation (adj p = 2.22 × 10⁻⁵⁵), translation elongation (adj p = 7.43 × 10⁻²⁶), tRNA aminoacylation (adj p = 1.01 × 10⁻²⁰), and intracellular transport (adj p = 5.30 × 10⁻³¹). Proteins associated with fatty acid β-oxidation were also enriched in the human fibroblast malonylome (adj p = 2.34 × 10⁻⁷; supplemental Table S5).

The molecular function analysis of mouse liver Kmal substrates showed enrichment in nucleotide binding (adj p = 9.35 × 10⁻³⁵), cofactor binding (adj p = 1.04 × 10⁻²⁷), and ATP binding (adj p = 1.86 × 10⁻¹⁵) (supplemental Fig. S3A, right). Kmal substrates in human fibroblasts were associated with nucleotide binding (adj p = 3.22 × 10⁻⁴⁵), nucleoside binding (adj p = 2.77 × 10⁻²⁹), ATP binding (adj p = 4.73 × 10⁻²⁶), and aminoacyl-tRNA ligase activity (adj p = 2.74 × 10⁻²⁰) (supplemental Fig. S3B, right), supporting the idea that Kmal may be involved in regulating protein translation.

There was no significant difference between GO (supplemental Fig. S4A) and KEGG pathway enrichments (supplemental Fig. S4B) of all the proteins identified in human fibroblasts versus light-only protein substrates derived from MCD−/− cells (supplemental Table S6). In addition, there was a significant overlap between the KEGG pathway analysis of mouse liver (supplemental Fig. S3C) and that of human fibroblasts (supplemental Fig. S3D). The top enriched categories of KEGG pathways for lysine-malonylated substrates were ribosome, valine/leucine/isoleucine degradation, proteasome, and fatty acid metabolism (supplemental Fig. S3, C and D). Twenty-nine of 45 key enzymes in mouse and 22 of 45 key enzymes in humans involved in regulation of fatty acid metabolism were lysine malonylated (supplemental Fig. S5, A and B and supplemental Table S5F). Among these, five enzymes (fatty-acid synthase, acetyl-CoA carboxylase 1, ATP-citrate lyase, AMP-activated protein kinase, and CPT1) are closely associated with malonyl-CoA metabolism (supplemental Fig. S5C).

Of particular note are a few proteins involved in fatty acid metabolism. We found that acetyl-CoA acetyltransferase 1, an enzyme participating in multiple metabolic pathways including fatty acid metabolism, was malonylated at seven sites: Lys¹⁷⁴, Lys¹⁹⁰, Lys²⁴³, Lys²⁵¹, Lys²⁶³, Lys²⁶⁸, and Lys²⁷³ (Fig. 4D). The Kmal level of (MCD+/+: MCD−/− SILAC ratio of 0.0044) Lys²⁶³ of acetyl-CoA acetyltransferase 1 was increased more than 200-fold in MCD−/− cells. Lys²⁶³ is in close proximity with the coenzyme A binding site and possibly makes two hydrogen bonds with coenzyme A, suggesting a possibility that this residue is important in regulating the protein's function. Lys²⁶³ was previously reported to be acetylated and succinylated as well (41, 47). In addition, among all the Kmal sites of acetyl-CoA acetyltransferase 1, Lys¹⁷⁴ is acetylated, and Lys²⁵¹ is succinylated (41). Therefore, these Kmal sites might also contribute to regulation of protein function depending on the type of modification. Hydroxymethylglutaryl-CoA lyase is malonylated at three lysine sites (Lys⁴⁸, Lys⁹³, and Lys¹³⁷) of which Lys⁴⁸ malonylation is increased roughly 39-fold in MCD−/− cells. Hydroxymethylglutaryl-CoA lyase exclusively localizes to mitochondria and is specifically responsible for leucine degradation as well as ketone production during fat breakdown. Hydroxymethylglutaryl-CoA lyase deficiency is a rare genetic disease that causes metabolic acidosis and hypoglycemia (52). A K-to-N mutation at Lys⁴⁸ of hydroxymethylglutaryl-CoA lyase ablates enzymatic activity, which suggests that K48 is a critical position for enzymatic function (53). Therefore, lysine malonylation of Lys⁴⁸ may lead to changes in enzymatic activity of this protein. ATP-citrate lyase catalyzes conversion of citrate to acetyl-CoA (supplemental Fig. S5C), which can be converted further to malonyl-CoA by acetyl-CoA carboxylase 1. Among the 14 Kmal sites in ATP-citrate lyase, Lys⁶⁸, which is located next to an ATP binding site (Lys⁶⁶-Lys⁶⁷), is malonylated and therefore might alter the ATP binding ability of the protein. Enrichment of fatty acid metabolism proteins in the malonylomes in both mouse liver and human fibroblasts suggests a possible feedback regulation of fatty acid biosynthesis by malonyl-CoA-mediated lysine malonylation.

Mitochondrial Function and Fatty Acid Oxidation Are Impaired in MCD−/− Cells

Integration of our bioinformatics analyses of lysine malonylated proteins identified in mouse liver and human fibroblasts demonstrates that among metabolic pathways proteins involved in fatty acid metabolism were preferentially heavily malonylated. Fatty acid synthesis, which utilizes malonyl-CoA as a substrate for synthesis and chain elongation, primarily occurs in the cytosol, whereas fatty acid oxidation occurs in mitochondria and peroxisomes. Because MCD−/− cells showed greatly increased Kmal immunostaining in mitochondria compared with MCD+/+ cells (Fig. 6) and MCD-deficient patients have been reported to present with pathologies similar to those of patients with fatty acid oxidation defects, we wanted to understand whether mitochondrial function and fatty acid oxidation are affected in MCD−/− cells. Long-chain fatty acids are broken down to medium- and short-chain fatty acids in mitochondria by VLCAD and medium-chain acyl-CoA dehydrogenase together with the mitochondrial trifunctional protein complex encoded by the Hadha and Hadhb genes. Mitochondrial trifunctional protein complex consists of LCHAD, long-chain enoyl-CoA hydratase, and long-chain keto-acyl-CoA thiolase enzymatic activities. In MCD−/− cells, multiple mitochondrial fatty acid oxidation proteins were heavily malonylated, and both VLCAD and trifunctional enzyme subunit alpha were substantially more malonylated than in wild-type cells (Fig. 7A and supplemental Fig. S7B). Many of the detectable malonylated lysine sites were only present in MCD−/− cells. Examination of the crystal structure of VLCAD (Protein Data Bank codes 2UXW and 3B96) reveals that sites of lysine malonylation are scattered across the polypeptide (Fig. 7B). Three of the Kmal sites (Lys²⁷⁸, Lys³³¹, and Lys⁴⁸⁰) occupy highly conserved amino acid positions among VLCAD orthologues. A majority of the lysine sites are surface-exposed, and their malonylation may impact different properties of the protein: three (Lys⁴⁸⁰, Lys⁴⁸², and Lys⁵⁵) are located at the putative surface of membrane attachment (54), two (Lys⁶³⁵ and Lys⁶³⁹) are found in proximity to the dimerization interface, and three (K²⁷⁶, K²⁷⁸, and K³³¹) are positioned near the active site where FAD and acyl-CoA molecules bind (Fig. 7B). We next analyzed whether VLCAD enzymatic activity was affected by malonylation in MCD−/− cells. Indeed, VLCAD activity was decreased 45% in MCD−/− cells as compared with MCD+/+ cells (Fig. 7C). Basal expression levels of VLCAD protein were similar in MCD+/+ and MCD−/− cells (supplemental Fig. S7A). Furthermore, the LCHAD activity of mitochondrial trifunctional protein was significantly decreased in MCD−/− cells as well (Fig. 7D).

Accumulation of cytosolic malonyl-CoA is known to inhibit CPT1, which is located on the outer membrane of the mitochondria. CPT1, together with carnitine-acylcarnitine translocase and CPT2, imports acyl-CoAs into the mitochondria for β-oxidation. Our data now suggest that an increase of lysine malonylation on proteins within the mitochondrial matrix can also inhibit fatty acid oxidation. To test whether mitochondrial function and fatty acid oxidation are indeed affected by lysine malonylation, we studied the impact of malonate on mitochondrial function in Fao liver cells. Previously, we have shown that malonate treatment induces significant lysine malonylation (15). To eliminate any confounding effects from direct interference of malonate itself on mitochondrial function, we treated cells with malonate for 1 day followed by a malonate-free overnight incubation prior to analysis of mitochondrial function. Malonate treatment of cells significantly reduced the OCR in the context of pyruvate, succinate, and octanoylcarnitine mitochondrial oxidation (Fig. 7, E, F, and G). CPT1 is not required for oxidation of octanoylcarnitine; hence, inhibition of CPT1 by malonyl-CoA cannot explain the observed decrease in OCR in the presence of octanoylcarnitine. Instead, this finding likely indicates that either oxidative phosphorylation or fatty acid oxidation activity is decreased by lysine malonylation.

Finally, we analyzed both succinate- and octanoylcarnitine-driven OCR in MCD−/− and MCD+/+ cells. Interestingly, succinate-driven OCR was only mildly reduced in MCD−/− cells (Fig. 7, H and J), whereas octanoylcarnitine-driven OCR was 40% decreased in MCD−/− cells as compared with MCD+/+ cells (Fig. 7, I and J). Together, these findings suggest that malonyl-CoA can inhibit mitochondrial fatty acid oxidation in MCD−/− cells, possibly through elevated lysine malonylation, independently of effects on CPT1.