Identification of Fungal Sphingolipid C9-methyltransferases by Phylogenetic Profiling*

Fungal glucosylceramides play an important role in plant-pathogen interactions enabling plants to recognize the fungal attack and initiate specific defense responses. A prime structural feature distinguishing fungal glucosylceramides from those of plants and animals is a methyl group at the C9-position of the sphingoid base, the biosynthesis of which has never been investigated. Using information on the presence or absence of C9-methylated glucosylceramides in different fungal species, we developed a bioinformatics strategy to identify the gene responsible for the biosynthesis of this C9-methyl group. This phylogenetic profiling allowed the selection of a single candidate out of 24–71 methyltransferase sequences present in each of the fungal species with C9-methylated glucosylceramides. A Pichia pastoris knock-out strain lacking the candidate sphingolipid C9-methyltransferase was generated, and indeed, this strain contained only non-methylated glucosylceramides. In a complementary approach, a Saccharomyces cerevisiae strain was engineered to produce glucosylceramides suitable as a substrate for C9-methylation. C9-methylated sphingolipids were detected in this strain expressing the candidate from P. pastoris, demonstrating its function as a sphingolipid C9-methyltransferase. The enzyme belongs to the superfamily of S-adenosylmethionine-(SAM)-dependent methyltransferases and shows highest sequence similarity to plant and bacterial cyclopropane fatty acid synthases. An in vitro assay showed that sphingolipid C9-methylation is membrane-bound and requires SAM and Δ4,8-desaturated ceramide as substrates.

Sphingolipids of fungi and plants can be grouped into different classes based on the nature of their polar headgroups. In the case of GIPC, 2 an inositol residue is phosphodiester-linked to the hydrophobic part of the molecule, the ceramide backbone. In the case of cerebrosides, the ceramide is directly linked to a glucosyl or galactosyl residue that can form the base for further glycosylations. The ceramide backbone usually carries several functional groups such as hydroxy groups, (E)-or (Z)-double bonds, methyl branches at aliphatic or olefinic carbon atoms, or even cyclopropane rings. Different patterns of functional groups are characteristic for GIPC versus cerebrosides as well as for different organisms (see Fig. 1) (1)(2)(3)(4)(5). The functional significance of these differences has yet to be investigated.
The genes responsible for most functional groups have been cloned (reviewed in Refs. 4, 6 -8), and now their biological functions will be characterized by reverse genetic, molecular, and cell biological approaches. A structural feature hardly studied so far is the C9-methyl branch at the (E)-⌬8-double bond of fungal cerebrosides. Many fungi share a canonical glucosylceramide structure with (E,E)-9-methylsphinga-4,8-dienine as sphingoid base, carrying (E)-double bonds at the ⌬4and ⌬8-positions and a methyl branch at the C9-position (4). Besides in fungi, a glucosylceramide with the same structure has been found in the sea anemone Metridium senile (9), the first organism described to contain C9-methylated sphingolipids, as well as in the protist Thraustochytrium globosum (10). Structurally related glucosylceramides containing an additional ⌬10-double bond have been found in T. globosum, in several species of starfish, in the marine sponge Agelas mauritianus, and in the sea squirt Phallusia fumigata (10 -15).
There are several studies pointing to important biological functions of fungal glucosylceramides. Fungal glucosylceramides carrying a C9-methyl group as well as certain plant glucosylceramides induce fruiting body formation in the fungus Schizophyllum commune (16). Human patients with a Cryptococcus neoformans infection as well as rabbits infected with Pseudallescheria boydii develop antibodies binding to C9-methylated fungal glucosylceramides. The purified antibodies inhibit fungal growth in culture (17,18). Fungal glucosylceramides can elicit defense responses in rice plants and cell suspension cultures of rice, and treatment with these fungal glucosylceramides protects rice plants against fungal infection (19 -21). The plant defensin RsAFP2 from radish seeds interacts specifically with fungal glucosylceramides (22). Pichia pastoris and Candida albicans strains deficient in the gene responsible for glucosylceramide synthesis are resistant to RsAFP2 (22). Because the C9-methyl group is the prime structural feature distinguishing fungal glucosylceramides from those of plants and mammals ( Fig. 1), it is likely that C9-methylation of fungal glucosylceramides plays an important role in the interactions between fungi and other organisms.
To enable the investigation of these proposed functions with molecular techniques, we were interested in identifying the gene encoding the methyltransferase that introduces the C9-methyl branch into fungal sphingolipids, as well as in directly demonstrating its biochemical function. We developed a bioinformatics strategy based on phylogenetic profiling that exploits the pattern of presence or absence of C9-methylated glucosylceramides in several fungi with completely sequenced genomes. After identifying candidate genes, we tested their involvement in sphingolipid C9-methylation by generating a knock-out mutant of the yeast P. pastoris. In a complementary experiment, a Saccharomyces cerevisiae strain was engineered to directly demonstrate the formation of C9-methylated sphingolipids. Finally, membrane association and substrate specificity of sphingolipid C9-methylation in P. pastoris were investigated by an in vitro assay.

Phylogenetic Profiling
Data Base Construction and Sequence Retrieval-The complete sets of protein sequences from several fungi, plants, and animals were downloaded from the web pages of the sequencing projects listed in Supplemental Table S1 and compiled into a single data base. Swiss-Prot species identification codes (www.expasy.org/cgi-bin/speclist) were included in the sequence titles as species identifiers.
Clustering of SAM-dependent Methyltransferase Sequences-SAMdependent methyltransferase sequences were compared pairwise using blastp (23) (e-value e Յ 10 Ϫ4 ). The resulting score matrix was used as input for the TRIBE-MCL protocol (Ref. 24, available at micans.org/ mcl) with parameter I ϭ 1.1. For each cluster containing more than four sequences, a phylogenetic tree was constructed with ClustalX (25).
Groups of Orthologous Sequences-Within the phylogenetic tree for each cluster, groups of orthologous sequences were identified by comparing the branching pattern of the tree with the phylogeny of the species from which the sequences originate. A group of orthologous FIGURE 1. Structures of typical glucosylceramides from mammals, plants, and fungi and of IPC from S. cerevisiae. A, mammalian glucosylceramide containing (E)-sphing-4enine as sphingoid base and 2-hydroxystearic acid as fatty acyl residue. In some tissues, the ⌬4-double bond is replaced by a C4-hydroxy group. B, plant glucosylceramide containing (E,E)-sphinga-4,8-dienine as sphingoid base and 2-hydroxypalmitic acid as fatty acyl residue. The ⌬4-double bond may be replaced by a C4-hydroxy group or be entirely missing, and the ⌬8-double bond may be either in the (E) or in the (Z) configuration. C, fungal glucosylceramide containing (E,E)-9-methylsphinga-4,8-dienine as sphingoid base and 2-hydroxystearic acid as fatty acyl residue. In some fungi, the fatty acyl residue additionally contains a ⌬3-double bond. D, IPC from S. cerevisiae containing 4-hydroxysphinganine as sphingoid base, 2-hydroxyhexadecanoic acid as fatty acyl residue, and a phosphoinositol headgroup. Its ceramide backbone differs from that of the glucosylceramides shown in A-C in containing a saturated sphingoid base and a very long chain fatty acyl residue with 24 or 26 carbon atoms. In all structures shown in A-D, the C2Ј-hydroxy group may be missing.
sequences is defined as a subtree of the phylogenetic tree in which the branches originate only from speciation events but not from gene duplications within the same species. Genes duplicated within a single species were counted as one gene if the duplicated sequences were more similar to each other than to any other sequence.
Groups of orthologous sequences as defined above may be nested, i.e. when sequences A, B, and C form a group of orthologous sequences with branching pattern ((A,B),C), then the sequences A and B form themselves a smaller group that also fits the definition. In such cases, only the enclosing group was considered for phylogenetic profiling.
Phylogenetic Profiles-The phylogenetic profile of a group of sequences is defined as the set of species from which the sequences originate (26). The difference between two phylogenetic profiles was quantified by counting how many sequences were present in only one of the two profiles. The number of differences was counted as the number of sequences present only in the first profile plus the number of sequences present only in the second profile.
Data Handling-Scripts for the handling of sequence data and phylogenetic trees were written in the programming language Python (www.python.org) with the Biopython libraries (www.biopython.org).

Knock-out of the Sphingolipid C9-Methyltransferase in P. pastoris
A knock-out cassette for an exact deletion of the candidate C9-methyltransferase CDS was constructed by triple fusion PCR as described in Ref. 27. All reactions were performed with PfuTurbo DNA polymerase (Stratagene). The primers used are listed in Supplemental Table S2. The underlined parts of the chimeric primers Up-R-Zeo-F and Down-F-Zeo-R are the reverse complement to the primers Zeo-F and Zeo-R, respectively. The Zeocin resistance cassette from the pGAPZ-B vector (Invitrogen) was amplified with primers Zeo-F and Zeo-R. 3099 bp of the genomic locus containing the candidate C9-methyltransferase CDS was amplified from genomic DNA of the P. pastoris strain GS115 (28) with primers Up-F and Down-R. From this PCR product, upstream and downstream sequences directly flanking the CDS were amplified with the nested primers Up-nested-F/Up-R-Zeo-F (530 bp) and Down-nested-R/Down-F-Zeo-R (516 bp), respectively. The upstream and downstream sequences and the Zeocin resistance cassette were fused in a single PCR for 12 rounds without primers. From this, the full-length fusion product was amplified in a separate reaction with primers Upnested-F and Down-nested-R. The knock-out cassette was ligated into the SmaI site of the pBluescript II KS(Ϫ)-vector (Stratagene) and checked by sequencing.
The knock-out cassette was excised from the plasmid with PstI/XbaI and used to transform electrocompetent cells of the strain GS115 (his4). Transformants were selected on YPD plates (34) containing 100 mg/liter Zeocin. Correct integration of the knock-out cassette was confirmed by PCR with the primer pairs Up-F/Zeo-int-R and Down-R/Zeo-int-F as positive and Up-F/Int-R and Down-R/Int-F as negative controls.

Knock-out of SUR2 in S. cerevisiae
A knock-out cassette for an exact deletion of the SUR2 CDS (YDR297W) was constructed by PCR with long primers as described in Ref. 29. The primers used are listed in Supplemental Table S3. The kanamycin resistance cassette of the plasmid pUG6 (30) was amplified with primers SUR2-kan-F and SUR2-kan-R. These primers contain Ϸ80 bp complementary to the sequences directly flanking the SUR2 CDS (underlined). The PCR reaction was performed with Taq DNA polymerase (Invitrogen). The knock-out cassette was ligated into the pGEM-T vector (Promega).
The knock-out cassette was excised from the plasmid with NotI/SacII and used to transform chemically competent cells of the strain UTL-7A (MATa ura3-52 trp1 leu2-3/112). Transformants were selected on YPD plates containing 200 mg/liter G-418. Correct integration of the knock-out cassette was confirmed by PCR with primer pairs SUR2-up-F/SUR2-up-R and SUR2-down-F/SUR2-Down-R as positive controls and primer pair SUR2-up-F/SUR2-Down-R as combined positive and negative control.

An S. cerevisiae Expression System
Identification of the ⌬4and ⌬8-Desaturases from P. pastoris-For PCR-based cloning of the P. pastoris ⌬4and ⌬8-desaturases, a genomic DNA library of the P. pastoris strain GS115 (28) was used as template. The primers used are listed in Supplemental Table S4.
To amplify the ⌬4-desaturase, degenerate primers were deduced from a multiple alignment of amino acid sequences from C. albicans, Neurospora crassa, Schizosaccharomyces pombe, Arabidopsis thaliana, Lycopersicon esculentum, Drosophila melanogaster, and Homo sapiens belonging to the DES family (31). PCR-amplification was carried out with the degenerate primers Delta4-deg-F and Delta4-deg-R at an annealing temperature of 48°C.
To amplify the ⌬8-desaturase, degenerate primers were deduced from a multiple alignment of amino acid sequences from ⌬5and ⌬6-fatty acid desaturases and ⌬8-sphingolipid desaturases described in Ref. 32. PCR amplification was carried out using the degenerate primers Delta8-deg-F and Delta8-deg-R at an annealing temperature of 40°C and resulted in PCR products of 177 bp for the ⌬4-desaturase and 564 bp for the ⌬8-desaturase.
For the identification of full-length ⌬4and ⌬8-desaturase sequences, the P. pastoris genomic DNA library was screened with these PCR products using the DIG System (Roche Molecular Biochemicals). The inserts of positive clones for each desaturase were sequenced, with CDS of 1083 bp (⌬4-desaturase) and 1629 bp (⌬8-desaturase) encoding polypeptides of 360 and 542 amino acids, respectively. 3 Cloning of the P. pastoris ⌬4and ⌬8-Desaturases-The complete CDSs of the P. pastoris ⌬4and ⌬8-desaturases were amplified from genomic clones by PCR with PfuTurbo DNA polymerase (Stratagene, ⌬4-desaturase) or Herculase DNA Polymerase (Stratagene, ⌬8-desaturase) with the primers Delta4-F and Delta4-R (⌬4-desaturase), and Del-ta8-F and Delta8-R (⌬8-desaturase) (Supplemental Table S4). The primers included adapter sequences containing the indicated restriction sites (underlined). The PCR products were ligated into the pESC-URA vector (Stratagene), with the CDS of the ⌬4-desaturase in the NotI and BglII sites of MCS1 and the CDS of the ⌬8-desaturase in the BamHI and XhoI sites of MCS2. In this way, three different plasmids were generated: One plasmid carrying the ⌬4-desaturase, one carrying the ⌬8-desaturase, and one carrying both desaturases. The cloned CDSs were checked by sequencing.
Cloning of a Candidate Sphingolipid C9-Methyltransferase-The complete CDS of a candidate sphingolipid C9-methyltransferase from P. pastoris was amplified from genomic DNA of the P. pastoris strain GS115 with the primers C9-F and C9-R (Supplemental Table S4). The primers included adapter sequences containing the indicated restriction sites (underlined). The PCR products were ligated into the BamHI and EcoRI sites of the pYES3-CT vector (Invitrogen). The resulting plasmid was checked by sequencing.
For the experiment shown in Fig. 4, chemically competent cells of the S. cerevisiae SUR2 knock-out strain were transformed simultaneously with the empty vector pYES3-CT (Fig. 4A) or the candidate C9-methyltransferase ligated in pYES3-CT (Fig. 4B), with the ⌬4-desaturase and the ⌬8-desaturase ligated into pESC-URA, and with the human glucosylceramide synthase (33) ligated into both MCS of the pESC-LEU vector (Invitrogen). The plasmid containing the glucosylceramide synthase was kindly provided by M. Leipelt, Universität Hamburg.

Lipid Analysis
Culture of P. pastoris-A preculture was grown in liquid YPD medium for 24 h at 30°C. 400 ml of YPD medium was inoculated 1:100 from the preculture and incubated again for 24 h at 30°C. The cells were harvested by centrifugation.
Culture of S. cerevisiae-Precultures were grown in complete minimal medium (34) containing 2% (w/v) glucose at 30°C for 3 days. To induce expression under the control of the GAL1 and GAL10 promotors, the cells were washed, and 100 ml of complete minimal medium containing 2% (w/v) raffinose and 2% (w/v) galactose was inoculated to give a starting optical density of A 600 ϭ 0.02. The cultures were incubated at 30°C for 48 h, and the cells were harvested by centrifugation.
Isolation of Glucosylceramides from P. pastoris-Approximately 10 g (fresh weight) of P. pastoris cells was resuspended in 20 ml of 0.45% (w/v) NaCl and boiled in a water bath for 15 min. The cells were sedimented by centrifugation, and the lipids were extracted by shaking in 30 ml of CHCl 3 /MeOH, 1:2 for 24 h at 8°C followed by 30 ml of CHCl 3 / MeOH, 2:1. The lipid extract was washed by phase partitioning with CHCl 3 /MeOH/0.45% (w/v) NaCl (8:4:3) and subsequently evaporated. The lipids were redissolved in CHCl 3 and applied to a 500 mg/6 ml Strata SI-2 column (Phenomenex). The column was flushed with 30 ml of CHCl 3 before the glycolipids were eluted with 30 ml of acetone/2propanol, 9:1. The glucosylceramides were purified from the acetone/ 2-propanol fraction by preparative TLC on Silica Gel 60 plates (Merck) in CHCl 3 /MeOH, 85:15.
Analysis of the Sphingoid Base Composition-The sphingoid base composition of P. pastoris glucosylceramides and of total S. cerevisiae cells was analyzed as described in Ref. 31. Briefly, sphingolipids were (35). The free sphingoid bases were extracted, washed, and converted to their DNP derivatives with one part of 0.5% (v/v) methanolic 1-fluoro-2,4-dinitrobenzene and four parts of 2 M boric acid/KOH, pH 10.5 (30 min at 60°C) (36). The DNP-derivatized sphingoid bases were extracted and purified by TLC in CHCl 3 /MeOH, 90:10.
Reversed-phase HPLC/MS with electrospray ionization (ESI) of DNP-derivatized sphingoid bases was performed on a MAT 95XL-Trap instrument (ThermoQuest) in negative ion mode. The eluate of the HPLC was split using a low dead volume T-piece with 5% entering the ESI-sprayer and 95% a UV detector (Hitachi).
Sphingolipid Methyltransferase Assay-P. pastoris cells from 50 ml of culture were harvested by centrifugation and resuspended in 2 ml of lysis buffer (100 mM Tricine/NaOH, pH 7.0, 15% (v/v) glycerol). The cells were disrupted by vortexing with glass beads (8 min) and ultrasonication (1 min). A cell-free homogenate was obtained by centrifuging at 1700 ϫ g (5 min). For investigating the membrane association of the sphingolipid C9-methyltransferase, the cell-free homogenate was centrifuged at 100,000 ϫ g (60 min) in a 70-Ti rotor (Beckman).
In a total volume of 100 l, the assay mixture contained 30 l of cell-free homogenate from P. pastoris cells (240 -360 g of protein), 5 l of S-adenosyl-L-[methyl-14 C]methionine solution in diluted H 2 SO 4 (Amersham Biosciences) (250,000 dpm at a specific activity of 2.26 GBq/mmol, final concentration 18 M), 5 l of sphingolipid solution in ethanol (5 g of lipid), and 0.1% (w/v) of Triton X-100. For investigating the membrane association of the enzyme, the cell-free homogenate was replaced by 100,000 ϫ g supernatant or 100,000 ϫ g pellet resuspended in lysis buffer. After 1 h at 30°C, 0.7 ml of 0.45% (w/v) NaCl was added, and the lipids were extracted by phase partitioning with CHCl 3 /MeOH/ 0.45% (w/v) NaCl (8:4:3). Products from assays with glucosylceramides or ceramides were analyzed by TLC in CHCl 3 /MeOH (80:20), whereas products from assays with free sphingoid bases as substrates were analyzed by TLC in CHCl 3 /MeOH/25% (w/w) NH 4 OH (50:30:1). Radioactivity was detected by radioscanning with a BAS-1000 BioImaging Analyzer (Fujifilm).
The 14 C-labeled ceramide spots were scraped off the TLC plate and hydrolyzed with Ba(OH) 2 . The free sphingoid bases were converted to their DNP derivatives and analyzed by reverse-phase HPLC as described above. Fractions corresponding to the individual sphingoid bases as well as to the intervals between them (background control) were collected and analyzed for radioactivity in an LS6000 IC liquid scintillation counter (Beckman).

RESULTS
Phylogenetic Profiling of SAM-dependent Methyltransferases-We employed a bioinformatics strategy based on phylogenetic profiling to identify candidate genes for the sphingolipid C9-methyltransferase. The phylogenetic profile of a gene is the set of species in which the gene is present. Because most genes in fungal genomes have not been experimentally investigated, we introduced groups of orthologous sequences to Fungal Sphingolipid C9-methyltransferases MARCH 3, 2006 • VOLUME 281 • NUMBER 9 find the equivalent genes in different species based on phylogenetic information. Phylogenetic profiles and groups of orthologous sequences are defined under "Experimental Procedures. " We identified candidate genes by comparing the phylogenetic profiles of groups of orthologous sequences with the phylogenetic profile expected for the sphingolipid C9-methyltransferase. The expected profile was deduced by assuming that the presence or absence of C9-methylated glucosylceramides in a particular species was reflected by the presence or absence of a sphingolipid C9-methyltransferase gene in that species. In addition, we assumed that the sphingolipid C9-methyltransferase belonged to the superfamily of SAM-dependent methyltransferases (see "Discussion"). The phylogenetic profiling strategy involved three steps.
First, a suitable set of species with completely sequenced genomes was selected. The selection included species with C9-methylated glucosylceramides as well as species without this functional group. C9-methylated glucosylceramides are present in the fungi C. albicans, C. neoformans, Magnaporthe grisea, and N. crassa, but they are absent in S. cerevisiae and S. pombe (4). To increase the number of species without C9-methylated glucosylceramides, three animals (H. sapiens, D. melanogaster, and Caenorhabditis elegans) and two plants (A. thaliana and Oryza sativa) were included as well. The predicted protein sequences of the corresponding genomes were downloaded (Supplemental Table S1) and assembled into a custom data base.
Second, as many protein sequences of known and hypothetical SAMdependent methyltransferases as possible were collected from the custom data base using PSI-Blast (23)  In the third step, groups of orthologous sequences were identified by analyzing the phylogenetic relationships within the collection of SAMdependent methyltransferases. To facilitate the phylogenetic analysis, the collection was clustered into families on the basis of sequence similarity using TRIBE-MCL (24). 10 families with more than four sequences were obtained. A phylogenetic tree was generated for each of these families with ClustalX (25). The branching patterns of the phylogenetic trees were analyzed to identify groups of orthologous sequences as described under "Experimental Procedures." The phylogenetic profile of each group was compared with the profile expected for the C9-methyltransferases. The ideal profile should contain only the species C. albicans, C. neoformans, M. grisea, and N. crassa.
Three groups came close to matching the expected phylogenetic profile (Table 1). Group 1 differed from the expected profile by containing S. pombe while lacking N. crassa. Group 2 matched the expected profile perfectly, 4 and Group 3 contained S. pombe while C. neoformans was missing.
Sequences from additional fungal species not included in the phylogenetic profiling were retrieved by a tblastn (23) search in GenBank TM and in the P. pastoris draft genome sequence 5 with the N. crassa sequence from Group 2 as query. These included sequences from yeasts, filamentous ascomycetes, and basidiomycetes. In most species, only one candidate sphingolipid C9-methyltransferase gene was found, but the genomes of the filamentous ascomycetes Aspergillus nidulans and Fusarium graminearum contain two sequences homologous to Group 2. As mentioned in the introduction, the sea squirt P. fumigata also contains C9-methylated glucosylceramides. However, in the sequenced genomes of two other sea squirts, Ciona intestinalis and Ciona savignyi, no candidate sphingolipid C9-methyltransferases were found. We selected the candidate sphingolipid C9-methyltransferase from P. pastoris to investigate its predicted biochemical function experimentally. 6 A Knock-out of the Candidate Sphingolipid C9-Methyltransferase from P. pastoris Is Lacking C9-methylated Glucosylceramides-A P. pastoris knock-out strain lacking the candidate C9-methyltransferase CDS was generated by homologous recombination. Glucosylceramides were isolated from this knock-out strain as well as from the corresponding wild-type strain, and their sphingoid base compositions were compared.
Construction of an S. cerevisiae Expression System for the Candidate C9-Methyltransferase-To obtain direct evidence that the candidate sphingolipid C9-methyltransferase is capable of methylating sphingolipids, it was functionally expressed in the yeast S. cerevisiae. We expected that native S. cerevisiae sphingolipids would not be suitable as substrates for C9-methylation, because they are lacking the ⌬4and ⌬8-double bonds present in the glucosylceramides of P. pastoris. Instead, S. cerevisiae contains GIPC with predominately C4-hydroxylated sphingoid bases while completely lacking glucosylceramides (Fig.  1). To provide a substrate that fulfills these anticipated requirements, an S. cerevisiae strain was engineered to produce glucosylceramides containing the sphingoid base (E,E)-sphinga-4,8-dienine. 4 Initially, Group 2 was lacking M. grisea. However, a re-inspection of the M. grisea genome on a nucleotide level with tblastn (23) 6 The nucleotide sequence of the sphingolipid C9-methyltransferase from P. pastoris has been deposited in the GenBank TM database with accession number DQ070247. The GenPept accession number of the deduced amino acid sequence is AAZ08581.

Fungal Sphingolipid C9-methyltransferases
For the functional expression of sphingolipid ⌬4and ⌬8-desaturases in S. cerevisiae, fragments of candidate ⌬4and ⌬8-desaturases from P. pastoris were amplified by PCR with degenerate primers, and the full-length sequences were subsequently retrieved by screening of a P. pastoris genomic library. An S. cerevisiae knock-out strain lacking the sphingolipid C4-hydroxylase SUR2p (39,40) was generated by homologous recombination. The resulting strain completely lacked C4-hydroxylated sphingolipids (Fig. 3A). This strain was used to express the P. pastoris ⌬4and ⌬8-desaturases individually or in combination, resulting in the formation of the monounsaturated sphingoid bases (E)sphing-4-enine and (E)-sphing-8-enine and of the di-unsaturated sphingoid base (E,E)-sphinga-4,8-dienine, respectively (Fig. 3, B-D). Finally, this strain was transformed with a plasmid containing the CDS of the human glucosylceramide synthase (33) to produce glucosylceramide. We expected this engineering to provide a suitable substrate for the candidate sphingolipid C9-methyltransferase from P. pastoris expressed in S. cerevisiae.
Functional Expression of the Candidate Sphingolipid C9-methyltransferase from P. pastoris in S. cerevisiae Resulted in the Biosynthesis of C9-methylated Sphingolipids-The CDS of the candidate C9-methyltransferase was expressed in the S. cerevisiae strain deficient in sphingolipid C4-hydroxylation together with the P. pastoris ⌬4and ⌬8-desaturases and the human glucosylceramide synthase. As described before (33), the expression of the human glucosylceramide synthase resulted in the formation of glucosylceramides in S. cerevisiae detected by TLC (data not shown). In the following experiment, total sphingoid bases from S. cerevisiae cells were analyzed.
The expression of the candidate C9-methyltransferase resulted in the formation of a small amount of sphingolipids containing the C9-methylated sphingoid base (E,E)-9-methylsphinga-4,8-dienine (Fig. 4). The (E,E)-9-methylsphinga-4,8-dienine formed accounts for 0.4 mol % of the total sphingoid bases. Because the substrate of the methylation reaction, C 18 -(E,E)-sphinga-4,8-dienine, accounts for 7.4 mol %, ϳ5% of the substrate have been methylated. This is a relatively low yield compared with P. pastoris, were 67 mol % of the sphingoid bases in glucosylceramides are C9-methylated (Fig. 2). This might indicate that, despite the engineering, S. cerevisiae sphingolipids are still not an ideal substrate for methylation, because, in contrast to P. pastoris glucosylceramides, they contain a very long chain fatty acid (Fig. 1). Alternatively, the enzyme might find a different lipid environment or different protein interaction partners in the S. cerevisiae expression system than in P. pastoris.
The peak corresponding to (E,E)-9-methylsphinga-4,8-dienine has the same retention time as (E,E)-9-methylsphinga-4,8-dienine from wild-type P. pastoris, and it is clearly absent in the control strain transformed with the empty vector (Fig. 4B). Methylation of the saturated sphingoid bases C 18 -and C 20 -sphinganine could not be detected, although they were present in much higher amounts than C 18 -(E,E)sphinga-4,8-dienine. This indicates that a desaturased sphingolipid substrate is indeed required for the activity of the sphingolipid C9-methyltransferase.  The identity of the sphingoid bases was further confirmed by HPLC/ MS. In the negative ion mode (m/z ϭ M Ϫ 1), a pseudomolecular ion with m/z ϭ 476 corresponding to the DNP-derivative of C 18 -(E,E)-9methylsphinga-4,8-dienine was detected at the expected retention time (Fig. 4, C and D). 7 This demonstrates that the candidate sphingolipid C9-methyltransferase is able to methylate C 18 -(E,E)-sphinga-4,8-dienine in a heterologous expression system. No pseudomolecular ion with m/z ϭ 504 corresponding to C 20 -(E,E)-9-methylsphinga-4,8-dienine could be detected, indicating that C 20 -(E,E)-sphinga-4,8-dienine is not accepted as a substrate or is methylated at a much lower efficiency.
The Sphingolipid C9-Methyltransferase from P. pastoris Is Membrane-bound and Uses SAM and ⌬4,8-Desaturated Ceramide as Substrates in Vitro-The substrate specificity of the P. pastoris sphingolipid C9-methyltransferase was investigated by an in vitro assay with [methyl-14 C]SAM and different sphingolipid substrates: free sphingoid bases, ceramides, and glucosylceramides. As mentioned before, the sphingoid bases of P. pastoris glucosylceramides have both a ⌬4and a ⌬8-double bond. Based on a hypothetical reaction mechanism ("Discussion"), we expected that at least the ⌬8-double bond should be required for methylation.
Because sphingolipids with ⌬4,8-desaturated sphingoid bases were not commercially available, the corresponding free sphingoid bases, ceramides, and glucosylceramides were prepared from natural sources.
With ceramides as substrate, a sphingolipid C9-methyltransferase activity could be detected in cell-free homogenates of wild-type P. pastoris but not in homogenates of the P. pastoris sphingolipid C9-methyltransferase knock-out strain (Fig. 5A). The activity could be detected with ceramides from wild-type P. pastoris, the P. pastoris sphingolipid C9-methyltransferase knock-out strain, and cucumber leaves but not with synthetic ⌬4-desaturated ceramide. This indicates that only ⌬4,8desaturated ceramides are acceptor substrates for the sphingolipid C9-methyltransferase. The activity with ceramides from the P. pastoris sphingolipid C9-methyltransferase knock-out strain was higher than with ceramides from wild-type P. pastoris, because 67 mol % of the latter was already methylated (compare with Fig. 2).
To determine which part of the ceramide molecule was methylated in the assay, the 14 C-labeled ceramide spots were hydrolyzed, and the sphingoid bases were analyzed by HPLC (Fig. 5B). Fractions corresponding to individual sphingoid bases were collected and analyzed for radioactivity. The 14 C labeling was detected only in the (E,E)-9-methylsphinga-4,8-dienine fraction, confirming that indeed the ⌬4,8-desaturated sphingoid base moiety of the ceramide was methylated. No sphin- golipid C9-methyltransferase activity could be detected with free sphingoid bases or with glucosylceramides as substrates, indicating that only ceramides are acceptor substrates.
Because many enzymes involved in sphingolipid biosynthesis are membrane-bound, the cell-free homogenate of wild-type P. pastoris was centrifuged at 100,000 ϫ g, and the assay was repeated using the supernatant or pellet as enzyme source. The methyltransferase activity was found in the pellet representing the membrane fraction but not in the supernatant, which is consistent with the presence of two hypothetical N-terminal transmembrane domains in the amino acid sequence of the enzyme (Fig. 6).
Fungal Sphingolipid C9-methyltransferases Show High Similarity to Cyclopropane Lipid Synthases from Plants and Bacteria-The sphingolipid C9-methyltransferases are proteins of 489 -525 amino acids. The most similar proteins are the cyclopropane fatty acid synthases from plants (Sterculia foetida, 52% similarity, 31% identity), and bacteria (Escherichia coli, 47% similarity, 29% identity; Mycobacterium tuberculosis, 40 -43% similarity, 25-27% identity) (42)(43)(44)(45). The similarity to the sterol methyltransferase Erg6p from S. cerevisiae is considerably lower (35% similarity, 20% identity). The similarity to the cyclopropane fatty acid synthases is confined to the C-terminal two-thirds of the C9-methyltransferase sequences. The N-terminal parts of the C9-methyltransferase sequences contain two predicted transmembrane domains (Fig. 6). A comparison of the sphingolipid C9-methyltransferases with the cyclopropane mycolic acid synthase CmaA1 from M. tuberculosis for which the crystal structure is available (46) shows that many of the amino acid residues involved in substrate and cofactor binding have been conserved between these proteins of different biochemical function (Fig. 6).
A structural model of the sphingolipid C9-methyltransferase from P. pastoris was generated with SWISS-MODEL (47) using the CmaA1 structure 8 as template (Supplemental Data). The structure of the sphingolipid C9-methyltransferase from P. pastoris has been deposited in the RCSB Protein Data Bank with PDB ID 2FAY. The sphingolipid C9-methyltransferase sequence fits very well onto the template structure with a root mean square deviation of 0.7 Å between all aligned ␣-carbons. Deviations Ͼ2 Å occur only in the loop regions where the alignment in Fig. 6 shows a gap in one of the sequences. Between ␣6 and ␤4, an additional loop is inserted in the sphingolipid C9-methyltransferase without disturbing the surrounding structure (Supplemental Fig.  S1, bottom left). In CmaA1, a short helix, ␣9, is part of a hydrophobic tunnel that could be involved in lipid binding (46). In the structural model of the sphingolipid C9-methyltransferase, ␣9 is missing, with ␤5 and ␣10 being directly connected through a loop (Supplemental Fig. S1, top right). This difference might be linked to the structural differences between sphingoid bases and mycolic acids, with the latter having a much longer hydrocarbon chain.

DISCUSSION
We developed a bioinformatics strategy based on phylogenetic profiling to identify candidate fungal sphingolipid C9-methyltransferases. This allowed the selection of a single candidate from the 24 -71 predicted methyltransferases present in each fungal species with C9-methylated glucosylceramides. A P. pastoris knock-out strain lacking the corresponding gene was devoid of C9-methylated glucosylceramides. In a complementary approach, expression of the P. pastoris enzyme in a specially engineered S. cerevisiae strain resulted in the biosynthesis of C9-methylated sphingolipids. Together with the in vitro generation of 14 C-labeled ceramide from [methyl-14 C]SAM and ⌬4,8-desaturated ceramides and the conserved SAM-binding motifs found in its amino acid sequence, this demonstrates the function of the P. pastoris enzyme as a membrane-bound sphingolipid C9-methyltransferase.
Phylogenetic Profiling-The term "phylogenetic profiling" was first introduced by Pellegrini et al. (26) with a definition equivalent to ours. It was shown that proteins with the same or similar phylogenetic profiles are more likely to be functionally linked than proteins with different phylogenetic profiles. Phylogenetic profiling has since been used to 8 RCSB Protein Data Bank ID 1KP9. identify plant-specific proteins (48) and to detect errors in genome annotations (49). In addition, its suitability to predict protein-protein interactions (50) and proteins responsible for missing steps in known biochemical pathways (51) has been suggested. In contrast to these proteome-scale applications, phylogenetic profiling has been used in the present study to specifically identify a protein responsible for one particular function. A similar strategy has independently been applied to identify bacterial proteins involved in flagellar motion (52).
Groups of Orthologous Sequences-For a phylogenetic profile to be meaningful, the sequences contained in it must be functionally equivalent. This can be assumed if they are orthologous to each other, while paralogous sequences often have functionally diverged. Orthology and paralogy can be inferred by comparing the evolutionary relationship between genes to the relationship between the species in which the genes are found (53,54). The idea of using phylogenetic information to infer protein function in completely sequenced genomes has been termed "phylogenomics" (55). In the present study, groups of orthologous sequences were defined based on phylogenetic trees of protein families. While the branching patterns of the phylogenetic trees were analyzed manually, there are algorithms that could be used to identify groups of orthologous sequences automatically (56,57). An alternative method uses a graph theoretical approach to define "clusters of orthologous groups" (58) without reference to a phylogenetic tree. Clusters of orthologous groups were, however, not suitable for finding the sphingolipid C9-methyltransferase, because they are less accurate than definitions based on phylogenetic trees (55). Quantifying the Differences between Phylogenetic Profiles-The phylogenetic profile of the gene of interest can differ from the expected profile if there are errors in the genome annotation, as was the case with the candidate sphingolipid C9-methyltransferase from M. grisea, 4 or if one of the species contains a pseudogene or a gene that is not expressed under the conditions investigated. In such cases, the group coming closest to the expected profile can be determined by quantifying the differences between two phylogenetic profiles as outlined in the "Supplemental Discussion." Comparing the Sphingolipid C9-methyltransferases with Other SAMdependent Methyltransferases-The sphingolipid C9-methyltransferase belongs to the superfamily of SAM-dependent methyltransferases, which can methylate a large variety of substrates at N, O, or C atoms. Although the methyl group can be added to the free electron pair of N or O atoms, C-methylation requires the existence of a double bond at the C atom to be methylated. The methyl group of SAM is transferred to this carbon atom in an S N 2 mechanism (59). The resulting carbocation intermediate is then deprotonated by an enzyme-bound base. The final reaction outcome is determined by which of the protons is elimi-nated (Fig. 7). In the sphingolipid C9-methyltransferase, the previously vinylic proton at C9 is eliminated to reintroduce the ⌬8-double bond at its original position. In the sterol methyltransferase Erg6p, one of the SAM-derived methyl protons is eliminated with a simultaneous hydride shift between the previously olefinic carbon atoms C24 and C25. This results in the formation of a ⌬24(28)-double bond to the SAM-derived methylene carbon in fecosterol, a biosynthetic precursor of ergosterol (59). In the cyclopropane fatty acid synthases from plants and bacteria, one of the SAM-derived methyl protons is eliminated without a hydride shift, which in this case results in the formation of a cyclopropane ring in place of the original double bond. The evolutionary relationships between the sphingolipid C9-methyltransferase, the cyclopropane fatty acid synthases, and the sterol methyltransferase are discussed in more detail in the "Supplemental Discussion." Which functions do C9-methylated glucosylceramides have? Because C9-methylation is typical for fungal glucosylceramides, the biological functions of C9-methylation and of glucosylceramides themselves are most likely connected to each other. The yeasts S. cerevisiae and S. pombe have no glucosylceramides at all, whereas all other yeasts FIGURE 7. Hypothetical reaction mechanisms involving the transfer of a methyl group from SAM to a lipid substrate. A, methylenation of the C24 of zymosterol by the enzyme Erg6p (38) as intermediate step in the ergosterol biosynthesis in S. cerevisiae. The reaction scheme has been adopted from Ref. 59. B, methylation of the C9 of the sphingoid base (E,E)-sphinga-4,8-dienine resulting in the formation of (E,E)-9-methylsphinga-4,8-dienine. C, introduction of the cyclopropane ring into oleic acid by the enzyme CPA synthase from the plant S. foetida (42) as an intermediate step in the biosynthesis of sterculic acid. It is currently not known if the C9 or the C10 of oleic acid is methylated before ring closure. The base that abstracts the proton (bold) from the proposed carbocation intermediate is part of the active site of the enzyme. In the case of bacterial cyclopropane lipid synthases, this base is a bicarbonate ion (46,63). Only the mechanism shown in A has been investigated experimentally (59); the mechanisms shown in B and C are hypothetical. and fungi investigated so far, including the closely related yeast Saccharomyces kluyveri, contain C9-methylated glucosylceramides (4). It seems that both sphingolipid C9-methylation and the ability to synthesize glucosylceramides are not essential and have independently been lost in a few yeast species. The P. pastoris knock-out strain generated in this study was viable and not impaired in growth compared with the wild-type strain. It is possible that C9-methylated glucosylceramides are only required at certain stages of the life cycle or under certain environmental conditions.
In the introduction, several studies have been cited that point to important functions of fungal glucosylceramides in the interactions between fungi and other organisms. At present, it is not known if the C9-methyl group is required for these interactions, but all fungal glucosylceramides investigated in these studies have a C9-methyl branch. The identification of the fungal sphingolipid C9-methyltransferase will now enable us to investigate the function of the C9-methyl branch in these contexts using knock-out strains and other molecular and genetic tools.