Self-Protection against the Sphingolipid Biosynthesis Inhibitor Fumonisin B1 Is Conferred by a FUM Cluster-Encoded Ceramide Synthase

The biosynthesis of fungal natural products is highly regulated not only in terms of transcription and translation but also regarding the cellular localization of the biosynthetic pathway. In all eukaryotes, the endoplasmic reticulum (ER) is involved in the production of organelles, which are subject to cellular traffic or secretion. Here, we show that in Fusarium verticillioides, early steps in fumonisin production take place in the ER, together with ceramide biosynthesis, which is targeted by the mycotoxin. A first level of self-protection is given by the presence of a FUM cluster-encoded ceramide synthase, Fum18, hitherto uncharacterized. In addition, the final fumonisin biosynthetic step occurs in the cytosol and is thereby spatially separate from the fungal ceramide synthases. We suggest that these strategies help the fungus to avoid self-poisoning during mycotoxin production.

crops, but they often exhibit toxicity to other eukaryotes, including fungi (3,4). This raises the question: how do fungi protect themselves from the mycotoxins they produce? In fungi, genes directly involved in the biosynthesis of a secondary metabolite are generally located next to one another in biosynthetic gene clusters (5). Such clusters often comprise one or more genes that confer self-protection against the toxic product of the cluster. Mechanisms of self-protection include (i) duplicated or resistant targets, generally enzymes of primary metabolism; (ii) detoxification by biotransformation to less harmful derivatives; and (iii) specific transporters that secrete toxins from the cytosol (6).
Among the most studied fungal mycotoxins are fumonisins of the B series (FBs), secondary metabolites produced by Fusarium and Aspergillus species (7). The predominant FB producer Fusarium verticillioides is the most prevalent fungus associated with contamination of corn and corn-derived products worldwide, and the consumption of FB-containing food and feed has been correlated with a number of human and animal diseases (8,9). FB are polyketide-derived aminopentol compounds with two tricarballylic esters (10,11), and particularly the analog FB 1 (Fig. 1) is an efficient inhibitor of ceramide synthase, a key enzyme in sphingolipid biosynthesis (12). Appropriately, FB 1 competes reversibly with both sphinganine and acyl coenzyme A (acyl-CoA) in the ceramide synthase active site (13). During plant infection, or after consumption of contaminated food and feed, inhibition of ceramide synthase results in the accumulation of free sphingoid bases, primarily sphinganine (9,14,15). Toxicity is attributed to apoptotic processes initiated in response to sphingoid base accumulation rather than reduced sphingolipid biosynthesis (16).
Complex sphingolipids consist of a ceramide backbone, i.e., sphingosine attached to a fatty acid (C 14 to C 26 ) in an amide linkage, with a polar headgroup (17). They are essential components of eukaryotic cellular membranes, which are involved in maintaining proper membrane structure and anchoring membrane-associated proteins. Sphingolipid biosynthesis is compartmentalized: the first steps take place in the endoplasmic reticulum (ER), while complex sphingolipids are produced in the Golgi apparatus (18). As indicated, precursors/degradation products of sphingolipids act as second messengers, therefore, sphingolipid metabolism is tightly controlled (16). Thus, when applied in a targeted manner, sphingolipid biosynthesis inhibitors can serve as treatment against severe human diseases, e.g., cancer, Alzheimer's, schizophrenia, multiple sclerosis, and diabetes, that are (among other defects) correlated with a deregulated sphingolipid metabolism (17,19,20).
In F. verticillioides, the FB biosynthetic gene cluster (FUM) includes 16 genes: FUM1 to FUM3, FUM6 to FUM8, FUM10, FUM11, FUM13 to FUM19, and FUM21 (11,21). FB biosynthesis is initiated when the polyketide synthase Fum1 catalyzes synthesis of an octadecanoic acid precursor, which then undergoes condensation with L-alanine, a reaction catalyzed by the aminotransferase Fum8 (Fig. 1). The cluster encodes a tricarboxylate transporter, Fum11, likely functioning as a mitochondrial carrier protein and providing the substrate for Fum7 and Fum10 to produce CoA-activated tricarballylic acid. Two of these molecules are attached to the polyketide backbone by Fum14, while the final biosynthetic step is performed by the dioxygenase Fum3 (22). In addition, the cluster encodes a transcriptional regulator, Fum21, belonging to the GAL4-like Zn(II) 2 Cys 6 transcription factor family (21). Despite all of the functional analyses of FUM cluster genes, the functions of FUM15 to FUM19 remain unclear (11). Also, the mechanism that protects F. verticillioides ceramide synthases from the inhibitory effect of FB has not yet been elucidated. While FB 1 has antifungal activity, in particular against FB nonproducers, F. verticillioides was more resistant to externally added FB 1 (23). Knowledge of this process has potential to provide new insights into methods to block FB production in crops and thereby improve food safety.
The similarity of Fum17 and Fum18 to ceramide synthases and of Fum19 to ATP-binding cassette (ABC) transporters suggests a role for these proteins in FB self-protection (11). In the present work, we demonstrate that Fum19 is involved in FUM cluster regulation, acting as a repressor of biosynthesis. Point mutations of the identified ATP-binding domains indicated that ATP hydrolysis and possibly transport are required for Fum19 activity. Moreover, confocal microscopy showed that the aminotransferase Fum8 colocalizes with Fum17, Fum18, and the fungal ceramide synthase Cer1 in ER-derived vesicles. The results presented here show that Fum18 constitutes a FUM cluster-encoded ceramide synthase as an additional toxin-targeted enzyme, which contributes to ceramide biosynthesis and thereby decreases self-poisoning.

RESULTS
The ABC transporter Fum19 acts as a repressor of the FUM gene cluster. In order to analyze the influence of FUM17, FUM18, and FUM19 on FB biosynthesis and fungal self-protection, we generated single deletion mutants in F. verticillioides strain M-3125. We also generated both double FUM17/18 and triple FUM17-19 deletion mutants to test for functional redundancy. Because FB 1 is the FB analog produced in greatest abundance by F. verticillioides, i.e., FB 1 constitutes approximately 80% of total FB (11), we used it as a marker for FB production. FB 1 production was quantified in shaking cultures by high performance liquid chromatography coupled to highresolution mass spectrometry (HPLC-HRMS). The identity of the chromatographic peak corresponding to FB 1 was verified by comparison with an FB 1 reference (see Fig. S1A and B in the supplemental material).
Single and double deletion of FUM17 and FUM18 did not affect FB 1 production ( Fig. 2A), confirming previously reported results (11). However, deletion of FUM19, in both wild-type (WT) and Δfum17/18 backgrounds, resulted in significant overproduction of FB 1 in the supernatant as well as in the extracted mycelium ( Fig. 2A; Fig. S1C). In contrast, FUM19 overexpression resulted in a significant reduction in FB 1 production ( Fig. 2A; Fig. S1C). Despite elevated FB 1 levels in Δfum19 mutants, fungal biomass was not decreased compared to the WT (Fig. S1D). The overall production of FB 1 correlated very well with FUM cluster gene expression. Indeed, the Δfum19 mutant exhibited elevated expression of FUM17 and FUM18, as well as FUM8, which served as a marker for FUM genes encoding FB biosynthetic enzymes (Fig. 2B). Together, these results indicated that FUM19 negatively impacts expression of other FUM cluster genes, which in turn indicated a repressive role of Fum19 in FB biosynthesis.
We hypothesized that this regulatory role of Fum19 may be linked to its ability to bind and transport FB 1 at the expense of ATP. In order to test whether ATP hydrolysis is required for this regulatory role, we performed site-directed mutagenesis of the two putative Fum19 ATP-binding domains (Fig. 3A). The domains were identified by comparison of the deduced amino acid sequence of Fum19 to the ATP-binding domain of the Salmonella enterica serovar Typhimurium histidine permease, HisP, which has been structurally resolved and thoroughly analyzed with respect to structure-function relationships (24,25). The comparison led to identification of two sets of amino acids (K631/K1260 and D743/D1397) predicted to be critical for binding of ATP to Fum19 (25). In an attempt to disable ATP-binding and thereby block the presumed transport function of Fum19, we introduced two sets of point mutations into cloned FUM19 that were designed to replace the K631/K1260 (FUM19 Kmut ) or D743/D1397 (FUM19 Dmut ) residues in the Fum19 protein ( Fig. 3A; Fig. S2A). In locus exchange of Δfum19 with the WT gene, FUM19 C , restored FB 1 production to close to WT levels (Fig. 3B). However, exchange with either FUM19 Kmut or FUM19 Dmut did not complement the Δfum19 phenotype (Fig. 3B). We confirmed that FUM19 expression was comparable for all complemented strains (Fig. S2B). These experiments suggested that ATP hydrolysis, and possibly transport, are tightly coupled to the regulatory role of Fum19. However, because extracellular FB 1 accumulated upon deletion or point mutation of FUM19, we can hypothesize that either F. verticillioides harbors additional FB 1 exporters or that the molecule is able to passively cross the plasma membrane.
Fum17 and Fum18 are partially regulated by Fum21. Expression analysis via quantitative real-time PCR (qRT-PCR) of FUM19 deletion and overexpression strains revealed that the cluster-specific transcription factor gene FUM21 was not coordinately up-or downregulated, respectively (Fig. 2B). This suggested that the negative feedback mediated by Fum19 does not occur via FUM21 repression on a transcriptional level.
To investigate this further, we deleted FUM21 in the WT and Δfum19 mutant strains of F. verticillioides. In both genetic backgrounds, the deletion completely abolished FB 1 production, confirming that Fum21 is essential for biosynthesis (Fig. 3B). However, in contrast to FUM8, FUM17 and FUM18 expression was only partially under the control of Fum21, as their expression was induced in Δfum19/Δfum21 double mutants (Fig. 3C). It is noteworthy that this expression occurred in the complete absence of FB 1 production, highlighting that cluster activation in the Δfum19 background is not a consequence of intracellular accumulation of FB 1 . Nonetheless, addition of exogenous FB 1 induced expression of FUM8, FUM17, FUM18, and FUM19, but not FUM21, in the WT background (Fig. S2C). Upregulation of FUM8, FUM17, and FUM18 by the addition of FB 1 was independent of the presence of a functional FUM19 but largely dependent on the presence of a functional FUM21 (Fig. S2C). The indicated strains were grown in ICI/6 mM Gln shaking culture for 7 days and analyzed by HPLC-HRMS. Culture fluids were measured directly, while metabolite extraction from washed mycelium was performed for intracellular FB 1 levels. The production level of the WT was arbitrarily set to 100%. The data are means Ϯ the standard deviations (n ϭ 3). (C) After 2 days of cultivation, relative expression (RE) was determined by qRT-PCR. The WT gene expression was arbitrarily set to 1. The data are means Ϯ the standard deviations (n ϭ 3). For statistical analysis, strains were compared as indicated using the Student t test (*, P Ͻ 0.05; **, P Ͻ 0.01; ***, P Ͻ 0.001). FB 1 Self-Protection by FUM-Encoded Ceramide Synthase ® Taken together, these experiments revealed complex and intertwined regulation mechanisms of the FUM gene cluster: FUM19 repressed FUM cluster genes in the WT background, whereas deletion of FUM19 strongly upregulated FB 1 production. FUM21 was essential for FB 1 production, and both FUM8 and FUM19 were tightly regulated by the transcription factor. In contrast, expression of FUM17 and FUM18 could still be induced in the Δfum21 background.
Phylogenetic relationships of Fum17 and Fum18 to other fungal ceramide synthases. An analysis of the deduced amino acid sequences of Fum17 and Fum18 revealed that both proteins have a TRAM/LAG1/CLN8 (TLC) protein domain (IPR006634), a domain that is present in all eukaryotic ceramide synthases described to date (26). To gain insight in the relationships of Fum17 and Fum18 to other fungal ceramide synthases, we inferred a phylogenetic tree using alignments of predicted amino acid sequences retrieved by BLAST analysis of selected genome sequences at the National Center for Biotechnology Information (NCBI)-GenBank and Joint Genome Institute. Because Fusarium is a member of the phylum Ascomycotina (class Sordariomycetes), most sequences were retrieved from other ascomycetous fungi, including species in classes Dothideomycetes, Eurotiomycetes, Leotiomycetes, Pezizomycetes, Saccharomycetes, and Sordariomycetes. We also retrieved sequences from five species in phylum Basidiomycotina. In the resulting phylogenetic tree, fungal ceramide synthases were resolved into two major clades, designated CS-I and CS-II (Fig. 4). Most species of the Ascomycotina and Basidiomycotina included in this analysis had at least one ceramide synthase in each of the major clades. However, the LAG1 and LAC1 genes from yeast species were an exception to this; LAG1 and LAC1 were both resolved within clade CS-I and were relatively closely related to one another. Fusarium species with the FUM cluster had three ceramide synthase genes (CER1, CER2, and CER3) located outside the cluster, in addition to the cluster genes FUM17 and FUM18. FUM17 homologs from Fusarium fujikuroi and Fusarium proliferatum were excluded from the analysis because the gene was pseudogenized in these species. In the tree, FUM17 and CER1 (FVEG_06971) were resolved within clade CS-I, while FUM18, CER2 (FVEG_12887), and CER3 (FVEG_15375) were resolved in clade CS-II (Fig. 4). Fusarium solani was unique in that it had a fourth non-FUM-cluster ceramide synthase gene that was resolved within clade CS-I and that was relatively distantly related to other Fusarium genes, but instead was most closely related to homologs in the sordariomycete species Gaeumannomyces tritici and Pyricularia oryzae (Fig. 4). Conclusively, phylogenetic analysis revealed that Fum17 and Fum18 are closely related to fungal ceramide synthases.
Fum17, Fum18, Fum8, and Cer1 colocalize in ER-derived vesicles. Next, we performed localization studies by confocal microscopy of Fum17 and Fum18, C-terminally fused to green and red fluorescent proteins (GFP/DsRed), respectively. We made sure that the fluorescent proteins did not give signals when excited with inappropriate wavelengths (Fig. S3). Fum17-GFP and Fum18-DsRed were shown to colocalize with each other in the perinuclear ER, and their localization around the nuclei was verified by nuclear visualization, i.e., staining with the DNA-binding dye Hoechst 33342. In addition, they colocalized in the membrane of small/medium-sized vesicles (ca. 1 to 2 m) which were independent of the nuclei (Fig. 5A). Full colocalization of both proteins with one of the three ceramide synthase homologs, Cer1 fused to blue fluorescent protein (BFP), confirmed their presence in the perinuclear ER and in ER-derived vesicles (Fig. 5B). The presence of DsRed-Fum8 (N-terminal fusion) in the vesicles, but not in the perinuclear ER, revealed that early FB biosynthesis takes place in these compartments (Fig. 5C). Intriguingly, the final FB biosynthetic step performed by the dioxygenase Fum3 takes place in the cytoplasm, as shown by C-terminal GFP fusion (Fig. 5D). In summary, these results strongly indicated that biosynthesis of both sphingolipids and inhibitors thereof partially overlap, while the final biosynthetic step is separate from Fum17, Fum18, and Cer1.
Fum18 is a functional ceramide synthase. In order to investigate whether Fum17 and Fum18 function as ceramide synthases, we established a yeast complementation (Continued on next page) FB 1 Self-Protection by FUM-Encoded Ceramide Synthase ® assay. As indicated above, the Saccharomyces cerevisiae genome encodes two paralogs of clade CS-I with redundant function, Lag1 and Lac1. LAG1/LAC1 double deletion mutants are nonviable or very slow growing (27,28). Therefore, we constructed a conditional null mutant, in which LAG1 is under the control of the inducible TET ON promoter (29) in the Δlac1 background. The resulting strain could grow only in the presence of doxycycline (Fig. 6A). As shown, the two S. cerevisiae ceramide synthase genes, either LAG1 or LAC1, could rescue the lethal phenotype when expressed from a plasmid under the control of the constitutive TEF1 promoter (Ptef1) (Fig. 6A).
We expressed the five F. verticillioides TLC domain proteins (CER1, CER2, CER3, FUM17, and FUM18) in the Δlac1/TET::LAG1 background. As shown, expression of CER1, CER2, or FUM18 fully complemented the S. cerevisiae null mutant in the absence of doxycycline (Fig. 6A). In contrast, FUM17 and CER3 did not complement the mutant, and the expression of CER3 was even toxic for S. cerevisiae ( Fig. 6A and B). Intriguingly, expression of a bicistronic FUM17::CER3 construct partially complemented Δlac1/TET:: LAG1 in the absence of doxycycline (Fig. 6A). This suggested that Cer3 is able to form nonfunctional heterodimers with both Lag1 and Lac1, while it can functionally dimerize with Fum17. These data indicated that the FUM gene cluster encodes a functional ceramide synthase, Fum18, while Fum17 can potentially contribute to ceramide biosynthesis by interaction with Cer3.
Fum18 contributes to resistance against FB 1 . In order to test whether Fum18, Cer1, and Cer2 are insensitive to FB 1 , we established a resazurin cell viability assay for of alignments of predicted amino acid sequences of selected ceramide synthases from the ascomycete classes Dothideomycetes, Eurotiomycetes, Leotiomycetes, Pezizomycetes, Saccharomycetes and Sordariomycetes. Sequences of five basidiomycete species were also included in the analysis, and the tree was rooted with rat and human sequences. Accession numbers for protein sequences are indicated after species names. Joint Genome Institute accession numbers are preceded by JGI; all other accessions are from NCBI/GenBank. In the Fum17 and Fum18 clades, the five-digit numbers after some of the Fusarium species names are NRRL strain designation. The genome sequences of these are present in GenBank, but they are not annotated. Numbers near branches are bootstrap values based on 1,000 pseudoreplicates. Values of Ͻ70 are not considered to be significant and therefore are not shown. S. cerevisiae and Fusarium strains generated during this study. The blue nonfluorescent dye resazurin is reduced to its pink highly fluorescent derivative resorufin, and this turnover is proportional to respiration processes in living cells, including fungi (30,31). In a previous study, real-time monitoring of bacterial growth revealed that fluorescence was proportional to cell density, allowing the assay to be used to compare treatments that differ in effect on growth (32). We adapted this protocol for S. cerevisiae and F. verticillioides strains exposed to FB 1 , where the difference in fluorescence maxima between nontreated and FB 1 -treated cells represented the growth-inhibitory effect for a given strain and was independent of the growth rate (Fig. S4). Both fungi exhibit an intrinsic tolerance to FB 1 ; therefore, relatively high concentrations of FB 1 (250 to 500 g/ml) were used in experiments to observe differences in inhibition (Fig. 6C to E).
In order to include human ceramide synthases in the susceptibility tests, CERS2, CERS4, CERS5, and CERS6 were amplified from cDNA of the colon cancer cell line HT29-MTX (33). Sequence analysis of the cDNA identified a previously unreported insertion in CERS6 that resulted in a frameshift and early termination, producing a predicted protein of 349 rather than 384 amino acids (K334 ¡ KVLVILTCFYSTGVQG). Although these human ceramide synthases harbor a homeobox (HOX) domain (IPR001356) and thereby differ structurally from the fungal homologs, CERS2/CERS4/ CERS6 fully, and CERS5 partially complemented Δlac1/TET::LAG1 (Fig. 6A).
We first analyzed Δlac1/TET::LAG1 strains complemented with either one of the fungal or human ceramide synthase genes and grown in the absence of inducer (Fig. S4A). Compared to yeast Lag1 and Lac1, human CERS4 and CERS6 were more strongly inhibited by FB 1 (500 g/ml), although the difference for CERS6 was not significant (Fig. 6C). Among the F. verticillioides homologs, Cer2 was more strongly  Fig. S4. In each case, inhibition of the control strain was arbitrarily set to 100%. The data are means Ϯ the standard deviations (n ϭ 3). For statistical analysis, strains were compared as indicated or with the control using the Student t test (*, P Ͻ 0.05; **, P Ͻ 0.01; ***, P Ͻ 0.001).
inhibited than Cer1 and Fum18. Coexpression of FUM17 and FUM18 did not further increase tolerance to FB 1 , suggesting that they do not cooperate in FB 1 resistance (Fig. 6C). Based on this result, we can assume that Fum18 is not an intrinsically more resistant ceramide synthase variant but that it contributes to fungal self-protection by providing additional cluster-encoded ceramide synthase activity. To confirm this assumption, we expressed CER1, CER2, and FUM18 in WT S. cerevisiae to generate strains with a third ceramide synthase gene in addition to LAG1 and LAC1. The results of growth assays indicated that heterologous expression of any of these F. verticillioides genes significantly increased tolerance of S. cerevisiae to FB 1 (Fig. 6D; Fig. S4B). The finding for CER2 was surprising given that initial assessment indicated that Cer2 was less tolerant to FB 1 than Cer1 and Fum18 (Fig. 6C).
Finally, we examined germinating conidia of the Δfum17, Δfum18, and Δfum17/18 mutant and WT strains of F. verticillioides in the presence or absence of FB 1 (250 g/ml). Young hyphae of the single and double deletion mutants of FUM18 were more susceptible to FB 1 , which could be complemented by reintroduction of the full-length gene in the Δfum18 background ( Fig. 6E; Fig. S4C). In summary, Fum18 is a functional ceramide synthase which confers additional cluster-encoded enzymatic activity, and thereby fungal self-protection against the produced ceramide synthase inhibitor FB 1 .

DISCUSSION
The biosynthesis of fungal secondary metabolites is highly regulated not only in terms of cluster gene expression but also concerning the spatial organization of the encoded enzymes. Like all eukaryotes, fungal cells contain small organelles and vesicles able to embed enzymes and potentially compartmentalize biosynthetic pathways. This mechanism has several advantages: (i) it concentrates enzymes in a smaller space, improving multienzymatic catalysis; (ii) it reduces cross-reactions with other biosynthetic pathways; and (iii) in the case of production of mycotoxins, it may protect cells from self-poisoning (34).
In the present study, we identified highly sophisticated compartmentalization, regulation and self-protection mechanisms, which ensure tightly controlled levels of the ceramide synthase inhibitor FB 1 in F. verticillioides. Compartmentalization of FB biosynthesis makes sense for two reasons. First, compartmentalization allows for spatially restricted and thereby controlled biosynthesis. The early FB pathway enzyme Fum8 was localized to ER-derived vesicles (Fig. 7), and indeed the duplicated target enzyme, the ceramide synthase Fum18, was found to colocalize with Cer1 in these vesicles. It is noteworthy that ceramide biosynthesis likely occurs on the perinuclear ER (35), which was nearly completely devoid of Fum8. Second, the last biosynthetic step, performed by Fum3, was localized to the cytoplasm, revealing that the most potent compound FB 1 is separate from the fungal ceramide synthases (36). Appropriately, this compartmentalization strategy likely helps to prevent self-poisoning.
Although Fum19 was not shown to be essential for FB 1 export, the results reported here indicate that this ABC transporter has an important role in the regulation of expression of FUM cluster genes and thereby self-protection. This was supported by several lines of evidence. Deletion (in two different backgrounds) and overexpression of FUM19 induced and reduced, respectively, the expression of other FUM genes, which is consistent with a role as repressor for Fum19 (Fig. 7). An earlier study on F. verticillioides Δfum19 reported a slight increase in FB 1 and a significant decrease in its direct precursor FB 3 , indicating a shift toward the final product and thereby an upregulation of the biosynthetic pathway (11). We suggest that FB 1 production in the WT background was not at its maximum under our cultivation conditions (fully synthetic 1-week culture), so that a more drastic increase could be observed for Δfum19. Noteworthy, FUM19 overexpression and a detailed analysis of intra-and extracellular FB 1 levels had not been performed. On a transcriptional level, FUM19 deletion in the Δfum21 background resulted in derepression of the resistance gene FUM18 (and FUM17), indicating a regulatory circuit that directly targets FUM gene expression in the complete absence of either Fum21 or the final product FB 1 . Taken together, this indicates that Fum19 is involved in a negative-feedback loop to moderate FB 1 levels, while Fum19 dysfunction, here mimicked by gene deletion and point mutations, gives a cue to upregulate expression of FUM cluster genes.
Point mutations of FUM19 confirmed that ATP hydrolysis and thereby possibly transport are tightly linked to its function, although final proof that Fum19 is a true exporter is still lacking. At this point, it cannot be excluded that ATP sensing is (one of) the vital role(s) of Fum19. Nevertheless, these data are consistent with earlier reports, in which deletion of the associated ABC transporter gene resulted in an upregulated secondary metabolite biosynthesis, i.e., of gibepyrone and beauvericin in F. fujikuroi and of sirodesmin in Leptosphaeria maculans (37)(38)(39). Noteworthy, the beauvericin ABC transporter Bea3 was successfully localized to the plasma membrane, although it was not essential for secretion (38). This implies either that these mycotoxins may passively cross the plasma membrane or that additional cluster-independent exporters are responsible for their secretion. As an example, redundancy of exporters has been observed for cercosporin in Cercospora nicotianae: although the gene cluster encodes a transporter of the major facilitator superfamily (MFS), two exporters outside of the FIG 7 Compartmentalization and self-protection strategies of the FB 1 producer F. verticillioides. Fum8 is localized to ER-derived vesicles, in which also Fum17, Fum18, and Cer1 are found, while Fum3 is cytoplasmic. Fum18 is a functional ceramide synthase required for self-protection by providing additional enzymatic activity and thereby supporting Cer1 and Cer2. Fum17 could contribute to self-protection by forming a functional heterodimer with Cer3. The ABC transporter Fum19 is formally a repressor of the FUM gene cluster. This was supported by the fact that FB 1 feeding triggered cluster gene expression, an effect which was mainly dependent on the cluster-specific transcription factor Fum21. FB 1 Self-Protection by FUM-Encoded Ceramide Synthase ® cluster were required for self-protection (40,41). Interestingly, cluster upregulation upon deletion seems to be specific to ABC transporter genes, while the loss of MFS transporters has been mainly correlated with a downregulation of the cluster (41,42).
Because the results from our analysis of FUM19 function provided new insights into FB biosynthesis, we decided to reassess the functions of FUM17 and FUM18. In the present study, we demonstrated that Fum18 can function as a ceramide synthase on its own and that it contributes to FB self-protection by providing supplementary ceramide synthase activity (Fig. 7). FB 1 inhibited growth of FUM18 deletion mutants during spore germination, but not when produced at a later stage in liquid cultures, since fungal biomass of FUM18 mutants was not reduced compared to the WT. Indeed, it has been suggested that inhibition of sphingolipid metabolism is likely to be more disturbing for actively growing cells (16,36), i.e., during spore germination in this case. In addition, Fum17 could contribute to FB self-protection by forming a heterodimer with Cer3. The later point is supported by the fact that homo-and heterodimers of Lag1/Lac1 have been reported for yeast ceramide synthase complexes (43).
The expression of additional ceramide synthase genes to decrease the toxic effect of sphingolipid biosynthesis inhibitors seems to be conserved. In this regard, FB 1 and AAL-toxin display a highly similar chemical structure, and their biosynthesis is conferred by homologous gene clusters (44,45). The F. fujikuroi FUM cluster and the Alternaria alternata AAL-toxin cluster harbor a FUM18 homolog but lack a functional FUM17 (45,46). Alt7 of the AAL-toxin cluster was suggested to be a ceramide synthase homolog, although direct functional tests have not been reported (45). However, prior to that, it was shown that resistance in tomato to AAL-toxin-producing isolates of A. alternata was mediated by the plant ceramide synthase Asc-1. Tomato Asc-1 prevented programmed cell death which was otherwise induced upon disruption of sphingolipid biosynthesis by AAL-toxin (47,48). Therefore, the presence of an additional target enzyme is a successful resistance strategy for both the plant and the fungal side.
The present study revealed that filamentous fungi have at least two ceramide synthase genes. Furthermore, phylogenetic analysis showed that they group into two distinct clades (CS-I and CS-II) and that within a given species, one is a member of CS-I and the other is a member of CS-II. Fusarium spp. and FB-producing Fusarium spp. can have three and (up to) five homologs, respectively. The phylogenetic relationships of FUM17 and FUM18 and their presence in the FUM cluster mirror the relationships and presence of two ceramide synthase genes in fungal genomes. The nesting of FUM17 and FUM18 within clades of sordariomycete genes suggests that they did not evolve from Fusarium genes, but instead evolved from genes in a sordariomycete ancestor. On a biochemical level, evidence suggests that fungal clade CS-I ceramide synthases provide precursors for glycosylinositol phosphorylceramide-type complex sphingolipids, while clade CS-II enzymes feed the glucosylceramide pool (49,50). Deletion of Fusarium graminearum BAR1, the FvCER3 ortholog, resulted in complete loss of glucosylceramide biosynthesis (51). Although Cer3 was not functional on its own in the yeast complementation assay, this does not exclude its functionality in F. verticillioides in vivo. It remains to be elucidated whether Cer2 and Fum18 homologs also contribute to the glucosylceramide pool in FB-producing Fusarium spp.
The yeast ceramide synthase assay developed in this study could be used to test novel (chemical) derivatives of sphingolipid biosynthesis inhibitors for improved or more specific ceramide synthase inactivation. The six human enzymes exhibit specificity with respect to carbon-chain length of fatty acyl-CoAs that they can use as the substrates (18). As a result, modification of FB 1 to mimic CoA precursors of a specific chain length has potential as a strategy to inhibit one ceramide synthase without affecting others. Targeting individual human enzymes would have powerful therapeutic applications, such as specifically inactivating CERS6 upregulated during development of multiple sclerosis (17). In addition, the disruption of fungal sphingolipid metabolism has been suggested as an efficient antifungal strategy (50).
In summary, we presented a comprehensive analysis of compartmentalization strategies and cluster-encoded self-protection mechanisms involved in FB biosynthesis in F.
verticillioides. The final step in FB biosynthesis, performed by the dioxygenase Fum3, was localized to the cytoplasm and does thereby not colocalize with fungal ceramide biosynthesis in the ER. In ER-derived membrane structures, Fum18 was shown to protect the fungal ceramide synthases Cer1 and Cer2 by providing FUM clusterencoded ceramide synthase activity. In addition, the ABC transporter Fum19 indirectly contributes to self-protection by modulating FB levels through its effects on expression of FUM cluster genes. These findings provide insight into FB self-protection in fungi and point to novel methods for ceramide synthase-related drug discovery.

MATERIALS AND METHODS
F. verticillioides media and growth conditions. F. verticillioides M-3125 (52) was used as parental strain for the analysis of the FUM gene cluster (FVEG_00315 to FVEG_00329) (11,21,22). General maintenance of fungal strains was performed on solidified complete medium (53). For the cultivation of strains in liquid culture, 100 ml of Darken preculture (54) in 300-ml Erlenmeyer flasks was inoculated with a piece of mycelium and shaken for 3 days at 180 rpm and 28°C. Then, 250 l of the preculture was transferred to 100 ml of synthetic ICI medium (Imperial Chemical Industries, Ltd., London, UK) (55) supplemented with 6 mM glutamine (Gln) and then shaken under the same conditions for 2 or 7 days for gene expression or FB 1 analyses, respectively.
Plasmid constructions. Vector assembly was achieved with yeast recombinational cloning (56,57). Deletion vectors harbored ϳ1 kb upstream and downstream flanks of the gene of interest, as well as a deletion cassette. Flanks were amplified using proofreading polymerase, primer pairs 5F/5R and 3F/3R (see Table S1 in the supplemental material), as well as F. verticillioides genomic DNA (gDNA). FUM17, FUM18 and FUM19 were deleted by exchange with the hygromycin B resistance gene under the control of A. nidulans PgpdA; for that purpose, hphR was amplified from pUC-hph (58) with hph_F/hph_R. Furthermore, deletion of FUM21 was achieved with the nourseothricin resistance gene under the control of A. nidulans PtrpC, and natR was amplified from pZPnat1 (GenBank accession no. AY631958.1) with nat1_F/nat1_R. S. cerevisiae BY4741 (Euroscarf, Oberursel, Germany) was transformed with the obtained fragments, as well as with the XbaI/HindIII-digested shuttle vector pYES2 (Life Technologies, Darmstadt, Germany), yielding the deletion constructs (Fig. S5 to S7).
Promoter exchange of S. cerevisiae LAG1 with the inducible TET ON promoter (Fig. S9) started out by amplifying upstream and (intragenic) downstream sequences with the primer pairs lag1_5F/lag1_5R and lag1_3F/lag1_3R (Table S1), respectively, based on BY4741 gDNA. The histidine HIS3 marker gene, including promoter and terminator sequences, was generated with his3_Prom_TET_F/his3_Term_R from gDNA of a prototrophic S. cerevisiae strain (Jena Microbial Research Collection, accession no. STI25222). TET ON (59) was amplified with TET_F/TET_R, and the complete vector was assembled under the use of doubly digested pYES2 as indicated above.
Finally, ceramide synthase yeast complementation vectors were constructed (Fig. S9A) by amplification of the inserts with the primer combination Tef1_F/Cyc1T_R (Table S1) based on F. verticillioides cDNA, S. cerevisiae gDNA, or human HT29-MTX (33) cDNA. Coexpression of two genes from the same promoter was achieved by inserting the viral 2A sequence in between, yielding one large polycistronic mRNA that gives individual proteins upon cotranslational cleavage (59). To this end, double-stranded DNA was gained from the 2A_F/2A_R primer combination (Table S1) via boiling and subsequent annealing at room temperature. The ceramide synthase genes were expressed under the control of the constitutive yeast promoter Ptef1 by cloning the obtained fragments into the SmaI/PvuII-digested vector FB 1 Self-Protection by FUM-Encoded Ceramide Synthase ® backbone pYES2::Ptef1 (59). All of the above-described expression vectors were verified by sequencing using primers listed in Table S1.
Expression analysis via qRT-PCR. Fungal strains were grown in ICI/6 mM Gln for 2 days, prior to harvest and lyophilization. RNA was isolated from ground mycelium under the use of TRI Reagent (Sigma-Aldrich, Steinheim, Germany). Next, 1.5 g of RNA was DNase I treated and transcribed into cDNA with a ProtoScript II first-strand cDNA synthesis kit (New England Biolabs, Frankfurt/Main, Germany) and included oligo(dT) primers according to the standard protocol. For qRT-PCR, MyTaq HS Mix (Biocat, Heidelberg, Germany) was used in combination with 5% (vol/vol) EvaGreen Dye (Biotium, Fremont, CA) and black nontranslucent plates (Applied Biosystems standard; 4titude, Ltd., Berlin, Germany). Reactions were run in an Applied Biosystems QuantStudio 3 real-time PCR system and analyzed with QuantStudio design and analysis software (Life Technologies, Darmstadt, Germany). Expression of the genes of interest, as well as of the three constitutively expressed reference genes (FVEG_11477 encoding ubiquitin, FVEG_02524 encoding actin, FVEG_09003 encoding a GDP-mannose transporter) (37), was determined in triplicate with primers listed in Table S1. The annealing temperature was set to 60°C, and primer efficiencies were between 90 and 110%. Relative expression was calculated with the ΔΔC T method (61). FB 1 analysis via HPLC-HRMS. For FB 1 analysis, the strains were grown in ICI/6 mM Gln for 7 days. The supernatant was separated from the mycelium by filtration through Miracloth (VWR, Darmstadt, Germany), and then culture fluids were fully cleared by centrifugation and mixed with 1:1 (vol/vol) methanol and 1% (vol/vol) naringenin as an internal standard (1 mg/ml in methanol; Sigma-Aldrich, Steinheim, Germany) to give 1 ml. For extraction of FB 1 from washed and lyophilized mycelium, 0.1 g was extracted with 1.5 ml of ethyl acetate-methanol-dichlormethane (3:2:1, vol/vol) by vigorous shaking for 2 h (37). Then, 1 ml was evaporated to dryness, taken up in 1 ml of methanol with 1% (vol/vol) naringenin, and subjected to HPLC-HRMS after filtration through 0.2-m PTFE filters (Carl Roth, Karlsruhe, Germany).
HPLC-HRMS measurements were conducted on a Thermo Fisher Q ExactivePlus hybrid quadrupole-Orbitrap mass spectrometer with an electrospray ion source operating in negative ionization mode with a full scan range of 100 to 1,000 m/z in combination with an Ultimate 3000 UHPLC system (Thermo Fisher Scientific, Dreieich, Germany). The system was equipped with a Kinetex C 18 column (150 ϫ 2.1 mm, 2.5 m, 100 Å; Phenomenex, Aschaffenburg, Germany). The following elution gradient was used (solvent A, H 2 O plus 0.1% [vol/vol] HCOOH; solvent B, acetonitrile plus 0.1% [vol/vol] HCOOH): 5% B for 0.5 min, 5 to 97% B in 11.5 min, and 97% B for 3 min; flow rate of 0.3 ml/min; injection of 10 l; and column oven at 40°C. Raw liquid chromatography-mass spectrometry data were analyzed using XCalibur (Thermo Fisher Scientific). FB 1 peak areas were normalized against the internal standard to account for variability between runs. FB 1 levels were related to the dry weight of the cultures, which were performed in biological triplicate.
Phylogenetic analysis. Homologs of ceramide synthase genes were retrieved by BLASTp and/or BLASTx analyses (62) of fungal genome sequence databases at NCBI and the Joint Genome Institute. Initially, the S. cerevisiae LAG1 and LAC1 genes and F. verticillioides FUM17, FUM18, CER1, CER2, and CER3 genes were used as query sequences. Subsequently, sequences from other fungal species were used to ensure that we obtained ceramide synthase sequences from diverse fungi. In some cases, gene sequences were manually annotated to correct obvious errors in automated predictions of translation start and stop sites and intron splice sites. Using BLAST, we retrieved the predicted amino acid sequences for 141 genes. We also retrieved a single homolog from both rat and human to use as an outgroup. The amino acid sequences were aligned with the MUSCLE algorithm as implemented in the program MEGA7 (63). The resulting alignment was then subjected to tree building analysis using the maximum-likelihood method as implemented in the program IQ-Tree (version 1.6.9) with ultrafast bootstrapping (64). The resulting tree was viewed and formatted with MEGA7.
Confocal microscopy. Fungal hyphae were grown from 10 4 conidia in 300 l of ICI/6 mM Gln as adherent cultures in ibidi dishes (ibidi, Gräfelfing, Germany) at 30°C for 16 h. For nuclear staining, NucBlue Live ReadyProbes reagent (Thermo Fisher Scientific, Schwerte, Germany) was used according to the manufacturer's instructions. All microscopy experiments were performed on an Axio Observer Spinning Disc confocal microscope (Zeiss, Jena, Germany) with the 63ϫ or 100ϫ oil objectives and analyzed with ZEN 2.6 software. Fluorescent stains and proteins were excited using the 405-, 488-, and 561-nm laser lines. Postprocessing of the images for brightness adjustments was done in ImageJ (https://imagej.nih.gov).
Yeast complementation assay. The generated Δlac1/TET::LAG1 conditional null mutant was transformed with the ceramide synthase-containing complementation plasmids (Fig. S9A), and positive colonies were selected on synthetic defined medium lacking uracil (SD-Ura) plus 8 g/ml doxycycline hyclate (AppliChem, Darmstadt, Germany). The empty vector pYES2::Ptef1 served as the control, as well as the progenitor S. cerevisiae BY4741 and Δlac1 strains (Euroscarf, Oberursel, Germany). At least two independent transformants were tested per strain, and verification of all strains by diagnostic PCR is summarized in Fig. S9C. For the drop assay, an overnight culture was prepared in 5 ml of SD-Ura plus 8 g/ml doxycycline, which was shaken at 180 rpm and 30°C. Then, 10 l of a 1:10 dilution series in water was spotted onto SD-Ura plates with or without doxycycline, starting with an optical density at 600 nm (OD 600 ) of 0.1. Growth was evaluated after 2 days at 30°C.
Resazurin cell viability assay. FB 1 growth inhibition of ceramide synthase-complemented yeast strains and F. verticillioides spores was evaluated via monitoring the reduction of resazurin to fluorescent resorufin every 30 min for up to 24 h. The experiments were conducted in a CLARIOstar plate reader (BMG Labtech, Ortenberg, Germany) with sterile black 96-well plates (BRANDplates; VWR, Darmstadt, Germany). Incubation was performed with SD-Ura at 30°C for yeast cells, while Fusarium spores were grown in the presence of ICI/6 mM Gln at 28°C. Each well contained 150 l: 10 l cells in medium (yeast, OD 600 ϭ 0.01; Fusarium, 5 ϫ 10 4 spores/ml final concentration), 10 l of FB 1 in water (250 to 500 g/ml final concentration; Biomol, Hamburg, Germany; lot 0542168) or 10 l of water as control, and 1.5 l of 0.002% (wt/vol) resazurin sodium salt in water (AppliChem, Darmstadt, Germany), adjusted to the final volume with medium. Resorufin was measured at the following wavelengths: excitation at 570 nm and emission at 615 nm (as previously performed) (30). Controls without cells did not show an increase in fluorescence. Measurements were done in biological triplicate, and fluorescence measurements of each sample were normalized against its lowest starting value. Growth inhibition was calculated by comparing fluorescence maxima between treated and nontreated samples (Fig. S4).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.

ACKNOWLEDGMENTS
We thank Daniela Hildebrandt for excellent technical assistance. Furthermore, we are grateful to Maria Joanna Niemiec and Ilse Jacobsen (Microbial Immunology, HKI Jena, Germany) for kindly providing HT29-MTX cell cultures.
This study was supported by the Leibniz Research Cluster in the frame of the BMBF Strategic Process Biotechnology 2020ϩ and by a grant of the European Social Fund ESF Europe for Thuringia projects SphinX and MiQWi.
We declare there are no competing interests. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.