Phylogenomics of the benzoxazinoid biosynthetic pathway of Poaceae: gene duplications and origin of the Bx cluster

Background The benzoxazinoids 2,4-dihydroxy-1,4-benzoxazin-3-one (DIBOA) and 2,4-dihydroxy-7- methoxy-1,4-benzoxazin-3-one (DIMBOA), are key defense compounds present in major agricultural crops such as maize and wheat. Their biosynthesis involves nine enzymes thought to form a linear pathway leading to the storage of DI(M)BOA as glucoside conjugates. Seven of the genes (Bx1-Bx6 and Bx8) form a cluster at the tip of the short arm of maize chromosome 4 that includes four P450 genes (Bx2-5) belonging to the same CYP71C subfamily. The origin of this cluster is unknown. Results We show that the pathway appeared following several duplications of the TSA gene (α-subunit of tryptophan synthase) and of a Bx2-like ancestral CYP71C gene and the recruitment of Bx8 before the radiation of Poaceae. The origins of Bx6 and Bx7 remain unclear. We demonstrate that the Bx2-like CYP71C ancestor was not committed to the benzoxazinoid pathway and that after duplications the Bx2-Bx5 genes were under positive selection on a few sites and underwent functional divergence, leading to the current specific biochemical properties of the enzymes. The absence of synteny between available Poaceae genomes involving the Bx gene regions is in contrast with the conserved synteny in the TSA gene region. Conclusions These results demonstrate that rearrangements following duplications of an IGL/TSA gene and of a CYP71C gene probably resulted in the clustering of the new copies (Bx1 and Bx2) at the tip of a chromosome in an ancestor of grasses. Clustering favored cosegregation and tip chromosomal location favored gene rearrangements that allowed the further recruitment of genes to the pathway. These events, a founding event and elongation events, may have been the key to the subsequent evolution of the benzoxazinoid biosynthetic cluster.


Background
Plants are sessile organisms which have evolved chemical and mechanical ways to defend against pathogens, herbivores and competitors. The synthesis of toxic compounds, generally arising from the so-called secondary metabolism is a hallmark of plant defense. Among the enzymes often involved in secondary metabolism and in particular in the synthesis of defense compounds and toxins in plants are the P450 enzymes [1]. These are heme-dependent oxidase enzymes that generally catalyze the insertion of one oxygen atom in a substrate after activation of molecular oxygen. The most common reaction catalyzed by this protein family is hydroxylation, but P450s are involved in a wide variety of catalyses such as dimerizations, isomerizations, dehydratations or reductions [2,3]. P450 proteins represent a large protein family very well represented in plants. For example 272 cytochrome P450 genes (CYP genes) are present in the Arabidopsis genome, including 26 pseudogenes [3]. This superfamily groups together proteins with as low as 20% sequence identity. Nevertheless secondary and tertiary structures are well conserved throughout the family. For instance P450 proteins share some conserved structures and sequences linked to properties such as oxygen or heme binding.
In grasses, P450s of the CYP71C subfamily are involved in the biosynthesis of the cyclic hydroxamic acids 2,4-dihydroxy-1,4-benzoxazin-3-one (DIBOA) and 2,4-dihydroxy-7methoxy-1,4-benzoxazin-3-one (DIMBOA). These natural compounds are well described as natural pesticides and allelochemicals, as chemical defense against microbial diseases and herbivory [4,5]. Their occurrence depends on the plant species [6]. For example, the predominant cyclic hydroxamic acid in rye Secale cereale is DIBOA whereas the main cyclic hydroxamic acid in maize Zea mays is DIMBOA, that differs from DIBOA by an additional methoxy group [5]. The presence of cyclic hydroxamic acids in grain crops has been known for over 50 years [7]. The four P450 genes involved in DIBOA biosynthesis, Bx 2 to Bx5 (known as CYP71C1 to CYP71C4 in the P450 nomenclature), were first described in maize by Frey et al. [8,9]. The final reaction steps leading to DIMBOA-glucoside were also elucidated in maize [10][11][12]. The full pathway involves 9 enzymes (BX1 to BX9) thought to act sequentially in the synthesis of DIMBOA-glucoside from indole-3-glycerol phosphate ( Figure 1). The Bx2 to Bx5 genes are clustered on the short arm of maize chromosome 4 [8,9]. Genetic analysis indicated that Bx1, Bx6, Bx7 and Bx8 are close to, or within this cluster, thus grouping genes of different families within a short chromosomal region [8][9][10]12]. Upon wounding, an additional O-methylation is activated, leading to HDMBOA [13,14] but the gene responsible for this reaction is still unknown. The same DI(M)BOA biosynthetic pathway has also been described in wheat Triticum aestivum, in rye S. cereale and in the wild barley Hordeum lechleri, the cultivated barley probably having lost the gene cluster [15][16][17]. The four CYP71C genes were cloned and characterized in diploid and hexaploid wheat and in wild barley [17][18][19].

Figure 1 The DIMBOA biosynthetic pathway and cellular compartment localizations.
Adapted from [8,12,20] A common evolutionary origin of this cluster of maize P450 genes by successive gene duplications has been proposed [8,16,20]. While it is no longer surprising to find biosynthetic gene clusters in bacteria and fungi [21], the nature and origin of such clusters in plants and animals is less studied. Large gene families such as the CYP family are often characterized by multiple gene duplications that leave a genomic trace as clusters of related genes [22,23]. However, these structural clusters such as the 13 CYP71B genes clustered on chromosome 3 of Arabidopsis thaliana, the 16 CYP6 genes clustered in the mosquito, or the 22 CYP2 loci clustered on mouse chromosome 7 [22,24,25] are not known to be co-regulated or to participate in a common pathway. Biosynthetic gene clusters are therefore different because they comprise non-homologous genes that function collectively. They have received increasing attention in plants [26], where known biosynthetic gene clusters serve in elaborating defense compounds such as phytoalexins from common precursors. They include the clusters for the biosynthesis of thalianol in A. thaliana, avenacin in Avena strigosa, momilactone and phytocassane in Oryza sativa [27] as well as the cyanogenic glucoside biosynthetic clusters in Lotus japonicus, Manihot esculenta and Sorghum bicolor [28].
The Bx gene cluster of maize is therefore of great interest, because it consists of an apparent structural cluster of four CYP71C genes in close genomic proximity with members of other gene families that, together, are known to direct the synthesis of DIMBOA glucoside from a common metabolic intermediate, indole 3-glycerol phosphate [20]. The sequence of the gene duplications, the nature of the ancestral genes and the mechanisms leading to the recruitment of several genes from different families into this biosynthetic cluster have not been determined in detail, yet these are the key questions in understanding the evolution of secondary metabolic gene clusters in plants [26,27].
Here, we take advantage of the information from newly sequenced genomes of Poaceae and of the known biochemical properties of the enzymes to describe the evolutionary origin of the DIMBOA biosynthetic pathway. We used a phylogenetic approach to establish the sequence of duplications of an ancestral CYP71C gene leading to the Bx2-Bx5 cluster in maize. We delineated the involvement of critical amino acids in achieving the current biochemical specificities of the P450 enzymes. We studied the syntenic relationships of genes at the interface between primary metabolism and the DIMBOA pathway: ZmBx1, ZmTSA and TSAlike, ZmIgl and Igl-like. We also searched for synteny around the other genes of the Bx cluster to gain insights into the origin of the entire DIMBOA pathway.

Sequence data
The BX2-5 protein and transcript sequences from maize [8] were used for BLAST approaches (blastn, blastp and tblastn). Sequences with more than 50% amino acid sequence identity with one maize CYP71C were used to identify equivalent CYP71C sequences in other Poaceae. Searches were made on the BLAST server of NCBI [29] for all Poaceae, on the maize genome website [30], on the Brachypodium distachyon website [31] and on the Phytozome database [32]. P450 sequences were manually checked and their annotation corrected when necessary based on known P450 motifs [3]. Incomplete sequences, and pseudogenes were removed from further analysis, resulting in 75 sequences representing the CYP71C subfamily in Poaceae. Five A. thaliana sequences were chosen to root the tree (CYP76C3, CYP76C7, CYP76C4, CYP76C1 and CYP76G1). For the studies on Bx1 (and Igl, TSA, TSA_like), Bx6, Bx7, Bx8 and Bx9, sequences were searched by BLASTP based on maize sequences with the same criteria (id% > 50) and on the same databases as for the P450 study. Intron/exon structures were determined whenever the genomic sequence was available.

Phylogenetic tree reconstruction
The selection of the best-fit model of protein evolution for the CYP71C subfamily was done with ProTest software version 2.4 [33] based on protein alignment obtained with MUSCLE software (available on the website phylogeny.fr [34]) using the default parameters [35][36][37]. No curation was needed because of the high identity between sequences and the correct alignment of P450 motifs. The JTT model [38] was then chosen as the best fit model of protein evolution for our sequences. The tree of the CYP71C subfamily in Poaceae was generated on the phylogeny.fr website [34] using maximum likelihood as reconstruction method (PhyML program). Bootstrapping (100 iterations) was done to document branch support. These criteria were also used for the reconstruction of the other BX phylogenetic trees.

Positive selection in coding sequences
The codeml program of the PAML software package was used to detect positive selection in the CYP71C subfamily [39]. cDNA sequences were aligned based on protein alignment by using the RevTrans 1.4 server [40]. We used branch and site models which allow ω to vary among branches in the phylogeny or among sites on the alignment [41,42]. First, we compared the one-ratio model (M0, one ω estimated for all sites) to the free-ratio model (independent ω ratios for each branch) to test for the hypothesis of variable ω among branches. Then, the M7 (beta distribution of ω ratios, with 0 < ω < 1) versus M8 (extension of M7 with a supplementary site class with ω > 1 estimated from the dataset) comparisons were done to test for positive selection among sites. Finally, the models M8 and M8a (ω = 1) were compared to determine if for a small fraction of sites the ω estimated under M8 was significantly higher than 1 [43,44]. Likelihood-ratio tests (LRT) were used to compare models. Twice the log-likelihood difference 2∆lnL was compared with a χ² distribution with degrees of freedom corresponding to the difference of free parameters numbers between the two models compared. The branch-site model MA was further applied to our data to detect positive selection affecting only a few sites on pre-specified lineages [45]. Four independent branches were studied with this model and Likelihood-ratio tests (LRT) were used to compare models with Bonferroni's multiple testing corrections.

Functional divergence analysis of BX2 to BX5 protein clades
The software DIVERGE2 (Detecting Variability in Evolutionary Rates among Genes) was used to identify critical amino acid residues involved in functional innovations after gene duplications [50]. The coefficients of type I and type II functional divergence (θI and θII, respectively) between two chosen clades were calculated. If these coefficients are superior to 1, it means that some amino acids were subject to altered selective constraints (type I functional divergence) or that a radical shift of amino acids physicochemical properties occurred after gene duplication and/or speciation (type II functional divergence) [50][51][52][53] . Thirty nine sequences comprising each BX clade and the sequences associated to the BX2 clade were aligned using MUSCLE (default parameters). This alignment and the equivalent portion of the tree were used as input parameters for analyses with DIVERGE2 software. We compared BX clades resulting from duplication events to each other. For each test, the posterior probability of each site to be under functional divergence was calculated. Sites with a posterior probability Qk > 0.67 (to select only radical cluster-specific sites [50][51][52][53]) were localized on the maize BX2-5 alignment.

BX2 molecular modeling and dockings
A model for BX2 was generated with the iterative threading assembly refinement server (I-TASSER [54]) [55,56] using the following templates: (pdb numbers) 3k9v, 3e6i, 3na0, 2hi4, 3czh and 1izo. We obtained a model with a C-score = −1.46, indicating a correct prediction. The top I-TASSER templates all presented a normalized Z-score > 1 with a large protein coverage (always superior to 76%), reflecting the high accuracy of the alignment. Predictions of binding sites were confident with a BS-score > 0.5 for all predictions. Dockings were assessed using the AutoDock4 software package [57,58]. Proteins were first prepared by removing water molecules, checking for missing atoms, adding of non-bonded hydrogens and computed Gasteiger charges under the AutoDockTool ADT 1.5.4 [59]. The protein model was then used to construct a grid box with a grid-point spacing of 0.375 Å. To dock the indole substrate, the input protein was the heme-containing Bx2. The grid centre position was positioned on top of the heme and included 40x40x40 = 64,000 points. Four independent Lamarckian genetic algorithm searches (250,000 and 2,500,000 evaluations) were run. For each analysis, the solutions were compared to determine if the results were reliable and reproducible. Interacting residues were also compared to determine the conserved residues in contact with indole.

Synteny of the Bx genes
Three approaches were used to analyze the synteny of maize Bx genes within available Poaceae genomes. We used the SyMAP v3.4 (Synteny Mapping and Analysis Program) to identify and display genome synteny alignments for Z. mays, O. sativa, B. distachyon, S. bicolor and S. italica genomes [60]. The Plant-Synteny site was also used to study synteny between Z. mays, Triticum, O. sativa, S. bicolor and B. distachyon [61]. Maizesequence.org was used to identify putative syntenies between maize and O. sativa and S. bicolor.

Phylogenetic analysis of the genes related to Bx2-5 in the CYP71C subfamily
We aligned all available CYP71C sequences and reconstructed a phylogenetic tree to determine the evolutionary origin of the P450 genes of the Bx cluster in Poaceae ( Figure 2 and Additional file 1). The tree we obtained showed that the maize BX2-5 sequences had orthologs distributed in four branches, boxed in Figure 2 as BX2 -Bx5 clades. Each of these four branches was strongly supported (bootstrap values 91-100%). The four clades contained sequences of biochemically characterized CYP71C enzymes such as those of maize [8,16] and wheat [19] as well as two sequences encoding P450s from rice that have not yet been biochemically characterized. Moreover, while the BX3, 4 and 5 clades were monophyletic, the BX2 clade was included in a parent clade that contained fourteen "BX2-like" sequences from non-benzoxazinoid producers: S. bicolor, S. italica and O. sativa. The rest of the tree was composed of at least four branches with more distant sequences of the CYP71C subfamily from maize, B. distachyon, S. italica, S. bicolor and O. sativa.

Multiple intron losses in the CYP71C subfamily
Frey et al. in their landmark paper on the Bx cluster [8] suggested that the position of introns in the CYP71C genes was indicative of a common evolutionary origin. We therefore located the position of introns in all the available genomic CYP71C sequences (Additional file 1). Only two intron positions were found, both in phase zero, and located at conserved positions, 192 (intron 1) and 336 (intron 2) of BX2 (hereafter all amino acid positions are given in terms of equivalent positions of maize BX2). Thirty four genes had both common introns while twenty five sequences had only intron 2. All genes in the BX3, BX4 and BX5 clades had both introns, except for maize BX4 that had only intron 2. The three genomic sequences of the BX2 clade had only intron 2. Intron 2 is an ancestral intron found in most plant P450s [22], while the conserved position of intron 1 suggested that it had the same origin among all the sequences. The presence or absence of the introns did not follow a simple phylogeny. This indicated independent intron losses or gains in the different branches. Identification of paralog/ortholog relationships in the phylogeny allowed us to place duplication and speciation events on the tree. Accordingly, the most likely hypothesis is that intron 1 was introduced after the first duplication in the CYP71C subfamily and that more than a dozen independent intron losses have occurred. The alternative hypothesis of an equally high number of independent gains of intron 1 occurring at the same position and in the same phase is unlikely.

Positive selection in the CYP71C coding sequences
We analyzed the evolution of the CYP71C subfamily and in particular the evolution of the maize Bx2-5 genes. We hypothesized that, after duplication and/or speciation events, the four CYP71C genes were subjected to selection. On one hand, the proteins maintained their P450 structure and overall catalytic activity (redox partner binding, dioxygen binding and activation), and on the other hand, they each acquired a distinct and high substrate specificity [8,16]. The conserved features of the four proteins i.e. helical structures, SRS regions, and conserved P450 motifs are illustrated in Figure 3. This balance of P450 structure conservation and substrate specificity would imply that genes are globally under purifying or neutral selection (ω < 1 or ω = 1 respectively), explaining the high sequence identity between proteins, and that some particular sites are under strong positive selection (ω > 1), leading to the substrate specificity. This hypothesis was evaluated by testing for positive selection among specific lineages, among sites and among specific sites in specific lineages on our phylogeny. Branch and site models were first used to test the hypothesis of heterogeneous levels of selection among lineages and to test for positive selection among sites. The branch model would allow us to test if positive selection existed among the various branches. The site models permit ω to vary among sites. As the positively selected sites in proteins are generally very few, focusing on the overall sequence would fail to detect these selected sites as the ω value on the overall alignment would be inferior to 1. These tests would thus allow us to detect sites under positive selection in the CYP71C subfamily. The changes at these sites being favored during evolution would be potentially important for the specificity of the proteins. The comparison between the one-ratio model M0 (one ω ratio for all sites calculated from the data) and the free-ratio model (independent ω ratios for each branch) revealed a heterogeneous selection level among lineages (2∆lnL = 725.58, p < 0.01). The three branches that define the BX3, 4 and 5 clades ( Figure 2) showed ω values superior to 1 supporting the hypothesis of adaptive evolution while the branch defining the BX2 clade did not. The test of positive selection among sites, M7/M8 was statistically significant (2∆lnL = 680.30, p < 0.01). This suggested that the ω ratio was variable among sites and that about 17% of the sites were under positive selection with ω = 1.69. The M8/M8a test was also significant, meaning that the estimated ω in M8 was statistically different from 1. The M8 model identified 52 sites under positive selection with a posterior probability superior to 0.95 (Additional file 2). We also used the branch-site model MA to go further into the specific evolution of each group of the 4 P450s of the DIMBOA-biosynthetic pathway. The use of this model allows the detection of positively selected sites on prespecified lineages. The phylogenetic tree shows the duplicated origin of the four genes, represented by the four Bx clades on Figure 2. Our aim was to detect positive selection that affected sites in each of the four clades by specifying the foreground branches during the tests. Bonferroni's corrections for multiple testing were done. The test returned sites under positive selection (posterior probablility >0.5) in the BX2 and BX4 clades but the associated ω2 was only statistically superior to 1 (α = 0.10) for BX2 sequences (ω2 = 6.44). Omitting sites in the membrane anchor not involved in substrate specificity, three sites were identified for BX2, including one site in the E/F loop and another one in SRS5 ( Figure 3).

Functional divergence analyses of the BX P450 enzymes
Tests of functional divergence with the DIVERGE2 software were realized to identify critical amino acid residues in each of the four BX clades. Functionally divergent sites may explain the substrate specificity of each protein. Thirty nine protein sequences, including the sequences of the four BX clades and the BX2-like sequences were aligned and the equivalent portion of the tree were used for analyses of type I and type II functional divergences among BX clades. Clades of BX proteins originating from gene duplications were compared to each other. We also compared the BX2 clade versus the larger (BX4(BX3/BX5) clade and the BX4 clade versus the (BX3/BX5) clade. Sites with significant posterior probability Qk > 0.67 were localized on the maize BX2-5 protein alignment ( Figure 3).
Comparisons with proteins from the BX4 clade showed no type I or II functional divergence. The comparison of the BX2 clade with the (BX4(BX3/BX5)) clade returned a statistically significant type I functional divergence (θI = 0.151 ± 0.067; LRT = 5.10; p > 0.05). Met 77 of maize BX2, localized within the first P450 motif, was returned with Qk > 0.67. The type I functional divergence test between the BX2 and BX3 clades was statistically significant (θI = 0.353 ± 0.084; LRT = 17.610; p > 0.05) and identified 5 sites. One site was on P450 motif 1, one in SRS1, one in loop E/F, another one in SRS2, and the last one was after SRS5.
Functional divergence of type II was also found between these two clades (θII = 0.140 ± 0.055). Twenty sites were identified, among which 14 amino acid residues differed between maize BX3 and BX2.
The type I functional divergence test between the BX2 and BX5 clades was statistically significant (θI = 0.294 ± 0.097; LRT = 9.216; p > 0.05) with one site (Glu 451) at Qk > 0.67. The type II functional divergence test (θII = 0.214 ± 0.054) returned twenty five sites. In the comparison of the BX3/BX5 clades, functional divergence of type I (θI = 0.417 ± 0.066; LRT = 39.688; p > 0.05) was found on 7 sites. All these sites differed between maize BX3 and BX5, suggesting a radical shift in the rate of evolution for these amino acids. The type II functional divergence test returned a θII equal to 0.222 ± 0.0356 and identified 25 sites which all differed between maize BX3 and BX5. They were placed throughout the protein alignment with two sites in SRS1, three sites in SRS2, two sites in SRS3, three sites in SRS4, one in SRS5, two in the E/F loop and two in the F/G loop.

Comparisons of maize BX2-BX5 sites subjected to positive selection and to functional divergence
Positions on the maize BX2-5 alignment with at least one site under positive selection and/or functional divergence were checked. Among them, we focused on sites showing either amino acid conservation or four different amino acids. These sites could be involved in the specific evolution of the maize BX2-BX5 functions. Thirty two positions were extracted. Among them, 23 involved one or more radical biochemical changes, potentially leading to substrate specificity. Nine sites were identified when focusing on sites within SRSs or in the F and G helices or in the F/G loop ( Figure 3). One site was added to these critical amino acids: the positively selected and under type II functional divergence Met 234 of BX2 just before the Fhelix. This site seems to be very important from a selection point of view as it implies a radical biochemical shift from positively charged to non-polar.

Maize BX2 modeling and indole docking
Homology modeling was used to understand the relationship between sites under positive selection and/or functional divergence and protein substrate specificity. The model was also used in docking approaches to gain insight into the specific interactions of the protein with its substrate indole. Indole was docked into the heme-containing BX2 protein model and the residues in contact with indole in the 40 computed docked conformations were compared. Four protein residues were repeatedly found, namely Leu 253 on 35/40 conformations, Ile 527 on 36/40 conformations and Ala 341, Thr 345 and the heme molecule on all docked conformations. The relative position of the indole secondary amine was oriented towards Ala 341 and Ile 527 on 31/40 conformations. The four protein residues (Figures 3 and 4A) are all in SRS regions, and three were identical in the other BX P450s. The fourth residue, the apolar Ile 527 was replaced by a polar Thr in BX4 and BX3 and by an apolar Met in BX5. When looking at the spatial localization of all sites previously identified as under positive selection and/or under functional divergence, two sites, Ser 156 and Cys 157, were oriented toward the active site. Both sites were identified above among the ten sites potentially important for the substrate specificities (Figures 3 and 4B).

Synteny of the Bx2-5 genes and their homologs among Poaceae
We analyzed the synteny of the maize Bx2-5 genes to determine whether it was conserved in the genomes of other Poaceae ( Figure 5). Although synteny blocks were detected between maize chromosome 4 and O. sativa, S. bicolor, B. distachyon and S. italica genomes, none of them contained genes homologous to the Bx2-5 genes. All the CYP71C homologs included in our phylogenetic analysis from those species were found in other genome locations. The cluster formed by S. italica CYP71C88, C89 and C92 for instance, was syntenic to the cluster of maize CYP71C36, C56 and C57 on chromosome 2 ( Figure 5A) and phylogenetically distant from the Bx cluster. Moreover, non-Bx genes in the vicinity of the Bx2-5 cluster did not show conservation of synteny. Therefore, in all currently available genomes, the position of the Bx2-5 genes is unique to maize. Bx6 and Bx6 paralog are in blue, Bx7 in green, Bx8 and Bx9 in grey. Arrows indicate the gene orientation. The chromosomes are taken from the maizesequence.org website [30]. B. The Bx gene cluster (~ 264 kb) on maize chromosome 4. Gene length is not to scale. 085303 is the GRMZM2G085303 gene of unknown function 085854 is the GRMZM2G085854 gene, an uncharacterized glucosyltransferase that has 27% amino acid identity with BX8. Bx7 is at about 15 Mb towards the centromere (right), separated from the cluster by about 360 other genes. Bx6 is not found in the current release of the maize genome. Genetic evidence places it within 7 cM of Bx3 and Bx4 [11,12] We also studied the phylogeny and syntenic relationships of the other maize Bx genes (ZmBx) involved in the DIMBOA-biosynthetic pathway. Their possible Bx orthologs in other Poaceae genomes were located on chromosomes (or scaffolds for S. italica) and their intron positions were mapped.

Phylogeny and synteny of the ZmBx1, ZmTSA, ZmIgl and ZmIgl_like genes
The Bx1 and Igl genes are thought to result from duplications of the TSA gene [62]. The twenty nine most closely related sequences from Z. mays, H. lechleri, T. aestivum, S. bicolor, S. italica, B. distachyon, O. sativa and Hordeum vulgare were analyzed. On the phylogenetic tree, IGL/BX1 and TSA/TSA_like sequences formed two distinct clades. In the IGL/BX1 clade, ZmBX1 and T. aestivum BX1 were grouped together ( Figure 6). The IGL sequences from Panicoideae were also closely related to BX1. A single branch included IGL from Pooideae and Ehrhartoideae. At first glance, these sequences followed the Poaceae phylogenetic history but this did not explain the origin of BX1 paralogs in T. aestivum, which belongs to the Pooideae (Figure 7). A likely explanation is that two duplication events of an Igl ancestor occurred before the separation of both Pooideae-Ehrhartoideae and Panicoideae ( Figure 6), followed by reciprocal losses of paralogs in the Panicoideae and in the Pooideae-Ehrhartoideae lineages. The ancestral intron pattern of ZmBx1 supports this view ( Figure 6), but additional genome sequences from Pooideae are needed to validate it. TSA and TSA_like proteins form the second clade on our tree. They probably originated from a more basal duplication of a "TSA ancestor", leading to the IGL/BX1 and TSA/TSA_like lineages.  [65,[89][90][91][92][93][94]. Boxes indicate species accumulating benzoxazinoids [6,17,19,20,95]. In the Hordeum genus, benzoxazinoids are detected in some of the H-clade species (such as H. flexuosum and H. lechleri) and not in some species of the I-clade (such as H. vulgare and H. spontaneum) [17,63,96]. Bold branches indicate species included in our CYP71C phylogenetic study (only B. distachyon is included when considering the common branch B. distachyon/Lolium). Species / genus are in bold when a complete genome is available.
We did not detect blocks of synteny in B. distachyon, S. italica, O. sativa and S. bicolor that contained any genes homologous to ZmBx1. However, the maize TSA_like/Igl and TSA regions on chromosomes 1 and 7, respectively, showed syntenic blocks in common within each of the four other Poaceae genomes. The results imply that these genes arose early during the evolution of Poaceae and maintained their syntenic relationships. Bx1 thus appears to be the only paralog of the TSA/IGL genes that lacks conserved synteny, and its location following rearrangement separated its evolutionary fate from that of the other TSA/IGL genes.

Phylogeny and synteny of the ZmBx6, 7, 8 and 9 genes
ZmBX6, the 2-oxoglutarate-dependent dioxygenase which catalyses the hydroxylation reaction at C-7 of DIBOA, was compared to the 15 closest sequences from Poaceae to reconstruct a phylogenetic tree (Figure 8). The ZmBX6 protein was most closely related to Sb10g006910 of S. bicolor (59.9% identity) and to Si010340 of S. italica (65.7% identity). No sequences of Bx6 orthologs are currently available from wheat or wild barley. Although genetic mapping studies place the ZmBx6 gene in the ZmBx cluster on maize chromosome 4 [11,12], we did not find ZmBx6 in the genomic sequence but found a close paralog (76.5% identity at the protein level) isolated on the long arm of chromosome 2. As the genomic sequence close to Bx4 contains gaps of undetermined length, it is likely that Bx6 is lacking in the current version of the maize genome. Therefore, synteny conservation around the ZmBx6 gene cannot be studied presently. ZmBX7 is a member of the large O-methyltransferase gene family, but paralogs were only found when lowering the BLASTP searches cutoff to 40% identity. The closest homologs were found in S. italica (Si022355, 62.9% identity and Si010415, 46.2% identity) and B. distachyon (Bradi1g47030, 49.9% identity). The corresponding genes had a common intron with ZmBx7 in phase 0. ZmBx7 was located on the short arm of maize chromosome 4, at about 15 Mb from Bx2, with over 360 annotated genes separating it from the Bx cluster. Its genetic proximity was previously reported [11,12]. The ZmBx7 paralog of B. distachyon was on chromosome 1 and the two paralogs of S. italica were present on scaffolds 3 and 7. No single copy orthologs were common to maize and these two species in the vicinity of the Omethyltransferases genes.
The two glucosyltransferases ZmBX8 and ZmBX9 are also members of a multigene family and the twenty closest homologs from Avena strigosa, O. sativa, S. cereale, T. aestivum, S. bicolor and S. italica were analyzed. As ZmBx8 was located between Bx2 and Bx5 where no synteny conservation was observed (see above), we focused on synteny blocks containing ZmBx9. However, the corresponding blocks in other genomes did not contain any genes homologous to ZmBx9. The reconstructed phylogenetic tree underlined the strong relationship between ZmBX8 and ZmBX9 (74.4% identity) and the presence of one close paralog in S. cereale (BAJ07107) and four in T. aestivum (BAJ07089, BAJ07091, BAJ07092 and BAJ07090) (Figure 9). These four proteins have been recently described as the glucosyltransferases involved in the DIMBOA-biosynthetic pathway in wheat [63]. The closest relatives to these glucosyltransferases were from S. italica (Si013725, Si015361 and Si013705), S. bicolor (Sb09g028320), O. sativa (Os11g0441500 and Os11g0444000) and A. strigosa (UGT710E5) (Figure 9). However none of these homologs were included in blocks of conserved synteny with the ZmBx9 region. We have mined the complete genomes of Poaceae that are currently available, and all available sequences for genes related to the known Bx genes of maize. Our study included both species that produce benzoxazinoids and species that do not (Figure 7). Our results are relevant to the origin of benzoxazinoids, the evolution of the four key P450 genes of this cluster, and the chromosomal arrangement of this cluster.

Origin of benzoxazinoids
Our analysis supports the hypothesis of Frey et al. [20] that the pathway of benzoxazinoid biosynthesis in Poaceae is monophyletic, at least until DIBOA-glucoside. Indeed the phylogeny of the enzymes involved in its synthesis (BX1 to BX5, and the glucosyl transferase BX8 and BX9) is congruent in benzoxazinoid producers for which the genes encoding these enzymes are known. It is likely that a basal ancestor of Poaceae in the Late Cretaceous, i.e. nearly 70 MYA [64], produced DIBOA glucoside. The origin of the last two steps, hydroxylation (BX6) and methylation (BX7) is less clear, because sequences orthologous to the maize genes are not currently available. Benzoaxinoids have also been detected in single isolated species of two distant related orders of dicots, the Ranunculales and the Lamiales [6]. The dicots diverged from the monocots about 150 to 300 million years ago [65]. As in Poaceae, the first step involves an indole-glycerol phosphate lyase (IGL), separating the DI(M)BOA biosynthetic pathway from primary metabolism. However, duplications leading to Igl were independent events in monocots and dicots [66]. The second step of the pathway leading to the production of indolin-2-one from indole is a P450dependent activity in the DIBOA-producing dicots [66] as in Poaceae, but as no CYP71C genes are found in dicots, the genes responsible cannot be orthologs. Independent evolution of benzoxazinoid biosynthesis in monocots and dicots is therefore most likely. This was also demonstrated recently for the biosynthesis of cyanogenic glucosides between monocots and dicots [28], and between plants and insects [67].

Duplications and neofunctionalization of the CYP genes
The phylogenetic relationship of each of the four BX-type P450s is robust and the Bx genes of the CYP71C subfamily genes in maize, wheat and wild barley are thus out-paralogs, their duplications occurring before speciation events. Interestingly, the intron-exon pattern of the CYP71C genes is not phylogenetically informative because of the many independent intron losses. The common ancestral origin of DIBOA glucoside biosynthesis in Panicoideae and Pooideae, and the availability of genome sequences from the two lineages allowed us to examine in greater detail the origin and evolution of the four P450 genes, the Bx2-Bx5 genes.
Although the similarity of the sequences and their clustering in maize would suggest that they are the product of a simple series of successive tandem duplications [27], our analysis shows a more complex evolutionary history. The first P450 gene, Bx2, is a member of a clade that contains many CYP71C sequences from a variety of Poaceae, including non-benzoxazinoid producers. Within this branch, all sequences from wheat and barley form a strongly supported monophyletic clade and all these sequences are biochemically characterized as encoding BX2 enzymes [8,16,19]. The other sequences, from rice, sorghum and millet have not been functionally characterized to date. These Bx2-like sequences might be active in other secondary metabolic pathway(s) in non-benzoxazinoid producers. We reconstructed with Codeml the sequence of the ancestor of the BX2/BX2-like clade, synthesized it and produced it in yeast to test our hypothesis of the original biochemical properties of BX2. However we were not able to express an enzymatically functional protein (results not shown). The function of the two close BX2 homologs of O. sativa, CYP71C15 and CYP71C16, is unknown. Moreover no benzoxazinoids have been found in rice ( [5] and unpublished data in [20]). Maize BX2 catalyzes N-demethylation of p-chloro-N-methylaniline [16] in addition to the hydroxylation of indole, suggesting that the ancestral enzyme may have had some catalytic versatility as well. The original function of the BX2/BX2-like P450 ancestor thus remains hypothetical.
If we assume that a common ancestor encoded a BX2-like protein catalyzing indole hydroxylation, then its duplication could explain the expansion of the pathway from indole to DIBOA (Figure 10), by a series of non-successive duplications. This sequence of events is based on our phylogenetic analysis, and on the catalytic properties of the current representatives of each branch in maize, wheat and barley [8,16,17,19]. The first duplication led to "a more recent BX2 ancestor" that maintained a C2 hydroxylase activity in all benzoxazinoid producers until the present time. Its duplicate was the common ancestor of BX3/4/5 whose neofunctionalization led to a new C3 hydroxylase activity that has been maintained in BX3. The BX3/4/5 ancestor was then duplicated first to give BX4 neofunctionalized to catalyze oxidative ring expansion and one more time leading to the present BX5 neofunctionalized to an N-hydroxylase. In this scheme ( Figure 10) Bx3 was duplicated twice, giving first Bx4 then Bx5, so that the pathway was not elongated by successive tandem duplications of the gene encoding the last step. The relative orientation and distance between the Bx2-5 genes in maize ( Figure 5B) also suggest inversions and rearrangements with only Bx3 and Bx4 present as a tandem array. The neofunctionalization of each new duplicate subtly modified the active site resulting in the specific regioselectivity of oxidation on the indole-like substrate ( Figure 10). Neofunctionalization as understood here is restricted to the substrate specificity of each enzyme with the conservation of their overall P450 characteristics. Both maize and wheat BX3 enzymes hydroxylate indoline-2-one as well as 1,4-benzoxazin-3-one [15,16], suggesting that the enzyme does not discriminate between its natural substrate and its ringexpanded analog. The use of a ring-expanded substrate (HBOA) is a feature of BX5 substrate specificity that supports the origin of Bx5 from Bx3.

Sites under selection in the P450 enzymes
The sequence identities at the amino acid level between BX2 to BX5 orthologs (e.g. from 76 to 79% identity between maize and wheat sequences, 76 to 81% between maize and wild barley sequences) and the BX enzyme substrate specificities are very high [8,[15][16][17]. This suggests that the neofunctionalization following duplication in the ancestral, basal poaceous species was accompanied by selection on only a few sites. This is indeed what our results show. The BX3, BX5 and BX4 clades in our phylogeny are under strong adaptive evolution and some sites are under positive selection. We have identified the specific sites in the maize BX2-5 proteins that have undergone positive selection and functional divergence (type I and/ or type II). More than a half of these sites are localized between the SRS1 and the I-helix and, in particular, among SRS1, 2 and 4 and between the E and G helices. These regions seem to have a major impact on the catalytic properties and the evolution / divergence of the BX2-5 enzymes. Such a non homogenous repartition of sites along the protein was previously described in a phylogenetic analysis of the CYP3 genes family [68]. That study proposed that SRS1, 5 and 6 were performing a universal CYP3A function and that SRS2, 3 and 4 were responsible in part for the functional differences among the enzymes of this family. In contrast, functional divergence in the vertebrate CYP2 family is not clustered in SRSs regions but distributed all along the CYP2 alignment [69].
We identified ten residues potentially important for the substrate specificity of each of the four maize P450s (Figures 3 and 4B). Furthermore, the docking of indole in our model of the active site of maize BX2 identified four residues in close contact with the substrate. One was inside SRS6 and the three other sites were localized in SRS2 and SRS4. The four sites we identified in maize BX2, among which one (Ile 527) differs in the BX3-BX5 proteins, could explain the specificities of the enzyme. Our study thus points to Ser 156, Cys 157, Met 234 and Ile 527 as first candidates for mutagenesis approaches to test their impact on the biochemical properties of maize BX2.
The limited number of sites shown to be under selection or standing out as important in our modeling is consistent with experimental evidence obtained with other P450 enzymes. Few amino acid changes are needed to significantly change substrate specificity of these enzymes. For example, four residues at positions 117, 209, 365 and 481 of Mus musculus CYP2A5 are sufficient to determine the steroid substrate specificity [70]. The specificity of this P450 is influenced by the hydrophobicity and residues size [71,72]. In the human CYP2C19, three residues at positions 99, 220, and 221 are key residues that determine the hydroxylase activity for omeprazole [73]. In CYP2B11 of Canis lupus three sites in putative SRSs (residues 114, 290 and 363) are important for the enzyme substrate specificity and regio/stereoselectivity [74]. In Papilio polyxenes,residues 116, 117, 371 and 484 of CYP6B1 are critical for substrate binding affinity [75]. While the CYP2 and CYP6 enzymes are predominantly xenobiotic-metabolizing enzymes with diffuse substrate specificity, the CYP71C studied here are thought to have a tight biosynthetic function. For plant biosynthetic P450s, mutagenesis of the mint limonene hydroxylases from the CYP71D subfamily showed that a single amino acid change is sufficient to convert a C6-to a C3-hydroxylase [76]. The limited number of critial residues identified in our study is therefore reasonable to explain the subtle shifts in substrate regioselectivity that accompanied the evolution of the BX2-BX5 enzymes but this will require experimental confirmation.

Origin of the Bx biosynthetic cluster: founding event
The assembly of a biosynthetic gene cluster in plants was discussed by Frey et al. [20] with the benzoxazinoid pathway as a prototype. They saw three essential and sequential modules: a branchpoint reaction, chemical modification leading to a biological active compound, and detoxification. Osbourn [26,77] assigned the branchpoint reaction to a signature enzyme, and chemical modification to tailoring enzymes, but did not recognize the importance and integrality of detoxification. Instead, chromosomal clustering was seen as a way to prevent the accumulation of toxic intermediates in a pathway [77]. Although the description of pathway assembly by the juxtaposition of three modules is a useful guide, our analysis suggests that in the case of the benzoxazinoid pathway clustering of the first two genes, Bx1 and Bx2, is the key event. Furthermore, we propose that both BX1 and BX2 are signature enzymes that only together constitute a branchpoint committing to benzoxazinoid biosynthesis. We propose to call their clustering the "founding event" of the biosynthetic cluster. The evidence for this view can be developed as follows: Indole as a product of a branchpoint reaction is not a committed precursor of benzoxazinoids. Phylogenetic analysis shows that an initial duplication of TSA led to an IGL ancestor that was further duplicated to Bx1. TSA is a subunit of tryptophan synthase in "primary metabolism", and current IGL enzymes are involved in the generation of biologically active volatile indole [62,78]. Thus IGL and BX1 perform the same reaction, albeit with diverged catalytic properties [8,78,79]. Our study of TSA, TSA_like, Igl and Bx1 demonstrated that Bx1 originated before the radiation of Poaceae. Although Grun et al. proposed that independent TSA gene duplication events have created Bx1-function in maize and wheat on one hand and in barley on the other [17], our phylogenetic analysis clearly shows that this is not the case. The sequence of H. lechleri named as "BX1" by Grun et al. [17] clearly falls within the IGL clade, with strong bootstrap support. Furthermore, its catalytic properties are not characteristic of BX1 but rather of an IGL [17,79]. Its kcat/KM (31 mM − 1 . s − 1 ) is much closer to that of Z. mays IGL (23 mM − 1 . s − 1 ) than to Z. mays BX1 (215 mM − 1 . s − 1 ) [17]. The sequence is therefore an ortholog of H. vulgare BAJ91226, and the H. lechleri Bx1 remains to be discovered. The absence of synteny between ZmBx1 and other Poaceae genome regions is quite surprising as we found synteny conservation for TSA, TSA_like and Igl. The synteny of ZmBx1 and Bx2 [8] is the only conserved feature in all benzoaxazinoid producers and points to the uniqueness of this clustering. The phylogenetic position of BX2 is similar to that of BX1, a sequence emerging from a background of several duplication events and remarkable only because it forms a monophyletic clade with the wheat and wild barley enzymes of identical function. Both BX1 and BX2 have close homologs that are not involved in benzoxazinoid biosynthesis, so that they are signature enzymes catalyzing branchpoint reactions only when considered together. Initial clustering of both genes enabled their subsequent coevolution and divergence from Igl_like and Bx2-like genes, respectively. In this view, genomic rearrangements that led to the random clustering of the newly duplicated ancestral Bx1 and Bx2 genes represents the "founding event" of the biosynthetic cluster. This key innovation is therefore a structural one, and it makes sense because it distinguishes a biosynthetic cluster of genes from an assemblage of genes (not necessarily clustered) that form a biosynthetic pathway. The terms signature/branchpoint and decoration/chemical modification can equally be applied to biosynthetic clusters as to genomically dispersed biosynthetic pathways, so a more specific nomenclature is required. What then would be the second step ? We propose to call it "elongation" in preference to recruitment, to emphasize the genomic feature over its functional aspect.

Origin of the Bx biosynthetic cluster: elongation
Conservation of the Bx1 -Bx2 synteny from maize to wheat and rye [80] confirms that the founding event of the biosynthetic cluster occurred in an ancestor of Poaceae. Elongation of the cluster to Bx5 by the P450 duplications and gene rearrangements described above led to cluster of 5 genes in maize. Is this the ancestral state or did the Bx3-5 genes integrate the cluster together, as a separate event ? In both rye and wheat Bx1-2 and Bx3-5 form two distinct clusters. In rye, ScBx1 and ScBx2 are located on chromosome 7R and ScBx3, ScBx4 and ScBx5 are on chromosome 5R. In wheat, TaBx1 and TaBx2 are closely located on group-4 chromosomes and TaBx3, TaBx4 and TaBx5 are closely located on group-5 chromosomes [80]. Rye 5R and 7R chromosomes have high conserved synteny with group-5 and group-4 chromosomes of wheat [80][81][82]. Nomura et al. [80] proposed that the ancestral Bx cluster was split in a common ancestor of rye and wheat. Moreover, the presence of microlinearity and partial orthology has been demonstrated between wheat group-7 chromosomes (containing the glucosyltransferase of the DIMBOA-biosynthetic pathway) and maize chromosomes 1 and 4 (including respectively ZmBx8 and ZmBx9). In rye, the glucosyltransferase is also found isolated on the 4R chromosome, consistent with the known synteny between rye and wheat [83]. The addition to the cluster of a glucosyltransferase gene necessary to convert DIBOA (the product of BX5) to DIBOA-glucoside resulted from an ancient gene rearrangement, and our phylogenetic analysis indicates the orthology of the rye, wheat and maize genes. We conclude that this cluster elongation was also an early event in an ancestral Poaceae species. It becomes difficult to distinguish detoxification as proposed by Frey et al. [20] and clustering (here of a glucosyltransferase) to prevent toxic intermediate accumulation as proposed by Osbourn [77]. Glucosyltransferases are integral parts of the cyanogenic glucoside biosynthetic clusters [27]. It has been proposed that physical disruption of the components of the cyanogenic glucoside metabolon can be a way to diversify the products of the pathway, as different intermediates are toxic to different targets [84]. There are therefore different ways to maintain integrity of a biosynthetic cluster: the genomic integrity that favors cosegregation of all components, and, at least for cyanogenic glucosides, the subcellular integrity as a metabolon. Is the glucosyltransferase the "last" enzyme in DIBOA-glucoside biosynthesis acting on the product of BX5 ? This is traditionally shown (Figure 1), and does not address the question of the earlier intermediates. Yet the products of BX3 and BX4 are observed as glucosylated metabolites in maize [85], Dutartre et al., in preparation], so that the contribution of a glucosyltransferase to the biosynthetic cluster may have preceded the last two duplications of Bx3. The maize glucosyltransferases have significant activity towards HBOA, the product of BX4 [10]. Significantly, Bx8 is closest to Bx1 and 2 in the cluster ( Figure 5B) and is the ortholog of the rye and wheat genes. The wheat sequences result from hexaploidization, with one duplication in the B genome [83]. Bx9 is a recent duplicate of Bx8 in the maize lineage as shown by our phylogenetic analysis and that of Sue et al. [83]. It is not located in the cluster, and its catalytic properties [10] indicate that it has lost considerable activity toward DIBOA. Following the Bx8/Bx9 duplication, the sequence divergence of Bx9 and its new location on another chromosome probably led to a new physiological role different from benzoxazinoid biosynthesis. The lack of QTL involved in DIMBOA synthesis associated with the Bx9 region [86] supports this conclusion.
Further elongation of the cluster corresponds to the aromatic hydroxylation and methylation of DIBOA glucoside by BX6 and BX7 [12]. The evolutionary history of this elongation, and indeed of the further methylation to HDMBOA-glucoside is difficult to ascertain at present, because Bx6 and Bx7 have not been sequenced in other benzoxazinoid producers, and the last methyltransferase gene is still unknown. Bx6 has a close paralog on chromosome 2, and Bx7, while close to the Bx cluster, is about 35 cM distant. Intriguingly, the closest homologs of Bx6 and Bx7 in S. italica are located on scaffold 7, in close proximity to four P450 genes, CYP71C81, and of the cluster of CYP71C88, 89,92. The latter is orthologous to the maize CYP71C36, 56, 57 cluster on chromosome 2. While the function of these genes is currently unknown, it is tempting to speculate that Bx6 and Bx7 are moonlighting in a different biosynthetic cluster. Their position on the outside of the Bx cluster may have prevented them from being lost when the S. italica ancestor lost the Bx1-Bx5 genes. We note that the Km of BX6 and BX7 towards their benzoxazinoid substrates is the poorest of all BX enzymes [12], and as they take glucosylated substrates and not their aglycone, the aglycone contribution to substrate specificity may be weak, supporting the alternative function hypothesis.
There are several examples of plant genes clusters located close to the tip of the chromosome as is commonly found in actinomycetes and ascomycetes [26]. This position is particularly favorable to adaptive evolution and to coordinated regulation [26,87]. The presence of genes in a cluster would favor their co-segregation and thus favor the rapid evolution of the linked genes. As no Bx cluster is found in S. bicolor, O. sativa and H.vulgare, it is likely that the original, complete cluster was lost in a single evolutionary event [80]. The chromosomal position of the Bx cluster may have favored through cosegregation the subsequent rearrangements of the Bx8, then Bx6 and Bx7 genes in close proximity. Additional genome sequences from benzoxazinoid producers may provide additional evidence for this sequence of events. Our analysis suggests that the key factor in the origin of biosynthetic gene clusters in plants, and perhaps in other organisms, is a "founder event" where the first two genes originating from random duplications and rearrangements find themselves linked and commit to a new pathway. Whether co-regulation or co-segregation is a most important result of this clustering remains to be ascertained. The genetic arguments for the primacy of co-segregation have long been known. In higher eukaryotes, the evidence and mechanisms of co-regulation of recently rearranged genes are less well established.

Conclusions
Our phylogenomic analysis of the origin of the Bx cluster in maize shows that the first two closely linked genes of the benzoxazinoid pathway are located at a chromosomal region that has no synteny conservation with the genomes of other Poaceae beyond Bx1/Bx2 themselves, and is therefore unique. Rearrangements following duplications of an IGL/TSA gene and of a CYP71C gene resulting in the clustering of the new copies (Bx1 and Bx2) constitute the founding event of the biosynthetic cluster. This founding event is a genomic character, different and perhaps more general than "branchpoint reaction" [20] or "signature enzyme" [26,77] that denote biochemical characters that would not adequately describe the importance of the tight clustering of Bx1 and Bx2. Elongation of the cluster involved duplications of a Bx2-like CYP71C gene and neofunctionalizations that involved positive selection at few distinct sites of these P450 enzymes. At least one glucosyltransferase gene was recruited into the pathway and rearranged into the cluster. Our data are consistent with our current understanding of biosynthetic clusters in plants [20,27,28,86], but highlight the importance of the founding event in seeding a biosynthetic cluster.   m  a  y  s   W           K  Q  H     R  L  P  P  T  P  P  G  K  L  P  I  I  G  H  L  H  L  I  G  S  H  P  H  V  S  F  R  D  L  H  A  K  Y  G  H  N  G  L  M  L  V  Q  V   B  X  3   _   Z   _  m  a  y  s   R           K  Q  Q  Q  R  L  P  L  P  P  S  P  P  G  K  L  P  I  I  G  H  L  H  L  I  G  A  E  T  H  I  S  I  R  D  L  D  A  K  H  G  R  N  G  L  L  L  L  R  I   B  X  5   _   Z   _  m  a  y  s   G           K  P      P  L  P  P  S  P  P  G  T  L  P  I  V  G  H  L  H  H  I  G  P  Q  T  H  I e  m  b  r  a  n  e  a  n  c  h  o  r   P  P  T  P  P  G  K  L  P  I  I   P  P  S  P  P  G  K  L  P  I  I   P  P  S  P  P  G  T  L  P  I