Genome analyses were performed on ten distinct genomes comprising four Dicots, three Monocots, the Pteridophytae Selagniella moellendorffii, the moss Physcomitrella patens and the Alga Chlamydomonas reinhardtii. Sequences and most annotations were downloaded from the Joint Genome Institute website. The predicted proteomes analysed 7 9 10 11 12 13 14 15 16 17 were deduced from the annotations given in Table 1.

The number of loci corresponds to the protein coding genes predicted by the annotation.

Genomic sets of predicted proteins were challenged with three Pfams HMMs 18 : PF00106, PF01370 and PF01073 using HMMER3. SDR Nomenclature Initiative HMMs were defined and updated as described previously 5 . The five SDR types (‘classical’, ‘extended’, ‘intermediate’, ‘divergent’, and ‘complex’) each has an HMM trained to identify sequences of respective type. The HMMs were created using HMMER3, with manually adjusted alignments of representative sequences as seed. Cutoffs are used to decide if a hit is significant or not: ‘classical’ — 138, ‘extended’ — 108, ‘intermediate’ — 162, ‘divergent’ — 160, and ‘complex’ — 140. In addition to the five types, an ‘unknown’ label is used for sequences with scores lower that these cutoffs but still high enough to safely predict the sequence as an SDR: ‘classical’ — 29, ‘extended’ — 75, and ‘divergent’ — 100. Scores below the cutoffs are considered not positive.

For the PLR/IFR family, an HMM was created and incorporated to the ‘SDR Nomenclature Initiative’ set of family HMMs. The procedure for training the HMM was the same as previously developed with iterative refinement of the model until no new members were found 5 .

For each sequence recognized by a HMM (hit), a score was assigned. Yet, several sequences were only recognized by one or two HMMs (either the Pfam derived HMM or the SDR-type HMM) and sometimes with a very low score. Thus, we defined a series of rules schematized in a decision tree (Figure 1). Sequences that were recognized by SDR nomenclature initiative were directly considered as positive. Sequences identified with both remaining sets of HMMs were also considered as positive. For the remaining sequences recognized only by one HMM, we first looked at the existence of strong homology with positive hits identified in the previous steps in order to include putative ‘truncated’ proteins (see Inventory refinement, below). Alternatively, we checked individually the existence of structural data in the scientific literature, which allowed either including some hits in our inventory or discarding certain families of enzymes, notably the medium-chain dehydrogenases, that display distinct structural motifs. In absence of structural data, the sequences recognized by a single HMM could not be classified and were included in a list of ambiguous sequences (Additional file 1: Table S1) that contains oxido-reductases that still await structural data before confirming or infirming their belonging to the SDR superfamily.

For the gene loci that are associated with several gene models and therefore with different protein predictions, a sole amino acid sequence was selected according two criteria: (1) the maximum HMM score and (2) the maximum alignment score deduced from a BlastP performed on other plant genomes. When the HMM and BlastP analyses led to contradictory predictions, a single protein prediction was manually selected after aligning the different gene models with its closest homologues. To include in the SDR classification the truncated proteins that failed to be recognized by the HMMs, a BlastP sequence search was performed on each genome using as query sequences the complete list of SDRs recognized in the first round of HMM searches. All sequences that displayed a segment of 60 amino acids with more than 50% identity were classified in the same type or family as its closest homologue.

Phylogenetic analyses and distance matrices were built using the Mega5 package 19 . Full length amino acid sequences were aligned using the ClustalW algorithm. Distance matrices evaluating the percentage of sequence identity were calculated on the basis of p-distance with the pairwise deletion option. Unless stated differently, phylogenetic trees were built using the Neighbor-Joining method. The percentage of replicate trees in which the associated taxa clustered together was calculated in the bootstrap test (500 replicates). Trees were drawn to scale, with branch lengths in the same units as those of the evolutionary distances used to infer the phylogenetic tree. The evolutionary distances were computed using the Poisson correction method and were expressed as the units of the number of amino acid substitutions per site.

Principal Component Analysis was performed on the distribution matrix given in Figure 2 using the R 2.14.1 software 20 with in-house developed scripts (Elie Maza, personnal communication). The robustness of the conclusions was checked by carrying the same analysis after removal of the individual exhibiting extreme values (SDR108E).

Initial HMM analyses were performed on ten complete genomes: 4 Eudicots (Arabidopsis thaliana, Populus trichocarpa, Vitis vinifera, Glycine max), 3 Monocots (Zea mays, Oryza sativa and Sorghum bicolor), the lycophyte Selaginella moellendorffii, the moss Physcomitrella patens and the unicellular green alga Chlamydomonas reinhardtii (Table 1). The predicted ‘proteomes’ deduced from the genome annotations were searched against three distinct sets of HMMs: the Pfam HMMs considered to encompass most SDR (PF00106, PF01370, PF01073), HMMs developed in the framework of the SDR nomenclature initiative 5 21 and a set of HMMs developed to predict the type (‘classical’, ‘extended’, ‘intermediary’, ‘divergent’ and ‘complex’) of SDR (see Methods).

This first analysis led to an exhaustive inventory of plant SDRs presented in supplemental data (Additional file 1: Table S1 and Additional file 2: Table S2). This inventory was divided into a main list (Additional file 1: Table S1), where the HMM scores or the high similarity with known SDRs were sufficient to establish a good prediction, and a complementary list of ambiguous SDR predictions (Additional file 2: Table S2), containing proteins with low HMM scores and absence of structural data (see decision tree in Figure 1 and Methods). Despite its very low HMM scores, we included in the main list a large family that comprises pinoresinol reductase (PLR), isoflavone reductase (IFR), vestitone reductase, phenylcoumaran benzylic ether reductase and eugenol synthase. Indeed, the structures of several members of this family were resolved by crystallography and the data revealed the presence of a SDR-typical Rossmann-fold 22 23 24 25 . Subsequently, an HMM was created and incorporated to the ‘SDR Nomenclature Initiative’ set of HMMs. The PLR/IFR family was named SDR460A, where the ‘A’ stands for ‘atypical’.

The number and the distribution of plant SDRs of different types are summarized in Table 2. As in other Eukaryotes, the major types consist of ‘classical’ and ‘extended’ SDRs. ‘Divergent’ SDRs are limited to one conserved family: an enoyl-ACP reductase (ENR) involved in lipid biosynthesis (AT2G05990.1 in Arabidopsis) 8 26 . While neither ‘intermediate’ nor ‘complex’ types are found in plants, we can notice a high number of ‘unknown’ types, meaning that the sequence patterns clearly differ from the other types. As previously noticed 2 , the SDR family is highly represented in the plant kingdom: while 73 SDRs were numbered in the human genome 27 and 39 in the cyanobacteria Synechocystis sp. PCC 6803 28 , the number of SDRs in land plants vary from 126 in the moss P. patens to 315 in soybean (G. max). Even if we consider the variations due to the genome sizes (Table 1), SDRs are more represented in the Angiosperms than in the alga C. reinhardtii or in P. patens, suggesting a relationship between the emergence of vascular plant and the apparent multiplicity of plant SDRs.

Families with low scores and no structural data (listed in Table S2) were omitted. *The presence of divergent SDRs in Sorghum bicolor was deduced from the Sbicolor_79_peptide annotation.

The HMMs developed by the SDR nomenclature initiative 5 aim at classifying the SDR superfamily into a large number of families, at a level where this classification would be informative regarding the functions of SDRs. In a first analysis, the HMMs defined in the frame of the SDR nomenclature initiative directly recognized 74% of plant SDRs. After performing similarity searches and associating truncated proteins with its closest homologues (see Methods), the proportion of non-classified SDRs dropped markedly since 94.5% of plant SDRs were categorized into 49 families. While the majority of these families are found in most Tracheophytes (Table 3), seven (SDR58C, 59C, 74C, 86C, 90C, 103U, 107C; Additional file 2: Table S2) are found only in P. patens or in C. reinhardtii. The occurrence of different family sequences in the analysed genomes was represented as a heat map (Figure 2).

Each family was associated with a representative gene and, when possible, with a specific SDR nomenclature initiative HMM. Information on the occurrence of SDRs in different genomes are reported by the taxon name (ViridP: Viridiplantae; LandP: Embryophytae; TracheoP: Tracheophytae; FlowerP: Magnioliophyta). Average pairwise identities were calculated from the sequences of plant genomes. Ambiguously predicted SDRs and families absent in flowering plants were omitted. *: occurrence in Viriplantae was deduced from the presence of homologues in other Green Algae genomes.

On the opposite, 5.5% of plant SDRs (from 4% in Angiosperms to 29% in C. reinhardtii) remained unclassified. The existence of these orphan SDRs lays in the conception of the ‘SDR nomenclature initiative’ HMMs. In order to achieve robust HMMs, the authors considered only families with sufficient number of representative and non-redundant sequences 5 , thus excluding SDR families with too few members. To circumvent this difficulty, we examined the possibility to define new families on the sole basis of amino-acid sequence conservation. Therefore, all the unclassified SDRs from the main inventory (Additional file 1: Table S1) were associated to its closest homologues using BlastP searches and sequence alignments. Interestingly, all the unclassified sequences from Angiosperms clearly matched with at least one Arabidopsis SDR, the e value obtained from a BlastP against Arabidopsis predicted proteome never exceeding 1 e-40. Thus, seven new clusters were defined on the basis of sequence conservation, four being common to all the Viridiplantae genomes while three were found only in land plants (Figure 2 and Table 3). Within these clusters, the average pairwise sequence identities ranged from 48% to 62%. These conservation rates are consistent with the average pairwise identities observed for the families defined by a ‘SDR nomenclature initiative’-HMM, that ranges from 37% to 82% identity (Table 3). All these clusters were represented by a limited number of sequences in each genome, supporting the explanation that the lack of ‘SDR-nomenclature-initiative’-HMMs is simply the consequence of an insufficient set of sequences and that these families might be defined in the future, with the release of new sequences in the UNIPROT database. To complete the plant SDR classification, each new cluster was assigned a representative gene, based on an Arabidopsis thaliana identifier. While all angiosperms SDRs could be categorized in a family, defined either by a specific HMM or by primary structure conservation, 15 sequences from C. reinhardtii, 4 sequences from P. patens and one sequence from S. moellendorffii were too distant to other SDR sequences and remained unclassified.

By extension, the ambiguous SDR sequences were also clustered on the basis of sequence homologies, allowing the definition of nine potential families (Additional file 2: Table S2). Yet, in absence of structural data confirming the existence of typical SDR structures, these sequences were not analysed further.

In a last step, plant SDR classification was combined with functional information. Taking advantage of our previous bibliographic research 8 and of the annotations found for Arabidopsis (TAIR10), we completed the classification by mentioning all the known functions described in the scientific literature in Table 3. Also, to each family, a representative gene was chosen according to three criteria: (1) favour Arabidopsis accessions with respect to the quality of TAIR annotations and its pertinence as a model plant; (2) when possible, opt for genes that have been functionally characterised; otherwise (3), priority was given to the accession that displayed the lowest average distance with other members of its family.

The distribution of the different families in the different taxa was further examined to understand the evolution of the plant SDR superfamily. We first addressed the question of potential origins of the different SDR families. Out of the 49 families listed in Table 3, 32 were found both in the alga C. reinhardtii and in the majority of land plants, suggesting that most plant SDRs families emerged prior to land plant radiation that started -460 Myear ago, in the Ordovician period 29 . For three additional families (SDR40C, SDR42E and SDR358U), the absence of a member in C. reinhardtii or even in P. patens predicted proteomes masked the occurrence of these families in other genus of green algae (Volvox, Micromonas, Chlorella and Ostreococcus), suggesting either that some genomic sets are incomplete or that the families are ancestral but the genes might have been lost in some taxa. In addition, 10 families absent in green algae are common to all land plants (Figure 2 and Table 3), indicating that 45 families are shared among land plants (embryophytes). 48 families are common to vascular plants as 3 additional families are specific to S. moellendorffii and Angiosperms. At last, a sole family, SDR115E, is found only in Angiosperms. The origins of some families may be very ancient: SDR1E, 2E, 6E and 7C families are found in all domains of life (Archea, Eukaryote, Prokaryote) while the SDR12C, 17C, 25C, 34C, 35C, 22E and 31E families are common to the majority of Eukaryotes 5 . Besides, several ancestral SDR families are close to Prokaryotic ‘homologues’. For example, the origin of the plastids is illustrated by the presence of chloroplastic SDRs similar to its cyanobacterial homologue. In a recent paper, Kramm et al. 28 listed 39 SDRs in the genome of the cyanobacteria Synechocystis sp. PCC 6803. 20 of these SDRs show clear homologies (>35% identity) with plant SDRs (data not shown). The SDRs clusters present both in cyanobacteria and plant genomes include the very ancient families (SDR1E, 2E, 3E, 6E) and several plastidial proteins involved in primary metabolism, such as sulfolipid biosynthesis protein (SDR52E), protochlorophyllide oxidoreductase (SDR73C), 3,8-divinyl protochlorophyllide a 8-vinyl reductase (SDR98U) or the members of the fatty acid synthase (FasII) complex (SDR152C and SDR87D).

The origin of these taxon-specificities probably results from three evolutionary mechanisms: horizontal gene transfers, differentiation of a novel family from a pre-existing SDR family and loss of genes. Indeed, Tarrio et al. 30 established that the Vein Patterning 1 (SDR75U) gene family had undergone five lateral gene transfer events, one occurring from bacteria to an ancestor of land plants. Conversely, extensive search of SDR homologues in the Genbank database revealed clear homologies between independent taxa, such as the similarities between the Tracheophyte SDR68C members and its Proteobacteria homologues or the close relationship between plant PLR-IFR family and Bacteria or Ascomycete isoflavone reductase-like proteins (data not shown), thus illustrating the possible importance of horizontal gene transfers. An original example of SDR differentiation is illustrated by the emergence of the Angiosperm-specific HC-toxin reductase (SDR115E) family, involved in the pathogen Helminthosporium carbonum (HC) toxin reduction 31 . Since previous phylogenetic analyses 32 showed the existence of significant homologies between HC-toxin reductase (SDR115E) and the large dihydroflavonol 4-reductase (4-DFR, SDR108E) family, we integrated SDR108E and SDR115E amino acids sequences in the same alignment and phylogenetic analysis (Figure 3 and Additional file 3: Figure S1A). The topology of the deduced tree (Figure 3 and Additional file 3: Figure S1A) suggests that the SDR115E branch belongs to a larger clade that includes 4-DFR (AT5G42800.1 cluster, 33 ), anthocyanidin reductase (AT1G61720.1 cluster, 34 35 ) and the brassinosteroid related 4-DFR-like protein BEN1 (AT2G45400.1 cluster, 36 ). The robustness of this topology was further checked using different phylogeny algorithms (Neighbour-Joining and Maximum Likelihood) or rooting the tree with external sequences from other SDR families (SDR1E, SDR6E, SDR31E). All trees displayed similar topologies, SDR115E members always clustering with 4-DFR, anthocyanidin reductase and BEN1 (data not shown), thus supporting the view that the HC-toxin reductase (SDR115E) branch evolved from an ancestor belonging to the SDR108E family. The divergences of sequences within the SDR108E-115E ‘clade’ are sufficient to establish two distinct HMM profiles. At last, two distinct features may illustrate the loss of genes in SDR evolution: (i) although found in Monocots, grapevine, poplar and soybean genomes, the SDR115E family is absent in Arabidopsis genome or ESTs database; (ii) some families found in P. patens or in S. moellendorffii genomes (SDR74C, 86C, 103U, Additional file 2: Table S2) are absent in all the Angiosperms genomes, suggesting that genes might have been lost during before flowering plants radiation.

The second obvious feature, when observing the distribution of SDR families (Figure 2), is the expansion pattern of the different families. A Principal Component Analysis (PCA) was performed on the distribution matrix used to build the heat map presented in Figure 2. It allowed the individualization of ten families displaying high values on the first axis (Figure 4A). All these families are characterized by a large number of members in contrast to the majority of SDR families represented in plant genomes with a limited set of sequences. Interestingly, the second axis is mainly driven by the vectors formed by P. patens and C. reinhardtii genomes (Figure 4B) and it discriminates two patterns of diversification: families expanded both in the moss P. patens and in vascular plants (SDR1E, SDR2E, SDR6E, SDR7C, SDR50E) and families expanded in vascular and flowering plants (SDR65C, SDR108E, SDR110C, SDR114C and SDR460A).

Remarkably, all the five families expanded in vascular plants comprise enzymes involved in secondary metabolism (Table 3): tropinone reductases (SDR65C) are known for their involvement in alkaloids biosynthesis; SDR110C NAD-dehydrogenases oxidize various phenolic or terpeninc compounds, including xanthoxin, a precursor of abscissic acid (ABA); SDR114C menthone and salutaridine reductase, are involved in monterpene and alkaloid metabolism respectively; the large SDR108E family members catalyze the reduction of several phenolic precursors (4-dihydroflavonol, anthocyanidin, cinnamoyl-CoA, phenylacetaldehyde or eutypine) and last, the atypical PLR/IFR family (SDR460A) is also involved in phenolic metabolism. On the opposite, several poorly diversified clusters (SDR52E, 73C, 152C, 87D, 357C) that contain highly conserved sequences participate in primary metabolism such as chlorophyll synthesis or degradation, lipid metabolism or vitamin synthesis.

For multigenic SDR families, the analyses can be conducted further with phylogenetic calculations. To illustrate the importance of this complementary approach, we focused on two large families involved in secondary metabolism: SDR110C (ABA2 xanthoxin dehydrogenase family) and SDR108E (4-DFR) family. For tropinone reductase (SDR65C) and menthone/salutaridine reductase (SDR114C) families, readers are referred respectively to Brock et al. 37 and Ziegler et al. 38 for complete phylogenetic analyses.

In our previous review 8 , we listed six different functions described for SDR110C in the scientific literature: Arabidopsis ABA2 xanthoxin dehydrogenase (abscissic acid biosynthesis) 39 40 , rice diterpenoid momilactone synthase A 41 , mint (Mintha sativa) isopiperitenol dehydrogenase 42 , Forsythia intermedia secoisolariciresinol dehydrogenase (lignan biosynthesis), the maize or rice feminization gene TASSELSEED2 43 and the Arabidopsis AtATA1 gene, involved in pollen and anther tapetal cells development 44 . A phylogenetic tree was established from the analysed genome sequences and completed with mint isopiperitenol dehydrogenase and forsythia secoisolariciresinol dehydrogenase sequences (Figure 5 and Additional file 3: Figure S1B). Remarkably, the six functions described in the literature were distributed on five different clades, thus giving valuable hypotheses regarding the putative function of the orthologues or paralogues in other Angiosperm species. On the different clades, we also observe that highly homologous SDRs are often clustered in specific chromosomal regions, illustrating the importance of gene duplication events in the diversification process. At last, the different accessions of Selaginella are distributed in three distinct clades. However, the bootstrap values of the different nodes were too low to clearly establish a clear relationship between lycophytes and Angiosperms SDR110C sequences.

For the highly variable SDR108E family, we included the SDR115E family in the analysis as both family are closely related (see above). As reported for the SDR110C family, several branches can be associated with functions described in the literature 8 : 4-DFR 33 , anthocyanidin reductase (AnR) 34 35 , HC-toxin reductase 31 , phenylacetaldehyde reductase 45 , cinnamoyl-CoA reductase (CCR) 46 or eutypine reductase 47 (Figure 3). In contrast to SDR110C, the tree is also informative concerning the evolution of SDRs among land plant since distinct sequences from S. moellendorffii and P. patens are clearly associated with independent clades. These associations are of special interest for certain classes of enzymes such as the CCR catalysing the first irreversible oxidation step leading to monolignol synthesis. Indeed, several enzymes involved in the lignin biosynthesis pathway appeared early in land plant evolution and the moss are believed to accumulate uncondensed monolignols 48 . Thus, the association on the same branch of sequences from P. patens and S. moellendorffii with Angiosperms bona fide CCR suggests that the enzyme anciently acquired its specificity and diverged rapidly from other SDR108E members. Last, as observed for SDR110C, several highly similar genes are clustered in specific chromosomal regions. Hence, with numerous members and a low conservation rate of amino acid sequences, the SDR108E family and its daughter branch SDR115E constitute a good example of a gradual and fast evolution of a multigenic family. Since the majority of the described enzymes reduce phenolic compounds, we may hypothesize that the SDR108E evolution accompanied tightly the complexification of phenolic and phenylpropanoid metabolism during land plant radiation.

Although essential for the functional study of large SDR families, phylogenetic analyses may be very informative for smaller families as well. This is the case of the Non-Yellow-Coloring 1 (NYC1) chlorophyllase b family, where the phylogenetic analyses clearly divide the family in two distinct clades: NYC1 and NOL (Non-yellow-coloring-Like) that diverged from a common ancestor (Figure 6). It was suggested that during evolution, the divergence led to the emergence of a functional hetero-oligomer, since both genes are necessary for chlorophyll b degradation 49 50 .

At last, we carried analyses on three SDR families involved in lipids primary metabolism: fatty-acid synthase (FAS-II)-β-ketoacyl synthase (SDR152C) 51 , (FAS-II)-enoyl-ACP reductase (SDR87D) 52 and the UDP-sulfoquinovose synthase (SDR52E) involved in sulfolipids biosynthesis 53 . By contrast with the families involved in secondary metabolism discussed above, the average sequence identity is high (Table 3), ranging from 68% to 75%. When the N-terminal chloroplast peptide signals are removed from sequence alignments, the average identities reach the scores of 79% (SDR152C) and 84% (SDR87D and SDR 52E). Phylogenetic trees were deduced from sequence alignments (Figure 7). Despite the presence of some duplication events observed for SDR152C (Figure 7A) and SDR87D (Figure 7B), the tree topologies are in good agreement with plant taxonomy for all the three SDR families, thus suggesting that the primary structure has been conserved under a high pressure of selection.