Bacterial production of long-chain omega-3 polyunsaturated fatty acids (PUFAs), such as eicosapentaenoic acid (EPA, 20:5n-3) and docosahexaenoic acid (DHA, 22:6n-3), is constrained to a narrow subset of marine γ-proteobacteria. The genes responsible for de novo bacterial PUFA biosynthesis, designated pfaEABCD, encode large, multi-domain protein complexes akin to type I iterative fatty acid and polyketide synthases, herein referred to as “Pfa synthases”. In addition to the archetypal Pfa synthase gene products from marine bacteria, we have identified homologous type I FAS/PKS gene clusters in diverse microbial lineages spanning 45 genera representing 10 phyla, presumed to be involved in long-chain fatty acid biosynthesis. In total, 20 distinct types of gene clusters were identified. Collectively, we propose the designation of “secondary lipids” to describe these biosynthetic pathways and products, a proposition consistent with the “secondary metabolite” vernacular. Phylogenomic analysis reveals a high degree of functional conservation within distinct biosynthetic pathways. Incongruence between secondary lipid synthase functional clades and taxonomic group membership combined with the lack of orthologous gene clusters in closely related strains suggests horizontal gene transfer has contributed to the dissemination of specialized lipid biosynthetic activities across disparate microbial lineages.
Citation: Shulse CN, Allen EE (2011) Widespread Occurrence of Secondary Lipid Biosynthesis Potential in Microbial Lineages. PLoS ONE 6(5): e20146. https://doi.org/10.1371/journal.pone.0020146
Editor: Jack Anthony Gilbert, Argonne National Laboratory, United States of America
Received: March 29, 2011; Accepted: April 19, 2011; Published: May 19, 2011
Copyright: © 2011 Shulse, Allen. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: A gift from the Legler Benbough Foundation: http://www.benbough.org/. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Bacteria have evolved the capacity for fatty acid biosynthesis for incorporation into membrane phospholipids in three distinct ways. The most common mechanism is the prototypical type II Fatty Acid Synthase (FAS II), well characterized in E. coli . In this system individual enzymatic activities reside on discrete enzyme products, encoded by the fatty acid biosynthesis, or fab, genes. An alternative pathway, albeit significantly less pervasive in bacterial lineages, is the type I FAS system (FAS I). The canonical pathway found in eukaryotic organisms, FAS I is also found in the Corynebacterineae of the order Actinomycetales . FAS I consists of a large, multifunctional biosynthetic complex containing all enzymatic domains necessary for acyl chain elongation and functional derivatization and is responsible for the production of both membrane phospholipid fatty acyl chains as well as precursor fatty acid molecules for elongation to long-chain mycolic acids in members of the Corynebacteriaceae, Mycobacteriaceae and Nocardiaceae families .
A third mechanism of de novo fatty acid synthesis coexists with the FAS II in a narrow subset of marine Gammaproteobacteria , . This pathway consists of a novel iterative FAS/PKS system and is responsible for the production of long-chain omega-3 polyunsaturated fatty acids (PUFAs) such as eicosapentaenoic acid (EPA, 20:5n-3) and docosahexaenoic acid (DHA, 22:6n-3) , , . The genes responsible for bacterial omega-3 PUFA production, designated pfaA–E, possess multiple fatty acid biosynthetic enzyme activities as integrated domains within operon-encoded gene products .
In addition to polyenoic fatty acyl products, related FAS/PKS gene clusters have been shown to synthesize other specialized long-chain fatty acid products. The C26 to C32 fatty acid alkyl chains containing hydroxyl or ketone moieties found in the heterocyst glycolipids of filamentous nitrogen-fixing cyanobacteria ,  and the C22 to C26 fatty acids of phenolic lipids comprising the dormant cysts of the gram-negative bacterium Azotobacter vinelandii  are both synthesized via an analogous iterative type I FAS/PKS mechanism. Their long chain length, typically containing ≥20 carbons, distinguishes these fatty acyl products from those produced by FAS II (≤18 carbons). We collectively term these specialized lipid products “secondary lipids”, to emphasize the accessory nature of these lipid molecules and distinguish these products from those synthesized via core, or primary, fatty acid biosynthetic mechanisms.
All three of these biosynthesis methods make use of highly conserved enzyme activities encoded on integrated domains (FAS I and Pfa synthases) or autonomous monofunctional gene products (FAS II) to accomplish the cycle of condensation, reduction, dehydration, and reduction necessary to produce a final fatty acid product. The biosynthetic reaction sequence of an elongation cycle includes the following activities: 1) Ketoacyl synthase [KS] catalyzes a condensing function responsible for chain-elongation; 2) Ketoacyl reductase [KR] catalyzes the reduction of the carbonyl group to a hydroxyl group; 3) Dehydratase/Isomerase [DH/I] catalyzes the dehydration of the β-hydroxyacyl intermediate generated by KR to a trans-2-enoyl derivative and the subsequent isomerization from the trans-2 to the cis-3 configuration; and lastly 4) Enoyl reductase [ER] catalyzes the reduction of double bond generated by DH to complete the chain elongation process. Other essential activities include acyl carrier protein [ACP] function which tethers the growing fatty acyl chain as a thioester as it is acted upon by other enzyme activities and phosphopantetheinyl transferase [PPTase] activity which converts ACP products from the inactive apo- form to the active holo-form via the posttranslational addition of a 4′-phosphopantetheine prosthetic group from acetyl coenzyme A. Acyltransferases [AT] catalyze the general transfer of a nascent acyl substrate from acyl-CoA to ACP for elongation of the fatty acyl chain. Malonyl-CoA:ACP transacylase [MAT] is a type of acyltranferase that specifically catalyzes the transfer of a 2C malonyl moiety from malonyl-CoA to ACP. A final component, specific to PKSs and Pfa synthases, is the chain length factor [CLF] domain shown to determine the ultimate chain length of the reaction product .
Evidence for additional widespread capacity for secondary lipid production potential via the Pfa synthase mechanism in environmental samples has recently been reported . In that study, culture-independent molecular surveys were used to identify 13 novel groups based on KS domain homology from disparate marine habitats. Beyond the marine environment, the phylogenetic extent and ecological breadth of secondary lipid biosynthetic potential has not been investigated.
The current capacity for inexpensive, rapid, whole genome sequencing has allowed for broad genome comparisons among diverse microbial lineages. Here, we expand upon previous studies of the distribution and diversity of secondary lipid production potential in the marine environment  by analyzing all sequenced microbial genomes for the presence of FAS/PKS gene clusters homologous to those involved in PUFA secondary lipid biosynthesis. We uncover and classify multiple previously unrecognized FAS/PKS gene clusters in diverse bacterial lineages representing varying physiologies and life histories, significantly expanding the palette and pervasiveness of gene products linked to specialized microbial metabolites.
Results and Discussion
In the following sections, we describe the diversity and organization of secondary lipid biosynthetic gene clusters identified in this study, first addressing those with characterized products then progressing to novel clusters with uncharacterized products. Additional analyses are presented to support the definition of secondary lipid synthases and differentiate these gene clusters from those involved in PKS or NRPS products based on PPTase domain conservation. Next, we analyze the genomes of secondary lipid synthase containing organisms for the presence of other lipid biosynthetic activities, including FAS II and ole gene functions, two systems whose products interact with pfa gene products. Lastly, we analyze the ecology and physiological properties of these organisms to provide insight into possible traits unifying secondary lipid production potential and present evidence showing that horizontal gene transfer has aided in the dissemination of these biosynthetic gene clusters.
Diversity and organization of FAS/PKS gene clusters
The presence of multiple acyl carrier protein (ACP) domains in a single gene product is a distinguishing characteristic of Pfa synthases. It has been shown that an increase in the number of ACPs increases the biosynthetic throughput of PUFA product synthesis . Most FAS/PKS gene clusters retrieved in this study contain multiple tandem ACP domains (Figure 1). However, although all gene clusters with tandem ACPs are presumed to produce fatty acyl products, not all fatty acyl-producing FAS/PKS gene clusters contain multiple ACPs (e.g. ars gene cluster responsible for the production of alkylresorcinols and alkylpyrones in Azotobacter vinelandii). Therefore we did not exclude gene clusters containing a single ACP if the domain content and organization was consistent within a candidate FAS/PKS cluster and phylogenetic analysis of the proximal ketoacyl synthase (KS) domain, harbored within the pfaA homolog (Figure 1), supported a common evolutionary relationship with validated FAS/PKS pathways.
Eighty-six microbial genomes with either the pfa gene cluster or a homolog were identified. Type designations are based on KS phylogeny, conserved domain structure and product information, if available.
A total of twenty distinct classes of FAS/PKS gene clusters, designated Types A–T (Figure 1), were identified in 86 finished or draft genomes available in GenBank and/or the Joint Genome Institute's Integrated Microbial Genomes (IMG) databases as of February 2011 (2.2% of 3839 genomes analyzed). The division of gene clusters into “Types” was strongly supported by three independent metrics: (i) domain count and organization as analyzed by non-parametric multidimensional scaling (NMDS; Figure S1); (ii) phylogenetic analysis of component enzymatic domains in each secondary lipid synthase identified (Figure 2 and Figure S2); and (iii) pathway-product information, if known. Alphabetic ordering of cluster “Types” is based on phylogeny of KS domains (Figure 2). All FAS/PKS gene clusters analysed were found to contain at least one KS domain, ACP domain, and ketoacyl reductase (KR) domain, and various combinations of malonyl-CoA:ACP transacylase (MAT), acyltransferase (AT), chain length factor (CLF), dehydratase/isomerase (DH/I), enoyl reductase (ER), and phosphopantetheinyl transferase (PPTase) domains. It is important to investigate both the domain content and organization of each gene cluster as these factors determine the ultimate length and functionality of the chemical product . The NMDS plot (Figure S1) is a visual representation of the similarity among all the gene clusters based on these two factors.
Gene cluster Types are colored and given a letter label. Asterisks represent Types first clustered and described in this study. Bootstrap values ≥50% are indicated by dots. The Escherichia coli DH10B FabF protein was used as the outgroup.
Secondary lipid synthase types with characterized products.
Type A and Type B produce omega-3 PUFAs and are found primarily in γ-proteobacteria of marine origin (Table 1). Type A represents the canonical secondary lipid synthesizing gene cluster responsible for eicosapentaenoic acid (EPA, 20:5n-3) synthesis and consists of five genes, pfaABCDE . Domain order within these genes is highly conserved: pfaA [KS-MAT-ACP(4–6)-KR], pfaB [AT], pfaC [KS-CLF-DH2], pfaD [ER], and pfaE [PPTase] (Figure 1). An exception to this conservation is found in Pseudoalteromonas sp. DS-12 where the PPTase domain is incorporated into pfaC . Intriguingly, analysis of the DH domains in Pseudoalteromonas sp. DS-12 reveals that the first DH domain is phylogenetically more similar to the second DH domain from all other gene clusters harboring two DH domains, and vice versa (Figure S2). This implies that at some point in the evolution of this gene cluster a section of pfaC and all of pfaE were translocated. It is unknown whether the Pseudoalteromonas gene cluster retains the ability to produce EPA .
The Type A cluster was previously found to be conserved in 15 sequenced Shewanella strains . In the present study, nine additional Shewanella genomes were analyzed and all were found to contain a coherent Type A gene cluster (Table 1). The presence of a TypeA pfa gene operon in all genomically characterized members of the Shewanellaceae (n = 24) suggests the genetic potential for EPA production is a defining characteristic of this lineage. The Shewanella are a genus of Gammaproteobacteria known for their ability to utilize a wide variety of electron acceptors and have been recovered from diverse environmental sources , , . In addition to the 24 analyzed Shewanella genomes, the EPA-synthesizing Type A gene cluster is also found in three Vibrio genomes (Vibrio sp. MED222, V. splendidus 12B01, and V. splendidus LGP32), two Photobacterium genomes  and one Pseudoalteromonas genome .
Type B represents the docosahexaenoic acid (DHA)-producing pfa gene cluster, which differs from the EPA-producing pfa cluster by the insertion of an additional KS domain in pfaB (Figure 1). An active site cysteine was found to be absent in all Type B pfaB KS domains examined in the current study. A recent study suggested that this pfaB KS domain may play a role in determining the final PUFA end product, as E. coli transformed with pfaA, pfaC, pfaD and pfaE from an EPA-producer and pfaB from a DHA-producer made both EPA and DHA . The Type B gene cluster is found exclusively in marine γ-proteobacteria representing three genera: Colwellia , Psychromonas , and Moritella . Although the Type B gene cluster retains a completely conserved domain structure in the genomes of the Colwellia, Moritella, and Psychromonas and the proximal pfaA KSs from these species group together at 74% amino acid sequence identity, Colwellia and Moritella are capable of producing significant quantities of DHA while the Psychromonas that have been investigated either do not produce DHA or produce it only in trace quantities , .
Type D is found in two Bacteriodetes strains (Table 1), of which Psychroflexus torquis ATCC 700755 is known to produce arachidonic acid (AA, 20:4n-6) and EPA . The Type D gene cluster has the same domain content as the Type A EPA-producing gene cluster, however the AT domain is rearranged and domains reside on split or fused gene products – the pfaA homolog is split into two genes, designated pfaA1 [KS-AT-ACP5] and pfaA2 [KR], and pfaB and pfaC are fused into one gene, pfaBC [KS-CLF-AT-DH2] (Figure 1).
The Type E gene cluster is found in the marine thraustochytrid Schizochytrium sp. ATCC 20888 (order Labyrinthulida) and is involved in the production of DHA and the omega-6 PUFA docosapentaenoic acid (DPA, 22:5n-6) . This gene cluster consists of three genes, designated PFA1, 2, and 3, and contains the same domains found in the five genes of the Type A cluster. Although the Type E gene cluster is clearly homologous to the Type A and B clusters, differences do exist. For example, in contrast to the Type A and B gene clusters, each containing one ER domain, the Type E cluster contains two, one located on PFA2 and the other on PFA3 (Figure 1). Sequence composition analyses indicate that these Schizochytrium ER domains share 89% identity at the amino acid level. Thus one of the ER domains may have resulted from a duplication and translocation event after transfer of an ancestral Pfa synthase gene cluster into the Schizochytrium genome. In addition to the Type E cluster for PUFA synthesis, Schizochytrium sp. ATCC 20888 also contains the canonical eukaryotic PUFA biosynthesis pathway, involving elongase and desaturase enzymes . The presence of a bacterial Pfa synthase in Schizochytrium is significant as it suggests evidence of lateral gene transfer between a bacterium and eukaryote. The integration and retention of a Pfa synthase in the Schizochytrium genome may contribute to the high PUFA content observed in certain labyrinthulids , . DHA- and DPA-producing thraustochytrids have been isolated from marine environments around the globe , although the relative contribution of each pathway to PUFA synthesis has not been determined in these strains.
Type F is found in the genomes of five nitrogen-fixing cyanobacteria. First characterized in the cyanobacterium Nostoc punctiforme strain ATCC 29133, the hgl genes, for heterocyst glycolipid , were shown to be involved in the production of the lipid moeity of heterocyst glycolipids. The other four cyanobacteria containing the Type F gene cluster, Nostoc azollae 0708, Anabaena variabilis ATCC 29413, Nodularia spumigena CCY 9414, and Nostoc sp. PCC 7120, also form heterocysts for nitrogen fixation. The Type F gene cluster is remarkably similar to Types A and B. One notable difference is the fusion of some of the domains contained on pfaB and pfaC on a single gene, hglC [KS-CLF-AT]. Additionally, the Type F cluster lacks the DH/I domains typically found on pfaC.
The final Type with a characterized pathway for which the chemical product has been verified is the Type P gene cluster, found in the genomes of two nitrogen-fixing Proteobacteria. This synthase is responsible for the production of the alkyl moiety of phenolic lipids in the cyst-forming Gammaproteobacterium Azotobacter vinelandii . The Type P gene cluster is also found in the Alphaproteobacterium Beijerinckia indica subsp. Indica ATCC 9039, which has not been found to produce cysts or phenolic lipids .
Secondary lipid synthase types with uncharacterized products.
In addition to the above pathways with characterized products, 14 additional gene clusters homologous to the pfa genes were discovered with putative fatty acyl end products (Table 2). The Type C cluster has been found in only one sequenced bacterial genome, that of the Deltaproteobacterium Desulfatibacillum alkenivorans AK-01, which has not been reported to produce PUFAs. This gene cluster contains a pfaBC fusion [KS-CLF-AT-DH2] and pfaD [ER] is located at the 3′ end of the gene cluster.
The Type G gene cluster, like the characterized Type F cluster, is found in cyanobacterial genomes. In fact, in the case of Anabaena variabilis ATCC 29413 and Nostoc sp. PCC 7120, the Type F and Type G gene clusters coexist in the same genome. Notably, the Type G gene cluster is always found to have two uncharacterized type I PKS genes directly upstream. The product of the Type G gene cluster may interact with the product of the linked type I PKS genes, producing a hybrid polyketide/fatty acid product. A precedent for the interaction of products from a PKS and a FAS/PKS exists, for example, in the case of phenolic lipid production in Azotobacter vinelandii (Type P). In this case, the phenolic functional group is produced by two type III PKS genes, which are flanked on both sides by the Type P FAS/PKS genes responsible for the production of the C22–26 fatty acid moiety of phenolic lipids. A similar head/tail functional group production relationship could be occurring between the Type G gene cluster and the upstream PKS genes.
Agrobacterium vitis S4 (class/subphylum Alphaproteobacteria) and Saccharopolyspora erythraea NRRL 2338 (phylum Actinobacteria) harbor the Type H gene cluster (Table 2). This gene cluster is defined by a split pfaA and a fused pfaBC homolog (Figure 1).
Type I contains a pfaBC fusion and is found in multiple genera of Deltaproteobacteria and one Planctomycete (Table 2). The Type J gene cluster is only found in one genome, that of Planctomyces limnophilus. It consists of just two genes, a pfaD homolog [ER] followed by a pfaA homolog [KS-MAT-ACP3-KR] (Figure 1). Phylogenetically, the Type J KS domain is closely related to the KS domain from the Type I gene cluster (Figure 2). This apparently partial gene cluster could have arisen after a partial transfer of a more complete Type I-like cluster. The Type K gene cluster also has a similar configuration to the Type I cluster (Figure 1) and is found in a single Deltaproteobacterium, Desulfococcus oleovorans Hxd3 (Table 2).
The Type L gene cluster is distinguished by a fused pfaBC [KS-CLF-DH2-AT] homolog, and a pfaD [ER] homolog at the 5′ end of the gene cluster. It has been suggested that this gene cluster plays a role in the production of polyunsaturated fatty acids in Streptomyces coelicolor A3(2)  however PUFA production has not been demonstrated in this strain.
The Type M cluster is conserved in all three finished genomes from strains representing the nitrogen-fixing genus Frankia (phylum Actinobacteria) (Table 2). This is especially significant given that these genomes range in size from 5.43 to 9.04 Mbp, indicating extensive genome expansion, contraction, and adaptation to specific niches .
The Type N cluster is found in the chitin-degrading species Chitinophaga pinensis DSM 2588, which forms spherical “resting bodies,” or microcysts, upon aging , as well as the clinical isolate Chryseobacterium gleum ATCC 35910. The composition of the C. pinensis DSM 2588 “resting body” has not been determined, and it is possible that the product of the Type N plays a structural role in these differentiated cells, a proposal consistent with the structural role of phenolic lipids (Type P) in differentiated cysts of A. vinelandii . However Chryseobacterium gleum ATCC 35910 is not known to produce differentiated cell types. The Type O gene cluster is found in the genomes of two marine Alphaproteobacteria, Dinoroseobacter shibae DFL 12 and Parvularcula bermudensis HTCC2503, and the rock-dwelling cyanobacterium Gloeobacter violaceus PCC 7421. The Type Q gene cluster is found in a single bacterial strain, Rhodococcus erythropolis PR4 (phylum Actinobacteria) (Table 2).
The Type R cluster is dispersed throughout several diverse microbial phyla, including the Acidobacteria (Candidatus Koribacter versatilis Ellin345 and Candidatus Solibacter usitatus Ellin6076), Elusimicrobia (Elusimicrobium minutum Pei191), and Chloroflexi (Roseiflexus castenholzii DSM 13941 and Roseiflexus sp. RS-1). The Type R cluster consists of one pfaA homolog containing the domain architecture KS-MAT-CLF-ACP(3–6)-KR followed by a downstream gene encoding phosphopantetheinyl tranferase (PPTase) activity. Although there is strong evidence that this gene cluster shares an evolutionary origin (Figure 2), it is possible that the fatty acyl products are modified to produce different end products in these physiologically diverse and ecologically distinct species. The Type R core gene cluster homologous to the pfa genes is often associated with other genes involved in lipid production. For example, in Korebacter versatilis Ellin345, the pfaA and pfaE homologs are separated by a gene encoding part of the acetyl-CoA carboxylase (ACC) complex, which catalyzes the carboxylation of acetyl-CoA to produce malonyl-CoA for fatty acid biosynthesis. This intervening gene is absent in the remaining four gene clusters comprising Type R.
Types S and T are each found in single representatives (Table 2).
Analysis of phosphopantetheinyl transferase domains
One characteristic distinguishing PKSs from lipid synthases is the molecular sequence of the associated PPTase products responsible for posttranslational modification and activation of ACP domains . To provide additional evidence that we have identified novel secondary lipid synthases rather than PKSs, we investigated PPTase domain diversity. The informative variable motifs are known as P0, P1a, and P1b in PUFA-specific PPTases and 1A, P1a′, and P1b′ in PKS/NRPS-specific PPTases. Secondary lipid synthase Types A, B, C, D, F, I, L, M, N, O, P, R, S, and T most often contain an associated PPTases in direct proximity to the Pfa synthase. Of these, Types A, B, C, D, F, I, and L contain the PUFA-associated P0 motif – defined as (L/V)Rx(L/V)LS in  and modified to hRxhLS in this comprehensive study of 69 secondary lipid-associated PPTases, where h = hydrophobic residue (Figure S3). Importantly, the PPTase from the Type P gene cluster, known to be involved in the production of phenolic lipids, does not contain the P0 motif. Consequently, the P0 domain may not be necessary for the production of secondary lipids in general. All secondary lipid Types investigated contain the PUFA-associated P1a motif – defined as K(G/D)KP in  and modified to x(G/D)xP in this study – rather than the PKS-related P1a′ motif. Exceptions were the Photobacterium PPTases which were previously shown to be EntD-like PPTases . EntD in E. coli is involved in the production of siderophore enterobactin  and is included in the PKS/NRPS-producing PPTase group. Examination of the final domain region revealed that Types A, B, C, D, F, I, and L contain the P1b domain – FNxSH in  and modified to (F/S)NxSH in this study – while Types M, N, O, P, R, T, and S contain the P1b′ domain – GSIxH in  and modified to hShxH in this study (Figure S3). This bifurcation between Types A–L and Types M–T is also reflected in the KS tree (Figure 2) and could have implications for the type of end product produced from these gene clusters.
Relationship of secondary lipid pathways with FAS II
The relationship between secondary lipid pathways and FAS II is of interest as both pathways draw from the same intracellular pool of precursor molecules for their biosynthetic activities. All bacteria possessing the archetypal Pfa synthase (Type A and B) also possess a complete FAS II and generally produce polyunsaturated fatty acids as a small percentage of their total fatty acids, with saturated and monounsaturated fatty acids comprising the majority of their phospholipids . However, several interesting exceptions exist among the bacteria harboring other putative secondary lipid synthases. The genome of the fish pathogen Renibacterium salmoninarum (Type M) lacks both fabA [DH/I] and fabZ [DH] homologs, at least one of which is necessary to perform the dehydration step involved in the biosynthesis of both saturated and monounsaturated fatty acids . Based on genome analysis and growth capabilities it has been suggested that this strain requires exogenous fatty acids, possibly scavenged from its host, for incorporation into membrane phospholipids . Consequently, the Type M pathway present in R. salmoninarum genome likely represents the only complete fatty acid biosynthesis pathway present in this strain. Likewise, Desulfobacterium autotrophicum HRM2 (Type S) is also missing fabA and fabZ homologs. Unlike the pathogenic R. salmoninarum, the completely oxidizing sulfate-reducing bacterium D. autotrophicum is free-living. Presumably, it could incorporate the fatty acids typically used as carbon sources into its membrane phospholipids. Given that D. autotrophicum harbors an incomplete FAS II, it is possible that the Type S secondary lipid synthase gene cluster is functioning as the “core” fatty acid synthase for this organism, thus the main producer of fatty acids for incorporation into phospholipids.
Co-occurrence of secondary lipid synthesis pathways with olefin-producing genes
Several major outstanding questions exist regarding the products synthesized by uncharacterized secondary lipid synthases described here (Types C, G, H, I J, K, L, M N, O, Q, R, S, T). It has previously been demonstrated that pfa gene products are necessary for the production of olefinic hydrocarbons in Shewanella onedensis MR-1 . pfa genes and several secondary lipid synthase types described here co-occur with the ole genes responsible for the head-to-head condensation of fatty acyl products and the formation of olefinic neutral lipids , in 44 of the 80 (55%) sequenced genomes examined (Pseudoalteromonas sp. DS-12, Shewanella livingstonensis, Shewanella sp. BR-2, Shewanella sp. SCRC-2738, Moritella marina, and Schizochytrium sp. ATCC 20888 were excluded from this analysis as their genomes have not been sequenced). Furthermore, Geobacter bemidjiensis produces the same olefinic hydrocarbon, hentriacontanonaene (C31H46), found in Shewanella strains harboring the pfa gene cluster. Therefore, we suggest that the Type I “pfa-like” secondary lipid synthase gene cluster in the Deltaprotoebacteria and Planctomycete strains may provide precursors necessary for olefin biosynthesis. Additionally, it is possible that the Type L cluster contributes to the production of olefinic hydrocarbons. Nakamurella multipartita DSM 44233 and Plesiocystis pacifica SIR-1 both contain ole genes .
These ole genes are often linked with the FAS/PKS gene clusters but can also reside elsewhere in the genome. In cases where unlinked pfa and ole genes exist, the transfer of these genes into the genomes was not coordinated, as bacteria possessing the same “Type” of pfa gene cluster often have differing ole gene configurations. An interesting question then arises regarding the primacy of these two biosynthetic gene clusters. One possibility is that the pfa homologs were originally producing a different product and were co-opted for the production of precursors for olefin biosynthesis over evolutionary time. It is also possible that the pfa homologs retain the ability to produce a stand-alone product and only contribute to olefin production under certain environmental conditions. Very little is known regarding the regulation and synthesis of olefin products.
Ecology of microorganisms possessing secondary lipid production potential
The microorganisms harboring the pfa genes and homologs belong to diverse phyla throughout the bacterial domain, exhibiting varied ecophysiologies and lifestyle strategies. Examination of this metadata provided important insight into the ecological basis of secondary lipid synthesis across the spectrum of biosynthetic diversity examined. A permutational analysis of variance (PERMOVA) was used to rigorously test the association between gene cluster Type and life history traits, for which there was adequate replication (n≥5, Types A, B, F, I, L, M, and R) (Figure S4). All gene clusters differed from each other in terms of life history traits (PERMANOVA pair-wise tests, P = ≤0.004), with the exception that L did not differ from M or R, and M did not differ from R (P = ≥0.05). Overall allocation success across all seven gene clusters equaled 60.6%, with all genes having a more distinct set of life history traits than expected by chance alone, with the exception of R. Type I was associated with being an obligate anaerobe. Many of the known fatty acid products produced by characterized Types are known to play a role in excluding oxygen from the cell (e.g. heterocyst glycolipids), or have been shown to provide antioxidative functions to the cell . Consistent with these functions, the product of the Type I pfa homologs in the Deltaproteobacteria may play a role in protecting these sensitive cells from lethal levels of oxygen in the environment. A complete PERMOVA analysis on all secondary lipid synthase types is presented in Figure S5.
Contribution of horizontal gene transfer to the dissemination of secondary lipid synthase pathways
Horizontal gene transfer (HGT) may help explain the patchy distribution of secondary lipid pathways among bacterial and eukaryotic phyla. Specifically, multiple secondary lipid Types may be found in the same phyla, while a single Type may be found across several phyla (Figure 3). To examine the possibility of HGT, all species of the same genus having an identified secondary lipid synthase gene cluster with two or more sequenced genomes were analyzed for evidence of HGT. For example, the pfa gene cluster (Type A) is present in three strains of Vibrio but absent in many closely related Vibrio genomes. A comparison of the pfa gene region of Vibrio splendidus 12B01 and corresponding regions of the genomes of multiple Vibrio species is shown in Figure 4. The two genomic regions immediately flanking the pfa gene cluster show 93% and 87% identity at the nucleotide level, respectively, to their homologs in the Vibrionales bacterium SWAT-3 genome. The pfa gene cluster clearly shows a perfect insertion characteristic of a genomic island.
Clades in red contain putative secondary lipid gene clusters. Tree modified from the Interactive Tree of Life.
Gray tracts indicate regions of homology between genomes. The pfa genes are labeled in the Vibrio splendidus 12B01 genome and are absent in the other Vibrio genomes.
The Type I gene cluster has an especially sporadic distribution among the Deltaproteobacteria. Of the two sequenced Pelobacter strains, Pelobacter propionicus DSM 2379 harbors a Type I cluster while Pelobacter carbinolicus DSM 2380 does not. Likewise, of the nine sequenced Geobacter, only six contain the Type I secondary lipid synthase. Furthermore, omega-3 PUFA production in marine Labyrinthulid protists (Eukaryota) of the genus Schizochytrium is known to proceed via the same FAS/PKS-related pathway providing additional evidence implicating HGT as an important contributor to chemical diversification , .
It has been noted that the pfaA homolog in Elusimicrobium minutum Pei191 (Type R) has an aberrant GC content (46%) compared to the rest of the genome (39%), prompting the suggestion that it entered the genome through horizontal gene transfer . To further investigate the potential novelty of these gene clusters within their host genomes, DNA compositional analyses were performed (%GC analysis and AlienHunter ). No conclusive evidence based on DNA compositional metrics was revealed, indicating these gene clusters are not aberrant within their respective genomes. Our analyses of all secondary lipid synthase gene clusters indicate that the GC content of the pfaA homolog is consistently higher than the genome average (data not shown), most likely a reflection of the amino acid composition of these proteins. While these results do not rule out possible ancient acquisition of these pathways, comparative genomic analysis, as presented above, provides better resolution in support of horizontal acquisition of these gene clusters. Improved genomic sampling within individual phylogenetic groups will enhance our ability to resolve the evolutionary history and mobility of secondary lipid production potential.
Secondary lipid synthase gene clusters have been detected in 10 microbial phyla, representing 86 species across two domains of life. The genetic potential to produce long-chain fatty acids via a FAS/PKS mechanism appears to be scattered throughout the bacterial domain and has been co-opted by some Eukarya. Results presented here demonstrate that these biosynthetic pathways are not relegated solely to a narrow group of marine bacteria, as previously believed. Instead, this third mechanism of bacterial fatty acid synthesis may be involved in the production of specialized lipid products across the bacterial tree. As additional genomes are sequenced and their physiologies explored, additional bacterial lineages harboring novel secondary lipid pathways and producing novel fatty acyl products will be revealed. The linking of these pathways to chemical products and determining the physiological role of these biosynthetic processes is a crucial next step.
Identification of FAS/PKS candidate sequences in sequenced microbial genomes
Genomes were downloaded from NCBI and the Joint Genome Institute's Integrated Microbial Genomes (IMG) database (http://img.jgi.doe.gov). A local TBLASTN of pfa genes known to produce EPA and DHA was performed against the downloaded genomes and results were curated by e-value (≤1e-30) to identify possible candidate sequences. To minimize false positive hits (e.g. polyketide synthase or nonribosomal peptide synthase genes), a refined set of query sequences was designed to exploit the conserved domain architecture of secondary lipid synthase gene products (e.g. PfaA–E). A subsequent TBLASTN of individual domains (KS, MAT, ACP, KR, DH, ER) was performed. These analyses provided a broad list of candidate genes. The most informative query sequences were tandem ACP sequences, a hallmark characteristic of secondary lipid products. All candidate genomes and specific gene neighborhoods were inspect for adjacent genes using the sequence visualization tool Artemis  and via IMG's Gene Ortholog Browser tool. Protein sequences were annotated using InterProScan  and manually curated. File S1 is presented as an Excel worksheet and provides a complete list of annotations and GenBank accession numbers for each protein sequence investigated in this study.
Proximal KS amino acid sequences were aligned using MAFFT  and viewed and edited in Jalview . Poorly aligned and/or divergent positions were excluded using Gblocks  with a minimum block of five and allowed gap positions equal to half. The best model for the resulting 364 amino acid alignment was evaluated using ProtTest , a program for selecting a model of protein evolution that uses PHYML  and PAL . ProtTest chose LG+G+F based on Akaike criterion. A maximum likelihood phylogenetic tree was constructed using RAxML rapid bootstrapping  through the CIPRES portal  cluster at the San Diego Supercomputer Center. The phylogenetic tree graphic was produced using the Interactive Tree of Life . The DH tree was constructed in an identical manner from the full alignment however using the LG+G model.
Multidimensional scaling analysis
Component domains from each gene cluster were letter coded (e.g. KS-MAT-ACP2-KR-ER-PPTase becomes ABCCDEF). Coded gene clusters were aligned using MAFFT and viewed and edited in Jalview. The alignment was used to generate a pairwise distance matrix in MOTHUR  and subsequently converted to a similarity matrix. This similarity matrix was imported into Primer 6  and used to create a multidimensional scaling (MDS) plot, using default parameters (25 random starts, Krustal fit scheme of 1, and a minimum stress value of 0.01). Subsequently, a cluster analysis of the data was performed, using average group linkage, and overlaid in order to define “Types” based on domain content and organization.
Permutational analysis of variance and canonical analysis of principal coordinates
A set of 33 life history traits were allocated to each bacterial species to create a binary matrix. A permutational analysis of variance (PERMANOVA) ,  was used to test the association between gene cluster type and bacterial life history traits. The PERMANOVA was based on unrestricted permutations of the raw data and a partial sums of squares. To visualize and test which life history traits were associated with which gene clusters we used a constrained canonical analysis of principal coordinates (CAP) , . CAP analyses were based on 10,000 random permutations of the raw data and a Bray-Curtis similarity matrix. Individual life history traits that might be responsible for any differences in multivariate space were investigated by calculating Spearman Rank correlations of canonical ordination axes with the original genera variables. Traits with strong correlations (defined as ≥0.4 in this study) were then overlaid as a bi-plot. We used the leave-one-out procedure in the CAP analysis to calculate allocation success for each a priori defined gene cluster. This essentially gave us a measure of distinctness for the life history traits associated with each gene cluster. Allocation success was considered indicative of a more distinct set of life history traits than expected by chance alone when values exceeded 14.3%. This threshold came from the possibility of each individual observation having a 14.3% chance of being placed into one of the 7 a priori defined groups (cluster Types). As replication within each a priori defined gene cluster group varied (ranging from n = 1 to n = 30), formal tests were only run on those gene clusters for which n≥5 (A, B, F, I, L, M, R).
Multidimensional scaling (MDS) plot of similarity matrix of domain order and count in pfa gene clusters and homologues. See File S1 for numerical key describing species abbreviations.
Maximum likelihood phylogenetic tree of dehydratase/isomerase protein domains (158 amino acid alignment). Gene cluster Types are colored and given a letter label. Asterisks represent Types first clustered and described in this study. Bootstrap values ≥50% are indicated by dots. The E. coli DH10B FabA protein was used as the outgroup.
Multiple sequence alignment showing sections of PPTase domains from representative organisms. Conserved domains are labeled at the top of the alignment. Motifs P2 and P3 do not show a clear pattern of variation among different secondary lipid types.
CAP ordination of the similarity among seven gene clusters (A, B, F, I, L, M, R) based on 33 life history traits. Group centroids are displayed for each gene cluster to ease interpretation. Vector lines in the bi-plot represent Spearman Rank correlations, with the direction indicating the relationship of each trait to the gene clusters in multivariate space. The length of each vector line is proportional to the strength of the correlation, with the blue circle showing the threshold for a correlation of one.
CAP ordination of the similarity among all gene clusters (A–T) based on 33 life history traits. Group centroids are displayed for each gene cluster to ease interpretation. Vector lines in the bi-plot represent Spearman Rank correlations, with the direction indicating the relationship of each trait to the gene clusters in multivariate space. The length of each vector line is proportional to the strength of the correlation, with the blue circle showing the threshold for a correlation of one.
The authors would like to sincerely thank Gareth Williams for statistical assistance with the PERMANOVA and CAP analyses. Additionally, we thank Sheila Podell for insightful discussion and providing comments on the manuscript.
Conceived and designed the experiments: CNS EEA. Performed the experiments: CNS EEA. Analyzed the data: CNS EEA. Wrote the paper: CNS EEA.
- 1. Magnuson K, Jackowski S, Rock CO, Cronan JE (1993) Regulation of fatty acid biosynthesis in Escherichia coli. Microbiological Reviews 57: 522–542.
- 2. Schweizer E, Hofmann J (2004) Microbial type I fatty acid synthases (FAS): Major players in a network of cellular FAS systems. Microbiol Mol Biol Rev 68: 501–517.
- 3. Kolattukudy PE, Fernandes ND, Azad AK, Fitzmaurice AM, Sirakova TD (1997) Biochemistry and molecular genetics of cell-wall lipid biosynthesis in mycobacteria. Mol Microbiol 24: 263–270.
- 4. Nichols DS (2003) Prokaryotes and the input of polyunsaturated fatty acids to the marine food web. FEMS Microbiol Lett 219: 1–7.
- 5. Nichols DS, McMeekin TA (2002) Biomarker techniques to screen for bacteria that produce polyunsaturated fatty acids. J Microbiol Methods 48: 161–170.
- 6. Metz JG, Roessler P, Facciotti D, Levering C, Dittrich F, et al. (2001) Production of polyunsaturated fatty acids by polyketide synthases in both prokaryotes and eukaryotes. Science 293: 290–293.
- 7. Yazawa K (1996) Production of eicosapentaenoic acid from marine bacteria. Lipids 31: S297–S300.
- 8. Tanaka M, Ueno A, Kawasaki K, Yumoto I, Ohgiya S, et al. (1999) Isolation of clustered genes that are notably homologous to the eicosapentaenoic acid biosynthesis gene cluster from the docosahexaenoic acid-producing bacterium Vibrio marinus strain MP-1. Biotechnol Lett 21: 939–945.
- 9. Allen EE, Bartlett DH (2002) Structure and regulation of the omega-3 polyunsaturated fatty acid synthase genes from the deep-sea bacterium Photobacterium profundum strain SS9. Microbiology 148: 1903–1913.
- 10. Campbell EL, Cohen MF, Meeks JC (1997) A polyketide-synthase-like gene is involved in the synthesis of heterocyst glycolipids in Nostoc punctiforme strain ATCC 29133. Arch Microbiol 167: 251–258.
- 11. Bauersachs T, Compaore J, Hopmans EC, Stal LJ, Schouten S, et al. (2009) Distribution of heterocyst glycolipids in cyanobacteria. Phytochemistry 70: 2034–2039.
- 12. Miyanaga A, Funa N, Awakawa T, Horinouchi S (2008) Direct transfer of starter substrates from type I fatty acid synthase to type III polyketide synthases in phenolic lipid synthesis. Proc Natl Acad Sci USA 105: 871–876.
- 13. Tang Y, Tsai SC, Khosla C (2003) Polyketide chain length control by chain length factor. J Am Chem Soc 125: 12708–12709.
- 14. Shulse CN, Allen EE (2011) Diversity and distribution of microbial long-chain fatty acid biosynthetic genes in the marine environment. Environ Microbiol 13: 684–695.
- 15. Jiang H, Zirkle R, Metz JG, Braun L, Richter L, et al. (2008) The role of tandem acyl carrier protein domains in polyunsaturated fatty acid biosynthesis. J Am Chem Soc 130: 6336–6337.
- 16. Fischbach MA, Walsh CT (2006) Assembly-line enzymology for polyketide and nonribosomal peptide antibiotics: Logic, machinery, and mechanisms. Chemical Reviews 106: 3468–3496.
- 17. Okuyama H, Orikasa Y, Nishida T, Watanabe K, Morita N (2007) Bacterial genes responsible for the biosynthesis of eicosapentaenoic and docosahexaenoic acids and their heterologous expression. Appl Environ Microbiol 73: 665–670.
- 18. Wang F, Xiao X, Ou HY, Gai YB, Wang FP (2009) Role and regulation of fatty acid biosynthesis in the response of Shewanella piezotolerans WP3 to different temperatures and pressures. J Bacteriol 191: 2574–2584.
- 19. Venkateswaran K, Moser DP, Dollhopf ME, Lies DP, Saffarini DA, et al. (1999) Polyphasic taxonomy of the genus Shewanella and description of Shewanella oneidensis sp. nov. Int J Syst Bacteriol 49: 705–724.
- 20. Moser DP, Nealson KH (1996) Growth of the facultative anaerobe Shewanella putrefaciens by elemental sulfur reduction. Appl Environ Microbiol 62: 2100–2105.
- 21. Perry KA, Kostka JE, Luther GW, Nealson KH (1993) Mediation of sulfur speciation by a Black Sea facultative anaerobe. Science 259: 801–803.
- 22. Orikasa Y, Tanaka M, Sugihara S, Hori R, Nishida T, et al. (2009) pfaB products determine the molecular species produced in bacterial polyunsaturated fatty acid biosynthesis. FEMS Microbiol Lett 295: 170–176.
- 23. Methe BA, Nelson KE, Deming JW, Momen B, Melamud E, et al. (2005) The psychrophilic lifestyle as revealed by the genome sequence of Colwellia psychrerythraea 34H through genomic and proteomic analyses. Proc Natl Acad Sci USA 102: 10913–10918.
- 24. Riley M, Staley JT, Danchin A, Wang TZ, Brettin TS, et al. (2008) Genomics of an extreme psychrophile, Psychromonas ingrahamii. BMC Genomics 9: 210.
- 25. Delong EF, Yayanos AA (1986) Biochemical function and ecological significance of novel bacterial lipids in deep-sea procaryotes. Appl Environ Microbiol 51: 730–737.
- 26. Auman AJ, Breezee JL, Gosink JJ, Kampfer P, Staley JT (2006) Psychromonas ingrahamii sp nov., a novel gas vacuolate, psychrophilic bacterium isolated from Arctic polar sea ice. Int J Syst Evol Microbiol 56: 1001–1007.
- 27. Bowman JP, McCammon SA, Lewis T, Skerratt JH, Brown JL, et al. (1998) Psychroflexus torquis gen. nov., sp. nov., a psychrophilic species from Antarctic sea ice, and reclassification of Flavobacterium gondwanense (Dobson et al. 1993) as Psychroflexus gondwanense gen. nov., comb. nov. Microbiology 144: 1601–1609.
- 28. Lippmeier JC, Crawford KS, Owen CB, Rivas AA, Metz JG, et al. (2009) Characterization of both polyunsaturated fatty acid biosynthetic pathways in Schizochytrium sp. Lipids 44: 621–630.
- 29. Singh A, Ward OP (1996) Production of high yields of docosahexaenoic acid by Thraustochytrium roseum ATCC 28210. J Ind Microbiol 16: 370–373.
- 30. Yokochi T, Honda D, Higashihara T, Nakahara T (1998) Optimization of docosahexaenoic acid production by Schizochytrium limacinum SR21. Appl Microbiol Biotechnol 49: 72–76.
- 31. Raghukumar S (2002) Ecology of the marine protists, the Labyrinthulomycetes (Thraustochytrids and Labyrinthulids). Eur J Protistol 38: 127–145.
- 32. Starkey RL, De PK (1939) A new species of Azotobacter. Soil Science 47: 329–343.
- 33. Bentley SD, Chater KF, Cerdeno-Tarraga AM, Challis GL, Thomson NR, et al. (2002) Complete genome sequence of the model actinomycete Streptomyces coelicolor A3(2). Nature 417: 141–147.
- 34. Normand P, Lapierre P, Tisa LS, Gogarten JP, Alloisio N, et al. (2007) Genome characteristics of facultatively symbiotic Frankia sp strains reflect host range and host plant biogeography. Genome Res 17: 7–15.
- 35. Sangkhobol V, Skerman VBD (1981) Chitinophaga, a new genus of chitinolytic myxobacteria. Int J Syst Bacteriol 31: 285–293.
- 36. Funa N, Ozawa H, Hirata A, Horinouchi S (2006) Phenolic lipid synthesis by type III polyketide synthases is essential for cyst formation in Azotobacter vinelandii. Proc Natl Acad Sci USA 103: 6356–6361.
- 37. Sugihara S, Orikasa Y, Okuyama H (2008) An EntD-like phosphopantetheinyl transferase gene from Photobacterium profundum SS9 complements pfa genes of Moritella marina strain MP-1 involved in biosynthesis of docosahexaenoic acid. Biotechnol Lett 30: 411–414.
- 38. Hantash FM, Ammerlaan M, Earhart CF (1997) Enterobactin synthase polypeptides of Escherichia coli are present in an osmotic-shock-sensitive cytoplasmic locality. Microbiology-Uk 143: 147–156.
- 39. White SW, Zheng J, Zhang YM, Rock CO (2005) The structural biology of type II fatty acid biosynthesis. Annu Rev Biochem 74: 791–831.
- 40. Wiens GD, Rockey DD, Wu ZN, Chang J, Levy R, et al. (2008) Genome sequence of the fish pathogen Renibacterium salmoninarum suggests reductive evolution away from an environmental Arthrobacter ancestor. J Bacteriol 190: 6970–6982.
- 41. Sukovich DJ, Seffernick JL, Richman JE, Hunt KA, Gralnick JA, et al. (2010) Structure, function, and insights into the biosynthesis of a head-to-head hydrocarbon in Shewanella oneidensis strain MR-1. Appl Environ Microbiol 76: 3842–3849.
- 42. Sukovich DJ, Seffernick JL, Richman JE, Gralnick JA, Wackett LP (2010) Widespread head-to-head hydrocarbon biosynthesis in bacteria and role of OleA. Appl Environ Microbiol 76: 3850–3862.
- 43. Nishida T, Orikasa Y, Ito Y, Yu R, Yamada A, et al. (2006) Escherichia coli engineered to produce eicosapentaenoic acid becomes resistant against oxidative damages. FEBS Lett 580: 2731–2735.
- 44. Herlemann DPR, Geissinger O, Ikeda-Ohtsubo W, Kunin V, Sun H, et al. (2009) Genomic analysis of “Elusimicrobium minutum,” the first cultivated representative of the phylum “Elusimicrobia” (formerly Termite Group 1). Appl Environ Microbiol 75: 2841–2849.
- 45. Vernikos GS, Parkhill J (2006) Interpolated variable order motifs for identification of horizontally acquired DNA: revisiting the Salmonella pathogenicity islands. Bioinformatics 22: 2196–2203.
- 46. Rutherford K, Parkhill J, Crook J, Horsnell T, Rice P, et al. (2000) Artemis: sequence visualization and annotation. Bioinformatics 16: 944–945.
- 47. Zdobnov EM, Apweiler R (2001) InterProScan - an integration platform for the signature-recognition methods in InterPro. Bioinformatics 17: 847–848.
- 48. Katoh K, Toh H (2008) Recent developments in the MAFFT multiple sequence alignment program. Briefings in Bioinformatics 9: 286–298.
- 49. Clamp M, Cuff J, Searle SM, Barton GJ (2004) The Jalview Java alignment editor. Bioinformatics 20: 426–427.
- 50. Castresana J (2000) Selection of conserved blocks from multiple alignments for their use in phylogenetic analysis. Mol Biol Evol 17: 540–552.
- 51. Abascal F, Zardoya R, Posada D (2005) ProtTest: selection of best-fit models of protein evolution. Bioinformatics 21: 2104–2105.
- 52. Guindon S, Gascuel O (2003) A simple, fast, and accurate algorithm to estimate large phylogenies by maximum likelihood. Syst Biol 52: 696–704.
- 53. Drummond A, Strimmer K (2001) PAL: an object-oriented programming library for molecular evolution and phylogenetics. Bioinformatics 17: 662–663.
- 54. Stamatakis A, Hoover P, Rougemont J (2008) A rapid bootstrap algorithm for the RAxML web servers. Syst Biol 57: 758–771.
- 55. Miller MA, Holder MT, Vos R, Midford PE, Liebowitz T, et al. (2009) The CIPRES Portals. CIPRES.
- 56. Letunic I, Bork P (2007) Interactive Tree Of Life (iTOL): an online tool for phylogenetic tree display and annotation. Bioinformatics 23: 127–128.
- 57. Schloss PD, Westcott SL, Ryabin T, Hall JR, Hartmann M, et al. (2009) Introducing mothur: open-source, platform-independent, community-supported software for describing and comparing microbial communities. Appl Environ Microbiol 75: 7537–7541.
- 58. Clarke K, Warwick R (2001) Change in marine communities: an approach to statistical analysis and interpretation. Plymouth: PRIMER-E.
- 59. Anderson MJ (2001) A new method for non-parametric multivariate analysis of variance. Austral Ecol 26: 32–46.
- 60. McArdle BH, Anderson MJ (2001) Fitting multivariate models to community data: A comment on distance-based redundancy analysis. Ecology 82: 290–297.
- 61. Anderson MJ, Willis TJ (2003) Canonical analysis of principal coordinates: A useful method of constrained ordination for ecology. Ecology 84: 511–525.
- 62. Anderson MJ, Robinson J (2003) Generalized discriminant analysis based on distances. Australian & New Zealand Journal of Statistics 45: 301–318.