Our findings show that START domain-containing proteins are amplified in plant genomes (Arabidopsis and rice) relative to animal genomes (Figures 1,2). Arabidopsis and rice contain 35 and 29 START proteins each, whereas the human and mouse genomes contain 15 each 11, and C. elegans and D. melanogaster encode seven and four, respectively. In comparison, bacterial and protist genomes appear to encode a maximum of two START proteins (see below).

START-domain minimal proteins comprising the START domain only, as well as START proteins containing additional sequence of unknown or known function appear to be conserved across plants, animals, bacteria and protists (Tables 1,2,3,4, Figure 1). However, only in plants, animals and multicellular protists (D. discoideum) are START domains found in association with domains having established functions in signal transduction or transcriptional control, consistent with the idea that START evolved as a regulatory domain in multicellular eukaryotes. The cellular slime mold D. discoideum, which progresses from unicellular to multicellular developmental stages, contains an unusual START-domain protein 8 which has so far not been found in any other organism: FbxA/CheaterA (ChtA), an F-Box/WD40 repeat-containing protein 1213. FbxA/ChtA is thought to encode a component of an SCF E3 ubiquitin ligase implicated in cyclic AMP metabolism and histidine kinase signaling during development 14. Mutant analysis shows that FbxA/ChtA function is required to generate the multicellular differentiated stalk fate 12.

Functional domains that were found associated with START in animals include pleckstrin homology (PH), sterile alpha motif (SAM), Rho-type GTPase-activating protein (RhoGAP), and 4-hydroxybenzoate thioesterase (4HBT) (Table 3), consistent with a previous report 11. The RhoGAP-START configuration is absent from plants, but is conserved across the animal kingdom from mammals to insects and nematodes. The RhoGAP-START combination in addition to an amino-terminal SAM domain is apparent only in proteins from humans, mouse, and rat, indicating that SAM-RhoGAP-START proteins are specific to mammals. Similarly, the 4HBT-START combination, also referred to as the acyl-CoA thioesterase subfamily 11, is found exclusively in proteins from humans, mouse and rat, and therefore seems to have evolved in the mammalian lineage.

In humans, about half of the START domain-containing proteins (6/15) are multidomain proteins, whereas in Arabidopsis and rice approximately three-quarters (26/35; 22/29) of START proteins contain an additional domain. The largest proportion of Arabidopsis and rice multidomain START proteins (21/26; 17/22) contain a homeodomain (HD), while a smaller group of proteins (4/26; 4/22) contain a PH domain together with a recently identified domain of unknown function 1336 (DUF1336) motif. In addition, a single START-DUF1336 protein of about the same size, but lacking strong sequence similarity to PH at its amino terminus, is present in both Arabidopsis and rice. It is striking that the sequence of the START domain correlates with the type of START protein, an indication that evolutionary speciation through duplication and subsequent sequence evolution of START domains took place after initial manifestation of novel protein architecture by domain shuffling.

The position of the START domain in proteins larger than 300 amino acids varies between plant, animal and protist kingdoms. For example, in human proteins, START is always near the carboxy terminus (1-55 amino acids from the end) (Table 3). In plant proteins, however, the START domain is not strictly confined to the carboxy terminus. In both Arabidopsis and rice the START domain can be positioned as much as approximately 470 amino acids from the carboxy terminus (HD-ZIP START proteins: Tables 1,2). Moreover, in a subset of plant proteins (PH-START DUF1336 proteins), the START domain is positioned centrally between two different domains. However, defined functional domains are typically amino terminal of the START domain in both animals and plants. By contrast, in the sole example of a START-domain protein in D. discoideum, FbxA/ChtA, the START domain is present at the amino terminus, with F-Box and WD40 domains positioned after it.

The START-domain proteins from Arabidopsis were classified into seven subfamilies according to their structures and sizes (Figure 2a). The majority of START domains are found in transcription factors of the HD family. HDs are DNA-binding motifs involved in the transcriptional regulation of key developmental processes in eukaryotes. However, only within the plant kingdom do HD transcription factors also contain START domains (Figure 1). Among around 90 HD family members in Arabidopsis 15, approximately one-quarter (21) contain a START domain. All HD-START proteins contain a putative leucine zipper, a dimerization motif that is not found in HD proteins from animals or yeast. Nuclear localization has been demonstrated for two HD-START proteins: GLABRA2 (GL2) 16 and REVOLUTA/INTERFASCICULAR FIBERLESS1 (REV/IFL1) 17. Furthermore, canonical DNA-binding sites are reported for GL2 18 and two other HD-START transcription factors, A. thaliana MERISTEM LAYER1 (ATML1) 19, and PROTODERMAL FACTOR2 (PDF2) 20.

A similar spectrum of START domain-containing proteins is found in Arabidopsis and rice, suggesting their origin in a common ancestor (Figure 2b). The size of the rice genome (430 Mb) is roughly four times that of Arabidopsis (120 Mb). Despite a twofold difference between the total number of predicted genes in rice (ssp. indica: 53,398 21) versus Arabidopsis (~28,000), the number of START domains per genome appears to be relatively constant: Arabidopsis and rice contain 35 and 29 START genes, respectively. Thus, START-domain genes belong to the subset of Arabidopsis genes (estimated at two-thirds) that are present in rice 21. However, one intriguing exception is the apparent absence of rice proteins orthologous to two unusual Arabidopsis START proteins (At4g26920 and At5g07260), which share sequence similarity to each other and to members of the HD-ZLZ START subfamily, but lack HD and zipper-loop-zipper (ZLZ) domains (Figure 2b; Tables 1,2). Their absence from rice makes them candidate dicot-specific START proteins.

Screening for expressed sequence tags (ESTs) by BLASTN was conducted to determine whether the types of START sequences from Arabidopsis and rice are also present in other plants (see Materials and methods). The screen detected 185 START domain-encoding sequences from a wide assortment of plants representing 25 different species. Consistent with our findings in Arabidopsis and rice (Tables 1,2), START domains were found in the plant-specific combinations (HD-START and PH-START) in both dicot and monocot members of the angiosperm division. ESTs for HD-START transcription factors were also identified from the gymnosperm Picea abies (AF328842 and AF172931), as well as from a representative of the most primitive extant seed plant, the cycad Cycas rumphii (CB093462). Furthermore, a HD-START sequence is expressed in the moss Physcomitrella patens (AB032182). Thus it appears that the HD-START plant-specific configuration evolved in the earliest plant ancestor, or alternatively has been retained in the complete plant lineage.

Sequence alignments and phylogenetic analysis revealed two distinct classes of HD-START proteins, which differ substantially in their leucine zippers and START domains (Figures 1,2,3). Both types of leucine zipper are unrelated in sequence to the homeobox-associated leucine zipper (HalZ), which is a plant-specific leucine zipper found in other HD proteins lacking START 22.

Most HD-START proteins (16/21 in Arabidopsis; 12/17 in rice) contain a leucine zipper with an internal loop (defined here as zipper-loop-zipper, ZLZ; also termed 'truncated leucine zipper motif' 23) immediately following a conserved HD domain (Figure 3a). The ZLZ motif appears to be less conserved than the classic basic region leucine zipper and seems to be plant specific. It was shown to be functionally equivalent to the HalZ leucine zipper domain for dimerization in an in vitro DNA binding assay 24.

The other HD-START proteins (5/21 in Arabidopsis; 5/17 in rice) contain a classic leucine zipper DNA-binding motif fused to the end of the HD, designated here as ZIP (Figure 3b). This leucine zipper shows strong sequence similarity to the basic region leucine zipper domains (bZIP and BRLZ) 2526, which have overlapping consensus sequences and are found in all eukaryotic organisms.

Despite these differences, it is likely that both types of HD-START transcription factors originated from a common ancestral gene. They share a common structural organization in their amino-terminal HD, leucine zipper (ZLZ or ZIP) and START domains (Figure 2a). Moreover, the carboxy terminus of HD-ZLZ START proteins (approximately 250 amino acids) shares sequence similarity with the first 250 amino acids of the approximately 470 amino acids at the carboxy terminus of HD-ZIP START proteins. This is exemplified by a comparison between the carboxy-terminal sequences of ATML1 (HD-ZLZ START) and REV (HD-ZIP START), which are 20% identical and 39% similar.

Several HD-ZLZ START genes correspond to striking mutant phenotypes in Arabidopsis, and for numerous HD-ZLZ START genes, functions in the development of the epidermis have been implicated. Proteins of the HD-ZLZ START subfamily share strong sequence similarity to each other along their entire lengths, including the carboxy-terminal sequence (approximately 250 amino acids) of unknown function that follows the START domain. The HD-ZLZ transcription factors ATML1 and PDF2 appear to be functionally redundant: double-mutant analysis shows that the corresponding genes are required for epidermal differentiation during embryogenesis 20. The rice HD-ZLZ START protein RICE OUTERMOST CELL-SPECIFIC GENE1 (ROC1) seems to have an analogous function to ATML1 in that its expression is restricted to the outermost epidermal layer from the earliest stages in embryogenesis 27. Another HD-ZLZ gene from rice, Oryza sativa TRANSCRIPTION FACTOR 1 (OSTF1), appears to be developmentally regulated during early embryogenesis and is also expressed preferentially in the epidermis 23. Mutations in Arabidopsis ANTHOCYANINLESS2 (ANL2) affect anthocyanin accumulation and the cellular organization in the root, indicating a role in subepidermal cell identity 28. The GL2 gene is expressed in specialized epidermal cells and mutant analysis reveals its function in trichome and non-root hair cell fate determination 2429. GL2 functions as a negative regulator of the phosopholipid signaling in the root 18, raising the possibility that the activity of GL2 itself is regulated through a feedback mechanism of phospholipid signaling through its START domain.

The HD-ZIP START genes characterized thus far are implicated in differentiation of the vasculature. Members of this subfamily are typically large proteins (more than 830 amino acids) that display strong sequence similarity to each other along their entire lengths, including the carboxy-terminal 470 or so residues of unknown function that follow the START domain. Mutations affecting PHABULOSA (PHB) and PHAVOLUTA (PHV), which have redundant functions, abolish radial patterning from the vasculature in the developing shoot, and perturb adaxial/abaxial (upper/lower) axis formation in the leaf 30. Mutant analysis reveals that REV 3132, isolated independently as IFL1 1733, is also involved in vascular differentiation. Although a mutant phenotype for A. thaliana HOMEOBOX-8 (ATHB-8) is not reported, its expression is restricted to provascular cells 34 and promotes differentiation in vascular meristems 35.

The presence of the START domain in HD transcription factors suggests the possibility of lipid/sterol regulation of gene transcription for HD-START proteins, as previously hypothesized 2. One advantage of such a mechanism is that the metabolic state of the cell in terms of lipid/sterol synthesis could be linked to developmental events such as regulation of transcription during differentiation. Changes in the activity of a HD-START transcription factor could be controlled via a lipid/sterol-binding induced conformational change. For instance, a protein-lipid/sterol interaction involving the START domain may regulate the activity of the transcription factor directly by affecting its DNA-binding affinity or interaction with accessory proteins at the promoter. Alternatively, or in addition, protein-lipid/sterol binding may positively or negatively affect transport or sequestration of the transcription factor to the nucleus.

A subset of animal and plant START proteins contain an amino-terminal PH domain, which is found in a wide variety of eukaryotic proteins implicated in signaling. PH domains are characterized by their ability to bind phosphoinositides, thereby influencing membrane and/or protein interactions 36. In some cases, phosphoinositide interactions alone may not be sufficient for membrane association, but may require cooperation with other cis-acting anchoring motifs, such as the START domain, to drive membrane attachment.

Although both plant and animal genomes encode START domains in association with an amino-terminal PH domain, the sequences of the PH-START proteins are not conserved between kingdoms (Figure 1; data not shown). In plants, the START domain is adjacent to the PH domain, whereas in animals the PH and START domains are separated by two serine-rich domains 11. The PH-START protein from humans, GPBP, has serine/threonine kinase activity and Goodpasture (GP) antigen binding affinity, two functions that involve the serine-rich domains. In contrast, the plant PH-START proteins contain a plant-specific carboxy-terminal domain (of around 230 amino acids) of unknown function, DUF1336 (Protein families database (Pfam)) 37. In addition, amino-terminal sequence analysis (TargetP; see Materials and methods) predicts that three PH-START proteins from Arabidopsis (At3g54800, At4g19040 and At5g45560) and two PH-START proteins from rice (BAD07818 and BAC22213) localize to mitochondria. This suggests a common lipid/sterol-regulated function of these proteins that is related to their subcellular localization.

Transmembrane segments may act to tether START-domain proteins to intracellular membranes. One START protein, MLN64 from humans, is anchored to late endosomes via four putative transmembrane segments that are required for its localization 3839. The carboxy-terminal START domain of MLN64 faces the outside of this compartment, consistent with a role in cholesterol trafficking to a cytoplasmic acceptor 38. Using sequence scanning and alignments with MLN64, putative transmembrane segments in the MLN64-related proteins from D. melanogaster (CG3522/Start1) 40 and C. elegans (3F991) are predicted (Table 3). No other START proteins from animals are predicted to contain transmembrane segments. A single putative amino-terminal transmembrane segment is predicted for one bacterial START protein (Xanthomonas axonopodis XAC0537), while one START protein from the unicellular protist Cryptosporidium parvum (1Mx.08) contains seven putative transmembrane segments amino-terminal to the START domain (Table 4). Transmembrane segments are also predicted for several Arabidopsis and rice START proteins (Tables 1,2). Among Arabidopsis START-domain proteins, four (At1g55960, At1g64720, At3g13062, and At3g23080) are predicted to encode a single transmembrane segment at the amino-terminal end of the protein. Two START proteins of very similar sequence (At4g14500 and At5g54170) contain two putative transmembrane segments, one at the amino terminus of the protein and the second very close to the carboxy terminus. This feature is also found in one rice protein (ID215312), and appears to be plant specific. Possible functions of these putative transmembrane proteins include lipid/sterol transport to and from membrane-bound organelles and/or maintenance of membrane integrity.

Cholesterol, phosphatidylcholine, carotenoid and ceramide are examples of lipids that are known to bind START domains from animals 3456. We explored the potential for predicting lipid/sterol ligands for plant-specific START domains by aligning plant START domains with related animal START domains having defined ligands. However, the mammalian MLN64 and StAR START domains, which have been shown to bind cholesterol, do not share convincing sequence conservation with any particular subfamily of Arabidopsis or rice START domains. Similarly, the START domains of CBP1, which binds the carotenoid lutein, and CERT, which binds ceramide, both show insufficient amino-acid conservation when compared in alignment to the most closely related Arabidopsis proteins. Knowledge of the precise ligand contacts within the cavity of these START domains is required to make more accurate predictions about their specific binding properties.

Although most START domains appear to have diverged in the evolution of plants and animals, one notable exception is the preservation of PCTP-like sequences (Figures 1,4a). The major function of mammalian PCTP, a START-domain minimal protein, is to replenish the plasma membrane with phosphatidylcholine and to allow its efflux. Mammalian PCTP binds phosphatidylcholine and its crystal structure in the ligand-bound form has recently been solved 4. Phosphatidylcholine is the most abundant phospholipid in animals and plants. It is a key building block of membrane bilayers, comprising a high proportion of the outer leaflet of the plasma membrane.

To identify plant START domains that are most similar to mammalian PCTP, we constructed a phylogeny with PCTP against the complete set of 64 START domains from Arabidopsis and rice. Figure 4a shows the branch from this phylogeny grouping mammalian PCTP with START domains from 10 plant proteins. Unlike PCTP, a protein of 214 amino acids, the plant PCTP-like proteins are larger (around 400 amino acids) and contain divergent amino-terminal and carboxy-terminal sequences in addition to the START domain. Most (8/10) of these proteins are predicted to encode an amino-terminal transmembrane segment while a subset (3/8) also encode a putative carboxy-terminal transmembrane segment (Tables 1,2). A sequence alignment of PCTP against the START domains from these 10 PCTP-like proteins is illustrated in Figure 4b. Among 27 amino-acid residues in PCTP that contact the phosphatidylcholine ligand 4 (Figure 4b), 63% (17/27) are similar to residues in one or more of the plant proteins.

Homology modeling was used to generate three-dimensional structures of the 10 PCTP-like START domains from Arabidopsis and rice, incorporating templates from PCTP and two other mammalian START domain structures (MLN64 3 and StarD4 7). The models proved most accurate for the At3g13062 and ID215312 START domains from Arabidopsis and rice, respectively. Figure 4c-f depicts the model of the 212-amino-acid START domain from At3g13062. Overall, the PCTP and At3g13062 START domain sequences are 24% identical and 45% similar.

The backbone of the At3g13062 START domain overlaps with the crystal structure of PCTP START 4. The model exhibits α helices at the amino and carboxy termini separated by nine β strands and two shorter α helices (Figure 4c). The curved β sheets form a deep pocket with the carboxy-terminal α helix (α4) acting as a lid, resulting in an internal hydrophobic cavity (Figures 4e-g). The amino-terminal α helix (α1), which is in a region of the START domain that is least conserved, is peripheral to the hydrophobic cavity, and is not predicted to contact the ligand. Lipid entry or egress would require a major conformational change, most likely opening or unfolding of the carboxy-terminal α helix lid. In the case of PCTP, the α4 helix has been implicated in membrane binding and phosphatidylcholine extraction 41. The overall conformation is distinct from other hydrophobic cavity lipid-binding proteins, such as the intracellular sterol carrier protein 2 (SCP2) 42 from animals, and the orthologous nonspecific lipid-transfer protein from plants 43. It is not clear how specific START is for its particular ligand, as the START domain of human StAR has been shown in vitro to have affinity for both cholesterol and sitosterol 44, the latter being the major plant sterol. However, modeling of highly similar START domains based on the crystal structure of PCTP implicates a similar molecular ligand, and suggests that these proteins have retained a common function in evolution.

Of about 100 bacterial genomes sequenced, only eight species encode proteins with similarity to START, whereas one-half (4/8) of unicellular protist genomes thus far sequenced appear to contain a START domain (Figure 1, Table 4). Expression data regarding the putative START-encoding genes from bacteria and unicellular protists is lacking to date. Most of the bacterial sequences represent START-domain minimal proteins of less than 245 amino acids. A START gene from the opportunistic pathogen Pseudomonas aeruginosa (PA1579) was previously hypothesized to have arisen by horizontal gene transfer 78. Here we show that similar START-domain genes also occur in one other pathogenic Pseudomonas species as well as in two related nonpathogens. The citrus-tree pathogen X. axonopodis and human pathogen Vibrio vulnificus also contain START genes. Nonpathogenic species that contain START genes include the anaerobic green sulfur bacterium Chlorobium tepidum and the anaerobic dehalogenating bacterium Desulfitobacterium hafniense. Among unicellular protists, the mammalian pathogens C. parvum, Giardia lamblia, Plasmodium falciparum, and Plasmodium yoelii yoelii contain START genes. By contrast, the genome of the unicellular green alga Chlamydomonas reinhardtii lacks START domains (Shinhan Shiu, personal communication).

START domains appear to be absent from the Archaea. Moreover the yeasts Saccharomyces cerevisiae and Schizosaccharomyces pombe, as well as another fungal species, Neurospora crassa, lack START domains. Perhaps START-domain genes are rarely found in prokaryotes and protists because their function is not required for normal unicellular proliferation. Their presence in a number of pathogenic species is consistent with the hypothesis of horizontal acquisition. This raises the possibility that in pathogens the START-domain proteins have a role in the interaction with host cells. For example, pathogens may recognize, transport and utilize lipids/sterols from the host organism via their START domains.

Putative START proteins were identified by BLASTP against databases at the National Center for Biotechnology Information (NCBI) 46, The Institute for Genomic Research (TIGR) 47, and the Munich Information Center for Protein Sequences (MIPS) 48. Putative START proteins from rice Oryza sativa ssp. indica were identified by BLASTP of sequences predicted on whole-genome shotgun deposited in MIPS Oryza sativa database (MOsDB) 49 using START-domain proteins from Arabidopsis (GL2, ATML1), rice (ROC1, AAP54082) and humans (PCTP) as query. START-domain proteins were verified using InterPro 50 and NCBI Conserved Domain (CD)-Search 5152, using cutoff E-values set to 1e-05. Arabidopsis proteins were checked against Arabidopsis EST databases using WU-BLAST2 53 to identify corresponding transcribed sequences. Supporting mRNA expression data was obtained from the microarray elements search tool 54 available through The Arabidopsis Information Resource (TAIR). Rice ESTs and cDNAs were identified using BLASTN at E-values of < 1e-60 (with most sequences showing an E-value = 0) and other plant ESTs were analyzed using TBLASTN at an E-value threshold of < 1e-80. Putative START protein-encoding plant cDNAs were identified from Sputnik EST database 55 and NCBI using Arabidopsis and rice START proteins as query. START domains were identified from the following plant species: Arabidopsis thaliana, Beta vulgaris, Capsicum annum, Cycas rumphii, Glycine max, Gossypium arboretum, Gossypium hirsutum, Helianthus annuus, Hordeum vulgare, Lactuca sativa, Lotus japonicus, Lycopersicon esculentum, Malus domesticus, Medicago sativa, Medicago truncatula, Oryza sativa, Phalaenopsis sp., Physcomitrella patens, Picea abies, Prunus persica, Solanum tuberosum, Sorghum bicolor, Triticum aestivum, Zea mays and Zinnia elegans.

Signal peptide and transmembrane analysis of the predicted START proteins was accomplished with TargetP 56 and TMHMM 57 using default settings. For the human MLN64 protein, and the MLN64-like proteins from D. melanogaster (CG3522/Start1) and C. elegans (3F991), sequence alignments were utilized to confirm the number and coordinates of the transmembrane segments. Experimental data for MLN64 38 was incorporated to predict the orientation of the transmembrane segments.

Alignments were made using ClustalW 58 with BLOSUM62 settings. Alignments of START domains were converted into the Molecular Evolutionary Genetics Analysis (MEGA) format in MEGA2.1 5960. For sequences originating from the same species, alignments were made using PileUp from Genetics Computer Group (GCG), screening for similarities and identities manually. After eliminating redundancies, datasets were realigned with ClustalW and a phylogenetic tree was derived. Phylogenies containing sequences from different organisms were constructed by the neighbor-joining method, assuming that sequences from different lineages evolve at different rates, followed by bootstrapping with 2,000 replicates 6061. The Poisson correction was used to account for multiple substitutions at the same site. Gaps were handled using complete deletion, assuming that different amino-acid sequences evolve differently. Alignments were constructed by GCG's PileUp to examine trends seen in the phylogenies. For analysis of proteins from different organisms, pairwise deletion was used instead of complete deletion. Domain alignments and consensus sequences shown in Figures 3 and 4 were constructed using GCG's Pretty 62.

Homology modeling of Arabidopsis and rice PCTP-like proteins was accomplished using the SWISS-MODEL web server 6364 and DeepView (Swiss-Pdb Viewer) 65. PDB templates of structures used for homology modeling were PCTP (PDB ID 1LN1A), MLN64 (1EM2A), and StarD4 (1JSSA). The WhatCheck summary reports for the highest likelihood models for Arabidopsis and rice PCTP-like proteins, At3g13062 and ID215312, respectively, were as follows: structure Z-scores: 1st generation packing quality: -1.779, -2.132; 2nd generation packing quality: -4.818, -4.425; Ramachandran plot appearance: -2.674, -2.393; chi-1/chi-2 rotamer normality: 0.138, 1.450; backbone conformation: -1.861, -1.446; and root mean square (RMS) Z-scores: bond lengths: 0.785, 0.859; bond angles: 1.389, 1.243; omega angle restraints: 1.007, 0.960; side chain planarity: 2.374, 1.346; improper dihedral distribution: 1.877, 1.544; inside/outside distribution: 1.324, 1.269. Coordinates and complete WhatCheck Reports are available on request. Cartoon images of the models were produced in RasMol Version 2.7.2.1.