Members of the bHLH superfamily have two highly conserved and functionally distinct domains, which together make up a region of approximately 60 amino-acid residues. At the amino-terminal end of this region is the basic domain, which binds the transcription factor to DNA at a consensus hexanucleotide sequence known as the E box. Different families of bHLH proteins recognize different E-box consensus sequences. At the carboxy-terminal end of the region is the HLH domain, which facilitates interactions with other protein subunits to form homo- and hetero-dimeric complexes. Many different combinations of dimeric structures are possible, each with different binding affinities between monomers. The heterogeneity in the E-box sequence that is recognized and the dimers formed by different bHLH proteins determines how they control diverse developmental functions through transcriptional regulation 10.

The bHLH motif was first observed by Murre and colleagues 11 in two murine transcription factors known as E12 and E47. With the subsequent identification of many other bHLH proteins, a classification was formulated on the basis of their tissue distributions, DNA-binding specificities and dimerization potential 12. This classification, which divides the superfamily into six classes, was initially based on a small number of HLH proteins but has since been applied to larger sets of eukaryotic proteins 1. More recently, an approach using evolutionary relationships was used to classify bHLH proteins into four major groups (A-D) 13, taking into account E-box binding, conservation of residues in the other parts of the motif, and the presence or absence of additional domains. The sequencing of new genomes has led to the identification of additional bHLH families, and this evolutionary classification has now been extended to include two additional groups (E and F; Table 1) 6. Parsimony analysis by Atchley and Fitch 13 of a phylogenetic tree derived from 122 sequences suggested that an ancestral HLH sequence most probably came from group B, and group B proteins are indeed the most prevalent type of bHLH proteins in animals. The situation is similar in the Arabidopisis genome, in which the G-box-binding bHLH proteins (part of group B) are the most abundant group 7.

One basis for the evolutionary classification shown in Table 1 is the presence or absence of additional domains, of which the most common are the PAS, orange and leucine-zipper domains. PAS domains, located carboxy-terminal to the bHLH region, are 260-310 residues long and function as dimerization motifs 14. They allow binding with other PAS proteins, non-PAS proteins, and small molecules such as dioxin. The PAS domain is named after three proteins containing it: Drosophila Period (Per), the human aryl hydrocarbon receptor nuclear translocator (Arnt) and Drosophila Single-minded (Sim) 15. The domain is itself made up of two repeats of approximately 50 amino-acid residues (known as PAS A and PAS B) separated by about 150 residues that are poorly conserved 16. PAS-domain-containing bHLH proteins (bHLH-PAS proteins) form phylogenetic group C. A distinct additional domain, the orange domain, is a 30-residue sequence that is also located carboxy-terminal to the bHLH region, from which it is separated by a short, variable length of sequence. Transcription factors with this additional domain, designated bHLH-O and forming part of phylogenetic group E, include the hairy-related proteins, called HEY1, HEY2 and HEYL in mouse and humans 17. The molecular function of the orange domain is still unclear; it has been proposed that it mediates specificity and transcriptional repression 18, but there is also evidence that it can play a role in dimerization 17.

A number of bHLH protein families, mostly in phylogenetic group B, have a leucine-zipper domain contiguous with the second helix of the HLH domain; like the HLH domain, this mediates dimerization. Proteins that have only a leucine-zipper domain coupled with a basic domain (denoted bZIP) and no HLH domain are a separate family of DNA-binding proteins in their own right (reviewed in 19). The sequence of the zipper consists of a repeating heptad, with hydrophobic and apolar residues occurring at the first and fourth positions and polar and charged residues at the remaining positions. Leucine is the residue that predominates at position 4; it thus lends its name to the zipper motif. One bHLH protein that has a leucine-zipper domain (and that is therefore denoted a bHLHZ protein) is Max, which forms the hub of a network of bHLH transcription factors. Max is known to form homodimers and heterodimers with the group B proteins Myc, Mad, Mnt and Mga, and these complexes each have sequence-specific DNA-binding and transcriptional functions 20.

The additional domains in bHLH proteins, such as the leucine zipper, are always carboxy-terminal to the bHLH region. The position of the bHLH and additional domains within the complete sequence of the protein varies widely between different families, however. This variable pattern of domain positioning has led to the proposal that bHLH proteins have undergone modular evolution by domain shuffling, a process that involves domain insertion and rearrangement 21.

In comparison with the volume of sequence data, structural data for the bHLH superfamily of transcriptional regulators are still relatively sparse. Just nine bHLH protein structures have been deposited to date in the Protein Data Bank (PDB; see Table 2) 22. The CATH 23 and SCOP 24 protein-structure classifications classify eight of these structures into one superfamily (Table 2; SREBP-2 has not been classified). A number of the structures (PDB codes 1an2,1ihlo, 1nlw, 1nkp, and 1am9) include an additional zipper domain that is carboxy-terminal to the HLH region. Two of the structures solved are heterodimers: a Max-Myc complex (PDB code 1nkp) and a Max-Mad complex (PDB code 1nlw). The remaining complexes are homodimeric, and all but one include the structure of the bound DNA double helix, giving insights into the binding specificity at the E box. Representatives of these bHLH structures are shown in Figure 1.

The structure of MyoD (Figure 1a) is typical of many bHLH proteins, comprising two long α helices connected by a short loop, which in the case of MyoD is 8 residues in length. The first helix (H1) includes the basic domain, which makes contact with the major groove of the DNA. MyoD is a homodimer in which the two monomers make identical contacts with the DNA. Comparisons of this structure with that of Max (which includes an additional leucine zipper domain; Figure 1d,1e,1f) reveal that the presence or absence of this domain does not significantly affect the structure of the bHLH segment 25.

Two interesting features revealed by the three-dimensional structure of the Pho4 bHLH domain (Figure 1b) are the existence of a short stretch of α-helix in the loop region that links helix H1 to helix H2 and the recognition of DNA bases outside the E-box sequence 26. The Pho4 protein binds DNA as homodimer, and its two subunits form a parallel four-helix bundle (Figure 1b). The short α-helix region in the loop lacks the stabilizing hydrogen-bonding network observed in other bHLH proteins. In the Pho4 structure, each half-site of the symmetrical E box is recognized by a triad of residues, but bases beyond the E box, including a GG sequence at the 3' end, are also recognized 26. Base recognition outside the E box is also observed for MyoD, but in this structure it occurs at the 5' end of the E box 25.

Sterol regulatory element binding protein la (SREBP-1a; Figure 1c) is an example of a bHLH structure that includes one of the additional domains, the leucine zipper. SREBPs are bHLHZ transcription activators that bind to a DNA target site as a homodimer and are essential for cholesterol metabolism 27. Unlike other bHLH proteins that recognize a symmetrical E box, SREBP-1a recognizes an asymmetrical sterol regulatory element. This asymmetric recognition is possible because of the presence of a tyrosine residue in the basic domain. The tyrosine replaces the arginine observed in other bHLH proteins such as Max, and this change results in the loss of polar interactions with the DNA 27. Recently, a crystal structure of another SREBP, SREBP-2, has been solved 28, in which SREBP-2 is bound in a complex with importin-β, a molecule that mediates the transport of molecules into and out of the nucleus; the structure reveals that SREBP-2 is imported into the nucleus as a homodimer.

Two of the most interesting structures to be solved to date are those of the Max-Mad (Figure 1d) and Max-Myc (Figure 1e) heterodimer complexes bound to double-stranded DNA 29. In each monomer, the amino-terminal α helix is a continuous secondary-structural element that includes the basic region and the α helix H1, and the carboxy-terminal α helix is made up of two continuous α-helical segments, helix H2 and the leucine-zipper region. The Myc-Max and Mad-Max complexes are quasi-symmetric heterodimers that have interfaces made up of hydrophobic and polar interactions involving residues in helices H1 and H2 and the leucine zipper. Mutation studies suggest that dimer specificity is controlled by the amino acids Gln91 and Asn92 (in the Max numbering) in the Myc-Max dimer. The studies also show that Glu125 controls Mad-Max heterodimer formation 29. One interesting feature of the Myc-Max crystal structure (Figure 1e,1f) is the presence of two heterodimers in the asymmetric unit of the crystals. The two structures form a heterotetramer in which the head-to-tail assembly of leucine zippers from different heterodimers results in the formation of an antiparallel four-helix bundle (Figure 1f). It has been shown previously that Myc-Max heterodimers can form higher multimeric structures 30, and there is evidence to suggest that the tetramer observed in the crystal also exists under physiological conditions 29.

The heterogeneity of DNA sequences recognized and dimers formed by the bHLH proteins enable them to function as a diverse set of regulatory factors. The bHLH proteins can be divided into those that are cell specific and those that are widely expressed. The cell-type-specific members of the superfamily are involved in cell-fate determination in many different cell lineages and form an integral part of many processes, including neurogenesis, cardiogenesis, myogenesis, and hematopoiesis (Table 3). The bHLH proteins involved in neurogenesis include Drosophila Atonal and other 'proneural' proteins 31. In vertebrates, Mash-1, Math-1 and the neurogenins are important in the initial determination of neurons, whereas Nero-D, NeuroD2, MATH-2 and others are differentiation factors 32. The bHLH transcription factors dHAND and eHAND are important in cardiac development in vertebrates 33. The myogenic regulatory factors, including MyoD, MRF-4, Myf-5 and myogenin, together regulate both the establishment and differentiation of the myogenic lineage 34. The stem cell leukemia (SCL) protein is a bHLH transcription factor that is essential for hematopoiesis and is associated with acute T-cell leukemia 35.

One family of bHLH proteins that is widely expressed in many different cell types is the Myc family. The Myc genes are among the most frequently affected genes in human tumors 36. Myc proteins are known to regulate translation initiation 37 and they also function as transcriptional activators when they form heterodimers with Max proteins (also members of group B) 38. There is some evidence, however, that these dimers may also operate as negative regulators of transcription (reviewed in 39). Max is also known to form homodimers and heterodimerize with other bHLH proteins including Mad 38. This dimerization network of Myc/Max/Mad transcription proteins has a large number of target genes involved in the cell cycle, and the network has been considered to function as a transcription module 20.

In summary, the bHLH superfamily constitutes a large and diverse class of proteins, with over 125 different proteins identified in humans and 145 in Arabidopsis. The discovery of their diverse functions in the cell cycle, cell-lineage development and tumorigenesis has elevated the interest in them in the 15 years since they were first identified by Murre and co-workers 11. So what do the coming years hold in store for this superfamily? With the sequencing of more genomes, it is expected that further superfamily members and new sequence families will be identified. With an increasing number of proteins targeted and solved by structural-genomics consortia, the structural data available for this superfamily will also grow. The knowledge gained from new sequences and novel high-resolution structures will offer further insights into the mechanisms by which they control such diverse processes. This increasing knowledge base may make them good targets for new drug therapies for conditions including heart disease and cancer.