Introduction

Neurotrophin proteins are fundamental developmental regulators for the vertebrate nervous system that promote the survival, growth, differentiation, and plasticity of neurons (e.g., Chao 2000; Patapoutian and Reichardt 2001). The neurotrophin gene family comprises six known constituents, the three genes included in the present study—nerve growth factor (NGF), brain-derived neurotrophic factor (BDNF), and neurotrophin-3 (NT-3)—and three other genes that are less well characterized: neurotrophin-4/5 (NT-4/5; also known as NT-4), which is absent from chickens (Carlos Ibañez and Finn Hallböök, personal communication) and presumably from other birds; and two that have been found in fish but not in other vertebrates, NT-6 and NT-7 (Hallböök 1999; Heinrich and Lum 2000). The differentiation of these now specialized neurotrophins occurred in concert with increases in complexity of the early vertebrate nervous system, probably via several gene duplication events from a single precursor neurotrophin found in the common ancestor of the Gnathostomata (Hallböök et al. 1998).

All known neurotrophin genes share a common structure. In vivo, they are translated as a pre-pro-protein which is proteolytically processed at a conserved arginine motif to yield the mature form. The presequence mediates the translocation of the proteins to the endoplasmic reticulum, and the pro-peptide region has been shown to promote folding of the mature region (Rattenholl et al. 2001a). Recent in vitro experiments have shown that a dimeric recombinant human NGF-pro-protein maintains biological activity (Rattenholl et al. 2001b), but less is known about the general function of the pre-proregions in vivo. Neurotrophin proteins bind to two classes of receptors, the p75 neurotrophin receptor (p75NTR) to which all neurotrophins bind and members of the Trk receptor family for which there is more ligand specificity (Kaplan and Miller 2000). The p75NTR receptor has been shown to mediate the apoptosis of neurons (Dobrowsky and Carter 2000) and oligodendrocytes (Beattie et al. 2002). The Trk receptors are high-affinity tyrosine kinase receptors that mediate growth and survival responses to the neurotrophins (Friedman and Greene 1999).

We sequenced the three avian neurotrophin loci from a set of 10 avian taxa that spans a broad range of evolutionary diversity, from representatives of the most ancient extant avian lineages to recently derived members of several suborders. Because the phylogenetic relationships of these taxa are known with reasonable certainty, we were able to reconstruct nucleotide and amino acid changes in the neurotrophins using an explicit phylogenetic context. This allows us to compare rates and patterns of nucleotide and amino acid evolution in the pro and mature regions of the three loci to assess whether their large differences in proregion variation stem from differences in mutation rate or from functional constraints. Recent experiments have shown that the proform, compared to the mature form, of NGF preferentially activates the p75NTR receptor (Lee et al. 2001; Chao and Bothwell 2002; Ibañez 2002; Beattie et al. 2002). The results presented here show that among the three avian neurotrophins, the proregion of BDNF has the highest degree of functional constraint and thus is likely to have a function not shared with the proregion of NGF and NT-3.

Materials and Methods

Laboratory Methods

We sequenced 587 nucleotides of the NT-3 gene corresponding to bases 172–759 of the human NT-3 sequence (Jones and Reichardt 1990 [GenBank M37763]), 602 nucleotides of the BDNF gene corresponding to bases 148–750 of the human BDNF sequence (Jones and Reichardt 1990 [GenBank M37762]), and 587 nucleotides of the NGF gene corresponding to bases 227–814 of the chicken NGF sequence (Ebendal et al. 1986 [GenBank X04003]) from up to 10 avian taxa and from the nonavian outgroup taxon Alligator mississipiensis.

Additional sequences were obtained from GenBank: Homo sapiens NGF (NM_002506), Homo sapiens BDNF (M37762), Homo sapiens NT-3 (M37763), Gallus gallus NGF (X04003), Gallus gallus BDNF (M83377), and Gallus gallus NT-3 (Z30092). In addition to obtaining these chicken sequences from GenBank, we sequenced the three Gallus gallus loci using the protocols described below, and in each case the nucleotide sequence we obtained was identical to the sequenced deposited in GenBank by previous investigators.

The taxa we sequenced and included in phylogenetic analyses (and their vernacular names used for convenience hereafter) were Alligator mississipiensis (alligator), Struthio camelus (ostrich), Gallus gallus (chicken), Anas platyrhynchos (mallard), Falco sparverius (kestrel), Halcyon malimbica (kingfisher), Coeligena torquata (hummingbird), Chloropipo holochlora (manakin), Mionectes striaticollis (flycatcher), Andropadus latirostris (bulbul), and Nectarinia olivacea (sunbird). Muscle samples from the alligator, ostrich, and chicken were obtained from commercial breeders; blood for the kestrel was kindly provided by L.A. Tell; DNA for the mallard was kindly provided by R.E. Ricklefs; and blood samples from the hummingbird, the kingfisher, and four passerines were kindly provided by T.B. Smith. Our phylogenetic analyses therefore included one nonavian outgroup, the alligator; one representative Palaeognathae, the ostrich; and nine Neognathae. These Neognathae represent six avian orders in the most widely used traditional avian classifications (e.g., Wetmore 1960). The order Passeriformes was represented by four taxa, two suboscines (manakin and flycatcher) and two oscines (bulbul and sunbird).

Purified genomic DNA was obtained from each sample using standard phenol–chloroform extraction protocols from blood or muscle tissue. Amplified DNA templates used for sequencing reactions were obtained via gene-specific PCRs using primers we designed from the published chicken sequences. All 5′ primers include the start codon of the protein-coding sequence, and all 3′ primers include the stop codon of the protein-coding sequence. NT-3 primers were ChickNT3-5′ (ATGTCCATCTTGTTTTATGTG) and ChickNT3-3′ (GTTCTTCCTATTTTTCTTGAC), BDNF primers were ChickBDNF5′ (ATGACCATCCTTTTCCTTACTATG) and ChickBDNF3′ (TCTTCCCCTTTTAATGGTTAATGTAC), and NGF primers were AllNGF5′ (GGTGC- ATAGCGTAATGTCCATG) and AllNGF3′ (ATAATTTAC- AGGCTGAGGTAG).

Each 25-µl PCR reaction mixture contained 10 mM Tris–HCl, pH 8.3, 50 mM KCl, 4.0 mM MgCl2, 0.001% gelatin, 200 µM of each dNTP, each primer at 1.25 µM, and 1 unit Amplitaq (Cetus–Perkin–Elmer). Reaction conditions were also invariant: an initial 5-min denaturation at 94°C; 35 cycles of 30 s of denaturation at 94°C, 30 s of annealing at 55°C, and 30 s of extension at 72°C; followed by 7 min of extension at 72°C. PCR products were checked via electrophoresis in 2% agarose and were purified using Qiagen spin-columns. Cycle sequencing reactions (with dye-terminator fluorescent labeling) were conducted using the amplification primers followed by electrophoresis using an ABI model 377 automated DNA sequencer (Applied Biosystems).

Strong PCR bands and clean chromatograms were readily obtained from all taxa for BDNF and NT-3 and from most taxa for NGF using the protocols described above. We encountered NGF amplification difficulties, however, for the alligator and the two oscine passerines, the bulbul and sunbird, and did not obtain NGF sequences from these taxa. This more variable amplification success for NGF is probably related to the much greater magnitude of proregion amino acid variation at that locus. Owing to the absence of the alligator, bulbul, and sunbird NGF sequences, most of our phylogenetic analyses were based only on BDNF and NT-3 sequences.

Sequence Alignment

Complementary sequences from each individual were aligned using Sequencer 3.1.1 (Gene Codes Corporation) and the chromatograms scanned by eye. All sequences used in phylogenetic analysis were confirmed by double-stranded sequencing, and all have been deposited in GenBank or EMBL nucleotide databases. Accession numbers listed here are in the order BDNF, NT-3, NGF: Struthio camelus, AF416632, SCA316235, SCA316265; Anas platyrhynchos, AF416633, APL316236, APL316266; Falco sparverius, AF416634, FSP316237, FSP316267; Halcyon malimbica, AF416635, HMA316238, HMA316261; Coeligena torquata, AF416636, CTO316239, CTO316262; Chloropipo holochlora, AF416637, CHO316240, CHO316263; and Mionectes striaticollis, AJ416888, AJ416887, MST316264. For the following three species, NGF was not sequenced (the order is BDNF, NT-3): Alligator mississipiensis, AF416634, AMI316234; Andropadus latirostris, AF416639. ALA316242; and Nectarinia olivacea, AF416640, NOL316243.

As described below, indels were rare and the sequences were aligned manually. In comparisons that included the alligator and all 10 avian taxa, no indels were present in the 600 nucleotides of the BDNF coding sequence, whereas two indels were present within the 591 nucleotides of NT-3 coding sequences. One of these NT-3 indels was a one-codon deletion restricted to the single coraciiforme taxon, the kingfisher H. malimbica. The other NT-3 indel was a one-codon deletion found exclusively in the two suboscine passerines, the manakin and flycatcher. In comparisons of NGF sequences that included eight avian taxa, we found only a single indel, a two-codon deletion present only in the suboscine passerine manakin C. holochlora.

Phylogenetic Relationships

To explore whether phylogenetic reconstructions based on the neurotrophin gene sequences were similar to those generated using other types of markers, we used maximum likelihood (ML) and maximum parsimony (MP) techniques to reconstruct relationships among taxa. Except where otherwise noted, analyses were conducted using PAUP*b.8 (Swofford 1999). These reconstructions were rooted using the alligator outgroup sequences.

ML analyses were conducted using MrBayes 1.11 (Huelsenbeck 2000), in which a Markov chain Monte Carlo search was parameterized using the general time reversible model (nst = 6), with site-specific rate variation partitioned by codon. Bayesian analyses were run for 250,000 generations and sampled every 1000 generations; inspection of the resulting ML scores suggested that stationarity was reached at approximately 10,000 generations in all searches, and we therefore conservatively discarded the topologies sampled from the first 20,000 generations. Consensus topologies from the remaining 2300 sampled trees were generated in PAUP*. MP reconstructions were generated in PAUP* via exhaustive searches, with all nucleotide substitutions weighted equally and with length variable sites excluded; MP bootstrap analyses were conducted for 1000 replications.

As discussed below, these reconstructions identified a single topology that was congruent with previous hypotheses of avian relationships based on independent molecular markers (e.g., DNA–DNA hybridization [Sibley and Ahlquist 1990]).

Comparisons of Rate Variation

We used three complementary methods to estimate levels of functional constraint in the pro and mature regions of the three neurotrophin loci. In all cases, comparisons were restricted to the eight avian taxa from which sequences of all three loci were obtained. All three analytical methods gave broadly congruent results.

First, we employed a simple but likely biased parsimony-based approach by importing the topology shown in Fig. 2 into the program MacClade (Maddison and Maddison 1992) and determining the minimum number of steps required to map the observed synonymous substitutions onto that tree. We quantified amino acid variation in the pro and mature regions of each gene in a similar way by determining the minimum number of steps required to map the observed amino acid changes onto that same topology. These parsimony-based character reconstructions provide estimates of the minimum number of nucleotide or amino acid changes that have occurred since these eight extant lineages shared a common ancestor, but this method is likely to underestimate the actual number of changes, particularly for classes of silent substitutions where some sites have likely undergone multiple changes. Second, we used the topology-based, maximum likelihood approach implemented in the CODEML program in the PAML package (Yang 1997) to estimate the mean d N/d S ratio (w) for each gene region. The search model was parameterized using the topology shown in Fig. 2, and we assumed a single d N/d S for each locus. Third, we used MEGA 2.01 (Kumar et al. 2001) to calculate d N/d S ratios from the aligned sequences using the Nei and Gojobori (1986) method as modified by Ina (1995) to include differing rates of transversion and transition substitutions.

Figure 1
figure 2

Phylogenetic relationships among 10 avian taxa based on analyses of combined BDNF and NT-3 nucleotide sequences. The tree shown is a phylogram from a Bayesian maximum likelihood search; thick lines indicate portions of this topology that were recovered across a variety of reconstruction techniques. Support indices for lettered internodes under these various reconstruction methods are given in Table 1.

Full size image

Results and Discussion

Nucleotide and Amino Acid Variation in Avian BDNF and NT-3

An analysis of the nucleotide sequences of each neurotrophin locus showed high levels of similarity among avian species and between the birds and the alligator outgroup. Figure 1 illustrates the alignment of the predicted protein sequences from the nucleotide sequences. The protein sequences of the mature regions of BDNF are identical among all the birds and the alligator, with only six amino acid substitutions separating the birds and human in the region we sequenced. Similarly, in the mature NT-3, the predicted protein sequence of alligator is identical to that of the chicken, but there are some conservative amino acid changes among the four passerine birds. NGF exhibited much greater variation among the avian taxa we examined (Fig. 1).

Figure 2
figure 1

Amino acid sequences of avian, alligator, and human BDNF, NT-3, and NGF neurotrophin genes. Dark vertical lines with arrowheads indicate the proteolytic processing sites that separate the prepro and maturse region of each protein. Shaded portions of mature region sequences indicate the variable regions (NH2, I, II, III, IV, V, and COOH) of the neurotrophins as described by Ibáñez et al. (1993). Darker-shaded regions in the proregions of the neurotrophins (I and II) indicate the two conserved domains necessary and sufficient for the biosynthesis of biologically active NGF as described by Suter et al. (1991). Asterisks identify sites known to be important for receptor binding (Ibáñez et al. 1992;Urfer et al. 1994). Indels are indicated by hyphens. The protein coding sequences of BDNF, NT-3, and NGF shown here begin at amino acid numbers 25, 26, and 6, respectively, of the published human sequences. The human sequences are included here for reference but were not used in phylogenetic analyses.

Full size image

Support for Phylogenetic Framework

As our comparisons of locus-specific patterns of molecular variation depend upon the phylogeny used to estimate the minimum number of nucleotide or amino acid substitutions at each locus, we examined whether phylogenetic reconstructions based on the neurotrophin sequences are congruent with one another and with previous reconstructions based on other markers. Our analyses of the combined NT-3 and BDNF sequences from 10 avian taxa suggest that nucleotide sequences of these loci are useful phylogenetic markers that provide broadly congruent estimates of avian phylogenetic relationships.

Phylogenetic reconstructions that included neurotrophin sequences from 10 avian taxa and the alligator outgroup all identified broadly congruent and generally well-supported trees (Fig. 2, Table 1). This topology was also obtained in reconstructions based on single gene sequences (Table 1). Most nodes in the tree shown in Fig. 2 also received consistently high support across reconstructions employing different subsets of the NT-3 and BDNF sequences and across different reconstruction techniques (Table 1). Topological features receiving universally high support included (1) the placement of the chicken/duck “galloanseriform” lineage basal to the remaining Neognathae, a relationship consistent with studies based on other molecular markers (e.g., Sibley and Ahlquist 1990; Caspers et al. 1997; Groth and Barrowclough 1999; Mindell et al. 1999; van Tuinen et al. 2000); (2) the monophyly of the four passerine taxa; and (3) the sister relationships of the bulbul/sunbird and the manikin/flycatcher within the passerine clade. The only region of the tree receiving consistently weak support involved the relationships among the four neognath lineages represented by the kestrel, kingfisher, hummingbird, and passerines. These lineages are each representatives of different avian orders in traditional taxonomic classifications, and there is growing molecular phylogenetic evidence (e.g., Sibley and Ahlquist 1990; Groth and Barrowclough 1999; van Tuinen et al. 2000) that these and other Neognathae lineages arose during a relatively short period of extensive radiation subsequent to the split between the galloanserifomes and the remaining neognaths.

Table 1 Support indices for the topology shown in Fig. 2 under different reconstruction techniques
Full size table

Comparisons of phylogenetic analyses that included all nucleotide sites with those that included only amino acid-invariant codons indicated that nucleotide substitutions at synonymous codon positions contributed most of the phylogenetic signal to the reconstructions. In particular, changes at synonymous sites do not appear to be more phylogenetically noisy than changes at nonsynonymous sites, as would be expected if synonymous sites had high levels of nucleotide saturation and exhibited concomitantly high levels of homoplasy. For example, posterior probabilities in the ML analysis restricted to amino acid-invariant codons were very similar to those in the ML analysis that included all codons, despite the potential loss of information caused by excluding the 43 variable codons. Similarly, in the MP analyses, measures of character consistency for the shortest trees based on the complete sequences (RCI = 0.423) and on the invariant amino acids were nearly identical (RCI = 0.400).

Molecular Evolution of Avian Neurotrophin Genes

Within each of the three neurotrophin loci, we noted large but expected differences in relative amino acid variation between the proregions and the mature regions: for all three loci, a much higher proportion of codons varied in the proregion than in the mature region (Fig. 1). Amino acid conservation was strict in all three mature regions: the mature region of the BDNF gene showed no amino acid variation, two codons of the NT-3 gene varied, and seven codons of the NGF gene varied (see also Fig. 3). In contrast, much greater variation in amino acid constraint was present among the proregions of the three loci: the BDNF proregion varied at 12 codons; the NT-3 proregion was intermediate, with variation at 29 codons; and NGF proregion had the highest variation, at 53 codons.

Figure 3
figure 3

Contrasting levels of amino acid constraint at three avian neurotrophin loci. A A parsimony-based index of variation calculated by dividing the number of steps required to map the observed nucleotide or amino acid changes onto the topology depicted in Fig. 2, divided by the total number of nucleotide sites (left-hand comparison) or amino acids (two right-hand comparisons) within a given protein region. B Nonsynonymous:synonymous substitution ratios (d N/d S) estimated via a topology-based, maximum likelihood approach (Yang 1997). C d N/d S ratios estimated via the modified (Ina1995) method of Nei and Gojobori (1986). In the bottom two panels, high d N/d S ratios suggest relaxed selective constraint, whereas low ratios indicate strong selective constraint.

Full size image

This pattern across loci of differing variability in the pro and mature regions is interesting primarily because it provides evidence for differences in the functions of the protein sequences encoded by the three proregions. The high amino acid conservation in the mature regions of each locus is likely to result from strong functional selection on the mature protein, but it is less obvious why the three proregions would show different levels of functional constraint. One explanation is that the three loci could be subject to different underlying mutation rates, but the similarity in the rates of synonymous nucleotide substitutions across all three loci (Fig. 3) argues against this possibility. The similarities in structure and composition of these three neurotrophin genes make them an unusually tractable system for comparing rates and patterns of molecular evolution among loci. In particular, we can test for intrinsic mutation rate differences among loci by comparing locus-specific rates of nucleotide substitutions at synonymous sites, because those sites are presumably free from selection on function at the amino acid level. We chose to base our estimates of underlying mutation rates on nucleotide changes at synonymous nucleotide positions within the mature regions of each protein. These degenerate sites are highly comparable across loci because the mature regions of the three loci have similar general structures, including a number of blocks of amino acids that are semiconserved across loci (Fig. 1); have similar codon use and therefore similar proportions of degenerate nucleotide sites; are nearly identical in length; and have nearly identical and highly even base compositions. All three loci have very similar rates of synonymous substitution (Fig. 3A).

All three mature regions also have similar (and very low) d N/d S ratios (Fig. 3B and C), indicating that they are all subject to high levels of selective constraint. In contrast, we found an order-of-magnitude difference in the d N/d S ratios at the three proregions (Fig. 3): the BDNF proregion is nearly as conserved as the NGF mature region, whereas the highly variable NGF proregion is under very relaxed constraint. We speculate that the BDNF proregion (and perhaps the intermediately variable NT-3 proregion) may have a function in addition to its role in protein folding. For example, the conservation of the proregions inversely correlates with the efficiency of proteolytic cleavage of the neurotrophins (NGF proteolytic processing is considerably more efficient than BDNF proteolytic processing [Mowla et al. 2001; Carlos Ibañez, personal communication]). Thus, in vivo the ratio of proBDNF to mature BDNF is likely to be higher than the corresponding pro:mature NGF ratio. In addition, recent studies have shown that proBDNF mediates TrkB phosphorylation and is thus biologically active, and that BDNF differs from NGF and NT-3 in its secretory pathway (Mowla et al. 2001). It seems likely that proBDNF may have a function that differentiates it from mature BDNF, with proteolytic processing being a possible mechanism that regulates the disparate biological activities. Lee et al. (2001) found that the proforms of NGF and BDNF are expressed in vivo, secreted, and cleaved extracellularly and that the proform of NGF preferentially activates the p75NTR. In addition, recent studies have shown that proNGF induces p75NTR-mediated apoptosis of oligodendrocytes following spinal cord injury (Beattie et al. 2002). Based on the patterns of variation among the three avian loci, we suggest that BDNF, with its apparent high degree of proregion functional constraints, may be the most likely candidate to operate in its proform and to be regulated by proteolytic processing in vivo. The apparent absence of the NT-4 protein in chickens (C. Ibañez and F. Hallböök, personal communication) further suggests that BDNF may play a compensatory role in the development of the avian nervous system.