The vertebrate innate immune system provides protection against a wide range of pathogenic microorganisms, and defensins are an important component of this response as well as having a role in adaptive immunity. In mammals, the defensins can be divided into the α- and β-defensin subfamilies on the basis of differences in the spacing of six, conserved cysteine residues. The α-defensins are produced by neutrophils and intestinal Paneth cells, whereas the β-defensins are mainly produced by epithelial cells in contact with the environment. The functions of human β-defensins seem to be disrupted in cystic fibrosis and inflammatory skin lesions such as psoriasis 12. A fuller knowledge of the human complement of β-defensins may therefore be useful in understanding human disease as well as in the design of novel, synthetic antimicrobial peptides.

The known human β-defensin genes show a conserved two-exon structure: the first exon encodes a signal peptide whereas the second exon encodes a short propiece and the mature defensin peptide with a characteristic six-cysteine motif and many basic amino-acid residues 3. The β-defensin genes are present at five syntenic loci in the human and mouse genomes, with the main locus on human chromosome 8p22-23 and mouse chromosome 8A3 4. All four, full-length, human β-defensins that are present in the public databases are from 8p22-23 (GenBank sequence accession numbers are human β-defensin 1 or DEFB1, Q09753; DEFB4 (formerly DEFB2), O15263; DEFB103 (formerly DEFB3), NP_061131; DEFB104 (formerly DEFB4), CAC85520), but there are substantial differences in their coding sequences, expression patterns and antimicrobial activities. DEFB1 is constitutively expressed in many tissues (respiratory tract, kidney, urogenital and oral cavity epithelia) whereas DEFB4 is expressed in response to bacterial infection or proinflammatory agonists in respiratory tract epithelial cells, and epidermal and gingival keratinocytes. Both DEFB1 and DEFB4 proteins have salt-sensitive, bactericidal activity against a spectrum of Gram-positive and Gram-negative enteric, urinary tract, and respiratory bacteria in vitro 5. DEFB103 is expressed in epithelial cells, adult heart, skeletal muscle, placenta and fetal thymus, it has broad-spectrum antimicrobial activity under conditions of low salt and (unusually among β-defensins) it retains activity against Staphylococcus aureus even in physiological saline. DEFB104 achieves highest expression in the testis (with lower levels in gastric antrum, neutrophils, uterus, thyroid, lung and kidney) and was found to be inducible in the respiratory epithelium upon exposure to Pseudomonas aeruginosa or Streptococcus pneumoniae 3.

The evolution of various genes involved in the vertebrate immune system has involved duplication followed by selection to provide responses to a wide range of pathogens, with well documented examples in immunoglobulin 6 and major histocompatibility complex genes 7. Hughes and Yeager 8 studied the evolution of α-defensins and found evidence for duplication followed by diversification driven by positive selection. Similar phenomena were also implicated in the evolution of bovine β-defensins 9 and amphibian antimicrobial peptides 10. In contrast, DEFB1 was found not to vary significantly across primates 11.

Here we describe a combined strategy to identify further β-defensin genes in the draft human genome sequence using computational techniques and verification using reverse transcription-PCR. Four full-length, novel genes (DEFB105, DEFB106, DEFB107, DEFB108) and a related pseudogene DEFB109p are reported, as well as their expression patterns and evidence for their evolution by duplication and positive selection.

All TBLASTX 12 matches to human bacterial artificial chromosome (BAC) clone sequences were in the 8p22-p23 region in a subsection of FPC contig ctg45 (1 April freeze WashU Accession Map Layout Files 13) bounded by the BAC clones RP11-161B1 (AC079018) and SCb-177K12 (AF252831) and consisting of 53 BACs in total. These 53 BACs were deemed to represent the human β-defensin gene family locus and were masked for repetitive sequences using RepeatMasker 14. This locus was the subject of a more sensitive search for the presence of novel human β-defensins, using a hidden Markov model constructed from an alignment of the GenBank β-defensin sequences mentioned above. As well as the known, full-length human β-defensins and the related epididymis-specific SPAG11 (formerly EP2) gene 15, two novel β-defensin genes, DEFB105 and DEFB106, were identified in this search and were then incorporated into the previous hidden Markov model. Further searches with this revised model identified a further three genes: DEFB107, DEFB108 and DEFB109p. None of these five genes was found in the EMBL sequence database (24 June 2002 release) or in the Ensembl genomic annotation database (version 6.28.1 16). The novel gene DEFB109p appears to be a pseudogene as it contains a premature stop codon within its first exon, as observed in three independently sequenced, overlapping 8p22-p23 BAC sequences (accession numbers AC068974, AC087203 and AF252830). Given the absence of premature stop codons in the other four genes, despite considerable divergence among them (they encode only 18-28% identical amino-acid residues), it is unlikely that they too are pseudogenes.

Many putative final exon fragments from novel mouse β-defensins were recently reported by Schutte et al. 4. However, their data were incomplete and were not supported by experimental verification. They used computational techniques to identify sequences matching a central portion of the mature defensin peptide including the six-cysteine motif characteristic of β-defensins. At best, this method could only identify incomplete final exons encoding this region of the peptide. No attempt was made to delineate precisely the boundaries of final exons or to identify first exons by these authors. In the present study we restrict our attention to complete genes, present in a BAC-clone-based map of the region and verified as encoding real transcripts by RT-PCR. Our full-length novel genes correspond to five final exon fragments (DEFB5, DEFB6, DEFB7, DEFB8 and DEFB9) reported by Schutte et al. 4 and we have adopted the official HUGO Human Gene Nomenclature Committee 17 names for these fragments: DEFB105, DEFB106, DEFB107, DEFB108 and DEFB109p (amended from DEFB109) respectively.

RT-PCR amplification confirmed the presence of the computationally predicted, functional genes, but sequencing of these products and subsequent alignment to genomic sequence also uncovered several instances where the actual gene structures differed from those that were predicted. In two cases, DEFB106 and DEFB107, splice sites other than those predicted were used and it was found that the predicted coding sequences had been respectively 36 base-pairs (bp) longer and 9 bp shorter than the actual coding sequences. In the case of DEFB105 RT-PCR verified the two predicted exons but also uncovered an additional, 42-bp intervening exon. The RT-PCR analysis also produced interesting results with respect to the coding sequence of DEFB108. In four independent, sequenced RT-PCR products from one mRNA sample, the amplified DEFB108 sequence was consistently found to differ from the computationally predicted sequence from the human public draft sequence at three nucleotide sites, but was otherwise identical. The three differences observed between predicted and amplified sequences respectively were an A to G at nucleotide 62 (which causes a conservative amino-acid change from lysine to arginine), a T to C at nucleotide 111 (which is a synonymous change) and a C to T at nucleotide 120 (also synonymous). This indicates novel human polymorphisms in DEFB108 and is consistent with observations of high degrees of polymorphism for other human β-defensins 18 (see also Additional data files 1-6 for the novel gene sequences confirmed by RT-PCR and the computationally predicted sequence of DEFB109p).

Figure 1a depicts the relative positions and orientations of the four novel genes, the novel pseudogene DEFB109p and four known genes in the vicinity: DEFB2-4 and the related epididymis-specific SPAG11. This cluster is about 350 kilobases (kb) centromeric of the DEFB1 gene. A phylogenetic tree relating functional human β-defensins (Figure 1b) reflects the spatial distribution of these genes, with the cluster of genes in Figure 1a appearing to form a clade separate from DEFB1. The simplest explanation for the origin of this cluster is a series of local duplication events followed by substantial divergence. In addition it seems that the four novel functional genes are more closely related to DEFB104 than to DEFB4 or DEFB103. Three of these genes, DEFB106, DEFB105 and DEFB108, encode mature peptides exhibiting the same spacing of conserved cysteine residues as seen in DEFB104, although in DEFB105 there is an extra cysteine residue (amino acid 43) towards the amino-terminal end of the mature peptide. DEFB105 also encodes an unusually long propiece peptide region in the second of its three exons. Strikingly, the DEFB107 protein lacks the first canonical β-defensin cysteine altogether and instead has a serine residue at the same position (Figure 2). The changes in the number of cysteine residues seen in DEFB105 and DEFB107 are likely to have important functional consequences. The predicted mature peptides for the four novel, functional β-defensins presented here have a similar proportion of cationic residues to human DEFB104, with higher proportions of anionic residues (10-13%) than are seen in human DEFB1, DEFB4 and DEFB103 (less than 4%). Indeed the pI of DEFB107 and DEFB108 are 6.74 and 6.89 respectively, whereas all other β-defensins described to date are cationic. This relative increase in anionic residues is expected to affect function, as the action of defensins initially involves interactions between the cationic mature defensin peptides and anionic membrane lipids 19. Expression analysis for the novel, functional human genes was carried out by RT-PCR on a panel of human RNA samples and the novel gene PCR products were confirmed by hybridization to an internal probe (Figure 3). Expression of all four novel genes was readily detected in testis. A longer exposure period revealed low levels of expression of DEFB108 in the liver (Figure 3, left-hand panel of DEFB108). Expression was not detected in any of the other tissues analyzed.

Six genes highly similar (85-98% identical at the amino-acid level) to three of the novel human β-defensins (DEFB105, DEFB106 and DEFB107) as well as to DEFB4, DEFB103 and DEFB104, were found within two olive baboon (Papio cynocephalus anubis) draft genomic sequences (GenBank accession numbers AC116558 and AC116559) using BLAST. Full-length sequences were obtained for each baboon gene except the putative DEFB4 ortholog (which lacks a first exon because of gaps in the draft genomic sequences); in all other cases the exonic structure of the putative baboon ortholog was identical to that of the human gene (Figure 2). It is very likely that these sequences originate from the baboon locus orthologous to the human region under study, but without more complete sequence or mapping data for the baboon genome it is impossible to be certain. These novel baboon gene sequences (see Additional data files for the accession numbers), together with the published sequence for olive baboon β-defensin 1 (AAK61474) and the full-length human β-defensins formed the basis for our evolutionary analyses.

Figure 4 shows d_N (number of nonsynonymous substitutions per nonsynonymous site) plotted against d_S (number of synonymous substitutions per synonymous site) for comparisons between the first exon (which encodes the signal peptide) and second exon (which encodes the mature defensin) from all human and baboon genes (the full d_N, d_S and d_N - d_S estimates for all first and second exon comparisons are available as additional data file 7). Two major trends are observable. In the vast majority of first exon comparisons d_S exceeds d_N (Figure 4a) and this excess is statistically significant in almost every case according to two-tailed Z-test results for all human and baboon genes, but there were no significant excesses of d_S according to the more rigorous Fisher's exact test (data not shown). Thus the rates of substitution in first exons indicate that they are evolving approximately neutrally, perhaps under weak purifying selection. The pattern seen in the second exon comparisons is quite different (Figure 4b). In the second exons d_N often exceeds d_S, and even in these short sequences, for certain comparisons this excess reaches statistical significance. Significant excesses of d_N over d_S are seen between DEFB1 and DEFB104 and between DEFB103 and DEFB107, with similar effects seen among the second exons of orthologous baboon genes (Table 1) using the method of Nei and Gojobori 20.

Moreover, in these comparisons d_S tends to be rather low relative to the rest of the data set (mean d_S = 0.464). Similar results are obtained using this method either modified to take account of the transition-to-transversion ratio R 21 or using the Jukes-Cantor correction, although the unmodified method is thought to be a more reliable basis for the detection of positive selection 22. If we assume that synonymous substitutions (which are selectively neutral or nearly so) have accumulated regularly with time, such a pattern of substitution suggests that duplication was followed by rapid nonsynonymous change that subsequently decelerated. Most comparisons involving the second exons of DEFB103, DEFB104 and DEFB107 with the second exons of other genes in the dataset show excesses of d_N over d_S, but these excesses fail to reach significance by either of the tests used (data not shown). Although some of the comparisons shown in Figure 4b indicate an excess of d_S over d_N, this is never significant by either of the two statistical tests used.

There is no detectable similarity between the introns of the genes under study except between the five putatively orthologous pairs of human and baboon genes where intronic sequence is available (DEFB103, DEFB104, DEFB105, DEFB106 and DEFB107): each pair shares 74-91% identity over 80-99% of their lengths in spite of many small indel events. In every case these pairs of orthologous introns show a substitution rate that is not significantly different from the value of d_S obtained for the coding sequences. In particular, the orthologous intron sequence comparisons for DEFB103, DEFB104 and DEFB107 give substitution-rate estimates of 0.1066 ± 0.010, 0.0816 ± 0.005 and 0.0666 ± 0.004 respectively, which are consistent with the d_S estimates for the coding sequences of these pairs: 0.1666 ± 0.077, 0.0856 ± 0.058 and 0.0536 ± 0.049. Thus, the excess of nonsynonymous substitutions observed for the second exons of these genes is not attributable to artificially low d_S estimates as a result of sampling error, but is caused by a real increase in d_N relative to d_S, which is the pattern expected to be generated by positive selection.

Averages for the ratio of radical to conservative amino-acid changes, p_R/p_C calculated over 47 mammalian genes were reported as 0.81 and 0.49 for charge and the Miyata-Yasunaga (polarity and volume) classification respectively 23. The equivalent values from Table 1, for comparisons between the second exons of genes showing evidence of positive selection, are all greater than these averages. Furthermore Table 1 shows that there has been a higher rate of change with respect to charge than with respect to polarity and volume (Miyata-Yasunaga amino-acid classification). That is, where there is evidence of positive selection, most nonsynonymous changes have tended to change the charges of the residues encoded but have tended to conserve the polarities and volumes of those residues.

Likelihood ratio tests (LRTs), as implemented by the PAML package 24 also indicate the operation of positive selection at sites within second exons. These tests indicate whether data (the substitutions inferred from an alignment) are best explained by one of two models of ω = d_N/d_S. Since ω is a measure of selective pressure on proteins, these models can be used to assess the evidence for variable selective pressures among sites. In a test for positive selection (the presence of sites at which ω > 1) two statistical distributions are compared: a null model that uses a distribution that does not allow for sites with ω > 1 and another model which does allow such sites. Three pairs of site-specific likelihood models were compared that assume variable selective pressure (as determined by the value of ω) among sites but no variation among sequences in the dataset: M0 (one-ratio) and M3 (discrete), M1 (neutral) and M2 (selection), and M7 (beta) and M8 (beta + ω) 25. The second exon of the pseudogene DEFB109p was omitted from the analysis. The discrete model (M3) with two site classes suggested that 41% of second-exon sites are under positive selection with ω= 1.67 and identified nine amino-acid sites under positive selection at the 95% cutoff. M3 was a significantly better fit to the data than the one-ratio model M0; the LRT statistic is 2Δl = -1061.79-(-1004.12) = 115.32, and p < 0.001 with two degrees of freedom. However the M0-M3 comparison is essentially a test of variability in the ω ratio among sites and does not constitute a rigorous test of positive selection. Model M1 (neutral) assumes two site classes with ω = 0 and ω₁ = 1 fixed and with the proportions p and p₁ estimated. Model M2 (selection) adds a third site class with the ratio ω₂ estimated, it suggests that about 47% of sites are under positive selection with ω₂ = 53.17 and identified 12 amino-acid sites under positive selection at the 95% cutoff. The two models can be compared using an LRT as follows, 2Δl = -1030.51 -(-1022.03) = 16.97; p < 0.001 with 2 df. So model M2 is significantly better than M1. Model M7 (beta) assumes a beta distribution for ω over sites. The beta distribution is limited to values between 0 and 1, providing the most flexible null hypothesis for testing positive selection. Model M8 (beta + ω) adds another site class to M7 (beta), with the ω ratio estimated from the data. However, the difference between M7 and M8 is not statistically significant, as indicated by the LRT: 2Δl = -1006.44 -(-1006.44) = 0. Nine particular sites were implicated (in both M2 and M3 models) as being under positive selection with greater than 95% confidence: positions 43, 44, 48, 52, 56, 57, 69, 70 and 73 in Figure 2a. All these positions are close or adjacent to a conserved cysteine residue and so it is possible they are important in determining β-defensin structure. In summary, the PAML analysis indicates that ω varies significantly between sites, and in two separate LRTs the parameters estimated suggest a substantial proportion of sites are under positive selection. In spite of this, the most stringent test (M7 versus M8) does not indicate a significant difference from neutrality. This result may be attributable to the extremely short lengths (33 codons aligned omitting positions with gaps) of the sequences aligned, such an effect was seen when analyzing short (130 codons) lysozyme sequences in the same way 25.

This study represents the most detailed study of a human β-defensin cluster to date, including the full-length sequences of four novel genes (DEFB105, DEFB106, DEFB107, DEFB108), and a novel pseudogene (DEFB109p), their expression patterns and sequences for the baboon orthologs of six genes from this cluster. The 8p22-23 defensin locus appears to have evolved by successive rounds of duplication followed by substantial divergence, to produce a diverse cluster of paralogous genes defined by these four novel genes and four known β-defensin genes (DEFB4, DEFB103 and DEFB104). Divergence has been most rapid within the second exons of these genes, which encode the mature β-defensin peptide, with many comparisons between paralogous genes showing an excess of nonsynonymous over synonymous substitutions. Statistically significant evidence of elevated nonsynonymous change is seen by two methods in the second exons, indicating the action of positive selection. By contrast, comparisons between the first exons of genes from this cluster, which encode a signal peptide, show an excess of synonymous substitutions consistent with neutral evolution or weak purifying selection. The duplication and subsequent positive selection of these genes predates human-baboon divergence more than 23 million years ago and is consistent with observations that DEFB1 has undergone very little change during the evolution of primates 11. The positive selection observed has tended to change the charges of residues encoded more than other qualities such as residue polarity or volume. As seems to be the case with other antimicrobial peptides, such as MHC receptors, immunoglobulins and α-defensins 678, this selection may be a response to the rapid evolution of pathogens.

In this study the computationally predicted gene structures were found by laboratory work to deviate from the actual structures in three out of five novel genes. These errors in the predictions arose in spite of the fact that the predictions were based on all the β-defensin protein-sequence data available and involved three completed BAC sequences rather than unfinished, gapped sequence. This has implications for purely computational approaches to novel gene discovery such as that of Schutte et al. 4.

DEFB104 and DEFB108 have no detectable orthologs in the mouse genome and therefore appear to have arisen by duplication since the divergence of rodents and primates; alternatively, there could have been a loss of these genes in the rodent lineage. They are also the best candidates for primate-specific β-defensins as they lack orthologs within all other known mammalian defensins, although our knowledge of mammalian defensins is currently incomplete. As we have shown, the evolution of the DEFB104 mature peptide has been driven by positive selection since its emergence, which is consistent with its novel antimicrobial properties 26. All defensins were thought to exist as monomers stabilized by three disulfide bridges between their three pairs of conserved cysteines 19. However, when compared with other known human β-defensins (DEFB1, DEFB4 and DEFB103), DEFB104 was found to have a different number of residues between its second and third, and between its fourth and fifth cysteine residues. Furthermore, DEFB104 was found to have bactericidal activity against Pseudomonas aeruginosa that was more than sixfold stronger than for any other β-defensin 26. Three of the novel genes described here (DEFB105, DEFB106 and DEFB108) encode mature peptides exhibiting the same spacing of conserved cysteine residues as seen in DEFB104 as well as sharing similar expression patterns, with highest expression in the testes. The functional divergence of β-defensin genes appears to have continued following human-baboon divergence, as exemplified by DEFB107 which displays a serine residue instead of the first canonical cysteine seen in the baboon ortholog. It is notable that a novel mouse β-defensin gene (Defr1), which also lacks the first canonical cysteine, has potent antimicrobial activity against a spectrum of pathogens 27. In addition a polymorphism in the DEFB1 gene which alters the first canonical cysteine to a serine residue has been shown to produce a peptide which is as active against the microorganisms tested as the usual form 28. The unusual amino-acid composition of the proteins encoded by the novel genes presented here suggests that they may possess novel functions, indeed the cationic nature of β-defensins is lost in DEFB107 and DEFB108. As shown recently, there may be more subtle consequences of variation in β-defensin protein sequences, affecting dimerization as well as net charge and disulfide bridges 29. It is worth noting that recent research has identified additional functions for β-defensins that link the innate and adaptive immune response. Both human and mouse β-defensins have been shown to be chemotactic for immature dendritic cells and memory T cells via the CCR6 chemokine receptor 30. The mouse β-defensin Defb2 has been shown to act directly on immature dendritic cells as an endogenous ligand for Toll-like receptor 4 (TLR-4), inducing upregulation of co-stimulatory molecules and dendritic cell maturation. These events, in turn, trigger robust, type-1 polarized adaptive immune responses in vivo, which suggests that β-defensins may have an important role in immunosurveillance against pathogens 31.

The expression of the novel antimicrobial peptides reported here in the human male reproductive tract is also of interest. Recent work has shown that the male urogenital tracts in mammals express a broad range of α- and β-defensins, with human DEFB1 expressed in testicular biopsies, seminal plasma and ejaculated spermatozoa 32. Together with our results demonstrating that DEFB105, DEFB106, DEFB107 and DEFB108 are predominantly expressed in the testes, these data suggest that the male reproductive tract has a complex innate defense mechanism.

The 8p22-23 β-defensin locus has evolved by duplication and subsequent divergence, to produce a diverse cluster of paralogous genes defined by four novel genes (DEFB105, DEFB106, DEFB107, DEFB108), a novel pseudogene (DEFB109p), and three known genes (DEFB4, DEFB103 and DEFB104). We present full-length sequences for the four novel genes, their expression patterns and the predicted sequences for the baboon orthologs of six genes from this cluster. Although comparisons among first-exon sequences of these human genes show little variation, the second-exon sequences that encode the mature β-defensin peptides show substantial divergence. Evolutionary analyses suggest that the divergence seen in second exons has involved positive selection disproportionately favoring alterations in the charge of amino-acid residues.

The following files are available with this article: the DNA sequences for olive baboon β-defensins (Additional data file 1), the DNA sequences for human β-defensins (Additional data file 2), the aligned baboon and human DNA sequences (Additional data file 3), the aligned first-exon baboon and human DNA sequences (Additional data file 4), the aligned second-exon baboon and human DNA sequences (Additional data file 5) and the DNA sequences for baboon and human introns (Additional data file 6) and an Excel file listing the full dN, dS and dN - dS estimates for all first and second exon comparisons shown in Figures 4a and b (Additional data file 7). The alignments are in MSF format. The sequence and alignment files are also available from 46.