We sequenced the 1,400 bp region immediately upstream of the DEFB1 translation start site (Figure 1) in 83 individuals with different ethnic origins (Yoruba from Nigeria 18 (YRI), Asians (AS), South American Indians (SAI), Australian Aborigines (AUA)); additional data derived from full gene resequencing of 47 subjects (24 African Americans (AA) and 23 European Americans (EA)) were retrieved from the Innate Immunity PGA (IIPGA) web site 19. A total of 27 single nucleotide polymorphism (SNPs) were identified and haplotypes (Additional data file 1) were inferred using PHASE 2021. The analyzed region encompasses all polymorphic variants previously shown to modulate DEFB1 expression levels. As a control for the AA and EA populations, data for 20 promoter regions were retrieved for 20 genes in the IIPGA. In particular, the 2 kb upstream of the translation initiation site of other innate immunity genes genotyped for AA and EA were retrieved only if the initial ATG was located in the first exon (as it is for DEFB1) and if it could be unequivocally identified. Also, promoter regions were discarded if located in recombination hotspots or in resequencing gaps. A total of 20 promoter regions finally constituted the control dataset. Data concerning the number of segregating sites and nucleotide diversity at the DEFB1 promoter region are summarized in Table 1 and indicate that both θ_W 22 and π 23 are definitely higher for DEFB1 compared to maximum values calculated for IIPGA gene promoters.

We excluded that the high degree of polymorphisms at the DEFB1 promoter is due to non-allelic gene conversion with other paralogous defensin genes on chromosome 8 by applying Sawyer's gene conversion algorithm 24.

Under neutral evolution, the amount of within-species diversity is predicted to correlate with levels of between-species divergence, since both depend on the neutral mutation rate 25. The HKA test 26 is commonly used to verify whether this expectation is verified. We performed both pairwise and maximum-likelihood (MLHKA) 27 tests with Rhesus macaque as an outgroup (instead of chimpanzee) so that greater divergence time results in more fixed differences and improves power to detect selection. For pairwise HKA tests we compared polymorphism and divergence level at the promoter region of DEFB1 with the 20 IIPGA genes; we consider these comparisons to be well-suited since lower sequence conservation and faster evolutionary rates are though to be a widespread feature of immune response genes 2829. Since IIPGA data refer to AA and EA, only these populations were used in the comparison; pairwise HKA tests (Table 2) yielded significant results (p < 0.05) in 11 out of 20 cases (with 5 additional tests yielding a p < 0.10), suggesting increased diversity at the DEFB1 promoter compared to most loci. For further confirmation, we performed a MLHKA test by comparing the DEFB1 5' region to all 20 promoter regions: a significant result was obtained (k = 3.31, p = 0.0018).

Another expectation for neutrally evolving genes is that values of θ_W and π are roughly equal; this is the case for the maximum values of innate immunity gene promoters but not for DEFB1, which shows greater π than θ_W, a finding consistent with an excess of intermediate frequency variants as a result of balancing selection 30. The statistics Tajima's D 31 and Fu and Li's D* and F* 32 are commonly used to evaluate the difference between θ_W and π and, therefore, to test departure from neutrality. As shown in Table 1, significantly positive values for the DEFB1 promoter of one or more statistics were obtained for all analyzed populations.

It should be noted that population history, in addition to selective processes, is known 31 to affect frequency spectra and, therefore, all related statistics such as Tajima's D and Fu and Li's D* and F*. In particular, positive values of the statistics are expected under a scenario of population contraction, while negative values are consistent with an increase in population size 3133. We performed all tests under the standard assumption of constant population size, which is unrealistic for human populations. Still, this approach is conservative when applied to African populations since they are thought to have undergone moderate but uninterrupted population expansion 34; in the case of non-African populations the effects of demography are more difficult to disentangle from balancing selection signatures since bottlenecks possibly occurred following migration out of Africa 34. One possibility to circumvent this problem is to exploit the fact that selection acts on a single locus while demography affects the whole genome. As shown in Table 1, Tajima's D, as well as Fu and Li's F* and D*, displays far higher values in the case of DEFB1 compared to the maximum values of innate immunity gene promoters in EA. In order to obtain a more extensive comparison, by including YRI and subjects of Asiatic ancestry we retrieved information concerning 231 genes resequenced in AA, EA, AS and YRI from the NIEHS SNPs Program (NIEHS panel 2) 35. In particular, for each gene a 5 kb region was randomly selected; the only requirement was that it did not contain any long (>500 bp) resequencing gaps, and if the gene did not fulfill this requirement it was discarded, as were 5 kb regions displaying less than six SNPs. The number of analyzed regions for AA, YRI, EA and AS were 209, 203, 177 and 172, respectively. We calculated the percentile rank of DEFB1 values in the distributions of Tajima's D and Fu and Li's F* and D* for this set of loci. In analogy to the results obtained above, values for DEFB1 ranked above the 95th percentile in all populations (except for Tajima's D in YRI, which ranked 93rd). It is worth mentioning that, as already noticed by other authors 36, resequenced genes in SNP discovery programs probably represent a sample biased toward non-neutrally evolving loci (in the case of the NIEHS SNPs Program, genes are selected on the basis of their having a role in organism-environment interactions), making comparison with their distribution a conservative test.

A second possibility to disentangle the effect of demographic history from selection is to apply calibrated population genetics models. In particular, one such model that has been proposed recently, cosi 37, is based on the ability to generate realistic data rather than relying on inference about population histories. We performed coalescent simulations using the cosi package 37 and its best-fit population parameters for YRI, AA, EA and AS. Data are reported in Table 1 and indicate that for Tajima's D, as well as for Fu and Li's D* and F*, application of a calibrated model allows rejection of neutrality for the four populations at the DEFB1 promoter region.

Population genetic differentiation, quantified by F_ST 38, can also be used to detect the signature of balancing selection. In particular, lower F_ST values are expected at loci under balancing selection compared to neutrally evolving ones 3940. F_ST among AA, EA and AS was 0.0057, much lower than the genome average of 0.123 40 and not significantly different from 0 (p = 0.25).

We next wished to verify that the evolution of the DEFB1 promoter is not influenced by the presence of a linked balanced polymorphism within, for example, the gene coding region. We exploited the availability of full resequencing data for the whole gene and calculated human-macaque divergence, nucleotide diversity, Tajima's D and F_ST in sliding windows for AA and EA. As shown in Figure 1, while inter-specific divergence is quite homogeneous along DEFB1, a peak in nucleotide diversity (expecially π) is observed at the promoter; consistently, in both AA and EA, the same region displays the maximum Tajima's D value and the minimum F_ST, with no other region showing evidence suggestive of balancing selection.

It should be noted that several defensin genes on 8p23.1, but not DEFB1, exhibit copy number variation (CNV) in humans 41; a more recent 42 genome-wide analysis of CNVs indicated that the 5' gene region of DEFB1 might be encompassed by a CNV, although the authors indicate that, since the breakpoints are difficult to establish, involved loci might flank rather than be encompassed by the CNVs. The authors studied HapMap subjects and reported a frequency for the CNV ranging from 6% to 14% in different populations. Since our YRI samples comprise a subset of HapMap YRI subjects, we checked whether any of them were reported to display a CNV in this region: two subject were retrieved, accounting for one gain and one loss. Electropherograms of these two subjects (as well as all other subjects in this study) revealed no evidence of unbalanced peaks at heterozygous SNPs and their removal from the sample did not affect the results for YRI. Previous 43 work had studied CNVs in the defensin cluster on chromosome 8 using real-time PCR assays and found that 24 American subjects with different ethnic origin had 2 copies of DEFB1. Taking these observations together, we consider that either DEFB1 lies outside the CNV or, in any case, that CNVs encompassing DEFB1 are very rare and do not affect the results reported here.

One effect of balancing selection is to preserve two or more lineages over an extended period of time, resulting in clades separated by long branch lengths. To examine the genealogy of DEFB1 promoter haplotypes, we built a median-joining network. The topology of this network (Figure 2) is unambiguous with no reticulations, a pattern consistent with the low level of recombination observed in this gene region (not shown). Two major clades (haplogroups 1 and 2) separated by long branch lengths are evident, each containing one common haplotype. We next wished to estimate the time to the most recent common ancestor (TMRCA) of the two haplotype clades, applying a phylogeny-based method 44 based on the measure ρ, the average distance of descendant haplotypes from a specified root. By using root 1 (Figure 2), ρ was equal to 9.45 so that, with a mutation rate based on 21 fixed differences between chimpanzee and humans and a separation time of 5 million years ago, we estimated a TMRCA of 4,489,791 years (standard deviation ±1,018,128).

In order to gain further insight into the evolutionary history of the DEFB1 promoter region, we resequenced those from three chimpanzees and one orangutan. These samples were obtained from the European Collection of Cell Cultures and the Pongo sequence was used in the median-joining network in order to root the phylogeny (Figure 2). A total of 5 polymorphic sites were identified in chimpanzees; one of them (-913 C/T in the human sequence) was shared with humans and, therefore, represents a trans-specific polymorphism. Trans-specific polymorphisms are an effect of long-term balancing selection, while they are highly unlikely under neutrality. Indeed, a neutral polymorphism is expected to persist for 4N_e generations (where N_e is the effective population size, estimated to be around 10,000 for humans) 45 and, therefore, the probability of observing a polymorphism shared between humans and chimpanzees, two species that diverged about 5 million years ago (around 20N_e generations), is extremely low 4647. Although the identification of a human/chimpanzee trans-specific SNP is consistent with the estimated TMRCA of the haplotype clusters (suggesting that balancing selection was established around the same time when the human and Pan lineages split), the possibility exists that the shared SNP is due to a coincidental mutation that occurred after speciation. Indeed, the location of the substitution at a CpG site makes the possibility of a recurrent mutation more likely and, therefore, taking into account the lack of functional data on this SNP, it is difficult to discriminate between the two possibilities.

Human genomic DNA was obtained from the European Collection of Cell Cultures (Ethnic Diversity DNA Panel plus additional samples for Australian Aborigine derived from HLA defined panels). From the same source we obtained the genomic DNA of three chimpanzees (Pan troglodytes) and one orangutan (Pongo pygmaeus). Additional DNA samples from South American Indians and Yoruba individuals were derived from the Coriell Institute for Medical Research.

The 1.4 kb region covering the promoter region of DEFB1 was PCR amplified (primer sequences are reported in Table 3). PCR products were treated with ExoSAP-IT (USB Corporation, Cleveland, OH, USA), directly sequenced on both strands with a Big Dye Terminator sequencing Kit (v3.1 Applied Biosystems, Monza, Italy) and run on an Applied Biosystems ABI 3130 XL Genetic Analyzer. All sequences were assembled using AutoAssembler version 1.4.0 (Applied Biosystems), inspected manually by two distinct operators, and singletons were re-amplified and resequenced.

DEFB1 genotype data for American subjects of either African or European descent were retrieved from the IIPGA website 19. From the same source, we derived resequencing data referring to promoter regions (2 kb upstream of the translation initiation site) of other innate immunity genes genotyped for AA and EA. Promoter regions were not selected if the initial ATG was not located in the first exon (as it is for DEFB1) or if it could not be unequivocally identified due to the presence of multiple 5' isoforms, which were identified through manual inspection of UCSC annotation tracks 83. Also, promoter regions were discarded if located in recombination hotspots (these were manually identified through the UCSC genome annotation tables snpRecombHotspotHapmap and snpRecombHotspotPerlegen 83) or in resequencing gaps. A total of 20 promoter regions finally constituted the control dataset.

Genotype data for 231 resequenced human genes were derived from the NIEHS SNPs Program web site 35. In particular, we selected genes that had been resequenced in populations of defined ethnicity, including Asians (NIEHS panel 2).

Haplotypes were inferred using PHASE version 2.1 2021, a program for reconstructing haplotypes from unrelated genotype data through a Bayesian statistical method. Haplotypes for AS, AUA, SAI and YRI individuals are available as supporting information (Additional data file 1).

Tajima's D 31, Fu and Li's D* and F* 32 statistics, as well as diversity parameters θ_W 22 and π 23 were calculated using libsequence 84, a C++ class library providing an object-oriented framework for the analysis of molecular population genetic data. Departure from neutrality was tested from coalescent simulations computed with ms software 85 fixing the mutation parameter, assuming no intra-locus recombination and a constant population size with 100,000 iterations. Calibrated coalescent simulations were performed using the cosi package 37 and its best-fit parameters for YRI, AA, EA and AS populations with 10,000 iterations. The F_ST statistic 38 estimates genetic differentiation among populations and was calculated as proposed by Hudson et al. 86. Significance was assessed by permuting 10,000 times the haplotype distribution among populations 87.

Pairwise HKA tests were performed using libsequence. The maximum-likelihood-ratio HKA test was performed using the MLHKA software 27 with multilocus data of 20 selected IIPGA promoter regions and Rhesus macaque (NCBI rheMac2) as an outgroup. In particular, we evaluated the likelihood of the model under two different assumptions: that all loci evolved neutrally and that only the DEFB1 promoter region was subjected to natural selection; statistical significance was assessed by a likelihood ratio test. We used a chain length (the number of cycles of the Markov chain) of 500,000 and, as suggested by the authors, we ran the program several times with different seeds to ensure stability of results.

In order to test for gene conversion events, we applied Sawyer's gene conversion algorithm 24 implemented in the GENECONV program. GENECONV assesses significance using two methods: permutations and an approximate p-value 8889. We performed several tests by varying the mismatch penalty from 0 to larger positive values and using 10,000 permutations. For all these runs and both methods, no pairwise or global p-value involving DEFB1 was significant, suggesting no inner or outer fragments showing past gene conversion.

The median-joining network to infer haplotype genealogy was constructed using NETWORK 4.2 44. The time to the most common ancestor (TMRCA) was estimated using a phylogeny based approach implemented in NETWORK 4.2 using a mutation rate based on 21 fixed differences between chimpanzee and humans in the 1.4 kb DEFB1 region.

All calculations were performed in the R environment 90.