DNA samples

The study was carried out according to the Declaration of Helsinki (2000) of the World Medical Association and was approved by the Ethical Review Boards of Kenya, Nigeria, Tanzania and Zimbabwe. Informed consent was obtained from volunteers of the following ethnic groups: Hausa (20), Ibo (20), Luo (30), Maasai (13), San (40), Shona (23), Venda (9), Yoruba (20) and Tanzanian Mixed Bantu (12). Ethnicity was assigned based on the submission that parents and grandparents of the volunteers were of the same self-identified ethnic group. The exact numbers of samples analysed per gene are shown in Tables S1-S4 (Tables 2-5) in the Appendix. DNA was extracted from whole blood samples stored in the AiBST Biobank of African Populations [21] using the QIAamp DNA Blood Mini Kit (Qiagen, KJ Venlo, The Netherlands).

PCR and sequencing

Primers were designed using SNPBox and Primer3 software [22, 23]. Their specificity for each gene studied was confirmed by a BLAST analysis search and comparison of genomic sequences in the National Center for Biotechnology Information (NCBI) databases http://blast.ncbi.nlm.nih.gov/Blast.cgi. Identical primers were used for the polymerase chain reaction (PCR) and sequencing, except where otherwise stated (Table S5 (Table 6)). First-step exon amplification mixtures (20 μl) contained 1x TiTaq buffer, 0.25 mM deoxyribonucleotide triphosphates, 0.5 μM of each primer and 0.25 units of TiTaq and DNA template (5 ng/μl). For PCR, an initial denaturation at 94°C for two minutes was followed by 35 cycles at 94°C for 30 seconds, 59°C for 30 seconds and 72°C for 60 seconds. Sequencing reactions were started at 96°C for one minute followed by 25 cycles at 96°C for ten seconds, 50°C for five seconds and 60°C for four minutes, and resolved on an ABI 3730 DNA Analyzer (Applied Biosystems, Brussels, Belgium).

Data analysis

Identification of SNPs was carried out using the novoSNP v2.1.9 software package [24]. Reference sequences were NC_000010.9 for CYP2C9, NC_000010.9 for CYP2C19, M33388 for CYP2D6 and NC_000008.9 for NAT2. All identified SNPs were compared with the NCBI Single Nucleotide Polymorphism database (dbSNP) [25]. As SNPs can cause the introduction of pre-microRNA (miRNA) sites, this was included as part of the annotation in the novoSNP analysis procedure. Frequencies of SNPs were calculated using Genepop [26].

HaploView v3.31 [27] was used to determine haplotypes from sequence genotype data. Linkage disequilibrium plots were generated to assess the extent to which SNPs were likely to be linked and hence likely to occur on the same haplotype with logarithm of odds (LOD) score > 3. The HaploView Tagger tool was used to estimate which SNPs were likely to be tagged by a single SNP in a predicted haplotype (threshold r² > 0.8).

Prediction of functional effects of non-synonymous SNPs

Functional effects of non-synonymous SNPs were predicted using the Polyphen prediction programme [28] based on position-specific independent counts (PSIC) scores of multiple sequence alignments, as well as structural information, if available. The programme predicts the functional effects of SNPs based on occurrence in active/binding sites or in transmembrane regions, interference with disulphide or other bonds, compatibility with homologous sequences at that position, as well as mapping to known three-dimensional protein structures or validated homology models. Protein sequence accession numbers were obtained from Swiss-Prot [29] as CYP2C9: P11712; CYP2C19: P33261; CYP2D6: P10635; NAT2: P11245.

Allele nomenclature

Allele nomenclature is assigned according to the Human Cytochrome P450 (CYP) Allele Nomenclature Committee [30] and the Arylamine N- acetyltransferase Gene Nomenclature Committee [31]. Alleles, as borne by specific SNPs, are assigned numbers -- for example, CYP2C9*9 to define the 10535A > G mutation, the presence of which results in the amino acid change H251R.

African populations

Certain SNPs or haplotypes that have been reported as prevalent and functionally important in other populations are rare or have not yet been detected in African populations. For example, CYP2C9*2 (R144C) and CYP2C9*3 (I359L), while extensively studied in Asian and Caucasian populations and identified as rare alleles in African Americans (frequency ~1 per cent), were not found in Africans, neither in this study nor in a Beninese population [38]. By contrast, CYP2C9*5 (D360E) and CYP2C9*6 (K273fs) have been identified in African populations, although at low frequency (frequency = 0.01; Table S1 (2)). CYP2C9*5 causes impaired enzyme activity,[6] and CYP2C9*6, first found in African Americans, is associated with phenytoin toxicity [39]. The importance of CYP2C9*8 (R150H), CYP2C9*9 (H251R) and CYP2C9*11 (R335W), which were detected in limited studies in Africans [17] (partly including the present study), and the distribution of poor metabolisers in African populations remain unclear.

The US Food and Drug Administration (FDA) has recommended genotyping for CYP2C9*2 and CYP2C9*3 with the aim of better use of warfarin [40]. Since these variants are practically absent in populations of African origin, their use in current pharmacodiagnostic kits that identify individuals carrying CYP2C9*2 and CYP2C9*3 may not be applicable in these populations. Test kits that detect CYP2C9*5, CYP2C9*6, CYP2C9*8, CYP2C9*9 and CYP2C9*11, as well as our novel SNPs, should be more predictive of the clinical response to CYP2C9 substrate drugs in Africans. Before such tools can be developed and deployed for clinical use, however, further studies are required to establish the frequencies of these alleles in larger African populations, in addition to genotype-phenotype studies to establish their functional relevance.

CYP2C19*2 (splicing defect) and CYP2C19*3 (W212X) have been recommended as biomarkers for the administration of certain CYP2C19 substrates [41]. The CYP2C19 poor-metaboliser phenotype is detected in two to four per cent of Caucasians and in about 20 per cent of Asians, and these two variants account for 99 per cent of these poor metaboliser phenotypes [42, 43]. Whereas CYP2C19*2 was the most frequent known defective variant in our study (frequency = 0.15; Table S2 (Table 3)), we and various genotype-phenotype correlation studies have found CYP2C19*3 to be rare in most African populations (frequency = 0.01; Table S2 (Table 3) [44]). We also identified one individual in the Maasai ethnic group who was heterozygous for this allele, and a few heterozygous individuals have previously been reported in a Tanzanian population [20]. Earlier data show that CYP2C19*2 accounts for over 70 per cent of slow metabolisers of S-mephenytoin [45]. The missing 30 per cent might be made up by CYP2C19*3 and other variants such as CYP2C19*12, CYP2C19*13 and CYP2C19*15, which would make these SNPs important contenders to include in genotyping panels for diagnostic purposes in Africans.

Our analysis of diverse African populations confirmed that CYP2D6*17 (T107I) and CYP2D6*29 (V136M, V338M) remain the most important functional SNPs in the metabolism of CYP2D6 substrate drugs. Together with other less prevalent haplotypes, they explain why African populations generally have a larger number of intermediate metabolisers (~40 per cent) compared with Caucasian populations (~15 per cent) [46].

Based on the highly polymorphic CYP2D6, we used principal component analysis to investigate inter-ethnic variability. The fact that no significant differences were detected across ethnicities (data not shown) could be due to our small sample sizes; however, our data are consistent with a recent study illustrating that CYP2D6 shows a high frequency of altered activity variants but no clear population structure [47]. It may also imply that the phenotype status of those populations is not significantly different either.

It has been speculated that the variation in acetylator (NAT2) status across major world populations reflects differences in dietary habits or the environment. There is a high prevalence of slow and intermediate acetylators in African populations, however, due to the common NAT2*5 (I114T), NAT2*6 (R197Q) and NAT2*14 (R64Q) alleles, which contribute largely to the slow acetylator phenotype. This is consistent with our data (Figure 1) and with a recent study of sub-Saharan populations which also indicates that the NAT2*5B and NAT2*6A haplotypes are more common than the wild-type haplotype NAT2*4 [1, 48].

Enzyme function

Some mutations in coding regions cause amino acid changes that result in alterations of enzyme activity, substrate selectivity and, sometimes, protein stability. Ensuing functional differences cause different metaboliser phenotypes. So far, over 30 such variants have been reported for CYP2C9, approximately 20 for CYP2C19 and over 60 for CYP2D6 [30].

We have predicted functional effects of novel non-synonymous SNPs discovered in this study (Table 1). These predictions were based on amino acid chemistry, conservation in the alignment of known sequences from the same protein families, and solved structures or homology modelling. Crystal structures of CYP2C9 and CYP2D6 have been reported [32, 34] and structures of the other enzymes have been approximated by homology modelling [49, 50]. It is assumed that such approximation is sufficiently accurate to predict functional effects in substrate recognition, binding and catalysis of reactions [51].

Amino acid changes with a PSIC score of less than 1 are assumed not to be involved in any functional sites and are predicted not to affect enzyme function (eg N474S in CYP2C9, V113I in CYP2C19, V119M in CYP2D6, and I158L and I270T in NAT2). Some changes with PSIC scores slightly above 1 may still have modest effects on enzyme function -- for example, R123L in CYP2D6 (PSIC score = 1.236). As shown in a previous study, in which the CYP2D6 sequence was aligned with Gotoh's sequence,[33] however, this residue is involved in the substrate recognition site SRS1 [52]. The T214I change in NAT2 (PSIC score = 1.257) seems to interfere with enzyme function because this residue is important for the interaction with the co-enzyme A ligand, according to homology model prediction.

The effect of the R186P change in CYP2C19 leads to a change in electrostatic charge and possibly geometry; hence, it is predicted to affect the protein dramatically, giving a high PSIC score. The high score observed for G445E in CYP2D6 might be due to its interaction with position 443, which is important for heme-ligand binding,[34] and therefore has a high probability of affecting enzyme function.

Whereas some defective splice site variants are well understood -- for example, CYP2D6*4 (1846G > A), which occurs at the zero acceptor position of exon 4 -- functional indications are less clear if mutations lie further away from splice site junctions. Rogan et al [53]. have used information theory analysis to show how other intronic and synonymous mutations may contribute to splice site effects in CYP genes [53]. For example, the defective allele CYP2C19*2 (19154G > A) results in a synonymous mutation (P227P), yet it has been associated with reduced enzyme activity. Further investigations showed that this mutation introduces a cryptic splice site 40 nucleotides downstream, resulting in a truncated protein. We used information theory to analyse novel synonymous SNPs and intronic SNPs within the splice sites (-25 to +2 for exon acceptor sites and -3 to +6 for exon donor sites) of CYP2C9, CYP2C19 and CYP2D6 but did not find any significant effects on splice site recognition (data not shown).

Pre-miRNA sequences are involved in the regulation of protein expression. Mutations in these sequences, as well as insertions of new pre-miRNA sequences, could affect enzyme expression, yet CYP1B1 is the only CYP that has been found to be miRNA regulated so far [54]. In the present study, we did not find any pre-miRNA sequences introduced in the 3' untranslated region (UTR) regions, yet in CYP2C19, 18818T > C in intron 4 and 19332G > A in intron 5 introduce miRNA binding sites for has-mir-139 and has-mir-448, respectively (Table S2 (Table 3)). Since miRNA binding sites mostly act within the 3' UTR, however, these mutations would not be expected to have any effects.

In summary, our data, in conjunction with other studies of sub-Saharan Africans and African Americans,[17, 19, 55, 56] indicate low heterogeneity in the frequency of functional mutations. In the genes studied, most functionally important SNPs have been found. What remains is to determine their prevalence across populations and to evaluate the functional effects of the novel SNPs. Expressing variant proteins and analysing their substrate turnover to show impaired enzymatic activity was beyond the scope of this study. We envisage that such analyses will strengthen our findings, however, and might become essential for the pharmacokinetic assessment of individual variants in order to meet regulatory requirements for diagnostic use.

Personalised therapy

Our data indicate the importance of CYP2C9, CYP2C19, CYP2D6 and NAT2 for genotype assessment, including the identified novel SNPs, so that optimisation of drug use in African populations can be considered under appropriate clinical scenarios. This could enable correct dose adjustment for individuals who are likely to experience ADRs owing to poor metabolism or an inadequate therapeutic effect owing to ultra-rapid metabolism. It is noteworthy, however, that other factors, which are not related to the newly identified SNPs but affect the clinical pharmacology of prescribed medications, may play a role in clinical ADRs or therapeutic failure.

The incorporation of CYP2C9 genotyping as part of pre-prescription diagnosis for individuals being treated with drugs metabolised by this enzyme [57] indicates the immediate utility of pharmacogenetics. Likewise, pre-prescription genotyping has been recommended for CYP2D6-metabolised drugs with a narrow therapeutic window, such as some antipsychotic agents [58]. NAT2 genotype information can be used to predict the phenotypic status of individuals to enable dose adjustment of anti-tuberculosis drugs such as isoniazid.