A worldwide correlation of lactase persistence phenotype and genotypes

An estimated 65% of human adults (and most adult mammals) downregulate the production of intestinal lactase after weaning. Lactase is necessary for the digestion of lactose, the main carbohydrate in milk [1], and without it, milk consumption can lead to bloating, flatulence, cramps and nausea [2]. Continued production of lactase throughout adult life (lactase persistence, LP) is a genetically determined trait and is found at moderate to high frequencies in Europeans and some African, Middle Eastern and Southern Asian populations (see Additional File 1 and Figure 1).

The most frequently used non-invasive methods for identifying the presence of intestinal lactase are based upon detecting digestion products of lactose produced by the subject (Blood Glucose, BG) or gut bacteria (Breath Hydrogen, BH). For both methods a lactose load is administered to the subject following an overnight fast. In individuals producing lactase this leads to a detectable increase in blood glucose. In individuals who are not producing lactase, the undigested lactose will pass into the colon where it is fermented by various gut bacteria, producing fatty acids and various gases, particularly hydrogen. Hydrogen passes through the blood into the lungs and so can be detected in the breath using a portable hydrogen analyser. Both the BG and the BH tests have asymmetric type I and type II error rates. Thus any study seeking association between a particular polymorphism and LP should take these error rates into account. In addition it should be noted that while in most cases the presence/absence of intestinal lactase in an adult is likely to be genetically determined, the loss of lactase can also be caused by gut trauma such as gastroenteritis [3–6]. Other non-invasive methods for detecting the presence/absence of lactase include assaying for urine galactose and detecting metabolites of Carbon-14-labelled lactose. These methods are rarely used today. The most reliable method is intestinal biopsy, which provides a direct determination of intestinal lactase activity. However, this procedure is very rarely used for diagnosing healthy individuals because of its invasive nature [7].

With the recent discovery of nucleotide changes associated with LP comes the prospect of direct genetic tests for the trait [8–10]. However, it has become clear that there are multiple, independently derived LP-associated alleles with different geographical distributions [1, 8, 11, 12]. LP is particularly common in Europe and certain African and Middle Eastern groups. As a consequence these are the regions where most genetic studies have been focused and all currently known LP alleles have been identified [7, 11, 12]. The first allelic variant that was shown to be strongly associated with increased lactase activity is a C>T change 13,910 bases upstream of the LCT gene in the 13^th intron of the MCM6 gene [13]. Functional studies have indicated that this change may affect lactase gene promoter activity and increase the production of lactase-phlorizin hydrolase mRNA in the intestinal mucosa [14–17] but, as with all LP-associated variants, there remains the possibility that linkage to as yet unknown causative nucleotide changes may explain observed associations. Haplotype length conservation [18], linked microsatellite variation [19] and ancient DNA analysis from early European farmers [20] later confirmed that this allele has a recent evolutionary origin and had been the subject of strong positive natural selection. Furthermore, a simulation model of the origins and evolution of lactase persistence and dairying in Europe has inferred that natural selection started to act on an initially small number of lactase persistent dairyers around 7,500 BP in a region between Central Europe and the northern Balkans, possibly in association with the Linearbandkeramik culture [21]. Another simulation study has inferred that it is likely that lactase persistence selective advantage was not constant over Europe, and that demography was a significant element in the evolution and spread of European lactase persistence [22].

However, the presence of this allele could not explain the frequency of LP in most African populations [8]. Further studies identified three additional variants that are strongly associated with LP in some African and Middle Eastern populations and/or have evidence of function, all are upstream of the LCT gene in the 13^th intron of the MCM6 gene: -13,907*G, -13,915*G and -14,010*C [11, 12, 23, 24]. Where data were sufficient, some of these alleles also showed genetic signatures of a recent origin and strong positive natural selection [12, 23].

Although at least four strong candidate causative alleles have been identified, only a small number of populations have been studied, and those are confined to Europe, Africa and the Middle East. It is therefore unlikely that all LP-associated or LP-causing alleles are currently known. As a consequence, genetic tests based on current knowledge would underestimate the frequency of LP in most world populations. As part of the first study to seek a genetic explanation for the distribution of LP in Africa [8], a statistical procedure (GenoPheno) was developed to test if the frequency of an LP-associated allele could explain reported LP frequency in ethnically matched populations. Crucially, this statistical procedure was designed to account for sampling errors and the asymmetric type I and type II error rates associated with different phenotype tests (BH and BG).

In this study we have sought to extend this approach to the whole of the Old World. However, while there is a rich literature on the frequencies of LP in different geographic regions [1] and a growing body of publications reporting the frequencies of candidate LP-causing alleles, in most cases the genetic and phenotypic data are not from the same people and often not of closely neighbouring groups. Thus, characterization of the extent to which LP frequency can be explained by current knowledge of LP-associated genotype frequencies is limited to populations where both data types are available. To overcome this problem we performed surface interpolation of various data categories (genetic, phenotypic, sample numbers, phenotype tests used and their associated error rates) and applied the statistical procedures described on a fine grid covering the Old World landmass. This has allowed us to identify regions where reported LP-associated allele frequencies are insufficient to explain the presence of LP. These regions should be good candidates for future genotype/phenotype studies.

In this study we have identified regions where the current data on LP-associated allele frequencies is insufficient to explain the estimated LP phenotype frequencies, by surface interpolating LP genotype and phenotype data. Our analyses also indicate regions where genotypic or phenotypic data is sparse or non-existent (see also the maps of Additional Files 3, 4, and 5). Data collection from these regions is likely to be of value in developing a fuller understanding of the distribution and evolution of LP. We suggest that regions where LP-associated genotypes are under-predicting LP are good candidates for further genetic studies.

We accept that surface interpolation can give misleading results when data are sparse and so urge caution in interpreting our results for such regions. The data used and maps generated will be regularly updated via the GLAD website http://www.ucl.ac.uk/mace-lab/GLAD/. We therefore expect that problems caused by interpolation over regions with sparse data will diminish as more data become available. However, to indicate regions where sampling is sparse we have generated three extra maps (Additional Files 3, 4, and 5) showing the density of sample sites for phenotypic data, for sites where -13,910*T allele data is available, and for sites where data on all 4 LP-associated alleles is available, respectively. These maps were generated using 2-dimentional-kernel density estimation [34–36] with a kernel bandwidth equal to half of the average nearest neighbour distance between sample sites.

While on a broad scale most regions of the Old World have been sampled for the -13,910*T allele, data on frequencies of the other three LP-associated alleles is localised mainly to Africa and the Middle East. It is likely that further studies will identify appreciable frequencies of the -13,907*G, -13,915*G or -14,010*C alleles, or reveal new LP-associated alleles, in other regions. To illustrate how data on the three non -13,910*T alleles have added to our knowledge of the genetics of LP, we have generated two additional maps: one is showing the correlation between the LP phenotype and the LP phenotype predicted by data where all 4 LP-associated alleles were sequenced (Additional File 6), and the second is showing the correlation between the LP phenotype and the LP phenotype predicted by data on the -13,910*T allele alone (Additional File 7). We then subtracted the surfaces of these two figures (Additional File 8).

Our analysis indicated a few regions (including Arabia and the Basque region) where the LP-associated allele frequency appears to over-predict LP phenotype frequency. If we assume that all four LP-associated alleles considered here are causative of the trait, or very tightly linked to causative variants, then it is likely that over-prediction is a result of population sampling problems. For example, the pastoralist Bedouin in Saudi Arabia have high frequencies of LP, while non-Bedouin Arabs from the same region typically have lower frequencies [37]. Alternatively, over-prediction may to an extent arise through post weaning non-genetic causes of LP, such as secondary lactose intolerance caused by gut trauma [3–6]. To an extent these problems of matching population groups from the same geographic regions applies to our whole analysis. However, it is notable from Figure 5 that where a lack of correspondence between LP phenotype and predicted phenotype frequencies occurs, it is usually under-prediction (maximum = 0.94; for West Africa), while over-prediction is rare and considerably smaller in scale (maximum = 0.29; for the Basque region). In some cases it was possible to identify genotype and phenotype data originating from the same populations or ethnic groups. For such populations we also performed a separate GenoPheno analysis (see Table 1, [8]). We note that the large discrepancies between LP phenotype frequencies and those predicted from allele frequency data, from this analysis of 'matched' populations, mostly occur in the same regions as where large discrepancies are indicated from the interpolation analysis (see Figure 5). This illustrates that - at a broad scale - the interpolation method that we have employed provides a reasonable approximation of where genotypic data are unable to explain the observed frequency of the LP phenotype, despite much of the data we use originating from different ethnic groups with different subsistence strategies [38, 39].

By applying the GenoPheno statistical procedure to interpolated layers of phenotype and genotype associated data (Figure 6), we have identified west and parts of east Africa, eastern Europe, and parts of western, central, and southern Asia as potential targets for further genetic studies. An absence of data for the -13,907*G, -13,915*G and -14,010*C alleles in many of these regions may partly explain under-prediction (Figure 5 and Additional Files 6, 7, and 8). Previous studies have noted the possibility of under-prediction in eastern Europe and proposed the presence of alleles other than -13,910*T [40, 41]. The population sampling problems described above may also explain the under-prediction we infer in parts of southern Asia, as in each of these regions, the locations where phenotype and genotype data were obtained are mostly well separated. This population data-matching problem is, however, unlikely to explain the lack of correspondence between LP and allele frequency-based predicted LP frequencies in the region around Pakistan and Afghanistan, as well as in west Africa and Italy. Further genetic studies in these regions should prove informative.

In this study we have demonstrated that lactase persistence genotype data is currently insufficient to explain lactase persistence phenotype frequency in western and eastern Africa and several other Old World regions. The identification of additional LP-associated or LP-causative alleles, especially in these regions, will help not only in developing a better understanding of the evolution of LP but also in elucidating the physiological mechanisms that underlie the trait. The interpolation and mapping approach that we have applied in this study may also be of value in studying the underlying genetic basis and evolution of other phenotypic variation that impacts on human health, such as the distribution of functional variation in drug metabolising enzymes [42].