*Correspondence: Einar Arnason, Institute of Biology, University of Iceland, Grensasvegur 12, 108 Reykjavik, Iceland. Phone: (354)-525-4613; Fax: (354)-525-4069. E-mail: email@example.com
Recently statements have been made about a special ‘genetic homogeneity’ of the Icelanders that are at variance with earlier work on blood groups and allozymes. To validate these claims an extensive reanalysis was undertaken of mtDNA variation by examining primary data from original sources on 26 European populations. The results show that Icelanders are among the most genetically heterogeneous Europeans by the mean number of nucleotide differences as well as by estimates of θ parameters of the neutral theory. The distribution of pairwise differences in general has the same shape as European populations and shows no evidence of bottlenecks of numbers in Iceland. The allelic frequency distribution of Iceland is relatively even with a large number of haplotypes at polymorphic frequencies contrasting with other countries. This is a signature of admixture during the founding or history of Iceland. Assumptions of models used to simulate number of haplotypes at sampling saturation for comparing populations are violated to different degrees by various countries. Anomalies identified in data in previous reports on Icelandic mtDNA variation appear to be due to errors in publicly accessible databases. This study demonstrates the importance of basing analyses on primary data so that errors are not propagated. Claims about special genetic homogeneity of Icelanders are not supported by evidence.
Recently claims have been made about the ‘genetic homogeneity’ of the Icelandic population. Iceland is said to be an ‘island so inbred that it is a happy genetic hunting ground’ (Ridley, 2000) inhabited by ‘a nearly homogeneous population’…‘carrying nearly the same genetic codes as the Viking explorers who settled here more than 1,100 years ago’ (Binyon, 1999) such that ‘nowhere else has such a pure—and predictable—genetic inheritance’ (Binyon, 1999). Although these statements about genetics are made in the popular press they derive from a scientific conjecture that, indeed, Icelanders are ‘genetically homogeneous’ (Gulcher & Stefansson, 1998; Gulcher et al. 2000). Variation of mitochondrial DNA (mtDNA) provides the supporting evidence that Helgason et al. (2000b) use to argue from simulations of the number of haplotypes expected at ‘sampling saturation’ (Helgason et al. 2000a) that the ‘Icelandic mtDNA gene pool contains relatively fewer distinct lineages than most other European populations.’ This implies a small ‘effective population size’ such that ‘founder effects and genetic drift may have had a considerable influence on the Icelandic gene pool’ (Helgason et al. 2000b).
These claims and interpretations are at variance with earlier work based on blood groups and allozymes (e.g. Bjarnason et al. 1973; Thompson, 1973) and with more recent results of mtDNA variation (Árnason et al. 2000). In order to validate these claims I undertook this study, which represents a larger, wider and more extensive analysis than previous studies and so is likely to reveal new patterns. I discuss the pros and cons of several measures of genetic diversity and the assumptions behind the measures because they can make a difference to the interpretation of results. Recently other workers have discussed how errors can enter into datasets (Röhl et al. 2001; Bandelt et al. 2001). Therefore I examined agreement among data sets in order to understand how errors can creep into datasets. With a better understanding of errors it is possible to guard against them so that they are not compounded by later studies. Anomalies identified were addressed by contacting people who maintain database sites. I have also reanalysed the primary mtDNA data on Icelanders and examined and reanalysed the primary data from the original sources on a large pool of data from other populations that are useful for comparison with Iceland. I present the outcome of this analysis and discuss the implications for future studies of Iceland and other countries. I also examine evidence from other genetic systems that have a bearing on the question of Icelandic homogeneity. This study highlights the importance of considering which genetic diversity measure is appropriate and illustrates the importance of performing analyses on primary data so that previous errors are not compounded.
Measures of Heterogeneity and Model Assumptions
Three measures are relevant for assessing homogeneity or heterogeneity of populations from mtDNA sequence variation (Nei, 1987). The mean number of pairwise sequence differences, , which take into account the frequencies of different haplotypes, p, and the mutational differences between them, Πij, is a straightforward measure of diversity. Other estimators are the number of different haplotypes or lineages k and the number of segregating sites S in a sample of size n. Using assumptions of the infinite-alleles or the infinite-sites models of the neutral theory the parameter θ, representing the effective population size Ne scaled by the mutation rate appropriate for the fragment μ, can be estimated from these basic statistics. The statistics and only depend on the sample size and the respective basic statistics, k and S (Ewens, 1972; Nei, 1987).
The mean number of pairwise sequence differences, , is high in Iceland (Sajantila et al. 1995; Richards et al. 1996; Helgason et al. 2000b; Árnason et al. 2000). However, positing that this high diversity is due to the female ancestors of Icelanders carrying mtDNA sequences which ‘differed by a relatively large number of bases,’ and that primarily provides information about demographic history older than that of Iceland (Gulcher et al. 2000; Helgason et al. 2000b), Helgason et al. (2000b) prefer the number of haplotypes k and the accompanying statistic derived from the infinite alleles-model of the neutral theory, that only depends on the relationship of the sample size and the number of distinct haplotypes observed in a sample (Ewens, 1972). The contention is that k is a sensitive indicator of demographic events such as a recent bottleneck because rare haplotypes would then tend to be lost. A shortage of rare haplotypes could then be taken as evidence for a bottleneck or drift. Thus a low k and in Iceland are thought to be sensitive indicators of bottlenecks and recent demographic history of the Icelanders (Helgason et al. 2000b). The higher the number of haplotypes relative to the sample size the higher the . However, the number of haplotypes increases at a rate of diminishing returns with increasing sample size. Therefore, the various countries are not directly comparable because they differ in sample sizes. To circumvent this the number of haplotypes expected if sample sizes were increased can be simulated based on the infinite-alleles model to arrive at the expected number of haplotypes at ‘sampling saturation’ (Helgason et al. 2000b). This is the number of haplotypes at which a further increase in haplotype numbers would be less than one for an incremental sample size increase of ten (Helgason et al. 2000b). This in effect represents an attempt to estimate by simulation the parameter for the number of haplotypes in a population using the observed and plugging this into the Ewens sampling formula of the infinite alleles model in reverse. The inference, however, is critically dependent on the model assumptions.
Helgason et al. (2000b) notice numerous homoplasies which violate mutational assumptions that every mutation generates a new allele (Ewens, 1972). They claim, however, that ‘violations of the assumptions … apply equally to different populations’ and, therefore, that the comparisons are valid. This is debatable. However, if this assumption is granted there remain two major assumptions that must be examined for this to be workable, namely, the assumption of selective neutrality and the assumption of steady state distribution. Violations of these assumptions may invalidate inferences based on or and simulated haplotype number at sampling saturation.
Under selective neutrality the infinite-alleles model predicts a steady state distribution of allelic classes in a population. Even though identities of alleles are changing through mutation and there is an evolutionary turnover of alleles due to random genetic drift, the distribution of allelic classes ranked by frequency remains stable at a steady state determined by the effective population size and the mutation rate (Ewens, 1972). If a population is at a steady state its effective population size can be estimated by Ewens (1972) formula and the expected numbers of haplotypes for a given sample size can be simulated. If populations that obey the steady-state differ in sample size their expected number of haplotypes at a sample size of saturation can be simulated facilitating comparisons. If the assumptions are violated to different extents by different populations the simulations simply restate the violations of assumptions and comparisons are not valid.
The and statistics estimate 2Neμ, the effective population size scaled by the mutation rate, but they utilise different aspects of the data. The Tajima (1989) test statistic for neutrality
provides a useful framework for analysing the evolutionary forces which give rise to violations of the above assumptions. The rationale of the test is that both the mean number of pairwise differences, , and the number of segregating sites, S, scaled by a sample size factor , are estimators of the parameter θ= 2Neμ. Since S (as well as the number of haplotypes k used for estimating ) ignores the frequencies of the alleles it is highly influenced by the existence of deleterious alleles which purifying selection keeps at a low frequency. In contrast, takes account of the frequencies of alleles and is therefore little affected by the presence of rare alleles in a sample which contribute little because they are rare. On the other hand alleles present at intermediate frequencies can increase considerably. Thus because it takes account of frequencies of haplotypes is sensitive to forces changing frequencies. If the sample includes some deleterious alleles D will tend to be negative due to purifying selection. Balancing selection will have the opposite effect, possibly maintaining several alleles at polymorphic frequencies, thus pushing D towards positive values. Admixture, the mixing of populations which differ in allele frequencies, may bring many alleles to high and even frequencies, mimicking the effects of balancing selection. Recent admixture can thus be a powerful force influencing .
Norway is one of the countries useful for comparison with Iceland for historical reasons. Frequently included in comparisons (e.g. Helgason et al. 2000b) is a Norwegian sample from a paper that bears the informative title: ‘Increased number of substitutions in the D-loop of mitochondrial DNA in the Sudden Infant Death Syndrome’ (SIDS; Opdal et al. 1998). The Norwegian sample has the lowest Tajima's D (Figure 1), significantly different from neutrality, implicating purifying selection in this sample over and above a potential demographic signature, with the most frequent allele being too frequent, the medium frequency alleles being too rare, and with a large number of singletons (c.f. Keith et al. 1985; Clark, 1987). SIDS may be implicated in mitochondrial disease (Opdal et al. 1998) and the Norwegian sample, therefore, is an example of comparative data violating the assumption of neutrality (Excoffier, 1990) in a direction expected under purifying selection. A comparison based on number of haplotypes failing to take notice of the apparent purifying selection in this sample might infer, incorrectly, a relatively large effective population size θk in Norway. The SIDS sample had a larger number of haplotypes relative to sample size than the control sample (Opdal et al. 1998) and therefore also a higher . By the arguments of sampling saturation this would be taken to mean that the ‘population’ of Norwegian SIDS individuals have a higher effective population size than Norway as a whole, which is obviously false. Thus failure to take notice of the apparent purifying selection in the sample would lead to incorrect conclusions.
A comparison of Iceland and Germany having large sample sizes (n= 520 and n= 423 respectively), reveals that Iceland has a more intermediate frequency distribution of haplotypes than Germany (Figure 1). The Icelandic sample has three haplotypes polymorphic by the 5% criterion, 22 polymorphic by the 1% criterion, 38 rare, and 65 singletons, whereas the German sample has one, nine, 37, and 170 respectively. Singleton haplotypes, which add most to the number of haplotypes, only start counting in Iceland at a cumulative frequency of 88% whereas they start at 60% in Germany. Thus the large number of polymorphic haplotypes in Iceland exhaust a considerable part of the frequency spectrum. Singletons are not found as readily there as for example in Germany (or Norway). The distributions are significantly different for these two countries (jackknife Kolmogorov-Smirnov test, P < 0.001). Germany has a higher homozygosity and both countries deviate significantly from the Ewens sampling distribution by both the exact test and the homozygosity test (Slatkin, 1994, 1996). These serve as examples of violations of the steady state assumptions. As expected from this, Iceland has a higher Tajima's D in the direction expected under balancing selection or admixture (Clark, 1987; Hedrick & Thomson, 1983; Markow et al. 1993) relative to, for example, Germany and Norway. In this instance the effect is likely to be due to admixture which will mimic the effects of balancing selection on the frequency distribution and hence on Tajima's D, rather than being due to balancing selection.
The assumptions of steady state and neutrality used in simulations of sampling saturation (Helgason et al. 2000a,b) likely do not hold for the various countries or are violated to different degrees by the different countries compared (e.g. Helgason et al. 2000b). Simulations of the expected number of haplotypes at a sampling saturation based on observed sample size and the estimated only restate the differences embodied in the violations of the model assumptions by the data available for the various countries.
There also is no evidence that Icelandic mtDNA haplotypes differ by a ‘large number of bases’ (Gulcher et al. 2000) relative to other countries. The maximum number of pairwise differences is 13 for both Iceland and Germany (boxplots in Figure 1) while the medians are 4 and 3 respectively. The high nucleotide diversity in Iceland therefore is due to the high frequency of intermediate differences, which results from Iceland having many more alleles at polymorphic frequencies than Germany. It is not due to a presence in Iceland of widely divergent haplotypes differing by a larger number of bases than haplotypes found in Germany or the other countries.
The interpretation that there are fewer haplotypes in Iceland than in other countries due to a founder effect and genetic drift with a small effective female population size (Helgason et al. 2000b; Gulcher et al. 2000) is therefore unlikely. Although rare alleles may be lost from a population under a founder event and drift, such sampling events are not expected to result in an even distribution of polymorphic alleles characteristic of Iceland. If the objective was to find more singletons in Iceland, sampling there would have to be much more intensive with considerably larger sample sizes because so little of the frequency spectrum is available for singletons. The large number of polymorphic alleles in Iceland is thus contrary to expectations of a founder effect and drift. Admixture during or subsequent to the settlement of Iceland is a much more likely explanation.
Quality of Data—Examining Primary Data
Genetic homogeneity or heterogeneity must be evaluated relative to other populations. Twenty-five European and Near-East populations are a useful comparison (e.g. Helgason et al. 2000b) with Iceland. However, data quality and accuracy must be ascertained before comparisons are made by examining the primary data.
Denmark, being a Nordic country and because of its long historical ties with Iceland, is a useful comparison to Iceland. Denmark shares a ‘Viking’ origin with Iceland and various comparisons are possible. For example the actual population size of Denmark has been higher than that of Iceland throughout the centuries. Richards et al. (1996) provide the data on mtDNA variation among 33 Danes reproduced in Table 1. Identical results were obtained from raw sequence data extracted from the EMBL Nucleotide Sequence Database and from the NCBI Genbank. Sequence data on the Danes ascribed to Richards et al. (1996) are also to be found in other publicly available databases. An example are the data in Table 2 obtained in this instance from the HvrBase database (Handt et al. 1998; Meyer et al. 1999). These data refer to the same individuals because both data sets carry the same encrypted individual identifiers (Tables 1 and 2) yet the two sets of sequences differ considerably. Statistics of genetic variation among the sequences belonging to each set summarise the differences between the sets (Table 3). The allelic configuration (Slatkin, 1994) of Richards et al. (1996) primary data from Table 1 is (8 3 3 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1) and the mean number of pairwise differences and both are relatively low. In stark contrast the data in Table 2 bear little similarity to Richards et al. (1996) primary data. The allelic configuration for the data in Table 2 is (3 2 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1) with being very high, second only to the Middle Easterners, and also relatively high (Table 3). The sequence data of Table 2 are clearly in error and do not represent the data reported by Richards et al. (1996). The estimation of statistics and relative rank of countries with respect to diversity and effective population size is materially affected by the errors (Table 3).
Sample excludes individuals GER234 and GER230 that had missing data in some sites. Fragment of the human mtDNA Control Region I between sites 16090 and 16365; site refers to number in reference sequence (Anderson et al. 1981) less 16,000.
The inference that Icelanders are relatively homogeneous genetically is at least in one instance based on a comparison with 25 other European and Near-East populations (Table 1 in Helgason et al. 2000b). In that comparison the summary statistics for Denmark are identical (Table 3) to those obtained using the erroneous data of Table 2.
Several other anomalies were identified in the comparative data used by Helgason et al. (2000b) that also appear to have been propagated in other papers (Helgason et al. 2000a,c). Another example is the anomalous results from Germany that also can be reproduced using erroneous data from the HvrBase database (erroneous results not shown). Other anomalies, such as in the data of Pult et al. (1994) on Switzerland were also addressed by contacting database operators and in some instances authors to confirm that there indeed were errors in sequence data and to correct them and prevent their further propagation. In addition a reanalysis was undertaken by examining the primary data from the original sources for 26 populations (references are given in table 1 in Helgason et al. 2000b). The results of the reanalysis are presented in Table 4 using the various statistics discussed above. The statistical significance of differences among all pairwise comparisons has not been assessed for these various statistics; many would not be significant. Instead the rank order of countries is noted for the three statistics because it suffices for the argument of this paper.
Table 4. Summary statistics of mtDNA variation for comparison of 26 European populations
Based on reanalysis of primary data from original sources. References to studies are given in Table 1 of Helgason et al. (2000b). Table is ordered by , the mean number of pairwise differences. N sample size, k number of haplotypes, S number of segregating sites, haplotype diversity, mean number of pairwise differences, estimate of 2Neμ based on number of haplotypes, estimate of 2Neμ based on number of segregating sites, D Tajima's test statistic.
1 Near East
3 Canary Islanders
Iceland ranked the ninth highest of the 26 countries by the mean number of pairwise differences , next to the French. The statistics estimate the effective population size by assuming a steady state distribution, based on the sample size and the number of haplotypes k and based on the sample size and the number of segregating sites S. Iceland ranked 11th by next to French, Finns, and Canary Islanders and ranked 13th by next to Russians and Canary Islanders. In contrast, after a reanalysis (Tables 3 and 4) of original data Danes for example rank the second lowest among the 26 countries, right above the Saami by the statistic and rank third lowest along with the Welsh and the Basque by . Thus, by these measures the Danes belong in the group of lower diversity nations such as the Saami, Welsh, Swedes and Swiss (Table 4). Thus examining the primary data from the original sources, and avoiding the use of erroneous data such as those exemplified by data in Table 2, Iceland ranks among highly heterogeneous populations in Europe for and in the top half of distributions on all measures of heterogeneity or effective population size. This contrasts with previous inference of ‘relative homogeneity’ of Icelanders based on mtDNA (Helgason et al. 2000b; Gulcher et al. 2000; Helgason et al. 2001).
High and heterogeneous mutation rates may affect inference about demographic events based on these statistics (Aris-Brosou & Excoffier, 1996). A high mtDNA mutation rate possibly could explain the apparent behaviour of the Icelandic data (Helgason et al. 2000b). To investigate this possibility using pedigree analysis Sigur ardóttir et al. (2000) estimated that the overall mutation rate in the two hyper-variable regions is 3/705, or 0.43%. The authors, however, also estimate that their experimental ‘handling- or labelling-error rate’ was 5/285, or 1.8% (Sigur ardóttir et al.2000). This high laboratory error rate makes it difficult to evaluate a mutational signal. Taken at face value and ignoring the error rate, the mutation rate appears too low to explain the patterns of the Icelandic distribution.
Thus variation of mtDNA does not support, and should not have been argued to support, homogeneity of Icelanders. It is, however, possible that this only applies to mtDNA. It is therefore appropriate to inquire whether this conjecture is supported by other evidence.
Evidence from Nuclear Genes and Population Changes
The various loci in the genome are under different selective constraints and differ in heterozygosity due to differences in the number and frequencies of segregating alleles. The technique used for detection also has an effect (e.g. iso-electric focussing or sequential electrophoresis, Keith et al. 1985). Thus the mean is not a robust estimator of genomic heterozygosity for it depends on the choice of particular loci. To further demonstrate this I report the heterozygosity of 15 loci (Table 5) from an extended dataset kindly provided by Dr. Alfred Arnason (personal communication). Part of these data were published by Jensson et al. (1989). Included is the TF locus studied by iso-electric focusing and TF heterozygosity in Iceland is high (Table 5) in contrast with the above results. Icelandic heterozygosity is in general high for these loci and similar to the nine other loci reported in Bjarnason et al. (1973). These 24 loci are commonly used for comparison and Icelandic heterozygosity falls right in with other northwestern European populations (Árnason et al. 2000). The conjecture of a special genetic homogeneity of Icelanders (Gulcher & Stefansson, 1998; Gulcher et al. 2000) is at odds with these earlier studies.
Table 5. Heterozygosity, , and sample sizes, N, of 15 nuclear encoded loci among Icelanders
The results are based on studies of healthy, unrelated Icelanders investigated in connection with paternity cases. The results were kindly provided by Dr. Alfred Arnason.
Studies have also failed to find much population structure within Iceland as shown by a relatively low FST (the standardised variance of allele frequencies) in Iceland when compared to neighbouring countries (Jorde et al. 1982; Williams, 1993). This indication of relatively high gene flow is in line with high mobility and internal migration rates (Jorde et al. 1982; Williams, 1993). It acts as a buffer on the effects of drift by increasing the effective population size of Icelanders.
Microsatellites represent another class of loci potentially useful for addressing these questions. Gulcher & Stefansson (1998) reported that the Icelandic population ‘is indeed more homogeneous as reflected in overall heterozygosity rates of microsatellite loci’ because the heterozygosity rate over 300 Genethon markers was 0.75 in Iceland compared to 0.79 in Europe as a whole. Previously, I (Árnason et al. 2000) pointed out that the Icelandic value was higher than the 0.70 for 5,264 loci spanning the entire genome in the French population (Dib et al. 1996). In reply Gulcher et al. (2000) objected to this comparison and stated that ‘the 298 markers were specifically selected for gene mapping on the basis of high heterozygosity.’ These are contradictory statements, however, for if the loci were specifically selected because of high heterozygosity they cannot give ‘overall heterozygosity’ because that implies a random selection of loci from the genome. Only by disclosing the identity of the loci could the results be independently verified. This remains to be done. The above differences, however, are relatively small and of doubtful significance without standard errors. It is thus unlikely that microsatellites can be used to argue for a special homogeneity of Icelanders.
Iceland was settled starting in 874 and an educated guess of the size of the population is between 20,000 and 70,000 at the end of the settlement in 930 (Thorarinsson, 1961). In 1095 Bishop Gizur Ísleifsson, levying tithes, counted about 4500 tithe paying farmers. From that the population has been estimated between 50,000 and 105,000 (Thorarinsson, 1961). Gulcher & Stefansson (1998) conjecture that there were strong founder effects that were later augmented due to bottlenecks resulting from the plague and from effects of volcanic eruptions. This is debatable.
Even if mortality was high at times the population was never reduced to very low numbers. In 1783 the Laki fissure eruption, an event of world significance (‘there existed a constant fog over all Europe and a great part of North America’Franklin, 1789) and one of the most important such events in Iceland, resulted in death of livestock, widespread famine and heavy mortality among people. The population of 49,500 in 1780 was reduced to 38,000 in 1785. However, the mortality may have hit non-reproducing members of the population the hardest because the population quickly bounced back and 16 years or half a generation later it was back up to 47,240 (Thorarinsson, 1961; Thorsteinsson & Jónsson, 1991). The effects of population reduction on genetic variability in populations depend on how small the population size in the bottleneck is, and the length of the bottleneck or the growth rate of the population following the bottleneck (Nei et al. 1975; Nei, 1987). These conditions were hardly met in Iceland, not even during events resulting from the Laki eruption. It is therefore doubtful that population reductions in Iceland had much effect on genetic variability of the Icelanders. At least the effects have not been detected in the heterozygosities of modern day Icelanders.
Examination of the published literature on blood group and allozyme variation does not provide any support for the notion of special genetic homogeneity of the Icelanders. Further studies of microsatellite variation are unlikely to do so.
A reanalysis of mtDNA variation using primary data from original sources shows that the Icelanders are among the most genetically heterogeneous Europeans by the mean number of nucleotide differences as well as by the estimates of parameters of neutral theory (Nei, 1987). The unimodal distribution of pairwise differences is similar to European populations in general and shows no evidence for a special bottleneck of the Icelandic population during or after its founding 11 centuries ago (Árnason et al. 2000). The allelic frequency distribution of Iceland shows an evenness in frequency with a larger number of haplotypes at polymorphic frequencies than in other countries, for example Germany. This is a signature of admixture during the founding or history of Iceland. Admixture generates random background linkage disequilibria making gene mapping by linkage disequilibria more difficult (Terwilliger & Weiss, 1998; Weiss & Terwilliger, 2000).
The assumptions of models used to estimate θ and simulate expected number of haplotypes at sampling saturation for comparing populations may be violated to different degrees by various countries. Anomalies were identified in data used in previous studies that are in some instances due to errors in publicly accessible databases. Steps were taken to correct them so that the errors are not propagated in future studies. By reanalysis using primary data from original sources the errors were avoided in this study. Claims about a special genetic homogeneity of Icelanders relative to European populations would be suspect to the extent that they depended on anomalous data instead of the primary data. In any case, one would not expect that meaningful patterns about homogeneity, founder effects and drift in different populations could emerge from analyses whose assumptions are violated and using erroneous data.
I thank Dr. Alfred Arnason for providing access to unpublished data and allowing me to publish them here. I also thank two anonymous reviewers for critical comments on the paper. The work was supported by grants from the Icelandic Science Foundation and the University of Iceland Research Fund.