Marriage does not relate to major histocompatibility complex: a genetic analysis based on 3691 couples

Abstract

Optimization of chances for healthy offspring is thought to be one of the factors driving mate choice and compatibility of the major histocompatibility complex (MHC) is assumed to determine the offspring's fitness. While humans have been claimed to be able to perceive information of MHC compatibility via the olfactory channel, it remains unknown whether humans use such information for mate choice. By investigation of 3691 married couples, we observed that the high polymorphism of MHC leads to a low chance for homozygous offspring. MHC similarity between couples did not differ from chance, we hence observed no MHC effect in married couples. Hormonal contraception at the time of relationship initiation had no significant effect towards enhanced similarity. A low variety of alleles within a postcode area led to a higher likelihood of homozygous offspring. Based on this data, we conclude that there is no pattern of MHC dis-assortative mating in a genetically diverse Western society. We discuss the question of olfactory mate preference, in-group mating bias and the high polymorphism as potential explanations.

1. Introduction

It is widely thought that mate choice evolves in an adaptive way, such as to optimize the chances for reproductive success and healthy offspring. One factor contributing to the offspring's health is the genetic predisposition for immunological defence. The key part of the immune system is encoded in the major histocompatibility complex (MHC) or—as called specifically in humans—the human leucocyte antigen (HLA) complex. This gene complex contains highly polymorphic loci [1] and contributes to adaptive biological functioning in two ways: first, the MHC is a critical part of the immune response as it binds peptides from different cellular compartments and presents those peptides to T-cells [2]. Presentation of peptides from various intracellular pathogens is carried out by MHC class I molecules, whereas MHC class II molecules are responsible for the presentation of peptides derived from extracellular pathogens, such as extracellular bacteria or eukaryotic parasites. Nevertheless, atypical patterns can be observed within both molecule classes, which is then referred to as cross-presentation (for an overview see [2,3]). Second, MHC molecules were found to be perceived by various species such as fishes [4], mice [5,6] and by humans [7] either by vomeronasal or olfactory neurons [8]. MHC-dependent peptides are observed at low concentrations in the urine of mice [9,10] and rats [11], and it has been suggested that individual MHC profiles are represented in human body sweat as well [12]. MHC heterozygosity enhances resistance to a broader pathogen spectrum and is hence beneficial for offspring [13]. As MHC genes are expressed co-dominantly, it is thought to be advantageous to mate with MHC dissimilar (i.e. different from oneself) and diverse (i.e. heterozygous) individuals. Indeed, individuals of various species have been found to prefer MHC dissimilar and diverse mates (for an overview see [14]).

In humans, it has been debated over the last 25 years whether the same mechanisms are in place and whether the MHC complex contributes to body odour preferences (for a review see [15]). In the first study on MHC and olfactory attraction conducted in humans, Wedekind et al. [16] reported that women rate body odours of MHC dissimilar men as more pleasant than odours of MHC similar men. This effect was limited to women who did not use oral contraception. Women using oral contraception, in contrast, preferred MHC similar male body odours. In the following study [17], the researchers conducted similar experiments but could only replicate trends regarding their previous findings: women without using oral contraception and men tended to rate MHC dissimilar odours as more pleasant. The first finding, however, raised attention towards olfactory mate choice, and subsequently, many studies involving 40 to 197 participants aimed to further clarify this effect.

Jacob et al. [18] found that women without using oral contraception preferred body odours of a moderate dissimilar range over body odours of higher or lower similarity. Santos et al. [19] could not reveal an effect of MHC similarity on pleasantness ratings using different olfactory experimental procedures (male versus female, sweat versus urine odour samples). Equally, studies testing men showed no effect of MHC similarity on olfactory ratings [20,21]. Another study reported mixed results—no effect of MHC on female odour preference, but on the male preference of MHC dissimilar female odours [22].

Roberts et al. [23] conducted a similar experiment to Wedekind et al.[16], but failed to replicate the initial findings. The researchers did not observe a significant relation between women's pleasantness ratings of male odours and MHC similarity. However, the study generated further hypotheses; after starting to use oral contraception, women in this study preferred more similar MHC odours compared to the time before using oral contraception. This result may point towards an impairment of adaptive mate preferences by means of pill use, similar to Wedekind et al. [16]. In addition, the ratings related to the relationship status of the women: women in relationships preferred MHC dissimilar male odours while single women favoured MHC similar odours. Our own results recently seemed to support the finding of Wedekind et al. [16] and showed that women rated male body odours as less attractive, when they shared at least one HLA-B allele and one HLA-C allele with the male subject. This effect was only present in women not using oral contraception and only towards odour probes that were similar in both loci [21]. However, it has to be considered that the analyses focused on different loci (-B, -C) than did Wedekind et al. [16] (-A, -B, -DR).

Taken together, the idea that MHC modulates olfactory partner attraction—the odour of an opposite-sex individual is on average rated more attractive when the individual is MHC dissimilar—is still under debate. A recently published meta-analysis [24], aggregating all currently available experimental studies, revealed no overall effect of MHC on olfactory preferences.

Nevertheless, the idea that MHC modulates mate choice is still not off the table. It is assumed, that the MHC profile might influence relationship characteristics, such as sexual satisfaction [25], rather than actual choice, although those findings have to be interpreted carefully [24]. In line with this idea, it has been demonstrated that higher MHC similarity predicts lower sexual responsiveness and higher attraction to extra-pair sexual partners [26].

Following the original idea, it has been expected that human mate choice is affected by MHC similarity and MHC dissimilar partners are preferred. Studies investigating MHC similarity in couples show, however, mixed results. A very similar pattern of MHC similarity in real couples as in random couples was observed in most studies [27–33]. An enhanced dissimilarity was observed for a genetically very homogeneous subgroup of 411 couples of Hutteries in North American [34], but not for a group of 194 couples of South Amerindian tribes [28]. Enhanced dissimilarity was also reported in a much smaller study for European couples but not for African ones [35]. A recent genome-wide study involving 883 couples from Europe and the Middle East furthermore showed MHC dissimilar mating in some but not all populations [36]. The largest studies performed so far, including 1017 to 1310 couples, indicated the opposite effect: couples differed in less MHC alleles than expected by random pairing, hence mate choice was skewed towards MHC similar partners [37,38]. This observation is supported by a genome-wide analysis revealing that 930 couples of European ancestry are genetically more similar than expected by chance [39]. Across all studies with human subjects, a recent meta-analysis found no major effect of MHC dissimilar mate choice [40]. Firstly, some variance in the study outcomes is probably explained by the different populations that were examined. Homogeneous populations like the Hutteries have a lower MHC diversity than populations with very heterogeneous members which may lead to an enhanced preference for MHC diversity. Secondly, cultural influences may be stronger in married than in unmarried couples. Finally, investigations differ in the selection of included MHC loci or classes and many studies do not control for confounders such as hormonal contraceptive use (for a review, see [15]).

In order to clarify the effects of MHC on human mating, we examined 3691 married couples, which were genotyped for the HLA class I loci A, B and C and the HLA class II loci DRB1, DQB1 and DPB1. We tested the following hypotheses: (i) married couples share less MHC alleles (higher MHC dissimilarity) than randomly assigned couples (by using permutation tests); (ii) the married couples, where the wife did not use oral contraception during the relationship formation, share less MHC alleles (higher MHC dissimilarity) than the couples where the wife used oral contraception at the time of relationship formation; and (iii) the variability of alleles in a geographical region is correlated to the variability between married couples. In addition, we explored the variety of alleles per locus in order to estimate the likelihood of marrying a MHC similar person.

2. Methods

(a). Sample

Potential haematopoietic stem cell donors registered with DKMS German Bone Marrow Donor Center are genotyped at recruitment for HLA-A, -B, -C, -DRB1, -DQB1, -DPB1. However, relations between individuals are not recorded. In order to compose a sample of couples, female donors have been contacted by a letter. Owing to potential language barriers and also to MHC variations, we restricted our sample to donors who self-identified as being of the German ethnic background. In this letter, they received a description of the study reporting that we aim to assess MHC similarity between married couples and invited them to participate if both inclusion criteria were met: (i) they were married and (ii) their husband was also registered with DKMS. We also asked a single question, namely ‘Did you use oral contraception when you started to date your partner? [yes, no]’. The procedure was reviewed and approved by the ethical board of the DKMS life science laboratory.

In total, 3691 couples answered and provided written informed consent of both, wife and husband. Genotyped data were available for all of them. The 3691 men were aged 23 to 58 years with a mean age of 42.4 and a standard deviation (s.d.) of 7.7 years. The 3691 women were aged from 25 to 58 years with a mean age of 40.1 and an s.d. of 7.8 years. The mean age difference per couple difference was hence 2.3 years with an s.d. of 3.4 years, which resembles the typical age difference distribution among married couples in Germany [41]. In 1341 couples (35.7%), the women reported not using oral contraception when starting to date their partner. All other female study participants (64.3%) reported to have used oral contraception. The couples differed insofar, as those reporting no oral contraception at dating were slightly older and had a lower age difference between the partners (without oral contraception: age men: 43.6 years (95% confidence interval (CI) 43.2–44.0); age women 41.5 years (CI 41.1–41.9); age difference 2.1 years (CI 1.9–2.3); with oral contraception: age men: 41.8 years (CI 41.5–42.1); age women 39.4years (CI 39.1–39.7); age difference 2.5 years (CI 2.3–2.6)). All of those differences were statistically significant, the effect sizes were, however, small (bootstrap test with 1000 permutations; age men: p = 0.001, [Cohen's]d = 0.22; age women: p = 0.001, d = 0.027; age difference: p = 0.006; d = 0.11).

Based on postcodes, it was possible to map the participants to 10 German first-digit postcode areas (166 < = n < = 548 couples per postcode area), more detailed mapping was not possible owing to considerations of data privacy. The regions differed in age of the men (F_9,3681 = 5.8, p < 0.001, effect size η² = 0.014) and women (F_9,3681 = 4.6, p < 0.001, η² = 0.012), but not in age difference between the couples (F_9,3681 = 1.0, p = 0.40, η² = 0.003).

(b). Major histocompatibility complex genotyping

DNA samples were analysed in the ASHI-accredited DKMS Life Science Laboratory (Dresden, Germany) using an amplicon-based workflow for high-resolution MHC typing (for details, see Lange et al. [42]; Schöfl et al. [43]). In particular, exons encoding the respective antigen recognition sites of HLA-A, -B, -C, -DRB1, -DQB1 and -DPB1 were sequenced by next-generation sequencing. For this study, alleles (including null alleles, i.e. genetic variations that are non-functional) that only differed by synonymous mutations or mutations outside the relevant exons have been joined into ‘g-groups’, as earlier described [44]. The distribution of alleles per locus was in accordance to the distribution of the German population as reported earlier [45] for HLA-A, -B, -C, -DRB1, -DQB1.

(c). Determination of major histocompatibility complex similarity

Our dependent variable was the MHC similarity between partners. This variable can be operationalized with different methods, which we named ‘minimal’ and ‘exact’ similarity and ‘probability of homozygous offspring’. Each locus consists of two alleles and depending on how the match of those two alleles between partners is counted, different similarity values result (compare figure 1 for an example). For the purpose of our study, we treated alleles that are in identical g-groups as ‘similar’. In the ‘minimal’ method, partners are considered as similar for a given locus, when at least one of the alleles is similar between partners. This approach has been used by several authors in several previous studies [21,25,37]. The method, we named ‘exact’, differentiates between similarity in one or both alleles at a given locus and has been used more frequently [7,16,17,19,21–23,28,29,35]. A high similarity (i.e. in many loci) between partners is considered disadvantageous because it enhances the likelihood for homozygous—and thereby potentially less immuno-competent—offspring. We therefore decided to directly compute this likelihood with a method, we called ‘probability of homozygous offspring’ (compare figure 1 and electronic supplementary material, figure S1). An advantage of this method is that not only the match between individuals, but also the homozygosity of each individual is considered.

Figure 1.

Example of MHC similarity between partners and the resulting count in the ‘minimal’ and ‘exact’ similarity and ‘probability of homozygous offspring’ method. For each partner, we examplified one locus with the respective two alleles. Geometric figures signify individual alleles. In the upper row, both partners are heterozygous and there is no overlap between any of the alleles. In the second row, partner A is homozygous and partner B heterozygous. Because one gene is always inherited from the father, the other one from the mother, these two pairings can not result in homozygous offspring. In row three, both partners share one allele. In the ‘minimal’ method, this results in the maximal count; the probabiltiy of homozygous offspring is however only 25%.

Comparison at different typing resolutions: besides determining MHC similarity from the sequencing of the respective genes or exons as for instance performed in [21,25], the MHC similarity can also be determined from the expressed proteins determined in the serum as performed in [7,16–19,22,23,30,34,38]. The advantage of the first method is its high precision of typing and the ability to collect samples from saliva instead of blood. A possible disadvantage of this method is over-precision by introducing distinct categories of DNA variations which express the same functional protein. In order to compare both methods, we translated the typing results to the respective serologically defined HLA antigens [46]. As antigens are not defined for all class II loci, we focused on the comparison of class I loci and calculated the probability of homozygous offspring as determined by g-group resolution from DNA sequencing and by HLA antigens over all three class I loci HLA-A, -B and -C for each of the 3691 couples. The resulting correlation was found to be 0.91 (CI 0.90–0.92).

(d). Statistical methods

First, we computed the frequency of alleles per locus over the whole sample. This sample comprised 3691 couples * 2 (individuals) * 2 (alleles per locus) = 14 764 data points per locus. Subsequently, we computed the frequency of ‘minimal’ and ‘exact’ MHC similarity between couples, as well as the probability for homozygous offspring. This was done per locus and as sum score over all class I loci (HLA-A, -B, -C), over all class II loci (HLA-DRB1, -DQB1, -DPB1) and over all loci (compare [26]).

Second, we compared the observed probability of homozygous offspring in married couples to the expected value probability of homozygous offspring in random couples. This was done by bootstrapping in the following way: for every man in the sample, we chose one woman randomly (with replacement) and computed the probability of homozygous offspring per locus in the resulting couples. The whole procedure was repeated 100 000 times and resulted in a Gaussian distribution of an expected value probability of homozygous offspring. The observed results in married couples were compared to the 95%CI of the random couples. Analyses were executed in Python.

Third, we determined whether oral contraception during relationship formation related to differences in the probability for homozygous offspring. This was done by using the same analysis as in the second step. Bootstrapping was performed within groups. This means that random couples for the condition ‘oral contraception’ were only drawn from the sample of women with oral contraception and also only from the sample of men whose women belonged to the oral contraception group.

Fourth, we compared the probability of homozygous offspring between the postcode areas, using the same analysis as in step two. Again, bootstrapping was performed within groups. The reason for this was that we assumed slightly different distribution of HLA homozygosity within different postcode areas (e.g. owing to proportion of rural areas versus cities) may relate to any potential effect of assortative/dis-assortative mate choice. Results were Bonferroni-corrected for the number of postcodes (Factor 10). Thereafter, we examined whether postcode areas with higher variety in MHC expressions had a different probability of homozygous offspring. We therefore computed the variety of allele expressions explained by the three most frequent alleles per locus and postcode. High values indicate a lower variety. This value was correlated with the probability of homozygous offspring per postcode.

3. Results

(a). Observed allele frequency

The variability of alleles per locus was very high. We observed 103 different variants for the locus with the highest variability (HLA-B) and 23 different variants for the locus with the least variability (HLA-DQB1) (figure 2a). Notably, in each locus, there are many variants which are very infrequent. For the locus with the highest variability (HLA-B), the three most frequent alleles were found in 29.6% of the individuals and low frequent alleles (<5%) were present in 40.6% of the individuals.

Figure 2.

Observed MHC values. In (a), the frequency of alleles over all 15 040 alleles (3.691 couples * 2individuals * 2alleles) is displayed per locus, e.g. for HLA A, three different alleles contribute to more than half of the total variance. The highest variety of individual alleles was observed for HLA B. In (b), the observed frequency of MHC similarity is displayed for the 3.691 couples. Under (i), the percentage of similarity in at least one, precisely one or precisely two alleles is displayed, as observed in the methods ‘minimal’ and ‘exact’ similarity. Under (ii), the probability of homozygeous offspring is displayed, e.g. for HLA A, 44.4% of couples are similar in at least one allele, thereof 40.7% in precisely one allele and 3.7% in both alleles. If the couples had offspring, in 1.1% of cases such ofspring would be homozygeous. In 12.7% of cases, such offspring would be homozygeous with a 50% chance and in 30.6% with a 25% chance. Taken together, the likelihood for homozygeous offspring in the sample is at 15.1%. Under (iii), the combined similarity values are summed over all class I and class II loci and over all six loci of class I and class II.

(b). Frequency of major histocompatibility complex similarity and probability for homozygous offspring in married couples

The observed frequency of MHC similarity between married couples varied between loci and was higher for those loci with a lower variety of alleles (HLA-DPB1, HLA-A, HLA-DQB1) than for those with a higher variety of alleles. The observed frequency depended of course on the statistical method used and is reported per method in figure 2bi,ii.

The percentage of couples which were similar according to the ‘minimal’ method varied from 20.8% (HLA-B) to 59.7% (HLA-DPQ1) and was naturally higher than the percentage of couples that were identical in two alleles according to the ‘exact method’. Such occurred in 1.0% (HLA-B) to 7.5% (HLA-DPB1) of the couples.

There were very few cases in our sample (<0.1 to 3.5%), where the couple actually had a 100% probability of homozygous offspring. The probability of homozygous offspring in married couples varied between 6.1% for HLA-B and 23.4% for HLA-DQB1. In the next step, we summed the observed percentages over the different loci. Those results are visualized in figure 2biii and show that the potential offspring of our married couples had a probability of 55.1% for being heterozygous in all six loci. The probability of being homozygous in one of the six loci was at 28.2%, the probability of being heterozygous in two loci was at 10.3% and at 2.8% in four or more loci. The probability of homozygous offspring was higher in class II than in class I loci. The percentages for the ‘minimal’ and ‘exact’ MHC similarity are also displayed in figure 2biii.

(c). Major histocompatibility complex similarity and probability of homozygous offspring in married versus random couples

The probability of homozygous offspring did not differ significantly in married couples compared to randomly permutated couples. As can be seen in figure 3, the observed probability of homozygous offspring fell within the 97.5% interval of 100 000 random permutations for each of the six loci. It was hence not significantly more or less likely for married couples to conceive homozygous offspring than it was for random couples. In line, the observed relative frequency of a 100%, 50%, 25% or 0% chance of homozygous offspring was similar to the likelihood of randomly permutated couples for each of the loci.

Figure 3.

Probability of homozygous offspring in married couples versus randomly permutated couples. (a,b,c,d,e,f) The expected values of married couples are depicted by the red line separately for each HLA locus. The x-axis indicates the fraction of couples. The histogram shows the distribution of expected values for 100 000 random permutations. The dashed lines indicate the 97.5% confidence interval of random permutations. It can be seen, that the red line never crosses the dashed line, hence married couples do not differ from random couples. (g,h,i,j,k,l) The observed relative frequency of a 0%, 25%, 50% or 100% chance of homozygous offspring is displayed for observed couples and permutated couples. Exact values can be extracted from the electronic supplementary material, Matlab figures.

Despite not reaching the formal level of significance, the observed distribution for married couples was shifted to the right in five out of six loci, indicating that it was slightly more likely to have homozygous offspring in married than in random couples. The same trend was observed when combining the loci into classes: for the combined allele match over all loci of class I or class II, no significant deviation from random permutations was observed, and the observed distribution was slightly shifted towards assortative mating.

Similar results were observed for the methods of ‘minimal’ and ‘exact’ MHC similarity between couples. Again, there were no significant deviations of married couples from random permutation, with one exception: for HLA-DPB1, married couples exhibited slightly more similar alleles than expected by chance (compare electronic supplementary material, figures S2 and S3).

(d). Probability of homozygous offspring in relation to oral contraception

Neither in women who did not use oral contraception nor in women who used oral contraception did the probability of homozygous offspring differ significantly from that obtained in a random permutation of couples (electronic supplementary material, figure S4). Visual comparison of both groups shows that women without oral contraception had a slightly higher tendency for assortative mating at HLA-A, -B and –C and a slightly higher tendency for dis-assortative mating at loci HLA-DPB1, -DRB1 and –DQB1. However, none of those observations crossed the 5% chance level.

(e). Probability of homozygous offspring in relation to postcode

The probability of homozygous offspring varied slightly between postcodes (electronic supplementary material, figures S5 and S6), but did not exceed the 5% chance level, except for 3 out of 60 calculations (5% as expected by chance) where the indication for dis-assortative mating was found (postcode 5 and postcode 9, locus HLA-DQB1; postcode 9, locus HLA-DRB1). None of the effects did hold for correction of multiple testing. It is noteworthy that a low variety of alleles per postcode correlated moderately (r > 0.3) with the enhanced probability value of homozygous offspring for the loci HLA-A, -B, -C and –DRB1 (compare electronic supplementary material, figure S7). In those postcode areas, where the variety of alleles was lower, the likelihood of homozygous offspring was higher.

4. Discussion

In contrast with our hypothesis, married couples did not share less MHC alleles than expected by chance. The probability of homozygous offspring of married couples did not differ from the range of probability of homozygous offspring observed in 95% of random permutations of couples. Hence, we found no significant pattern of MHC dis-assortative mating. This observation is in line with several previous studies [27–33] and with a meta-analysis [40]. However, several other studies reported MHC dis-assortative [34–36] as well as MHC -assortative [37–39] mating.

Despite not being statistical significant at the 5% level, it is noticeable, that married couples in our sample tended towards more assortative mating (hence towards more homozygous offspring) than the random permutations. This pattern was observed in each locus, except for HLA-DQB1. One potential reason is the in-group bias observed in studies with participants from mixed ethnicity. The first study showing MHC positive assortative mating was performed in 1017 couples with mixed ethnicity. The authors computed several comparisons within and between ethnicities and concluded that their results may be ‘entirely explainable by ethnic and racial self-preference of subgroups characterized by distinctive frequencies of A and B locus antigens' [38, p. 92]. This observation is supported by a genome-wide analysis revealing that 930 couples of North European ancestry are genetically more similar at MHC class I than expected [39], and again, the authors discuss this finding against the background of ethnicity. By contrast, our study was restricted to participants of German nationality based on participants' respective self-assessment. As self-assessment is not free of bias (e.g. [47], we further investigated regional subgroups as provided by first-digit postcode area thus reducing the potential for subpopulations in our study cohort. Nevertheless, we found a similar pattern as in the whole sample.

A different results pattern was found in a recent genome-wide study which showed higher dissimilarity in MHC loci than one would expect from genome-wide analysis. This indicates conserved MHC dis-assortative mating against the background of in-group mating bias [36]. For North European (but not for other) populations, the authors even observed larger MHC dissimilarity than expected by chance. The sample of this population was, however, biased by the fact that all couples in this study had children and the authors consequently discuss that the pattern of results might be restricted to fertile couples.

The sample size of our study is approximately three times larger than that of the largest study to date. This high powered study allowed detection of effects contributing to only 0.167% of the variance (for α = 0.05 and 1 − β = 0.8; two-tailed correlation test of bivariate normal model [48]). If there is any effect of MHC similarity on marriage in German society, those do hence account for less than 0.17% of the variability of partner choice. We assume that this would also apply to other large-scale societies with a comparable level of genetic admixture.

Our study hence indicates that there is no substantial effect of MHC dissimilarity patterns on human marriage in populations with a high genetic admixture. Previous studies showed that humans above chance prefer MHC dissimilar body odour probes (for review compare [15]). However, the effects are weak and found only for subgroups of women in a relationship [23] or for women not using oral contraception [16,21]. In our previous study, designed for effects of HLA-B or –C similarity, we found that the effect was only present for probes that were similar in both, HLA-B and in HLA-C [21]. According to the most recent meta-analysis, the sum of evidence shows no overall effect of MHC on olfactory preferences [24]. In the light of this meta-analysis and our study, one may consider the possibility that there is no substantial effect of MHC patterns on human mate choice.

An alternative explanation is the following: there is an MHC effect for human mate choice, but the effect was masked by confounding factors in our study. As potential confounders to olfactory studies, relationship status and oral contraception are known. Further confounders concern cultural expectations, local mating and polymorphism.

The potential effect of relationship status can be ruled out for our study, as all participants were married. For oral contraception, two studies found that this shifts body odour preferences towards higher MHC similarity [16,21]. Our data show a different picture. We found no significant effect of oral contraception of MHC-related mate choice.

Cultural expectations, such as socioeconomic status, common interests or family arrangements influence mate choice and marriage. Thus, it may be less likely that an MHC effect is evident when investigating married couples, as compared to studies in which subjects are freely able to choose the odour sample they find most attractive. Cultural restrictions of marriage are higher in Israel than in Northern Europe, and this is reflected in MHC similarity which is higher in Israeli (possibly owing to culturally imposed endogamy) than in Northern European couples [36]. Nevertheless, MHC dis-assortative effects have also been reported for established couples in rather liberal countries, such as Germany and the USA. Those effects relate to partnership satisfaction. In a study conducted in 252 couples, mainly students, those who were similar in HLA-B or HLA-C reported to like the odour of their partner less and reported a reduced sexual satisfaction [25]. This fits to the enhanced extra-pair mating desire observed in MHC similar couples [26]. These two studies support the idea that assumed ancestral MHC-related preferences are conserved.

Local mating can drive the tendency towards assortative mating and it fits to this argument, that established couples—including German ones—are genetically more related than one would expect by chance, as shown in a genome-wide analysis [36].

MHC alleles are highly polymorphic and although nearly 90% of our couples shared at least one allele, it is not very likely that two individuals possess a similar MHC combination in the majority of alleles. Hence, the probability for homozygous offspring was rather low and the probability for highly homozygous offspring (in four or more loci) was only 2.8%. The high polymorphism of MHC observed in countries with a rich history of migration raises the question of whether MHC dis-assortative mating is biologically relevant. This is different in very homogeneous groups, such as the Hutterites in North America, where MHC dis-assortative mating has been observed [34]. It has therefore been suggested to focus studies on MHC-related mate choice in homogeneous populations [40]. We agree that such populations are very interesting for understanding the evolutionary significance of this phenomenon. However, studies in heterogeneous samples like ours indicate that there is a degree of population polymorphism in which MHC dis-assortative mating might be of little relevance. In principle, high population polymorphism can lead to higher MHC similarity because of the in-group mating bias or local mating. Or it can lead to higher MHC dissimilarity because the larger pool of partners one can choose from increases the likelihood of MHC dissimilar partners. Our study indicated both phenomena: the enhanced similarity of married as compared to random couples speaks for an in-group bias or local mating. The positive relation between polymorphism and MHC dissimilarity observed over different postcode areas indicates an enhanced likelihood of dissimilarity driven by greater MHC diversity.

We conclude that MHC similarity does not affect marital choice in the German society and we assume that this result might be generalized to all complex Western societies. Based on our frequency estimations, we believe that the very high polymorphism of the MHC complex leads to a low probability of meeting someone who is so similar in terms of MHC similarity that a potential offspring would be highly homozygous. An important next step would be a simulation analysis determining the amount of group polymorphism sufficient to make MHC dis-assortative mating superfluous from a theoretical viewpoint and testing MHC similarity in couples from more genetically homogeneous small-scale societies.

Ethics

The procedure was reviewed and approved by the ethical board of the DKMS life science laboratory.

Data accessibility

All data are presented in the electronic supplementary material and available from the authors upon request.

Author contributions

A.H.S., T.H., I.C., J.S. and G.E. designed the study; A.H.S. and J.S. carried out the genetic analysis; I.C., G.R., D.K.K. and J.S. analysed the data; I.C. drafted the manuscript; L.S., J.H. and J.S. helped draft the manuscript; T.H., A.H.S., J.S., G.R. and J.H. gave critical input into data interpretation. All authors gave final approval for publication and agree to be held accountable for the work performed therein.

Competing interests

None of the authors’ reports competing for interests.

Funding

This work was supported by the Deutsche Forschungsgemeinschaft (DFG) for the project 424634632. J.H. is supported by the Czech Science Foundation grant no. (18-15168S).