Royal Society Publishing

In search of the chemical basis for MHC odourtypes

Jae Kwak, Alan Willse, George Preti, Kunio Yamazaki, Gary K. Beauchamp

Abstract

Mice can discriminate between chemosignals of individuals based solely on genetic differences confined to the major histocompatibility complex (MHC). Two different sets of compounds have been suggested: volatile compounds and non-volatile peptides. Here, we focus on volatiles and review a number of publications that have identified MHC-regulated compounds in inbred laboratory mice. Surprisingly, there is little agreement among different studies as to the identity of these compounds. One recent approach to specifying MHC-regulated compounds is to study volatile urinary profiles in mouse strains with varying MHC types, genetic backgrounds and different diets. An unexpected finding from these studies is that the concentrations of numerous compounds are influenced by interactions among these variables. As a result, only a few compounds can be identified that are consistently regulated by MHC variation alone. Nevertheless, since trained animals are readily able to discriminate the MHC differences, it is apparent that chemical studies are somehow missing important information underlying mouse recognition of MHC odourtypes. To make progress in this area, we propose a focus on the search for behaviourally relevant odourants rather than a random search for volatiles that are regulated by MHC variation. Furthermore, there is a need to consider a ‘combinatorial odour recognition’ code whereby patterns of volatile metabolites (the basis for odours) specify MHC odourtypes.

1. Introduction

Research on the role of the genes that make up the major histocompatibility complex (MHC) in chemical communication in mice was initiated by Lewis Thomas's speculation (Thomas 1975), followed by the serendipitous finding by Yamazaki and colleagues, that MHC genes are associated with an individual's odour signature (Yamazaki et al. 1976). Many subsequent studies showed that MHC genetic variation expressed in urinary volatile patterns modulates social and reproductive behaviours such as mate choice and parent–infant recognition (reviewed in Beauchamp & Yamazaki 2003; Yamazaki & Beauchamp 2007).

From the beginning of these studies, the chemical nature and structures of the MHC-regulated chemosignals have been the objects of intense investigation. Two different classes of compounds have been proposed: small volatile molecules and non-volatile MHC peptide ligands (Boehm & Zufall 2005; Restrepo et al. 2006). Volatile molecules are non-ionic and have a finite vapour pressure between ambient and 250°C, and their molecular weights are generally less than 300 Da, whereas MHC peptide ligands are ionic and composed of nine amino acids, with molecular weights that exceed 1000 Da. The peptide hypothesis is attractive since it can explain one mechanism of MHC regulation of the discriminative chemosignals. However, while MHC-derived protein fragments have been identified in murine urine (Singh et al. 1987), peptide ligands have not. In contrast, hundreds of volatile compounds have been detected in urine. However, as detailed below, it is still unclear which compounds are regulated by MHC genes, and how, although several mechanisms have been proposed (Restrepo et al. 2006 and references therein).

As will be described in detail in what follows, chemical studies designed to specify the chemical identity of MHC odourtypes have typically focused on searches for qualitative and/or quantitative differences in volatile profiles between individual mice that differ only in MHC genes. Many quantitative chemical differences have been identified, although different studies have identified different volatiles. However, we will argue that none of these have been demonstrated to actually underlie the odour differences between such mice as perceived by other mice. Moreover, a number of studies have found that urinary volatile profiles that appear to be regulated by MHC genes are also influenced by non-MHC genes, as well as environmental variation. Although these observations are interesting and important, they provide further evidence that the identified MHC volatiles cannot underlie the ‘true’ MHC odourtype since in behavioural tests this remains consistent across different genetic backgrounds. That is, animals can identify the MHC type of an animal even when the background genotype is varied. Following a review of these studies, we speculate on the reasons for this lack of success and suggest potential alternative approaches that may provide a better chance of identifying MHC odourtypes.

2. Chemical investigations of MHC odourtype in mice

The results of chemical investigations on MHC odourtypes in inbred laboratory mice that have been published following their discovery in the 1970s is summarized in the electronic supplementary material, table S1. The chemical investigations of MHC odourtypes have been carried out almost exclusively on mice. Whereas it has been suggested that other species also detect the individual odour difference due to the MHC difference (reviewed in Yamazaki & Beauchamp 2007), the chemical studies in other species are very limited. In this review, we focus on the chemical investigations in mice, while those in other species are summarized in the electronic supplementary material.

While mice can recognize the MHC-driven odour cues from whole body odours, urine and blood (Yamazaki et al. 1979, 1999; Yamaguchi et al. 1981), it has been demonstrated that urine is the strongest and most effective odour source (Yamaguchi et al. 1981). Therefore, most chemical investigations have focused on urine-derived compounds.

All published studies on the chemistry of mouse MHC odourtypes have been conducted on inbred strains. This makes sense since it is in these strains that one can control for differences in other parts of the genome. Nevertheless it will be important to investigate the chemistry of wild mouse MHC odourtypes. In this regard, there is as yet no direct evidence for the existence of MHC odourtypes in wild mice. Several studies have failed to find specific behavioural effects of MHC variation in wild mice (Cheetham et al. 2007; Sherborne et al. 2007; Thom et al. 2008; but see Potts et al. 1991). However, this cannot be taken as proof that they do not exist. Indeed, since MHC odourtypes have been identified in many different mouse strains (Boyse et al. 1991), and in other species (see the electronic supplementary material), it seems highly probable that they will be identified. Their possible functional significance in regulating wild mouse behaviour (e.g. mother–infant recognition; mate attraction from a distance) and/or physiology remains to be determined.

Prior to reviewing MHC odourtype chemistry, a brief introduction to mouse inbred strains and their nomenclature is necessary. An inbred strain is defined as the one resulting from mating brother and sister pairs for at least 20 consecutive generations. Within the same inbred strain, its genotype is nearly identical. Substrains are defined here as subcolonies that have been separated from their parent strain and maintained in different laboratories for more than 20 generations of inbreeding. For example, the C57BL/6 strain developed by Clarence Cook Little in 1921 was introduced to the Jackson Laboratory in 1948, which became the C57BL/6J substrain. The C57BL/6N substrain diverged from the C57BL/6J substrain transferred from the Jackson Laboratory to National Institute of Health in 1951. The substrains C57BL/6J and C57BL/6N are more than 200 generations apart (Taft et al. 2006). Substantial genetic and behavioural differences caused by genetic drift in substrains have been reported (Crawley et al. 1997; Simpson et al. 1997).

A congenic strain is formed by repeated backcrosses of a donor strain to an inbred (background) strain with selection for a locus of interest from a donor strain. The B6-H2k strain consists of the C57BL/6 genetic background, but has an H2k derived from the AKR strain (rather than H2b, which is the inherent H2 type of the C57BL/6 inbred strain) in the MHC region. The detailed rules covering mouse nomenclature are available on the Jackson Laboratory Website (International Committee on Standardized Genetic Nomenclature for Mice 2010).

The first analytical report of urinary volatile compounds from mice with different MHC types was published by Novotny et al. (1980). They compared the volatile profiles of male inbred mouse strains AKR-H2b, AKR-H2k and C57BL/6-H2k (B6-H2k). Their data indicated that the background genetic effect was larger than the MHC effect on the expression of the volatile metabolic phenotypes. No specific MHC-associated compound was reported.

The same research group (Schwende et al. 1984) then investigated the urinary volatile profiles collected from immature and oestrogen-treated mature female C57BL/10 (B10) congenic strains with different MHC haplotypes (see the electronic supplementary material, table S1 for details of this and many other studies). While quantitative differences were observed in the immature congenic females, no MHC-specific compounds were reported.

Singer et al. (1993) subsequently adopted a bioassay-guided urine fractionation procedure to obtain chemical fractions responsible for the mouse's ability to discriminate MHC-congenic mice by odour cues. First, mice were trained (we refer to these trained mice as sensor mice) in a standard Y-maze to discriminate between inbred strains of mice that differed only in the MHC region. In this Y-maze the sensor mice were rewarded for going into one of the arms scented by urine of one of the two MHC-congenic strains being tested. Then these trained mice were tested with various urine fractions (e.g. of different molecular weight or polarity) to identify those fractions responsible for successful strain discrimination. In these studies, urines of B6-H2b and B6-H2k mice were fractionated by either gel permeation chromatography, dialysis, ultrafiltration or solvent extraction. In addition, degradation of protein by Pronase, as well as denaturation of protein using perchloric acid, was employed to investigate whether urinary proteins are involved in the MHC odour cues. Except for the case of gel permeation chromatography, all other methods consistently confirmed that compounds with low molecular weight were responsible for the discriminative MHC signals. Although the bioassay-active fraction obtained using gel permeation chromatography was the protein fraction, it is probable that volatiles were the active compounds because most of the protein—the major urinary proteins (MUPs)—strongly sequester volatile compounds and release them slowly (Bacchini et al. 1992; Hurst et al. 1998). In addition, the sensor mice were not allowed to contact the samples and therefore it is probable that only airborne compounds released from the protein fraction were delivered to the Y-maze for the bioassay. Consequently, the discriminative chemosignals in this protein fraction were probably made up of volatile compounds.

Some of these fractionation techniques were modified and two separate chemical analyses of the urines from B6-H2b and B6-H2k mice were conducted in subsequent studies by our group (Singer et al. 1997; Willse et al. 2005). The extraction method was the same, but the extracted material in the earlier study was analysed by gas chromatography equipped with a flame ionization detector (GC/FID), whereas the later study analysed the extracted material by gas chromatography/mass spectrometry (GC/MS). The analytical results employing GC/FID (Singer et al. 1997) showed that 8 out of 32 compounds were quantitatively different between the MHC-congenic mouse strains (i.e. mice differing only in MHC type). One of these compounds was identified as phenylacetic acid. In addition, a separate experiment using anion exchange chromatography confirmed that acidic compounds were responsible for the MHC odour cues, suggesting that some volatile organic acids are regulated by MHC genetic variation (Singer et al. 1997).

The Willse et al. (2005) study employed a comprehensive statistical method to examine all detectable components in the GC/MS data. This ‘metabolomic’ approach permitted unbiased analyses of a broad spectrum of metabolites in the search for MHC-regulated compounds. Using this method, several hundred components in the GC/MS data were resolved. Variation in the relative amounts of about 80 compounds was found to be MHC-associated, including two reported mouse pheromones: 2,5-dimethylpyrazine and 2-sec-butyl-4,5-dihydrothiazole.

The abovementioned studies focused on inbred mice that differed only in MHC genes. Since the background strain of the mice was held constant, the identified MHC-associated volatile metabolites were restricted to this particular mouse strain. However, behavioural studies using animal sensors have shown that an MHC-associated odour cue can be recognized in spite of variation in background strains (Beauchamp et al. 1990; Yamazaki et al. 1994; Eggert et al. 1996; Willse et al. 2006). These behavioural results indicate that an MHC-regulated metabolic odour signal must have a background-independent invariant component. That is, there should be MHC-regulated odourants that are common to the strains examined in these behavioural studies.

To evaluate this conjecture, several groups examined the urinary volatile profiles of MHC-congenic mice derived from different background strains (Eggert et al. 1996; Willse et al. 2006; Novotny et al. 2007; Zomer et al. 2009). Eggert et al. (1996) investigated the urinary volatile metabolic profiles of BALB/c (H2d), BALB/k (H2k) and C3H (H2k) mice. Unlike virtually all other studies to date, where almost all components were detected in the mouse urines, only 36 out of 55 peaks were found to be present in all three strains. Both qualitative and quantitative differences were reported in the urinary volatile profiles of the three strains, indicating that background genes, MHC genes and their interaction regulate the urinary volatile profiles. No compounds, however, were structurally identified.

Willse et al. (2006) examined the urinary volatile metabolic profiles of two MHC haplotypes (H2b and H2k) and their heterozygous cross (H2b × H2k) in two different background strains (B6 and BALB/c). Although a large proportion of the background genes are shared between these strains, thereby limiting potential between-strain genetic variability, substantial genetic variation is observed that is mainly derived from the variability in founders from the main wild source, Mus musculus domesticus (Yang et al. 2007). In this study, we used a different method of isolation for volatile compounds: solid phase microextraction (SPME) headspace analysis, followed by GC/MS and statistical analyses. SPME is an attractive method for investigation of the MHC-associated volatile metabolites because it is simple, fast and reasonably comprehensive. Importantly, volatile signals collected by this method carry sufficient information for sensor mice to discriminate odours of different MHC types (Kwak et al. 2009). Of 148 compounds examined, 108 (73%) significantly varied in different genotypes. When considering only the homozygotes (B6-H2b, B6-H2k, BALB-H2b and BALB-H2k), background genes accounted for about 40 per cent of total genetic variability while MHC genes accounted for 20 to 25 per cent, based on multivariate redundancy analyses. This relatively large MHC contribution is striking given that MHC genes make up less than 0.5 per cent of the mouse genome.

Unexpectedly, nearly 40 per cent of the genetic variability was due to the interaction between MHC and background genetic variation. That is, as shown in the electronic supplementary material, figure S1, a compound was found to be more prominent in one MHC type in one strain and less prominent in that same MHC type in the other strain. In some cases the pattern was even reversed (e.g. 2-sec-butyl-4,5-dihydrothiazole; see the electronic supplementary material, figure S1c; Willse et al. 2006). Thus, for compounds such as these, MHC regulation of volatile metabolic phenotypes was modulated by background genes. As a result of these widespread interactions, only a few compounds were shown to be consistently regulated by MHC genes in the same direction for both background strains. This result appears to be inconsistent with behavioural studies, described above, that demonstrate that animals are able to identify MHC type in the face of background variation.

In an attempt to understand this apparent contradiction, Willse et al. (2006) developed a multivariate or combinatorial odour-recognition approach (olfaction is also thought to employ an analogous combinatorial code; reviewed in Johnson & Leon (2007) and references therein). Individual volatile metabolites influenced by the interaction of MHC and background genes, if considered individually, may not underlie MHC odourtype recognition, but they may do so in combination. For example, the relative ratios of certain combinations of any two compounds may be consistent for MHC type over the different background strains even though each compound by itself is not. Using this approach, 43 non-overlapping compound pairs representing 86 compounds were found to be significantly different in MHC groups across background strains.

The mode of inheritance was also investigated in the same study (Willse et al. 2006) by breeding heterozygous mice carrying two different sets of MHC genes (one set from female parent and the other set from male parent) in the two different background strains. It was not known whether the metabolic phenotypes of MHC heterozygote are an additive trait (an average of the two different MHC groups) or exhibited a different form of inheritance. A previous behavioural study (Yamazaki et al. 1984) showed that sensor mice could easily distinguish the scent of MHC heterozygote urine from the scent of equal parts of urine mixture from two MHC homozygotes, indicating that the odour phenotypes of MHC heterozygote are not an additive trait. Inheritance patterns (additive; dominant; recessive) were investigated by comparing the urinary metabolic phenotypes of three MHC types (H2b, H2bk and H2k) for two different background strains (B6 and BALB/c). Substantial heterozygous effects on the metabolic phenotypes were identified (a heterozygous effect is defined as occurring where the amount of a metabolite in the heterozygote is greater or lesser than that in either homozygote; e.g. see the electronic supplementary material, figure S1d,g). Most of the heterozygous effects were also modulated by background strain. While numerous metabolites were identified with a heterozygous pattern, only two of them had the same heterozygous pattern in the two strains. This result also suggests either that the chemically analysed volatiles were not those that specify the behaviourally determined MHC odourtypes or that their relative concentrations must account for odourtype stability across inbred strains.

Novotny et al. (2007) investigated the urinary volatile metabolic profiles of two B6 MHC-congenic strains, four B10 MHC-congenic strains, two BALB MHC-congenic strains and three B6-derived mutant strains (see the electronic supplementary material, table S1 for details). The majority of the 16 compounds they examined were affected by either MHC or background genes. Surprisingly, no substantial alteration in the volatile profile was found by antibiotic treatment.

Recently, Zomer et al. (2009) investigated the profiles of volatile compounds and microorganisms associated with the scent-marks collected from two BALB MHC-congenic strains and two B10 MHC-congenic strains. Since the volatile profiles of the real scent-marks were highly variable, urine samples (labelled as ‘simulated scent-marks’) were used for the chemical analyses. These authors reported that both MHC and background genes influenced the profiles of volatile compounds and micro-organisms; background genes had a larger effect than MHC genes, but no specific compounds were identified. Based on the fact that the results of the chemical analyses were consistent with the microbial results, they suggested that the volatile compounds regulated by MHC and background genes are influenced by commensal micro-organisms. This suggestion is based on the assumption that microflora are a source of some volatile compounds (Lanyon et al. 2007). However, they provided no direct evidence that volatile compounds associated with MHC and/or background genes are regulated by commensal micro-organisms. Indeed, results of previous studies (Yamazaki et al. 1990; Novotny et al. 2007) provided evidence against the importance or necessity of commensal micro-organisms in determining MHC odourtypes.

Montag et al. (2001) analysed volatile compounds of the urines collected from three B10 MHC-congenic strains (B10.A-H2a, B10.BR-H2k and B10.D2-H2d) and B6 mice. Quantitative but not qualitative differences in the volatile profiles were observed. Two compounds (3-methylbutanal and 2-pentanone) were MHC-associated. The same study further demonstrated that urinary volatiles from different mouse strains could be differentiated by a chemical sensor device (electric nose).

Kwak et al. (2008) examined the effect of diet on MHC-regulated urinary metabolic phenotypes. Volatile profiles of urine samples from two B6 mouse groups that differ only in MHC genes (H2b and H2k), each on two different diets (laboratory diet and synthetic diet), were evaluated. The laboratory diet was ‘laboratory rodent diet 5001’ and the synthetic diet was ‘basal diet 5755’, both purchased from Purina Mills. Of 50 identified metabolites, 20 were influenced by diet variation and 20 were influenced by MHC variation. Ten compounds were significantly affected by the interaction of diet and MHC. Of total variability in metabolic phenotype, 53 per cent could be attributed to diet and 35 per cent to MHC. The interaction of diet and MHC accounted for about 12 per cent of the variability. Although the metabolic phenotypes were influenced by diet, MHC types could be accurately classified using a statistical algorithm, a result consistent with behavioural studies using trained sensor mice (Kwak et al. 2008).

Röck et al. (2007) investigated the urinary volatile profiles of three substrains of B6 mice (two variants of C57BL/6J and C57BL/6NCrl), a β2m-deficient mutant (B6.129P2-B2mtm1Unc/J) and two unrelated inbred strains (BALB/cCrl and DBA/2Crl). They observed that the difference in volatile profiles of the unrelated strains compared with any B6 substrain was larger than that of the B6 substrain compared with the other B6 substrains. However, the volatile profile of the β2m-knockout mice (β2m is necessary for expression of some MHC gene products) was very similar to that of all B6 substrains examined. Furthermore, differences among the B6 substrains were more prominent than differences between the β2m-deficient mutant and any B6 substrains. In light of the relatively large quantitative differences within the substrains, which are probably due to genetic drift over the generations as well as to different environmental conditions, they concluded that differences in urinary volatile profiles in mice cannot be interpreted as being robustly regulated by MHC genes.

In summary, with one exception (Eggert et al. 1996), all published chemical investigations have failed to find qualitative differences in volatile compounds that are associated with MHC types. Instead, the patterns (or relative ratios) of volatile metabolites vary according to MHC types, but in an inconsistent and complex way.

3. Why do published studies show little agreement on the identity of MHC-associated compounds?

(a) The problem of individual variation among presumably genetically identical mice

Significant individual variation within an inbred strain can often obscure the results, especially when the sample size is small. For example, Röck et al. (2007) reported that the standard deviation of the mean for the intensity of each component detected in the GC/MS data collected from 15 genetically identical individual inbred mice under identical environmental conditions varied between 18 and 308 per cent.

Several factors can be postulated to explain the individual variation within supposedly genetically identical individuals of an inbred strain. Different volatile metabolic phenotypes between dominant and subordinate mice have been reported (Novotny et al. 1990). The position of a mouse within the mother's uterus (whether it is adjacent to a same-sex sibling) can affect its behavioural phenotype after birth, and perhaps this is also the case for odourtype (reviewed in vom Saal 1989; Ryan & Vandenbergh 2002). An individual's health status and difference in body weight may account for some variation in volatile profiles. Behavioural differences among individuals could result in different patterns of food and water consumption, and hence urination frequencies, which could impact proportions of volatiles. Individual variation in volatile profiles may also be affected by variation in MUP content, which may change volatile profiles since MUPs retain volatile compounds and release them slowly (Bacchini et al. 1992; Hurst et al. 1998).

(b) The problem of inconsistency across studies

Based on behavioural results reviewed above, one would expect there to be a set of MHC-regulated compounds that could be identified in many inbred (and ultimately in wild) mice. However, chemical studies have not yet reached a consensus on such a set and have instead shown striking inconsistencies.

Three types of inconsistency in the MHC-associated compounds have been described. First, compounds have been reported to be MHC-regulated in one study but not even detected in other studies. As an example, phenylacetic acid was reported as an MHC-associated compound (Singer et al. 1997), but its presence in urine has not been documented in other studies. This inconsistency is probably due to differences in analytical methods used to collect and isolate the urine samples. Solvent extraction (the technique used by Singer et al. (1997)) may bias the analyses in favour of less volatile compounds, whereas highly volatile compounds such as methylthiomethanethiol may be lost during the removal of solvent. In contrast, techniques that collect and concentrate the headspace above a sample (e.g. SPME) may bias the analyses in favour of more volatile compounds and lose some odourants having a low vapour pressure (e.g. phenylacetic acid). In addition, most studies have selected a limited number of compounds, mostly ones with high abundance (although not necessarily high odour potency), for the statistical analyses. As a result, many true MHC-associated compounds have probably been excluded in these analyses.

The second type of inconsistency is that some compounds were reported to be MHC-associated in one study but not in others. For example, phenylacetic acid was reported to be MHC-regulated in one study (Singer et al. 1997), but not in a second study employing the same extraction method (Willse et al. 2005). These results were not necessarily contradictory since, as Willse et al. (2005) discussed, twice as many urine samples were analysed in the first study compared with the second study, thereby providing greater power to detect differences. Several compounds (e.g. dimethyl sulphone) that have been reported as MHC-regulated metabolites in later studies (Willse et al. 2005, 2006; Kwak et al. 2008) were previously ignored based on bioassay results of the fraction obtained by anion exchange chromatography, which theoretically captures only negatively charged compounds such as organic acids (Singer et al. 1997).

The third and most puzzling inconsistency is that certain compounds were reported to be MHC-related, but their patterns in different MHC types were opposite in different studies. For example, Willse et al. (2005, 2006) and Kwak et al. (2008) reported that the concentration of 2-sec-butyl-4,5-dihydrothiazole was higher in urines of B6-H2k compared with B6 mice. However, Novotny et al. (2007) reported the opposite pattern. One possible explanation for this discrepancy is that slightly different inbred mouse strains were used in the experiments. The B6-H2k strains used in the different laboratories carry the same H2k haplotype derived from the AKR strain, but the genetic backgrounds are slightly different (C57BL/6Boy versus C57BL/6JFlaEg). The B6-H2k mice used for our studies were born in our laboratory, while the other B6-H2k mice (B6.AK-H2k/FlaEg; Stock # 001148) used by Novotny et al. (2007) were purchased from the Jackson Laboratory. It is possible that these accompanying minor genetic variations, along with different environments and their interactions with the MHC genes, influence the expression of 2-sec-butyl-4,5-dihydrothiazole. However, a complete reversal of relative amounts is hard to understand and seems inconsistent with behavioural results.

In a somewhat different example, one study employing the solvent extraction method found a higher level of p-cresol in B6 urine compared with that of B6-H2k urine (Willse et al. 2005). An opposite pattern, however, was found in the same laboratory in a study using SPME (Kwak et al. 2008). Different extraction methods and individual variation of mice may be responsible for this opposite pattern. The solvent extraction method included a centrifugal ultrafiltration step that removes the majority of MUPs (Osada et al. 2008), while intact urine was used for the SPME method. Since MUPs retain various volatile compounds and release them slowly, certain compounds may be partially lost when MUPs are removed by the ultrafiltration.

4. MHC odourtypes and complexity of individual odour signatures

Chemical studies reviewed above have documented the complexity of volatile urine profiles influenced by variation in MHC genes. In addition to the MHC-associated chemosignals, mice release diverse chemical signals that communicate not only their unique odour signatures but also information such as gender, age, motivational state, sexual readiness and health status. Some of the sources and chemical identity of these signals in mice have been studied, including MUPs (Hurst et al. 2001; Chamero et al. 2007), exocrine glands secreting peptides (Kimoto et al. 2005), urinary signals of virus infection (Yamazaki et al. 2002) and the animal's age (Osada et al. 2003, 2008), and volatile semiochemicals that are involved in sexual communication (e.g. 2-sec-butyl-4,5-dihydrothiazole, 3,4-dehydro-exo-brevicomin and methylthiomethanethiol; Jemiolo et al. 1986; Lin et al. 2005).

Like the MHC-associated volatile compounds, these chemosignals are also affected by genetic and environmental variations, as well as their interactions, and they may also underlie individual recognition in some cases. In one prominent example, distinctive profiles of MUPs are expressed in different strains, as well as in different subspecies, of mice (Robertson et al. 1996; Stopková et al. 2007). An atypical MUP that preferentially binds to 2-sec-butyl-4,5-dihydrothiazole is male-specific (Armstrong et al. 2005). Variation in MUPs is extensive in wild populations, where they are involved in individual recognition (Hurst et al. 2001; Cheetham et al. 2007), inbreeding avoidance (Sherborne et al. 2007) and evaluation of genetic heterozygosity of potential mates (Thom et al. 2008). However, since large amounts of proteins are not generally observed in the urine of other mammals, including humans, MUP-like molecules probably have a more limited role in olfactory communication in those species. Nevertheless, it is noteworthy that 3-methyl-2-hexenoic acid, a characteristic underarm odour compound in humans, is carried in underarm secretions by apolipoprotein D, a member of the same lipocalin family as MUPs (Zeng et al. 1996).

Background genetic variation, as well as its interaction with MHC genes, and other factors influence the concentration of the putative mouse pheromones. For example, 2-sec-butyl-4,5-dihydrothiazole in urine is influenced by MHC type and mouse strain (Novotny et al. 1980; Willse et al. 2006; Röck et al. 2007; Zhang et al. 2007). The amount of this pheromone in urine is also regulated by gender (Liebich et al. 1977) and age (Osada et al. 2008). The level of another pheromone in urine, 3,4-dehydro-exo-brevicomin, is also affected by background strain, diet and age (Novotny et al. 1980; Willse et al. 2006; Zhang et al. 2007; Kwak et al. 2008; Osada et al. 2008). Finally, we detected methylthiomethanethiol in the urines collected from the BALB mice (BALB/b, BALB/k and BALB/bk) but not in those from the B6 mice (B6, B6-H2k and B6-H2bk) used for the studies conducted by Willse et al. (2006), and its concentration was significantly higher in the MHC heterozygote (H2bk) of the BALB mice than in the homozygotes (H2b and H2k; J. Kwak & G. Preti 2006, unpublished data).

5. Discussion

Why has it been so difficult to specify the odourants involved in MHC-based recognition? As indicated above, some of the variation in findings is a consequence of differences in strains of mice studied and methods used to isolate and identify constituents. But the problem is obviously deeper than this. The most telling critique of the MHC-odourant hypothesis comes from the work of Röck et al. (2007). These investigators reported that the differences in the urinary volatile profiles among the B6 substrains were more prominent than those between the β2m-deficient mutant and any of B6 substrains. In light of the relatively large quantitative differences within the substrains, probably due to genetic drift over the generations, as well as to different environmental conditions, they concluded that small differences in relative urinary volatile concentrations due to MHC variation cannot be considered a robust genetic trait and that current evidence does not support the proposition that MHC class I variation influences urinary volatile composition.

It is important to keep in mind the fundamental observation motivating all chemical studies reviewed here: the incontrovertible evidence, obtained independently in many laboratories, that MHC variation influences urine and other body odour as demonstrated in behavioural experiments (reviewed in Beauchamp & Yamazaki 2003). This body of evidence includes not only training studies but also many behavioural studies that do not employ traditional training paradigms (reviewed in Beauchamp & Yamazaki 2003). This is not to deny that other sorts of variation (genetic, environmental) also impact body odour, but when this is controlled or even when it is systematically varied (Yamazaki et al. 1994; Willse et al. 2006; Kwak et al. 2008) MHC influences on odour remain robust.

Although it had been assumed that MHC-congenic strains differ only in the MHC region, it is now known that due to genetic drift these strains have accumulated small differences in non-MHC regions that may influence an animal's odour (Carroll et al. 2002). While this could contribute to difficulties identifying the chemical nature of MHC odourtypes by providing an uncontrolled source of variation, it seems very unlikely that this can account for the fundamental problems we have identified. Behaviourally, the MHC contribution is robust against any potential uncontrolled variation in background genes. That is, MHC odourtypes are recognized in F2 segregants (where any background differences are randomly distributed), as demonstrated in several different paradigms (e.g. Yamaguchi et al. 1981; Yamazaki et al. 1994). And, as described above, they are evident even when background is specifically confounded. We conclude that the odourants that MHC variation impacts upon are easily detected by mice; why then is it so difficult to specify their chemical identity?

First, it is crucial to recognize the distinction between chemically measured volatile levels and perceived odour. The potency of an odourant (i.e. the concentration at threshold and the shape of its concentration–intensity function) is often unrelated to its relative concentration in a mixture. For example, the odourant 2-isobutyl-3-methoxypyrazine (the predominant odour of green pepper; Buttery et al. 1969) is highly potent and, while it might appear as a small peak in a complex mixture, its role in defining the odour of that mixture may be profound (see also O'Connell et al. 1979 for a similar point concerning the active odourant in hamster vaginal secretions). Thus, analysing small sets of most prominent peaks with the expectation that these will define the odour of this mixture is naive. Moreover, the olfactory system processes mixtures in complex ways. Human psychophysical studies indicate that mixtures of more than four or five odourants cannot be deconstructed into their components and that these mixtures are highly sensitive to relative potencies of the odourants in the mixture (Livermore & Laing 1996; see also Brennan & Kendrick 2006). Therefore, for complex mixtures, analysis of individual volatiles, based on what the instrumentation and methods find to be the most prominent, is likely to fail to identify the odours as detected by a mammalian olfactory system.

Consider a parallel evolved system for communicating individual identity: visual face recognition. There is now considerable evidence that humans, other primates and even sheep are able to recognize individuals by distinctive visual characteristics of their faces (reviewed in Calder & Young 2005; Tate et al. 2006; McKone et al. 2007). This ability seems to reside in some sort of holistic processing of relational visual information. For example, inverted faces are processed in different ways than faces as normally viewed, and this does not seem to be merely the result of common experience. Thus, no single visual characteristic of a face is definitive—a large nose or small eyes do not constitute the code for individual identity. Also, many investigators believe that there is an innate face template that is fine-tuned during early human development. This recalls findings that early experiences are potent in forming preferences for MHC-regulated mate choice (Yamazaki et al. 1988). Moreover, this innate component is thought to be shaped by the evolutionary importance of individual recognition in a species like humans (or sheep) who have a highly developed visual system (McKone et al. 2007). However, for most mammals, olfaction is the primary communicative sense and so it would seem probable that there is an innate template for recognizing MHC odourtypes.

6. Prospectus

In spite of the methodological and theoretical difficulties discussed here, we still wish to chemically specify the pattern of volatiles that characterizes individual MHC odourtypes. How can this enterprise productively proceed?

We suggest that the search for the chemical basis for MHC odourtypes must envision the goal not as an attempt to find volatiles that distinguish MHC types but instead as a search to specify the chemistry underlying the odours that animals perceive when they recognize an MHC odourtype. This focus on odours has two important implications for how the problem is approached.

First, identification and evaluation of the most prevalent (and hence most easily analysed) volatile compounds are likely to miss behaviourally relevant odourants. Volatiles at low concentration but with high odour impact are likely to be behaviourally relevant. Second, odours are virtually always comprised of mixtures of odourants wherein the patterns (i.e. relative amounts of odourants) characterize the perceived odour. Analyses should focus on these patterns of odourant molecules (see Willse et al. 2006).

Although these implications are not difficult to identify, knowing how to address them successfully is much more problematic and entails enormous practical problems. How should one search for these hypothetical groups of odourants that specify an MHC odourtype? A bioassay-driven approach may be useful since it focuses directly on the phenomenon of interest—the odour rather than odourants. A problem with this approach, however, is that in fractionating a complex mixture one may destroy the pattern and hence no fraction will remain active (Beauchamp et al. 1976; Singer et al. 1993). A related approach would be to differentially mix fractions from different inbred or congenic lines to isolate groups of volatiles critical to individual identification prior to attempting to identify specific compounds. While neither of these approaches can assure success, they at least refocus the search towards the identity of the odours as perceived.

Regardless of what approach is used to focus on MHC-related odourants, their structural identification would still require isolation from the urine matrix, which, as described above, is subject to various errors. The difficulty in chemically specifying critical volatile pattern differences that underlie MHC odourtypes starkly contrasts with the apparent ease with which animals accomplish this task. This highlights the importance of understanding olfactory system function as we attempt to decipher an animal's olfactory-based communication system. As we begin to understand how odour mixtures are processed and develop computational models to accomplish what the animal so effortlessly does, we should also make progress in chemically defining MHC odourtypes.

Acknowledgements

This work was supported by ARO contract DAAD19-03-1-0109. Opinions, interpretations, conclusions and recommendations are those of the authors and are not necessarily endorsed by the United States Government. We are grateful to Dr Bruce A. Kimball and Dr Johannes Reisert for useful suggestions on earlier versions of this manuscript.

Footnotes

    • Received January 26, 2010.
    • Accepted March 9, 2010.

    References

    View Abstract