Brain gene expression differences within and between populations
In this experiment, we studied variations in gene expression and methylation in brains of RJF and domesticated White Leghorns (WL), and their offspring. We focussed on thalamus and hypothalamus, brain regions involved in fear and stress responses, both of which have changed significantly during domestication [3
]. Within each population, we selected parental animals with divergent phenotypes in order to maximise the within population genetic variation. Specifically, we used two pairs of each population, with pairs within population differing in their behaviour in a series of previously validated tests of stress reactions in chickens [6
]. From these, totally 73 offspring were hatched and reared until three weeks of age, when they were tested in a fear test, similar to that used in the parents.
In both breeds, body weights differed between families in both generations, and behavioural scores, as measured in the fear tests, differed between families in both generations of WL, but not RJF (Additional file 1
). Hence, morphological, and to some extent behavioural, phenotypes showed a significant and transgenerationally stable variation in the animals used for the present study. It should be noted that phenotyping was done at different ages in the two generations, which may have been the reason for the lack of transgenerational correlation in fear behaviour in RJF. All eight parents were sacrificed at an age of 373 days, and 48 offspring (12 from each pair) at 21 days, and from each brain, the thalamus-hypothalamus region was removed for extraction of both DNA and mRNA. For the offspring, eight pools of both were prepared, each consisting of six same-sex samples within families. Hence, there were in total eight parental single-animal samples, and eight pools of offspring samples. The mRNA was hybridized to a 38K Affymetrix chicken gene expression microarray, and the DNA was used for subsequent tiling array analysis of methylation. Between populations, there were in total 281 significantly (FDR-corrected P < 0.05) differentially expressed (DE) genes in the parents, and 1674 in the offspring. The lower number of DE genes in the parents could possibly be an effect of the lower power of detection given the smaller biological sample size in this generation. Between families within populations, only a few genes were significantly DE, and also DM was less frequent between families (Additional file 2
). This indicates that expression and methylation profiles are relatively stable within breeds, but both may have changed considerably during domestication.
Transgenerational stability of gene expression profiles
Out of the significantly DE genes in the parents (comparing populations), 86% percent (n = 242) were also significantly DE in the offspring (Additional file 3
), and there was a distinct similarity in the expression differences in both generations (Figure ). The overall pattern of fold-change levels between populations (regardless of whether they were significant) was strongly correlated over generations (Figure ), further showing a transgenerational stability in gene expression profiles. Also within populations, the overall pattern of fold-change levels between families was highly correlated across generations (Figure ).
Figure 1 Gene expression and methylation differences between populations and across generations. a. Heat map, showing the clustering of 242 differentially expressed genes, comparing parental Red Junglefowl (RJF) and White Leghorn (WL) layers, and their offspring. (more ...)
Figure 2 Correlations of differential expression and methylation of genes between generations. a-c, Correlations between generations of differential gene expression, comparing Red Junglefowl and White Leghorn (a); families within Red Junglefowl (b); and families (more ...)
We further used signalling intensities of individual probesets on each microarray to correlate global expression levels between parents and their own offspring, compared to offspring of other birds, and found a significantly higher correlation within families than between (mean difference in correlation coefficients 0.0017 ± 0.0002 (SEM), t = 8.2, P < 0.001). This was true both for RJF and WL, and further supports that specific brain gene expression profiles are indeed inherited.
Gene methylation: inheritance and differences between populations
For analysis of differential methylation (DM), we selected 3623 genes from the list of genes which had the highest fold changes in DE in both generations, both in the between- and within-population comparison. Note that only 281 of these were significant in parents and 1674 in offspring. For each of these genes, 50-75 bp-probes representing a region spanning from -7.25 kb upstream to +3.25 kb downstream of the transcription start point (hence mostly covering promoter regions and other cis-acting regulatory elements) were placed on a custom made tiling array. Methylated DNA immune precipitation (MeDIP) was used to enrich methylated DNA fragments, and after labelling and hybridisation, the relative levels of methylated to un-methylated DNA was assessed for each probe.
Out of the 3623 selected genes, 239 were significantly DM (FDR-corrected P < 0.05) when comparing RJF and WL parents, and 821 were DM in the corresponding comparison in the offspring. A smaller number were classified as DM when comparing between families within population (Table S2). A heat map of the genes classified as DM in both generations showed a highly consistent pattern across generations (Figure ). Furthermore, DM levels were significantly correlated between generations when comparing RJF with WL (Figure ), and also to a lesser degree when comparing WL, but not RJF families (Figure ).
Of the 145 genes which were significantly DM in both generations (Additional file 4
; Additional file 5
), 79% were hypermethylated in WL (Figure ). This is a highly significant bias (χ2
= 49.8, P < 0.0001), indicating that this breed has acquired novel methylation patterns during its selection history.
We further analysed the relationship between DM and DE on the 3623 selected genes. There was no overall correlation between the level of DM of a gene (% of DM probes) and the degree of DE of the same gene (Additional file 4
). Furthermore, there was no overrepresentation of DE genes among the top 100 DM promoters when compared to a random sample of 100 DM genes (χ2
= 2.1, P > 0.05). This is contrary to the common notion that methylation causes down-regulation of gene expression, but similar findings have recently been reported from other species, for example humans [16
]. The finding is quite surprising, and indicates that the specific sites of methylation may be of major importance for gene regulation. For example, there may be a substantial difference between methylation of transcription factors compared to insulator sequences. Since we only analysed a 10 kb region around the transcription start site of each gene, we can not exclude that DM in other, more distant regulatory regions may be more closely connected to the expression level.
To illustrate examples of the transgenerationally stable methylation patterns observed, we show methylation graphs for four genes (ABHD7, GAB1, KSR1
) in Figure . In all four, the methylation pattern was reliably inherited, shown by the fact that the DM pattern was highly similar in parents and offspring. ABHD7
showed extensive DM ranging several kb downstream of the transcription start site. In none of the four genes, the significantly DM loci were in CpG-islands, so methylation must have targeted cytosines in other genomic contexts. Extensive methylation of non-CpG regions have recently also been reported for the human methylome [16
], and it remains unknown which functions these epigenetic variants may serve.
Figure 3 Transgenerational stability of methylation patterns in specific genes. a-h, Differential methylation levels (Log2 fold change) of promoter regions, comparing Red Junglefowl and White Leghorn, are shown with a resolution of 50-75 bp-regions in parents (more ...)
Verification of differential methylation with independent animals, tissues and method
To verify the results of the array-based methylation analysis, we arbitrarily selected four genes, which were DM on the tiling arrays in either parents or offspring, FUCA1, PCDHAC1, TXNDC16, and RUFY3, and replicated the findings for those, using a different technique and a different animal material. Hypothalamus/thalamus regions from eight five-weeks old RJF and eight WL (same strains as earlier, but different parents) were dissected and treated as described above. The DNA was bisulfite-treated, and the degree of methylation was determined in the regions that were significantly DM on the tiling array using methylation sensitive high resolution melting (MS-HRM) analysis.
All four genes were significantly DM in the same direction as found on the tiling array (FUCA1 and PCDHAC1 hypermethylated in WL; RUFY3 and TXNDC16 hypomethylated) (Figure ). This suggests that the tiling array produced reliable results and that the observed methylation differences are representative for the population differences at large.
Figure 4 Verification of differential methylation of four arbitrarily chosen genes. a. Methylation differences between Red Junglefowl and White Leghorn in three different tissues, estimated by methylation sensitive high resolution melting (MS-HRM) analysis. The (more ...)
In order to check for tissue-specificity of the DM, we also performed HRM on the same four genes, using DNA-pools prepared from cerebellum and blood from the offspring samples included in the tiling arrays. All four genes were significantly DM in cerebellum. In blood, FUCA1 and PCDHAC1 were significant, while RUFY3 showed a tendency for DM (P = 0.08) (Figure ). The fact that TXNDC16 was not DM in blood indicates that this gene shows tissue-specific, heritable methylation.
Genetic stability of methylation differences
There is a risk that the methylation differences detected by the MeDIP technique could be affected by sequence differences in the promoter regions used for the arrays. To exclude this possibility, we used the recently published resequencing data of Red Junglefowl and domestic chickens [18
] to check the 145 significantly DM probes in both parents and offspring for possible deletions, insertions and SNP density. Apart from occasional SNPs (Additional file 5
), no major sequence differences were detected.
The methylation differences observed may be a result of either inheritance of the epigenetic changes independently of genetic changes, or result from sequence differences which secondarily affect methylation at close or remote loci. This is more difficult to differentiate, since it would require extensive resequencing data of the individuals actually used in the study, combined with, for example, methylation QTL-studies.
To suggestively analyse whether differential methylation of specific loci are caused by sequence differences we decided to study its genetic stability and segregation over several generations. For this purpose, we used a total of 18 birds from the eighth generation of an intercross between RJF and WL. In this population, genetic recombinations in each generation have broken up the linkage between adjacent loci, and we could therefore check for both stability of the methylation sites, and for possible cis- or trans-regulation of these.
From this group of advanced intercross birds, we selected individuals, which were homozygous for either the WL or RJF-allele, or heterozygotes, of an SNP located within 176-1449 kb of the locus showing DM. Using HRM analysis on DNA from blood, extracted from these different genotypes, we again analysed the methylation on FUCA1, PCDHAC1 and RUFY3 in these individuals (Figure ). For FUCA1, we found two different non-significant, but distinct, methylation levels, where the birds homozygous for the WL-marker were hypermethylated, and heterozygotes were similar to the ones homozygous for the RJF-marker (P = 0.07). With respect to PCDHAC1, the three genotypes were significantly different (P < 0.001), with heterozygotes having a methylation level falling between the hypermethylated WL homozygotes, and the RJF genotypes. RUFY3 showed a high level of methylation, which was not significantly different between the three genotypes. Hence, two of the three DM loci were stable over the eight generations of intercrossing, and tended to segregate according to genotype at the locus. This is consistent with a cis-regulating mechanism, showing a dominant inheritance of hypomethylation in genotypes with RJF alleles for FUCA1, and an intermediate, codominant inheritance in PCDHAC1. RUFY3 may possibly be under control of trans-acting loci, which have segregated during the intercrossing.
Although these results are not conclusive, they suggest that sequence differences may determine the DM for at least two of the three loci, possibly for all of them. This further suggests that selection during domestication may have targeted genotypes which modify the epigenomes, perhaps affecting phenotypes indirectly.
To examine which genetic pathways and functions that may have been affected by DE and DM, we performed a gene ontology (GO) analysis. We analysed the DM and the DE genes in each generation separately, and then selected those GO-terms and KEGG pathways (P < 0.1), which were significantly enriched in both generations (Additional file 6
A majority of the enriched GO terms are related to phosphorylation and kinase activity, important aspects of intercellular signalling. Looking specifically at the KEGG pathways enriched among DM and DE genes in offspring only (where the biological sample is considerably larger), the analysis shows that MAPK signalling pathway (which, for example, is associated with stress responses), long-term potentiation (affecting memory consolidation), neurotrophin signalling pathway (involved in neural differentiation) and GnRH signalling pathways (related to reproduction) are enriched. All these are potentially interesting from a domestication perspective, in that they may be related to well documented differences between RJF and WL in stress tolerance, behaviour and reproduction.
Over-representation of epigenetically affected genes in selective sweep regions
We considered that the epigenetic differences between the layer breed and their ancestor could reflect general effects of selection during domestication, as suggested above, perhaps being related to differences in the domestication induced phenotypes, such as growth, feeding behaviour and social tolerance. If so, we would expect the epigenetic differences to be accumulated in genomic regions which have been under selection during domestication. Therefore, we compared our data to one of our earlier, and recently published, datasets on chickens [18
]. This dataset consists of an extensive list of selective sweeps related to chicken domestication, based on resequencing of populations of RJF and a number of domesticated breeds. In total, 149 selective sweeps present in all domestic chickens, and 134 present in egg laying breeds only, were used. A sweep was defined as a 40 kb region where the heterozygosity Z-score was below -4.
There were 216 DE genes (DE in both generations) with annotated loci within the 975 Mb of the genome covered by the sweep analysis. Five of them were situated within 50 kb of selective sweeps present in all domestic chickens (non-significant association, based on a permutation test; P > 0.1), and nine in the laying breed sweeps (significantly more than expected by chance; P < 0.05).
We performed the same analysis on 134 DM loci, and found that four were within 50 kb of sweeps in all domestic chickens (non-significant; P > 0.1), and six in laying breed sweeps (P < 0.05). The significant overlapping genes in laying breed sweeps are shown in Additional file 7
It is interesting to note that ABHD7
, which was the strongest DM and one of the strongest DE genes in our experiment, is positioned in a laying breed sweep. This gene is named EPHX4
in humans, and is related to detoxification of exogenous chemicals [19
]. Based on its position in a selective sweep, and its differential methylation and expression, it would appear that the epigenetic variant of the gene (or the genotype affecting the epigenetic state of it) may have been selected during domestication. KSR1
, an important gene in MAPK/Ras dependent signalling [20
], as well as ADRA2C
, an alpha adrenoreceptor that may be related to egg laying [21
] and regulation of the sympathetic stress reaction [22
], are also situated in laying breed sweeps.
Although our data do not allow us to conclude on which genes and which sweeps that are associated with specific phenotypes, they suggest that selection of epigenetic variation may have been an important part of chicken domestication.
Our findings show that differential methylation and gene expression in hypothalamus/thalamus are abundant in a comparison between domesticated White Leghorn layers and their wild ancestors, the Red Junglefowl. Many of these epigenetic differences are inherited, demonstrating transgenerational stability. It is possible that these differences are a result of selection during domestication, targeting either sequence differences which affect epigenetic states of specific loci, or epigenetic states which are not related to sequence differences.
The causal relationship between methylation and gene regulation is not clear, since differential methylation was associated with both up- and down-regulation of the gene expression, or did not affect it at all. Since similar dissociation between methylation and gene expression has recently been found in the human genome as well [17
], this indicates that epigenetic regulation is more complex than previously assumed. Whereas it is often believed that methylation of promoter regions is associated with down-regulation of gene expression, our results indicate that gene regulation is more complex than so. For example, chromatin structure may be more important than commonly assumed. Furthermore, we found that CpG-islands are not always methylated, so there may also be evolutionary contraints on methylation sites, hence affecting the rate with which epigenetic adaptations may occur in different parts of the genome. Although speculative, these issues should be considered in future research.
Some of the methylation differences observed appear to be tissue-specific, whereas others affect a wider range of cells. The mechanism whereby differential methylation at a particular locus only in, for example, the brain can be transferred from parents to offspring remains elusive. In Drosophila
, similar observations have been made with respect to gene expression, where induced differences specifically in the brain can be transmitted via sperm and cause tissue specific effects in the next generation [24
]. Possibly, microRNA regulation may be involved [25
], and both sperm and eggs may also transfer specific histone variants [26
]. There is also a close link between genetics and epigenetics, in that the epigenetic state of a particular locus is determined by both genetic and epigenetic variations at other loci [23
Stable inheritance of epigenetic variants has been demonstrated in plants [7
]. Also in mammals (mainly in rodents and humans), there is increasing evidence that this occurs widely [10
]. Our results are the first to demonstrate the same in birds, and furthermore show a long-term stability over several generations of specific methylation states.
Although we have only studied one population each of Red Junglefowl and domesticated chickens, the observations in this experiment could indicate that selection of favourable epigenomes, or genotypes favourably affecting the epigenome, may have been an important aspect of chicken domestication. However, further studies are needed, where methylation of specific genes are analysed in a wide range of domesticated populations, analogous to the recent study of sequence variation in the domesticated chicken genome [18