Our data demonstrates that genome-wide prediction of imprinted genes is not yet reliable, at least for expression in the adult brain, and that validation of candidate imprinted genes is very important. Of the genes we tested, only the previously known imprinted genes, Grb10
, were imprinted in the brain and none of the reported candidates from the literature2, 4
on chromosome 11 showed evidence of imprinting. Our data support previous findings that only a small percentage of the candidate imprinted genes identified by Nikaido et al are confirmed as imprinted in normal hybrid embryos.25 Grb10
has been shown to have complex imprinting patterns in different tissues, being paternally, maternally, or biallelically expressed depending on the tissue studied19–21
so it certainly remains possible that these candidate genes show imprinted expression in tissues other than the brain. It is important to note that it is not possible to prove that a gene is never imprinted, since imprinted expression could be specific to a different tissue, time of development, or possibly to the strain background. However, our results provide evidence that these genes are not universally imprinted and can show biallelic expression in the brain. This presents the interesting possibility that as more candidate imprinted genes are validated in different tissues it may be possible to develop machine learning programs to identify the tissue-specific imprinting signatures surrounding these genes. Studies of parthegenotes and androgenotes will not distinguish these tissue-specific imprinted genes, and so it is important to develop methods such as the one we present here to look at imprinted expression in specific tissues.
It is interesting to note that for the genes on chromosome 11 identified as imprinted by the studies of Nikaido et al and Luedi et al, there were only two genes found in both datasets, suggesting very little overlap of these two methods. We were unable to test Mtmr3
due to a lack of exon SNPs polymorphic for the strains we studied. We tested Ikzf1
and found no imprinting or strain bias. The computer model of Luedi et al predicted that Ikzf1
should be maternally imprinted, whereas the experimental data from Nikaido et al demonstrated expression from the paternal genome. In addition, a recent study26
examined the differential expression of genes in mice with maternal or paternal duplications of chromosome 11 and also found paternal expression of Ikzf1
in the brain. The Nikaido study uses androgenotes and parthegenotes that may not undergo normal maintenance of imprinted genes through the process of manipulating embryos,27
whereas the Schulz study uses translocated chromosomes to generate uniparental disomies. In the case of uniparental disomies it is not clear whether the translocated chromosome pieces are packaged properly in the nucleus with the homologous chromosome or whether they are located with the centromeric chromosome in the nucleus, and whether this could potentially lead to subtle changes in gene expression.28
It is therefore formally possible that the expression of Ikzf1
is more sensitive to changes in chromatin packaging. There is precedence for this in the work of Ling et al who showed that normal expression of the paternal Nf1
allele depends on the stable interaction of the paternal mouse chromosome 11 and the maternal mouse chromosome 7 through the CTCF insulator protein, whereas expression of the maternal Nf1
allele is not affected by this interaction29
It is also unclear how a mixed strain background could affect the interpretation of imprinted candidates in these different experiments, given we see very clear effects of strain polymorphisms on expression on many genes.
The parental expression bias we see in our data ranges from very subtle (3% difference in allelic expression) to quite substantial (54% difference). Although these differences are statistically significant, it is unlikely that the smaller differences affect tumor susceptibility or other biological function. We report these subtle differences here because they may reflect subtle differences in the epigenetic marks surrounding these genes that could explain why they were identified in the studies of Nikaido et al and Luedi et al. For most of the genes that show a strain-specific bias, there are polymorphic SNPs between B6 and A/J or 129S1/129X1 in the 5′UTR and intron 1 (Jackson Laboratory Mouse Phenome Database SNP website, http://phenome.jax.org/pub-cgi/phenome/mpdcgi?rtn=snps/wiz1
), suggesting that the difference in expression levels could be due in part to cis-acting polymorphisms affecting transcription. Two exceptions are Havcr2
that has SNPs between B6 and A/J in intron 2, but not intron 1 or the 5′UTR, and Serpinf1
that has a synonymous coding SNP polymorphic between B6 and A/J in exon 8. These two genes could be differentially expressed due to polymorphisms in trans-acting factors, or other more indirect effects.
We have previously shown using quantitative PCR and SYBR green detection methods that the level of expression of Nf1
in wild-type B6 brains is 30% higher than wild-type 129 brains.24
In these experiments using Taqman quantitative PCR on B6X129S4 F1 brains we find higher expression from the B6 allele than from the 129S4 allele ranging from 5–16% depending on the experimental run. In both experiments the B6 allele is expressed at a higher level than the 129S4 allele, however the size of the effect is smaller in the F1 hybrid than in the inbred strains. This may be due to trans-acting factors that are matched to the strain of the Nf1
gene, such that there is a mixture of trans-acting alleles in the F1 hybrid and a dampening of the expression difference.
It is interesting to note that we see a similar strain effect on the Trp53
gene, such that the B6 allele is expressed 24–54% higher on average than the 129S4 allele depending on the experimental run. It has been shown that the phenotype of Trp53
null mice differs on a 129 inbred strain background compared to a B6,129 mixed strain background30
and this data suggests that there may be additional strain specific effects in the presence of p53. In the case of both the Nf1
tumor suppressors, expression is higher in the brains of mice that are more susceptible to brain cancer16
suggesting that the higher level of the tumor suppressor gene is not protective against cancer. We hypothesize that the B6 strain may have evolved higher levels of oncogenic factors in balance and opposing Nf1
, such that when the tumor suppressor genes become mutated or lost the oncogenic effects are more pronounced in the B6 strain, increasing susceptibility to cancer. Additional experiments are needed to test whether this accounts for the difference in strain susceptibility to cancer.
Although we have not identified any novel fully imprinted genes in this study, the validation of candidates and demonstration of their biallelic expression in the adult brain will be useful for refinement of future computational predictors of imprinted genes. In particular, this method of expression analysis distinguishes imprinting from other variations in expression, and provides support to pursue further proof of imprinting by bisulfite analysis of genomic DNA. These data also emphasize the importance of assaying imprinted genes on well-defined genetic backgrounds to separate out strain-specific effects from parental origin effects. There remain 120 candidate genes on chromosome 11 from the studies of Nikaido et al and Luedi et al that may be imprinted; however, because 30 out of the 30 we tested were biallelically expressed, it may be worth refining the methods of predicting imprinted candidates before analyzing these additional genes. Finally, our data suggest that Grb10 remains a top candidate for the imprinted effect on chromosome 11 responsible for variation in tumor susceptibility in Nf1−/+;Trp53−/+cis mice.