Of the 35,105 putative promoters assayed in our ChIP-seq analysis of M. domestica fibroblasts, only ~46% (16,320) were marked by H3K4me3. This fraction is considerably smaller than the 74% and 71% of promoters marked by this expression-associated modification in cultured human and mouse cells, respectively [29, 45], and is most likely an artifact of inaccuracy in the annotation of the M. domestica gene set. The initial set of predicted protein-coding and non-coding genes was produced by analyzing similarity with well-annotated eutherian gene sets, a practice that is expected to underrepresent or overlook diverged orthologs, paralogs, and marsupial-specific genes [46, 47]. Further annotation has relied on individual sequencing of genes-of-interest, as well as a small number of RNA-seq data sets that are enriched for the 3’ ends of genes, leaving the 5’ annotation of many genes incomplete or inaccurate. This issue was underscored by a recent, comprehensive RNA-seq study of the M. domestica X chromosome  in which we found that the 5’ ends of nearly half of the genes on the X chromosome are incorrectly annotated in the MonDom5 assembly, with ~30% having a transcription start site more than 5 kb upstream from the first annotated 5’ exon. Annotation issues of this kind, especially at the 5’ ends of genes, pose a significant challenge for correlating promoter histone modification states with transcriptional states. In light of this limitation, our results likely underestimate the number of opossum promoters marked, either independently or concurrently, by MOAs and/or H3K9me3.
We were, nevertheless, able to identify 179 genes that were concurrently marked by MOAs and H3K9me3 within 5 kb of an annotated 5’ exon. Twenty-one of these were expressed in fibroblasts and had an informative SNP in each reciprocal cross. Importantly, only six of them showed 100% overlap of significant peaks of H3K4me3, H3K9Ac, and H3K9me3, and half of these exhibited strongly biased allele expression, of which at least one, Meis1, was clearly expressed in a parent-of-origin specific manner; i.e., is imprinted. None of the 11 candidate genes with less than 100% peak overlap exhibited monoallelic or strongly skewed expression. The high frequency (50%) of monoallelic expression among genes with 100% overlap of transcriptionally opposing histone marks suggests that complete peak overlap be adopted as an essential criterion in future ab initio searches for imprinted genes in non-eutherian species.
It is important to note that opossum Meis1 expression occurs in vivo as well as in cultured fibroblasts. For example, in a study unrelated to the present one (and unfortunately uninformative for imprinting analysis), Meis1 transcripts were found to be abundant in RNA-seq reads in cDNA prepared from gestational day 13.5 (E13.5) opossum brain and extra-embryonic membranes (X Wang, KC Douglas, AG Clark, PB Samollow, unpublished data). In addition, expression of the Meis family of genes in eutherians is strongly developmental-stage and cell-type specific, and inspection of transcript contig assemblies in the genome browser features of OpossumBase (http://opossumbase.org/?q=genome_browsers)  indicate that expression of Meis1 is also developmentally variable and tissue-specific in opossum. The OpossumBase contigs were constructed from normalized cDNA libraries and are not useful to gauge expression levels, but they do verify that Meis1 transcripts were sufficiently abundant to construct full-length mRNA transcript contigs from E9.5 embryo and E12.5 fetus samples, and from 1-day and 12-day post-partum newborns, but not from 25-day newborns. Similarly, transcripts sufficient for full-length contig assembly were present in adult ear pinna, thyroid, eye, tongue, heart, pancreas, stomach, colon spleen, ovary, and skeletal muscle, but not in adipose, brain, lung, diaphragm muscle, liver, kidney, or testis samples. In the present study, using standard PCR protocols, we unable to detect Meis1 3'UTR transcripts in adult opossum liver, kidney, and heart in cDNA prepared from four F1 animals from our reciprocal crosses (Figure 4B). This outcome agrees with the contig profiles in OpossumBase for liver and kidney, but not for heart; which suggests that Meis1 is expressed in heart, but at levels too low for detection by routine PCR amplification from non-normalized cDNA.
Of the two remaining monoallelically expressed genes, Cstb clearly showed allele-biased expression that was independent of parent of origin, while imprinted vs. allele-biased expression of Rpl17 could not be distinguished due to lack of reciprocal allelic transmission data. The possibility of different underlying causes for monoallelic expression emphasizes the importance of conducting reciprocal crosses to detect genuine parent-of-origin-specific expression patterns, a practice that has been absent from many past studies of marsupial imprinted genes.
Assessment of the transcriptional state of these three monoallelically expressed genes reveals the first case of an imprinted gene in a marsupial that is not known to be imprinted in any other organism, and suggests a role for histone modification states in the occurrence of monoallelic-expression of genes in the opossum and perhaps other marsupial genomes. Contrastingly, methylation analysis of gDNA from these fibroblasts failed to find evidence of DMRs at annotated CpG islands in the promoter regions of this novel imprinted gene or either of the other monoallelically expressed genes, Cstb and Rpl17. This is consistent with past reports that DMRs are rare or absent from marsupial orthologs of eutherian imprinted genes.
Examination of the four previously known annotated opossum imprinted genes, Igf2r, Htr2A, L3mbtl, and Mest failed to detect transcriptionally opposing histone modifications at their respective promoters or their gene bodies. Igf2r is not imprinted in humans but is imprinted in mouse, sheep, dog, and marsupials (wallaby and opossums). In mouse, the transcriptional regulation of Igf2r is controlled by a DMR in intron 2 and by an antisense transcript (Air). Interestingly, the DMR at intron 2 is present in human, mouse, and sheep, but absent in dog and marsupials [49, 50]. Transcriptionally opposing histone states have been associated with the imprinted state, or lack thereof, in human and mouse; but the full-length Air antisense transcript has only been described in mouse [51, 52]. Htr2A, L3mbtl, and Mest show variation of imprinted status in human organs sampled, and are associated with certain disease states that correlate with aberrant DMRs, but no studies of associated histone states have been reported for these loci [53–55].
We were able to assess the imprinting status at the Igf2r locus, but a lack of suitable SNP variants in our animals prevented us from analyzing expression patterns of Htr2A, L3mbtl, and Mest. It is possible that these genes are not imprinted in opossum fibroblasts, in which case the absence of transcriptionally opposing histone modifications would be expected. Alternatively, any or all of these three genes could be imprinted in opossum fibroblasts but not marked or regulated by the specific histone modifications we examined, or DMRs, but rather by some yet-to-be-identified genomic elements or regulatory mechanisms such as non-coding RNA transcripts. If so, there could be additional imprinted loci in fibroblasts that went undetected by our strategy relying on only four histone modifications.
Although Meis1 showed parent-of-origin-specific allele expression in three individual fibroblast cell lines, there was ‘leaky’ expression of the paternal allele in some samples. Leaky expression of the repressed allele has been observed for some imprinted genes in eutherians and for some paternally imprinted X-linked genes in marsupials [7, 8, 56–58]. At the G6pd locus, the degree of paternal allele leakiness is age-dependent, with adults showing greater levels of paternal leakage than fetuses and newborns . Similarly, studies in eutherians have demonstrated a loss of allele-specific gene regulation for X-linked genes in a passage-number-dependent manner in primary cell lines . Although we used low passage fibroblast cell lines, the cells were originally grown from adult tissue, and the combination of adult source and increasing passage could have resulted in higher levels of leakiness. Alternatively, it is possible that the epigenetic regulation of imprinted loci in marsupial cells is not as stable as in eutherians due to the apparent lack of differential DNA methylation at these loci. Furthermore, most studies of marsupial imprinted gene expression have not utilized highly sensitive assays, such as pyrosequencing, to measure allele-specific expression of imprinted genes; so leaky expression of the repressed allele could be more prevalent than previously believed.
The vertebrate Meis gene family comprises three homeobox genes, which act as cofactors for a wide range of Hox genes. Acting alone or in combination with other Hox cofactors, especially members of the Pbx transcription factor family, Meis family genes influence myriad early developmental processes that are essential for body axis patterning and organogenesis [61–66], neurologic (brain, eye) development [62, 64, 66–69], cardiac development and cariomyocyte regeneration [70, 71], hematopoiesis and angiogenesis [61, 69, 71–74], and more in mouse, zebrafish, chicken, and Drosophila. In the absence of protein functional data, we are unable to determine experimentally which ortholog of the vertebrate Meis gene family is represented by the imprinted opossum Meis locus. Nevertheless, a reciprocal blast search strategy using the opossum predicted mRNA and amino acid sequences from the Ensembl annotation indicates that the opossum-imprinted Meis gene shares the greatest sequence similarity with Meis1 orthologs of human, mouse, and rat. Moreover, this locus was matched by the Illumina reads to the exclusion of other Meis paralogs, and comparative synteny analysis shows gene content and order of the genomic region flanking this locus to be strongly conserved with that containing the human MEIS1 and mouse Meis1 genes. Hence we feel confident that this is the ortholog of MEIS1/Meis1.
Although Meis1 expression is crucial in many fundamental embryonic and fetal developmental processes, how its functions relate to current views on the evolutionary advantages of genomic imprinting is not obvious. The highly developed and widely accepted Parental Conflict Model (a.k.a. Kinship Model) proposes that imprinting will be favored when the extraction of resources from a parent by its offspring occurs in such a manner that the fitness benefits of provisioning the offspring differ between the two parents [75–78]. Applied primarily to therian mammals because of their universal placental/lactational provisioning strategy (both eutherians and marsupials form placental attachments), the Conflict Model juxtaposes the reproductive strategies of males and females by noting that offspring of different fathers in multiple-paternity litters compete for the same maternal resources. Maximization of fitness for any one father is achieved by his progeny extracting maternal resources more effectively than the progeny of other fathers, whereas for the mother, the best strategy is to provide resources equitably among all her offspring. This creates "conflict" between paternal and maternal genomes for genes that influence resource allocation and is generally couched in terms of fetal growth regulation, and/or in neurologic development that can enhance or inhibit postnatal growth rates through variation in feeding competence [75, 77–79]. The Conflict Model is pleasingly consistent with the known roles of several imprinted genes in fetal growth and postnatal nutritionally related behaviors, but for most imprinted genes agreement with the Conflict Model has been assumed rather than demonstrated .
The known influences of Meis1 in vertebrate (and Drosophila) development are ambiguous with regard to the Conflict Model. Proper Meis1 expression is essential for body and limb bud axis patterning, brain segmentation, angiogenic patterning, and the proliferation of stem cells in developing organ systems such as retina, heart, and hematopoietic centers. However, Meis1 expression also has inhibitory effects that induce cell-cycle arrest and stem cell quiescence in ways that enable non-proliferative cellular differentiation during and after organogenesis while simultaneously preserving pools of multipotent stem cells for lineage renewal [61–64, 66, 68–74]. Experimental alterations of Meis1 expression are associated with severe, often fatal, prenatal developmental defects in neurologic patterning, the differentiation of hematopoietic stem cells and establishment of definitive hematopoiesis, peripheral angiogenesis, cardiomyocyte cell-cycle regulation, and hematopoietic cancers; but in no case has Meis1 expression been associated with fetal or postnatal growth rates or nutrient acquisition per se. Together with the absence of Meis1 imprinting in human and mouse, the conflicting developmental functions of this gene suggest no obvious reason why it should be imprinted in opossum. Perhaps the explanation for Meis1 imprinting in opossum needs to be sought outside of the confines of the Conflict Model.
In view the logical power and broad acceptance of the Conflict Model, few alternative hypotheses for the advantages of imprinted gene expression have been proposed, and the few that have were quickly dismissed as evolutionarily unstable or logically flawed. Nevertheless, as there is no a priori basis for believing that the Conflict Model must explain every case of imprinting in all species, there could be as-yet-unidentified biological advantages for the imprinting of genes that are not involved in embryonic and perinatal growth and development. It seems prudent, therefore, to remain open to, and actively seek, alternative hypotheses for the evolutionary advantages of imprinting on a locus-by-locus basis, especially in non-eutherian species.