By diverting high-throughput genotyping arrays, we explored the putative imprinted status of candidate human genes.
Historically, the first maps of imprinted genes have been established thanks to the use of mice harboring uniparental disomies.24
These large and complex chromosomal rearrangements together with ad hoc crosses allowed the production of mice carrying a large genomic region in the expected double copies but inherited from only one or the other parent. The resulting phenotypes were severe and allowed the identification of large clusters of imprinted genes, controlled by an imprinting control region (ICR). Putative imprinted genes, disseminated across the genome in opposition to those located in clusters, might have been missed by this screen as they are unlikely to participate to such large phenotypic syndromes and as they are probably not under the control of a long range ICR. In addition, imprinted genes having a restricted expression profile, either spatially or temporally, or with an imprinted pattern limited to a precise developmental stage, organ, or isoform, are probably very difficult to identify, particularly in humans. For example, in spite of controversial reports suggesting or excluding an imprinted status for the well-studied retinoblastoma gene RB1, it is only very recently that it was shown to undergo genomic imprinting.25
Attempts to exhaustively identify imprinted genes have been developed, either using computational approaches or with high throughput strategies.9,10,13-15
If bioinformatical screenings have suggested that imprinted genes might be much more numerous than the actual known list, up to 600 in mice12
and 150 in humans,11
validation analyses to confirm the imprinted status of the candidate genes have not been very successful and the list has not increased much. A previous approach using genotyping arrays failed to reveal new imprinted genes.26
However, a similar screening led to the identification of the random monoallelic pattern of lymphocytes.27
Difficulties to identify new imprinted genes partially rely on the lack of strict characteristics of sequence, gene organization, etc., common to all imprinted genes and that could be used as criterion to identify new candidates.
High-throughput techniques as in this work allow the screening of a huge number of genes simultaneously, though the location of these SNPs, not gene-focused, is not ideal. Therefore, the background is high, particularly for the hybridization of cDNA and needs a careful selection of pertinent data. We used filtering steps to concentrate on the more robust set of data in terms of quality and reproducibility of the results and performed a complete validation step.
By sequencing the cDNA of placentas previously genotyped as heterozygous for SNPs located within exons, we could distinguish true or false candidates. Our validation step allowed the exclusion of some false positive genes. These might be due to a lack of sensibility of the technique; indeed, the software is designed to provide the genetic composition of genomic DNA (heterozygosity/homozygosity) whereas cDNA signals might deviate from this clear-cut situation and lead to a misinterpretation. In addition, our validation approach concentrates on the analysis of the major isoform or all isoforms of a given gene and might therefore miss some imprinting effects that could be specific of a particular alternative isoform of the gene (potentially targeted by the array’s probe), as already observed for GRB10
, for example.28
In our screen, ZFAT
, the best candidate from the arrays, was targeted by six monoallelic SNPs, in at least two of the five placentas under study. It is noteworthy that these six SNPs are located in intronic regions of the ZFAT
gene. Re-examination of the complete data showed that six additional SNPs, also located in intronic sequences, presented a monoallelic pattern in the only informative of the five placentas. It is therefore likely that, as we hypothesized prior to undertaking this study and as already mentioned,27
a significant portion of pre-messengers is present in the RNA pool. This contributes efficiently to the hybridization process and produces consistent and robust results leading to the identification of monoallelic genes. This result strengthens the choice of the Affymetrix genotyping arrays that unfortunately harbor only 5% of coding SNPs, but that contain a large proportion of intronic SNPs.
This approach led to the confirmation of known imprinted genes such as INPP5F
. This latter gene, a zinc finger transcription factor, had been cited as imprinted in lymphocytes21
and was recently identified by screens for differentially methylated regions in uniparental disomies29
and hydatidiform moles.16
We confirmed that this gene was also imprinted in the human placenta and that the expressed allele was the maternally inherited one, as recently reported.29
In the murine placenta, the Zpf597
gene displayed a biallelic expression. Homozygous mice invalidated for this gene (also known as HIT-4) are early embryonic lethal whereas heterozygous mice exhibit abnormal behaviors and abnormal structures in the brain.30
However, the precise role of this gene in both placental and brain function remains to be analyzed.
Besides, seven novel human imprinted genes were found, which represents an increase of about 10% relative to the today list of human imprinted genes. Little is known about the genes we identified and their precise role in the placenta generally remains to be characterized. Neurotrimin (NTM) is a cell adhesion molecule that seems to participate in synaptogenesis.31
MAGI2 is involved in infantile spasm.32
ZFAT is a widely expressed transcription factor from the zinc finger family that appears to play a role in immune functions and apoptosis.33,34
GLIS3 is implicated in neonatal diabetes and pancreatic development.35,36
This pathway has a considerable importance in placental physiology, since glucose transport and insulin regulation are considerably affected in growth restricted fetuses, as reviewed in.37,38
In addition, GLIS3 is strongly expressed in endothelial cells, an issue of relevance for the normal placental physiology, also encountered for ZFAT. LIN28B controls the expression of the let-7 miRNA and is implicated in cancer progression and glucose metabolism.39,40
was found associated with age at menarche in various populations.41
, the paralogous gene of LIN28B, has been shown to control IGF2 expression level at the post-transcriptional level.42,43
The expression of LIN28B
is in addition correlated with IGF2
expression, at the mRNA and peptidic level in ovarian cancers, an interesting connection between imprinted genes.44
gene has been shown to regulate inflammatory pathways in mice.45
The balance between pro- and anti-inflammatory pathways is an issue of extreme importance in human placental diseases, where pro-inflammatory protein encoding genes are abnormally upregulated in pathological pregnancies, especially in preeclampsia.46,47
The imprinted status of inflammatory factors may thus be relevant to adjust the amount of protein synthesized. Some of these seven genes have a very strong expression in the placenta (ZFAT
) while others are barely expressed in this tissue (NTM
) (data obtained using Unigene and Nextbio). A cerebral expression profile is also common. Population studies have highlighted SNPs in some of them in association with growth related phenotypes.41,48-50
Among several independent placenta samples, a perfectly monoallelic expression could be seen for five of the seven candidate genes, showing a strict imprinting. In these cases, either one or the other allele could be observed in the cDNAs, rejecting the hypothesis that one of the nucleotidic variations could be responsible for a functional effect resulting in the decrease of the expression level of one copy. For two genes (NTM
), however, variability among individuals was observed, as some exhibited a strict monoallelic expression, while others maintained a biallelic profile of the same gene, tested under the same conditions. This phenomenon of polymorphic imprinting had already been observed for the IGF2R gene that is imprinted in mice but shows a variable pattern in humans.51
The mechanisms underlying this variability are so far unknown.
Genes identified by this screen are located away from known imprinted genes or clusters. As no other signals in their neighborhood could be deduced from the arrays, they could therefore a priori belong to the class of isolated imprinted genes. It is striking to find four out of seven new imprinted genes coding for transcription factors with zinc finger domains. Analysis of 3′ exons of genes from this family had highlighted the coexistence of different and contradictory histone marks that could be comparable to those of imprinted genes.52
The expression of a selection of ZF genes was however found biallelic by these authors and the presence of these marks was rather correlated with the repeats of ZF domains. In addition to the genes identified in this screen (ZFAT
), other ZF genes are known to be imprinted: ZNF264
, among others.
Conservation of the imprinted status is not strict between mammalian species. We explored murine placentas and observed a biallelic profile for most of the newly identified imprinted genes. Therefore, the placental imprinted profile of these genes seems to be specific to the human species. The L3MBTL1
gene, coding for a zinc finger transcription factor and potential tumor suppressor, is the only other case of an imprinted gene (also expressed from the paternal copy in human tissues) but escaping imprinting in mice.53
However, it is difficult to refute or affirm categorically the imprinted status of a gene, particularly in the human species, as this phenomenon can be very restricted in its temporal and/or spatial profile.
gene, however, seems to be also imprinted in the murine and bovine species. This gene produces three alternative isoforms, all of which seem to be imprinted from our results. A partial knockout mouse was performed that lacks the expression of the longest isoform (S-SCAMα) but maintains the two others (S-SCAMβ and γ).54
No reproduction failure was reported but these mice die shortly after birth and seem to have abnormal signaling responses in dendrites. It is likely that this isoform, whose function is necessary in the brain, is not crucial in the murine placenta and that the presence of the two other isoforms is sufficient for the placental function of the Magi2
gene, whatever it is. In humans, deletion of this gene was associated with infantile spasms.32
It was a surprise to see that the human MAGI2
gene expressed the paternal copy while in murine placenta and bovine tissues, the active copy was the maternal one. This situation has already been described for the ZIM2
and some imprinted genes also show variable imprinting effects according to different isoforms and tissues.
Imprinted genes have in common a complex epigenetic regulation involving different mechanisms including antisense, miRNAs, differential methylation and histone modifications. Though not a strict requirement, the presence of antisense RNA is a hallmark of imprinted genes. ZFAT-AS1
is a 2-exon gene transcribed from the opposite strand and partially overlapping some ZFAT
exons. We could show that this antisense gene also presents a monoallelic expression in the placenta and that the expressed copy is also of paternal origin. According to the ncRNAimprint database, all antisense RNAs described in the context of imprinted genes present this paternal specificity.56
The antisense ZFAT-AS1
RNA together with the described antisenses for GLIS3
seem to be human specificities absent in other mammalian species, from available sequence data of these genomic regions. The nearest genes distal to ZFAT
, located about 80 kb upstream, are two miRNAs. The ZFAT
gene has been hypothesized to be a potential target of these miR30B
). These miR30 RNAs, however, putatively target other genes, including the known imprinted genes RB1
. The small size and the paucity of frequent SNPs within these sequences make it difficult to analyze the expression profile and the potential imprinted status of these miRNAs, whose interactions with ZFAT
might be of great interest.
The LIN28B and ZFAT genes were further studied in order to characterize their potential role in placenta function, as an imprinted gene.
We then explored the methylation status of the regions surrounding ZFAT and LIN28B. If the epigenetic control of known imprinted genes, generally located within clusters under the control of an ICR, is difficult to fully characterize, it appears even more difficult to anticipate how an isolated imprinted gene is regulated and at which distance are the crucial elements responsible for the differential imprints affixed during gametogenesis. In spite of this apparent lack of differentially methylated regions in ZFAT CpG islands, binding sites for transcription factors known to regulate imprinted genes can be detected in the ZFAT promoter, such as CTCF and YY1. In addition, other mechanisms than DNA methylation could be involved in the regulation of the ZFAT expression profile. For LIN28B, the CpG island located around the promoter showed, on the contrary, a differentially methylated profile that is likely to be an important regulatory region, a DMR as observed in most imprinted genes.
We then explored the imprinted status of ZFAT
in other human tissues but failed to observe a monoallelic expression in other tissues than the placenta: in lymphocytes and endometrial tissue, ZFAT
is strictly biallelic while LIN28B
is not expressed. Therefore, the imprinted profile of ZFAT
seems to be rather specific and may be restricted to the placenta. However, ZFAT
could be successfully amplified from different tissues and also seems expressed in many tissues including spleen, thymus and brain.57
Other genes are also known to have a restricted imprinted profile whereas others maintain their imprint within the complete body58
; even IGF2
escapes imprinting in the brain.59
Null mice for Zfat
present an embryonic lethality and do not survive after E8.5.60
The phenotype is not observed in heterozygous mice, in agreement with our finding of a biallelic expression of Zfat
in mouse. A placental phenotype was observed, as the spongiotrophoblast does not develop correctly. At E8, Zfat
is normally expressed in the endothelial and hematopoietic progenitors of blood islands but the authors show in null mice an impaired differentiation of hematopoietic progenitors in blood islands of the yolk sac. ZFAT is also suspected to play a role in angiogenesis and hematopoiesis.22
By immunohistochemistry we could find ZFAT expression in endothelial cells and syncytiotrophoblast. In addition, the gene was consistently downregulated in placentas from preeclampsia and/or IUGR, whereas this gene was described as stable along the human gestation period.61
As a loss of imprinting would rather be synonym of increased expression, this deregulation is likely due to a transcriptional mechanism in the pathological context. Very interestingly, a recent study revealed that SNP variants within the ZFAT
gene are associated with hypertension.62
Our data, together with the literature coincide to propose ZFAT as a regulator of the differentiation of endothelial cells within and outside the placenta. This is consistent with the role of this gene in the assembly of endothelial cells during the angiogenesis process, and the formation of capillary networks, particularly in HUVECs.22
appears as an interesting candidate to have a role in placental development that needs to be further characterized. Labeling of slides from pathological placentas might bring additional information on this abnormal expression profile.
In conclusion, in this work we could validate and identify new human imprinted genes whose function in the placenta remains to be explored. Increasing the number of identified imprinted genes and completing exhaustively their list will help to extract the specificities of this class of particular genes in terms of epigenetic regulation, physiological function and pathological alterations. These data are of particular interest in the present context, where approaches using new technologies such a RNA-seq to identify imprinted genes in mice tend to question some previously known imprinted genes and failed to identify many new imprinted genes.63-65
Indeed, some genes exhibiting a maternal expression could have been misinterpreted as imprinted after contamination of the murine placental tissue by highly expressing maternal uterine cells. The exploration of human placental tissue circumvents this problem, as its size and volume make it easier to avoid maternal contaminations.