Breeding scheme and maternal small RNAs affect the efficiency of transgenerational inheritance of a paramutation in mice

Paramutations result from interactions between two alleles at a single locus, whereby one induces a heritable change in the other. Although common in plants, paramutations are rarely studied in animals. Here, we report a new paramutation mouse model, in which the paramutant allele was induced by an insertional mutation and displayed the “white-tail-tip” (WTT) phenotype. The paramutation phenotype could be transmitted across multiple generations, and the breeding scheme (intercrossing vs. outcrossing) drastically affected the transmission efficiency. Paternal (i.e., sperm-borne) RNAs isolated from paramutant mice could induce the paramutation phenotype, which, however, failed to be transmitted to subsequent generations. Maternal miRNAs and piRNAs appeared to have an inhibitory effect on the efficiency of germline transmission of the paramutation. This paramutation mouse model represents an important tool for dissecting the underlying mechanism, which should be applicable to the phenomenon of epigenetic transgenerational inheritance (ETI) in general. Mechanistic insights of ETI will help us understand how organisms establish new heritable epigenetic states during development, or in times of environmental or nutritional stress.

ETI is contradictory to the established dogma pertaining to developmental global epigenetic reprogramming events, which occur during preimplantation embryogenesis and primordial germ cell (PGC) development in the fetal gonads 29,30 . It is believed that during the two waves of global reprogramming, epigenetic alterations gained during the lifetime of an individual, are erased and reset, thus preventing potential epimutations from being transmitted to subsequent generations. However, subsequent studies have demonstrated that neither of the two reprogramming events is complete, because many genomic loci, e.g., imprinted loci and retrotransposons (IAPs), appear to be resistant to the reprogramming events 29 . The fact that many environmentally induced epimutations appear to be transmitted across multiple generations suggests that some unknown mechanisms exist to protect those epimutations from being corrected by the global epigenetic reprogramming 16,18,29 .
The Kit locus has been found to be susceptible to paramutations 5,12 . An earlier study reported that an insertional mutation in one Kit allele (a LacZ gene cassette inserted into exon 1 of the Kit gene) caused altered Kit expression from the other allele, leading to a ''white-tail-tip'' (WTT) phenotype in genetically wild type (WT) progeny 12 . Interestingly, direct injection of RNAs isolated from Kit 1/LacZ somatic tissues, or sperm, into WT zygotes, which were derived from parents completely unrelated to the paramutation family, could induce the WTT phenotype, suggesting RNAs are involved in the formation of Kit LacZ paramutation 12 . Subsequently, it was reported that microinjection of miR-1 and miR-124 into WT zygotes induced paramutation-like effects on Cdk9 and Sox9, leading to cardiac hypertrophy and embryonic overgrowth, respectively 10,11 . These phenotypes appear to be transmissible through either the male or female germline for three generations 6,10,11 . These paramutation mouse models not only demonstrate the existence of non-Mendelian inheritance in mammals, but also serve as an excellent tool for studying the underlying mechanism.
We report, here, another paramutation in the Kit locus in mice, which was induced by an insertion of a copGFP gene cassette into the start codon in exon 1 of the Kit gene. We show, here, that the WTT phenotype, although non-specific to the Kit copGFP -induced paramutation, can be used to track the paramutation, because the incidence of this phenotype is far greater than the baseline incidence in regular laboratory C57BL/6J mouse colonies. We also demonstrate that the Kit copGFP -induced paramutation could be transmitted through either the male or the female germline. More interestingly, we demonstrated that this Kit paramutation could be corrected in 3-4 generations if an outcrossing scheme was used, whereas the paramutation phenotype persisted for many more, if not infinite, generations if WT mice with the paramutation phenotype were intercrossed. Moreover, we found that both paternal (sperm-borne) and maternal (oocyte) RNAs could induce the WTT phenotype, but the RNA-induced WTT phenotype failed to be transmitted through the germline.

Results
An insertional mutation in Kit locus induces a paramutation phenotype. We previously generated a knock-in mouse line, in which a DNA fragment containing copGFP (from the copepod Pontellina plumata), an Hprt-PGK cassette, and the SV40 polyA signal, was inserted in-frame, immediately downstream of the start codon in exon 1 (Fig. 1A) 31 . The knock-in allele has been officially named Kit tm1(copGFP)Rosan . For simplicity, we called it Kit copGFP hereafter. The Kit copGFP allele is null, as Kit copGFP/copGFP are not viable and the heterozygotes (Kit 1/copGFP ) display white tail tips, white bellies, and white paws (Fig. 1B, left panel) in either 129Sv/Ev: C57BL/6J hybrid, or pure C57BL/6J background, with 100% penetrance. This line underwent .10 generations of backcrossing onto the C57BL/6J background when the present study was conducted.
Distribution of the Kit copGFP allele among offspring derived from heterozygous breeding pairs followed the Mendelian ratio 31 . However, we observed that ,60% of the genotypically WT progeny derived from the heterozygous breeding pairs (Kit 1/copGFP 3 Kit 1/copGFP ) displayed WTT with various patterns (Fig. 1C, D). Unlike Kit 1/copGFP mice, these WT mice showed neither white bellies nor white paws (Fig. 1B, middle panel). The WTT phenotype is similar to that reported in Kit tm1Alf -induced paramutant mice 12 . To determine whether the WTT phenotype could be diluted through outbreeding, we then set up breeding pairs between heterozygous (Kit 1/copGFP , called HET hereafter for simplicity) and pure black WT C57BL/6J mice (i.e., WT with black tail tip, called WT BTT hereafter) (Fig. 1E, F). Similar to the heterozygous breeding pairs, ,55-57% of the F1 WT progeny displayed WTTs, and these mice are called 1 st WT WTT hereafter for simplicity.
The ''white-tail-tip'' phenotype is not unique to the Kit paramutation family. The WTT phenotype in Kit tm1Alf -induced paramutant mice has been questioned because normal WT laboratory mice of different strains (including C57BL/6J) display WTTs 32 . Indeed, we noticed that many of our pure WT mice, which were on either C57BL/6J or 129/SvEv (Fig. 1B, right panel) background and were totally unrelated to the Kit copGFP line, also displayed WTTs albeit at a much lower incidence. To determine the baseline incidence of the WTT phenotype among WT C57BL/ 6J mice, we set up four types of breeding pairs between WT males and females that were completely unrelated to the Kit copGFP line, including WT BTT males mated with WT BTT ( Fig. 2A) or WT WTT (Fig. 2B) females, and WT WTT males mated with WT BTT (Fig. 2C) or WT WTT (Fig. 2D) females. ,30% of the F1 WT progeny produced by WT BTT mating pairs displayed WTTs ( Fig. 2A), whereas WTTs were observed in ,38-40% of the F1 WT offspring derived from the breeding pairs with one of the parents that were WTT-positive ( Fig. 2B-D). These data suggest that ,30-40% of WT laboratory C57BL/6J mice display WTTs.
Although the WTT phenotype is not specific to Kit paramutation, we did observe that ,55-60% of the progeny derived from one or both heterozygous parents (Kit copGF ) displayed this phenotype ( Fig. 1D-F), and the incidence of the WTT phenotype was significantly greater than the baseline levels (55-59% vs. 30-40%, x 2 test, p , 0.01). Thus, the WTT phenotype represents a convenient and reliable readout for tracking the Kit paramutation phenotype, as long as the baseline levels in the general population are taken into consideration.
Transgenerational inheritance of the Kit paramutation. The paramutation phenotype, i.e., WTT, is present in F1 progeny of heterozygous parents (F0) with a penetrance of ,60% (Fig. 1D-F). However, it remains unknown whether the paramutation phenotype can be transmitted to subsequent generations. To determine the transgenerational inheritance of this novel Kit paramutation, we bred F1 paramutant mice using two breeding schemes: outcrossing and intercrossing. In the outcrossing scheme, we bred 1 st WT WTT mice with WT BTT mice that were totally unrelated to the Kit paramutation family for up to 4 generations (Fig. 3). In contrast, for the intercrossing scheme, 1 st WT WTT siblings derived from HET (Kit copGFP ) parents were intercrossed to obtain F2s WT mice; F3 and F4 WT mice were obtained through intercrossing WT WTT F2 and F3 siblings, respectively (Fig. 4).
When F2 WT WTT females were further outcrossed with WT BTT males, ,64% of the F3 mice displayed WTTs (Fig. 3C), whereas WTTs were seen in only ,45% of the F3 mice derived from WT BTT females mated with F2 WT WTT males (Fig. 3D). The incidence of WTT was significantly decreased from F2 to F3 when the paramutation was transmitted through either the female (from 72% to 64%, x 2 test, p , 0.01) (Fig. 3A, C), or the male germline (from 56% to 45%, x 2 test, p , 0.01) (Fig. 3B, D), suggesting the paramutation phenotype is being ''diluted'' in subsequent generations in the outcrossing scheme. Since F3 mice derived from outcrossing of WT WTT males already displayed WTT levels close to the baseline in WT unrelated C57BL/6J populations (45% vs. 40%), we further outcrossed only the F3 WT WTT females with WT BTT males, which led to the baseline levels of the WTT phenotype (39%) (Fig. 3E).
The outcrossing scheme appeared to ''dilute'' the paramutation. To determine the outcome of intercrossing on transmission efficiency of the paramutation, we bred F1 WT WTT siblings derived from Kit copGFP heterozygous parents. F2 mice from the F1 intercrossing displayed an incidence of WTT at ,70% (Fig. 4A). When F2 WT WTT siblings were further intercrossed, ,76% of the WT F3s displayed the WTT phenotype (Fig. 4B). Similarly, the incidence of the WTT phenotype persisted at a similar rate (,67%) in WT F4s when the WT WTT F3 siblings were further intercrossed (Fig. 4C). Interestingly, when F2 mice derived from outcrossing of F1 WT WTT female were used for intercrossing, ,73% of the F3 WT mice displayed WTT (Fig. 4D). In contrast, when F2 mice from the outcrossing of F1 WTT-bearing males were intercrossed, only ,63% of the F3 mice had the WTT phenotype (Fig. 4E). The difference between the two intercrossing schemes was statistically significant, suggesting that ''stricter'' intercrossing between siblings derived initially from the heterozygous parents can maintain a higher transmission efficiency, whereas any outcrossing in between intercrossing would lead to a decrease in the transmission efficacy.
Stochastic changes in levels of Kit coding and noncoding isoforms in mice carrying the paramutant allele. A paramutant allele usually displays altered gene expression profiles, which are often associated with a phenotype. Therefore, offspring that inherit the paramutant allele may display the phenotype in the absence of the paramutagenic allele (i.e., the copGFP allele in this study). To determine changes in Kit gene expression profiles, we conducted qPCR analyses on five tissues (brain, intestine, skin, testis, and ovary) collected from WT BTT (negative control for the Kit paramutation), WT WTT (Kit paramutation-unrelated, baseline WTT control), HET (mice with both the paramutagenic and the paramutant alleles), and 1 st WT WTT (mice carrying only the paramutant allele) mice. Kit can produce six transcript isoforms, among which only isoforms 1 and 2 encode KIT protein, whereas isoforms 3-6 do not contain ORFs www.nature.com/scientificreports and thus, are noncoding (based on ENSMUSG00000005672.8 in Ensembl Genome Browser). Since large noncoding RNAs derived from an mRNA-coding gene may affect the transcriptional activity of their host gene 33,34 , we examined the levels of all six Kit transcript isoforms using isoform-specific qPCR analyses (Fig. S1). Overall, levels of the protein-coding isoforms (isoforms 1 and 2) were not significantly altered in testis, ovary and brain samples collected from HET (Kit 1/copGFP ) and 1 st WT WTT mice, as compare to the controls, which were either WT WTT or WT BTT mice, both of which were totally unrelated to the Kit paramutation family. A minor, but statistically significant increase in levels of isoform 1 was observed in intestine between WT WTT and WT BTT mice (Fig. S1). Similarly, a moderate, but statistically significant decrease in isoform 2 levels was detected in the brain samples between WT WTT and 1 st WT WTT mice (Fig. S1). Levels of all four noncoding isoforms (isoforms 3-6) were highly variable among samples of three biological replicates, which may explain a lack of statistically significant differences among most of the five organs and four types of mice analyzed. However, a significant increase in levels of isoform 3 was detected in the ovary samples between 1 st WT WTT and WT BTT mice. In skin samples, isoform 4 levels were drastically decreased in HET mice, and levels of isoforms 4 and 5 were increased significantly in 1 st WT WTT mice, as compared to control (WT BTT) mice (Fig. S1). Overall, changes in all six Kit transcript isoforms appeared to be stochastic with a fairly high degree of variation among all five organs and all four types of mice analyzed. Together, these data suggest that the paramutant Kit allele, which is derived from either of the heterozygous parents, may display altered expression in both coding and noncoding isoforms of Kit, but the changes are likely subtle and stochastic in different tissues.
No methylation changes in Kit promoter in sperm DNA, but changes occur in tail DNA samples. We then examined CpG methylation levels of the CpG-rich Kit promoter through bisulfite sequencing. We observed very low levels of CpG methylation in this region, with little variation across individual biological replicates from WT BTT, WT WTT, HET, and 1st WT WTT mice (Fig.  S2A). Only one CpG was consistently methylated across all genotypes.
Sperm DNA possesses different methylation patterns, compared to DNA derived from somatic tissues 35 . However, it has been shown that many murine genes involved in zygotic development remain hypomethylated in sperm 35 . To determine whether this was the case for Kit, we assayed CpG methylation of the Kit promoter in tail DNA samples. Interestingly, we observed slightly higher levels of methylation throughout the Kit promoter region in WT BTT, WT WTT, and HET mice, whereas no methylation above the baseline was observed in 1 st WT WTT (Fig. S2B). However, these patterns were not entirely consistent between individual biological replicates, and were not strictly correlated with the presence of the paramutant allele in HET and 1 st WT WTT mice.
We next examined the ,1.5 Kb CpG-rich region surrounding the Kit promoter through MeDIP-qPCR (Fig. S3A). Both 5 mC and 5 hmC (5-hydroxymethylcytosine) methylation was examined in sperm DNA. 5 hmC modifications are associated with transient or dynamically regulated CpG methylation, and are often the intermediate products in the biochemical reactions that facilitate 5 mC removal or incorporation, and thus, the presence of 5 hmC modifications would be indicative of dynamic methylation regulation in this region 36 . We found no significant differences in DNA enrichment, which was calculated as percent enrichment of diluted input controls, among WT BTT (negative control for Kit paramutation), WT WTT (control for baseline WTT), HET (paramutant allele in the presence of paramutagenic allele) and 1 st WT WTT (paramutant allele in the absence of paramutagenic allele) mouse sperm (Fig.  S3B, C). These data are consistent with our sperm DNA bisulfite sequencing data, supporting the idea that the Kit promoter is hypomethylated in sperm (Fig. S2). Taken together, we failed to detect significant changes in DNA CpG methylation patterns in the Kit promoter region that correlate with the WTT phenotype in mice carrying the paramutant allele.
Both paternal and maternal RNAs can induce the paramutation phenotype although it is non-heritable. It was reported that injection of RNAs isolated from Kit 1/tm1Alf sperm into zygotes from mice unrelated to the Kit tm1Alf paramutant mice led to offspring with the WTT phenotype 12 . This finding suggests that RNA might be the mediator for paramutation establishment, and potentially its transmission through the germline. To test whether this finding applies to the Kit copGFP paramutation, we injected total RNA contents, including both small and large RNAs, isolated from not only sperm of HET (Kit 1/copGFP ) mice (as done in the previous study) 12 , but also oocytes, into the zygotes derived from WT BTT mating pairs that were totally unrelated to the Kit copGFP paramutant line. As a control, we injected paternal or maternal RNAs isolated from WT mice that were totally unrelated to the Kit copGFP paramutant  line. Injection of either sperm-borne (i.e., paternal) or oocyte (maternal) RNAs isolated from WT, unrelated mice into unrelated WT zygotes led to close-to-baseline levels of WTT phenotype in offspring (25-29% vs. 30%) (Fig. 5A, B). However, injection of either paternal or maternal RNAs derived from Kit 1/copGFP mice into unrelated WT zygotes resulted in offspring with an incidence of the WTT phenotype at between 52-62% (Fig. 5C, D), which is much higher than that in the control group and the baseline levels. Injection of sperm RNAs appeared to have a greater effect, as compared to injection of oocyte RNAs (62% vs. 52%), both of which were isolated from the HET (Kit 1/copGFP ) mice.
Both maternal and paternal RNAs appeared to be able to induce the WTT phenotype. To determine whether the induced paramutation phenotype could be transmitted to subsequent generations, we further outcrossed the WTT-positive, sperm RNA-induced mice (a total of 9 males and 15 females) with WT pure black C57BL/6J mice. To our surprise, offspring yielded by outcrossing of either male or female WTT-positive, sperm RNA-induced mice, displayed the baseline levels of the WTT phenotype (Fig. 5E). These data imply that although sperm-borne RNAs can induce the paramutation phenotype, the induced phenotype cannot be transmitted through the germline to subsequent generations.
Maternal miRNAs and piRNAs affect the inheritance of the Kit copGFP -induced paramutation. DROSHA is a nuclear RNase III that cleaves primary miRNA transcripts into precursor miRNAs in the nucleus and thus, is essential for miRNA biogenesis 37 . Previous studies have demonstrated that oocyte-specific Drosha conditional knockout (Zp3-Cre; Drosha lox/2 , hereafter called Drosha cKO) mice display normal oocyte development and fertility, although those Drosha-null oocytes contain neither precursor nor mature miRNAs 37 . Mov10l1 encodes a protein that is required for PIWI-interacting RNA (piRNA) biogenesis, and global Mov10l1 knockout mice do not produce piRNAs in any of their cells 38,39 . Therefore, these two KO lines provided an excellent opportunity to test whether maternal miRNAs or piRNAs could affect paramutation formation and transmission.  We crossed Kit 1/copGFP male mice with either Zp3-Cre; Drosha lox/D or Mov10l1 -/females, which were both on pure C57BL/6J background, and examined the WTT phenotype among all genotypically WT offspring (Fig. 6). When Kit 1/copGFP male mice were bred with WT BTT, unrelated female mice, ,55% of offspring displayed WTTs (Fig. 1F). Interestingly, when Kit 1/copGFP male mice were bred with Drosha cKO (pure black and miRNA-deficient in oocytes) and Mov10l1 KO (pure black and piRNA-deficient in all cell types including oocytes) females, ,90% of offspring showed the WTT phenotype, which is significantly higher than the incidence seen when crossed with pure black WT females (55% vs. 88-89%, x 2 test, p , 0.01). Because the major difference between the pure black (WT BTT) females and the Drosha cKO or Mov10l1 KO females lies in that the oocytes from the KO females contained no miRNAs or piRNAs, our data suggest that in normal oocytes, miRNAs and piRNAs can suppress the transmission of the Kit paramutation. Maternal miRNA or piRNA deficiency appears to enhance the transmission of the paramutation phenotype in this case.

Discussion
Mutations in the Kit allele are often associated with coat color changes, e.g., mice carrying Kit w/wv allele are entirely white, Kit 1/tm1Alf mice display white tail tips and white paws 12,32 . Kit 1/CreERT2 mice also display white bellies in addition to white paws and white tail tips 40 . We generated one null Kit allele, Kit copGFP , with the copGFP cassette inserted into exon 1 immediately after the start codon 31 . Heterozygotes all display white bellies, white paws and white tail tips 31 . The white spot phenotype results from the impaired Kit expression from the WT kit allele when the other is null 12 . Moreover, WT progeny of heterozygous parents tend to display WTTs, but without white bellies and white paws.
An earlier study has demonstrated that the WTT represents a phenotype associated with a paramutation induced by the Kit /tm1Alf mutation 26 . However, the validity of the study has been challenged because mice with WTTs are fairly common in most, if not all, of the lab mouse colonies of various strains 32 . In our own WT or transgenic mouse colonies, which are mostly on C57BL/6J or 129 Sv/Ev:C57BL/ 6J hybrid background, ,30% of the WT mice that are totally unrelated to the Kit mutant lines display WTTs, suggesting that the WTT phenotype is not unique to either the Kit genetic mutants or Kit paramutants. Despite the relatively common WTT phenotype in general lab mouse populations, it remains elusive how the WTT phenotype is formed. Nevertheless, the incidences of the WTT phenotype in general lab mouse populations represent baseline levels of this phenotype, and thus, should be defined so that the validity of using WTT as a phenotypic readout of the paramutation can be assessed more reliably. Our data suggest that the WTT can be used to evaluate the Kit paramutation effects because among WT progeny of Kit copGFP heterozygous parents, the WTT incidence is much higher than that in the general mouse population (30-40% in general population vs. 60-70% in WT offspring from Kit copGFP heterozygous parents). The elevated incidence in the WTT phenotype must be due to Kit copGFP -induced paramutation. It is of great importance to know the baseline incidence of the WTT phenotype because non-paramutant effects, otherwise, would have been taken into account, leading to potentially exaggerated conclusions. Equally important is that all analyses must be carried out using a large number of individual paramutant mice because some of them may be those with the baseline WTT phenotype, which is totally irrelevant to the Kit paramutation. However, when a large number of animals are analyzed, both Kit paramutation-specific and nonspecific WTT phenotypes will both be included, thus allowing for more reliable assessment of the paramutation effects.  Several lines of evidence suggest that the Kit copGFP paramutation is epigenetic by nature: First, the WTT phenotype exists in 60-70% of genetically WT mice derived from heterozygous parents, suggesting that the WT allele that they inherited from either or both of their parents has been modified, not in DNA sequence, but in functional status, and thus, represents the paramutant allele. Second, the WTT phenotype can be transmitted through the germline to subsequent generations; however, segregation of the phenotype does not follow the Mendelian Law, suggesting the modifications are not genetic, but epigenetic. Third, both maternal and paternal RNAs can induce the WTT phenotype in offspring of parents that are totally unrelated to the paramutation family, insinuating that the paramutation can be established by gamete RNAs during early embryonic development. Lastly, although the paramutation phenotype can be transmitted to subsequent generations, the penetrance decreases in later generations and eventually returns to the baseline levels after 3-4 generations when the paramutant mice are outcrossed. The transgenerational decrease in the penetrance of the paramutant phenotype most likely results from the global reprogramming events, which occur during preimplantation embryonic development and during PGC development. It has been shown that neither of the two global reprogramming events is complete because many imprinted loci and repetitive elements retain their epigenetic marks afterwards 29 . The persistence of the paramutation phenotype (i.e., WTT) across several generations, however, does suggest that the Kit copGFP paramutation is partially resistant to reprogramming.  In the outcrossing schemes, the male germline appears to be more capable of correcting the Kit copGFP paramutation because it takes less than three generations to bring the incidence of the paramutation phenotype down to the baseline levels when the Kit paramutation is transmitted through the male germline. In contrast, four generations are needed to do the same if the paramutation is transmitted through the female germline. This difference may result from the differential reprogramming mechanisms between the paternal and maternal genomes during the post-fertilization development, e.g., the former undergoes active demethylation, whereas passive demethylation occurs to the latter 29,41 . Therefore, a paternal paramutation may have a greater chance to be reprogramed than a maternal paramutation.
Compared to the outcrossing strategy, intercrossing appears to maintain the penetrance of the paramutation phenotype beyond four generations, if not indefinitely. The striking difference in penetrance of the paramutation phenotype between intercrossing and outcrossing schemes implies that a paramutation, or an epimutation in general, can persist in a population indefinitely in the case of intercrossing; it can be corrected and eventually ''fade away'' when outcrossed. The discovery that outcrossing leads to a dilution effect, while intercrossing causes persistence of the paramutation phenotype, is similar to phenotypes caused by genetic mutations. However, the distribution of the paramutation phenotype does not follow the Mendelian ratio, and can be explained by the developmental reprogramming. Nevertheless, this finding does imply that both outcrossing and intercrossing schemes should be evaluated in animal experiments on epigenetic transgenerational inheritance. Also, it suggests that in human populations, paramutations, or other types of epimutations, indeed can be corrected, or at least diluted during a period of several generations because human populations are largely outbred. However, paramutations, or epimutations in general, can persist for many more generations in a relatively inbred human population.
The previous study has demonstrated that injection of total RNAs isolated from Kit 1/tm1Alf sperm to WT unrelated zygotes can induce the WTT phenotype, and this phenotype can be transmitted to the next generation 12 . Although not tested in the previous study, injection of maternal RNAs should yield the same results given that both male and female paramutant mice can transmit the WTT phenotype. Indeed, we demonstrate that both paternal (i.e., sperm-borne) and maternal (oocyte) total RNAs can induce the WTT phenotype when injected into WT, unrelated zygotes. These results further validate the notion that gamete RNAs can induce the Kit copGFP paramutation phenotype. However, the WTT phenotype fails to be transmitted to the next generation, which is contradictory to the previous study 27 . One possible explanation would be that we took the baseline WTT levels into consideration when drawing our conclusions. The failure in transmitting the germline RNA-induced paramutation phenotype suggests that gamete RNAs can induce the paramutation, but its stable transmission requires other factors. In other words, gamete RNAs may be sufficient to induce a paramutation, but are insufficient for the inheritance of the induced paramutation.
It would be critical to identify the factors essential for successful transmission of the gamete RNA-induced paramutation. A recent report demonstrates that maternal and early embryonic expression of DNMT2, a methyltransferase that mainly methylates tRNAs, is required for Kit tm1Alf -induced paramutation and miR-124-induced Cdk2 paramutation 13 . However, one cannot determine, based on that study, whether the sperm-borne, methylated tRNA-derived small RNAs are required for the establishment or the transmission of paramutation. Moreover, it remains unknown whether the requirement for Dnmt2 is a direct effect on germline sncRNAs, or an indirect effect on the stability of other factors essential for paramutation transmission.
Given that paramutation represents a special type of epimutation, the underlying mechanism for paramutation transgenerational inheritance could well be applicable to general epigenetic transgenerational inheritance. The finding that RNA injection into zygotes can induce a phenotype also suggests that although RNA is not generally considered genetic material, it can indeed alter the phenotype if introduced into early embryos. One can then imagine that supplementation of RNAs during IVF or ICSI in human fertility clinics could potentially cause phenotypic changes, although no DNA is introduced to the test-tube babies. One assuring thing, however, is that our data suggest that the RNA-induced phenotypic changes may not be transmittable to subsequent generations. Overall, RNA-induced phenotypic alterations may represent an ethical issue in reproductive medicine, although RNA is generally not considered genetic materials in general.
sncRNAs have been implicated in the establishment and transmission of paramutations 2,4,6 . However, miRNAs and piRNAs are both essential for spermatogenesis and a lack of either causes disrupted spermatogenesis, leading to no sperm or the production of defective sperm that cannot fertilize eggs [42][43][44] , thus precluding studies of the effects of sperm-borne sncRNAs on paramutations. However, a lack of either miRNAs or piRNAs in oocytes appears to be compatible with normal folliculogenesis and female fertility 37,39 , which provides an excellent opportunity for us to test the effects of maternal/oocyte miRNAs and piRNAs on the paramutation transmission. The significant increase in the incidence of the WTT phenotype (from 55% to 88-89%) in progeny of females with Drosha-or Mov10l1-null oocytes suggests that the maternal miRNA and piRNA pathways have a suppressive role in paramutation transmission. This finding implies that the maternal miRNA and piRNA machineries normally inhibit transmission of the paramutation, which could either be directly involved in sncRNA biogenesis during post-fertilization development, or acting indirectly on other factors essential for paramutation transmission through post-transcriptional or epigenetic regulations.
In summary, we report another paramutation mouse model, and demonstrate that the paramutation can be transmitted across multiple generations and the breeding scheme can drastically affect the transmission efficiency. Both paternal and maternal RNAs from paramutant mice can induce the paramutation phenotype, but effective transmission requires yet-to-be-defined additional factors. Whole genome/transcriptome approaches are needed to identify the molecular changes responsible for the establishment, maintenance, memory and transmission of the paramutation in the near future.

Methods
Use of mouse lines. All mice were maintained in a temperature and humiditycontrolled, specific pathogen-free facility under a light-dark cycle (10 h-light/14 hdark) with food and water ad libitum. Breeding and all experimental procedures were performed according to the mouse use protocols approved by the Institutional Animal Use and Care Committee (IACUC) of the University of Nevada, Reno.
Kit 1/copGFP mice were generated as described in our previous report 31 . Zp3-Drosha cKO (Zp3-Cre; Drosha lox/2 ) and Mov10l1 KO female mice were generated as described 37,45 . All three lines were backcrossed for at least 10 generations to the C57BL/6J background before used for the experiments reported here.
Breeding scheme. Male and female Kit 1/copGFP mice were bred to get the 1 st generation (F1) paramutant WT mice, and the number of F1 WTT-positive mice and the number of all F1s were recorded. In the ''intercrossing'' scheme, F1 WT WTT siblings were bred to obtain F2s; breeding of F2 WT WTT siblings led to the production of F3s; breeding of F3 WT WTT siblings led to the production of F4s, and the number of F2, F3 and F4 mice with the WTT phenotype was counted against the total number of F2s, F3s and F4s, respectively.
In the outcrossing strategy, F1 WT WTT males and females were bred with WT BTT, totally unrelated WT females and males, respectively, to obtain F2s. The number of WTT-positive F2 among the total number of F2s was determined. Similarly, F2 WT WTT males and females were further bred with totally unrelated WT BTT females and males, respectively, to obtain F3s, and the number of WTTpositive F3 among the total number of F3s was determined. F4s were obtained using the same outcrossing scheme.
In the mixed breeding scheme, WTT-positive F1s were outcrossed with unrelated WT BTT mice to obtain F2s. Subsequently, F2 WT WTT siblings were bred to yield www.nature.com/scientificreports SCIENTIFIC REPORTS | 5 : 9266 | DOI: 10.1038/srep09266 F3s, and the number of WTT-positive F3 and the total number of F3s were then counted.
To test the effects of maternal miRNAs and piRNAs, Kit 1/copGFP male mice were bred with Zp3-Drosha cKO and Mov10l1 KO female mice, respectively. A trained observer determined the WTT phenotype based on the presence of the white spots on the tails, as illustrated in Fig. 1C.
PCR-based genotyping. Kit 1/copGFP mice can be genotyped using the primers as described 31 . Since these mice display white tail tips, white paws and white belly, Kit 1/copGFP mice could be easily recognized.
Preparation of sperm and oocytes. Caudal epididymal sperm were collected into HTF medium from Kit 1/copGFP males. Followed by washing several times using PBS, the sperm pellet was stored in 280uC for subsequent analyses. For oocyte collection, 4-6 week-old Kit 1/copGFP females were superovulated by intraperitoneal injection of 5IU of Pregnant Mare's Serum Gonadotropin (PMSG), followed by intraperitoneal injection of 5 IU of human Chorionic Gonadotropin (hCG) 48 h later. Mature oocytes were then collected from oviducts 14-16 h after hCG injection, and freed from cumulus cells by treatment in M2 medium containing 0.1% bovine testicular hyaluronidase. Followed by washing using the M2 medium, the oocytes were transferred into centrifuge tubes, and snap-frozen in liquid nitrogen for subsequent RNA extraction.
RNA isolation. Total RNA was isolated from sperm, oocytes and different organs using the mirVana TM miRNA isolation kit (Ambion, Grand Island, NY) according to the manufacturer's instructions.
RNA injection and embryo transfer. For collecting zygotes, WT BTT females (C57BL/6J) superovulated with 5 IU of PMSG (i.p.) followed by 5 IU of hCG (i.p.) 48 h later were mated with WT BTT males, and zygotes were collected by flushing the oviducts with M2 medium. For RNA injection, ,1-2 picolitre (pl) of total RNA (0.5 ng/ml) isolated from HET (Kit 1/copGFP ) or control (WT BTT) mouse sperm or oocytes were injected into the zygotes under inverted microscope using a microinjector (Cat# 930000043, Eppendorf). After injection, the survived embryos were transferred into oviducts of pseudopregnant CD1 (albino) females that had been mated with vasectomized males of the same strain during the night before. Cesarean section was performed on day 19 after embryo transfer to obtain live pups, which were then transferred to surrogate mothers.
Bisulfite sequencing. Genomic DNA (gDNA) was isolated from either sperm or tail samples, using the GenElute Mammalian Genomic DNA MiniPrep Kit (Sigma). For sperm samples, the mirVana RNA extraction lysis buffer treatment was performed prior to this extraction (Life Technologies). After gDNA was isolated, 1 mg was bisulfite treated, following the instructions with the NEB Epimark Bisulfite Conversion Kit (NEB). Following cleanup, bisulfite-specific PCR amplicons were amplified using GoTaq 2X, covering the promoter region (Promega). These PCR products of 170 bp. were run on 2% agarose gels, and the bands of interest were isolated and gel purified with the Qiaquick PCR purification kit (Qiagen). The resulting DNA extracts were cloned into the pGEM T-easy subcloning vector (Promega), and 2 ml of ligation reaction was then transformed into 50 ml of 5a competent cells (NEB), and grown overnight on ampicillin-LB-Agar plates at 37C (NEB). Individual colonies were grown in ampicillin liquid culture and their DNA was extracted using the Zyppy plasmid Miniprep Kit (Zymo Research). Plasmid DNA was then digested with EcoRI (to confirm the insert) and sequenced, using the SP6 sequencing primer. Sequencing was performed at the Nevada Genomics Center. Individual CpGs were then quantified for bisulfite conversion and the totals were calculated.
Kit isoform qPCR. RNA was isolated from the five tissues of interest, including brain, intestine, skin, testis and ovary, using the mirVana protocol (Life Technologies). RNA was reverse transcribed to obtain cDNAs, which were then diluted to 50 ng/ml and used for qPCR analyses, using FAST SYBR Green (Applied Biosystems). Each reaction contained 50 ng of cDNA and 1 ml of primer mix, in a total of 20 ml reactions. Gapdh and Hprt were used as endogenous controls. qPCR analyses were performed on a qPCR machine (7900HT Fast Real Time PCR System, Applied Biosystems). Relative quantification was performed with Gapdh serving as the endogenous control, and expression levels were further normalized against control wild type BTT samples. Primers used are listed in Table S1.
Kit promoter MeDIP qPCR. Genomic DNA (gDNA) samples were prepared using the Mammalian Genomic DNA Miniprep Kit (Sigma). Eluted gDNA samples were sonicated to achieve DNA fragmentation with sizes ranging between 200-500 bp using the Bioruptor (Diagenode), with the high setting, with 3X (5X 0530 On/0530 Off) repetitions. DNA fragments were confirmed for their integrity by running ,5 ml on 2% agarose gels. For 5 mC IPs, an MeDIP Kit was used following the manufacturer's instructions (Active Motif, Cat. #55009). MeDIP products were subject to qPCR on a qPCR machine (7900HT Fast Real Time PCR System, Applied Biosystems). Primers used are listed in Table S1. Sample enrichment was relative to equivalent dilutions of sample input DNA, measured as percent enrichment. DCt values were obtained relative to the No Template Control (NTC) samples. For 5 hmC MeDIP, the same protocol was followed, but the 5 hmC antibody (Zymo, Cat. #A4001-25), was used in place of the 5 mC antibody, for the pull-down.
Statistics. The x 2 test was used to evaluate differences in the frequency of the paramutation phenotype between the control and paramutant groups. The two-tailed student t-test was used to assess statistical significance for RT-qPCR analysis. Data were presented as mean 6 SEM. Significance was set at p , 0.05 for all tests.