The way that the embryo of a mammal, such as a mouse or a human, develops from a fertilized egg is a complicated process that relies on controlling: which genes are active; when these genes activate; and for how long they are active. In broad terms, there are four ways that this control can be achieved:

First, inside the sperm or egg, genes can be marked with small chemical tags that flag these genes to be activated (or remain inactive) after fertilization, depending on whether the modification was made by the father (in the sperm) or the mother (in the egg); this process is known as ‘imprinting’. Second, the mother can alter the gene activity in her offspring via the placenta; this process is known as ‘maternal effect’. Third, instructions encoded within the embryo's DNA can directly control if, and when, a nearby gene becomes activated; this is known as ‘cis-regulation’. Finally, similar instructions can also control genes that are situated elsewhere in the embryo's DNA through indirect mechanisms; this is known as ‘trans-regulation’.

Now, Spies, Smith et al. have investigated these four processes in the offspring of two different strains of mice, one originally from Europe and the other from Southeast Asia. The two strains were crossbred and the resulting embryos were analyzed to see which of the four processes affected gene activity. This analysis revealed 31 imprinted genes and 331 genes that exhibited a maternal effect—which sometimes changed gene activity by as much as 52%. Spies, Smith et al. also found over a thousand DNA instructions in the embryo's DNA that could directly influence the activity of nearby genes, but fewer instructions that could indirectly control genes that were further away.

These results suggest that direct control of genes, which affects only the genes closest to the DNA instruction, can vary a lot between individual embryos of the same species. However, indirect control of embryonically active genes, which affects many genes across the genome at the same time, appears much more tightly constrained by evolutionary forces. Which genes in the mother are responsible for the molecular signals that drive the maternal effect is an important question for future work, with implications for the genetic basis of embryonic development and disease.

Multicellular animals develop via complex pattern formation processes that unfold during embryogenesis, with gene regulation playing a central role. Surprisingly, the impact of genetic variation on gene regulation and the relative importance of the genetic mechanisms underlying regulatory variation in mammalian embryogenesis remain undefined. There are four broad categories of genetic mechanisms of gene expression regulation: imprinting, parental effects, cis-regulation, and trans-regulation. All of these are potentially subject to natural selection and provide genetic windows into the forces that shape gene regulatory evolution. Here, we quantify their impact and relevance in mammalian embryogenesis by asking which of these genetic mechanisms reveal evolutionary divergence or constraint since the last common ancestor of two distantly related mouse strains, C57BL/6 (B6) and Cast/Ei (Cast).

Genomic imprinting is a well-studied form of parent-of-origin-dependent gene expression that produces strongly biased allelic expression patterns in offspring according to the parental source of the inherited allele. The progeny's imprinted gene expression patterns result from epigenetic modifications put in place in the previous generation and reflect the particular genetic architecture at these specialized loci. A recent genome-wide survey estimated that fewer than 200 genes are imprinted in mice and humans (Babak et al., 2008). Imprinting is thought to play a critical role in early embryonic and placental development in mammals and has been shown to influence post-natal health and behavior [Reviewed in Peters (2014)].

Related to but distinct from imprinting is the concept of parental effects. Parental effects produce phenotypic variation that depends, directly or indirectly, on the genotype of either the mother or father, rather than that of the offspring. Therefore, the resulting phenotypic patterns lag a generation behind the genetic transmission of the causal variants. The most well-studied parental genetic effects are caused by deposition of maternal transcripts into the egg prior to fertilization, resulting in differences in early embryonic development depending on the genotype of the mother. Certain genes have also been shown to respond to maternal influence after birth through genetically defined maternal behaviors (Weaver et al., 2004). Because of the small contribution, through the sperm, of the paternal transcriptome to the fertilized zygote, and because of the stronger maternal contribution to child rearing in most model organisms, parental effects are typically thought of as synonymous with maternal effects, although true paternal effects are known to exist (Rando, 2012).

Maternal effects have been shown to be important during embryonic development, leading to differences in the birth weight of mice depending on the genotype of the mother (Cowley et al., 1989; Wolf et al., 2011). However, neither the causal molecular mechanism for these effects in the mother nor the responding genes in the embryo have been identified and few studies to date have discriminated between prenatal and postnatal effects. Maternal effects likely contribute to the heritability of complex traits but have been confounded with family structure in mouse as well as human studies (Mott et al., 2014).

Genetic variants affecting gene expression are classified into cis or trans, depending on whether they act on the gene copy on the same chromatid (cis) or on both chromatids equally (trans). Regulatory variants near their target genes usually act in cis, by modifying the activity of a promoter or enhancer, but very occasionally act in trans if they affect a gene product that regulates a nearby locus. One approach to quantifying cis and trans regulation in model organisms involves assaying gene expression in two parental lines compared to allele-specific expression (ASE) in the hybrids (Wittkopp et al., 2004) and asking whether alleles are regulated differently in the parentals compared to the hybrids. In adult flies and mice, these analyses have demonstrated that cis and trans regulatory variants frequently target the same genes, often with opposite, compensatory effects on the target's expression (McManus et al., 2010; Goncalves et al., 2012). Recent genome-wide studies of ASE patterns across many individuals have demonstrated the ability of high-throughput sequencing methods to directly identify genes with cis-regulatory variants (Montgomery et al., 2010; Pickrell et al., 2010).

Another method to quantify cis- and trans-regulatory variation is analysis of expression quantitative trait loci (eQTL) (Schadt et al., 2003). eQTL are identified as genetic loci whose genotypes correlate with gene expression changes across a number of genetically heterogeneous individuals. Genome-wide eQTL studies have demonstrated that the strongest genetic variation in expression regulation generally occurs in variants located proximally to their target genes, presumably acting in cis. While typically fewer in number, trans-regulatory variants affect both alleles of each target gene and can affect expression of many genes simultaneously, resulting in a large aggregate effect.

ASE and eQTL approaches have been applied widely in adult tissues in humans and model organisms. However, these approaches have not yet been applied to understand the global relationship between genetic variation and gene expression during mammalian embryogenesis, and while imprinting has been studied extensively, the contribution of maternal effects on mammalian embryonic gene expression has never been addressed. For several reasons, it is likely that patterns of regulatory divergence differ between embryos and adults. First, eQTL studies have demonstrated dynamic trans-regulatory variation during Caenorhabditis elegans larval development (Francesconi and Lehner, 2014). Second, recent studies in flies and fish have demonstrated higher expression of more ancient conserved genes as well as more conserved gene expression patterns during embryogenesis compared to the adult (Domazet-Lošo and Tautz, 2010; Kalinka et al., 2010). This suggests that the number of cis- as well as trans-eQTL should be lower in embryos compared to adult.

Here, we quantify evolutionary gene regulatory divergence between B6 and Cast during embryogenesis, addressing each of the four regulatory architectures. We focus on embryonic day 11.5 (E11.5), which is equivalent to roughly 40 days of gestation in human. It is a time characterized by organogenesis, rapid growth of the brain and limbs, increasingly complex and localized pattern formation, and consequently substantial constraints on allowable phenotypic variation. An ASE/eQTL hybrid approach facilitates quantifying the different types of regulation and their effects on expression variation, revealing unique and important features of gene regulatory architecture during mammalian embryonic development.

For this study, we chose two inbred strains derived from geographically separate subspecies of Mus musculus: C57BL/6J (B6), the classic inbred mouse reference strain, which was derived from European mice, and CAST/EiJ (Cast), which was derived from the Southeast Asian Mus musculus castaneus. Inbreeding caused random subsets of alleles segregating in their ancestral populations to become fixed, and as a consequence the two strains exhibit one single nucleotide polymorphism (SNP) difference every 120 bp on average (Keane et al., 2011).

We collected 16 E11.5 F1 embryos obtained from reciprocal crosses between B6 and Cast (Figure 1). We profiled genome-wide gene expression using a 3′-biased RNA sequencing method, 3SEQ (Beck et al., 2010), in order to maximize SNP coverage in the 3′ UTRs, where the average SNP density is ∼1 SNP per 170 bp (compared to 1 SNP per 265 bp in the coding sequence). Mapping of reads separately against the Cast (Keane et al., 2011) and B6 genomes allowed us to eliminate reference mapping bias (Figure 2—figure supplement 1) as the average fraction of reads supporting the B6 allele (B6 allele fraction, ‘BAF’) was 0.502. Overall, out of a possible 6917 genes with sufficient reads covering SNPs, 1594 (23%) showed significant ASE (p < 0.01, paired t-test, Bonferroni corrected; Figure 2A,B). We next backcrossed F1 mice (derived from female B6 by male Cast crosses) with B6 parentals to obtain N2 embryos. We harvested 154 N2s at E11.5, performed 3SEQ, and quantified ASE in the approximately 50% of embryos that were heterozygous at each gene. There was excellent agreement between average BAF estimates from F1s and N2s (Figure 2C; rho = 0.77).

Figure 1

Download asset Open asset

Figure 2 with 3 supplements see all

Download asset Open asset

To quantify developmental progression for each embryo at the time of harvest, we counted the number of somites, which form every 90–120 min (Saga and Takeda, 2001) and ranged from 42 to 57 in our embryos. Of the 7712 genes with moderate expression of at least 5 reads per million (RPM), 1796 genes showed significant correlation between total expression and somite number (q < 0.05, Spearman correlation). Genes changing over this developmental time course were enriched for a number of gene ontology (GO) categories indicative of important developmental processes occurring at E11.5, including synaptogenesis (q = 2.2 × 10⁻⁸, GOrilla) and eye lens formation (q = 5.8 × 10⁻⁷). A third significant GO category, blood microparticles (q = 9.6 × 10⁻⁷), includes the globins, which are beginning to undergo the transition from fetal to adult form over the developmental time period we assayed (Noordermeer and de Laat, 2008) (Figure 2D).

By correlating ASE with somite number, we identified twelve genes that show significant differential regulation between B6 and Cast alleles as a function of developmental timing (q < 0.10, Spearman rho). Globins and functionally related genes were well represented in this set, which included both major (Hbb-b1) and minor (Hbb-b2) adult beta globins. Both genes increase in absolute level and relative expression compared to fetal beta globins (Figure 2D; Figure 2—figure supplement 2), but the B6 allele of the major form is delayed compared to the Cast allele (Figure 2E; rho = 0.86, q < 10⁻¹⁴), whereas the opposite occurs for the minor form (Figure 2F; rho = −0.51, q = 0.02). By the end of our time course, the alleles are equally expressed (Figure 2E,F). Transferrin also increases in absolute expression but the Cast allele is delayed compared to the B6 allele (rho = −0.46, q = 0.02; Figure 2—figure supplement 2). These results suggest that compensatory mechanisms have evolved in one or both strains to keep the overall regulatory dynamic of oxygen physiology stable during this important switching period.

Comparison of embryos from either side of the reciprocal cross allowed us to identify parent-of-origin dependent ASE patterns indicative of imprinting. Consistent with previous studies in mouse embryos (Babak et al., 2008), we found a total of 31 genes with significant imprinting (Figure 2G–I; p < 0.01, binomial test, Bonferroni corrected), all of which had been previously identified. We note that the clear separation of most of the imprinted loci from the noise of the bulk of all expressed genes, with few exceptions (Figure 2I; Figure 2—figure supplement 3), suggests that most imprinting generally acts uniformly across the embryo, as expected from the methylation-based mechanism that sets the expression state of the regulated allele pre-zygotically (Reik et al., 1987).

Imprinting is not the only molecular mechanism that results from the mother–offspring resource allocation conflict in mammals (Moore and Haig, 1991). To explore other genetically-encoded parental effects, we took advantage of the backcross design, in which N2 offspring have genotypically distinct parents. We identified a large number (n = 331) of embryonically expressed genes whose total expression levels differed significantly depending on the genetic background of the mother (Mann Whitney U test, p < 0.01, Bonferroni corrected), irrespective of the genotype of the embryo (Figure 3A,B; Supplementary file 1). For example, Angptl4 expression increases 52% from an average of 107 RPM in embryos with an F1 mother to an average of 163 RPM in embryos with a B6 mother (Figure 3C). As for the magnitude of the effect, these maternal effect target genes exhibited up to a 3.5-fold (e.g., Polm: 2.5–11-fold at a 95% confidence interval) difference in total expression level depending on genotype of the mother.

Figure 3

Download asset Open asset

We were able to exclude a number of alternative mechanisms to explain these maternal effects. All F1s used for the backcross were derived from B6 female by Cast male matings, and therefore all N2 embryos carry the B6 mitochondrial genome, ruling out a confounding mitochondrial effect. The maternal effects were unchanged when considering sex, X chromosome genotype, litter, and developmental stage. Furthermore, for the vast majority of genes with SNP coverage, we were able to verify that maternal effect target genes were not imprinted or the result of contamination by maternally expressed transcripts. We found that maternal genotype by itself was a better predictor of the maternal effects than maternal genotype in conjunction with embryonic genotype at any of the imprinted loci, indicating that the maternal effects were not the result of trans effects downstream of imprinted genes. To exclude the possibility of a batch effect in our 3SEQ data, we used quantitative RT-PCR (qPCR) to confirm 4 genes as maternal effect target genes.

The most significant of the maternal effect target genes, Angptl4 (Figure 3C; p = 8 × 10⁻¹³ by 3SEQ, p = 1 × 10⁻¹¹ by qPCR), functions in lipid metabolism and energy homeostasis, suggestive of a role in responding to the placental nutrient supply (Yoshida et al., 2002). Similarly, Gpx3 (Figure 3B; p = 5 × 10⁻¹² by 3SEQ, p = 6.6 × 10⁻⁴ by qPCR), a glutathione peroxidase, is an antioxidant enzyme that acts in fat and glucose metabolism and has been associated with obesity and type 2 diabetes in mouse and human (Lee et al., 2008; Chung et al., 2009). Both of these genes are transcriptionally regulated by the peroxisome proliferator-activated receptor PPARγ.

To examine how genetic variation carried by the embryo itself affects gene expression phenotype, we performed eQTL analyses by correlating expression levels of each measurably expressed gene (‘target’) with N2 embryo genotypes (either B6/Cast or B6/B6) at each recombination-defined locus (‘controlling locus’). Genotyping and recombination analysis based on the expressed SNPs (Figure 4A) divided the autosomes into 986 marker region blocks, each separated by one or more recombination events that define candidate controlling loci. We tested every controlling locus for correlation with gene expression of every gene. At a genome-wide significance level of q < 0.01, we identified a total of 1034 eQTL.

Figure 4 with 1 supplement see all

Download asset Open asset

1017 of these eQTL (98%) mapped close to the target gene (Figure 4B), suggesting cis-regulation as the underlying mechanism. A gene's difference in expression between heterozygotes and homozygotes could accurately predict its allelic ratio as assayed by ASE (Figure 4C). This high correlation (rho = 0.58, n = 4821; rho = 0.70 when considering only the n = 1000 genes with the highest allele-specific read coverage) provides a systematic validation of the large effect of embryonic cis-regulatory polymorphisms on expression phenotype. Note that our study has sufficient power to identify substantial numbers of eQTL when performing this all–loci–by–all–genes analysis. If, instead, we were focused specifically on only studying cis-regulation, we could have increased our power to identify cis-eQTL by only comparing genes to their nearby loci, thereby reducing the number of statistical tests being performed.

In contrast to the large number of cis-eQTL, we found few trans-eQTL regardless of FDR cutoff (Figure 5A). After the top 3 (each comprising a controlling locus–target gene pair, e.g., Figure 5B), significance drops off rapidly compared to background, unlike for the cis-eQTL (Figure 5C). To explore this lack of trans effects further, we reasoned that a set of genes with a high prior probability of having downstream effects might be enriched for trans-controlling loci. Our cis-eQTL/ASE loci comprise a large set of such candidates because differences in expression levels of a gene product may propagate to downstream targets (Yvert et al., 2003). However, we found that neither strong cis-eQTL nor ASE loci are enriched for trans-eQTL. Finally, application of a random forest model (Michaelson et al., 2010; Francesconi and Lehner, 2014) that simultaneously considers developmental progression (number of somites) and genotype also did not enrich for trans-eQTL (Figure 4—figure supplement 1). We estimate that the number of comparable trans-eQTL is about 1% that of cis-eQTL.

Figure 5

Download asset Open asset

Because a trans-regulatory effect is potentially distributed over several target loci, we asked whether there was evidence for ‘hotspots’, or controlling loci whose genotypes correlate with small expression differences at numerous target genes (Brem et al., 2002; Schadt et al., 2003). Permutation tests revealed a significant signal (p < 10⁻²⁴ KS test, Figure 5D) in support of hotspots, but no specific candidates. Reinforcing the above result that ASE genes do not act as trans-controlling loci, strong ASE genes were not substantially more likely to act as hotspots than weaker ASE genes (Figure 5D; p = 0.20, KS test). Therefore, at this stage of embryogenesis, we found no evidence for downstream propagation of genetically controlled expression level differences.

We hypothesized that the paucity of trans-eQTL was due to increased evolutionary constraint of regulatory networks during embryonic development. Since trans-eQTL have previously been found to have weaker effects than cis-eQTL (Montgomery and Dermitzakis, 2011), an alternative hypothesis is that we were simply underpowered to find trans-eQTL. To distinguish these possibilities, we obtained genotype and expression data from previously published eQTL studies performed in adult mouse tissues (Wang et al., 2006; Yang et al., 2006; Chen et al., 2008). After sampling individuals to match the same number of total eQTL found at an FDR of 0.05, we found that all previously published studies conducted in adult tissues showed a substantially higher proportion of trans-eQTL than did our embryos (Figure 5E,F). This result was robust to varying the sample size from the previously published data, to whether p-values or FDR was used to assess significance, and to varying the thresholds for comparing proportions of trans-eQTL across the ranked eQTL. Based on this result and our high power to detect cis-eQTL even at low FDR, we conclude that embryonic trans-eQTL are subject to elevated selective constraint compared to adult tissues.

If cis-regulatory divergence in embryos is strongly buffered and does not lead to downstream trans-regulatory effects, why does embryogenesis tolerate such a large number of cis-effects in the first place? We hypothesized that divergent cis-regulation between strains occurs preferentially in genes with lower evolutionary constraint. In support of this hypothesis, ASE genes were significantly depleted of GO categories related to organ development (p = 6 × 10⁻⁵, q = 0.05), as well as gene regulation (p = 1 × 10⁻⁴, q = 0.07), and transcription factors (p = 3 × 10⁻⁵, q = 0.03), demonstrating constraints on variation in core regulatory genes. Furthermore, we found that evolutionary constraint at the DNA sequence level (Cooper et al., 2005; Davydov et al., 2010) was lowest in the most significant cis-eQTL (Figure 5G; rho = −0.14, p = 6 × 10⁻¹³; n = 2788). Similarly, ASE magnitude of effect correlated negatively with evolutionary constraint (rho = −0.10, p = 9 × 10⁻⁸, n = 2612), whereas cis-eQTL in adult mouse tissues showed no correlation (|rho| < 0.05). Thus, the strong enrichment for cis-eQTL/ASE in less conserved genes is unique to embryos.

We have performed the first, to our knowledge, comprehensive global analysis of the effects of genetic variation on mammalian gene expression during embryogenesis. Our experimental design allowed us to simultaneously assay the effects of genetic variation on imprinting, maternal effects, cis-regulation, and trans-regulation. Supporting the high quality of our data and analytical methods, we found good agreement between related measurements of cis-regulatory variation when comparing ASE in F1 embryos (sequenced at high coverage) with ASE in N2 embryos (sequencing many individuals), and when comparing ASE with cis-eQTL.

Our two mouse strains exhibit SNPs at 17.7 million sites, about sixfold more than the heterozygosity in African humans (1000 Genomes Project Consortium, 2012). The vast majority of these polymorphisms have accumulated as a result of genetic drift (Kimura, 1983) over the time since the present-day European and Southeast Asian M. musculus populations were separated from their common ancestral population. Of the small subset of polymorphisms that affect molecular function, some may produce an advantageous phenotype subject to positive selection, some are deleterious to organismal fitness and are therefore subject to purifying selection, and yet others are selectively neutral (subject to drift). Missense polymorphisms are severely underrepresented genome-wide and exhibit lower derived allele frequencies relative to other genomic sites (1000 Genomes Project Consortium, 2012), reflecting the global predominance of purifying selection. This signal of purifying selection is weaker but still substantial for noncoding gene regulatory sites (ENCODE Project Consortium, 2007).

Given these known dynamics, our study sheds light on the microevolution of gene regulation. Gene expression divergence of the type we quantified in this work occurs as a result of polymorphisms in regulatory regions or in expressed regulatory gene products. The former manifests as cis-effects, the latter as trans-effects. Because we focused on embryonic gene expression, two competing hypotheses as to what patterns we might find were plausible. Under the first hypothesis, the substantial DNA sequence divergence between mouse strains would be reflected in many expression differences. The alternative hypothesis is that few expression changes would be detected between strains. This hypothesis is suggested by the developmental hourglass model (Domazet-Lošo and Tautz, 2010; Kalinka et al., 2010), which posits that gene expression patterns are most highly conserved in the middle of embryonic development (the so-called ‘phylotypic stage’).

In support of the first hypothesis, we found substantial numbers of significant eQTL, demonstrating that a sizeable subset of all segregating polymorphisms causes expression changes. However, we found that nearly all acted in cis, rather than trans. This depletion of trans-regulatory variation is substantially stronger than that observed across several adult mouse tissues. In contrast to adult tissues, eQTL acting in our embryos were correlated with lower DNA sequence conservation, supporting the conclusion that evolutionary constraint on gene expression regulation is markedly stronger during mammalian embryogenesis than in the adult. In addition, the identified cis-expression changes do not themselves cause detectable trans-effects. These findings support the idea of the developmental hourglass model (Domazet-Lošo and Tautz, 2010; Kalinka et al., 2010) while suggesting that substantial neutral divergence of molecular phenotype (gene expression) can accumulate in the form of cis-regulatory variation even under strong evolutionary constraint at the organismal (embryonic) level.

Previous studies have shown enrichment for trans-regulation when using an eQTL model that considers both genotype and stage during C. elegans larval development. Perhaps because of the nearness of our embryos to the mouse phylotypic stage, or because of fundamental differences between mice and worms, we found that incorporating developmental timing (i.e., somite number) into our eQTL model did not alter the balance between cis- and trans-regulation. Somite number was, however, correlated with significant changes in ASE, indicating differential developmental timing of gene expression between the Cast and B6 strains. In particular, we found that alleles of several hemoglobin-related genes, including the adult beta globins and transferrin, were expressed at differing levels between the strains at early stages of the fetal-to-adult hemoglobin transition, but then equalize by the end of our developmental time course. These results demonstrate our ability to identify subtle tissue-specific expression differences between strains, despite sequencing from whole embryos containing a mix of tissues.

It is formally possible that some of the significant eQTL identified here act in trans, even though their target genes are located in genomic proximity. However, the good agreement between eQTL and ASE suggests that the majority of our local eQTL are indeed cis-eQTL. Furthermore, because there is no reason to believe that the proportion of locally-situated but trans-acting eQTL is higher in embryos than in adults, it does not materially alter our conclusions if a small number of our local eQTL do act in trans.

Because of the large number of embryos collected and individually sequenced for our study, we were able to accurately quantify ASE mean and standard deviation for thousands of genes. When stratifying embryos by side of cross, we found that nearly all imprinted genes showed very strong allelic bias, meaning that the epigenetically silenced allele is expressed at only low levels across the embryo. Consistent with previous studies that found few genes exhibiting tissue-specific imprinting [apart from those found in the extra-embryonic tissues; reviewed in Prickett and Oakey (2012)], we observed no weak allelic bias that might result from averaging effects across tissues.

In addition to the clear signature of parent-of-origin-dependent allelic expression, we also found strong maternal effects on total gene expression levels for hundreds of genes. Whereas most previous studies have employed F1 intercrosses in which parental genotypes are indistinguishable, we used a backcross design, enabling us to track the potentially important sources of genetic variance contributed by the parents. The observed maternal effects are restricted to pre-natal influences and separate from imprinting and mitochondrial effects.

Our results demonstrate that the genotype of the mother plays an important role in regulating gene expression, not just after birth, as has been previously suggested, but also during embryogenesis, affecting the expression of hundreds of genes (Bult and Lynch, 1997; Weaver et al., 2004; Wolf et al., 2011). The implications of this finding are significant for human genetic twin, familial, and genome-wide association studies. What is usually thought of as familial environment [e.g., Grundberg et al. (2012)] may have a substantial heritable genetic component exerting an effect not on the generation carrying it but on the offspring.

Parental genetic influences on offspring phenotype have been implicated in growth regulation and adult-onset obesity- and diabetes-related diseases but the gene targets of such effects have not been identified (Jarvis et al., 2005). While we cannot rule out an epigenetic or RNA-mediated mechanism underlying this regulation, it is more likely acting through the placenta's role of transmitting nutrients, hormones or other factors to the embryo. This hypothesis is supported by the known functions of the most significant target gene, Angptl4, which responds to lipid metabolism and energy homeostasis, and participates in angiogenesis (Koster et al., 2005; Scott et al., 2012). We hypothesize that these maternal effect target genes continue to be expressed differently depending on maternal genotype, and may be at least partly responsible for the previously observed effect of maternal genotype on pre-natal growth (Cowley et al., 1989; Wolf et al., 2011).

In summary, using our hybrid ASE/eQTL approach we were able to assay all four types of genetic variation in gene expression regulation. We found that cis-regulatory variation caused the largest divergence in overall gene expression in mouse embryos, affecting over 1000 genes with up to sevenfold differences in expression levels between the strains. These cis-regulatory variants did not translate into trans-regulatory divergence, and we identified only ∼10 trans-eQTL genomewide. Imprinting affected the allelic balance of dozens of genes, generally resulting in complete or nearly complete silencing of the inactive allele. Finally, we identified over 300 maternal effect target genes. The number of maternal effect targets, and their overall significance, are only slightly more modest than cis-regulatory effects, but nonetheless substantial, with approximately 1/3 as many genes involved, and maximal effect size of up to 3.5-fold difference or approximately 1/2 as large. Given this context, it is clear that maternal effects can play a substantial genetic role in determining embryonic gene expression patterns. We speculate that maternal effects contribute to evolutionary divergence, allowing expression heterogeneity in the embryo to build as a result of differences in the adult mother, where gene regulatory networks show less evolutionary constraint than in the developing embryo.

1. 1. Smith CL
   2. Spies N
   3. Rodriguez JM
   4. Baker JC
   5. Batzoglou S
   6. Sidow A

   (2015) Raw sequencing data

   Publicly available at NCBI Gene Expression Omnibus (GSE62967).

   http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE62967

1. 1. 1000 Genomes Project Consortium

   (2012) An integrated map of genetic variation from 1,092 human genomes

   Nature 491:56–65.

   https://doi.org/10.1038/nature11632

   • Google Scholar
2. 1. Babak T
   2. Deveale B
   3. Armour C
   4. Raymond C
   5. Cleary MA
   6. van der Kooy D
   7. Johnson JM
   8. Lim LP

   (2008) Global survey of genomic imprinting by transcriptome sequencing

   Current Biology 18:1735–1741.

   https://doi.org/10.1016/j.cub.2008.09.044

   • Google Scholar
3. 1. Beck AH
   2. Weng Z
   3. Witten DM
   4. Zhu S
   5. Foley JW
   6. Lacroute P
   7. Smith CL
   8. Tibshirani R
   9. van de Rijn M
   10. Sidow A
   11. West RB

   (2010) 3'-end sequencing for expression quantification (3SEQ) from archival tumor samples

   PLOS ONE 5:e8768.

   https://doi.org/10.1371/journal.pone.0008768

   • Google Scholar
4. 1. Brem RB
   2. Yvert G
   3. Clinton R
   4. Kruglyak L

   (2002) Genetic dissection of transcriptional regulation in budding yeast

   Science 296:752–755.

   https://doi.org/10.1126/science.1069516

   • Google Scholar
5. 1. Bult A
   2. Lynch CB

   (1997) Nesting and fitness: lifetime reproductive success in house mice bidirectionally selected for thermoregulatory nest-building behavior

   Behavior Genetics 27:231–240.

   https://doi.org/10.1023/A:1025610130282

   • Google Scholar
6. 1. Chen Y
   2. Zhu J
   3. Lum PY
   4. Yang X
   5. Pinto S
   6. MacNeil DJ
   7. Zhang C
   8. Lamb J
   9. Edwards S
   10. Sieberts SK
   11. Leonardson A
   12. Castellini LW
   13. Wang S
   14. Champy MF
   15. Zhang B
   16. Emilsson V
   17. Doss S
   18. Ghazalpour A
   19. Horvath S
   20. Drake TA
   21. Lusis AJ
   22. Schadt EE

   (2008) Variations in DNA elucidate molecular networks that cause disease

   Nature 452:429–435.

   https://doi.org/10.1038/nature06757

   • Google Scholar
7. 1. Chung SS
   2. Kim M
   3. Youn BS
   4. Lee NS
   5. Park JW
   6. Lee IK
   7. Lee YS
   8. Kim JB
   9. Cho YM
   10. Lee HK
   11. Park KS

   (2009) Glutathione peroxidase 3 mediates the antioxidant effect of peroxisome proliferator-activated receptor gamma in human skeletal muscle cells

   Molecular and Cellular Biology 29:20–30.

   https://doi.org/10.1128/MCB.00544-08

   • Google Scholar
8. 1. Cooper GM
   2. Stone EA
   3. Asimenos G
   4. Program NC
   5. Green ED
   6. Batzoglou S
   7. Sidow A

   (2005) Distribution and intensity of constraint in mammalian genomic sequence

   Genome Research 15:901–913.

   https://doi.org/10.1101/gr.3577405

   • Google Scholar
9. 1. Cowley DE
   2. Pomp D
   3. Atchley WR
   4. Eisen EJ
   5. Hawkins-Brown D

   (1989)

   The impact of maternal uterine genotype on postnatal growth and adult body size in mice

   Genetics 122:193–203.

   • Google Scholar
10. 1. Davydov EV
    2. Goode DL
    3. Sirota M
    4. Cooper GM
    5. Sidow A
    6. Batzoglou S

    (2010) Identifying a high fraction of the human genome to be under selective constraint using GERP++

    PLOS Computational Biology 6:e1001025.

    https://doi.org/10.1371/journal.pcbi.1001025

    • Google Scholar
11. 1. Domazet-Lošo T
    2. Tautz D

    (2010) A phylogenetically based transcriptome age index mirrors ontogenetic divergence patterns

    Nature 468:815–818.

    https://doi.org/10.1038/nature09632

    • Google Scholar
12. 1. Eden E
    2. Navon R
    3. Steinfeld I
    4. Lipson D
    5. Yakhini Z

    (2009) GOrilla: a tool for discovery and visualization of enriched GO terms in ranked gene lists

    BMC Bioinformatics 10:48.

    https://doi.org/10.1186/1471-2105-10-48

    • Google Scholar
13. 1. ENCODE Project Consortium

    (2007) Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project

    Nature 447:799–816.

    https://doi.org/10.1038/nature05874

    • Google Scholar
14. 1. Francesconi M
    2. Lehner B

    (2014) The effects of genetic variation on gene expression dynamics during development

    Nature 505:208–211.

    https://doi.org/10.1038/nature12772

    • Google Scholar
15. 1. Goncalves A
    2. Leigh-Brown S
    3. Thybert D
    4. Stefflova K
    5. Turro E
    6. Flicek P
    7. Brazma A
    8. Odom DT
    9. Marioni JC

    (2012) Extensive compensatory cis-trans regulation in the evolution of mouse gene expression

    Genome Research 22:2376–2384.

    https://doi.org/10.1101/gr.142281.112

    • Google Scholar
16. 1. Grundberg E
    2. Small KS
    3. Hedman ÅK
    4. Nica AC
    5. Buil A
    6. Keildson S
    7. Bell JT
    8. Yang TP
    9. Meduri E
    10. Barrett A
    11. Nisbett J
    12. Sekowska M
    13. Wilk A
    14. Shin SY
    15. Glass D
    16. Travers M
    17. Min JL
    18. Ring S
    19. Ho K
    20. Thorleifsson G
    21. Kong A
    22. Thorsteindottir U
    23. Ainali C
    24. Dimas AS
    25. Hassanali N
    26. Ingle C
    27. Knowles D
    28. Krestyaninova M
    29. Lowe CE
    30. Di Meglio P
    31. Montgomery SB
    32. Parts L
    33. Potter S
    34. Surdulescu G
    35. Tsaprouni L
    36. Tsoka S
    37. Bataille V
    38. Durbin R
    39. Nestle FO
    40. O'Rahilly S
    41. Soranzo N
    42. Lindgren CM
    43. Zondervan KT
    44. Ahmadi KR
    45. Schadt EE
    46. Stefansson K
    47. Smith GD
    48. McCarthy MI
    49. Deloukas P
    50. Dermitzakis ET
    51. Spector TD
    52. Multiple Tissue Human Expression Resource (MuTHER) Consortium

    (2012) Mapping cis- and trans-regulatory effects across multiple tissues in twins

    Nature Genetics 44:1084–1089.

    https://doi.org/10.1038/ng.2394

    • Google Scholar
17. 1. Jarvis JP
    2. Kenney-Hunt J
    3. Ehrich TH
    4. Pletscher LS
    5. Semenkovich CF
    6. Cheverud JM

    (2005) Maternal genotype affects adult offspring lipid, obesity, and diabetes phenotypes in LGXSM recombinant inbred strains

    Journal of Lipid Research 46:1692–1702.

    https://doi.org/10.1194/jlr.M500073-JLR200

    • Google Scholar
18. 1. Kalinka AT
    2. Varga KM
    3. Gerrard DT
    4. Preibisch S
    5. Corcoran DL
    6. Jarrells J
    7. Ohler U
    8. Bergman CM
    9. Tomancak P

    (2010) Gene expression divergence recapitulates the developmental hourglass model

    Nature 468:811–814.

    https://doi.org/10.1038/nature09634

    • Google Scholar
19. 1. Keane TM
    2. Goodstadt L
    3. Danecek P
    4. White MA
    5. Wong K
    6. Yalcin B
    7. Heger A
    8. Agam A
    9. Slater G
    10. Goodson M
    11. Furlotte NA
    12. Eskin E
    13. Nellåker C
    14. Whitley H
    15. Cleak J
    16. Janowitz D
    17. Hernandez-Pliego P
    18. Edwards A
    19. Belgard TG
    20. Oliver PL
    21. McIntyre RE
    22. Bhomra A
    23. Nicod J
    24. Gan X
    25. Yuan W
    26. van der Weyden L
    27. Steward CA
    28. Bala S
    29. Stalker J
    30. Mott R
    31. Durbin R
    32. Jackson IJ
    33. Czechanski A
    34. Guerra-Assunção JA
    35. Donahue LR
    36. Reinholdt LG
    37. Payseur BA
    38. Ponting CP
    39. Birney E
    40. Flint J
    41. Adams DJ

    (2011) Mouse genomic variation and its effect on phenotypes and gene regulation

    Nature 477:289–294.

    https://doi.org/10.1038/nature10413

    • Google Scholar
20. 1. Kent WJ
    2. Sugnet CW
    3. Furey TS
    4. Roskin KM
    5. Pringle TH
    6. Zahler AM
    7. Haussler D

    (2002) The human genome browser at UCSC

    Genome Research 12:996–1006.

    https://doi.org/10.1101/gr.229102

    • Google Scholar
21. Book

    1. Kimura M

    (1983)

    The neutral theory of molecular evolution

    Cambridge University Press.

    • Google Scholar
22. 1. Koster A
    2. Chao YB
    3. Mosior M
    4. Ford A
    5. Gonzalez-DeWhitt PA
    6. Hale JE
    7. Li D
    8. Qiu Y
    9. Fraser CC
    10. Yang DD
    11. Heuer JG
    12. Jaskunas SR
    13. Eacho P

    (2005) Transgenic angiopoietin-like (angptl)4 overexpression and targeted disruption of angptl4 and angptl3: regulation of triglyceride metabolism

    Endocrinology 146:4943–4950.

    https://doi.org/10.1210/en.2005-0476

    • Google Scholar
23. 1. Langmead B
    2. Salzberg SL

    (2012) Fast gapped-read alignment with Bowtie 2

    Nature Methods 9:357–359.

    https://doi.org/10.1038/nmeth.1923

    • Google Scholar
24. 1. Lee YS
    2. Kim AY
    3. Choi JW
    4. Kim M
    5. Yasue S
    6. Son HJ
    7. Masuzaki H
    8. Park KS
    9. Kim JB

    (2008) Dysregulation of adipose glutathione peroxidase 3 in obesity contributes to local and systemic oxidative stress

    Molecular Endocrinology 22:2176–2189.

    https://doi.org/10.1210/me.2008-0023

    • Google Scholar
25. 1. McManus CJ
    2. Coolon JD
    3. Duff MO
    4. Eipper-Mains J
    5. Graveley BR
    6. Wittkopp PJ

    (2010) Regulatory divergence in Drosophila revealed by mRNA-seq

    Genome Research 20:816–825.

    https://doi.org/10.1101/gr.102491.109

    • Google Scholar
26. 1. Merkin J
    2. Russell C
    3. Chen P
    4. Burge CB

    (2012) Evolutionary dynamics of gene and isoform regulation in Mammalian tissues

    Science 338:1593–1599.

    https://doi.org/10.1126/science.1228186

    • Google Scholar
27. 1. Michaelson JJ
    2. Alberts R
    3. Schughart K
    4. Beyer A

    (2010) Data-driven assessment of eQTL mapping methods

    BMC Genomics 11:502.

    https://doi.org/10.1186/1471-2164-11-502

    • Google Scholar
28. 1. Montgomery SB
    2. Dermitzakis ET

    (2011) From expression QTLs to personalized transcriptomics

    Nature Reviews. Genetics 12:277–282.

    https://doi.org/10.1038/nrg2969

    • Google Scholar
29. 1. Montgomery SB
    2. Sammeth M
    3. Gutierrez-Arcelus M
    4. Lach RP
    5. Ingle C
    6. Nisbett J
    7. Guigó R
    8. Dermitzakis ET

    (2010) Transcriptome genetics using second generation sequencing in a Caucasian population

    Nature 464:773–777.

    https://doi.org/10.1038/nature08903

    • Google Scholar
30. 1. Moore T
    2. Haig D

    (1991) Genomic imprinting in mammalian development: a parental tug-of-war

    Trends in Genetics 7:45–49.

    https://doi.org/10.1016/0168-9525(91)90230-N

    • Google Scholar
31. 1. Mott R
    2. Yuan W
    3. Kaisaki P
    4. Gan X
    5. Cleak J
    6. Edwards A
    7. Baud A
    8. Flint J

    (2014) The architecture of parent-of-origin effects in mice

    Cell 156:332–342.

    https://doi.org/10.1016/j.cell.2013.11.043

    • Google Scholar
32. 1. Noordermeer D
    2. de Laat W

    (2008) Joining the loops: beta-globin gene regulation

    IUBMB Life 60:824–833.

    https://doi.org/10.1002/iub.129

    • Google Scholar
33. 1. Peters J

    (2014) The role of genomic imprinting in biology and disease: an expanding view

    Nature Reviews. Genetics 15:517–530.

    https://doi.org/10.1038/nrg3766

    • Google Scholar
34. 1. Pickrell JK
    2. Marioni JC
    3. Pai AA
    4. Degner JF
    5. Engelhardt BE
    6. Nkadori E
    7. Veyrieras JB
    8. Stephens M
    9. Gilad Y
    10. Pritchard JK

    (2010) Understanding mechanisms underlying human gene expression variation with RNA sequencing

    Nature 464:768–772.

    https://doi.org/10.1038/nature08872

    • Google Scholar
35. 1. Prickett AR
    2. Oakey RJ

    (2012) A survey of tissue-specific genomic imprinting in mammals

    Molecular Genetics and Genomics 287:621–630.

    https://doi.org/10.1007/s00438-012-0708-6

    • Google Scholar
36. 1. Rando OJ

    (2012) Daddy issues: paternal effects on phenotype

    Cell 151:702–708.

    https://doi.org/10.1016/j.cell.2012.10.020

    • Google Scholar
37. 1. Reik W
    2. Collick A
    3. Norris ML
    4. Barton SC
    5. Surani MA

    (1987) Genomic imprinting determines methylation of parental alleles in transgenic mice

    Nature 328:248–251.

    https://doi.org/10.1038/328248a0

    • Google Scholar
38. 1. Saga Y
    2. Takeda H

    (2001) The making of the somite: molecular events in vertebrate segmentation

    Nature Reviews. Genetics 2:835–845.

    https://doi.org/10.1038/35098552

    • Google Scholar
39. 1. Schadt EE
    2. Monks SA
    3. Drake TA
    4. Lusis AJ
    5. Che N
    6. Colinayo V
    7. Ruff TG
    8. Milligan SB
    9. Lamb JR
    10. Cavet G
    11. Linsley PS
    12. Mao M
    13. Stoughton RB
    14. Friend SH

    (2003) Genetics of gene expression surveyed in maize, mouse and man

    Nature 422:297–302.

    https://doi.org/10.1038/nature01434

    • Google Scholar
40. 1. Scott CA
    2. van Huyen D
    3. Bany BM

    (2012) Angiopoietin-like gene expression in the mouse uterus during implantation and in response to steroids

    Cell and Tissue Research 348:199–211.

    https://doi.org/10.1007/s00441-012-1337-4

    • Google Scholar
41. 1. Stegle O
    2. Parts L
    3. Piipari M
    4. Winn J
    5. Durbin R

    (2012) Using probabilistic estimation of expression residuals (PEER) to obtain increased power and interpretability of gene expression analyses

    Nature Protocols 7:500–507.

    https://doi.org/10.1038/nprot.2011.457

    • Google Scholar
42. 1. Wang J
    2. Williams RW
    3. Manly KF

    (2003) WebQTL: web-based complex trait analysis

    Neuroinformatics 1:299–308.

    https://doi.org/10.1385/NI:1:4:299

    • Google Scholar
43. 1. Wang S
    2. Yehya N
    3. Schadt EE
    4. Wang H
    5. Drake TA
    6. Lusis AJ

    (2006) Genetic and genomic analysis of a fat mass trait with complex inheritance reveals marked sex specificity

    PLOS Genetics 2:e15.

    https://doi.org/10.1371/journal.pgen.0020015

    • Google Scholar
44. 1. Weaver IC
    2. Cervoni N
    3. Champagne FA
    4. D'Alessio AC
    5. Sharma S
    6. Seckl JR
    7. Dymov S
    8. Szyf M
    9. Meaney MJ

    (2004) Epigenetic programming by maternal behavior

    Nature Neuroscience 7:847–854.

    https://doi.org/10.1038/nn1276

    • Google Scholar
45. 1. Wittkopp PJ
    2. Haerum BK
    3. Clark AG

    (2004) Evolutionary changes in cis and trans gene regulation

    Nature 430:85–88.

    https://doi.org/10.1038/nature02698

    • Google Scholar
46. 1. Wolf JB
    2. Leamy LJ
    3. Roseman CC
    4. Cheverud JM

    (2011) Disentangling prenatal and postnatal maternal genetic effects reveals persistent prenatal effects on offspring growth in mice

    Genetics 189:1069–1082.

    https://doi.org/10.1534/genetics.111.130591

    • Google Scholar
47. 1. Yang X
    2. Schadt EE
    3. Wang S
    4. Wang H
    5. Arnold AP
    6. Ingram-Drake L
    7. Drake TA
    8. Lusis AJ

    (2006) Tissue-specific expression and regulation of sexually dimorphic genes in mice

    Genome Research 16:995–1004.

    https://doi.org/10.1101/gr.5217506

    • Google Scholar
48. 1. Yoshida K
    2. Shimizugawa T
    3. Ono M
    4. Furukawa H

    (2002) Angiopoietin-like protein 4 is a potent hyperlipidemia-inducing factor in mice and inhibitor of lipoprotein lipase

    Journal of Lipid Research 43:1770–1772.

    https://doi.org/10.1194/jlr.C200010-JLR200

    • Google Scholar
49. 1. Yvert G
    2. Brem RB
    3. Whittle J
    4. Akey JM
    5. Foss E
    6. Smith EN
    7. Mackelprang R
    8. Kruglyak L

    (2003) Trans-acting regulatory variation in Saccharomyces cerevisiae and the role of transcription factors

    Nature Genetics 35:57–64.

    https://doi.org/10.1038/ng1222

    • Google Scholar

Author details

1. Noah Spies

   1. Department of Pathology, Stanford University School of Medicine, Stanford, United States
   2. Department of Genetics, Stanford University School of Medicine, Stanford, United States

   Contribution

   NS, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Contributed equally with

   Cheryl L Smith

   Competing interests

   The authors declare that no competing interests exist.

2. Cheryl L Smith

   1. Department of Pathology, Stanford University School of Medicine, Stanford, United States
   2. Department of Genetics, Stanford University School of Medicine, Stanford, United States

   Contribution

   CLS, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Contributed equally with

   Noah Spies

   Competing interests

   The authors declare that no competing interests exist.

3. Jesse M Rodriguez

   1. Department of Computer Science, Stanford University, Stanford, United States
   2. Biomedical Informatics Program, Stanford University School of Medicine, Stanford, United States

   Present address

   Google Inc., Mountain View, United States

   Contribution

   JMR, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

4. Julie C Baker

   Department of Genetics, Stanford University School of Medicine, Stanford, United States

   Contribution

   JCB, Conception and design, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

5. Serafim Batzoglou

   Department of Computer Science, Stanford University, Stanford, United States

   Contribution

   SB, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

6. Arend Sidow

   1. Department of Pathology, Stanford University School of Medicine, Stanford, United States
   2. Department of Genetics, Stanford University School of Medicine, Stanford, United States

   Contribution

   AS, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   arend@stanford.edu

   Competing interests

   The authors declare that no competing interests exist.

• 2,226

  views

• 402

  downloads

• 3

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.