There is now considerable evidence for non-genomic transmission of induced phenotypic traits between generations [4]. Such processes may provide an important mechanism for the inheritance of disease traits in humans [5], [6]. For example, mortality from diabetes was increased in men if the paternal grandfather had been exposed to abundant nutrition during their pre-pubertal slow growth period [7]. Also, the daughters of women exposed to nutrient restriction and stress during pregnancy as a result of the Dutch Hunger Winter showed decreased birth weight and an increased risk of insulin resistance, and, in turn, their granddaughters also were born with a lower birth weight despite adequate nutrition during their daughter's pregnancy [8], [9]. Similarly, metabolic dysregulation induced by nutritional or hormonal interventions during F0 pregnancy in animal models can be transmitted to more than one generation of offspring. For example, feeding pregnant rats a protein-restricted (PR) diet during pregnancy in the F0 generation induced elevated blood pressure and endothelial dysfunction [10], and insulin resistance [11], [12] in the F1 and F2 offspring, despite adequate nutrition following weaning and during pregnancy in the F1 generation. Adverse effects of feeding a PR diet during F0 pregnancy on glucose homeostasis have been detected in F3 offspring [13]. Finally, the administration of dexamethasone to dams in late pregnancy induced an increased expression of the glucocorticoid receptor (GR) and of its target gene phosphoenolpyruvate carboxykinase (PEPCK) in the liver of the F1 and F2 offspring, although these effects were lost in F3 offspring [14].

Although the passage of induced traits between generations is well known, the underlying mechanisms are poorly understood. Epigenetic processes have been suggested to be involved [6]. Supplementation of the diet of F0 A^vy mice with methyl donors induced phenotypic changes through altered methylation of the A^vy locus in F1 and F2 offspring [15]. Hypomethylation and increased mRNA expression of the hepatic PPARα and GR promoters has been reported in F1 and F2 offspring of F0 dams fed a PR diet during pregnancy [15]. Female germ cells which give rise to the F2 generation are formed in F1 offspring when they were in utero and thus are exposed to the environmental challenge to which the F1 offspring are exposed. Thus it is unclear whether the results of these studies represent transmission of induced phenotypes between generations. Therefore, conclusive demonstration of such transmission between generations requires that induced changes are detected in F3 offspring [17]. Anway et al. [18] showed transmission essentially without further change of altered male fertility and changes in promoter methylation of specific genes in testis up to F4 generation offspring of F0 dams exposed to the endocrine disruptor vinclozolin. Such germline effects involved altered expression of DNA methyltransferases [19].

The twenty most up- or down- regulated genes in each generation are listed in Table 1. One gene, lymphocyte activation gene-3, was among the most up-regulated genes in all three generations. Pancreatic amylase was among the most up-regulated genes in F1 and the most down-regulated genes in F3, but was not among the twenty most altered genes in F2. Trefoil factor-2 was among the most up-regulated genes in F2 and F3, but was not among the twenty most altered genes in F1. No other genes were among the twenty most altered genes in more than one generation.

113 Genes differed by at least 2 fold between F0 dietary groups in all three generations. The number of up-regulated genes in F1 was 49/113, in F2 was 69/113 genes and in F3 was 71/113. 35/113 was up-regulated in all three generations (Figure 1) and 25/113 were down-regulated in all three generations in PR compared to PS offspring (Figure 2). However, 53/113 transcripts showed directionally opposite differences in expression between generations (Figure 3).

We validated the array by real time RTPCR for seven genes which differed in the patterns of difference between PS and PR offspring in samples from all three generations. Although the magnitude of difference between PS and PR offspring varied between real time RTPCR and the array, all seven genes showed the same direction of difference between PS and PR lines and the same pattern of change between generations when analysed either technique (Table 2). Linear regression analysis showed a significant positive association (P=0.013, r=0.86) between the difference between PR and PS samples in F3 detected by microarray and by real time RTPCR.

Assessment of the number of genes present in the Biological Processes and Molecular Functions ontologies showed differential changes within individual sub-ontologies. The data shown in Figure 4 do not indicate enrichment within individual categories. The total number of genes which showed differential expression between PS and PR offspring increased between F1, F2 and F3 generations in 15 of the 25 sub-ontologies of the Biological Processes ontology (Figure 4A). 3/25 showed a decrease in F2, but similar numbers in F1 and F3. The number of genes which differed by at least 2 fold between PR and PS offspring in the remaining ontologies was similar in all three generations. In the Molecular Functions ontology, 3/9 sub-ontologies showed an increase between F1 and F3 offspring in the number of genes which differed by at least 2 fold between PR and PS offspring (Figure 4B). One sub-ontology, Transporter Activity, showed a lower number of altered genes in F2 with a similar number of altered genes in F1 and F3. The remaining sub-ontologies showed similar numbers of altered genes in all three generations.

The data shown in Figure 5 show the proportions of genes within each category which were either up or down regulated, and do not indicate enrichment within categories. In the Biological Processes ontology, all but two of the sub ontologies, Reproductive Processes and Immune System Processes, showed an increase in the proportion of up-regulated genes in liver from F2 offspring compared to F1 (Figure 5A). In F3 offspring, the proportion of up-regulated genes was increased in all of the sub-ontologies compared to F1 offspring (Figure 5A). Compared to F2 offspring, 24/25 sub-ontologies showed an increase in up-regulated genes, while the proportion of down-regulated genes was increased in Rhythmic Processes, compared to F2 offspring (Figure 5A). In the Molecular Function ontology, the proportion of up-regulated genes was greater in F2 offspring in 7/9 sub-ontologies, while the proportion of up-regulated genes was decreased in Enzyme Regulator Activity and Electron Carrier Activity, compared to F1 offspring (Figure 5B). In F3 offspring, the proportion of up-regulated genes was greater in 8/9 sub-ontologies, but decreased in Enzyme Regulator Activity, compared to F1 offspring, but was increased in all sub-ontologies compared to F2 offspring (Figure 5B).

The results of KEGG analysis of pathways which are relevant to hepatic function which contained at least ten genes which differed by 2 fold or more between PS and PR lines are shown in Figure 6. In the F1 generation, Calcium Signalling, Cell Adhesion, Adherens Junction, Jak-STAT Signalling and Wnt Signalling pathways were over-represented amongst the up-regulated genes. Tight junction and Steroid Hormone Biosynthesis pathways were over-represented amongst the down-regulated genes, while Jak-STAT Signalling and Wnt Signalling pathways were under-represented amongst the down-regulated genes. In contrast to F1, the Tight Junctions pathway was over-represented amongst the up-regulated genes and under-represented amongst the down-regulated genes in F2. MAP Kinase Signalling and ECM Receptor Interaction pathways were over-represented amongst the up-regulated genes in F2, but were not altered in F1. In contrast to F2, but similar to F1, the Tight Junctions pathway was over-represented amongst the down-regulated genes in F3. Cell Adhesion Molecules and Retinol Metabolism pathways were altered in F3, but not F1 or F2. The Adherens Junction pathway was over-represented amongst the up-regulated genes in all three generations.

Previous studies have shown that phosphoenolpyruvate carboxykinase (PEPCK) activity and mRNA expression increased in the offspring of dams fed a PR diet [21], [22]. PEPCK mRNA expression is regulated by the methylation status of its promoter [23]. Fasting plasma glucose was increased in F1 and F2, but was lower in F3, PR offspring compared to PS offspring (F0 diet, generation, interaction all P<0.0001) (Figure 7). We, therefore, measured the level of PEPCK mRNA expression and the methylation of individual CpG dinucleotides located within or proximal to response elements for transcription factors which are known to regulate PEPCK transcription [24] (Figure 8) in adult female offspring liver. There was a significant interaction between generation and F0 diet on PEPCK mRNA expression (P<0.0001) (Figure 7). There was no difference between generations in PEPCK expression in PS offspring. PEPCK expression was higher in PR offspring in F1 and F2, but lower in F3, than PS offspring. There was no significant effect of F0 diet on the methylation status of six of the nine CpGs which were measured in the PEPCK promoter (data not shown). However, CpG −508, was hypomethylated in F2 and F3 PR offspring compared to PS offspring (generation P=0.009, F0 diet P<0.0001, interaction P=0.002) (Figure 7). CpG −440 was hypomethylated in all three generations of PR offspring compared to PS offspring (generation not significant (NS); F0 diet P=0.0023, interaction NS) (Figure 7). CpG −90 was hypomethylated in F1 PR offspring, did not differ significantly from PS offspring in F2, but was hypermethylated in F3 offspring compared to PS offspring (generation P<0.0001, F0 diet P=0.015, interaction P=0.022) (Figure 7).

The differences in the expression of the liver transcriptome between PS and PR lines differed between F1, F2 and F3 generations in terms of the number of genes which showed altered expression, the magnitude of difference, the distribution of up and down regulated genes and the ontologies affected. Furthermore, the number of genes which differed between PS and PR lines increased between F2 and F3 generations. Of the genes which differed between PS and PR offspring in all three generations, 47% differed between generations in the direction of the difference between PS and PR lines, while the remainder maintained the same direction of difference in all three generations. Together these findings do not support the suggestion that alterations in the transcriptome in F1 are passed without change to subsequent generations. Thus the expression of the transcriptome in the offspring of any generation following a challenge during F0 pregnancy cannot be assumed simply to reflect that of preceding or subsequent generations.

These findings are consistent with some reports which show variation or loss of induced phenotypes between generations. For example, feeding a PR diet during F0 pregnancy induced impaired glucose homeostasis in male F1 offspring which was exacerbated in F2, but then fell to a level below control offspring in F3 [13]. This is in contrast to stable prevention of inter-generational drift in body weight in offspring of F0 A^vy mice supplemented with methyl donors [28]. Together these findings suggest that the effects of altered nutrition on the phenotype of subsequent generations of offspring may be more complex than those mediated by a single exposure to an endocrine disruptor [18]. The present study showed that some genes, ontologies and pathways differed by a similar amount and in the same direction in each generation. These findings are consistent with those of Anway et al. [18], [19] and so suggest that for specific genes transmission of altered expression of may have occurred unchanged through the germline.

The majority of genes which were altered in all three generations showed variation in the magnitude and direction of the effect of the maternal PR diet between generations. Transgenerational divergence in body weight has been shown between control A^vy mice and those supplemented with methyl donors, which implies an interaction between gene mutations and epigenetic processes [28], although the mechanism underlying the amplification of obesity is not known. One possible explanation is differences in the environment provided by the mother during development between generations providing different signals to the developing offspring in each generation (Figure 9). For example, the environmental signals received by the F1 fetus would reflect an interaction between the phenotype of the F0 dam with the PR diet leading to an altered phenotype in the F1 offspring. In turn, signals received by the F2 fetus would reflect the interaction of the phenotype of F1 dam with the environment, which differs from that of the F0 dam. The signals received by the F3 fetus would, therefore, reflect an interaction between the phenotype of the F2 dam and the environment, which differs from those of the F0 and F1 dams. Furthermore, germ cells destined to become F2 offspring would be exposed to signals from the F0 dams and those from the F1 dams, and those destined to become F3 offspring would be exposed to signals from the F1 dam as wells the F2 dam. Together, such complex interactions between mother and offspring may provide a mechanism by which the number of altered genes is greater in F3 than F1, and by which ontologies and pathways are altered in F3, but not in previous generations. Such maternal effects on the phenotype of the offspring are well-established and have been suggested to be a mechanism for generation of novel phenotypes [29], although the molecular basis of such changes has not been shown previously. This model also suggests a mechanism by which single phenotypic traits induced in F1 offspring are apparently lost in future generations [14] and so emphasises a need for comprehensive analysis of the phenotype of offspring in transgenerational studies.

The twenty genes which showed the greatest difference between PR and PS offspring in each generation had diverse function, some of which, for example surfactant protein C, have uncertain functional significance in liver. Two of the genes which differed between PR and PS offspring in more than one generation, trefoil factor-2 and lymphocyte activation gene-3, are associated with tissue repair and limitation of immune response, respectively, although whether they exhibit these function in liver is unclear. Amylase-2 activity has been demonstrated in liver, although its contribution of glycogenolysis relative to glycogen phosphorylase is unclear [30]. Amyase-2 was up-regulated in F1 PR offspring which had raised fasting blood glucose concentration but was down-regulated in F3 PR offspring with normal blood glucose. This suggests that altered hepatic amylase-2 activity may contribute to differences between generations in glucose homeostasis between PS and PR offspring.

Fasting plasma glucose was increased in F1 and F2, but not F3 offspring. Because gluconeogenesis is the major source of plasma glucose in rats fasted for 12 hours [31] and F1 offspring of dams fed a PR diet show increased gluconeogenic activity [22], we measured the mRNA expression and methylation of the PEPCK promoter in the liver of the offspring. The pattern of PEPCK mRNA expression between generations followed that of plasma glucose concentration, with the exception that PEPCK expression was lower in PR than PS offspring in F3. This suggests that, in agreement with previous findings of the effect of prenatal exposure of F1 offspring to dexamethasone [14], changes in the regulation of PEPCK transcription are an important process underlying differences in plasma glucose concentrations between generations. Of the nine CpGs measured in the PEPCK promoter, three CPGs showed differential methylation between PS and PR lines. All three are located within known transcription factor response elements [32] and thus changes in the level of methylation may be expected to be associated with altered transcriptional activity. CpG −508 is located proximal to a heat-shock factor response element, while CpG −440 is within a PPAR response element and CpG −90 within a cAMP response element. The PEPCK promoter contains multiple response elements and hence overall capacity for transcription represents the overall regulation of its activity. Thus the level of PEPCK mRNA cannot be assumed to simply reflect the methylation status of single CpGs. However, CpGs −508, −440 and −90 each showed specific changes between generations in the relative level of methylation between PS and PR offspring. These differences in the methylation of individual CpGs are consistent with the models described above for the transmission of induced phenotypes between generations. Thus methylation of CpG −440 may be transmitted unchanged through germ cells, although this would require the preservation of these epigenetic marks through, albeit incomplete, genome-wide demethylation at fertilisation [33]. In contrast, the patterns of change in CpGs −508 are consistent with differences in maternal signals between generations. Thus these findings support the suggestion that epigenetic processes may be involved in the transmission of induced phenotypes between generations.