Paternally induced transgenerational inheritance of susceptibility to diabetes in mammals

We first generated a nongenetic prediabetes mouse model that closely simulates the metabolic abnormalities of human prediabetes. Insulin resistance was induced in male mice by feeding a high-fat diet, and impaired fasting glucose was induced by injecting these mice with a low dose of STZ, which does not cause diabetes in chow-fed mice, as described previously (17). As expected, prediabetic male founders displayed increased body weight, energy intake, adiposity, plasma glucose, insulin, leptin, cholesterol, and triglyceride levels (Fig. S2 and Table S1). Furthermore, upon glucose and insulin tolerance tests (GTTs and ITTs), they showed decreased insulin sensitivity as well as impaired glucose tolerance compared with controls (Fig. S2 C–E). The homeostasis model assessment of insulin resistance index (HOMA-IR) was also increased (Table S1).

To determine the potential effects of paternal prediabetes on offspring, we mated male founders of prediabetes or control with females and examined physiological and metabolic changes in their offspring at 16 wk of age. Paternal prediabetes did not alter body weight, fat mass, or energy intake in offspring (Fig. 1 A and B, Fig. S3 A and B, and Table S1). Most serum biomarkers were unaffected, although leptin and insulin both showed trends to increased levels in offspring of prediabetic fathers (P = 0.06 and 0.07 in males, and P = 0.18 and 0.09 in females). Next we assessed glucose tolerance and insulin sensitivity in offspring. Paternal prediabetes showed increased blood glucose rise during a glucose tolerance test at 16 wk of age (Fig. 1C and Fig. S3C). A similar pattern was observed at 30 wk, but with a further impairment of glucose tolerance evidenced by a lower P value (Fig. S3 E and G). Concomitantly, the insulin tolerance test revealed a significant decrease in insulin sensitivity in offspring of prediabetic fathers, which worsened with increasing age of the offspring (Fig. 1D and Fig. S3 D, F, and H). Moreover, when these mice reached 12 mo, they exhibited increased fasting blood glucose levels (female control 5.23 ± 0.19 mM, prediabetes 7.38 ± 0.45 mM, P < 0.01; male control 5.15 ± 0.16 mM, prediabetes 7.27 ± 0.33 mM, P < 0.01), accompanied by more serious glucose intolerance and insulin insensitivity (Fig. 1 E and F) in males as seen in their fathers.

To determine the mechanisms of the glucose intolerance and insulin insensitivity observed in offspring, we performed genome-wide microarray analyses. Paternal prediabetes altered the expression of 402 genes (97 up-regulated and 305 down-regulated, P < 0.05, fold change >2; Fig. 2A and Dataset S1) in the pancreatic islets of offspring. A large proportion of these genes had enriched gene ontology terms associated with insulin and glucose metabolism, including GTPase activity, GTP and ATP binding, sugar binding, and calcium binding (Dataset S2).

We further focused on three groups of genes that have been shown to affect glucose-stimulated insulin secretion: (i) maturity-onset diabetes of the young (MODY) genes, (ii) glucose uptake and metabolism genes, and (iii) insulin signaling genes. Of the six MODY genes, Hnf1a was significantly decreased in offspring of prediabetic fathers, with 57% expression compared with the control (Fig. 2B). Expression of other MODY genes did not differ significantly between groups. mRNAs for the glucose transporters Glut1 and Glut2 were all detectable in islets, but there were no significant differences in expression (Fig. 2C). In contrast to glucose transporter expression, one of the downstream enzymes in the glycolytic pathway, Pfk, was significantly decreased (Fig. 2C). Expression of G6pc and Pgm1 did not differ significantly. Among the insulin signaling genes, there were no significant differences in expression of insulin receptor (Insr) and Akt2 between the two groups (Fig. 2D). However, phosphatidylinositol (PI) 3-kinase subunits (Pik3ca and Pik3r1) mRNAs were significantly lower in the offspring of prediabetic fathers. On the other hand, expression of protein tyrosine phosphatase non-receptor type 1 (Ptpn1), a tyrosine phosphatase that inhibits insulin signaling, was significantly increased. Proinsulin (Ins2) showed a trend toward decreased expression (P = 0.09). These changes would be likely to lead to impaired insulin signaling in offspring of prediabetic fathers.

Another group of genes that has been reported to be associated with insulin resistance is the basic helix–loop–helix Per/AhR/Arnt/Sim (bHLH-PAS) family. Among the bHLH-PAS family members, Hif3a was significantly decreased and Arnt showed a trend toward decreased expression (P = 0.08; Fig. 2E). We confirmed our results by quantitative real-time PCR (qRT-PCR), which showed a similar expression difference in four genes normalized to two housekeeping genes across 12 animals examined (Fig. S4). Overall, these molecular findings are consistent with metabolic changes in offspring.

As cytosine methylation is a widespread DNA modification that carries heritable information between generations (6, 18), we therefore turned to genome-wide studies to search for differentially methylated loci between control and prediabetic offspring. We performed methylated DNA immunoprecipitation sequencing (MeDIP-Seq) to characterize cytosine methylation across the entire genome, and fraction of methylated cytosine was calculated for a variety of elements including upstream2k, downstream2k, 5′ untranslated region (5′UTR), 3′UTR, coding sequence (CDS), and intron. Paternal prediabetes altered the methylation of 446 upstream2k, 373 downstream2k, 173 5′UTR, 272 3′UTR, 1,634 CDS, and 5,602 intron element-associated genes (Dataset S3) in the pancreatic islets of offspring, respectively. However, changes in cytosine methylation did not globally correlate with changes in gene expression (Datasets S1 and S3), indicating that gene expression is unlikely to be epigenetically specified at each individual gene. Importantly, we found substantial increases in methylation at intragenic regions of Pik3r1 and Pik3ca (Fig. 3A), methylation at these loci is likely to inhibit transcriptional activity, as it is associated with their low expression levels (Fig. 2D). Bisulfite sequencing for these genes further confirms the methylation status (Fig. 3B). In addition, Ptpn1, which is up-regulated in the pancreatic islets of prediabetic offspring, exhibited decrease in cytosine methylation (Fig. 3 A and B), further suggesting that epigenetic regulation could drive a substantial fraction of the observed expression changes in offspring. Together, these results identify several differentially methylated loci that are strong candidates to be upstream controllers of the gene expression changes in offspring.

The above results led us to consider the hypothesis that paternal prediabetes affects cytosine methylation patterns in sperm. To globally investigate effects of paternal prediabetes on sperm methylation, we isolated sperm from control and prediabetic males, and surveyed cytosine methylation patterns across the entire genome by MeDIP-Seq. Notably, global cytosine methylation profiles were altered in prediabetes samples compared with controls, and the methylation of 263 upstream2k, 278 downstream2k, 121 5′UTR, 247 3′UTR, 1,299 CDS, and 4,354 intron element-associated genes were changed, respectively (Dataset S4). These results indicate that the sperm epigenome is largely responsive to environmental differences provided by males. Of particular interest, we observed that a large proportion of differentially methylated genes identified in sperm overlapped with that of pancreatic islets. In sperm, of the identified 3,020 intragenic element-associated (including intron, CDS, 5′UTR, and 3′UTR) hypermethylated genes, 1,189 (∼39%) were also hypermethylated in pancreatic islets, and of the 3,001 intragenic element-associated hypomethylated genes, 1,080 (∼36%) were also hypomethylated in pancreatic islets (Fig. 4A). On the other hand, of the 287 intergenic element-associated (including upstream2k and downstream2k) hypermethylated genes, only 9 (∼3%) were also hypermethylated in pancreatic islets, and of the 254 hypomethylated genes, only 7 (∼3%) were also hypomethylated in pancreatic islets (Fig. S5A). These results indicate that cytosine methylation in sperm (especially for intragenic regions) is a strong factor in determining methylation status in somatic tissues.

Most interestingly, we found a substantial increase in methylation at the same region of Pik3ca in sperm (Fig. 3A), which has been found to be hypermethylated in pancreatic islets. We then assayed the methylation status of this locus by bisulfite sequencing, and found an average difference of 35.8% methylation between control and prediabetes sperm (Fig. 4C). Given that MeDIP is unlikely to identify small differences (such as 10% or 20%) in methylation at a small number of cytosines, we performed bisulfite sequencing to determine the methylation status of Pik3r1 and Ptpn1 in sperm. The methylation level of Pik3r1 was increased in prediabetes samples compared with controls (78.6% vs. 96.1%; Fig. 4B). In contrast, methylation of Ptpn1 was unaltered between control and prediabetes sperm (Fig. S5B), indicating that differential methylation at this locus observed in islets is established at some point during development. To test whether genes differentially methylated are de novo methylated or potentially inherited DNA methylation from sperm, we performed bisulfite sequencing on embryonic day (E) 3.5 blastocysts. Both Pik3r1 and Pik3ca showed higher methylation levels in prediabetes samples compared with controls (Fig. 4 B and C), suggesting that these genes partially inherit methylated alleles from sperm. Together, these results indicate that cytosine methylation status in sperm strongly predisposes toward methylation in blastocysts, perhaps by incomplete postfertilization demethylation of methylated cytosines.

In our study, the offspring (F₁ generation) derived from the prediabetic fathers (F₀ generation) exhibited a prediabetes phenotype. We then asked whether the metabolic and epigenetic changes in the F₁ generation can be passed to the next generation (F₂ generation). We mated male F₁ generation of prediabetes or control mice that were aged 12 mo with normal females and then examined metabolic and epigenetic changes in their offspring. We first assessed metabolic changes in the F₂ generation at 16 wk of age and found that the F₂ generation also exhibited impaired glucose tolerance and decreased insulin sensitivity (Fig. 5 A and B and Fig. S6 A and B), accompanied by nonalteration in fasting blood glucose levels (female control 4.53 ± 0.11 mM, prediabetes 4.64 ± 0.15 mM, P = 0.74; male control 4.51 ± 0.18 mM, prediabetes 4.72 ± 0.23 mM, P = 0.52) as seen in the F₁ generation at 16 wk of age. We then assessed epigenetic changes in these animals. We randomly selected 10 regions distributed on different chromosomes that were most affected by paternal prediabetes from the islets MeDIP-Seq dataset. MeDIP-qPCR analysis showed that all of these regions were significantly affected in the F₂ generation (Fig. 5B). Moreover, we examined the methylation status of Pik3r1, Pik3ca, and Ptpn1, and also found a similar methylation status compared with the F₁ generation (Fig. 5C). Because the F₁ generation animals lived in standard laboratory housing without any treatment and their offspring exhibited similar phenotypic and epigenetic changes, the observed effects of epigenetic inheritance are most likely due to the prediabetes-associated physiological and metabolic conditions in fathers, although we cannot completely rule out the effects of treatment on the F₀ generation males.

Although STZ is an alkylating agent, it is recognized and transported by Glut2; thus it is mainly targeted to pancreatic β cells, which express high levels of this protein (19, 20). Furthermore, the half-life of STZ is very short (∼5–15 min in vivo), which indicates that it cannot function for a prolonged time in vivo, but mating of animals occurs post 4 wk of STZ injection. These facts do not favor the low dose of injected STZ as the reason for sperm epigenetic alterations. To further assess STZ’s effects, an identical amount of STZ was injected in chow-fed mice, which does not cause diabetes but has the same effects of STZ on sperm DNA (17). Western blot analysis showed that expression of the sensitive DNA-damage marker γH2AX was unchanged post 1 d, 1 wk, or 4 wk of STZ injection (Fig. 6A), indicating that the low dose of STZ does not cause DNA damage in sperm. Next we asked whether the low dose of STZ causes epigenomic changes of sperm DNA. We randomly selected 10 regions distributed on different chromosomes that were most affected by paternal prediabetes from the sperm MeDIP-Seq data. MeDIP-qPCR analysis showed that none of these regions were affected in sperm of STZ-alone treated nondiabetic animals (Fig. 6B). Bisulfite sequencing analysis showed that the methylation status of Pik3r1 and Pik3ca was also unaffected by STZ-alone treatment (Fig. 6C). These results indicate that the low dose of STZ does not cause epigenomic alterations in sperm. To further clarify this issue, we mated STZ-alone–treated nondiabetic males with females and examined epigenetic changes in islets of offspring at 3 wk of age. Similarly, we used the randomly selected 10 regions distributed on different chromosomes that were most affected by paternal prediabetes as used in the F₂ generation. MeDIP-qPCR analysis showed that none of these regions were affected by STZ-alone treatment (Fig. 6D). Bisulfite sequencing analysis showed that the methylation status of Pik3r1, Pik3ca, and Ptpn1 was also unaffected by STZ-alone treatment (Fig. 6E). These results largely exclude the possibility that the effects of epigenetic inheritance are due to the action of STZ on sperm DNA.

All animal care and use procedures were in accordance with guidelines of the Institutional Animal Care and Use Committee of the Institute of Zoology, Chinese Academy of Sciences. All experiments were performed with Institute of Cancer Research mice. We generated prediabetes mouse models according to a previous study, with slight changes (17). Male F₀ founders were weaned from mothers at 3 wk of age, and sibling males were placed into cages with HFD (33% energy as fat) or control diet until 12 wk of age, at which point mice fed with HFD were injected intraperitoneally with a subdiabetogenic dose of STZ (100 mg/kg body weight) and kept on the same diet for 4 wk. Fasting blood glucose (2 h of fasting) was examined each week post-STZ for 4 wk, and only glucose levels at ∼7–11 mM were considered as prediabetes. At 16 wk, male F₀ founders were mated with females. At 3 wk of age, a portion of the offspring were killed and islets were generated, each from a different father, which were used for further expression and methylation analysis. Phenotype data from one offspring per father, chosen at random, were generated at 16 wk of age. GTT and ITT were also performed in offspring at 30 wk and 12 mo of age. Additional details are provided in SI Methods.

Pancreatic islets were isolated as previously described (31). Samples from six control and six prediabetic offspring (three females and three males for each group), each from different fathers, were chosen for microarray analysis using NimbleGen Mouse Gene Expression 12× 135K array (Roche NimbleGen), according to the manufacturer’s instructions. Additional details are provided in SI Methods.

We analyzed mRNA levels by qRT-RCR after reverse transcription as described before (32). The same samples used for microarray analysis (one offspring per father; n = 6, prediabetes; n = 6, control) were used for qRT-PCR. Additional details are provided in SI Methods.

For offspring, the same animals used in the microarray analysis were used for the MeDIP study. For sperm, the animals used were exactly the fathers of offspring used for microarray analysis. The MeDIP method was adapted from a previous study (33). For qPCR following MeDIP, recovered DNA fractions were diluted 1/50 and measured using real-time PCR with an Applied Biosystems7500 system and SYBR Green SuperMix UDG (Invitrogen). Primers are summarized in Dataset S5. Additional details are provided in SI Methods.

Bisulfite genomic sequencing was performed as previously described (32). For offspring, the same samples used in microarray were used for bisulfite sequencing. For sperm, the animals used were exactly the fathers of offspring used for the microarray analysis. For bisulfite sequencing, pooled DNA of six animals (equally from each animal) for each group was used for the analysis. For E3.5 blastocysts, each bisulfite treatment was performed on 18 pooled blastocysts derived from three animals (6 blastocysts randomly selected from each animal) for each group. Additional details are provided in SI Methods.

Phenotype data were analyzed by SPSS 16.0 after log transformation or square-root transformation unless raw data were normally distributed. Measurements at single time points were analyzed by ANOVA, or, if appropriate, by two-tailed Student t test. Time courses were analyzed by repeated-measurements ANOVA. All data are shown as mean ± SEM; P < 0.05 was considered statistically significant.