Prolactin Receptor Signaling Regulates a Pregnancy-Specific Transcriptional Program in Mouse Islets

Abstract

Pancreatic β-cells undergo profound hyperplasia during pregnancy to maintain maternal euglycemia. Failure to reprogram β-cells into a more replicative state has been found to underlie susceptibility to gestational diabetes mellitus (GDM). We recently identified a requirement for prolactin receptor (PRLR) signaling in the metabolic adaptations to pregnancy, where β-cell–specific PRLR knockout (βPRLRKO) mice exhibit a metabolic phenotype consistent with GDM. However, the underlying transcriptional program that is responsible for the PRLR-dependent metabolic adaptations during gestation remains incompletely understood. To identify PRLR signaling gene regulatory networks and target genes within β-cells during pregnancy, we performed a transcriptomic analysis of pancreatic islets isolated from either βPRLRKO mice or littermate controls in late gestation. Gene set enrichment analysis identified forkhead box protein M1 and polycomb repressor complex 2 subunits, Suz12 and enhancer of zeste homolog 2 (Ezh2), as novel candidate regulators of PRLR-dependent β-cell adaptation. Gene ontology term pathway enrichment revealed both established and novel PRLR signaling target genes that together promote a state of increased cellular metabolism and/or proliferation. In contrast to the requirement for β-cell PRLR signaling in maintaining euglycemia during pregnancy, PRLR target genes were not induced following high-fat diet feeding. Collectively, the current study expands our understanding of which transcriptional regulators and networks mediate gene expression required for islet adaptation during pregnancy. The current work also supports the presence of pregnancy-specific adaptive mechanisms distinct from those activated by nutritional stress.

Gestational diabetes mellitus (GDM) is a metabolic disorder that emerges only during pregnancy, yet it presents serious health outcomes to mother and offspring even decades after birth (1). As with other forms of diabetes mellitus, a relative deficiency of functional β-cells is a major factor in the pathogenesis of GDM (2). Controversy exists, however, regarding the underlying etiology of GDM when considering its relationship with obesity and nutrition. Although glycemic management of GDM can be achieved by nutritional therapy (3), only approximately half of the cases can be attributed to the metabolic effects of obesity (4, 5). Furthermore, the risk of both recurrent GDM and type 2 diabetes mellitus is exceedingly high and occurs independently of obesity (6).

Unlike various other metabolic stressors that stimulate β-cell expansion, the physiologic adaptation to pregnancy comprises both an interaction between the mother and the fetoplacental unit and a rapidly fluctuating hormonal milieu that induces stereotypic changes in systemic metabolism spanning gestation and postpartum (2). Whereas the proliferative phenotype and augmentation of β-cells seen throughout pregnancy are rapidly reversed postpartum, the adverse consequences of disturbed maternal glucose homeostasis impact both mother and developing offspring during pregnancy and in the long-term (1, 7). These unique features of pregnancy and GDM form the premise for additional studies into the genetic relationship between GDM and type 2 diabetes and their shared or distinct pathophysiologic mechanisms (8).

Among the diverse extracellular and intracellular mechanisms known to contribute to gestational β-cell adaptation, the lactogens are perhaps the most well studied (9). Most mammals, including rodents and primates, produce pituitary-derived prolactin and placental lactogens, which both signal through the prolactin receptor (PRLR) (9–12). We and others have established a critical role for PRLR signaling in rodent gestational β-cell proliferation (13), with the lack of PRLR expression in β-cells resulting in a failure of islets to proliferate during gestation (14). Conversely, β-cell–specific overexpression of placental lactogens increases pancreatic islet proliferation and mass (15). Taken together, these studies illustrate that increases in circulating lactogens during pregnancy are essential and sufficient to orchestrate key components of metabolic adaptations to pregnancy. Downstream targets of PRLR signaling that contribute to gestational proliferation include tryptophan hydroxylase (Tph) 1 (Tph1), the rate-limiting enzyme for islet serotonin synthesis (16), transcription factors such as Foxm1 (17) and MafB (14), osteoprotegerin (Tnfrsf11b), menin (Men1), as well as cyclins, cyclin-dependent kinases, cell cycle inhibitors, and Prlr itself (14, 16, 18–20). However, a global assessment of how PRLR signaling regulates β-cell transcription during gestation is lacking.

To address this question, in the current study we examined how loss of PRLR signaling within β-cells alters pancreatic islet gene expression during pregnancy. Our findings identify both known and novel PRLR signaling targets, as well as candidate upstream regulators of β-cell transcription. Additionally, we found that the transcriptomic program driven by PRLR signaling and its target genes is distinct from the compensatory response of islets to high-fat diet (HFD) feeding, indicating pregnancy specificity of key downstream mediators of this pathway.

Materials and Methods

Mouse strains, husbandry, breeding, and experimentation

Female transgenic β-cell–specific PRLR knockout (βPRLRKO) mice were used in the current study, as previously developed and described (14). For gestational studies, 8-week-old virgin females were mated with C57BL/6J males and vaginal plugs were scored as gestational day (GD) 0.5. Plugged females were single-housed for the duration of pregnancy. For nutritional studies, mice were maintained on standard chow (4.7 kcal percent fat, catalog no. 7917; Envigo, Indianapolis, IN), then given ad libitum access to standard chow or an HFD (58.0 kcal percent fat, catalog no. D12492; Research Diets, New Brunswick, NJ) for either 4 or 12 weeks. Mice were group-housed (five per cage) on a 12-hour light/12-hour dark cycle at 25 ± 1°C and constant humidity with free access to food and water except as noted. All procedures involving mice were approved and conducted in accordance with the University of Alabama at Birmingham Institutional Animal Care and Use Committee.

Glucose tolerance testing

Glucose tolerance tests (GTTs) were performed by IP injection of glucose at 1 g/kg (chow-fed and 12-week HFD mice) or 2 g/kg (4-week HFD mice) of body weight using a glucose stock solution of 20% (w/v) d-glucose (Sigma-Aldrich, St. Louis, MO) in 0.9% saline to 16-hour overnight fasted mice. Blood glucose was determined using a Bayer Contour Next glucometer.

Islet isolation and cell culture

Islets were isolated from donor mice using retrograde perfusion of the pancreatic duct with collagenase and purified using density centrifugation as previously described (14). Donor mice included βPRLRKO mice and littermate controls or C57BL/6J mice either from virgin females (nonpregnant) or at GD16.5 (pregnant). For in vitro experiments, hand-picked islets were cultured overnight in RPMI 1640 supplemented with 10% FBS. Prior to prolactin treatment, islets from individual mice were picked and divided equally into fresh media.

MIN6 and αTC6 cells were incubated in standard tissue culture conditions at 37°C with 5% CO₂ in growth media containing 10% fetal bovine serum; MIN6 cells were additionally supplemented with β-mercaptoethanol (1.75 µL/500 mL of media). Both cell culture lines and islets were treated with fresh growth media supplemented with recombinant mouse prolactin at a final concentration of 500 ng/mL (R&D Systems, Minneapolis, MN) or an equal volume of vehicle for 24 hours. Culture media were changed daily.

Microarray analysis

Microarray-based mRNA quantification was performed using the Affymetrix GeneChip mouse 2.0 array (Thermo Fisher, Waltham, MA). Hybridization and microarray scanning were performed by the Stanford Functional Genomics Facility (Stanford, CA). All microarray data have been deposited within the Gene Expression Omnibus repository (accession no. GSE118134). Subsequent microarray annotation, normalization, and differential expression analysis were completed in the R statistical computing environment (version 3.4.3). Briefly, raw microarray data across all samples were compiled and corrected using the robust multiarray average of background-adjusted and quantile-normalized microarray expression data using R package Oligo (version 1.42.0) as previously described (21). Gene annotations were then generated using the MoGene 2.0 transcript cluster (mogene20sttranscriptcluster.db). Once expression data were examined for RNA quality, the R package limma (version 3.34.9) was used to generate differential gene expression using generalized linear modeling (knockout vs wild-type) with a Benjamini–Hochberg (BM) P value adjustment (22). Owing to sample size limitations (n = 3), dispersion estimates were first determined via maximum likelihood, assuming that genes of similar average expression strength possess similar dispersion, as previously described (23). Gene-wise dispersion estimates were then shrunk according to the empirical Bayes approach, providing normalized count data for genes proportional to both the dispersion and sample size. Differential expression was represented as log₂ (fold change) for each annotated gene. Statistical significance was estimated at a BH-adjusted P value (Q value) <0.05.

Bioinformatics and data visualization

Details of the R coding scripts and other bioinformatics tools used in the current study are available in an online repository (24). Functional gene set enrichment analyses were performed with the interactive Web-based platform Enrichr (25) using the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database. Gene network analysis and visualization were performed using Cytoscape (version 3.7.0), with literature-curated pathway enrichment being achieved with Qiagen’s Ingenuity Pathway Analysis (IPA^®; Qiagen, Redwood City, CA) on differentially expressed genes (DEGs) using a low-stringency statistical threshold of P < 0.05. Heat map and hierarchical clustering generation was performed using the pheatmap package (version 1.0.8) within R, and VennPlex (25) was used to create the Venn diagrams and determine overlapping genes.

RNA isolation for microarray and quantitative RT-PCR studies

Total RNA was isolated using TRIzol and reverse transcribed using the Ambion RETROscript kit according to the manufacturers’ instructions. Differential expression by quantitative PCR (qPCR) was performed with FastStart Universal Probe Master (Roche LifeSciences, Indianapolis, IN) and Thermo Fisher TaqMan™ assays (26) on a Bio-Rad CFX96 Touch™ machine (Bio-Rad Laboratories, Hercules, CA). Assays were performed in technical replicates and normalized to mouse actin (Actb) as a reference standard.

Statistical analysis

To determine the significance in overlap between the Venn diagram of DEGs identified in our study and those of previous publications, we used a hypergeometric probability distribution with a Bonferroni correction to generate an adjusted P value. For all targets, an unpaired two-tailed Student t test was performed using a Tukey correction for multiple comparisons. Statistical significance was concluded based on a false discovery–adjusted P value (Q) <0.05. Functional and network gene set enrichment analyses, along with curated literature-supported candidate upstream regulators, were performed using Qiagen’s IPA unless otherwise specified. GTTs were considered significant for a repeated measures ANOVA P value <0.05.

Results

Differential gene expression in pancreatic islets from βPRLRKO mice during gestation

To identify genes regulated by PRLR signaling in maternal pancreatic islets during pregnancy, we examined global transcriptomic profiles at late gestation using commercial oligonucleotide microarrays (as described in Materials and Methods). RNA purified from pancreatic islets isolated from GD16.5 βPRLRKO mice (Prlr^f/f;RIP-Cre) and littermate controls (Prlr^f/f and Prlr^f/+) identified 2542 DEGs at P < 0.05; of these, 70 reached a Bonferroni-adjusted P value (Q) <0.05 (26). Unsupervised principal components analysis revealed that the two eigenvectors most responsible for sample variance (51.7%) were sufficient to describe a separation between βPRLRKO mice and controls, supporting that PRLR knockout confers a true and measurable shift in global gene expression (27). Gene set enrichment analysis identified tryptophan metabolism and prolactin signaling as enriched pathways [Fig. 1(a)], both of which are associated with the metabolic adaptation to pregnancy (13, 16, 32, 33). We then compared the DEGs reaching Q value significance with published studies reporting DEGs in murine islets during pregnancy in wild-type mice. We incorporated microarray data of islet gene expression during pregnancy compared with nonpregnant controls [Schraenen et al. (28), GD9.5; Layden et al. (29), GD12.5; and Rieck et al. (30), GD14.5] and generated a Venn diagram of these studies with the 70 DEGs found in βPRLRKO mice. This comparison revealed a significant overlap of 26 DEGs (hypergeometric P = 2.8 × 10⁻⁶¹) [Fig. 1(b) (28–30)]. Among these were genes responsible for enriching pathways in Fig. 1(a), including tryptophan metabolism (Tph1 and Tph2) and prolactin receptor signaling (Prlr), as well as diverse cellular physiologic processes [Table 1 (14, 16, 20, 30, 34–39)] (26).

Figure 1.

Identifying the prolactin-dependent transcriptional effects of pregnancy. (a) KEGG pathway enrichment using DEGs of isolated pancreatic islets from pregnant βPRLRKO mice relative to littermate controls (n = 3). (b) Comparison of genes differentially regulated in pregnant βPRLRKO mice to published studies of islets from pregnant females at various stages: Schraenen et al. (28), GD15.5; Layden et al. (29), GD13.5; Rieck et al. (30), GD14.5; and the current study (βPRLRKO). (c) Cross-comparison of βPRLR-dependent DEGs identified in the current study with those identified by Goyvaerts et al. (31) using a constitutive whole-body PRLR knockout mouse model (Q < 0.05). (d) qPCR validation of genes found to be induced by pregnancy and suppressed in βPRLRKO mice, reported as average relative fold change ± SEM. Significance was assumed using a Bonferroni-adjusted P value of <0.05. *P < 0.05.

Table 1.

Genes Induced by Pregnancy That Require β-cell PRLR Signaling

Gene	Description	Log2 (Fold Change)	Q Value	References Relevant to β-Cells or Islets
Cldn8	Claudin 8	−4.7	0.0009
Ivd	Isovaleryl coenzyme A dehydrogenase	−2.4	0.002	(34)
Tph1	Tryptophan hydroxylase 1	−5.3	0.002	(14, 16, 35)
Matn2	Matrilin 2	−3.1	0.002	(34)
Cish	Cytokine inducible SH2-containing protein	−2.4	0.004	(36)
Tph2	Tryptophan hydroxylase 2	−2.6	0.004	(16)
Socs2	Suppressor of cytokine signaling 2	−1.7	0.005	(37)
Igfbp5	IGF binding protein 5	−2.6	0.005	(30, 34, 38)
B3galnt1	UDP-GalNAc	−2.5	0.008
Slc2a13	Solute carrier family 2 (facilitated glucose transporter)	−1.4	0.008
Apobec1	Apolipoprotein B mRNA editing enzyme, catalytic polypeptide 1	−1.6	0.011
Znrf2	Zinc and ring finger 2	−1.3	0.013
Tnfrsf11b	TNF receptor superfamily	−2.2	0.019	(20)
Fmo1	Flavin containing monooxygenase 1	−1.8	0.021
Slc40a1	Solute carrier family 40 (iron-regulated transporter)	−1.5	0.021
Ehhadh	Enoyl–coenzyme A, hydratase/3-hydroxyacyl coenzyme A dehydrogenase	−1.8	0.023	(34)
Prlr	Prolactin receptor	−1.8	0.031	(14, 16, 39)
Car15	Carbonic anhydrase 15	−1.6	0.033
Enpp2	Ectonucleotide pyrophosphatase phosphodiesterase 2	−1.1	0.033
Igfals	IGF binding protein, acid labile subunit	−0.9	0.033
Gcdh	Glutaryl–coenzyme A dehydrogenase	−0.9	0.041
Wipi1	WD repeat domain, phosphoinositide interacting 1	−1.3	0.043
Aqp4	Aquaporin 4	−2.1	0.045
Chgb	Chromogranin B	−1.4	0.046	(34)
Slc6a8	Solute carrier family 6 (neurotransmitter transporter, creatine)	−1.0	0.047
Gnb1l	Guanine nucleotide binding protein (G protein), β polypeptide 1-like	−0.9	0.049

Thus, this approach identified 26 genes that are induced by pregnancy in a PRLR-dependent manner. We also mined microarray data of islet gene expression in global PRLR-knockout females [Prlr^−/−; created by Ormandy et al. (40)] examined by Goyvaerts et al. (31) at early gestation (GD9.5). Although the global Prlr^−/− and βPRLRKO models differ in several key respects (9, 14, 40, 41), this cross-model comparison showed a significant overlap (hypergeometric P = 2.9 × 10⁻⁵⁰) in differential gene expression (defined as Q < 0.05) [Fig. 1(c) (31)]. Validating this approach, the genes in Table 1 included well-established PRLR signaling target genes within β-cells: Prlr (13, 14, 41), Tph1 and Tph2 (16, 28), Cish (36), and Opg (20). To further validate our putative PRLR signaling target genes, we examined gene expression using qPCR on an independent set of wild-type GD16.5 and nonpregnant C57BL/6J female mice, which revealed significant changes in gestational expression within islets consistent with array data [Fig. 1(d)]. Altogether, these results indicate that the effects of PRLR signaling on β-cell gene activation occur independently of PRLR signaling in other tissues and are present at both early and late gestation.

Prolactin induces β-cell–specific expression of PRLR-dependent genes

Our analysis establishes that PRLR signaling is necessary for the induction of a subset of genes in islets during healthy gestation. However, PRLR signaling may itself interact with the hormonal milieu of pregnancy to regulate transcription. To determine whether PRLR signaling is sufficient to regulate genes in an islet-autonomous fashion, we treated cultured islets for 24 hours with recombinant prolactin. This treatment significantly upregulated many of the PRLR-dependent DEGs [Fig. 2(a)], including Tph1 and Tph2, as previously reported (28). Our findings therefore demonstrate that PRLR signaling is both sufficient and required to regulate a subset of DEGs normally induced by pregnancy.

Figure 2.

Prolactin induces β-cell–specific gene expression associated with gestational adaptation. (a) qPCR quantification of PRLR-dependent pregnancy genes in primary isolated pancreatic islets treated with/without 500 ng/mL recombinant mouse prolactin (+prl) for 24 h (n = 4). (b) Expression of DEGs expressed at significantly higher levels in MIN6 relative to αTC6 cells, and response of both cell lines to treatment with/without 500 ng/mL recombinant mouse prolactin for 24 h (n = 3). (c) Expression of genes comparably expressed in MIN6 and αTC6 cells at baseline and response to treatment with/without 500 ng/mL recombinant mouse prolactin for 24 h (n = 3). All values are reported as average relative fold change ± SEM. Significance was assumed using a Student t test at P < 0.05. *P < 0.05.

Although we and others have established that PRLR is expressed specifically in β-cells of rodent islets (14, 42), the transcriptional response to loss of prolactin signaling in β-cells could reflect secondary or compensatory changes induced within other islet cell types. Therefore, to examine whether PRLR target genes were specifically expressed within β-cells and responsive to prolactin in a cell-autonomous manner, we tested the baseline and prolactin-stimulated expression of a subset of DEGs in α-cells and β-cells using MIN6 and αTC6 cell lines, respectively. Most PRLR-dependent DEGs examined were expressed at significantly higher levels in MIN6 than αTC6 cells at baseline [Fig. 2(b)]. Treatment with recombinant prolactin robustly induced DEG expression in MIN6 cells, whereas αTC6 cells exhibited either an attenuated or absent response to prolactin treatment [Fig. 2(b)]. A minority of DEGs were expressed at comparable levels in unstimulated MIN6 and αTC6 cells, but they were induced only in MIN6 cells following the 24-hour prolactin treatment [Fig. 2(c)]. In sum, all of the PRLR DEGs examined were stimulated by prolactin treatment in MIN6 cells. Thus, these data indicate that a large majority of PRLR DEGs are disproportionately expressed within β-cells, and that lactogen-mediated activation of PRLR signaling is sufficient to robustly induce their expression in a β-cell–specific manner.

Prolactin and high-fat feeding stimulate divergent transcriptional cascades

Metabolic stressors, including nutrition and pregnancy, have been theorized to converge on the β-cell, leading to its dysfunction and ultimately diabetes mellitus; however, empirical evidence supporting this mechanistic convergence remains inconclusive (1, 43). Because pregnancy initiates an adaptive metabolic response via profound increases in lactogenic and other hormones, we hypothesized that PRLR-mediated adaptations activate a pregnancy-specific adaptive mechanism in pancreatic islets. To test this, we first compared our βPRLRKO data set to published RNA-sequencing data of gene expression changes following an HFD, a widely used model system of metabolic stress (44) [Fig. 3(a)]. Surprisingly, the overlap of DEGs was minimal and not significant (hypergeometric P = 0.4), comprising only four genes (Matn2, 2410021H03Rik, Igfbp5, and Cntn3) that were differentially coexpressed by both βPRLRKO and following 30-day (∼4-week) HFD feeding. To determine whether the PRLR is required for β-cell physiologic adaptation to HFD feeding, we performed a 4-week HFD study using female βPRLRKO and littermate control mice. This approach also allowed us to define whether HFD feeding displayed sexually dimorphic effects on the 70 DEGs reaching Q value significance in βPRLRKOs. Unlike the requirement for intact PRLR signaling in maintaining gestational glucose homeostasis, both βPRLRKO female mice and littermate controls developed comparable degrees of weight gain [Fig. 3(b)] and glucose tolerance following GTTs [Fig. 3(c) and 3(d)] after just 4 weeks of HFD feeding. Prolonged HFD feeding to 12 weeks to promote obesity in addition to glucose intolerance displayed the expected further increases in weight and glucose intolerance, yet it failed to reveal any difference between βPRLRKO and littermate control mice (27). Lastly, gene expression analysis via qPCR demonstrated insignificant effects of an HFD on differential gene expression [Fig. 3(e)]. Thus, for both sexes, HFD feeding failed to reproduce the induction of PRLR-dependent DEGs. These results support our hypothesis that both transcriptional regulation and physiologic adaptation of β-cells by PRLR signaling employs pregnancy-specific mechanisms that differ from those of HFD adaptation.

Figure 3.

Pregnancy and HFD feeding activate divergent transcriptional programs. (a) Venn diagram illustrating the degree of overlap between βPRLR-dependent DEGs and those induced by an HFD based on an RNA sequencing data set generated by Cruciani-Guglielmacci et al. (44). (b) Body weight (grams) of βPRLRKO mice and littermate controls (CON) at baseline (t = 0) and following 4 wk of standard chow or 60% HFD feeding (n = 6). (c) IP glucose tolerance testing and (d) area under the curve (AUC) of HFD and CON mice. (e) qPCR quantification of PRLR-dependent pregnancy genes in primary isolated pancreatic islets from chow- or HFD-fed mice (n = 4), reported as average relative fold change ± SEM. Significance was assumed using a Student t test at P < 0.05. *P < 0.05; **P < 0.01; ****P < 0.0001. n.s., not significant.

Pathway analysis of differential gene expression

To identify potential mechanisms underlying pregnancy-specific adaptations in islets, we next sought to understand the global transcriptional landscape that is regulated by PRLR signaling in pregnancy. Because the false discovery rate (FDR)–adjusted threshold yielded only 70 DEGs, we considered whether reducing the statistical stringency of our analysis (P < 0.05) would capture nodal networks and regulatory clusters. Hierarchical clustering and heat map visualization justified this approach, displaying a distinct expression profile even at this cutoff [Fig. 4(a)]. KEGG pathway enrichment of the 2540 genes (P < 0.05) revealed a clear signature of cellular proliferation among all of the top five most enriched networks: G₂/M DNA damage checkpoint (P = 1.7 × 10⁻⁵), GADD45 signaling (P = 5.6 × 10⁻⁵), regulation of cellular mechanics (P = 2.3 × 10⁻⁴), estrogen-mediated S-phase entry (P = 2.8 × 10⁻⁴), and mitotic roles of polo-like kinase (P = 4.9 × 10⁻⁴). Volcano plot revealed a negatively skewed signature of transcriptional changes by βPRLRKO relative to littermate controls [Fig. 4(b)], consistent with known upregulation of β-cell genes during pregnancy (29, 30).

Figure 4.

Gene expression analysis of βPRLRKO identifies proliferative impairment. (a) Hierarchical clustering of normalized beta values via Wald.D2 test and dendrogram constructed by Euclidean distance with heat map visualization of DEGs. (b) Volcano plot illustrating the distribution of genes based on −log₁₀ (P value) as a function of log₁₀ (fold change), labeling the most differentially regulated genes (|fold change| > 2, FDR < 0.05). (c) Protein–protein interaction network generated from DEGs based on the STRING database. (d) Gene set enrichment of the gene module with highest degree of protein interactivity using KEGG pathways of DEGs. A significance threshold of P < 0.05 was used for gene expression, followed by BH-adjusted P (“Q”) <0.05 for subsequent gene set enrichment analysis.

Our gene set enrichment analysis identified established pathways differentially affected by loss of PRLR signaling. However, to identify novel networks affected in βPRLRKO, we used the STRING database (45) to generate a network of DEGs based on known and predicted interactions among their encoded proteins [Fig. 4(c)]. After ranking nodes by degree of interaction, the largest cluster was identified to functionally enrich pathways associated with cellular proliferation and developmental programs [Fig. 4(d)]: cell cycle (BH-adjusted P = 10⁻⁹), progesterone-mediated oocyte maturation (BH-adjusted P = 10^−6.2), p53 signaling (BH-adjusted P = 10^−5.3), FoxO signaling (BH-adjusted P = 10^−5.2), and oocyte meiosis (BH-adjusted P = 10^−4.3). In contrast, when protein–protein interaction analysis was performed on the data set from mice fed an HFD (44), the network instead enriched pathways mediating fatty acid metabolism and signaling, with peroxisome proliferator–activated receptors (PPARs) as the most enriched transcriptional regulators (27). These data demonstrate that βPRLRKO impairs a pregnancy-specific program of β-cell proliferation during pregnancy.

EZH2/PRC2 and FOXM1 are PRLR-dependent candidate transcriptional regulators of β-cell adaptation in pregnancy

To identify upstream regulators capable of driving this proliferative signature, we used the IPA causal network analysis to leverage the directionality of gene changes when performing gene set enrichment (46). IPA upstream functional analysis disproportionately enriched genes known to interact with PRLR itself, both directly and indirectly [Fig. 5(a)]. Other upstream regulators included growth and DNA damage inducible 45γ (Gadd45g, 2.1-fold suppressed) and DNA methyltransferase 3B (Dnmt3b, 2.2-fold suppressed), both regulators of DNA methylation turnover (47, 48). Examination of the genes reported to interact with PRLR revealed increased expression of upstream factors and decreased expression of PRLR target genes [Fig. 5(b)]. Of the PRLR target genes, we noted that the histone methyltransferase Ezh2 was suppressed in βPRLRKO.

Figure 5.

EZH2 and Foxm1 as likely upstream regulators of PRLR-dependent gene expression. (a) PRLR identified as top differentially expressed upstream regulator via the Ingenuity database (IPA). (b) Regulatory network of DEGs associated with PRLR (blue indicates downregulated; yellow indicates upregulated). (c) Upstream DNA-binding protein enrichment of upregulated DEGs and (d) downregulated DEGs in βPRLRKO relative to wild-type was performed using published ChIP sequencing data sets based on the Encyclopedia of DNA Elements (ENCODE). (e) Upregulated DEGs were enriched for histone modifications using the ENCODE database. (f) Circular genome plot of differentially regulated direct downstream targets of EZH2. (g) Coregulatory network of downstream DEGs of EZH2-FOXM1 regulation. *P < 0.05 for pathway and network enrichment.

As a transmembrane receptor resident at the cell surface, PRLR indirectly regulates transcription through signal transduction cascades, including the canonical JAK2–STAT5 pathway (49). However, as mice with a β-cell–specific deletion of Stat5 do not exhibit defective glucose homeostasis or GDM during pregnancy (50), we hypothesized that multiple transcription factors coordinately regulate transcription for PRLR signaling-mediated gestational adaptations. To identify candidates, we inspected the proximal promoter region (−1 kb to +500 kb) of differentially regulated genes (P < 0.05) for disproportionate enrichment of transcription factor binding sites. Using the ENCODE chromatin immunoprecipitation (ChIP) sequencing database (51), we identified promoter binding sites for Foxm1 and E2f4 among the top downregulated DEGs [Fig. 5(c)], a finding consistent with prior studies (14, 17). Stat5a was also statistically enriched (FDR < 0.05), although not among the 10 most enriched putative regulators. Conversely, promoter enrichment of genes expressed at higher levels in βPRLRKO identified promoter binding sites for Suz12 (BH-adjusted P = 10^−7.8) and Ezh2 (BH-adjusted P = 10^−4.4), two components of the polycomb repressor complex (PRC2) [Fig. 5(d)]. Colocalization of Ezh2 and Suz12 promoter binding sites DEGs was also noted, demonstrating a 90% (28 out of 31) overlap. To determine whether PRC2 methyltransferase activity likely controls gene expression in βPRLRKO, we then enriched DEG promoters using ChIP-sequencing analysis of chromatin modifications using the human histone modification database (52). Histone modification enrichment identified H3K27 trimethylation (H3K27me3), the modification enzymatically mediated by EZH2/PRC2, as the only histone modification localized to upregulated DEG promoters [Fig. 5(e)]. Reciprocally, the promoter regions of downregulated DEGs contained sites of H3K27 acetylation, the counterregulatory modification of H3K27me3. Consistent with other research (53, 54), these observations support that the PRC2-mediated gene silencing that occurs during pregnancy is de-repressed in the setting of βPRLRKO.

Ezh2 has been classically identified as a transcriptional repressor via histone methyltransferase activity in the PRC2 (55). However, we and others have identified a noncanonical role for Ezh2 as a positive regulator of gene expression upon heterodimerization with Foxm1 and other transcriptional regulators (56–58). Therefore, to understand whether this dual role of Ezh2 might be simultaneously present in pancreatic islets, we plotted the 75 DEGs (P < 0.05) that are targets of Ezh2 according to the Ingenuity^® database [Fig. 5(f)]. The resulting distribution depicted a broad array of target genes both increased and decreased by βPRLRKO. Additionally, genes cotargeted by Ezh2 and Foxm1 feature key regulators of cellular proliferation [Fig. 5(g)]: Top2a (log₂ fold change = −1.4, P = 0.001), Mki67 (log₂ fold change = −1.4, P = 0.001), Ccne2 (log₂ fold change = −0.30, P = 0.04), Myc (log₂ fold change = −0.81, P = 0.0007), Esr1 (log₂ fold change = −0.60, P = 0.005), Cdk1 (log₂ fold change = −0.6, P = 0.001), and Cdkn1a (log₂ fold change = 0.87, P = 0.0096). In sum, our analyses support a role of Foxm1 and Ezh2 as coregulators of the proliferative signature within β-cells during pregnancy.

Discussion

Prior studies have identified transcriptional changes within pancreatic islets during pregnancy, a period of rapid β-cell proliferation and mass expansion (28–30). Many hormonal and metabolic changes accompany gestation, with lactogens serving a critical role in physiologic adaptation of β-cells (14, 15). Although the mitogenic effects of lactogen signaling have been widely studied, the specific gene targets and transcriptional networks under lactogen control remain poorly understood. Using a βPRLRKO mouse model, we confirm the requirement of intact β-cell PRLR signaling for induction of both well-established and novel genes and gene networks during pregnancy.

We identified and cross-validated a number of known and novel pro-proliferative genes as induced by pregnancy in a PRLR-dependent manner (Table 1). Our findings are consistent with a recent proteomic analysis of islets during pregnancy, which independently found proteins encoded by our DEGs to be differentially regulated in late gestation: Ivd, Matn2, Ehhadh, Myc, Igfbp5, and Chgb (34). Among these, we confirmed the requirement of intact β-cell PRLR signaling for induction of well-established gestational proproliferative genes (e.g., Tph1, Tph2, Prlr, Opg). In addition to previously identified genes of pregnancy, we identified novel gene targets of PRLR signaling that provide exciting avenues for future study into the mechanisms of islet gestational adaptation. For instance, pregnancy increases gap junction coupling between β-cells (11), which has been theorized to regulate insulin secretion and the alterations in insulin secretion during pregnancy (59). Claudin-8 (Cldn8), here identified as induced by PRLR in pregnancy, is an integral membrane protein and tight junction component that has not been studied in β-cells in any context.

Similarly, we identified Igfbp5 as a novel PRLR signaling target in β-cells. Previous studies found Igfbp5 in FACS-sorted β-cells during pregnancy (30), as well as a target of Akt1 within β-cells (38). Furthermore, IGFBP5 protein levels increase within islets during pregnancy (34) and are linked to glucose metabolism and diabetes risk (60). Therefore, our findings connecting PRLR signaling and Igfbp5 gestational regulation merit further investigation.

In our search for regulatory networks impacted by PRLR disruption in pregnancy, we identified Foxm1 and Ezh2 as putative nodal upstream coregulators of gestational gene expression and cellular proliferation (Fig. 6). Within the pancreatic islet, Ezh2 has previously been shown to silence the tumor suppressor Ink4a/Arf (p16) in aging and diabetes to increase β-cell proliferation (61, 62). Conversely, selective β-cell Ezh2 knockout in mice has been shown to produce hyperglycemia in mice (61). In the current study, we found that pregnancy induces Ezh2 in a PRLR-dependent manner. Therefore, gestational induction of Ezh2 likely creates a proproliferative state required for metabolic adaptation. Our results thus broaden our understanding of Ezh2 as a regulator of β-cell function.

Figure 6.

Proposed model of pregnancy-specific transcriptional reprogramming. Placental and pituitary lactogens activate PRLR signaling within β-cells during pregnancy. Downstream factors, including Ezh2 and Foxm1, regulate transcription of target genes. In turn, PRLR target genes promote gestational proliferation, including induction of Tph1, responsible for increased islet serotonin, which promotes β-cell proliferation through autocrine-paracrine effects. In contrast, HD feeding induces transcriptional reprogramming via PPARs independent of prolactin receptor and serotonin signaling.

Although canonically associated with gene silencing via H3K27me3, Ezh2 has also been shown to heterodimerize with transcription factors, such as Foxm1, to activate gene expression in a methylation-independent mechanism (58, 63). In breast and prostate tissue, Ezh2 is induced by PRLR activation via STAT5, and it mediates the mitogenic effects of prolactin in these tissues (64, 65). Our results therefore support a broader role for Ezh2 in the regulation of β-cell proliferation during pregnancy, analogous to its role in other tissues (58).

We and others have identified the transcription factors MafB, FoxM1, and Stat5 as downstream of PRLR signaling and critical for gestational β-cell proliferation (9, 14, 17). How these proteins coordinately regulate individual PRLR target genes is poorly understood. For example, Tph1, whose induction by lactogens is well established, is transcriptionally activated via convergence of Stat5, PI3K, and Erk signaling (66). However, although genetic analysis demonstrates a requirement for MafB in Tph1 induction (14), precisely how MafB interacts with the Tph1 promoter, or with Stat5, is unknown. Nevertheless, gestational increases in serotonin following Tph1 induction promote β-cell proliferation through autocrine-paracrine effects mediated by Htr2b (16, 39) (Fig. 6). Thus, further studies are needed to assess how these PRLR-responsive transcription factors, along with Ezh2, coordinate gene expression of Tph1 and other PRLR target gene loci following lactogen stimulation.

As GDM and type 2 diabetes share many similarities and risk factors, it has been widely assumed that they share underlying pathophysiologic mechanisms. However, our data demonstrate that the reprogramming of pancreatic islets during pregnancy is distinct from that of nutritional stressors. We found that the robust PRLR-dependent transcriptional changes in gestation were essentially nonoverlapping with those induced by HFD feeding. Coexpression network analysis further identified distinct functional enrichment in pregnancy associated with cellular proliferation. In contrast, HFD feeding enriched fatty acid metabolism and PPAR signaling, consistent with known roles for these pathways in dietary adaptation in β-cells (67–69) (Fig. 6). Our analysis therefore provides several avenues for investigation whereby disease specificity may be realized. We previously found Foxm1 and Ezh2 coregulated genes involved in structural remodeling and metabolic reprogramming in humans with ischemic heart failure, suggesting this regulatory complex may represent a broader transcriptional regulatory mechanism beyond gestational adaptation within islet β-cells (58). In this study, we identify many PRLR-dependent epigenetic modifiers that regulate chromatin remodeling (e.g., Ezh2 and Suz12) and/or differential DNA methylation (e.g., Gadd45g and Dnmt3b). Therefore, the pregnancy-specific response to PRLR signaling likely requires many genetic and epigenetic regulators to coordinate β-cell adaptation. Nevertheless, future studies are needed to mechanistically define these programs.

Although our findings underscore the importance of PRLR signaling as a mediator of β-cell function in pregnancy, we recognize several limitations to our studies. We and others have established that PRLR is expressed specifically in β-cells of rodent islets (14, 42). However, we cannot exclude the possibility of non–β-cell effects of βPRLRKO. Therefore, future studies should employ single-cell or cell-sorted approaches to analyze gestational islet transcription. The computational approach we used to define transcriptional regulators of PRLR signaling provide preliminary support for their role in gestational adaptation; however, future mechanistic studies are needed to determine the nature of their influence. Additionally, although gestational induction of DEGs in our data is likely to be directly downstream of PRLR signaling, we recognize that parallel compensatory signaling mechanisms exist that may contribute to the differential regulation of gene expression. Consistent with this possibility, we observed a subtle increase in GHR expression on microarray analysis (1.3-fold, P = 0.03), a known inducer of the prolactin receptor (70).

In conclusion, the current study examines novel pregnancy-specific gene networks and PRLR-dependent transcriptional programs that describe a state of augmented cellular proliferation. Despite the stated limitations, we think that the computational approach, combined with both in vivo and in vitro experimentation, serves to significantly advance our understanding of how β-cells respond to the metabolic demands of pregnancy. These lactogen-mediated regulatory networks likely reflect broader adaptive mechanisms that apply to other contexts. We propose PRLR-dependent transcriptional targets and upstream regulators as exciting new candidates to examine in gestational diabetes pathogenesis.

Glossary

Abbreviations:

BH

Benjamini–Hochberg

ChIP

chromatin immunoprecipitation

DEG

differentially expressed gene

Ezh2

enhancer of zeste homolog 2

FDR

false discovery rate

GD

gestational day

GDM

gestational diabetes mellitus

GTT

glucose tolerance test

HFD

high-fat diet

H3K27me3

H3K27 trimethylation

IPA

Ingenuity Pathway Analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes

PPAR

peroxisome proliferator–activated receptor

PRLR

prolactin receptor

qPCR

quantitative PCR

Tph

tryptophan hydroxylase

βPRLRKO

β-cell–specific PRLR knockout