High-Fat Diet During Mouse Pregnancy and Lactation Targets GIP-Regulated Metabolic Pathways in Adult Male Offspring

Gastric inhibitory polypeptide (GIP) is released from the duodenum after nutrient intake and regulates postprandial insulin secretion (1). GIP is known for its anabolic effects in adipocytes (2) and has been shown to stimulate the activity of lipoprotein lipase in adipose tissue in vitro (3) (4). Genetic ablation of the GIP receptor (GIPR) protects from high-fat diet (HFD)–induced obesity and insulin resistance in mice (5). GIPR knockout (Gipr^−/−) mice use fat as energy substrate rather than storing it in adipocytes. Moreover, central appetite-regulating pathways are downregulated in ovariectomized female Gipr^−/− mice (6), and energy expenditure is increased in Gipr^−/− mice exposed to a high–glycemic index diet (7) or an HFD (8).

Since Gipr^−/− mice are protected from obesity when exposed to a postweaning HFD, we hypothesized that this phenotype may be lost when these mice were exposed to the same HFD during pregnancy (intrauterine [IU]) and lactation (L). The IU milieu has a persisting influence on the development of adult diseases (9,10). Maternal caloric malnutrition as well as overnutrition during IU and L can induce metabolic disturbances later in life in offspring (11,12). Animal studies have shown that mice exposed to a maternal HFD during IU/L and weaned onto a standard chow developed adiposity, insulin resistance, and hepatosteatosis later in life (13,14). Not only the diet during IU/L but also the postweaning diet is important for the manifestation of metabolic diseases. Studies investigating the interaction of an HFD during IU/L and postweaning usually apply the HFD continuously from premating throughout the end of the study in adulthood. However, in terms of programming effects of metabolic diseases, studies investigating a “second hit” later in life (e.g., a HFD) after a time frame of regular diet might better reflect the adult onset of obesity, as shown recently (15). Recent work suggests that the IU milieu induces epigenetic alterations, such as DNA methylation and histone modifications, which affect gene transcription that persists throughout life (16). Pathways targeted by epigenetic programming are of eminent importance to develop strategies to reduce the metabolic risks faced by offspring of obese mothers. We therefore chose a model in which extensive research has already identified some pathways involved in the protection from obesity, the Gipr^−/− mouse.

Our results demonstrate that Gipr^−/− mice exposed to an HFD during IU/L, weaned onto a regular chow for 22 weeks, and then reexposed to the HFD lose their metabolic protection, which is associated with hypermethylation of certain CpG dinucleotides in the promoter regions of genes involved in energy balance and with changes in central insulin sensitivity. This phenotype was similar to wild-type (WT) mice that were exposed to a control chow during IU/L but also challenged with an HFD later in life. We therefore conclude that the protection from diet-induced obesity in Gipr^−/− mice is “overruled” by deleterious and persistent fetal programming effects of an HFD up to at least 45 weeks of life and identify metabolic pathways addressed by GIP.

Animals were maintained on a 12-h light-dark cycle with ad libitum access to chow and water. Age- and body weight–matched female mice heterozygote for the GIPR (Gipr^+/−) on a C57BL/6 background were maintained on either a control (C) diet D12450B (10% kcal fat as soybean oil and lard, 20% kcal protein, 70% kcal carbohydrate, 3.85 kcal/g) or an HFD D12492 (60% kcal fat as soybean oil and lard, 20% kcal protein, 20% kcal carbohydrate, 5.24 kcal/g) 2 weeks prior to mating with a male Gipr^+/− mouse and throughout IU and L (all diets from Research Diets, Inc., New Brunswick, NJ). Offspring were genotyped at 2–3 weeks of life. Male Gipr^−/− offspring on C diet or HFD during IU/L and WT offspring on C diet during IU/L were individually housed and placed onto C diet at postnatal day 21 and kept on this diet until the age of 25 weeks, followed by a further 20 weeks on HFD. This resulted in Gipr^−/− mice exposed to C diet during IU/L and to HFD at the age of 25 weeks (KO C-C-H), Gipr^−/− mice exposed to HFD during IU/L and to HFD at the age of 25 weeks (KO H-C-H), and WT mice exposed to C diet during IU/L and to HFD at the age of 25 weeks (WT C-C-H). Results obtained for these groups during the first 25 weeks of life are indicated with the first two diet symbols (KO C-C, KO H-C, and WT C-C). At the age of 45 weeks, mice were killed using cervical dislocation. Tissues were harvested, immediately frozen in liquid nitrogen, and stored at −80°C. Figure 1 illustrates the study design. Animal protocols were approved by the local governmental animal ethic review board (State of Brandenburg).

Fasting blood samples were collected after an overnight fast at 45 weeks of age from the submandibular vein plexus. Serum glucose, triglycerides, total cholesterol, and nonesterified fatty acids (NEFAs) were measured using commercial kits (glucose HK 125, ABX Diagnostics, Montpellier, France; triglycerides and total cholesterol, ABX Pentra, Montpellier, France; and NEFA HR, Wako Chemicals, Neuss, Germany) by using an autoanalyzer (Cobas Mira S, Hoffmann-La Roche, Basel, Switzerland). Plasma insulin levels were determined as previously described (17).

Body weight and cumulative food intake were measured weekly. Food consumption was expressed in kilocalories per gram of body weight for the indicated age.

A glucose tolerance test (GTT) was performed at the age of 43 weeks. Overnight-fasted mice received an injection of D-(+)-glucose (2.0 g/kg body wt i.p.) (18). Blood was taken from the submandibular vein plexus and added to heparinized collecting tubes for plasma glucose and insulin measurements at 0, 10, 30, 60, and 120 min after glucose injection.

Body composition was determined by nuclear MRS (minispect MQ 10 NMR Analyzer; Bruker, Karlsruhe, Germany) in conscious mice (6) at the age of 13, 22, 31, and 42 weeks of life.

mRNA expression was performed in epididymal white adipose tissue, gastrocnemius muscle, and hypothalamus as previously described (6,19). The hypothalamus was dissected as a whole without any further subfractionation. Primers were designed using Primer Express software (Applied Biosystems). Primer sequences are listed in Supplementary Table 1. The constitutive gene HPRT was used as a housekeeping gene.

The hypothalamic insulin signaling pathway was analyzed by performing the PathScan Intracellular Signaling Array Kit (Cell Signaling Technology, Danvers, MA) that allows a quantitative analysis of the phosphorylation status of key proteins involved in insulin and energy signaling. Total protein lysate (75 µL) was processed following the manufacturer’s instructions. Signal density was visualized using LI-COR Biosciences Odyssey (LI-COR Bioscience, Bad Homburg, Germany).

Tissue was fixed in 10% buffered formalin, embedded in paraffin, sectioned at 5 µm, and stained with hematoxylin-eosin. Adipocyte size was measured using CellProfiler 2.0 (www.cellprofiler.org). One visual field with at least 40 cells of three samples of epididymal adipose tissue from one animal was analyzed for cell surface size and expressed in micrometers squared. The average adipocyte cell size for each group was calculated, and the frequency of cell sizes sorted by categories was determined for each group.

For evaluation of crown-like structures, 2-µm sections were dewaxed using 3% hydrogen peroxide and incubated overnight (4°C) with an F4/80 anti-mouse antibody (1:8,000; Serotec, Puchheim, Germany), followed by an anti-rat secondary antibody for mouse tissue that includes Histofine (Nichirei Biosciences Inc., Tokyo, Japan). Images were acquired by Mirax slide scanner (Carl Zeiss, Göttingen, Germany). Crown-like structures were determined in three randomly chosen areas within the slides.

Genomic DNA was extracted from muscle and hypothalamus using a commercial kit (Machery-Nagel, Dueren, Germany) following the manufacturer’s protocol, and quality was checked using a NanoDrop ND-1000 Spectrophotometer (PEQLAB, Erlangen, Germany).

Genomic DNA (700–1,000 ng) was bisulfite treated using the EZ-96 DNA Methylation Kit (Zymo Research, Orange, CA) according to the manufacturer’s protocol.

Methylation was determined for five loci by MALDI-TOF (matrix-assisted laser desorption ionization time-of-flight) mass spectrometry using EpiTYPER by MassARRAY (Sequenom, San Diego, CA) as previously described (20,21). Briefly, the target regions were amplified from bisulfite-treated DNA in duplicate PCR reactions. Primers were designed using the Sequenom EpiDesigner Web resource (http://www.epidesigner.com). Primer sequences for all amplicons are listed in Supplementary Table 2. Methylation data were generated by the MassARRAY EpiTYPER v1.2 software (Sequenom).

Four samples were measured in duplicate for four amplicons. Quality control of the data was performed using EpiTyper and the statistical software R (version 3.0.2), eliminating all samples and all CpG sites with a call rate <80%, respectively. In total, 182 CpG dinucleotides were measured (single sites and composite sites with two or more adjacent CpG sites falling within one analyzed fragment). CpG dinucleotides relevant for this study are named regarding their distance to the transcription start site.

Myoblast nuclear extracts from mouse C2C12 cells (no. 36078; Active Motif) were used for electrophoretic mobility shift assay (EMSA). Cy5-labeled and unlabeled oligonucleotides (Metabion) containing the methylated or unmethylated CpG from respective EpiTYPER experiments (Supplementary Table 3) were annealed and purified in a 12% polyacrylamide gel. Binding reaction was carried out with or without different concentrations of unlabeled competitor oligonucleotides. In each reaction, 6.5 µg nuclear extract was incubated in 5× binding buffer (4% vol/vol glycerol, 1 mmol/L MgCl₂, 0.5 mmol/L EDTA, 0.5 mmol/L dithiothreitol, 50 mmol/L NaCl, and 10 mmol/L Tris HCl, pH 7.5) with 0.5 µg Poly[d(I-C)] (Roche Diagnostics) and 1 ng labeled probe in a total volume of 10 µL for 25 min at 4°C. Protein-DNA complexes were separated on a 5.3% polyacrylamide gel by electrophoresis in 0.5× trisborate EDTA running buffer. The gels were visualized by scanning with the Trio Typhoon plus Typhoon imager (GE Healthcare).

Data are represented as means ± SEM. Statistical analyses were performed using one-way ANOVA with Tukey post hoc test (version 20; SPSS, Chicago, IL). Data were tested for normal distribution. Non–normally distributed data were transformed to the logarithmic form to approximate the normal distribution.

Statistical analysis on methylation data was performed using R (version 3.0.2). For DNA methylation analysis, inferential statistics was used to analyze the 182 CpG sites of the three groups. For inferential statistics, Student t test was performed and resulting P values were adjusted for multiple testing. For descriptive analysis, quotients of mean values from each two groups were calculated. For verification we also performed a correlation analysis of methylation data and gene expression data.

During the first 25 weeks of life, body weight was not different except for weeks 4, 13, and 14, when KO H-C was heavier than KO C-C (Fig. 2A). When an HFD was fed at week 25 of life, KO C-C-H showed a significantly lower increase in body weight than WT C-C-H (Fig. 2B). Body weight of KO H-C-H was in between the range of WT C-C-H and KO C-C-H. Before exposure to the HFD, adiposity of KO C-C was significantly lower compared with WT C-C (Fig. 2C). After 12 weeks on the HFD, KO C-C-H showed a significant decrease in adiposity compared with WT C-C-H, but now KO H-C-H had significantly increased adiposity compared with KO C-C-H, similar to WT C-C-H.

When offspring were on a regular diet, there was no difference in cumulative food intake for all groups at the age of 25 weeks (Fig. 2D). When mice were fed an HFD, KO C-C-H consumed significantly more kilocalories per grams of body weight than WT C-C-H. In KO H-C-H, there was a significant decrease in cumulative energy consumption compared with KO C-C-H, which was similar to the level of WT C-C-H. Despite increased food intake, KO C-C-H had significantly reduced serum levels for total cholesterol compared with WT C-C-H (3.33 ± 0.33 vs. 5.22 ± 0.26 mmol/L, respectively; P < 0.005) (Table 1). In KO H-C-H, total cholesterol levels were 4.44 ± 0.40 mmol/L, but this level was not significantly different compared with KO C-C-H. Serum levels for triglycerides and NEFA were unchanged.

The intraperitoneal GTT after 18 weeks on HFD showed that KO C-C-H had significantly lower fasting blood glucose levels compared with WT C-C-H (Fig. 3A and Table 1). However, KO H-C-H showed significantly increased fasting blood glucose levels compared with KO C-C-H (P < 0.05), reaching levels similar to those of WT C-C-H. Accordingly, KO C-C-H had significantly lower fasting insulin levels compared with WT C-C-H (P < 0.05) (Table 1). KO H-C-H showed severe glucose intolerance similar to WT C-C-H (Fig. 3A). KO C-C-H, however, had significantly lower glucose levels after 30 min (P < 0.05 and P < 0.005 compared with KO H-C-H and WT C-C-H, respectively) and after 90 min (P < 0.05 compared with KO H-C-H and WT C-C-H). Additionally, the area under the curve for glucose (Fig. 3B) was lower in KO C-C-H compared with WT C-C-H (P < 0.05) but increased again in KO H-C-H (P < 0.05 compared with KO C-C-H, and P = not significant compared with WT C-C-H). During the intraperitoneal GTT, Gipr^−/− mice showed slightly lower insulin levels compared with WT, with no significant difference between KO C-C-H and KO H-C-H (Fig. 3C and D).

KO C-C-H had significantly smaller adipocytes compared with WT C-C-H (P < 0.001) (Fig. 4A, C, and D). HFD during IU/L caused an increase in adipocyte size in KO H-C-H (P < 0.001 compared with KO C-C-H) (Fig. 4A, D, and E). Figure 4B shows that KO C-C-H had more small adipocytes (<2,000 µm²) compared with KO H-C-H and WT C-C-H, which both had more cell counts in categories with large adipocytes (≥8,000 µm²).

Ccl2 mRNA expression was 48% decreased in KO C-C-H compared with WT C-C-H mice (P < 0.05) (Fig. 5A). However, in KO H-C-H, Ccl2 mRNA expression was increased 2.73-fold compared with KO C-C-H (P < 0.05). Ccl3 mRNA levels were significantly downregulated by 43% in KO C-C-H (P < 0.05 compared with WT C-C-H) and again 1.39-fold upregulated in KO H-C-H compared with KO C-C-H; however, the difference was not significant. The same pattern was observed for mRNA expression of Emr1: it was decreased by 44% in KO C-C-H compared with WT C-C-H (P < 0.05) and 1.35-fold upregulated in KO H-C-H (P = not significant compared with KO H-C-H). However, we could not detect differences in crown-like stuctures (Supplementary Fig. 1). There was no difference in gene expression for Tnf and Il6.

A strong increase in gene expression levels of Ppara (2.45-fold, P < 0.01), Ppargc1a (1.78-fold, P < 0.01), Cpt1a (1.39-fold, P < 0.05), and Cpt1b (1.53-fold, P < 0.01) was observed in KO C-C-H compared with WT C-C-H (Fig. 5B). This increase was abolished when Gipr^−/− mice were exposed to an HFD during IU/L (KO H-C-H). These mice had the same gene expression levels as WT C-C-H. We did not observe any significant difference in gene expression for Cd36, Lpl, Acadm, and Acadl.

Hypothalamic gene expression of phosphatidylinositol 3-kinase (PI3K)-p85α (Pik3r1) was 22% downregulated in KO C-C-H compared with WT C-C-H (P < 0.01) (Fig. 6A) but was, however, again upregulated in KO H-C-H similar to WT C-C-H. Phosphorylation of hypothalamic Akt at Thr³⁰⁸ or Ser⁴⁷³ was not significantly different between the groups (Fig. 6B). For phosphorylation of mTOR at Ser²⁴⁴⁸, a downregulation of 18% (P < 0.05) was observed in KO C-C-H compared with WT C-C-H that was again upregulated in KO H-C-H similar to the level of WT C-C-H. A similar regulation was observed for the phosphorylation of S6 at Ser^235/236.

Gene expression of Npy was 1.27-fold upregulated and of Lepr was 32% downregulated in KO H-C-H compared with KO C-C-H (both P < 0.05). There were no significant changes in expression for other anorexigenic and orexigenic genes (Supplementary Fig. 2).

DNA methylation analyses were performed for the promoter regions of Ppara, Ppargc1a, Cpt1b, and Cpt1a in muscle and for the promoter region of Pik3r1 in hypothalamus. For Cpt1b, we identified three CpG sites (CpG_+227/224, CpG₋₇₂, and CpG₋₂₀₂) that had a loss in DNA methylation in KO C-C-H compared with WT C-C-H (Fig. 7A–C). In KO H-C-H, methylation was again increased as seen in WT C-C-H. These changes in methylation were significantly inversely correlated with gene expression levels of Cpt1b (r = −0.540 for CpG_+227/224, P < 0.05; r = −0.593 for CpG₋₇₂, P < 0.05; r = −0.629 for CpG₋₂₀₂, P < 0.005). For Ppara, we identified one CpG site at CpG₋₁₄₀ (Fig. 7D) that also showed a significant inverse correlation with gene expression levels (r = −0.658, P < 0.005). We did not see any significant changes in DNA methylation for the promoter of Ppargc1a or Cpt1a in muscle or of Pik3r1 in hypothalamus. To identify the biological relevance of DNA methylation at the associated CpG sites, EMSA were performed. We detected methylation-specific binding of transcription factors at the Ppara promoter region using nuclear extracts from C2C12 mouse myoblasts in four independent EMSA. We observed additional protein-DNA complex formation at the methylated Ppara CpG₋₁₄₀ site (Fig. 7E). For the other three tested CpG sites, no methylation-specific protein-DNA was detected (Supplementary Figs. 3–5).

This study shows that Gipr^−/− mice are no longer protected from the adverse effects of an HFD in early adulthood when this diet was applied during IU and L. The phenotypic alterations were evident by increased adiposity, enlarged adipocytes, impaired glucose tolerance, and increased proinflammatory gene expression in adipose tissue in KO H-C-H compared with KO C-C-H mice. KO H-C-H mice behaved similar to WT mice on a normal chow during IU/L and exposed to an HFD later in life. Mechanistically, this phenotype is most likely due to central inhibition of PI3K causing decreased insulin sensitivity in the hypothalamus and to decreased peripheral fatty acid oxidation in skeletal muscle. Moreover, our study suggests that alterations in DNA methylation might be at least partially responsible for the changes in energy consumption. WT mice on HFD during IU/L and reexposed to this diet later in life (WT H-C-H) showed a metabolic phenotype similar to that of WT C-C-H. Results are shown in Supplementary Figs. 6–11.

We did not see a relevant difference in body weight or body fat between KO H-C and KO C-C mice up to the age of 25 weeks. After the HFD was started at the age of 25 weeks, differences for genotypes and maternal diet became evident. KO C-C-H showed significantly lower body weight and total body fat compared with WT C-C-H as shown previously (5,8). The growth curve of KO H-C-H mice was in between the WT C-C-H and KO C-C-H mice, but KO H-C-H mice had significantly increased total body fat after 6 and 12 weeks upon reexposure to an HFD compared with KO C-C-H mice. We show that even after a time window of 22 weeks on the C diet, Gipr^−/− mice on an HFD during IU/L (“first hit”) are programmed to overreact toward the “second hit” of the HFD compared with Gipr^−/− mice on a C diet during IU/L. This is also reflected by the increased size of adipocytes in KO H-C-H compared with KO C-C-H mice, which is important, since the size of adipocytes is correlated with adipose tissue dysfunction and insulin resistance (22). This strong programming phenomenon was shown recently in a CID1 mouse model (15,22): WT male offspring of damns fed an HFD during IU/L were kept on normal chow after weaning until the age of 25 weeks and then reintroduced to the HFD for 19 weeks. These mice showed higher body weight, larger adipocytes, and reduced glucose tolerance compared with mice exposed to the HFD at 25 weeks of age only.

The GIP signaling pathway is necessary for adipocyte development (23). This might explain why lacking the GIPR protects Gipr^−/− mice from diet-induced obesity (5,24). However, if the HFD appears as a “second hit” later in life, this protection disappears and the fetal programming effect of an HFD overrules the Gipr^−/− phenotype. Nevertheless, there was less weight gain in the KO H-C-H mice than in WT controls suggesting that some protection remained despite fetal programming effects of an HFD. One possible component might relate to the moderate impairment of insulin release in the Gipr^−/− mice, which reduces the obesogenic effect of insulin. It is most likely that KO H-C-H as well as WT C-C-H mice have severe β-cell damage, since serum insulin concentrations were dramatically reduced during the intraperitoneal GTT, whereas KO C-C-H were able to minimally increase insulin release at the beginning of the intraperitoneal GTT.

Obesity is associated with increased gene expression of proinflammatory cytokines or MCP-1 (25). MCP-1 seems to be responsible for macrophage infiltration in adipose tissue (26,27). Macrophage infiltration can be quantified by measuring gene expression of macrophage inflammatory protein 1-α (Ccl3) and Emr1 (28,29). We saw a significant decrease of gene expression of Clc2, Ccl3, and Emr1 in KO C-C-H compared with WT C-C-H. In KO H-C-H, gene expression of Clc2 significantly increased, whereas for Emr1 and Ccl3 there was a trend for upregulated gene expression in these mice.

Even though the GIPR is absent in skeletal muscle (30,31), genes involved in fatty acid oxidation were significantly upregulated in KO C-C-H compared with WT C-C-H but were downregulated in KO H-C-H. DNA methylation analysis correlated with the transcription data and indicated the induction of epigenetic changes in Gipr^−/− mice in muscle for Cpt1b and Ppara once they are exposed to an HFD during IU/L. EMSA experiments suggest a functional relevance of the identified Ppara CpG₋₁₄₀ site for transcriptional regulation.

It has been shown that the IU milieu is able to induce epigenetic changes. A study in rats demonstrated that an HFD during IU/L resulted in DNA hypomethylation in liver in young male offspring that seemed to be associated with long-term hepatic dysfunction (32).

Miyawaki et al. (5) reported no differences in food intake between WT and Gipr^−/− mice on an HFD over a period of 4 days. We clearly see an increased cumulative food intake in KO C-C-H compared with WT C-C-H, which matches the observation that Gipr^−/− mice use fat for energy metabolism rather than storing it in adipose tissue (5). This phenomenon was reversed when Gipr^−/− mice were placed on an HFD during IU/L, indicating that GIP physiologically prevents increased energy expenditure in response to increased energy intake as shown earlier (7).

Energy intake is regulated in the hypothalamus, and hypothalamic insulin sensitivity is necessary for balanced food consumption (33). It has been shown that a reduction in the catalytic subunit p85α of PI3K enhances insulin-stimulated Akt activity (34). It was reported that PI3K subunit p85α was upregulated in the hypothalamus of mice that were exposed to an HFD throughout IU/L and the first 120 days of life (35). HFD-induced obesity was shown to require activation of PI3K-dependent hypothalamic pathways (36). Our results indicate that KO C-C-H mice have enhanced hypothalamic insulin sensitivity compared with WT C-C-H mice, since we saw a significant reduction of hypothalamic p85α PI3K gene expression in the former, which correlated with the glucose challenge test. This was reversed in KO H-C-H mice similar to WT C-C-H mice. These results were further supported by increased phosphorylation of the hypothalamic kinase mTOR and S6 protein in the mTOR signaling pathway downstream of Akt. Increased mTOR signaling in the hypothalamus is associated with decreased food intake (37), which was also observed in our study. Thus, central modulation of insulin sensitivity is proposed as a novel mechanism involved in GIP action. However, we could not detect any DNA methylation changes in the p85α PI3K gene promoter.

In summary, we used the approach of fetal programming by HFD to identify GIP-regulated metabolic pathways that alter hypothalamic insulin sensitivity and the PGC-1α–driven and peroxisome proliferator–activated receptor α–driven fat oxidation in skeletal muscle and adipose tissue inflammation.

Acknowledgments. The authors thank Elisabeth Meyer, Susann Richter, and Kerstin Weinert, of the German Institute of Human Nutrition, and Nicole Spada, of the German Research Center for Environmental Health, for excellent technical assistance. Johannes Beckers and Peter Huypens, of the German Research Center for Environmental Health, are acknowledged for valuable comments on mouse genetics. Sonja Kunze, of the German Research Center for Environmental Health, and Elke Rodriguez and Hans-Jörg Baurecht, of the University Hospital, Kiel, Germany, kindly gave advice on methylation analysis. The authors thank Bernard Thorens, of the Department of Physiology and Center for Integrative Genomics, University of Lausanne, Lausanne, Switzerland, for providing the Gipr^−/− mouse model.

Funding. This work was supported by a grant from Deutsche Diabetes Gesellschaft to M.K. and a grant from Deutsche Forschungsgemeinschaft (Pf 164/14-2) to A.F.H.P.

Duality of Interest. No potential conflicts of interest relevant to this study were reported.

Author Contributions. M.K. designed experiments, researched data, wrote and edited the manuscript, and obtained funding supporting the research. F.K.-N. researched data and wrote and edited the manuscript. F.I. designed experiments and researched data. B.N., A.K., E.R., and H.G. designed experiments, researched data, and wrote and edited the manuscript. T.d.l.H.G. and M.A.O. analyzed data. A.F.H.P. designed experiments, wrote and edited the manuscript, and obtained funding supporting the research. A.F.H.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.