Fetal iron deficiency induces chromatin remodeling at the Bdnf locus in adult rat hippocampus

Abstract

Fetal and subsequent early postnatal iron deficiency causes persistent impairments in cognitive and affective behaviors despite prompt postnatal iron repletion. The long-term cognitive impacts are accompanied by persistent downregulation of brain-derived neurotrophic factor (BDNF), a factor critical for hippocampal plasticity across the life span. This study determined whether early-life iron deficiency epigenetically modifies the Bdnf locus and whether dietary choline supplementation during late gestation reverses these modifications. DNA methylation and histone modifications were assessed at the Bdnf-IV promoter in the hippocampus of rats [at postnatal day (PND) 65] that were iron-deficient (ID) during the fetal-neonatal period. Iron deficiency was induced in rat pups by providing pregnant and nursing dams an ID diet (4 mg/kg Fe) from gestational day (G) 2 through PND7, after which iron deficiency was treated with an iron-sufficient (IS) diet (200 mg/kg Fe). This paradigm resulted in about 60% hippocampal iron loss on PND15 with complete recovery by PND65. For choline supplementation, pregnant rat dams were given dietary choline (5 g/kg) from G11 through G18. DNA methylation was determined by quantitative sequencing of bisulfite-treated DNA, revealing a small alteration at the Bdnf-IV promoter. Chromatin immunoprecipitation analysis showed increased HDAC1 binding accompanied by reduced binding of RNA polymerase II and USF1 at the Bdnf-IV promoter in formerly ID rats. These changes were correlated with altered histone methylations. Prenatal choline supplementation reverses these epigenetic modifications. Collectively, the findings identify epigenetic modifications as a potential mechanism to explicate the long-term repression of Bdnf following fetal and early postnatal iron deficiency.

early-life micronutrient deficiencies profoundly affect brain development and function, leading to reduced educational and job potentials (41). Among these, iron deficiency anemia is the most prevalent, affecting 20–30% of pregnant women and their offspring (38). Early-life iron deficiency anemia causes negative and long-lasting effects on learning and memory, emotion, and social behavior (31). Of great concern is the finding that these learning and memory deficits in humans persist into adulthood despite prompt iron treatment (6, 29, 49).

The hippocampus is an important component of the neural circuit responsible for learning and memory (36, 56). During late fetal and early neonatal life, the rapidly developing hippocampus demands substantial iron import (48, 59) and is vulnerable to iron deficiency. Indeed, animal models of fetal-neonatal iron deficiency anemia show behavioral abnormalities analogous to those in humans that are accompanied by changes in neurotransmission, neuronal structure, and neurochemistry (14, 16, 30). These changes include long-term dysregulation of molecules critical for synaptic plasticity (44, 61). Moreover, similar hippocampal pathologies are observed in the nonanemic genetic models of hippocampus-specific iron deficiency, suggesting that these effects are specific to the loss of neuronal iron (7, 15), which underscores the critical requirement for adequate iron during hippocampal development.

Consistent with learning and memory deficits, early-life iron deficiency causes persistent downregulation of Bdnf, a gene critical for hippocampal plasticity (62, 66). The Bdnf gene consists of eight 5′-noncoding exons, a single 3′-coding exon, and two differential poly-A tail signals, all of which potentially generate 16 transcript variants (1). The Bdnf-IV transcript variant is robustly expressed in the postnatal rat hippocampus (1). The long-term downregulation of Bdnf resulting from early-life iron deficiency potentially implicates an epigenetic mechanism, such as changes in DNA methylation, histone modifications, or chromatin protein binding (25). Bdnf is epigenetically regulated in the hippocampus in a number of contexts, including memory formation, environmental enrichment, perinatal exposure to methylmercury, and seizures (24, 32, 34, 43). Moreover, repression of hippocampal Bdnf by early-life iron deficiency is reversed with choline supplementation during late gestation (23). As a methyl donor, choline supplementation can epigenetically alter gene transcription (26, 42) and may improve Bdnf expression by reversing the epigenetic modifications caused by early-life iron deficiency. However, whether early-life iron deficiency induces chromatin remodeling at the Bdnf locus has yet to be determined. Thus, we examined the specific effects of early-life iron deficiency on DNA and histone methylation at the Bdnf-IV promoter.

MATERIALS AND METHODS

Animals.

Gestational day 2 (G2) pregnant Sprague-Dawley rat dams were obtained from Charles River (Wilmington, MA) and kept on a 12:12-h light-dark cycle with ad libitum access to food and water. Fetal-neonatal iron deficiency was induced by dietary manipulation, as described previously (46) with purified diets. In brief, pregnant dams were given a purified iron-deficient (ID) diet (4 mg/kg Fe, 1.1 g/kg choline chloride, TD 80396; Harlan Teklad, Madison, WI) from G2 to postnatal day (PND) 7; thereafter, nursing dams were given a purified iron-sufficient (IS) diet (200 mg/kg Fe, 1.1 g/kg choline chloride, TD 01583; Harlan Teklad). Iron treatment commenced at PND7 to mimic the full-term human neonate in terms of hippocampal development (http://www.translatingtime.net). This dietary manipulation generated iron-deficient (ID) pups with ∼60% hippocampal iron loss at PND15 (20), a time when rapid dendritic arborization and synaptogenesis commence in CA1 (45), and the hippocampus is iron-repleted by PND65 (formerly iron-deficient, FID) (21). Control iron-sufficient rats (IS) were generated from pregnant dams maintained on a purified IS diet. To supplement the choline-fortified diet, pregnant rat dams were given ID+choline (4 mg/kg Fe, 5.0 g/kg choline chloride) or IS+choline (200 mg/kg Fe, 5.0 g/kg choline chloride) diets during gestational day 11 through 18, as previously described (23). All litters were culled to eight pups (six males and two females) at PND0 and weaned at PND21. Same sex littermates were housed together with 4 rats/cage. Only male offspring were used in the experiments. The University of Minnesota Institutional Animal Care and Use Committee approved all experiments in this study.

Hippocampal dissection.

PND65 rats were killed by an intraperitoneal injection of pentobarbital sodium (100 mg/kg). Brains were removed from the cranium and bisected along the midline on an ice-cold metal block. Both left and right hippocampi were dissected separately and immediately flash-frozen in liquid nitrogen and stored individually at −80°C.

Bisulfite-conversion and pyrosequencing.

DNA was extracted using the standard phenol chloroform method, and the amount was measured using a NanoDrop spectrophotometer. One microgram of DNA was bisulfite-treated using the EpiTect Bisulfite Kit (Qiagen, Valencia, CA) following the manufacturer's protocol. Pyrosequencing was performed to analyze the methylation profiles at Bdnf using custom-designed oligonucleotides that target the rat Bdnf-IV (NW_047763: 251886–252199) promoter (Bdnf-P4). The PyroMark PCR kit (Qiagen) was used in a 25-μl reaction, according to the manufacturer's protocol. PCR conditions were 95°C for 15 min followed by 45 cycles of 95°C for 15 s, 55°C for 30 s, and 72°C for 15 s. Ten microliters of the biotinylated PCR product was used for each sequencing assay. Pyrosequencing was done using the PyroMark Q96MD (Qiagen) system following the manufacturer's protocol and the PyroMark Gold 96 reagents kit (Qiagen). Methylation was analyzed using Qiagen's Pyro Q-CpG software. 7 CpG sites were analyzed for Bdnf-P4.

Chromatin immunoprecipitation assay.

Chromatin was prepared from hippocampal tissue following the manufacturer's recommendation (Millipore, Temecula, CA). All left hippocampi were homogenized in ice-cold PBS (500 μl) using a motorized plastic pestle and pelleted by centrifugation at 13,000 rpm (30 s). The pellets were resuspended in PBS and fixed in 1% formaldehyde solution with intermittent mixing at 37°C (5 min) and room temperature (5 min). Formaldehyde fixative was removed with a PBS (1 ml) rinse by pelleting suspension. Pellets were then resuspended in 500-μl lysis buffer [1% SDS, 10 mM EDTA, 50 mM Tris pH 8.1, 1 mM PMSF, 10 μl 10× protease inhibitor cocktails (Roche, Indianapolis, IN)] and cooled in an ice bath for 10 min. Lysates were then sonicated (Branson Sonifier 250; VWR Scientific, Batavia, IL) to shear DNA (200–1,000 bp), 1/10 of which was used for validation by agarose gel electrophoresis following a cross-linking reversal (0.2 M NaCl, 65°C overnight). Sonicated lysates were diluted 10-fold with a chromatin immunoprecipitation assay (ChIP) dilution buffer (0.01% SDS, 1.1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris-pH 8.1, 167 mM NaCl) and precleared with 75 μl of protein A agarose/salmon sperm DNA (50% slurry; Millipore). Following removal of Protein A agarose, precleared lysate was immunoprecipitated by ChIP-grade antibody with an end-over-end rotation (4°C, overnight). The antibody-histone complex was collected by the addition of 50 μl of protein A agarose/salmon sperm DNA slurry with mixing (4°C, 1 h). Following washes (per manufacturer's protocol; Millipore), immune-histone complex was eluted in 200 μl of elution buffer (1% SDS, 0.1 M NaHCO₃). Reverse cross-linking was achieved by incubation in NaCl (0.2 M, 65°C, overnight). A protease digestion (20 μg proteinase K, 20 mM EDTA, 100 mM Tris-pH 6.5, 45°C, 1 h) was performed to recover DNA, which was further purified using ChIP DNA clean and concentration kit (Zymo Research, Irvine, CA). Antibodies used in this study included anti-H4ac (GAM-0202; SA Biosciences, Frederick, MD), anti-H3K4me3 (ab8580; Abcam, Cambridge, MA), anti-H3K4me1 (ab8895; Abcam), anti-H3K27me3 (ab6002; Abcam), anti-USF1 (sc-229X; Santa Cruz Biotechnology, Santa Cruz, CA), anti-HDAC1 (CS200577; Millipore, Billerica, MA), and anti-RNA pol II (8WG16; Pierce, Rockford, IL). Normal IgG was used as a negative control. Levels of precipitated GAPDH (active) and MyoD1 (inactive) loci were used to validate the ChIP experiments.

Real-time PCR.

For analysis of precipitated DNA from ChIP, SYBR-Green PCR (Fast SYBR Green master mix; Applied Biosystems, Foster City, CA) was used to amplify Bdnf-P4 using the following oligonucleotides: Bdnf-IV forward (GATGAAAGGTTTGGCTTCTGTG) and reverse (TCGGTGAATGGGAAAGTGG). Oligonucleotides were validated for amplification efficiency (−3.2 slope). These oligonucleotides amplify a 250-bp Bdnf-IV (NW_047673:251697–251947) promoter. Input DNA (10%) was used as a normalizer to account for input amount (ΔCt). Data were expressed as a ratio to IS control (2^−ΔΔCt) using one of the IS sample as a calibrator (ΔΔCt). For Bdnf mRNA quantitation, TaqMan PCR was used with primers and probes obtained from Applied Biosystems (Foster City, CA). Real-time PCR was performed with Stratagene 3000P instrument.

Western blot analysis.

To validate the mRNA data, we used Western blot to quantify levels of BDNF protein using a previously described protocol (63). In brief, 30 μg of hippocampal protein lysate was separated using a 4–20% SDS-PAGE gel (Novex, Life Technologies, Carlsbad, CA). Proteins were blotted onto a nitrocellulose membrane (Pierce), blocked with blocking buffer for fluorescent Western blotting (Rockland, Gilbertsville, PA), and incubated with antibodies against BDNF (rabbit IgG; Abcam), and β-actin (mouse monoclonal; Sigma, St. Louis, MO). Following PBS ± 0.1% Tween-20 washes, blots were incubated in Alexa-Fluor 700 anti-mouse (Invitrogen, Grand Island, NY) and IR Dye-800 anti-rabbit (Rockland), and analyzed by Odyssey Infrared Imaging System (Li-Cor Biosciences, Lincoln, NE). Integrated intensity of BDNF normalized to β-actin was determined using Photoshop CS5.1 (Adobe, San Jose, CA).

Statistics.

Group differences were analyzed by unpaired t-tests with alpha set at 0.05. Data were graphed and statistically analyzed using Prism 5 GraphPad (GraphPad Software, San Diego, CA).

RESULTS

Early-life iron deficiency induces changes in DNA methylation at Bdnf-IV promoter.

Consistent with our previous findings (4, 62), early-life iron deficiency caused a persistent decrease in hippocampal Bdnf-IV (previous nomenclature BDNF-III) mRNA and BDNF protein (Fig. 1A). To determine whether decreased expression was associated with changes in DNA methylation, we assessed DNA methylation at the promoter region proximal to the transcription start site of this transcript variant (Fig. 1B). Within this promoter region, there are multiple CG sites (Fig. 1C) with methylation levels that are differentially affected by early-life iron deficiency (Fig. 1D). Overall, the PND65 FID hippocampus showed a small reduction in percentage of DNA methylation at the Bdnf-IV (P4) promoter (Fig. 1D). To determine whether early-life iron deficiency affects expression of DNA methyltransferases, we measured hippocampal Dnmt1 and Dnmt3a mRNA and found no difference in either ID PND15 or FID PND65 hippocampus compared with IS controls (data not shown).

Fig. 1.

Early-life iron deficiency decreased BDNF expression and altered DNA methylation at the Bdnf-IV promoter in PND65 rat hippocampus. A: reduced Bdnf-IV mRNA and BDNF protein levels in PND65 formerly iron-deficient (FID) hippocampus compared with iron sufficient (IS) control. B: schematic representation of the rat Bdnf locus with multiple noncoding 5′-exons. C: selected sequence of Bdnf-IV promoter (P4) is shown with CpG islands (bold and italic), qPCR oligonucleotide (arrow), and identified USF1 binding site (italicized and stylized letters). D: quantitative analysis of methylated CpG showed a lower overall methylation rate, particularly at two CpG islands within promoter P4 of FID hippocampus. Values are expressed as means ± SD; n = 5 or 6/group. *P < 0.05 and **P < 0.01, using an unpaired t-test.

Increased presence of HDAC1 at the Bdnf-IV promoter.

Given the implication of HDAC1 in modulating Bdnf-IV expression (24, 65), we quantified the presence of HDAC1 at the Bdnf-P4. At PND65, levels of HDAC1 were increased in the FID hippocampus compared with IS controls (Fig. 2A). Consistent with previous findings relating HDAC1, histone H4 acetylation, and RNA polymerase II in modulating expression of genes with multiple splice variants (17), the PND65 FID hippocampus showed a decrease in H4ac at the Bdnf-P4 (Fig. 2B). We further analyzed the binding of RNA polymerase II (pol II) and found a decrease in pol II at the Bdnf-P4 in FID hippocampus (Fig. 2C). Finally, binding of the activity-dependent transcription factor USF1 at the Bdnf-P4 (8, 68) was similarly decreased in the adult FID hippocampus (Fig. 2D).

Fig. 2.

Early-life iron deficiency reduced histone acetylation and binding of transcription regulators at the Bdnf-IV promoter in PND65 FID hippocampus. ChIP analyses showed an increased HDAC1 binding (A), decreased histone H4 acetylation (B), decreased binding of RNA polymerase II (C), and USF1 (D). Values are expressed as means ± SE; n = 4–8/group. *P < 0.05 and **P < 0.01, using an unpaired t-test.

Abnormal histone H3 methylation.

To identify a possible mechanism behind the increased HDAC1 concomitant with a decreased USF1 binding at the Bdnf-IV promoter in the FID hippocampus, we analyzed histone H3 modifications (marks) that modify chromatin structure and influence transcriptional activity. The adult FID hippocampus showed a significant increase in K27me3 and K4me1 marks accompanied by a decrease in the K4me3 mark (Fig. 3).

Fig. 3.

Early-life iron deficiency altered histone H3 methylation at the Bdnf-IV promoter in PND65 hippocampus. ChIP analyses showed an increased enrichment of K27me3 and K4me1 and a decreased enrichment of K4me3. Data were normalized by 10% input and calibrated to IS control. Values are expressed as means ± SE; n = 6–8/group. *P < 0.05, ***P < 0.001, using an unpaired t-test.

Dietary choline supplementation during late gestation reverses early-life iron deficiency-induced epigenetic modifications.

We have previously shown that dietary choline supplementation during late gestation normalizes hippocampal Bdnf-IV expression in adult FID rats (23). This transcriptional recovery is accompanied by a partial recovery of hippocampal BDNF protein (Fig. 4). To determine whether this reversal is associated with remodeling of chromatin structures, we analyzed histone modifications at the Bdnf-IV promoter following prenatal choline supplementation. Consistent with the findings presented in the Figs. 2 and 3, HDAC1 binding and enrichment of the K27me3 mark were increased in the FID hippocampus (Fig. 5, A and B), while USF1 binding and enrichment of the K4me3 mark were decreased (Fig. 5, C and D). These epigenetic modifications induced by early-life iron deficiency were partially or completely reversed by prenatal dietary choline supplementation (Fig. 5, FID-C group).

Fig. 4.

Western blot analysis showed a partial recovery of BDNF protein level in PND65 FID hippocampus following prenatal choline supplementation. Values are expressed as means ± SE; n = 4–6/group. **P < 0.01, using an unpaired t-test.

Fig. 5.

Dietary choline supplementation during late gestation reverses early-life iron deficiency induced epigenetic modifications at the Bdnf-IV promoter. ChIP analyses of K27me3 enrichment (A), HDAC1 binding (B), K4me3 enrichment (C), and USF1 binding (D) showed consistent changes induced by ID, which recovered with choline supplementation. Values are expressed as means ± SE; n = 6–8/group. *P < 0.05, and ***P < 0.001, using an unpaired t-test.

DISCUSSION

The long-lasting impairments in cognition and socio-emotional behaviors induced by early-life iron deficiency can be ascribed, in part, to the persistent dysregulation of hippocampal Bdnf (30, 62). Consistent with this notion, we have previously documented evidence that altered BDNF is associated with abnormal CA1 dendritic growth and maturation and BDNF-mediated MAPK signaling in the PND65 FID hippocampus (5, 21, 47, 61). Furthermore, there are strong links between BDNF signaling and learning and memory, which are known to be sensitive to epigenetic modifications (35, 64). However, the possibility of an epigenetic mechanism underlying the early-life iron deficiency induced long-term gene repression has not been investigated. The present study shows that early-life iron deficiency induced stable histone modifications at the Bdnf promoter. The findings reveal a potential epigenetic mechanism by which iron deficiency induces chromatin remodeling and, thereby, renders genes such as Bdnf-IV less transcriptionally responsive.

Chromatin remodeling is a dynamic process involving multiple modifications of DNA and histones (25) and is regulated, in part, by nutrient availability (50, 60). Since iron is a necessary cofactor for multiple JmjC ARID domain-containing (JARID) histone demethylases (9, 11, 54, 55, 58), a loss of iron during hippocampal development would likely affect chromatin structures. It has not previously been demonstrated whether iron deficiency affects chromatin remodeling. Our findings that early-life iron deficiency induced significant histone modifications and small changes in DNA methylation at the Bdnf locus in the adult rat hippocampus suggest that histone modifications likely play a key role in the long-term Bdnf repression. This observation differs from other models of early-life insults (e.g., stress and neglect), which demonstrate that increased DNA methylation at CpG islands around the Bdnf-IV promoter is associated with long-term repression of hippocampal Bdnf (51, 52). The disparate outcomes likely relate to a different mode of early-life insults. Our early-life iron deficiency model is a metabolic insult. Thus, our findings may uncover a dynamic interaction between iron and histone modifications in mounting an adaptive response to early-life iron deficiency.

Methylation and acetylation of histones are interdependent histone modifications. K4me3 enrichment is associated with increased histone acetylation and recruitment of histone acetyltransferases (HATs) at active loci (2, 19). In contrast, K27me3 enrichment is associated with reduced histone acetylation and increased histone deacetylases (HDACs) binding at inactive loci (2, 19, 22, 25, 27, 37). Thus, increased HDAC1 binding concomitant with decreased H4ac enrichment at the Bdnf-IV promoter in the FID hippocampus is in line with increased enrichment of K27me3 mark. Increased HDAC1 binding may lead to a lower enrichment of H4 acetylation, since hippocampal mRNA levels of HATs or HDACs were not different between IS and FID groups (transcriptomic data not shown), suggesting an unaltered de novo histone H4 acetylation. Alternatively, the impaired hippocampal learning and memory in the FID group may cause a lower H4ac enrichment at the Bdnf-IV promoter given that learning and memory regulates histone acetylation (12, 57). Moreover, neural stimulation can mediate changes in H4 acetylation and USF1 binding (8, 13). Abnormal hippocampal neurotransmission (20) may cause the lower H4ac enrichment and USF1 binding at the Bdnf-IV promoter in FID hippocampus. Collectively, these chromatin modifications likely contribute to the repression of hippocampal Bdnf.

Enrichment of the K27me3 mark is associated with repressive promoters, whereas enrichment of the K4me3 mark is associated with active promoters (2). The increased enrichment of the K27me3 mark and decreased enrichment of the K4me3 mark at the Bdnf-P4 in the FID hippocampus provides evidence that early-life iron deficiency reduces the transcriptional activity of Bdnf-P4. These changes in histone methylation marks may directly link to our previous observation of dysregulated iron-dependent JARIDs in the adult FID hippocampus (4). Analysis of JARID function in the context of early-life iron deficiency will be important in establishing a link among iron deficiency, epigenetic modifications, and transcriptional regulation. Interestingly, we also found increased enrichment of the K4me1 mark at the Bdnf-P4 of FID hippocampus. This finding appears counterintuitive because the enrichment of this mark is typically found at active promoters (2). However, emerging evidence suggests that increased enrichment of the K4me1 mark and reduced recruitment of RNA pol II at the promoter regions are associated with repression of genes that are inducible (10). In the context of early-life iron deficiency, our finding of lowered pol II binding at the Bdnf-P4 further corroborates this mechanism of Bdnf repression.

Finally, methyl diets (e.g., choline, betaine) are known to influence epigenetic modifications (26, 42, 67). Choline supplementation during the critical period of hippocampal development has been shown to improve hippocampal function (28, 39, 40). Importantly, it diminishes neurobehavioral abnormalities induced by early-life insults, including early-life iron deficiency (23, 53). The reversal of early-life iron deficiency induced epigenetic modifications by prenatal choline supplementation supports the concept that epigenetics is one mechanism by which iron deficiency causes long-term gene dysregulation in the adult rat hippocampus. Whether provision of dietary choline during an early period of hippocampal development resolves the abnormal CA1 dendritic morphogenesis and BDNF-mediated MAPK signaling in the adult PND65 hippocampus remains to be determined in a future study. We predict that the results are likely affirmative given the evidence of recovered learning behavior and BDNF expression (23). The findings in the present study form a mechanistic basis for the use of dietary choline as an adjunctive therapy to complement iron treatment for early-life iron deficiency. Interestingly, significant reductions in repressive (K27me3) and active (K4me3) marks were detected in the control group supplemented with choline (Fig. 5, A and C; ISC group). Additional study is needed to determine whether these changes underlie the reduced Bdnf-IV expression in this group (23). The finding implicates a potential negative impact of too much choline.

Perspectives and Significance

Our data provide evidence that early-life iron deficiency remodels chromatin structure to alter gene expression. This reprogramming may reflect a cellular adaptation to a metabolic insult during development and may play a key role in iron deficiency-induced long-term gene dysregulation. Previous investigations into the mechanistic underpinnings, which focused on the concept of form and function, postulated that altered neural circuitry development is a basis for impaired hippocampal function in adulthood (3, 5, 20). Here, we demonstrated that early-life iron deficiency induces epigenetic modifications at a key gene locus critical for regulation of hippocampal plasticity, providing an additional mechanism by which iron modulates hippocampal development and function. The continued deficits in cognitive function and emotional behaviors, as well as increased risk of adulthood schizophrenia and depression, are concerning in spite of prompt iron treatment of early-life iron deficiency (18, 30, 33). These long-term negative effects illustrate a need to identify adjunctive remedial therapy. Our findings form a molecular basis to test the use of methyl diets, such as choline, as a potential adjunctive therapy.

GRANTS

This study was supported by National Institute of Child Health and Human Development Grant R01-HD29421–17 to M. K. Georgieff.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

Author contributions: P.V.T., R.A.S., and M.K.G. conception and design of research; P.V.T., B.C.K., and Y.-C.L. performed experiments; P.V.T., B.C.K., Y.-C.L., and R.A.S. analyzed data; P.V.T., B.C.K., Y.-C.L., R.A.S., and M.K.G. interpreted results of experiments; P.V.T. and Y.-C.L. prepared figures; P.V.T. drafted manuscript; P.V.T., B.C.K., R.A.S., and M.K.G. edited and revised manuscript; P.V.T., B.C.K., Y.-C.L., R.A.S., and M.K.G. approved final version of manuscript.