FGF-FGFR Mediates the Activity-Dependent Dendritogenesis of Layer IV Neurons during Barrel Formation

Abstract

Fibroblast growth factors (FGFs) and FGF receptors (FGFRs) are known for their potent effects on cell proliferation/differentiation and cortical patterning in the developing brain. However, little is known regarding the roles of FGFs/FGFRs in cortical circuit formation. Here we show that Fgfr1/2/3 and Fgf7/9/10/22 mRNAs are expressed in the developing primary somatosensory (S1) barrel cortex. Barrel cortex layer IV spiny stellate cells (bSCs) are the primary recipients of ascending sensory information via thalamocortical axons (TCAs). Detail quantification revealed distinctive phases for bSC dendritogenesis: orienting dendrites toward TCAs, adding de novo dendritic segments, and elongating dendritic length, while maintaining dendritic patterns. Deleting Fgfr1/2/3 in bSCs had minimal impact on dendritic polarity but transiently increased the number of dendritic segments. However, 6 d later, FGFR1/2/3 loss of function reduced dendritic branch numbers. These data suggest that FGFs/FGFRs have a role in stabilizing dendritic patterning. Depolarization of cultured mouse cortical neurons upregulated the levels of several Fgf/Fgfr mRNAs within 2 h. In vivo, within 6 h of systemic kainic acid administration at postnatal day 6, mRNA levels of Fgf9, Fgf10, Fgfr2c, and Fgfr3b in S1 cortices were enhanced, and this was accompanied by exuberant dendritogenesis of bSCs by 24 h. Deleting Fgfr1/2/3 abolished kainic acid-induced bSC dendritic overgrowth. Finally, FGF9/10 gain of function also resulted in extensive dendritogenesis. Together, our data suggest that FGFs/FGFRs can be regulated by glutamate transmission to modulate/stabilize bSC dendritic complexity. Both male and female mice were used for our study.

SIGNIFICANCE STATEMENT Glutamatergic transmission plays critical roles in cortical circuit formation. Its dysregulation has been proposed as a core factor in the etiology of many neurological diseases. We found that excessive glutamate transmission upregulated mRNA expression of Fgfrs and their ligands Fgfs. Deleting Fgfr1/2/3 not only impaired bSC dendritogenesis but also abolished glutamate transmission-induced dendritic overgrowth. Overexpressing FGF9 or FGF10 in cortical glutamatergic neurons results in excessive dendritic outgrowth within 24 h, resembling the changes induced by excessive glutamate transmission. Our findings provide strong evidence for the physiological role of fibroblast growth factors (FGFs) and FGF receptors (FGFRs) in establishing and maintaining cortical circuits. Perturbing the expression levels of FGFs/FGFRs by excessive glutamatergic neurotransmission could lead to abnormal neuronal circuits, which may contribute to neurological and psychiatric disease.

Introduction

One remarkable feature of cortical glutamatergic neurons is their distinctive dendritic arbor, where most of their presynaptic connections reside. The spatial arrangement of dendritic segments affects how synaptic inputs are propagated and integrated at the soma and subsequently affect synaptic plasticity and behavior (Häusser et al., 2000; Sjöström et al., 2008). During early brain development, both intrinsic and extrinsic signals can modulate dendritogenesis (McAllister, 2000; Gao, 2007; Jan and Jan, 2010; Joo et al., 2014). In particular, neuronal activity plays a crucial role in modulating dendritic morphogenesis (McAllister, 2000; Cline, 2001; Wong and Ghosh, 2002; Chen and Ghosh, 2005).

The rodent whisker-barrel map in the primary somatosensory (S1) cortex is a popular model system to explore the role of neural activity in sculpting cortical sensory circuits (Simons and Woolsey, 1979; Feldman and Brecht, 2005; Petersen, 2007; Wu et al., 2011). Barrel cortex layer IV spiny stellate cells (bSCs) project their dendrites toward the clusters of thalamocortical axons (TCAs) relaying sensory information from their corresponding whiskers. Studies using several “barrelless” mutant mice lacking barrel cytoarchitecture, indicate a key role for glutamate transmission in bSC dendritogenesis (Datwani et al., 2002; Espinosa et al., 2009; Ballester-Rosado et al., 2010; Narboux-Nême et al., 2012; Matsui et al., 2013; Mizuno et al., 2014). For example, deleting presynaptic active zone protein RIM1/2 in the thalamus (Narboux-Nême et al., 2012) or deleting postsynaptic glutamate receptors NR1, NR2B, or mGluR5 impair the dendritic morphogenesis of bSCs (Espinosa et al., 2009; She et al., 2009; Ballester-Rosado et al., 2010, 2016; Mizuno et al., 2014). Despite these striking results, the cellular mechanisms by which glutamate receptors regulate dendritic morphogenesis are poorly understood.

FGF signaling plays diverse roles in neurogenesis, brain patterning, and establishing of axonal projections during embryonic development (Thisse and Thisse, 2005; Iwata and Hevner, 2009; Dorey and Amaya, 2010; Guillemot and Zimmer, 2011). The majority of fibroblast growth factors (FGFs; 22 known members) act through FGF receptors (FGFRs) 1–3, all of which contain an intracellular tyrosine kinase domain that can trigger several different signaling pathways, thus having diverse impacts on the cell (Itoh and Ornitz, 2004; Guillemot and Zimmer, 2011). The functions of FGFs/FGFRs in neural circuit formation during postnatal development are only just now being revealed (Umemori et al., 2004; Terauchi et al., 2010; Singh et al., 2012). For example, Umemori et al. (2004) found that different FGF ligands exerted differential impacts on dendritic branching, axon elongation, and vesicle aggregation/synapse formation in chicken embryonic motor neurons. In addition, FGF1 application promotes neurite branching of cultured spinal ganglion cells (Aletsee et al., 2003) and intracerebroventricular FGF2 infusion promotes dendritic growth of adult-born hippocampal granule cells (Rai et al., 2007). Furthermore, cultured FGF2 knock-out neurons exhibit shorter dendritic length (Baum et al., 2016). Together, these findings suggest that FGF–FGFR signaling can regulate dendritic growth; however, whether endogenous FGF/FGFRs are regulators in dendritogenesis in vivo is unknown.

In this study, we explored whether FGF–FGFR signaling is involved in bSC dendritogenesis and whether neural activity regulates FGF/FGFR levels. Using both in vitro and in vivo approaches, we found that Fgf9, Fgf10, Fgfr2c, and Fgfr3b mRNAs were present in the developing S1 cortex and were upregulated by neural activity. Mosaic animals were generated to conduct both gain-of-function and loss-of-function (LOF) experiments. We found that deleting Fgfr1/2/3 in bSCs not only perturbs their dendritic outgrowth but also impairs the maintenance of a stable dendritic pattern. In addition, FGFR1/2/3 LOF abolishes the dendritogenesis induced by excessive glutamate receptor activation. Furthermore, expressing FGF9 or FGF10 in bSCs results in exuberant dendritic branching, similar to the morphology of bSCs following excessive synaptic activity. Together, our results suggest that FGFs/FGFRs play a key role in activity-dependent dendritic patterning of glutamatergic neurons in the developing cortex.

Materials and Methods

Chemicals and antibodies.

All reagents and chemicals were purchased from Sigma-Aldrich, unless otherwise stated. Alexa Fluor-conjugated secondary antibodies were purchased from Invitrogen. Rabbit anti-red fluorescent protein antibody was purchased from Rockland Antibodies & Assays. 2-Methyl-6-(phenylethynyl) pyridine hydrochloride (MPEP), and 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide (NBQX) were purchased from Tocris Bioscience. d-(−)-2-Amino-5-phosphonopentanoic acid (d-APV) and nifedipine were purchased from Abcam.

Experimental design and animals.

The conditional alleles for Fgfr1 (Fgfr1^flox/flox), Fgfr2 (Fgfr2^flox/flox), and Fgfr3 (Fgfr3^flox/flox) were generated as described previously (maintained in an FVB background; Trokovic et al., 2003; Yu et al., 2003; Su et al., 2010). Male and female triple transgenic mice homozygous for Fgfr1^flox/flox;Fgfr2^flox/flox;Fgfr3^flox/flox (abbreviated as Fgfr1/2/3 triple conditional mice) were originally generated in Dr. Jean M. Hébert's laboratory (Kang and Hébert, 2015) and transferred to the Lu laboratory to establish breeding colonies, initially at Baylor College of Medicine and then at Indiana University. Wild-type ICR mice from Envigo were used for cortical neuron culture and FGF overexpression studies. Both male and female mice were used for data analysis. Animals were treated in compliance with the U.S. Department of Health and Human Services, Baylor College of Medicine, and Indiana University, Bloomington, guidelines and procedures approved by the relevant institutional animal care and use committee.

Genotyping.

Tail lysates were prepared by immersing tail pieces in tail digestion buffer (50 mm KCl, 10 mm Tris-HCl, Triton X-100, 0.1 mg/ml proteinase K, pH 9.0) and vortexing gently for 3 h at 60°C to lyse the tail tissue. Tail lysates were heated to 94°C for 10 min to denature the proteinase K (Thermo Fisher Scientific) then centrifuged at 16,100 × g for 15 min. The supernatants were used as DNA templates for PCRs (EconoTaq Plus Green 2X Master Mix, Lucigen). The genotyping primers are as previously described (Trokovic et al., 2003; Yu et al., 2003; Su et al., 2010).

Administration of kainic acid.

Kainic acid (KA) was dissolved in sterilized saline (0.9% NaCl) and administered intraperitoneally or subcutaneously to neonatal ICR pups [postnatal day 6 (P6)] at a dose of 1.5 mg/kg, according to previous studies that have demonstrated that this dose reliably induces electrographic and behavioral seizures (Stafstrom et al., 1992; Lynch et al., 2000; Howland et al., 2004). Pups were separated from their dam immediately before saline or KA administration. Littermates injected with saline served as controls. All pups injected with KA displayed swimming seizure-like behaviors within 15–30 min of injection as previously reported for rat pups (Stafstrom et al., 1992; Lynch et al., 2000; Howland et al., 2004). The seizing behavior lasted for 3–4 h. During this time, saline-injected pups were placed together with their littermates treated with KA. Once seizing behaviors stopped in all pups, they were returned simultaneously to their dam.

Plasmid construction.

The full-length cDNA coding regions for mouse Fgf9 (mFGF9) and Fgf10 (mFGF10) were PCR amplified from pMD18-mFGF9 and pMD18-mFGF10 (Sino Biological) using the following primers for mFGF9: forward-5′-CTAGCTAGCATGGCTCCCTTAGGTGAAGTTGGG-3′, reverse-5′-TTGGCGCGCCTCAGCTTTGGCTTAGAATATCCTTA-3′; and for mouse FGF10: forward-5′-CTAGCTAGCATGTGGAAATGGATACTGACACATT-3′, reverse-5′-TTGGCGCGCCCTATGTTTGGTATCGTCATGGGGAG-3′. (Italics indicate the restriction sites for NheI and AscI.) The PCR fragment was cloned between NheI and AscI sites of the pAAV-EF1-DIO-hCHR2(H134R)-EYFP-WPRE plasmid (a gift from Mingshan Xue, Baylor College of Medicine and Jan and Dan Duncan Neurological Research Institute at Texas children's hospital, Houston, TX) and designated as pAAV-EF1α-DIO-mFGF9 and pAAV-EF1α-DIO-mFGF10.

In utero electroporation.

In utero electroporation (IUE) was performed as previously described (Shimogori, 2006; Rice et al., 2010). Briefly, pregnant female dams 14.5 d postgestation were anesthetized by isoflurane inhalation, and a small incision was made in the abdominal wall to expose the uterine horns. Approximately 0.5–1 μl of the DNA solution was injected into the lateral ventricle of the embryo using a pulled glass micropipette. For deleting FGFR1/2/3 from cortical neurons, pAAV-EF1α-icre-P2A-tdTomato (0.5 μg/μl) was electroporated into the right hemisphere of 50% of the Fgfr1/2/3 triple conditional embryos. pAAV-EF1α-tdTomato (0.25 ∼ 0.5 μg/μl) was electroporated into the left hemisphere of the other 50% Fgfr1/2/3 triple conditional mice that served as littermate controls. For the FGF9/FGF10 gain-of-function experiments, a DNA mixture of pCAG-ER^T2CreER^T2 (Matsuda and Cepko, 2007), pAAV-EF1α-DIO-tdTomato (a gift from Dr. Mingshan Xue), and pAAV-EF1α-DIO-mFgf9/10 (mixed in a molar ratio of 10:1:10, with a final concentration of 1 μg/μl) was infused into the right hemisphere of 50% wild-type embryos (ICR mice). The remaining embryos received a DNA mixture of pCAG-ER^T2CreER^T2 and pAAV-EF1α-DIO-tdTomato to serve as littermate controls (mixed in a molar ratio of 10:1). Each embryo within the uterus was placed between platinum tweezers-type electrodes (5 mm diameter; Harvard Apparatus). Square electric pulses (30–35 V, 50 ms) were passed five times at 1 s intervals using an electroporator (catalog #ECM830, Harvard Apparatus). After electroporation, the wall of the abdominal cavity and skin were then sutured, and embryos were allowed to develop to term, depending on the experiment.

Primary neuronal culture.

Cerebral cortices were dissected from embryonic day 15.5 (E15.5) ICR mouse embryos and dissociated, and the dissociated cells were seeded at 8 × 10⁵/cm² and maintained in Neurobasal medium (Invitrogen) supplemented with 2% B27 supplement (Invitrogen), 50 U/ml penicillin, and 50 μg/ml streptomycin, as described previously (Huang et al., 2009). Neurons were cultured for 7 d in vitro (DIV7) for pharmacological experiments.

Total RNA extraction and RT-PCR.

Total RNA was extracted from neuronal cultures or brain tissue by RNeasy Mini Kit (Qiagen) and followed by on column DNase digestion according to the manufacturer instructions. One microgram total RNA was converted to cDNA by using an iScript TM cDNA Synthesis Kit (Bio-Rad). For real-time PCR, cDNA was mixed with PCR master mix solution (Bio-Rad) and 0.4 μm of each forward and reverse primer in a final volume of 20 μl. Real-time PCR primers were as previously described (Fon Tacer et al., 2010). The PCR was reacted in the CFX96 Touch Real-time PCR Detection System (Bio-Rad). The Δ Δ Ct method was used to calculate relative expression levels.

Immunostaining.

Mice were anesthetized and perfused with PBS followed by 4% paraformaldehyde (PFA)/PBS. Brains were harvested and postfixed with 4% PFA/PBS overnight at 4°C. Fixed brains were sectioned into 100-μm-thick sections in the coronal plane by using a vibrating microtome (VT-1000 Vibrating Microtome, Leica Microsystems). Free-floating sections were used in all subsequent steps. Sections were permeabilized with 0.2% Triton X-100 in PBS for 20 min at room temperature, incubated with blocking solution (3% normal goat serum prepared in PBS with 0.01% Triton X-100), and then incubated overnight with rabbit anti-red fluorescent protein antibody (1:2000 dilution in blocking solution) to identify tdTomato protein. Goat anti-rabbit IgG-conjugated with Alexa Fluor 555 antibody (1:2000 dilution) and 4′,6′-diamidino-2-phenylindole dihydrochloride (5 μg/ml) were used to detect primary antibody and visualize nuclei, respectively.

Reconstruction and quantification of neuronal morphology.

In brain sections, tdTomato-expressing layer IV bSCs were identified as those neurons with nuclei located near barrel borders and lacking apical dendrites projecting toward the pial surface. The z-stack images of these layer IV cortical neurons were acquired with a Leica SP8 confocal microscope with a 25× numerical aperture 0.95 objective. The z-stacks were taken at 0.5 μm intervals to image all the dendritic segments connected to the cell bodies. Neuronal morphology was traced and reconstructed using Imaris software (Bitplane) with the Neurofilament module. Dendritic order was defined using the centrifugal method, as described previously (Uylings et al., 1975). Polarized and nonpolarized neurons were defined as previously described (Espinosa et al., 2009; Ballester-Rosado et al., 2010). The greatest sum of dendritic length in a continuous 180° sector was defined as the dendritic length inside the barrel, while the remainder was defined as the dendritic length outside the barrel. All imaging and analysis was performed in a blinded manner.

Statistical analysis.

Data were acquired from both male and female animals and were combined for data analysis. Data analysis and determination of statistical significance were performed using GraphPad Prism 6.0 software (GraphPad Software). In the figures, data are expressed as the mean ± SEM. We used the unpaired t test, Pearson correlation coefficients, one-way ANOVA, or two-way ANOVA to examine data, as presented in the figure legends. Original data values and detail statistical comparisons for all figures are contained in extended data.

Results

Simultaneously eliminating Fgfr1/2/3 perturbed activity-guided dendritogenesis in vivo

Fgfr1/2/3 are expressed in the developing S1 cortex based on in situ hybridization data shown in the Allen Brain Atlas (http://developingmouse.brain-map.org/), while Fgfr4 expression has not been detected in neocortical precursors during development (Kang et al., 2009). Fgfr5 lacks an intracellular kinase domain and is thought to act as a negative regulator of FGF–FGFR signaling (Sleeman et al., 2001). Despite the differential binding affinities of an individual FGF for FGFRs, it is proposed that different FGFRs can compensate for each other's functions (Itoh and Ornitz, 2004; Beenken and Mohammadi, 2009; Guillemot and Zimmer, 2011; Hébert, 2011). Considering the redundant function of FGFR1/2/3, in vivo loss-of function experiments that simultaneously remove Fgfr1/2/3 were conducted to explore the potential contribution of FGFR signaling in bSC dendritogenesis. Specifically, Fgfr1/2/3 was deleted from a small number of bSCs by using IUE to introduce icre (nuclear cre) and tdTomato-expressing constructs into cortical layer IV precursor cells of Fgfr1/2/3 triple conditional embryos (Fgfr1^flox/flox;Fgfr2^flox/flox;Fgfr3^flox/flox; Fig. 1A). Control tdTomato-positive bSCs were generated by electroporating a tdTomato-expressing construct with the IUE procedure.

Figure 1.

Eliminating Fgfr1/2/3 perturbed activity-guided dendritogenesis processes in vivo. A, Schematic representation of experimental procedures. B–G, Examples of original images and computer-aided reconstructions. B1–G1 show the locations of barrels (dashed lines) and reconstructed neurons (white arrows). II–V, Cortical layers. The projected images from each confocal image stack are shown in B2–G2. B3–G3 show the traced images of neurons in B2–G2. B4–G4 show color-coded segments according to their branch orders. H–J, Quantitation of total length (H), branch points (I), and mean length (J) of dendrites. K–N, Summaries of segment length (K, M) and number (L, N) per branch order. Cell numbers and animal numbers for each group are as listed in the figure, and there are 28 cells/4 animals in the P6-LOF group. One-way ANOVA with post hoc Tukey's multiple-comparisons test was conducted for H to J. Two-way ANOVA with post hoc Tukey's multiple-comparisons test was conducted for K to N. The statistical analysis (#) compared the indicated groups with control (Ctrl) group with younger age: #p < 0.05; ###p < 0.001. The statistical analysis (*) compared the indicated groups: *p < 0.05; **p < 0.01; ***p < 0.001. Original data values and detailed statistical comparisons for this part of the figure are contained in extended data file Figure 1-1.

Figure 1-1

Original data values and detailed statistical comparison for Figure 1. Download Figure 1-1, XLSX file ^{(37.3KB, xlsx)}

bSCs project the majority of their dendrites toward the barrel hollow, where they form synaptic connections with their corresponding TCAs. The polarized dendritic pattern of bSCs is established before P6 and is followed by substantial addition and outgrowth of dendritic segments (Espinosa et al., 2009; Matsui et al., 2013; Mizuno et al., 2014). The dendritic morphologies of individual tdTomato-positive bSCs were reconstructed in three dimensions (Fig. 1B–G) and analyzed for the following characteristics: total dendritic length, total branch point number, average dendritic length, segment number/length per branch order, and polarity (see Materials and Methods). We found that control bSCs acquired a polarized pattern by P6 (Huang and Lu, 2017) and progressively increased their total dendritic length (Fig. 1H), similar to previous studies (Espinosa et al., 2009; Ashby and Isaac, 2011). This increase in total dendritic length is the result of both developmental additions of branch point numbers (Fig. 1I) and the augmentation of mean dendritic length (Fig. 1J). The progressive increase of mean dendritic length during development for most branch orders suggests that the overall dendritic coverage of bSCs expands from P6 to P15 (Fig. 1K). Interestingly, bSC dendritic segment complexity increases substantially from P6 to P9, with no further increase after P9 (Fig. 1I,L). For dendritic segments, de novo branches were added at multiple branch orders between P6 and P9 (p < 0.0001; Fig. 1L). Together, our detailed morphological characterization revealed at least three distinctive phases of dendritic morphogenesis, as follows: (1) bSCs orient their dendrites toward TCAs by the end of the first postnatal week, (2) bSCs add de novo dendritic segments and reach their final dendritic branch pattern by P9, and (3) individual segments get progressively longer throughout the first 2 postnatal weeks.

FGFR1/2/3 LOF neurons deviate from this pattern of dendritic morphogenesis as they develop (Fig. 1H–J,M–N). FGFR1/2/3 LOF bSCs display normal total dendritic length, branch points, and mean segment length at P6 (Fig. 1H–J). By P9, FGFR1/2/3 LOF bSCs have significantly more branch points (p = 0.0037), but total dendritic length is similar to that in P9 controls (p = 0.4928). Surprisingly, between P9 and P15 in FGFR1/2/3 LOF bSCs, the branch point number (p < 0.0001; Fig. 1I) is significantly reduced. At P15, FGFR1/2/3 LOF bSCs are less complex than P15 control bSCs with regard to total dendritic length (p = 0.0001; Fig. 1H) and dendritic branch numbers (p = 0.0274; Fig. 1I). The mean dendritic length of FGFR1/2/3 LOF bSCs is comparable to that of control bSCs at P6 (p = 0.9795), P9 (p = 0.1298), and P15 (p > 0.9999; Fig. 1J). There is also no significant difference in dendritic polarities between P15 control bSCs (0.7935 ± 0.0221) and FGFR1/2/3 LOF bSCs (0.7691 ± 0.0192; p = 0.9586). Our finding supports the endogenous role of FGFR1/2/3 in regulating dendritic segment addition from P6 to P9 and also in maintaining the dendritic complexity after P9.

To examine the temporal profiles of Fgfr1/2/3 mRNA expression during barrel map formation, real-time PCR experiments were conducted with P3–P9 S1 cortex (Fig. 2A,B) using the experimental procedures described in Huang et al. (2017). The threshold cycle value of real-time PCR amplification curve (Ct value; a lower Ct value corresponds to higher expression) was used to represent relative expression. Ct values of the Fgfr1/2/3-b isoforms (Ct value range, 28–31; Fig. 2A) are much higher than those of the Fgfr1/2/3-c isoforms (Ct value range, 22–25; Fig. 2B). These data suggest that Fgfr1/2/3-c isoforms are much more abundant compared with the b isoforms in the developing S1 cortex. Moreover, Fgfr1b (r = −0.7913), Fgfr2b (r = −0.7930), Fgfr2c (r = −0.8397), and Fgfr3c (r = −0.9135) expression levels increase, while Fgfr3b (r = 0.9953) and Fgfr1c (r = 0.953) decrease from P3 to P9 (Fig. 2A,B). Together, our results suggest that Fgfrs are dynamically expressed in developing somatosensory cortex. The dendritic deficits in Fgfr1/2/3 LOF bSCs support a physiological role of FGF–FGFR signaling in instructing bSC dendritogenesis.

Figure 2.

Developmental expression of FGFs and FGFRs. A–C, The mRNA expression of the b splicing variant of Fgfr1/2/3 (A); the c splicing variant of Fgfr1/2/3 (B); as well as Fgf7, Fgf9, Fgf10, and Fgf22 (C) were measured at the indicated developmental stage. The threshold cycle values of the real-time PCR amplification curve [Ct value (the lower the Ct value corresponds to a higher expression level)] are plotted to represent the relative expression of each gene. Pearson correlation coefficients were calculated to examine the correlation between gene expression level and developmental stage. Original data values and detailed statistical comparisons for this part of the figure are contained in extended data file Figure 2-1.

Figure 2-1

Original data values and detailed statistical comparison for Figure 2. Download Figure 2-1, XLSX file ^{(14.6KB, xlsx)}

Several FGFs and FGFRs are upregulated by synchronized glutamate transmission

Our 3D reconstruction studies point to the involvement of FGFR1-3 in dendritogenesis after P6. This timing coincides with the stage when there is a substantial increase in spontaneous activity due to synaptogenesis and augmentation of synaptic strength (AMPAfication; Mierau et al., 2004; Allène et al., 2008; Yang et al., 2009; Unichenko et al., 2015). Our previous studies found that deleting mGluR5 in cortical glutamatergic neurons results in an increase of calcium-permeable AMPAR in the thalamocortical synapses (Huang and Lu, 2017). Interestingly, mRNA levels of Fgf 7/9/10/22 and Fgfr1, Fgfr3 are increased in the S1 cortex of mGluR5 knock-out mice (Huang and Lu, 2017). Here we also found dynamic regulation of Fgf7/9/10/22 expression from P3 to P9. The abundance of Fgf9 (r = −0.9963), Fgf10 (r = −0.6873), and Fgf22 (r = −0.6058) mRNAs gradually increased, while the Fgf7 (r = 0.8361) mRNA level gradually decreased (Fig. 2C). Hence, we hypothesized that neural activity regulates FGF/FGFR expression in developing cortical neurons and modulates dendritic morphogenesis.

To test this hypothesis, primary cortical cultured neurons at DIV7 were treated with 30 mm KCl to trigger depolarization and glutamate release. Fgf7/9/10/22 and Fgfr1/2/3b/c mRNA levels were examined after 1–24 h treatment. We found that 4 h of KCl treatment significantly upregulated Fgf7 (p = 0.0184), Fgf9 (p = 0.0003), and Fgf10 (p = 0.0003; Fig. 3A). Fgf7/9/10 mRNA upregulation lasted at least until 24 h (the last time point examined). KCl treatment also upregulated Fgfr1–3b and c splicing variants with different temporal dynamics: Fgfr2b/c were significantly upregulated after 2 h, while Fgfr1b/c were significantly upregulated only after 8 h (Fig. 3A,B).

Figure 3.

Glutamate transmission is required for KCl-induced upregulation of Fgf7, Fgf9, and Fgf10 mRNA expression. A–C, Time course of Fgf7, Fgf9, Fgf10, and Fgf22 mRNA expression (A), b splicing variant of Fgfr1/2/3 (B), and c splicing variant of Fgfr1–3 (C) were measured at the indicated time points after KCl (30 mm) treatment in DIV7 primary cortical neurons. An unpaired t test was used to assess statistical significance. In A, the statistical analysis indicated by black (Fgf7), blue (Fgf10), red (Fgf9), and purple (Fgf22) symbols refers to the KCl-treated group compared with the control (Ctrl) group at that time point. In B, the statistical analysis indicated by black (Fgfr1b), red (Fgfr2b), and blue (Fgfr3b) symbols referred to the KCl-treated group compared with the Ctrl group at that time point. In C, the statistical comparisons indicated by black (Fgfr1c), red (Fgfr2c), and blue (Fgfr3c) symbols compared the KCl-treated group and the Ctrl group at the indicated time point. *p < 0.05; **p < 0.01; ***p < 0.001. Primary cortical neurons were pretreated with an NMDAR antagonist (d-APV, 50 μm), an AMPAR antagonist (NBQX, 10 μm), an mGluR5 antagonist (MPEP, 2 μm), an extracellular calcium chelator (EGTA, 1 mm), nifedipine (l-type calcium channel inhibitor, 10 μm), or DMSO (vehicle of nifedipine, 0.01%) for 30 min, which was followed with KCl (30 mm) treatment for 8 h. D–F, Real-time PCR was conducted to measure the mRNA expression of Fgf7 (D), Fgf9 (E), and Fgf10 (F). One-way ANOVA with post hoc Tukey's multiple-comparisons test was used to assess statistical significance. * refers to the indicated group compared with control group: *p < 0.05; **p < 0.01; ***p < 0.001. # refers to the indicated group compared with KCl-treated neurons: #p < 0.05; ##p < 0.01; ###p < 0.0001. Original data values and detailed statistical comparisons for this part of the figure are contained in extended data file Figure 3-1.

Figure 3-1

Original data values and detailed statistical comparison for Figure 3. Download Figure 3-1, XLSX file ^{(71.1KB, xlsx)}

KCl treatment increases extracellular calcium influx through l-type calcium channels and enhances glutamate neurotransmission (Wang and Zhuo, 2012). To determine whether calcium influx or glutamate receptors are required for KCl-induced Fgf-Fgfr mRNA changes, we pretreated cortical neurons with l-type calcium channel blocker nifedipine (10 μm), AMPAR antagonist NBQX (10 μm), NMDAR antagonist d-APV (50 μm), or mGluR5 antagonist MPEP (2 μm) before KCl treatment. We found that blocking l-type calcium channels or NMDARs significantly attenuated Fgf7/9/10 upregulation after 8 h of KCl treatment, while inhibiting AMPAR or mGluR5 had no impact (Fig. 3D–F). Interestingly, Fgf7 upregulation, but not Fgf9/10 upregulation, was also affected by an extracellular calcium blocker, EGTA (1 mm; p = 0.0206).

To explore whether Fgf-Fgfr mRNA levels can be regulated by neural activity in vivo, we further examined whether increasing glutamate transmission can upregulate Fgf/Fgfr expression in the developing S1 cortex. P6 pups were injected once intraperitoneally or subcutaneously with KA (a glutamate analog; 1.5 mg/10 ml/kg) or saline. We found that mRNA expression of Fgf9 (p = 0.0208), Fgf10 (p = 0.0133), Fgfr2c (p = 0.004), and Fgfr3b (p = 0.0162) in the S1 cortex was significantly upregulated in KA-treated neonatal mice 6 h after KA administration (Table 1, Fig. 4). We did not observe significant changes in Fgf7 (p = 0.6407), Fgf22 (p = 0.5531), Fgfr1b (p = 0.8646), Fgfr1c (p = 0.295), Fgfr2b (p = 0.5099), and Fgfr3c (p = 0.7034) expression levels (Table 1). Bdnf (p = 0.0031) was also upregulated (Table 1), as previously reported (Zafra et al., 1990; Rocamora et al., 1996). Together, these data suggest that mRNA expression of Fgfr2c, Fgfr3b, Fgf9, and Fgf10 is upregulated in response to enhanced glutamatergic transmission in vivo.

Table 1.

Summary of normalized mRNA levels of Fgfs and Fgfrs in S1 cortex

Gene	Ctrl group (n = 5)	KA group (n = 5)	p value
Fgf7	1 ± 0.13	1.11 ± 0.12	0.6407
Fgf10	1 ± 0.08	1.32 ± 0.05	0.0133*
Fgf22	1 ± 0.07	1.11 ± 0.15	0.5531
Fgf9	1 ± 0.02	1.34 ± 0.12	0.0208*
Fgfr1 b	1 ± 0.05	1.04 ± 0.17	0.8646
Fgfr1 c	1 ± 0.08	1.29 ± 0.15	0.295
Fgfr2 b	1 ± 0.12	0.91 ± 0.11	0.5099
Fgfr2 c	1 ± 0.08	2.18 ± 0.28	0.004**
Fgfr3 b	1 ± 0.06	1.32 ± 0.09	0.0162*
Fgfr3 c	1 ± 0.05	1.04 ± 0.05	0.7034
Bdnf	1 ± 0.10	2.82 ± 0.41	0.0031**

Somatosensory cortex samples were collected after KA injection for 6 h and subjected to quantitative PCR experiments. The statistical analysis refers to the control (Ctrl) group compared with the KA group. Bold highlights the p values that are statistically significant. Original data values and detailed statistical comparisons for this part of the figure are contained in extended data file Figure 4-1 available at https://doi.org/10.1523/JNEUROSCI.1174-17.2017.f4-1.

*p < 0.05,

**p < 0.01 by unpaired t test.

Figure 4.

KA treatment upregulated Fgf9, Fgf10, Fgfr2c, and Fgfr3b mRNA expression in S1 cortex. Wild-type pups were administered KA at postnatal day 6. S1 cortices were collected after 6 h treatment, and real-time PCR was conducted to measure mRNA expression (n = 5 for each group). An unpaired t test was used to assess statistical significance. The statistical analysis compared the KA-treated group to the control (Ctrl) group: *p < 0.05; **p < 0.01. Original data values and detailed statistical comparisons for this part of the figure are contained in extended data file Figure 4-1.

Figure 4-1

Original data values and detailed statistical comparison for Figure 4 and Table 1. Download Figure 4-1, XLSX file ^{(15.8KB, xlsx)}

KA promotes dendritogenesis in bSCs through FGFR in vivo

The fact that glutamate transmission modulated the dendritogenesis of bSCs (Datwani et al., 2002; Espinosa et al., 2009; Mizuno et al., 2014; Ballester-Rosado et al., 2016) motivated us to investigate whether the massive glutamate transmission elicited by KA administration promoted dendritogenesis in an FGFR-dependent manner. Simultaneous removal of Fgfr1/2/3 was achieved as described in Figure 1 followed by KA injection at P6 (Fig. 5A). Afterward, detailed dendritic morphology was examined at P7 (Fig. 5B). We first examined the impacts of KA-induced seizures on bSC dendritic morphology. Scattered bSCs in the S1 cortex of FGFR1/2/3 conditional mice were labeled with tdTomato using IUE as described above, and their dendritic morphology was examined 1 d after KA administration at P6. We found that KA injection significantly increased total dendritic length (p < 0.0001; Fig. 5C) and branch point number (p < 0.0001; Fig. 5D) but did not alter average segment length (p = 0.3399; Fig. 5E). Branch order analysis revealed that KA significantly increased segment number (p < 0.0001; Fig. 5F) but had no impact on the segment length of any branch orders (p = 0.5492; Fig. 5G). KA administration did not affect the polar ratio (p = 0.3124; Fig. 5J), it promoted dendritic growth both inside the barrel (p < 0.0001; Fig. 5H) and outside the barrel (p < 0.0001; Fig. 5I). Notably, KA failed to induce dendritic outgrowth in Fgfr1/2/3 LOF bSCs. These results suggest that FGFR1/2/3 signaling can be upregulated to promote dendritic outgrowth by excessive neural activity.

Figure 5.

KA treatment promoted bSC dendritogenesis within 24 h in an FGFR1/2/3-dependent manner. A, Schematic representation shows the experimental procedures and developmental stages for IUE and KA treatment. B, Examples of computer-aided reconstructions. C–E, The total dendritic length (C), branch point number (D), and mean dendritic length (E) are plotted. F, G, Summary of number (F) and length (G) per branch order. Two-way ANOVA with post hoc Tukey's multiple-comparisons test was conducted for F and G. The statistical analysis indicated by the red asterisk refers to the KA group compared with the control (Ctrl) group. The statistical analysis indicated by the purple asterisk refers to the KA+FGFR LOF group compared with the KA group: ***p < 0.001. H–J, Quantitative analysis of dendritic length inside (H) and outside (I) of the barrel that was used to calculate dendritic polarity (J). One-way ANOVA with post hoc Tukey's multiple-comparisons test was conducted for C–E and H–J. The statistical analysis compared the indicated groups: ***p < 0.001. Original data values and detailed statistical comparisons for this part of the figure are contained in extended data file Figure 5-1.

Figure 5-1

Original data values and detailed statistical comparison for Figure 5. Download Figure 5-1, XLSX file ^{(48.8KB, xlsx)}

Postnatal FGF9 and FGF10 overexpression promoted dendritic growth in vivo

If Fgf/Fgfr upregulation induced by KA leads to dendritic extension, FGF9 and FGF10 overexpression at P6 should result in a similar dendritic extension. To overexpress Fgf9 or Fgf10 in bSCs after P5/P6, we used the combination of IUE and an inducible Cre/loxP approach (Fig. 6A). Specifically, we used the Cre-On system to activate Fgf9 and Fgf10 expression by using inducible Cre (ER^T2-Cre-ER^T2) and Cre-dependent activation constructs (Fig. 6A; Matsuda and Cepko, 2007; Atasoy et al., 2008; Cardin et al., 2010). A Cre-On expression cassette was engineered by placing Fgf9 or Fgf10 cDNA into the double-floxed inverted orientation (DIO) construct (see Materials and Methods). Upon Cre-mediated recombination, the orientation is reversed to allow Fgf9 or Fgf10 transcription driven by the EF1α promoter. We conducted IUE to introduce DIO-Fgf9/DIO-Fgf10 and DIO-tdTomato (reporter) constructs into wild-type embryos at E14.5 to target bSCs (Fig. 6A). ER^T2-Cre-ER^T2 was activated by tamoxifen injection (100 mg/10 ml/kg, i.p.; single injection) at P5 to initiate Fgf9 and Fgf10 transcription in cortical neurons (Fig. 6A). Two days after tamoxifen treatment, the impact of Fgf9 or Fgf10 overexpression on bSCs was examined through a detailed morphological analysis of tdTomato-positive spiny stellate cells at P7 (Fig. 6B–D).

Figure 6.

FGF9 and FGF10 expression promoted dendritogenesis of layer IV cortical neurons. A, Schematic representations of the electroporation and tamoxifen treatment protocols that were used to express FGF9 and FGF10. B–D, Examples of original images and computer-aided reconstructions. B1–D1 show the locations of barrels (dashed lines) and reconstructed neurons (white arrows). II–V, Cortical layers. The projection images from confocal image stacks are shown in B2–D2. B3–D3 show the traced images of neurons in B2–D2. B4–D4 show color-coded segments according to their branch orders. E–G, Quantitation of total length (E), branch point numbers (F), and mean length (G) of dendrites. H, I, Summaries of segment number (H) and length (I) per branch order. Two-way ANOVA with post hoc Bonferroni's multiple-comparisons test was conducted for H and I. The statistical analysis indicated by the red asterisk refers to the FGF9 group compared with the control (Ctrl) group. The statistical analysis indicated by the blue asterisk refers to the FGF10 group compared with the Ctrl group. J–L, Quantitative analysis of dendritic length inside (J) and outside (K) the barrel that was used to calculate dendritic polarity (L). A Mann–Whitney test was conducted for E, F, G, J, K, and L. The statistical analysis compared the indicated groups: *p < 0.05; **p < 0.01; ***p < 0.001. Original data values and detailed statistical comparisons for this part of the figure are contained in extended data file Figure 6-1.

Figure 6-1

Original data values and detailed statistical comparison for Figure 6. Download Figure 6-1, XLSX file ^{(34.3KB, xlsx)}

We found that Fgf9 and Fgf10-expressing bSCs displayed significantly more dendritic outgrowth compared with their control neurons. Total dendritic length (p = 0.0007 and p = 0.034; Fig. 6E) and dendritic branch point number (p = 0.0037 and p = 0.0081; Fig. 6F) were significantly increased, while average dendritic length (p = 0.4979; p = 0.9674; Fig. 6G) was unchanged. The segment numbers per branch order were significantly increased in Fgf9-overexpressing (p = 0.0006) and Fgf10-overexpressing (p = 0.01) bSCs (Fig. 6H). Neither FGF9 (p = 0.7536) nor FGF10 (p = 0.6535) had significant impacts on segment length per branch order (Fig. 6I). Regarding polarity, FGF9 or FGF10 overexpression significantly increased dendritic growth inside the barrel (p = 0.0025; p = 0.0095; Fig. 6J), but not outside the barrel (p = 0.0581; p = 0.4857; Fig. 6K); thus, bSCs retained their polarized morphology (p = 0.6115; p = 0.7375; Fig. 6L). Together, we found that FGF9 or FGF10 overexpression is sufficient to promote dendritic growth of bSCs.

Discussion

Here we first show that the dendritic pattern of bSCs is established in several distinctive phases during the first 2 postnatal weeks. Fgfr1/2/3 are dynamically expressed in S1 cortex, and deleting these receptors in bSCs results in a reduction in dendritic branch numbers and total dendritic length from P9 to P15, while control bSCs maintain their dendritic branch pattern and increase their total dendritic length by the elongation of individual segments. Interestingly, mRNA levels of several Fgfs and Fgfrs can be upregulated by glutamate transmission. Glutamate transmission-driven dendritic growth depends on FGFR1/2/3, while FGF9 or FGF10 overexpression results in a similar dendritic growth. Together, our data suggest that FGF–FGFR signaling can be regulated by glutamate transmission to both establish and maintain the dendritic patterns of bSCs.

FGFRs in bSCs are required for activity-induced dendritogenesis during whisker-barrel map formation

Our dendritogenesis study revealed distinctive phases of dendritic patterning. First, as reported by Espinosa et al. (2009), bSCs acquire a TCA-oriented pattern by P6. Our previous studies show the importance of NGF–TrkA signaling in the establishment of the dendritic polarity of bSCs (Huang and Lu, 2017). Next, from P6 to P9, branch numbers increase more than twofold and then remain stable from P9 to P15. The segment addition at multiple branch orders from P6 to P9 indicates de novo branch addition (Fig. 1L). Despite the establishment of a basic dendritic pattern by P9, individual dendritic segments at almost all branch orders, except the first order, continue to elongate from P9 to P15 (Fig. 1K).

At P6, FGFR1/2/3 LOF bSCs (0.8264 ± 0.0161, p = 0.9798) acquire dendritic polarity similar to control bSCs (0.8479 ± 0.0267) and exhibit no detectable deficits. At P9, FGFR1/2/3 LOF bSCs display a small but significant increase in dendritic branch point numbers (Fig. 1I). However, this increase in dendritic complexity is transient. By P15, dendritic segment numbers in FGFR1/2/3 LOF bSCs are reduced compared with both P15 control bSCs and P9 FGFR1/2/3 LOF bSCs. The substantial reduction of FGFR1/2/3 LOF bSC dendritic branch point numbers from P9 to P15 argues (Fig. 1N) for the importance of FGFR1/2/3 bSCs in maintaining the established dendritic pattern. In other words, these data reveal the importance of neuronal/dendritic maintenance immediately after the establishment of dendritic morphogenesis.

FGFR signaling is commonly initiated by ligand-dependent dimerization of FGFRs, followed by the activation of several signaling cascades, including& extracellular signal-regulated kinases 1/2, phosphatidylinositol 3-kinase-AKT, protein kinase C, and small GTPases (Eswarakumar et al., 2005; Iwata and Hevner, 2009; Dorey and Amaya, 2010; Guillemot and Zimmer, 2011; Chuang et al., 2015). Our understanding of the function of each specific FGF ligand and FGFR remains limited. This is partially due to the highly redundant functions of the 22 FGF family members and complex interactions between ligands and b/c splicing variants of FGFR1/2/3. The high sequence similarity between FGFRs together with the promiscuity of FGF binding implies redundancy within the family members. However, targeted gene inactivation of different FGF and FGFR members yields a specific phenotype for each knockout. For example, FGF7 and FGF22 independently regulate inhibitory and excitatory synapse formation, respectively (Terauchi et al., 2010; Dabrowski et al., 2015). Our mosaic analysis of FGFR1/2/3 LOF bSCs within the wild-type environment supports the cell-autonomous requirement of FGFR1/2/3.

In addition to the studies revealing the endogenous role of FGF–FGFR signaling in dendritogenesis of bSCs during barrel map formation, we also find that FGFR1/2/3 are required for KA-induced dendritic outgrowth (Fig. 5). Taking a complementary approach of overexpressing FGFR ligands by using the Cre-ER/loxP system, we were able to reveal that FGF9 or FGF10 expression for 1–2 d had a dramatic impact on bSC dendritic outgrowth after P5/P6 (Fig. 6). Together, our data provide strong support for the involvement of FGF/FGFR in mediating neuronal activity-dependent dendritogenesis. Several important questions remain to be answered in future studies: which FGFRs are the key receptor; which FGFs or FGFs activate FGFRs on bSCs; what cells secrete FGFs; and what are the downstream mechanisms underlying FGFR-mediated dendritic formation and maintenance.

Neural activity regulated Fgf/Fgfr mRNA levels

During embryonic development, the expression of each individual FGF is tightly regulated regarding their expression levels, timing, and location (Ford-Perriss et al., 2001; Yaguchi et al., 2009). Little is known on what controls Fgf/Fgfr mRNA expression in postmitotic neurons. Our studies with primary cortical neurons show that neuronal depolarization (KCl treatment) upregulated Fgf/Fgfr mRNA levels very quickly in an NMDAR- and calcium-dependent manner. Interestingly, the activity regulates different Fgf/Fgfr mRNAs with distinct dynamics. It has been shown that epileptiform activity upregulates Fgf/Fgfr (Gómez-Pinilla et al., 1995; Kondratyev et al., 2002; Kim et al., 2010). Both Fgf3 and Fgfr1 transcription was increased 1 h after seizure onset in hippocampi of the KA-induced seizure model (Hermey et al., 2013). FGF2 protein expression was increased in whisker stimulation (Whitaker et al., 2007).

Several synaptic events are notable during the transition from the first to the second postnatal week for thalamocortical connections: synaptic surface AMPARs increase (AMPAfication); NMDAR subunits switch from NR2B dominant to NR2A dominant; and the critical period for whisker barrel map plasticity and synaptic plasticity closes (Wu et al., 2011). By combining whole-cell recording and single-cell resolution 2P glutamate uncaging, Ashby and Isaac (2011) mapped out local functional connectivity in layer IV barrel cortex during P3–P12 on a daily basis. They found that during the first postnatal week, the connectivity between bSCs was ∼4%. Interestingly, between P8 and P9, the connectivity between bSCs increases abruptly, approximately threefold, which persisted until P12, the oldest stage recorded. Before P8, bSC dendrites are almost entirely aspiny (Ashby and Isaac, 2011). However, from P8 to P13, dendritic spine counts for bSCs increase ∼250-fold. Sensory experience from whiskers is required for the increase in intraconnectivity but not for spinogenesis or total dendritic length. Thus, it is plausible that the rise in intracortical glutamatergic connectivity enhances FGF/FGFR signaling after P8/P9 to regulate and maintain dendritic morphology.

In this study, we provide the first in vivo demonstration for neural activity regulation on Fgf9, Fgf10, Fgfr2c, and Fgfr3b mRNA levels in the cortex. The precise mechanism underlying activity-regulated Fgf/Fgfr transcription remains to be determined. Using stromal cells, Chuang et al. (2006) showed that prostaglandin E2 upregulated Fgf9 mRNA expression via activating ELK1 (Chuang et al., 2006). Interestingly, ELK1 can be activated by increased glutamate transmission in either cortical neurons or brain slices (Vanhoutte et al., 1999; Madabhushi et al., 2015). Regarding Fgf10 transcription, studies using mouse embryos, trigeminal ganglion, and chick embryonic limb buds show that Tbx5 and POU transcriptional factors can upregulate Fgf10 expression by binding to its promoter region (Agarwal et al., 2003; Cox et al., 2006; Nishimoto et al., 2015).

Cortical circuit plasticity triggered by sensory/motor experience or learning has been documented in many species using different paradigms (Globus and Scheibel, 1967; Hensch, 2005). In the barrel cortex, the impacts of whisker trimming or whisker lesions on bSC dendritic morphology have been studied mostly using classic Golgi staining and 2D reconstruction of dendritic orientation and overall dendritic coverage (Steffen and Van der Loos, 1980; Lee et al., 2009). Here we triggered massive neurotransmission by applying KCl or kainic acid, which are very strong paradigms. It remains to be determined whether natural neural activity, for example through whisking or whisker stimulation, will similarly upregulate FGF/FGFR signaling and, if so, whether the dendritic changes induced by this level of neural activity will be long lasting.

Interestingly, saline-treated control bSCs exhibited more branch points and total length than the values extrapolated from naive P6 and P9 controls (compare Figs. 1H,I, 5C,D). Saline-treated pups encountered different sensory experiences than naive pups ∼20 h before being killed. As a part of the experimental procedures, saline-treated pups underwent maternal deprivation for 3–4 h, experienced handling, and were also housed together with their KA littermates undergoing seizures for 3–4 h (see Materials and Methods). It remains to be determined whether handling and extra sensory stimuli caused the differences noted between the two groups. Interesting, the values for saline-treated P7 FGFR1/2/3 LOF bSCs are in the range of expected values from naive P6 and P9 FGFR1/2/3 LOF bSCs. The differences between control and FGFR1/2/3 LOF bSCs are in the same direction we observed for KA-treated pups, providing additional support for our hypothesis that activity-driven dendritic morphological changes require FGFRs. Alternatively, the disparity between dendritic data from Figure 1 and Figure 5 could suggest a transient dendritic outgrowth between P6 and P7 that is refined between P7 and P9. Such transient overgrowth failed to occur in FGFR1/2/3 LOF neurons.

In summary, our data here suggest that FGF–FGFR signaling can be enhanced by glutamate transmission to promote dendritic arborization and physiological levels of glutamate transmission that are sufficient for FGF–FGFR signaling to maintain dendritic structures. Dysregulated FGF/FGFR expression (Gaughran et al., 2006; Turner et al., 2006, 2012, 2016; Terwisscha van Scheltinga et al., 2010; Williams and Umemori, 2014) has been observed in depression, anxiety, and schizophrenia. Thus, abnormal FGF/FGFR levels in these neurological disorders may account for the abnormal dendritic arborization patterns observed in these diseases (Kulkarni and Firestein, 2012).