Genetic deletion of fibroblast growth factor 14 recapitulates phenotypic alterations underlying cognitive impairment associated with schizophrenia

Cognitive processing is highly dependent on the functional integrity of gamma-amino-butyric acid (GABA) interneurons in the brain. These cells regulate excitability and synaptic plasticity of principal neurons balancing the excitatory/inhibitory tone of cortical networks. Reduced function of parvalbumin (PV) interneurons and disruption of GABAergic synapses in the cortical circuitry result in desynchronized network activity associated with cognitive impairment across many psychiatric disorders, including schizophrenia. However, the mechanisms underlying these complex phenotypes are still poorly understood. Here we show that in animal models, genetic deletion of fibroblast growth factor 14 (Fgf14), a regulator of neuronal excitability and synaptic transmission, leads to loss of PV interneurons in the CA1 hippocampal region, a critical area for cognitive function. Strikingly, this cellular phenotype associates with decreased expression of glutamic acid decarboxylase 67 (GAD67) and vesicular GABA transporter (VGAT) and also coincides with disrupted CA1 inhibitory circuitry, reduced in vivo gamma frequency oscillations and impaired working memory. Bioinformatics analysis of schizophrenia transcriptomics revealed functional co-clustering of FGF14 and genes enriched within the GABAergic pathway along with correlatively decreased expression of FGF14, PVALB, GAD67 and VGAT in the disease context. These results indicate that Fgf14−/− mice recapitulate salient molecular, cellular, functional and behavioral features associated with human cognitive impairment, and FGF14 loss of function might be associated with the biology of complex brain disorders such as schizophrenia.


INTRODUCTION
Psychiatric diseases such as schizophrenia, depression and bipolar disorder are associated with cognitive deficits thought to arise from an imbalance between the excitatory and inhibitory (E/I) tone in cortical circuits. 1,2 Fast-spiking GABAergic parvalbumin (PV) interneurons have a fundamental role in maintaining E/I balance, controlling excitability and shaping the synaptic plasticity of principal neurons. 3 Even though disrupted function of other types of inhibitory neurons has been associated with the etiology of brain disorders, changes in PV neurons are one of the most consistent associations. In animal models, reduced function of PV neurons results in desynchronized network activity, decreased gamma frequency oscillations and cognitive deficits, phenotypes that mimic clinical presentation and post-mortem tissue studies of virtually all human brain disorders associated with disrupted cognition. 1,[4][5][6][7][8][9][10] In schizophrenia and bipolar patients, the number of PV interneurons and the expression level of molecular components of GABAergic synapses, such as glutamic acid decarboxylase 67 (GAD67) and vesicular gamma-amino-butyric acid (GABA) transporter (VGAT), are found decreased in post-mortem brains leading to the 'GABA hypofunction hypothesis' as a potential etiology. 1,[11][12][13][14][15][16] Yet, the molecular understanding of how such a detrimental loss of the GABAergic system might lead to corrupted cortical networks manifesting in disease, remains poorly explored.
Highly expressed in the central nervous system, FGF14 is an accessory protein of voltage-gated Na+ (Nav) channels at the axonal initial segment (AIS), [30][31][32][33][34][35][36] a regulator of neuronal excitability, 30,37-39 a presynaptic organizer of glutamatergic synapses, 40,41 a scaffolding molecule for kinase signaling pathways 33,34,42,43 and a regulator of synaptic plasticity. 41,44,45 At the circuitry and behavioral level, genetic deletion of FGF14 in Fgf14 − / − mice results in presynaptic structural deficits of CA3-CA1 hippocampal synapses, decreased long-term potentiation, and cognitive deficits, aberrant responses to epileptic agents and decreased threshold for seizure induction. 41,44,45 Furthermore, Fgf14 − / − mice exhibit abnormal locomotor activity explained by a reduced response to dopamine receptor D2 agonists in the basal ganglia. 45 Both phenotypes in the hippocampus and the basal ganglia could imply an impaired GABAergic tone supporting the emerging view of a tight correlation between the dopamine and the GABAergic systems. 46,47 Yet, evidence of FGF14 control of GABAergic inhibitory transmission is lacking.
Using a combination of confocal microscopy, patch-clamp electrophysiology, in vivo local field potential recordings and behavioral studies, we discovered that genetic deletion of Fgf14 in rodents leads to a reduced number of PV interneurons, decreased expression of GAD67 and VGAT, and reduced GABAergic transmission in CA1 pyramidal neurons associated with impaired gamma oscillations and working memory. Further, bioinformatics analysis of schizophrenia transcriptomics confirms FGF14 functional clustering with GABAergic synaptic signaling and identifies genetic covariance of FGF14, PVALB, GAD67 and VGAT in the disease, supporting Fgf14 − / − mice as an attractive model to interrogate the biology of complex brain disorders associated with disrupted cognitive circuitry such as schizophrenia.

MATERIALS AND METHODS Animals
The primary Fgf14 −/− colony was maintained at the animal facilities of the University of Texas Medical Branch and of the University of Torino following approved protocols. Genotypes were confirmed by in-house PCR analysis or by Charles River Laboratories International (Houston, TX, USA). Description of the animal husbandry is given in the Supplementary Material.

Immunofluorescence
Brain tissue derived from Fgf14 − / − and Fgf14 +/+ 4-to 5-month-old male littermates was processed for immunofluorescence staining using either 4% paraformaldehyde or liquid nitrogen vapor fixation followed by permeabilization, blocking, and primary and Alexa-conjugated secondary antibody staining as previously described 48 and outlined in the Supplementary Material.

Data acquisition and image analysis
Multichannel confocal images were acquired using a Zeiss LSM-510 META (Carl Zeiss Microscopy, Jena, Germany) confocal microscope as previously described. 49 Detailed information on image quantification can be found in the Supplementary Material.

Western blotting
Tissue homogenate preparation and western blotting were carried out as previously described 50 and outlined in the Supplementary Material.

Whole-cell patch-clamp electrophysiology
Horizontal hippocampal slices were prepared from 4-to 5-month-old Fgf14 − / − and Fgf14 +/+ male littermates, and whole-cell patch-clamp recordings were performed at room temperature from CA1 pyramidal neurons using standard parameters previously described 51,52 and outlined in the Supplementary Material.

In vivo electroencephalogram
Extracellular field potentials were recorded in freely behaving mice using stainless steel electrodes stereotaxically implanted in the CA1 of the right hippocampus. 53 Spectrograms were calculated using the software NeuroExplorer (Nex Technologies, Madison, AL, USA).

Behavioral tests
Tests were performed in Fgf14 − / − and Fgf14 +/+ 2-to 4-month-old males using the 4-8 version of the eight-arm radial maze. 54 A detailed description of behavioral experiments and analysis are part of the Supplementary Material.

Human studies
As a computational gene co-expression search engine, SEEK (http://seek. princeton.edu/) was applied to identify a gene network co-expressed with FGF14. Functional annotation and pathway enrichment was performed with a built-in Gene Ontology and KEGG terms. Enriched data sets were downloaded from NCBI Gene Expression Omnibus, and the differential gene expression was analyzed by R Limma packages described in the Supplementary Material. We also applied GeneMANIA and STRING to the analysis of pathway enrichment and protein-protein interaction for the coexpressed genes of FGF14 in hippocampus. The method and result are detailed in the Supplementary Material.

RESULTS
Anatomical abnormalities found in the CA1 hippocampal region of post-mortem brains from schizophrenia patients have been identified as an index of the disease severity and treatment responsiveness. 55 These studies along with the reported association between hippocampal PV neurons with cognitive function in the normal brain and in schizophrenia, prompted us to begin our investigations in the CA1 hippocampal region. 15,16,46,47,55 Thus, we first asked whether FGF14 was expressed in PV interneurons in the CA1 region. Confocal imaging confirmed that FGF14 is expressed at the AIS of cells across all CA1 sublayers (Figure 1a), including PV-positive neurons ( Figure 1b). Therefore, we posited whether genetic ablation of Fgf14 might have physiological consequences for these inhibitory neurons.
To test this hypothesis, we first examined whether the overall number of PV interneurons in the CA1 area was affected by genetic loss of FGF14. Cell count quantification based on immunofluorescence staining revealed that the total number of PV-positive cells was significantly reduced in Fgf14 − / − mice compared with wild-type controls (Figures 1c-e; 74.21% ± 5.02, 100% ± 4.89, P o 0.001) with a most pronounced and significant reduction in the stratum oriens (SO; 71.42% ± 6.96, 100% ± 8.10, Po 0.05) and in stratum pyramidalis (SP; 77.94% ± 4.96, 100% ± 7.06, P o0.05; Figure 1f). Notably, the expression level of PV per cell was not significantly different across groups (P = 0.78, n = 4 littermate mice; Supplementary Figures 1F-I); the neuropil composed of PV-positive neurites in either SP or SO was also unaffected by Fgf14 deletion (P = 0.49 and 0.28; Supplementary Figures 1A-E). Furthermore, the total number of cells in the CA1 SP, which primarily includes pyramidal neurons, was also unchanged (P = 0.24; Figures 1j-l). Thus, we concluded that deletion of Fgf14 leads to a cell-and subfield-specific reduction in the number of PV neurons in the CA1 hippocampal region.
PV interneurons are the primary source of GABAergic synapses in the CA1 region. Reduction in GABA synthesis and its synaptic release machinery accompany loss of PV neurons in schizophrenia and other psychiatric disorders. [4][5][6] Consistent with this, GAD67 mean fluorescence intensity in PV soma, the primary source of the enzyme pool, and in puncta across SP was significantly reduced in Fgf14 − / − mice compared with littermate control (89.85% ± 3.08, 100% ± 3.96, P o0.05; and 91.86% ± 0.10, 100% ± 0.08, Po 0.0001, respectively; Supplementary Figures 2A-D and Figures 2a-c). The area per puncta and total puncta number, though, were unchanged (P = 0.10 and 0.88, respectively; Figure 2c). An additional marker and key regulatory protein at presynaptic GABAergic inputs is VGAT, the vesicular transporter that loads presynaptic vesicles with GABA, another marker associated with brain disorders. [56][57][58] Along with GAD67, the mean fluorescence intensity of VGAT in PV soma and its mean content per puncta in  (Figures 2g and h). PV interneurons contribute greatly to the integrity of the GABAergic circuit in the CA1 region, posing the question of whether Fgf14 deletion might result in functional consequences for inhibitory transmission. Using patch-clamp electrophysiology, we recorded spontaneous and miniature inhibitory postsynaptic currents (sIPSCs and mIPSCs, respectively) from visually identified CA1 pyramidal neurons in Fgf14 − / − and wild-type control mice (Figures 3a and f). We found that genetic deletion of Fgf14 led to a rightward shift in the probability distribution of sIPSC frequency (apparent loss in the higher frequency domain) accompanied by a reduction in the largest-and smallest-size synaptic event population (Figures 3c-e; P o0.001, Kolmogorov-Smirnov test). Spontaneous IPSCs are action potential-driven synaptic events and as such represent a compound readout of the firing status of presynaptic interneurons and their neurotransmitter release machinery. To examine GABA release mechanisms independently from interneuron spontaneous firing, we recorded mIPSCs in the presence of tetrodotoxin and found that, similarly to sIPSCs, the frequency distribution histograms showed lowered probability of short inter-event intervals in Fgf14 − / − when compared with wildtype control mice (Figures 3f-h, P o 0.001 with Kolmogorov-Smirnov test). This phenotype was paralleled by a loss of largeand small-amplitude mIPSCs in Fgf14 − / − compared with wild-type control mice (Figures 3f, i and j). In both sIPSCs and mIPSCs, the averaged sIPSC frequency and amplitude between the two groups were not significantly different (Supplementary Figure 3) and no changes in rise and decay time were found across synaptic events (both sIPSCs and mIPSCs) of different genetic groups (Supplementary Figure 4). Thus, genetic deletion of Fgf14 leads to functional changes in the CA1 inhibitory circuitry that support the structural alterations demonstrated in Figures 1 and 2. Moreover, it suggests loss of interneuron firing (shift in sIPSC frequency) and pre-and postsynaptic modifications at GABA synapses (shift in frequency and amplitude distribution of mIPSCs, respectively).
Reduced PV neuron function can desynchronize the CA1 network resulting in reduced gamma oscillations and impaired cognition. Thus, we postulated that gamma oscillations might be impaired upon ablation of Fgf14. To test this, gamma oscillations were recorded in the CA1 SR layer in Fgf14 − / − and control mice by in vivo local field potential (Figure 4a). Spectral analysis within the 30-100 Hz range revealed that Fgf14 − / − mice had a strong reduction in gamma oscillation power (2.94 ± 0.11 μV 2 , n = 7 compared with 7.92 ± 0.17 μV 2 , n = 7 in wild-type mice,  Figure 4d). Thus, consistent with disrupted GABAergic transmission, spectral analysis confirmed that FGF14 is required for the integrity of the cognitive circuitry. The   (Figures 1-4) has been associated with deficits in spatial working memory in animal models and patients afflicted with schizophrenia. 1,58,59 Thus, we next evaluated the spatial working memory performance of Fgf14 − / − mice using the eight-arm maze test. Analysis of the latency to perform the task showed that Fgf14 − / − mice required a longer time to complete the task than wild type (n = 20 wild type and n = 19 Fgf14 − / − , P o 0.001; Figure 4e). Furthermore, revisiting errors (Figure 4f), which are directly linked to working memory performance, were more frequent in Fgf14 − / − mice than in wild types (P o 0.05), corroborating the cellular and functional phenotypes associated with genetic deletion of Fgf14.
To provide translational value to our studies, we examined large transcriptomic data sets from schizophrenia post-mortem tissues deposited in the NCBI Gene Expression Omnibus, seeking genes whose expression might covary with that of FGF14. Through the SEEK-based gene co-expression search engine with built-in functional annotation and pathway enrichment by Gene Ontology terms (Supplementary Table 2) and by KEGG ortholog analysis, we found that FGF14 was enriched within the 'GABAergic synapse' pathway (Supplementary Table 1) and its expression profile correlated with that of PVALB (P = 0.004 in hippocampus; P = 0.0059 in prefrontal cortex), GAD67 (P = 0.0009 in hippocampus; P = 0.0003 in prefrontal cortex) and VGAT (P = 0.04 in     Table 3) in tissue/disease-specific conditions including schizophrenia. Western blot analysis confirmed a significant reduction of GAD67 and VGAT protein expression in the hippocampus (Figures 2g and h) as well as in the prefrontal cortex ( Supplementary Fig.5A, B) in Fgf14 − / − mice compared with controls, providing further correlations between our mouse model and human studies. We subsequently analyzed expression of the FGF14, PVALB, GAD67 and VGAT genes and their correlations in two schizophrenia-enriched data sets and matched controls from the dorsal lateral prefrontal cortex. 60,61 Both data sets showed a significant decreased expression of FGF14, PVALB, GAD67 and VGAT (Figure 5a and Supplementary Figure 6), and a highly significant correlation between FGF14, PVALB, GAD67 and VGAT was found in all samples and in schizophrenia alone with the largest effect size in schizophrenia and controls for GAD67 (Figure 5b). The top enriched pathway such as 'GABAergic synapse' was also confirmed by another pathway enrichment tool, GeneMANIA and the GABAergic gene cluster was identified by STRING as showed in Supplementary Table 4 and Supplementary Figure 8.

DISCUSSION
Here we provide new evidence for FGF14 in maintaining GABAergic activity in the CA1 hippocampal region, an area critical for cognitive function. 65 Genetic deletion of Fgf14 leads to a decrease in the number of PV interneurons and in the expression level of the presynaptic GABAergic markers GAD67 and VGAT.
These changes are associated with reduced inhibitory tone of pyramidal neurons, decreased gamma frequency oscillations and deficits in working memory. Bioinformatics analysis from human transcriptomics identified FGF14 as a component of GABAergic synaptic signaling and revealed a correlated decrease in FGF14, PVALB, GAD67 and VGAT gene expression in schizophrenia post-mortem tissues compared with matched controls. These results provide a new mechanistic role for FGF14, an emerging neuropsychiatric disease-associated gene [17][18][19][20][21][22][23][24][25][26][27] in the context of human brain disorders. The identification of FGF14 immunoreactivity at the AIS of CA1 PV interneurons suggests this protein contributes to interneuron structural and functional diversity 66 and as such might be part of the repertoire of signaling molecules dictating the cell cardinal and definitive specifications of PV neurons in the neural circuitry. 1,67 The cell type and sublayer-specific loss of PV neurons observed upon Fgf14 deletion confirms this hypothesis and supports the notion that FGF14 is indispensable for the development, maintenance and/or survival of PV interneurons in the CA1 region. Our post-mortem human tissue study identifies a significant correlation between the Fgf14 and PVALB genes, which might contribute to the loss of PV interneurons observed in We also found that the expression level of GAD67 and VGAT, two well-characterized disease-associated proteins critical for GABA synthesis and differentiation in addition to synaptogenesis of PV neurons, [68][69][70][71] are decreased in PV-positive somas and at inhibitory presynaptic terminals in Fgf14 − / − animals. This phenotype is consistent with a diminished total pool of the two proteins, Whether loss of PV neurons and deficits in GABAergic markers in Fgf14 − / − brains occur through causative loops or are separate coincidental events remains to be determined. Evidence exists for two separate functions of FGF14 as a regulator of intrinsic excitability at the AIS 30,33,34,37,38,42,43 and a presynaptic organizer. 40,41 Thus, phenotypes observed in Fgf14 − / − hippocampi might arise from disruption of two independent functions of FGF14: one impairing intrinsic firing of PV neurons, causing cell death, arrested development and/or aberrant circuitry integration, and one disrupting the presynaptic GABA machinery (synthesis and loading) via downregulation of selective markers (that is, GAD67 and VGAT). Convergence of these disrupted functions might have fatal consequences for the final specifications and circuitry integration in PV neurons in the CA1 region. In an Fgf14 null condition, these uncompensated functions could be aggravated by concomitant loss of excitatory inputs, leading to a global remodeling of PV innervation fields, 69 shifting cortical networks into a high-PV status with reduced plasticity. 68 Given that the majority of GABAergic inputs in the CA1 area arise from PV interneurons, and that GAD67 and VGAT are essential for the synthesis and loading of GABA at presynaptic terminals, we posited that Fgf14 genetic deletion might impair hippocampal inhibitory transmission. Consistently, the distribution of sIPSC frequency (and amplitude) in Fgf14 − / − CA1 pyramidal neurons was found populated by longer inter-event intervals compared with Fgf14 +/+ suggesting that in Fgf14 − / − mice the remaining inhibitory neurons might fire less than in control animals. sIPSCs are action potential-driven synaptic events and as such represent a compound readout of the firing status of presynaptic interneurons and the neurotransmitter release machinery. To examine GABA release mechanisms independent Owing to significant deviation from the mean (42 s.d.) in FGF14 gene expression (221310_at), five samples were removed from the GSE12649 data set. GPL96 represents Affymetrix Human Genome U133A Array; GPL570 represents Affymetrix Human Genome U133 Plus 2.0 Array. GPL96 had no probeset selected for VGAT (SLC32A1). The original P-value was adjusted by Benjamini and Hochberg (false discovery rate). The R-value represents a Pearson Correlation with significance at the 0.01 level (two-tailed). (b) Forest plot illustrates the effect size (logFC) of differential gene expression of FGF14, PVALB, VGAT and GAD67 between controls and schizophrenia patients in post-mortem dorsolateral prefrontal cortex (DLPFC) (BA46). Linear model and empirical Bayes method (Limma) was applied for assessing the differential gene expression of FGF14 and its co-expression genes, including PVALB, GAD67 and VGAT, in two independent data sets (GSE21138 and GSE12649) deposited in NCBI GEO. LogFC40 suggest decreased gene expression in patients with schizophrenia. The forest plot was created by R rmeta package. VGAT, vesicular gamma-amino-butyric acid transporter. [61][62][63][64] Fibroblast growth factor 14 and cognitive function TK Alshammari et al of interneuron spontaneous firing, we isolated mIPSCs. The distributions of frequency and amplitude of mIPSCs in Fgf14 − / − mice were shifted with loss in short inter-event intervals and in smallest and largest amplitude events, respectively. These results highlight a combination of pre-and postsynaptic deficits in GABAergic transmission possibly induced by lower quantum content, probability of vesicle release or reduced number of vesicle, which might arise from structural and/or functional loss of a subset of inhibitory terminals. The lack in change of rise and decay time in the Fgf14 − / − mice argues against significant changes in the mechanism of diffusion of GABA across the synaptic cleft, or in the composition of postsynaptic ionotropic GABA receptors. However, high-resolution structural analysis is required for confirmation.
In cortical areas loss in PV neuron function and changes in GABAergic activity can desynchronize the E/I network leading to reduced gamma oscillations, a phenotype associated with schizophrenia and other psychiatric disorders. 1,3,72 In vivo electroencephalogram local field potential recordings in the CA1 region revealed that in Fgf14 − / − animals the total, slow and fast, gamma band power was suppressed compared with control mice. These temporally segregated gamma oscillations reflect information carried by CA3 Schaffer collaterals and perforant path inputs, respectively. [73][74][75] Thus, a reduction in both slow and fast gamma implies that both intra-and extra-hippocampal synaptic inputs might be compromised, possibly reflecting more widespread anomalies in Fgf14 − / − brains, arising from the prefrontal cortex-thalamic-hippocampal loop. 76 At the behavioral level, we show that Fgf14 − / − animals exhibit impaired spatial working memory, complementing the array of deficits commonly found in psychiatric disorders associated with cognitive impairment. 59 Human transcriptomics data confirmed functional clustering of FGF14 with GABAergic signaling and identified a highly correlated decrease of FGF14, PVALB, GAD67 and VGAT in schizophrenia postmortem tissues, indicating possible genetic co-regulation of these genes. Thus, diminished expression of FGF14 in humans might be a risk factor for complex brain diseases associated with cognitive impairment such as schizophrenia. These findings extend the original studies linking the dominant negative FGF14 F145S missense mutation to the inherited, rare disorder SCA27 to a much broader set of human brain diseases.
The range of phenotypes from molecular to behavioral observed in Fgf14 − / − mice along with our corroborating human studies lay the groundwork for new mechanistic hypotheses on the biology and potential risk factors of cognitive impairment in schizophrenia and other complex brain disorders associated with E/I tone imbalance and disrupted development and plasticity of GABAergic signaling. 77 These findings further strengthen the emerging role of the AIS and its molecular components in the biology of diseases such as schizophrenia, bipolar disorder and depression. 42

CONFLICT OF INTEREST
HYM has served as a consultant for Forum, Lundbeck, Reviva and Sunovion; he owns stocks in Acadia Pharmaceuticals, GlaxoSmithKline and SureGene. The remaining authors declare no conflict of interest.