Adult microbiota‐deficient mice have distinct dendritic morphological changes: differential effects in the amygdala and hippocampus

Abstract Increasing evidence implicates the microbiota in the regulation of brain and behaviour. Germ‐free mice (GF; microbiota deficient from birth) exhibit altered stress hormone signalling and anxiety‐like behaviours as well as deficits in social cognition. Although the mechanisms underlying the ability of the gut microbiota to influence stress responsivity and behaviour remain unknown, many lines of evidence point to the amygdala and hippocampus as likely targets. Thus, the aim of this study was to determine if the volume and dendritic morphology of the amygdala and hippocampus differ in GF versus conventionally colonized (CC) mice. Volumetric estimates revealed significant amygdalar and hippocampal expansion in GF compared to CC mice. We also studied the effect of GF status on the level of single neurons in the basolateral amygdala (BLA) and ventral hippocampus. In the BLA, the aspiny interneurons and pyramidal neurons of GF mice exhibited dendritic hypertrophy. The BLA pyramidal neurons of GF mice had more thin, stubby and mushroom spines. In contrast, the ventral hippocampal pyramidal neurons of GF mice were shorter, less branched and had less stubby and mushroom spines. When compared to controls, dentate granule cells of GF mice were less branched but did not differ in spine density. These findings suggest that the microbiota is required for the normal gross morphology and ultrastructure of the amygdala and hippocampus and that this neural remodelling may contribute to the maladaptive stress responsivity and behavioural profile observed in GF mice.


Introduction
A rapidly growing body of evidence points to a role of the microbiota in the regulation of brain and behaviour. The microbiota-gutbrain axis represents a bidirectional network of communication between the intestinal microbiota and the brain (Rhee et al., 2009;Bercik et al., 2012;Sampson & Mazmanian, 2015). The gut microbiota is required for development of the hypothalamic-pituitary-adrenal (HPA) axis, optimal stress responsivity and social cognition (Neufeld et al., 2011;Clarke et al., 2013;Desbonnet et al., 2014). Dysregulation of the microbiota-gut-brain axis may contribute to the development of psychiatric and gastrointestinal diseases, a link supported by the comorbidity found between anxiety disorders and irritable bowel syndrome (Fond et al., 2014) as well as the abnormal composition of gut microbiota in patients with autism (Mayer et al., 2014).
Studies have documented altered amygdalar and hippocampal structure in both humans and animals with exaggerated stress responses. Changes in amygdalar and hippocampal volume have been associated with anxiety disorders in humans and with early-life stress in rodents (De Bellis et al., 2000;Isgor et al., 2004;Salm et al., 2004;Salzer & Weniger, 2010;Machado-de-Sousa et al., 2014). There is also evidence showing long-lasting dendritic hypertrophy of excitatory neurons in the BLA of rodents repeatedly exposed to stressors (Vyas et al., 2002(Vyas et al., , 2004(Vyas et al., , 2006. In contrast, repeated stress induces dendritic atrophy in hippocampal neurons (Magariños & McEwen, 1995a;Vyas et al., 2002). Thus, although the mechanisms underpinning the ability of the gut microbiota to influence stress responsivity and behaviour remain to be elucidated, many lines of evidence point to the amygdala and hippocampus as likely targets.
This study is based on the hypothesis that the GF phenotype could be due to morphological changes in amygdalar and hippocampal subregions with implications for neural output. To this end we determined if the presence of a gut microbiota is necessary for normal structure of the amygdala and hippocampus. We firstly utilized design-based stereology to compare the volumes of well-defined brain regions between adult GF and conventionally colonized (CC) animals. Secondly, we evaluated the dendritic morphology of single neurons in the BLA and hippocampus, regions known to undergo neuronal remodelling and to regulate both anxiety and social behaviours (Vyas et al., 2002;Allsop et al., 2014).

Materials and methods
Animals GF and CC Swiss Webster breeding pairs were supplied by Taconic (Germantown, New York, USA) and first-generation male offspring were used in all experiments. GF mice were housed 3-5 per cage in flexible film gnotobiotic isolators held under controlled conditions (temperature 21 AE 1°C, 55-60% relative humidity) and under a 12 h light/dark cycle. CC mice were housed 2-5 per cage in the standard animal facility under the same controlled conditions and light/dark cycle as the GF animals. Only one pair of CC animals was housed 2 per cage. Both GF and CC mice received the same autoclaved pelleted diet (Special Diets Services, product code 801010). Cohorts of GF and CC mice were randomly allocated to two groups: one for stereology and one for dendritic morphology. Animals were killed at 9-10 weeks of age. A total of 26 mice were used in this study. All experiments were performed in accordance with the guidelines of European Directive 86/609/EEC and the Recommendation 2007/526/65/EC and were approved by the Animal Experimentation Ethics Committee of University College Cork.

Stereological volume estimation
Amygdalar, hippocampal and total brain volume were estimated in both hemispheres using Cavalieri's principle (Gundersen et al., 1988). Brain structures were identified using the Paxinos & Franklin (2001) atlas as a guide. For all analyses, preliminary studies were carried out to establish inter-rater reliability with a disagreement of less than 10% on each brain structure volume.

Histological preparation
GF and CC animals were terminally anesthetized with sodium pentobarbital and perfused transcardially with 0.1 M phosphate buffer (pH 7.4), followed by 4% paraformaldehyde in 0.1 M phosphate buffer. After perfusion, the brains were carefully removed and postfixed for 24 h in a 4% paraformaldehyde 0.1 M phosphate buffer solution after which the brains were cryoprotected in a 30% sucrose solution for 48 h and flash frozen in isopentane. The brains were stored at À80°C before being sliced. Brains were sectioned coronally at 40 lm on a cryostat and stained with thionin. Slides were then coded to obscure the experimental group of each animal until the final step of statistical analysis.

Amygdalar volume
The volumes of the lateral (LA; LAVL + LAVM + LADL), basolateral (BLA; BLA + BLP) and central nucleus of the amygdala (CeA; CeL + CeM) were estimated. The first rostrocaudal appearance of the BLA (Bregma À0.59 mm) was chosen as the sample starting point and the termination of the LA was chosen as the end point (Bregma À2.45 mm). For each amygdala, 7 to 12 evenly spaced serial sections with a randomized start were studied (periodicity of every third or fourth slice; section increment of 0.12 or 0.16 mm). The region of interest was digitally outlined in each section (49 magnification; N.A. 0.1; Fig. 1A) and converted into area values (mm 2 ) using Fiji Is Just ImageJ (FIJI) software (Schindelin et al., 2012). The area values were then combined with the cut section thickness (0.04 mm) and section increment to compute the volumes of the sampled amygdala nuclei. In three GF animals, only one hemisphere was evaluated due to damaged amygdalar tissue.

Hippocampal volume
The volumes of the dorsal and ventral hippocampal formation, the granular layer of the dentate gyrus (GrDG) as well as the hippocampal fields were estimated. The hippocampal formation included the hippocampus proper (Ammon's horn), the dentate gyrus and the subicular complex. The first rostrocaudal appearance of CA3 (Bregma À0.94 mm) was chosen as the sample starting point and the termination of the GrDG (Bregma À4.03 mm) was chosen as the end point. The transition point between the dorsal and ventral hippocampus was taken to be the first full ventral extension of CA3 (Bregma À2.70 mm). The GrDG and the hippocampal field volumes were estimated within the rostrocaudal coordinates containing the total hippocampal formation. The hippocampal fields included the stratum oriens, stratum pyramidale and stratum radiatum. CA2 and CA3 were grouped together (CA2/3) as their anatomical borders are not easily distinguishable using nissl staining. For each hippocampal structure, 17-25 evenly spaced sections with a randomized start were studied (periodicity of every third slice, section increment of 0.12 mm). The regions of interest were outlined (49 magnification; N.A. 0.10; Fig. 1B) and volume was calculated as described above. In one CC animal, only one hippocampus was evaluated due to damaged tissue.

Total brain volume
The volume of the fore-and midbrain (the whole brain excluding the olfactory bulb and cerebellum; approx. Bregma 2.80-4.84 mm) was estimated as described above. Briefly, for each brain, 13-15 evenly spaced serial sections with a randomized start were studied (periodicity of every 10th slice; section increment of 0.4 mm). The perimeter of each hemisphere was digitally outlined in each section (29 magnification, N.A. 0.08) and the total brain volume was calculated. Two GF animals were excluded from the brain volume analysis because tissue damage was extensive and prevented the accurate measurement of the perimeter of one hemisphere.

Dendritic morphology and spine density
Golgi-Cox stain GF and CC animals were terminally anesthetized with sodium pentobarbital and perfused transcardially with a vascular rinse (0.9% saline solution). Immediately following perfusion, brains were removed from the skull and Golgi-Cox staining was performed with a wholebrain staining kit (Bioenno Tech, Irvine, CA, USA) using a 10-day impregnation period. Using a vibratome, 150 lm coronal sections were collected serially and mounted on gelatin subbed slides. To visualize amygdala cytoarchitecture, slices were lightly stained for Nissl bodies with thionin. Slices were then dehydrated (alcohol and xylene) and coverslipped with DPX. Slides were coded to obscure the experimental group of each animal until the final step of statistical analysis.

Classification of neuronal subtypes
Spiny pyramidal neurons and aspiny interneurons in the BLA were selected for analysis on the basis of morphological criteria described in the literature (McDonald, 1982;Sah et al., 2003;Klenowski et al., 2015). Approximately 80% of the neurons in the BLA are excitatory projection cells, and can be accurately subclassified as pyramidal or stellate based on distinct morphological and electrophysiological characteristics alone (McDonald, 1982(McDonald, , 1992Millhouse & DeOlmos, 1983;Washburn & Moises, 1992;Rainnie et al., 1993;Faber et al., 2001;Sah et al., 2003). BLA pyramidal neurons (also termed class I, principal or pyramidal-like) have a large triangle shaped soma and three to seven spiny dendrites emanating from the soma, with one or two of these dendrites being more prominent and similar to the apical dendrite of cortical neurons (McDonald, 1982;Sah et al., 2003;Klenowski et al., 2015). However, unlike the pyramidal neurons in the cortex or hippocampus, these cells are not oriented in one plane, the apical and basilar dendrites are of equivalent lengths, and the distal dendrites do not have a terminal elaboration but taper rapidly. Only BLA pyramidal neurons with a single primary apical dendrite were chosen for analysis. The second group of cells found within the BLA have a smaller cell body, two to six primary dendrites lacking spines, and a spherical dendritic arbour (McDonald, 1982;Millhouse & DeOlmos, 1983;Sah et al., 2003;Klenowski et al., 2015). These neurons (here termed aspiny interneurons but also known as aspiny stellate, or class II) are GABAergic local circuit interneurons which express calciumbinding proteins and an inhibitory electrophysiological response profile (McDonald, 1992;Washburn & Moises, 1992;Rainnie et al., 1993;Lang & Par e, 1998;Kemppainen & Pitk€ anen, 2000;McDonald & Betette, 2001;Sah et al., 2003). We used the presence of an . The volumes of the defined (black lines) subregions of interest were estimated using Cavalieri's principle. The arrow indicates the border between CA1 and CA2/3. (C) The LA, BLA and CeA were significantly enlarged in GF compared to CC mice. (D, F) When compared to controls, the total volume of the hippocampus was larger in GF mice. (E) There was no difference in total brain volume between CC and GF mice. CC, n = 5-6; GF, n = 5-8. *P ≤ 0.05, **P < 0.01. apical dendrite, soma diameter and presence of spines as a wellvalidated means to distinguish pyramidal dendrites from aspiny interneurons. This method of morphological classification between BLA neuron subtypes has been widely utilized in the literature (McDonald, 1982;Millhouse & DeOlmos, 1983;Vyas et al., 2002Vyas et al., , 2006Rubinow et al., 2009;Qin et al., 2011;Bringas et al., 2013;Klenowski et al., 2015).
In the ventral hippocampus, CA3 pyramidal neurons were selected for analysis based on their distinct morphological features. Hippocampal pyramidal neurons have distinct apical and basilar dendritic arbours and a triangular-shaped cell body (Spruston, 2008). Both short-and long-shaft pyramidal neurons were analysed (Fitch et al., 1989).
In the dorsal dentate gyrus of the hippocampus, granule cells were identified by their cone-shaped dendritic tree and lack of basilar dendrites. They were differentiated from basket cells by their elliptical, markedly smaller cell body (Amaral et al., 2007). Granule cells were selected from the outer two thirds of the granule cell layer to minimize the potential for analysing an immature neuron.

Analysis of dendritic arbourization
The analysis of BLA neurons was restricted to those located between Bregma À0.71 mm and À2.45 mm. The analysis of ventral hippocampal neurons was restricted to those located in CA3 between Bregma À2.70 mm and À3.52 mm. The analysis of dentate granule cells was restricted to those located in the dorsal hippocampus, between Bregma À0.94 mm and À2.70 mm. Both apical and basilar dendrites were characterized for pyramidal neurons. To be included in the analysis, the Golgi-impregnated neurons had to satisfy the following criteria as described by Vyas et al. (2002): (i) absence of truncated dendrites, (ii) dark and uniform staining along all dendritic projections and (iii) relatively isolated from neighbouring stained neurons to avoid interference with dendritic reconstruction.
Three pyramidal and aspiny inhibitory neurons were reconstructed in each hemisphere of each animal, with the exception of one CC animal for which all six BLA pyramidal neurons were from the right hemisphere due to tissue damage to the opposite BLA. For the same reason, all six hippocampal pyramidal neurons from one CC and one GF animal were located in the right hemisphere. Two granule cells, one from the infrapyramidal blade and one from the suprapyramidal blade, were reconstructed in each hemisphere of each animal. The neurons were viewed using an Olympus AX70 Provis upright bright-field microscope with an Olympus DP50 camera (Mason, Ireland). Images of each dendritic arbour were taken at 409 magnification (N.A. 1.0) with an oil immersion objective at 1 or 2 lm intervals throughout the entire dendritic arbour. Colourinverted images were then stacked and reconstructed manually in 3D using the Neurofilament tool in Imaris (Bitplane, Switzerland). Total dendritic length and branching were calculated by the software automatically. Sholl analysis, the measurement of the extent and complexity of dendritic material as a function of radial distance from the soma, was performed on 2D images of the traced dendritic arbours using the plug-in for FIJI (Schindelin et al., 2012;Ferreira et al., 2014). The radius step size was 10 lm for aspiny interneurons and dentate granule cells and 20 lm for pyramidal neurons.

Analysis of spine density
Spine density was measured in all BLA and hippocampal pyramidal and granule neurons whose dendritic trees were reconstructed to quantify dendritic arbourization. This analysis was not performed on BLA interneurons as they have few to no spines (McDonald, 1982;Millhouse & DeOlmos, 1983;Sah et al., 2003;Klenowski et al., 2015). Dendritic spines were counted manually from image stacks taken at a magnification of 1009 (N.A. 1.4). A dendritic segment of approximately 20 lm (branch order ≥ 2) was selected that possessed consistent and dark impregnation along the entire extent of the dendrite. All protrusions, regardless of their morphological characteristics, were counted as spines if they were continuous with the dendritic shaft. Two to three apical and basilar dendritic segments from each BLA and hippocampal pyramidal neuron were analysed. Three dendritic segments from each dentate granule cell were analysed. Spines were classified using RECONSTRUCT image analysis software developed by Risher et al. (2014). Briefly, each individual spine's length and width were measured and, based on these measurements, spines were classified in the following subtypes: filopodia (length value > 2 lm), mushroom (width value > 0.6 lm), thin (length-to-width ratio > 1) or stubby (length-to-width ratio ≤ 1). All values were calculated by averaging the spine densities of the sampled segments for each neuron.

Statistical analysis
All data are expressed as means + or AE 1 SEM. Animal means were used for all analyses of dendritic morphology and spine density data. The unpaired Student's t-test (a = 0.05) was used to assess the reliability of group differences (CC vs. GF; right vs. left hemisphere) in volume, total dendritic length, number of branches and spine morphology and density. For Sholl analyses, group differences were tested for significance with a two-way mixed-design ANOVA, with radial distance from the soma as a within-group factor. Post hoc comparisons were made using a Bonferroni's correction, with a significance set at P < 0.05. Percentage changes were calculated with respect to corresponding control values.

Stereological volume estimation
Data were obtained from mice with at least one intact and undamaged amygdala or hippocampus (CC: n = 5-6; GF: n = 5-8). There were no significant hemispheric differences in amygdalar or hippocampal volume in either CC or GF mice (P > 0.05 for all regions), therefore, hemisphere means were used for these analyses.

Amygdalar volume
When compared to CC mice, the amygdala of GF animals showed significant volumetric expansion of all three subregions studied (Fig. 1C). In these animals, the mean volume was increased by 13% in the LA (t 12 = 2.21, P = 0.047), 16% in the BLA (t 12 = 3.13, P = 0.0087) and 20% in the CeA (t 12 = 2.74, P = 0.018). Importantly, there was no significant difference in total brain volume between groups (t 10 = 0.75, P = 0.47; Fig. 1E).

Dendritic morphology and spine density
Morphometric analyses were performed on the entire dendritic arbour of Golgi-stained aspiny interneurons and pyramidal neurons with cell bodies in the BLA and the ventral CA3 of the hippocampus. For each animal, 4-6 neurons of each type were analysed for dendritic length and branching (CC: 48 BLA interneurons, 36 BLA pyramidal neurons, 30 hippocampal pyramidal neurons, 24 dorsal dentate granule cells; GF: 48 BLA interneurons, 42 BLA pyramidal neurons, 36 hippocampal pyramidal neurons, 24 dorsal dentate granule cells).
Dorsal hippocampal dentate granule cell dendritic morphology and spine density The dorsal dentate granule cells of GF mice were less complex compared to controls (Fig. 5A-F). In GF mice, there was no change in granule cell dendritic length between groups (t 10 = 1.29, P = 0.23; Fig 5D); however, there was a 21% reduction in the number of branch points (t 10 = 2.29, P = 0.045; Fig 5E). Two-dimension Sholl analysis revealed a significant interaction of group (CC vs. GF) and distance from the soma on the distribution of granule cell dendrites (F 87, 870 = 1.48, P = 0.0042); however, post hoc comparisons uncovered no statistically significant dendritic regions in which the reduction in complexity occurred (Fig. 4F).

Discussion
This study explored how the microbiota affects the morphology of the amygdala and hippocampus, key brain regions implicated in regulating the range of behavioural and physiological endpoints which are abnormal in GF animals. Our data show, for what is to our knowledge the first time, that mice lacking a gut microbiota displayed an enlarged amygdala and hippocampus. Importantly, there was no difference in total brain volume between GF and CC mice, ruling out the possibility that these enlargements were due to whole-brain expansion. Germ-free status induced dendritic hypertrophy in BLA inhibitory aspiny interneurons and excitatory pyramidal neurons. BLA pyramidal neurons also showed increased thin, stubby and mushroom spine density. When the increase in dendritic length and spine density is taken together, we estimate that there are 81% more axospinous synapses on the BLA pyramidal neurons of GF mice. In contrast, the absence of a gut microbiota induced dendritic atrophy in both hippocampal pyramidal neurons and dentate granule cells. In GF mice, there was also a loss of stubby and mushroom spines on hippocampal pyramidal neurons. We estimate that there are 32% fewer synaptic connections on the hippocampal pyramidal neurons of GF mice when the decreases in dendritic length and spines are combined. Taken together, our findings demonstrate that the gut microbiota is necessary for the normal morphology of the amygdala and hippocampus and suggest that structural changes could underlie the altered stress responses and behavioural profile observed in GF mice.

Morphological impact of growing up germ free
Increasing evidence from GF mice points to a role for the microbiota in regulating social and stress-related behaviours relevant to emotional and neurodevelopmental disorders such as anxiety, depression and autism spectrum disorder (Sudo et al., 2004;Heijtz et al., 2011;Neufeld et al., 2011;Clarke et al., 2013;Desbonnet et al., 2014;Luczynski et al., 2016). However, the neurobiological mechanisms underlying these phenotypes are unclear. To this end, we compared the striking morphological changes observed in GF mice to those found in animal models of stress-related disorders and autism.

Amygdalar and hippocampal volumes
Disruption of social behaviour, a characteristic of both animal models of autism and the GF mouse (Desbonnet et al., 2014;Arentsen et al., 2015), is associated with a alterations in amygdala and hippocampus volume. The effect of prenatal valproate exposure, a risk factor for autism spectrum disorder, on the structural development  The dendrites of ventral hippocampal pyramidal neurons of GF mice were significantly shorter (D) and had less branching compared to controls (E). (F) In GF mice, Sholl analysis revealed a retraction in intermediate dendrites (160-380 lm from the soma). (G, H) The spines on basilar dendrites were thinner in GF mice (G), but there was no significant difference in the length of spines between groups (H). (I) GF mice had less spines on all portions of the dendritic arbour of ventral hippocampal pyramidal neurons compared to CC animals. (J, K, L) This reduction in spine density was due to a decrease in stubby spines (K) and mushroom spines (L), but not of thin spines (J). CC, n = 5; GF, n = 6. *P ≤ 0.05, **P < 0.01. increases hippocampal cell density (Edalatmanesh et al., 2013), whereas maternal immune activation, another model of autism in animals, reduces amygdalar volume and hippocampal thickness (Fatemi et al., 1999(Fatemi et al., , 2008Abazyan et al., 2010).
Changes in amygdalar and hippocampal size have also been documented in rodents subjected to stressors. Prenatally stressed rats have increased LA volume (Salm et al., 2004) and both chronic stress and corticosterone treatment induce hippocampal atrophy (Isgor et al., 2004;Murray et al., 2008;Lee et al., 2009). It remains unclear how gross changes in amygdalar and hippocampal volume relate to stress-related brain function. However, the fact that changes in the size of these structures are frequently observed in humans with anxiety disorders or autism clearly indicates that volumetric alterations of limbic structures either influence or are a result of maladaptive stress responsivity (De Bellis et al., 2000;Schumann et al., 2004Schumann et al., , 2011Salzer & Weniger, 2010;Machado-de-Sousa et al., 2014). To further explore what the changes in amygdalar and hippocampal volume could mean for the stress-related brain function of GF mice, we analysed the dendritic morphology and synaptic connectivity of single neurons in these regions (see below).

Amygdalar and hippocampal dendritic morphology
We found dendritic hypertrophy of BLA aspiny interneurons in GF mice. Dendritic hypertrophy of GABAergic inhibitory interneurons has also been observed in the prefrontal cortex of chronically stressed rodents (Gilabert-Juan et al., 2013). However, further investigation is required to determine if dendritic remodelling of interneurons is limited to the BLA and if it is induced by other treatments.
There exists a substantial literature documenting morphological changes to excitatory pyramidal neurons in the BLA of rodents with similar behavioural profiles to GF mice. In the autism mouse model induced by exposure to valproate, BLA neurons are longer but have reduced spine densities (Bringas et al., 2013). Chronic stress induces a long-lasting hypertrophy and increased spine density of rat BLA pyramidal neurons (Vyas et al., 2002(Vyas et al., , 2004(Vyas et al., , 2006. Dendritic remodelling in hippocampal neurons has been observed in rodent models which, like GF mice, display altered stress responsivity and sociability. Most notably, repeated stress induces apical dendritic atrophy and spine loss in CA3 pyramidal neurons (Watanabe et al., 1992;Magariños & McEwen, 1995b;Vyas et al., 2002;Sandi et al., 2003;Christian et al., 2011). Interestingly, repeated stress and high fat diet act synergistically to induce dendritic atrophy in CA3 pyramidal neurons, raising the possibility that diet, metabolism and perhaps gastrointestinal function can impact brain structure (Baran et al., 2005). Although CA3 neurons have not yet been investigated in rodent models of autism, the CA1 neurons of these animals exhibit structural changes: maternal immune activation reduces dendritic length (Baharnoori et al., 2009), whereas prenatal exposure to valproate induces spine loss (Takuma et al., 2014).
Neurogenesis has a functional role in memory formation and stress responsivity (Snyder et al., 2011;Mar ın-Burgin & Schinder, 2012), and is influenced by the microbiota (Ogbonnaya et al., 2015). A substantial portion of newly generated granule cells migrate into the granule layer of the dentate gyrus and become integrated into the hippocampal network (Hastings & Gould, 1999;Cameron & Mckay, 2001). We thus investigated if the morphology of mature dentate granule cells is also different in these animals, uncovering reduced branching in GF mice. Dentate granule cells appear to be resistant to structural remodelling induced by repeated corticosterone treatment (Woolley et al., 1990); however, maternal immune activation reduces dendritic complexity (Zhang & van Praag, 2015).

HPA-axis signalling
The amygdala has many downstream targets that modulate neuroendocrine stress responses (Herman et al., 2003;Ulrich-Lai & Herman, 2009). The BLA is activated by psychological stressors and BLA lesions dampen HPA-axis response to restraint (Cullinan et al., 1995;Bhatnager et al., 2004). The CeA, however, is not thought to be involved in HPA-axis signalling to psychological stressors, but instead regulates autonomic stress responsivity (Dayas et al., 2001). In contrast to the amygdala, numerous studies have implicated the hippocampus with the inhibition of the HPA axis (Dunn & Orr, 1984;Sapolsky et al., 1984;Herman et al., 2005). GF mice show increased HPA-axis signalling in response to acute restraint stress (Sudo et al., 2004;Neufeld et al., 2011;Clarke et al., 2013). Thus, the increase in HPA-axis signalling in GF mice could be due to concomitant BLA hyperactivation and hippocampal hypoactivation. Indeed, we found that the BLA of these animals is larger and neurons have increased spine density, suggesting that these cells receive more excitatory inputs. In the hippocampus, we found atrophy of both dentate granule cells and CA3 pyramidal cells in GF mice, implying that these cells receive less excitatory inputs. Taken together, we uncovered contrasting patterns of dendritic morphology in the amygdala and hippocampus of GF mice, which has marked implications for the neural regulation of HPA-axis signalling.
Anxiety-like and social behaviours GF mice have reduced anxiety and deficits in social cognition (Heijtz et al., 2011;Neufeld et al., 2011;Desbonnet et al., 2014). There appears to be a significant link between anxiety and social behaviours. Recent studies have utilized optogenetics to confirm the involvement of both the amygdala and the ventral hippocampus in the regulation of these behaviours (Allsop et al., 2014). In addition to its role in anxiety (Deacon et al., 2002;Kjelstrup et al., 2002;Bannerman et al., 2004;McHugh et al., 2004), the ventral hippocampus is involved in sociability, with hippocampal lesions leading to abnormal responses engagement in social situations (Cadogan et al., 1994;Machado & Bachevalier, 2006;Kheirbek et al., 2013). Moreover, the ventral hippocampus has robust reciprocal connections with the amygdala, another brain structure involved in anxiety and sociability (Allsop et al., 2014).
The amygdalar subregions have distinct roles in the regulation of anxiety and social behaviours: the LA and BLA integrate sensory and aversive information and send projections to the CeA (Fanselow & LeDoux, 1999;Sah et al., 2003;Ledoux, 2007;Pape & Pare, 2010;Allsop et al., 2014). In mice, stimulation of the projections from BLA to the CeA produced an anxiolytic phenotype (Tye et al., 2011). Importantly, stimulation of the BLA as a whole produced the opposite effect (Tye et al., 2011), suggesting that the majority of BLA neurons project to other sites and have an anxiogenic effect. Indeed, stimulation of BLA projections to the ventral hippocampus induced a state of increased anxiety but a reduction in sociability (Felix-Ortiz et al., 2013;Felix-Ortiz & Tye, 2014).
In this study, we report dendritic hypertrophy in the BLA and atrophy in the ventral hippocampus of GF mice. It is noteworthy that mushroom spine density was drastically impacted by microbial status: GF mice had one third more and less mushroom spines on BLA and ventral hippocampus pyramidal neurons, respectively. Mushroom spines are crucial for the formation of mature, long-lasting glutamatergic synapses (Matsuzaki et al., 2001(Matsuzaki et al., , 2004Kasai et al., 2003;Holtmaat et al., 2005); therefore, changes in their number likely have a significant influence on the number of excitatory inputs the neurons of GF mice receive. In keeping with this idea, our data suggest that the BLA is hyperactive, whereas the ventral hippocampus is hypoactive. Indeed, transcriptomic analysis revealed that the amygdala of GF mice is in a hyperactive state (Stilling et al., 2015). However, the GF behavioural profilereduced anxiety and social cognitionclearly cannot be explained by either excitation or inhibition of the BLA-ventral hippocampal circuit alone but instead may reflect the cumulative impact of relevant pathways.
The functional relationship between amygdalar and hippocampal activity and social and anxiety behaviours is likely more complex; the BLA and ventral hippocampus project and receive inputs from many other brain areas involved in these behaviours including sensory regions such as sensory thalamus and sensory cortices, the prefrontal cortex, the lateral septum, the supraoptic nucleus, the paraventricular nucleus, the periaqueductal grey and the bed nucleus of the stria terminalis (Herman et al., 2003;Ledoux, 2007;Ulrich-Lai & Herman, 2009;Janak & Tye, 2015). Interestingly, GF mice show hypermyelination of the prefrontal cortex (Hoban et al., 2016), further enforcing the idea that neural communication is altered in several key brain structures.

Future directions
This study serves as a proof of principle experiment which raises the intriguing possibility that the gut microbiota is required for normal brain structure. However, the developmental time point during which the microbiota exerts its influence remains to be determined. The structural brain changes observed may not be the result of a defective brain-gut axis during postnatal life, but rather could be due to the absence of microbiota in the dam during embryogenesis and gestation. Indeed, the peripheral hormones of the mother, which could also include gut hormones, are known to play key organizational roles in the brain during gestation (Bale et al., 2010;Buss et al., 2012). Moreover, prenatal stress has been shown to affect the microbiota in both rodents and humans (Golubeva et al., 2015;Ja sarevi c et al., 2015;Zijlmans et al., 2015). Recent evidence has demonstrated that the maternal microbiota during pregnancy also impacts the immune function of the offspring (Gomez de Ag€ uero et al., 2016). Alternatively, the observed changes in brain structure in GF mice could be occurring during the postnatal period. To tease apart the pre-versus postnatal influence of the microbiota on brain development, the amygdala and hippocampus structure of GF and CC mice should be investigated at several time points. To eliminate the possibility that the microbiota could influence brain development prenatally, future studies should also deplete the microbiota with antibiotics at different postnatal time points. These experiments would allow researchers to determine if the alterations in dendritic spines in the amygdala and hippocampus occur because of deficits in the initial synaptogenesis or later pruning of spines during development (Tau & Peterson, 2009).
An intriguing question that arises from this study is whether colonizing GF mice with bacteria could prevent or reverse the observed structural brain changes. This is especially of interest given that chronic treatment with Lactobacillus rhamnosus reduces anxiety and alters GABA receptor expression in the amygdala (Bravo et al., 2011). Colonization experiments have been performed in GF mice, with certain behavioural and physiological abnormalities proving reversible (Sudo et al., 2004;Heijtz et al., 2011;Clarke et al., 2013;Desbonnet et al., 2014) and others irreversible (Clarke et al., 2013;Desbonnet et al., 2014;Ogbonnaya et al., 2015;Stilling et al., 2015). Moreover, there appear to be critical time windows in which HPA-axis signalling can be normalized in GF mice (Sudo et al., 2004); therefore, the time point in which bacteria is introduced should be carefully considered (Borre et al., 2014;Neufeld et al., 2016).

Conclusions
Our results show altered brain morphology in GF animals and suggest that the amygdala and hippocampus are brain regions whose structural integrity is contingent on the presence of a gut microbiota. This has implications for the expression of limbic system-mediated behaviours and physiology relevant to the stress response, anxiety-like behaviours and social interactions. Further studies are warranted to establish if amygdalar and hippocampal structure are malleable at a structural level following microbiotadirected interventions across the lifespan. An increased understanding of the impact of the microbiota on these brain structures may ultimately inform novel strategies for the treatment of multiple neuropsychiatric disorders.