Imbalance of flight–freeze responses and their cellular correlates in the Nlgn3^−/y rat model of autism

A rat model of NLGN3 deficiency was created by zinc-finger nuclease targeting of exon 5 of Nlgn3, leading to a 58 bp deletion (Envigo [30], Fig. 1A). RNA sequencing revealed a ~ 25% loss of Nlgn3 mRNA in Nlgn3^−/y rats vs WTs (Fig. 1C). It revealed the presence of two novel mRNA variants, a truncated “short” isoform, caused by transcription through the deletion site until a stop codon is reached in the adjacent intron, and a long isoform, caused by a cryptic splice site upstream of the targeted deletion, with ~ 25% of RNA-seq reads splicing at this locus to the next exon (Fig. 1D). The short isoform could encode a ~ 30kD protein, whereas the long isoform encodes a predicted protein product 17 amino acids shorter than the full-length protein (Fig. 1B). As both these abnormal potential Nlgn3 isoforms are predicted to contain the N-terminus, but not the C-terminus, of NLGN3, we utilised Western blotting to probe for these (Fig. 1E, Additional file 1: Fig. S9(A–B). We found no presence of NLGN3 protein in Nlgn3^−/y cortical homogenate using either C-terminus or N-terminus-specific NLGN3 antibodies (Fig. 1E), indicating the novel mRNA variants are not translated to a protein product or generate a highly unstable protein.

Nlgn3^−/y mice have been reported to show reduced freezing during both cued and contextual fear conditioning tasks [53]. We found that relative to WT controls, Nlgn3^−/y rats also exhibit reduced freezing behaviour during the conditioning (p < 0.0001, F_{(1, 22)} = 6.61), recall (p = 0.001, F_{(1, 22)} = 13.36), and extinction (p = 0.0009, F_{(1, 22)} = 14.61) phases of an auditory-cued fear conditioning task (Fig. 2A–D) and during a contextual fear conditioning task (Additional file 1: Fig. S1A–D), conditioning p = 0.025, F_{(1, 25)} = 5.67,recall p < 0.0001, F_{(1, 25)} = 26.61). Whilst Nlgn3^−/y rats did not exhibit high levels of freezing during recall, defined as no movement except for breathing, they did appear to respond to the tone by decreasing their overall movement. Therefore, we redefined fear response as immobility of the paws and torso but allowing for head movements. This reanalysis revealed that Nlgn3^−/y rats show a more similar fear learning and extinction profile to WTs, although still significantly reduced (Fig. 2E–F, p < 0.0001, F_{(1, 22)} = 3.23, Additional file 1: Fig. S1D, p < 0.0001, F_{(1, 25)} = 20.65, Additional file 1: Fig. S2A, p = 0.008, F _{(1, 22)} = 8.333). Examination of freezing and paw immobility during the first 5 CS presentations shows significantly reduced recall in Nlgn3^−/y rats relative to WTs (Fig. 2F, p = 0.012, F_{(1, 22)} = 7.52). However, Nlgn3^−/y rats display a significantly higher response to the CS when considering immobility of paws only in comparison with classic freezing (Fig. 2F, p < 0.0001). This effect was not seen in WT animals (Fig. 2F, p = 0.24). As hyperactivity could be a confounding the interpretation of these data, we tested locomotion in the same cohort of WT and Nlgn3^−/y rats in an open field arena before running the fear conditioning paradigm. Analysis of open field behaviour revealed no differences in movement between WT and Nlgn3^−/y rats on any of the four days tested (Additional file 1: Fig. S3A, p = 0.29, F_{(1, 22)} = 1.19. Furthermore, on a marble interaction task, time interacting with marbles was not different between WT and Nlgn3^−/y rats (Additional file 1: Fig. S3E, p = 0.09). These findings indicate that Nlgn3^−/y rats, despite showing reduced freezing behaviour, still form the association between tone and shock but may be expressing their fear in a different manner.

To further explore a potential role for NLGN3 in fear learning, we employed the active place avoidance (APA) task (Fig. 3A, C). During training, response to the low-ampere foot-shock differed between genotypes. Over the course of the 8 trials, 8/9 Nlgn3^−/y rats responding by jumping and escaping the arena altogether. Only 1/9 WT rats showed this behaviour (Fig. 3B, p = 0.0034). Once an animal escaped the arena the trial had to be ended as it was not possible to measure the time to learn the location of the shock-zone.

When testing was repeated in a modified arena with a lid to discourage jumping/escape behaviour, naïve cohorts of both WT and Nlgn3^−/y rats displayed no escape behaviour in response to the foot-shock, in addition to learning the location of the shock-zone by the end of the training sessions and could avoid it by actively remaining in the safe zone (Fig. 3D–H). Nlgn3^−/y rats displayed enhanced performance in this task; throughout training sessions 1 (TS1) and 2 (TS2) Nlgn3^−/y rats entered into the shock-zone significantly fewer times across trials (Fig. 3E, G, TS1: p = 0.0045, F_{(1, 21)} = 10.09, TS2: p = 0.044, F_{(1, 21)} = 4.60), and spent significantly less time in this zone (Fig. 3F, TS1: p = 0.027, F_{(1, 21)} = 5.68, TS2: p = 0.025, F_{(1, 21)} = 5.80) in comparison with WT rats. During habituation to the arena, distance travelled changed over habituation days (Additional file 1: Fig. S3C–D, p = 0.008, F_{(3, 42)} = 0.53), however Nlgn3^−/y rats showed no hyperactivity in comparison with WT (Additional file 1: Fig. S3C–D, Trial 1: WT vs Nlgn3^−/y, p = 0.99, Trial 2: WT vs Nlgn3^−/y, p = 0.90). Furthermore, during training in the presence of foot-shocks the locomotion was not different between WT and Nlgn3^−/y rats (Additional file 1: Fig. S3D, p = 0.59, F_{(1, 21)} = 0.29). During the probe trial, Nlgn3^−/y rats displayed significantly prolonged avoidance of previous shock-zone relative to WT animals, despite no shock being applied. Nlgn3^−/y rats entered the previous shock-zone fewer times on average (Fig. 3J, p = 0.0039, F_{(1, 21)} = 10.51), and spent less total time in this zone (Fig. 3K, p = 0.045, F_{(1, 21)} = 4.53) in comparison with WTs.

The ability of the Nlgn3^−/y rats to successfully learn the location of the shock-zone indicates that spatial memory is unaffected by the loss of NLGN3. However, the exaggerated escape behaviour of Nlgn3^−/y rats seen in the unmodified arena, along with the increased avoidance during the probe trial, suggests NLGN3 loss results in altered fear expression to the shock.

One possible explanation for the data described thus far is Nlgn3^−/y rats are hypersensitive to electrical shocks, and this difference in sensitivity leads to atypical fear response behaviour. To test this, we examined the response of naïve WT and Nlgn3^−/y rats to increasing intensities of foot-shocks (0.06 to 1 mA). Backpedalling and paw withdrawal were the most common initial behaviours observed when an animal first responded to a foot-shock (Fig. 4A). The minimum shock required to elicit any response, or to elicit a backpedalling response, was not different between Nlgn3^−/y and WT rats (Fig. 4B, p = 0.13, Fig. 4C, p = 0.26). This indicates Nlgn3^−/y rats are not hypersensitive to foot-shocks. Additionally, thermal stimulus-induced tail-flick response in Nlgn3^−/y rats was increased compared to WT rats (Additional file 1: Fig. S4B, p = 0.036), suggesting a decreased sensitivity to thermal pain in Nlgn3^−/y rats.

However, Nlgn3^−/y rats exhibited significantly more jumping behaviour than WT rats in response to the higher amplitude shocks (Fig. 4D, p = 0.0081, F_{(1, 23)} = 8.39), suggesting that Nlgn3^−/y rats tend to exhibit flight behaviour in response to foot-shocks. The greater incidence of jumping behaviour here in comparison with the fear conditioning paradigm shown in Fig. 1 is likely due to the repetitive, increasing nature of the foot-shocks given in the paradigm here. At the end of the ramp phase, the shock amplitude was reduced to assess sensitivity changes of the animals induced by the paradigm. Number of jumps to this lower shock intensity was not significantly different between WT and Nlgn3^−/y animals (Additional file 1: Fig. S4A, p = 0.35). These data further suggest that the loss of NLGN3 leads to increased flight behaviour in response to fearful stimuli.

We hypothesised that the increase in shock-elicited flight response in Nlgn3^−/y rats is due to altered physiological properties in the periaqueductal grey (PAG), a midbrain region previously shown to control fear expression [2, 18, 22, 36, 39, 41, 56, 57, 68, 75].

Using whole-cell patch-clamp recordings from acute slices, we found that cells in the dPAG fired an increased number of action potentials to incremental depolarising current injections in Nlgn3^−/y rats compared to WTs (Fig. 5B, p = 0.018, F_{(1, 17)} = 6.87). Rheobase current was also significantly decreased in Nlgn3^−/y cells (Fig. 5C, p = 0.014). No changes in the passive membrane properties or action potential threshold (Additional file 1: Fig. S5) were observed in Nlgn3^−/y rats, however the fast-afterhyperpolarisation potential was significantly decreased (Additional file 1: Fig. S5H, p = 0.0047). Conversely, vPAG cells recorded from Nlgn3^−/y and WT rats fired an equivalent number of action potentials (Fig. 5F, p = 0.54, F_{(1, 17)} = 0.38) and had an average rheobase current comparable to that of WT rats (Fig. 5G, p = 0.40). Nlgn3^−/y vPAG neurons did, however, display increased membrane time constants (Additional file 1: Fig. S5, p = 0.0095). In order to achieve optimal slice health, these recordings were performed in slices from rats ages 4–6 weeks, however a small dataset was also recorded in slices from animals ages 8–10 weeks in order to confirm the hyperexcitability in the dPAG perseveres into young adult Nlgn3^−/y rats. Indeed, we observed an increase in excitability of neurons in the dPAG of Nlgn3^−/y rats in comparison with WT (Additional file 1: Fig. S6A, p = 0.0094, F_{(1, 9)} = 10.82), but not in vPAG neurons (Additional file 1: Fig. S6B, p = 0.92, F_{(1, 13)} = 0.0097). This observed hyperexcitability of dPAG cells in Nlgn3^−/y rats may explain the increased flight and decreased freezing behaviour seen in these rats.

In addition to intrinsic excitability, the excitability of a neuron depends on the synaptic input it receives. We measured mEPSC amplitude and frequencies in dorsal and ventral PAG cells using whole-cell patch-clamp recordings in acute slices from Nlgn3^−/y and WT rats. We found that cells recorded from Nlgn3^−/y and WT rats had comparable mEPSC amplitudes and frequencies in both dPAG (Fig. 5D, amplitude: p = 0.28, frequency: p = 0.61) and vPAG (Fig. 5H, amplitude: p = 0.78, frequency: p = 0.88). Together, these data suggest that dPAG cells are intrinsically hyperexcitable, but do not appear to receive altered excitatory synaptic input.

We did not observe alterations in the excitatory synaptic properties of PAG neurons ex vivo (Fig. 5D, H), however, the synaptic inputs to neurons within the fear circuitry may still be altered in Nlgn3^−/y rats in vivo. Indeed, reduced freezing behaviour during fear recall and extinction has been shown to be correlated with reduced CS-evoked local-field potential (LFP) amplitudes in the PAG [75]. Therefore, we predicted CS-evoked LFPs (or “event-related potentials”, ERPs) would be reduced in Nlgn3^−/y rats, despite overall excitatory synaptic inputs being unchanged. We recorded LFPs in the PAG during auditory fear recall from naïve WT and Nlgn3^−/y rats (Fig. 6). As was seen in non-implanted animals (see Fig. 2), implanted Nlgn3^−/y rats showed reduced freezing behaviour during recall in comparison with WTs (Fig. 6E, F_(1,13) = 17.05, p < 0.001). Nlgn3^−/y rats again displayed a significantly higher response to the CS when examining immobility of the paws only in comparison with classic freezing during fear recall (Fig. 6F, p = 0.001, F_{(1, 7)} = 29.75), an effect that was not seen in WT animals (p = 0.12, F_{(1, 6)} = 3.39). WT and Nlgn3^−/y rats did not show any difference in paw immobility behaviour during the conditioning phase of this task (Additional file 1: Fig. S2B, p = 0.95, F_(1,11) = 0.004).

However, despite the decreased freezing behaviour of Nlgn3^−/y rats, we observed robust ERPs in the PAG of both Nlgn3^−/y and WT rats during fear recall (Fig. 6H–K), the amplitude of which did not differ between WT and Nlgn3^−/y rats (Fig. 6L, p = 0.42, F_{(1, 13)} = 0.73). It was noted, however, that the ERP peak-to-trough duration was significantly shorter in Nlgn3^−/y rats in comparison with WTs (Additional file 1: Fig. S7C–D, p = 0.042, F_{(1, 13)} = 5.09), although the biological relevance of this finding is currently unclear. We furthermore found that dPAG ERP amplitude or duration and percentage of time spent freezing was not correlated on an individual rat level (Additional file 1:  S7A, B, amplitude WT: p = 0.63, r = −0.22; amplitude KO: p = 0.41, r = −0.34; duration WT: p  = 0.61, r = 0.23; duration KO: p = 0.23, r = 0.47;   p 0.84, r = −0.56). In a subset of the same rats (WT n = 5, KO n = 7), we recorded LFPs during the tone habituation session to determine if ERPs were triggered by the unconditioned tone. Contrary to the ERPs seen after conditioning, no ERPs were observed during tone habituation (Fig. 6G, WT: p = 0.25, Nlgn3^−/y: p = 0.093), and no freezing behaviour was exhibited by either genotype (Fig. 6C). These results indicate that Nlgn3^−/y rats display robust ERPs in the PAG during fear recall, of comparable amplitude to WTs, despite showing significantly reduced freezing behaviour. These data suggest that ERPs in the PAG reflect overall fear state elicited by the tone and are not indicative of the type of fear response behaviour (i.e. freezing, flight) exhibited.

Several studies have reported electrical/chemical stimulation of the dPAG evokes robust escape responses such as running and jumping [4, 5, 23, 24, 60, 67, 86], followed by periods of freezing and 22 kHz ultrasonic vocalisations (USVs) [39]. Therefore, we examined whether electrical stimulation of the dPAG would promote greater flight responses in Nlgn3^−/y rats compared to WT controls. Bilateral dPAG stimulation (Fig. 7A) resulted in an immediate post-stimulation hyperactivity of the animals that lasted 1–5 s, followed by freezing (reviewed in [9]). A significantly higher percentage of Nlgn3^−/y rats escaped the arena altogether during the increasing dPAG stimulations (Fig. 7C, p < 0.0001), in addition to a higher percentage exhibiting jumping behaviour at dPAG stimulations of 60, 65, and 70 µA in comparison with WT rats (Fig. 7D, p = 0.0065). This indicates a lowered threshold for dPAG stimulation-induced flight behaviour in Nlgn3^−/y rats. Nlgn3^−/y rats also displayed reduced overall classic freezing and freezing reanalysed as immobility of paws in comparison with WT controls (Fig. 7E, p = 0.025, F_{(1, 12)} = 6.58, Additional file 1: Fig. S2C, p = 0.008, F_{(1, 12)} = 9.86). Additionally, Nlgn3^−/y rats produced fewer 22 kHz USVs relative to WT controls (Fig. 7F–I, p = 0.034).

To control for the possibility that non-specific brain stimulation causes increased escape behaviour in Nlgn3^−/y rats, a small cohort of animals (3 WT, 3 Nlgn3^−/y) were bilaterally implanted with stimulating electrodes in primary somatosensory cortex (S1). Given incremental stimulation of S1, rats displayed no jumping or escape-like behaviour at any time during the behavioural assessment, irrespective of genotype. Furthermore, stimulation of S1 did not induce freezing behaviour (Additional file 1: Fig. S8). This indicates that the increase in flight behaviour in Nlgn3^−/y rats is specific to stimulation of the dPAG.

Together with our findings in Fig. 5B, these data support our hypothesis that increased intrinsic excitability of dPAG cells results in a circuit bias that favours flight over freezing behaviours.

Fear conditioning and recall are often used to assess emotional learning in ASD/ID models, using the quantification of freezing behaviour as a proxy for the memory of the CS–US association. We find Nlgn3^−/y rats display less freezing behaviour (defined as no movement except for respiration) during both auditory and contextual fear recall than WT rats. Taken in isolation, these data could be interpreted as reduced fear learning and/or memory in Nlgn3^−/y rats. However, reanalysis of these data revealed that Nlgn3^−/y rats stop exploratory behaviours following onset of the tone and respond by staying fixed in the same location within space but moving the head and neck. This type of fear behaviour has been reported before in rats confronted with a snake [11, 70], and suggests Nlgn3^−/y rats do form an association between the CS and US, but are expressing their fear differently to WT rats. Two alternative explanations for this behaviour are a change in exploratory activity due to altered anxiety levels, or an increase in repetitive, stereotypic behaviours. However, we found no change in locomotion during open field testing, or in tests believed to reflect stereotypic behaviour (marble burying) in Nlgn3^−/y rats. Hence, the most parsimonious explanation for this head movement is a change in flight-related fear responses. We did not observe escape behaviour during this task, likely because the arena was fully enclosed with no possible escape route. A previous study [53] reported reduced freezing in the Nlgn3^−/y mouse, however no further investigation was made into the fear responses of these mice, so it is not known whether these two models of Nlgn3 deficiency display converging phenotypes. Interestingly, a study on social interactions of Nlgn3 R451C mice [32] reported increased jumping behaviour of these mice, consistent with our findings.

Further insight into the fear responses and learning of Nlgn3^−/y rats was seen in direct response to electrical foot-shocks. Active place avoidance (APA) and shock-ramp paradigms revealed Nlgn3^−/y rats exhibit escape behaviours in response to foot-shocks much more readily than WT controls. However, Nlgn3^−/y rats were able to efficiently learn the location of a shock-zone in the APA task once escape routes were blocked. Moreover, shock sensitivity testing revealed that Nlgn3^−/y rats are not hypersensitive to electrical shocks, but again show increased flight responses. These data further support our hypothesis that Nlgn3^−/y rats do not display associative learning impairments, but preferentially exhibit flight over freezing behaviour in response to fear.

Control of flight and freeze responses to fear are known to involve the dorsal and ventral PAG. Low-intensity electrical stimulation of the dorsal PAG has been shown to elicit freezing responses [60, 72], and higher stimulation to elicit flight responses [4, 5. 60, 67, 85, 23, 24, 72], whereas stimulation of the ventral PAG has been shown to elicit freezing responses rather than flight responses [86, 23, 24]. Correlating with the increased flight behaviour seen in Nlgn3^−/y rats, we observe increased intrinsic cellular excitability in the dorsal, but not ventral, PAG in slices from naïve Nlgn3^−/y rats. These changes in intrinsic excitability in the dorsal PAG are likely to affect the excitation/inhibition balance within the PAG, bringing the resting state of Nlgn3^−/y rats closer to the “threshold” of eliciting an escape response [22]. Altered inhibition could also be contributing to the freeze/flight imbalance observed in Nlgn3^−/y rats, however, and understanding of the relative contribution of altered excitatory and inhibitory circuitry will require a much more detailed of the fear circuit involved.

The increase in firing frequency in the dPAG of Nlgn3^−/y rats appears to be a result of reduced fast-afterhyperpolarisation potential (fAHP). fAHP is mediated by Ca²⁺-activated large-conductance K⁺ channels (BK) which act to hyperpolarise the membrane and reduce neuronal firing [62]. BK channel open-probabilities have been shown to be decreased in another model of ASD/ID, the Fmr1^−/y mouse, leading to increased neuronal excitability [19]. This presents an interesting future research avenue into the function of BK channels in the Nlgn3^−/y rat.

Several studies have reported a strong positive correlation between synaptic input to a neuron and LFP magnitude [1, 29, 77]. We found that miniature excitatory postsynaptic currents (mEPSCs) were not altered in either dorsal or ventral PAG cells recorded ex vivo from Nlgn3^−/y rat slices, suggesting that excitatory synaptic input to these PAG neurons was not altered. Consistent with this, CS-evoked LFPs (or “event-related potentials”, ERPs) recorded from the PAG during fear recall were of comparable amplitude in WT and in Nlgn3^−/y rats. However, we note that the peak-to-trough duration of ERPs in the dPAG of Nlgn3^−/y rats was significantly shorter than those in WTs. As voltage-gated ion channels have been suggested to affect LFP waveform [55, 48, 49], the altered BK channel conductance implicated by the reduced fAHP observed in dPAG neurons ex vivo may be contributing towards this phenotype. The shorter ERP we observe in the dPAG of Nlgn3^−/y rats during fear recall may be reflective of faster, less sustained activity in the PAG. Further experimentation is required to understand this.

ERPs recorded from the PAG during fear recall have been reported to reduce in amplitude during extinction, correlating with reduction in freezing behaviour [75]. We observe very little extinction behaviour in WT rats, however, we also observe no decrease in PAG ERP amplitude across the repeated CS presentations in WT rats. This agrees with Watson et al. [75] in that PAG ERP amplitude is associated with freezing level. However, we observe that despite exhibiting significantly less freezing behaviour than WT rats, the PAG ERP amplitudes in Nlgn3^−/y rats do not differ from WTs. This suggests that ERP amplitude in the PAG reflects the presence of fear, but is unrelated to the type of behavioural response the rat is exhibiting. The presence of robust amplitude ERPs in Nlgn3^−/y rats supports our hypothesis that these rats acquire learned fear of the tone despite the significantly reduced freezing behaviour they exhibit. It is possible that the shorter duration ERPs seen in the Nlgn3^−/y rats instead reflect the differences in freezing behaviour observed between genotypes.

Finally, we show that in vivo dPAG stimulation elicits flight responses in a significantly higher percentage of Nlgn3^−/y rats than WTs. If intrinsic excitability of dPAG neurons is increased in Nlgn3^−/y rats, additional stimulation of this brain region may cause flight responses to be elicited at a lower threshold than that of WT rats. Together, these results suggest that intrinsic changes within the dPAG neurons of Nlgn3^−/y rats underlie the preference for flight responses seen in their behaviour. In addition, compared to WT, Nlgn3^−/y rats emit fewer 22 kHz distress calls during freezing induced by dPAG stimulation, which further indicates potential dysfunction within the dPAG circuitry [39]. Reduced > 50 kHz calls have previously been reported in Nlgn3^−/y mice [53], suggesting that neuroligin-3 loss may cause altered USV emissions in both mice and rats.

Sprague–Dawley Nlgn3^−/y transgenic rats created by Horizon Discovery, now Envigo (RRID: RGD_11568700) were housed on either a 14/10 h (Bangalore Biocluster) or 12/12 h (University of Edinburgh) light/dark cycle with a 21 ± 2 °C room temperature and food/water ad libitum. Animal husbandry was carried out by University of Edinburgh or Bangalore Biocluster technical staff. Rats were housed 4 per cage (2 WT, 2 Nlgn3^−/y, littermates where possible) in conventional non-enriched cages, except for rats that had undergone surgeries, which were single-housed in individually ventilated cages. Body weight was monitored throughout experiments.

Experiments carried out in Edinburgh included: RNA sequencing and Western Blotting (Fig. 1), acute slice whole-cell electrophysiology recordings (Fig. 5, Additional file 1: Fig. S5), and in vivo electrophysiology and behaviour experiments (Figs. 6, 7, Additional file 1: Figs. S6–8).

Experiments carried out in Bangalore included: Western Blotting, auditory fear conditioning (Fig. 2), contextual fear conditioning (Additional file 1: Fig. S1), active place avoidance (Fig. 3, Additional file 1: Fig. S3), shock-ramp test (Fig. 4), open field (Additional file 1: Fig. S3A), marble interaction time (Additional file 1: Fig. S3E) and tail-flick test (Additional file 1: Fig. S4B).

Rats were handled for a minimum of 3 days prior to behavioural testing. Animals undergoing fear conditioning and active place avoidance tasks underwent marble burying, open field, object recognition memory tasks and three-chamber task prior to those shown in this study.

Male littermates were assigned to experimental groups based on genotype to achieve balanced cohorts. Genotyping was carried out by Transnetyx Inc. All experiments and analyses were performed blind to genotype.

Fear behaviour was scored as either “classic freezing”, defined as no movement except for respiration [7], or “paw immobility response”, defined as all 4 paws unmoving, however allowing for movement of the head and neck. Behaviours were scored if lasting > 1 s. For the shock-ramp paradigm, paw withdrawal responses, backpedalling, forward/backward running, and jumping were scored. For dPAG stimulation experiments, response behaviours were scored as freezing, startle, attention, running, or jumping, according to criteria described previously [11]. All behaviour was manually scored using BORIS [27] or in-house software Z-score (created by O. Hardt).

Stimfit software [28] combined with custom-written Matlab scripts (A. Jackson) were used for whole-cell patch-clamp data analysis. mEPSCs were analysed for the final 3 min of the 10-min recording. Events were detected using template-matching and filtered to 3 × standard deviation of baseline [14].

Data collected from LFP recordings were analysed using custom-written MATLAB scripts (F. Inkpen, A. Jackson). Raw traces of 3 tones were averaged, and then z-scored to normalise data to baseline noise. Peak and trough of the LFPs were manually selected.

Raven Lite software (Cornell Lab, Centre for Conservation Bioacoustics) was used to generate spectrograms and to manually quantify USVs.

Throughout, all data is shown as mean ± SEM, or as percentages where appropriate. Statistics were carried out using GraphPad Prism software 8.0, SPSS, or RStudio. Two-way ANOVAs with Holm–Sidak post hoc repeated measures test (Figs. 2, 3, 4, 5, 6, 7, Additional file 1: Figs. S1, 2, 3, 5–7), Pearson’s R correlation (Additional file 1: Fig. S7), unpaired t-tests (Fig. 4, Additional file 1: Fig. S4), paired t-tests (Additional file 1: Fig. S4), Fisher’s exact tests (Figs. 3, 7), three-way ANOVAs (Figs. 2, 6, Additional file 1: Fig. S1), or generalised linear mixed modelling (GLMM) (Fig. 5, Additional file 1: Fig. S5) were employed. N was taken to be animal average in all cases to avoid pseudoreplication, except for when GLMM statistical analysis was employed. R packages lme4 and car were utilised to perform GLMMs. p values of < 0.05 were taken to be significant, and one star (*) represents all p values < 0.05 throughout. Full details of statistical tests and results are described in Additional file 2: Tables S1, 2.