Topography and Ensemble Activity in the Auditory Cortex of a Mouse Model of Fragile X Syndrome

Autism spectrum disorder (ASD) is often associated with social communication impairments and specific sound processing deficits, for example, problems in following speech in noisy environments. To investigate underlying neuronal processing defects located in the auditory cortex (AC), we performed two-photon Ca2+ imaging in FMR1 (fragile X messenger ribonucleoprotein 1) knock-out (KO) mice, a model for fragile X syndrome (FXS), the most common cause of hereditary ASD in humans. For primary AC (A1) and the anterior auditory field (AAF), topographic frequency representation was less ordered compared with control animals. We additionally analyzed ensemble AC activity in response to various sounds and found subfield-specific differences. In A1, ensemble correlations were lower in general, while in secondary AC (A2), correlations were higher in response to complex sounds, but not to pure tones. Furthermore, sound specificity of ensemble activity was decreased in AAF. Repeating these experiments 1 week later revealed no major differences regarding representational drift. Nevertheless, we found subfield- and genotype-specific changes in ensemble correlation values between the two times points, hinting at alterations in network stability in FMR1 KO mice. These detailed insights into AC network activity and topography in FMR1 KO mice add to the understanding of auditory processing defects in FXS.


Introduction
Autism spectrum disorder (ASD) patients display persistent impairments in social communication and social interactions accompanied by restricted, repetitive behavioral patterns at various severity combined with increased sensitivity to sensory stimuli (American Psychiatric Association, 2013).Auditory deficits are prevalent in ASD, most often described as impaired auditory filtering and difficulties of understanding speech in noise (Alcántara et al., 2004;Tomchek and Dunn, 2007;Ashburner et al., 2008;Groen et al., 2009;DePape et al., 2012;Schafer et al., 2013;Schelinski and von Kriegstein, 2020).Loss-of-function mutations in the FMR1 gene leading to fragile X syndrome (FXS) are the most frequent monogenetic cause (6-8%) of ASD in humans (Muhle et al., 2004;Fyke and Velinov, 2021), with a general prevalence of ASD at ∼2% of children (Baio et al., 2018).Furthermore, various forms of ASD cause misexpression or malfunction of fragile X messenger ribonucleoprotein (FMRP; Assaf et al., 2010;Just et al., 2012).Therefore, FMR1 knock-out (KO) mice have been used frequently as a mouse model of ASD to investigate deficits in neuronal processing caused by altered expression of the FMRP encoded by the FMR1 gene locus.Loss of FMRP causes misregulations in translational processes and alterations in direct protein interaction partners of FMRP (Bülow et al., 2022).At the synapse, these functions include indirect and direct control of presynaptic Ca V activity by activating BK Ca channels or directly interacting with Ca V (Deng et al., 2013;Ferron, 2016) and translation and localization of diacylglycerol lipase-α which uses the endocannabinoid pathway at the postsynaptic side, again influencing presynaptic Ca V activity, leading to increased transmitter release (Jung et al., 2012;Fyke and Velinov, 2021).Furthermore, in the cortex of FMR1 KO mice, misregulation of the matrix metalloprotease-9 (MMP-9), caused by a lack of FMRP, leads to decreased perineuronal net formation around parvalbumin (PV) interneurons, altogether disturbing E/I balance and leading to hyperexcitability (Gibson et al., 2008;Contractor et al., 2015;Wen et al., 2018).
At the level of the auditory brainstem, a wide array of alterations have been reported in ASD patients and FMR1 KO rodents, including disrupted branching and orientation of dendritic trees (Kulesza and Mangunay, 2008), decreased neuron size (Ruby et al., 2015), hyperexcitability (El-Hassar et al., 2019), altered expression of K V 3.1 (Strumbos et al., 2010), altered synaptic transmission and connectivity (Garcia-Pino et al., 2017;McCullagh et al., 2017;Curry et al., 2018), and decreased frequency selectivity (Garcia-Pino et al., 2017).These studies are important contributions to understanding auditory defects in ASD, as this disorder disrupts auditory perception in the temporal envelope (Robertson and Baron-Cohen, 2017), which is highly dependent on fast and precise synaptic transmission in the auditory brainstem.However, deletion of FMRP exclusively in the forebrain of mice also causes some of the FMR1 KO phenotypes, suggesting that local cortical circuitry is directly affected by genetic causes of ASD (Lovelace et al., 2019).Moreover, FMR1 KO mice display impaired auditory cortical plasticity (Kim et al., 2013;S. Yang et al., 2014) and impaired auditory-related learning (Zhao et al., 2005;Reinhard et al., 2019).In the auditory cortex (AC) of FXS patients and FMR1 KO mice, electroencephalogram studies revealed increased amplitudes in event-related potentials and reduced habituation to repeated sounds, indicating a higher noise level in AC activity (Castrén et al., 2003;Ethridge et al., 2016;Lovelace et al., 2018;Wen et al., 2019).On the single-unit level, AC neurons in FMR1 KO mice exhibit stronger responses to brief tones, broader frequency tuning, and reduced frequency modulation selectivity (Rotschafer and Razak, 2013), again indicating spectrotemporal processing deficits.However, very little is known about cortical network activity alterations at the cellular level in FMR1 KO or ASD.
We here performed an in-depth study of sound-evoked activity patterns within the AC of FMR1 KO mice using two-photon Ca 2+ imaging to maintain single-cell resolution while observing hundreds of neurons within one experiment.We reveal decreased bandwidth (BW) of neurons, altered topography of frequency-related activity, and alterations in ensemble activity in response to simple and complex sounds.Our results contribute to the understanding of altered AC activity patterns in FMR1 KO mice that might underlie impaired sound processing in FXS.

Materials and Methods
Animals.Male hemizygous FMR1 KO mice were generated by inserting the homologous recombination target vector pMG5 into exon 5 of the FMR1 gene, resulting in a loss of FMRP (Bakker et al., 1994), and obtained from Jackson Laboratories (B6.129P2-Fmr1tm1Cgr/J, Strain #:003025).They were bred in the animal facility of the RPTU University of Kaiserslautern-Landau with animals of the same genetic background to obtain heterozygous females.These were bred either with KO males to obtain more hemizygous KO males and wild-type (WT) males, as well as homozygous KO females, or with WT males to also obtain WT females.Food and water were provided ad libitum, and all animals were group housed at a 12 h light/dark cycle.Genotyping was carried out on a material left over from ear punch markings.The study was performed in accordance with the guidelines of the German Animal Welfare Act and the European Directive 2010/63/EU for the protection of animals used for scientific purposes.Animal experiments were approved by the regional council of Rhineland-Palatinate (Landesuntersuchungsamt Rheinland-Pfalz) under the file numbers G15-2-076, G19-2-032, and G21-2-072.For in vivo experiments, surgical procedures started at Postnatal Day (P)32, and succeeding imaging experiments were performed up to P70.Slice experiments were conducted between P50 and P70.Animals of both sexes were used (WT, 4× female, 4× male; KO, 3× female, 2× male), with data pooled.
Injection of viral vectors, habituation, and window implantation.In order to provide sufficient analgesia, mice were injected with carprofen (Rimadyl®, Zoetis; 5 mg/kg body weight) intraperitoneally 30 min prior to the initial anesthesia with isoflurane (3-5%, Fluovac system, Harvard Apparatus).The anesthetized mouse was placed on a heating mat (TC-1000 Temperature Controller, CWE) which maintained the body temperature monitored through a rectal thermometer and fixed in a stereotactic frame (Model 900, David Kopf Instruments).The head holder was connected to the Fluovac system which allowed continuous supply of isoflurane (1-3%) during the whole operation.The level of anesthesia was checked regularly by the paw withdrawal reflex, triggered by an intertoe pinch, and isoflurane concentration was adjusted accordingly.Drying-out of the mouse's eyes was prevented by applying an eye ointment.After the fur was shaved from the scalp, Braunol® (B.Braun Melsungen) was applied for disinfection purposes, and 50 µl of lidocaine (Lidocainhydrochlorid 20 mg/ml, bela-pharm) was injected subcutaneously for local analgesia.After 5 min waiting time, the scalp was opened along the midline and removed over the right hemisphere.Afterward, the head was tilted by 45°to the right, and the skin over the left hemisphere was pushed aside to reveal the underlying skull and muscles.The musculus temporalis was then partly removed to access the skull area over the AC.The area of the AC was then approximated by topographic structures, and two injection sites were chosen.A small hole was drilled with a dental drill.The tip of a syringe (NanoFil, World Precision Instruments) containing 3.8 × 10 12 GC/ml of the adeno-associated viral vector (AAV)1-hSyn-jGCaMP7f (pGP-AAV-syn-jGCaMP7f-WPRE was a gift from Douglas Kim and GENIE project (Addgene viral prep #104488-AAV1)) which was inserted through the hole.Seven hundred fifty nanoliters of the vector solution was injected at a rate of 80 nl/min at 500 µm depth.After successful injection, the syringe was kept in place for 5 min, and the procedure was subsequently repeated for the second injection site.A titanium anchor was attached using dental cement (C&B Metabond; Parkell).The remaining skin from the left side of the scalp was then again pulled above the injection sites and cemented to the head plate, sealing the operation site.The animal was then retracted from the stereotactic frame, placed on a heating mat until fully awake, and then brought to its home cage for recovery.To prevent dehydration during the operation, 0.5 ml of 0.9% NaCl solution was administered subcutaneously 60 min after the start of the surgery.For analgesia and to prevent inflammation, carprofen (5 mg/kg body weight) was administered for 2 subsequent days.The well-being of the animal was monitored daily.
Before performing imaging experiments in awake mice, animals were first habituated to the experimental situation, starting earliest 3 d after AAV injection.In each session, animals were brought to a treadmill (LN treadmill, Luigs & Neumann) under the imaging setup and were allowed to freely explore their surroundings for 15 min.Subsequent head fixation lasted initially for 10 min and increased by 25-30 min each day until 2 h were reached, resembling the maximal time for one imaging session.Animals were not habituated on the day of window implantation surgery and on the 2 following days.After five habituation sessions, none of the animals showed any signs of stress, for example, cowering, clinging, and sudden fast running, and were therefore used for subsequent imaging experiments.
Eight days after AAV injection, a window was implanted into the skull, following the general surgical procedures described above.The joint between the cement and remaining skin over the left hemisphere was reopened.The boundary of a round piece of the skull, ∼3 mm in diameter, over the AC was thinned down by tracing the edge with a dental drill.Once the remaining skull encircled by the furrow was loose, it was removed with a fine kinked probe.The dura mater was removed with a small needle and fine forceps.A stack of 2 × 3 mm round cover glass (thickness #0, Warner Instruments) with 1 × 4 mm round cover glass (thickness #1), glued together by a UV-curing adhesive (NOA 60, Norland Optics), was lowered onto the brain, sealing the opening in the skull.The stack was then fixed with dental cement.The remaining exposed tissue and skull were sealed with dental cement as well.The animal was then retracted from the stereotactic frame, placed on a heating mat until fully awake, and then brought to its home cage for recovery.To prevent dehydration during the operation, 0.5 ml of 0.9% NaCl solution was administered subcutaneously 60 min after the start of the surgery.For analgesia and to prevent inflammation, carprofen (5 mg/kg body weight) was administered for 2 subsequent days.The well-being of the animal was monitored daily.
Sound stimulation.The light box as well as the microscope and treadmill position was covered with a sound-attenuating foam (Basotect®, BASF) protecting the recording site from external background noise as well as scanner noise.With this, the sound pressure level (SPL) of ambient noise in the relevant hearing range for mice (1-90 kHz) and in the range of frequencies used for sound stimulation (4-100 kHz) approximated 35 dB SPL and 20 dB SPL at max, respectively.Ambient noise was recorded with a high-sensitive microphone (378A06, 3-40,000 Hz, 12.6 mV/Pa, inherent noise: 22 dB(A) re 20 µPa, PCB Piezotronics), and the corresponding signal was amplified by an analog amplifier [MA3, Tucker-Davis Technologies (TDT)], coupled with an analog-to-digital multifunction processor (RX6, TDT) controlled by SigCalRP (v4.2,TDT).Daily calibration of the speaker was done using the same equipment except for a less sensitive microphone, which could record higher frequencies up to 100 kHz (Model 378C01, 4-100,000 Hz, 2.01 mV/Pa, inherent noise: 42 dB(A) re 20 µPa, PCB).The SPL during tone presentation (4-100 kHz) was recorded and used to adjust driving voltages of the speaker for each frequency according to the desired SPL.These normalized values were then exported to MATLAB, and filter coefficients were calculated to be used for the online finite impulse response filter of sound presentation during experiments.During two-photon imaging, two groups of acoustic stimuli were presented.Group 1 consisted of 17 different pure tones (PTs) (4-64 kHz, four equivalent steps per octave; 250 ms tone length followed by a 1 s pause), AM tones (same carrier frequencies and time course as PTs, 20 and 40 Hz modulation frequency, 70 dB SPL), and complex acoustic stimulations, consisting of mouse vocalization mimic (3.8 kHz fundamental frequency and second and third harmonic), AM mouse vocalization mimic (same as vocalization mimics but 1 kHz modulation frequency), 12 natural animal vocalizations (1 s pause between vocalizations, 70 dB SPL; all vocalizations downloaded from http://www.avisoft.com/animal-sounds/), and an overlay of all 12 natural animal vocalizations.We thank Matthias Göttsche (Stocksee, Germany) for allowing us to use the recordings of the Blasius's horseshoe bat.To provide enough data for a suitable analysis, only vocalizations with a duration of ≥0.55 s were chosen for analysis, that is, 10 vocalizations.They were randomly presented within their sound group (PTs, AM tones, vocalization mimics, or natural animal vocalizations) during each of the 10 repetitions.Group 2 was used to create frequency response areas (FRAs).Therefore, PT sequences were presented with five different SPLs (30-70 dB SPL increasing in 10 dB steps if each tone/vocalization was high-pass filtered at 4 kHz).For widefield imaging, five PTs (4, 8, 16, 32, 64 kHz, 500 ms tone; 5 s pause between tones) were randomly presented during each of the 16 repetitions and played at 50, 60, and 70 dB SPL.All stimuli were created with MATLAB 2020a (MathWorks) controlling a script written in RPvdsEX (v88, TDT) and loaded to the RX6 digital-to-analog converter.The output signal of the RX6 was passed by an electrostatic speaker driver (ED1, TDT) and finally transformed into acoustic signals by an electrostatic free field speaker (ES1, TDT), positioned ∼10 cm from the ear of the mouse contralateral to the window.
In vivo Ca 2+ imaging.For awake in vivo Ca 2+ imaging, the cranial window was covered with an ultrasound gel (Anagel, Ana Wiz) for recordings with a 10× water immersion objective (IMPPLFLN, Olympus K.K.) and 16× water immersion objective (CFI75 LWD, 0.8 NA, Nikon).All recordings were obtained using the Ultima Investigator microscope (Bruker AXS SAS) with the objective tilted by 45°, maintaining the animal in an upright position.During all recordings, the animal was awake and able to move on a treadmill but was fixed with the head anchor.As an excitation source for two-photon imaging, a Chameleon Vision II Titan:Sapphire laser (Coherent), controlled by Chameleon Vision (v2.83,Coherent), with 140 fs pulse duration and 80 MHz repetition rate, with built-in precompensation, was used, tuned to 940 nm.The laser intensity was adjusted with a Pockels cell (Model 302RM Driver, Conoptics) and was in the range between 13 and 65 mW.Imaging was performed using a Galvo-Resonant 8 kHz scanner, recording a 512 pixel × 512 pixel field of view (FOV), covering ∼820 µm × 820 µm, at 29.76 frames/s.The emitted fluorescence was collected by the 16× objective and guided through a green emission bandpass filter onto a GaAsP photomultiplier tube (Bruker).All imaging components were controlled by Prairie View (v5.5.64.500,Bruker), and parameters were set by MATLAB 2020a via Prairie Link (v5.5.0.48,Bruker).During all recordings, the movement of the mouse on the treadmill was registered by two hall sensors.
On the first day of imaging, the FOVs were chosen to cover as much area as possible of the cranial window, depending on the area of transfected tissue.The first FOV was chosen as the origin of a coordinate system, and coordinates of each FOV were saved and together with a reference image used for guidance on the subsequent days.During Day 1, PTs, AM tones, and complex sounds were presented at 70 dB SPL.On the next day, coordinates and vasculature were used to find the same FOVs on a coarse scale.On a finer scale, the same position was searched by accurately matching pixel coordinates of cells from the live image with the reference image and scrolling on the z-axis until most cells from Day 1 were detectable.During imaging on this day, PTs with different SPLs (30-70 dB) were presented.To increase the number of neurons per subfield, on Days 4 and 5 of imaging, the same XY-coordinates per FOV were used, but a different depth was chosen.It was assured that no cells from the first depth were visible, optimizing the increase in the number of cells.The resulting depths ranged from 180 to 270 µm.The sound stimulation paradigm on Days 4 and 5 was identical to Days 1 and 2, respectively.The same imaging pattern was then repeated 7 ± 1 d later for most FOVs and depths.In some cases, recordings could only be performed in the first week, due to the worsening of the optical conditions of the window.In cases of statistical comparison of datasets across the 2 weeks, only FOVs recorded in both weeks were included.However, analysis was not limited to neurons only visible/ active on both experimental days.At the end of each session, the animal was brought back to its home cage.
For widefield imaging, performed on Day 3, illumination with blue light was achieved by a LED (470 nm, Thorlabs).The emitted fluorescence was passed through a green emission filter onto a scientific CMOS camera (Prime 95B, Teledyne Photometrics), controlled by Micro-Manager (v2.0, "https://micro-manager.org").The corresponding FOV size covered ∼1.5 mm × 1.5 mm, resulting in a 1,024 pixel × 1,024 pixel image.Data were collected, after focusing roughly 200 µm below the pial surface, at a rate of 10 Hz with an exposure time of 60 ms.
Analysis of in vivo data.For analysis of widefield imaging data, the procedure of image processing was adapted from Romero et al. (2019).Raw images were downsampled to a 256 pixel × 256 pixel resolution.Small drifts in fluorescence signal were removed by computing a temporal baseline (F0) for each pixel from a polynomial fit (Degree 3) of a 15 s sliding window (Chronux toolbox, MATLAB).The change in fluorescence was calculated for each frame as percent change from the temporally smoothed signal (ΔF/F•100).These amplitudes were used for further analysis.Baseline activity levels for each stimulus were defined for each pixel by creating a histogram of amplitudes of all frames during the 2 s prestimulus period.To check for tone-evoked responses, the maximum amplitude was picked from the 750 ms period after tone onset and averaged with the preceding and following frame.In cases where the resulting value exceeded the prestimulus baseline activity distribution by at least two standard deviations (z-score >2), the response was characterized as tone evoked.A frequency-specific response amplitude was only calculated when a tone-evoked response occurred in a minimum of 4 of 16 repetitions.The frequency-specific response was then calculated as the mean of all significant tone-evoked response amplitudes to the respective frequency.The frequency eliciting the highest mean response amplitude within a pixel was set as the best frequency (BF) of that given pixel.As one FOV acquired with the 10× objective covered only a part of the AC, multiple overlapping FOVs were necessary in order to create a gapless BF map.The frequency-specific response amplitudes of each pixel within a FOV were normalized to provide comparability.In cases where one pixel was represented more than once (due to overlapping FOVs), the BF with the higher normalized mean response amplitude was chosen.Next, a vector-based calculation of reversal points, similar to the analysis in Romero et al. (2019), was provided as follows to assist subfield parcellation.First, centers of existing low-frequency hubs were identified.From each of these, a set of 1,440 radial vectors from 0 to 360°(0.25°stepsize) were drawn.The mean BFs along each radial vector (±1°) were smoothed with a moving average (window size 10 frames).The smoothed values were then fitted with a Gaussian filter (Degree 3), so that reversal points (first maxima) could be marked in the BF map.The end of the AC was defined as the point, where 10 pixels in a row showed no sound-evoked response at all.The marked reversal and end points within the BF map served as a template for the "drawassist" function of MATLAB.Thereby, the subfield borders could be drawn by hand, but the outline was corrected by the information of the underlying BF map.Assignment of A1, AAF, and A2 was performed, based on existing knowledge from earlier studies (Tsukano et al., 2015(Tsukano et al., , 2016;;Romero et al., 2019).
For two-photon data, recordings were processed with suite2p (https://suite2p.readthedocs.io/;Pachitariu et al., 2017), first correcting for rigid as well as nonrigid movement shifts.Next, region of interest (ROI) detection, signal extraction, and local neuropil signal extraction were carried out.ROI fluorescence traces, subtracted by 0.7 times neuropil traces, were then deconvolved using the OASIS algorithm (Friedrich et al., 2017), and the resulting spiking probabilities were used for most of later analyses.ROIs were grouped as "cell" or "noncell" by a trained classifier depending on the activity parameter and the parameter of the ROI shape.This automatic classification of ROIs as cells or noncells was manually reviewed.The data were then exported to MATLAB for further processing.To rescale ROI fluorescence traces and neuropil traces, first a wavelet denoising was carried out by utilizing the MATLAB function "wdenoise" (Wavelet Toolbox, MATLAB 2022a).Denoised traces were then scaled by 0.86, and the subtracted noise was added again.Deconvolution was performed by the OASIS algorithm, and following analyses were identical to the unscaled trace analyses.
AC activity is influenced during movement by inhibiting neuronal activity (Nelson et al., 2013).Therefore, phases of running which exceeded 1 cm/s were excluded from the activity traces for analysis.Furthermore, unresponsive neurons were removed from analysis if their peak signal-to-noise ratio (Eq. 1) was below 36 dB for the whole activity trace: where F raw , F n , and σ n as the ROI trace, the neuropil trace, and the standard deviation of the neuropil trace, respectively.Neurons were defined as PT responsive by comparing the mean spiking probabilities 400 ms prestimulus and 400 ms poststimulus onset.A one-way ANOVA compared both distributions for each frequency-intensity distribution, and if p < 0.01 in at least one PT-SPL combination, neurons were classified as PT responsive, and others were excluded.In case of the animal running during sound stimulations, the given repetition was removed from analysis, and the complete FOV was disregarded for analysis in case fewer than five repetitions without running were recorded.Mean poststimulus responses of each frequency were averaged across SPLs, resulting in a single mean value per frequency.These points were then fitted with a unimodal and bimodal Gaussian (Eqs. 2 and 3, respectively) fit function to check for single-or double-peak tuning, respectively: To adjust for the different number of parameters of the two fits, the adjusted coefficient of determination (R 2 adj) was used to decide which fit was more suitable.If both fits resulted in R 2 adj < 0.4, the neuron was classified as "irregular tuned"; otherwise, the fit that resulted in a higher R 2 adj was used to assign a "single-" or "double-peak tuning".Furthermore, the tuning BW of each peak was determined as the full width at half maximum (Eq.4): In each FRA, the frequency that elicited the highest response was defined as the BF, regardless of SPL.Next, local FOV coordinates from all neurons were converted into a global coordinate system.Global coordinates from each neuron were used to calculate the local BF distribution.For each neuron, its BF and the BFs of all neurons within 100 µm were extracted.Then, the interquartile range (IQR) of the distribution was calculated as a measure of local heterogeneity.If less than five neurons were within 100 µm radius (including the center neuron), no IQR was calculated.
To analyze which sounds evoke activity in different ensembles of neurons, a correlation analysis was performed related to Bathellier et al. (2012).Hence, the activity of each neuron in a time window of 0.4 s (0.55 s in case only complex sounds were analyzed) from the start of acoustic stimulation was averaged, and the resulting values were combined to a "cell vector".Vectors were then correlated across sounds (using Pearson's correlation), determining the similarity, and these correlation values were averaged across repetitions, describing the reliability of responses.As for FRA analysis described above, in case of the animal running during sound stimulations, the given repetition was removed from analysis, and the complete FOV was disregarded for analysis in case fewer than five repetitions without running were recorded (for details see above).The hierarchical ordering of the resulting correlation matrix was carried out using the unweighted average distance.The correlation matrix was exported in R (R Language, v4.1.2,https://www.R-project.org/)where clusters were defined by a hierarchical cluster tree using dynamic tree cut (Langfelder et al., 2007, method "hybrid", "deepsplit" set to 2.5).To determine the similarity of cluster content between 2 experimental days 1 week apart, for each FOV, the content of a given cluster observed in Week 1 was compared with the content of all clusters for Week 2, and the cluster with the most similar content was chosen as the counterpart.The fraction of sounds shared was then averaged across all clusters observed in Week 1 to determine one similarity value for the FOV.For neuron correlations across the week, the two cell vectors described above were correlated for each given sound and repetition (but not between sounds), and the values were averaged for one FOV.
Code accessibility.The code/software described in the paper is freely available online at https://github.com/HirtzLab/Imaging_auditory_cortex_fmr1_KO.The code is available as Extended Data.In the present study, the code was run on standard PCs using Windows 10 operating systems.
Electrophysiological recordings, accompanied by single-cell Ca 2+ imaging, were performed on an electrophysiological rig.Acute brain slices containing the AC were transferred into a recording chamber and fixed with a U-shaped platinumiridium grid stringed with nylon strands.The chamber was then mounted on an upright microscope (Eclipse E600FN, Nikon), equipped with differential interference contrast optics, appropriate objectives (Nikon 4× CFI Achromat, 0.1 NA; Nikon 60× CFI Fluor W, 1.0 NA), and a scientific CMOS camera (Iris 9, Teledyne Photometrics), controlled by Micro-Manager (v2.0).During imaging experiments, a blue light LED (470 nm, Thorlabs) was used to illuminate the whole slice (1.1 mW/cm 2 at maximum), and the emitted fluorescence was guided through a bandpass filter (500-550 nm) and captured by the camera.Once transferred to the microscope, the slices were continuously perfused with aCSF (room temperature, pH 7.4, bubbled with carbogen) using a peristaltic pump (ISM796B, Ismatec).Pipettes pulled from borosilicate glass capillaries with a filament (GB150F-8P, Science Products) using a horizontal puller (Flaming Brown Micropipette Puller P-87, Sutter Instruments) had resistances ranging from 2.5 to 4 MΩ when filled with an internal solution [in mM: 140 K-gluconate, 10 HEPES, 1 MgCl 2 , 2 ATP-Na 2 , 0.3 GTP-Na 2 , 0.05 Oregon green BAPTA-1 (OGB-1)].
Whole-cell recordings were obtained using a patch-clamp amplifier (EPC9, HEKA Electronics) and a micromanipulator (SM-I, Luigs & Neumann) linked to the head stage.Capacitive transients were neutralized, and series resistance was compensated by 50-70%.The liquid junction potential (15.4 mV) was corrected online.During the 10 min filling time of the cell with the internal solution, parameters for AP generation, that is, intensity and duration of the injected current, were determined.Injected currents ranged from 0.5 to 0.8 nA with a duration of 2-8 ms.Once a suitable concentration of OGB-1 was achieved, rectangular current injections were performed at different frequencies while simultaneously recording fluorescent responses.The protocols were repeated up to three times.The obtained electrophysiological data were digitized with a sampling rate of 50 kHz and low-pass filtered at 8.3 kHz.Imaging data were sampled at 40 Hz with an exposure time of 25 ms.Image sequences recorded from OGB-1-filled cells were analyzed using a custom-written MATLAB GUI allowing to display electrophysiological and fluorescence traces in parallel.Baseline for ΔF/F values was calculated by taking the mean of 75 ms prior to the peak.
Data were analyzed using a custom-written MATLAB (MathWorks) code with the HEKA Patchmaster Importer (Keine, 2019).To normalize changes in fluorescence across animals and slices, ΔF/F values were calculated by averaging fluorescence 330 ms prior to each peak yielding a local baseline, which was used to normalize the corresponding peak.Hence, ΔF/F values for each stimulation intensity could be obtained.
Data visualization and statistics.Bar graphs and values in text present mean ± standard error of mean with number in bar depicting n-number.Normal distribution was tested with Kolmogorov-Smirnov test.Normally distributed datasets were compared using paired or unpaired, two-tailed t tests.Distribution-free datasets were compared using Wilcoxon signed rank tests or Mann-Whitney U tests for paired or unpaired datasets, respectively.Significance levels are as follows: p < 0.05 *, p < 0.01 ** and p < 0.001 ***.

Results
To study topographic order and ensemble activity patterns of neurons within the AC of FMR1 KO mice and WT littermates, we used two-photon imaging at single-cell level of GCaMP7f-expressing L2/3 neurons.We presented randomized PTs at increasing SPLs (4-64 kHz, 30-70 dB SPL), to characterize frequency tuning and topography.From a total of 82,141 identified WT neurons and 63,409 KO neurons, 20,757 and 11,916 neurons were responsive to PTs (see Materials and Methods for details).These were further classified as "single-peak", "double-peak", and "irregular"-tuned neurons (Fig. 1A, Extended Data Fig. 1-1), as done by Gaucher et al. (2020) before.In general, irregular-tuned neurons made up the biggest fraction, followed by single-peak neurons and double-peak neurons in both genotypes (WT: irregular, 74%; single-peak, 23%; double-peak, 4%; KO: irregular, 80%; single-peak, 17%; double-peak, 3%).Their location was aligned with the borders of AC subfields as determined by conventional widefield fluorescence imaging at low magnification.Subfield parcellation resulted in three subfields, namely, A1, AAF, and A2.In some cases, following the nomenclature from Romero et al. (2019), five subfields could be identified.To streamline analysis, in such cases, the suprarhinal auditory field was termed A2, and the dorsoposterior field and ventral posterior auditory field, if present, were added to A1.Furthermore, tone-responsive regions adjacent to subfields were added to them (Extended Data Fig. 1-2).
Neurons in the AC exhibit a local BF heterogeneity which can span multiple octaves (Romero et al., 2019).This local BF heterogeneity is an indirect method to assess tonotopic organization (Bowen et al., 2020).We analyzed the IQR of all single-peak neurons within a 100 µm radius of a given neuron and repeated this procedure for each neuron in a subfield for single-peak neurons in FMR1 KO and littermates (Fig. 2B).Single-peak neurons in FMR1 KO exhibited an increased IQR in A1 and AAF compared with control (A1: WT: 0.8 ± 0.01 oct, KO: 0.93 ± 0.02 oct, p = 2.2 × 10 −3 , Mann-Whitney U test, AAF: WT: 0.98 ± 0.01 oct, KO: 1.22 ± 0.03 oct, p = 1.1 × 10 −5 , Mann-Whitney U test, WT vs KO, respectively; Fig. 2B, right), whereas in A2, IQR showed a nonsignificant tendency to be higher in control (WT: 1.06 ± 0.01 oct, KO: 0.95 ± 0.01 oct, p = 0.055, Mann-Whitney U test).These findings indicate a less structured local tonotopic organization in A1 and AAF of FMR1 KO.FMRP regulates several voltage-gated Ca 2+ channel types (Fyke and Velinov, 2021).Thus, AP-evoked fluorescence changes at the soma of KO animals might differ from those observed in WT, leading to the possibility of differences in detection thresholds for the experiments described above.To test this, we prepared acute cortical slices from the brains of both genotypes and performed patch-clamp recordings of single neurons in layer 2/3 AC, filling them with OGB-1 through the patch pipette.Amplitudes of AP-induced changes in fluorescence did not differ significantly between WT and KO animals, both for single APs (WT: 0.024 ± 0.002 dF/F, n = 14, KO: 0.028 ± 0.003 dF/F, n = 12, p = 0.2034, unpaired t test; Fig. 3A) and trains of 10 APs evoked at 5 Hz (WT: 0.074 ± 0.008 dF/F, n = 10, KO: 0.095 ± 0.014 dF/F, n = 9, p = 0.1937, unpaired t test) or 10 Hz (WT: 0.08 ± 0.009 dF/F, n = 9, KO: 0.1 ± 0.016 dF/F, n = 9, p = 0.2467, unpaired t test; Fig. 3B).It should be noted though that a trend toward higher amplitudes in KO animals was observed.To assess whether this might influence our results, we employed a rescaling algorithm to decrease the Ca 2+ trace amplitudes recorded in KO animals while keeping their noise level original (see Materials and Methods for details).We decreased the signals to 86% in accordance with fluorescence signals observed in slices when eliciting single APs.Next, we reanalyzed BW for all neurons in our dataset, as this parameter should be strongly affected by changes in activity levels.While there was a slight increase of BW in A1 (0.71 oct to 0.75 oct) and a minor reduction in A2 (0.79 oct to 0.80 oct), differences in A1 and AAF between WT and KO were still present (A1: WT: 0.85 oct ±0.01 oct, n = 2,495, rescaled KO: 0.75 oct ±0.01 oct, n = 935, p = 8.8 × 10 −10 , Mann-Whitney U test, AAF: WT: 0.93 oct ±0.02 oct, n = 744, rescaled KO: 0.75 oct ±0.02 oct, n = 589, p = 3.3 × 10 −9 , Mann-Whitney U test; Extended Data Fig. 1-3), and still no difference could be observed in A2 (WT: 0.80 oct ±0.01 oct, n = 1,395, rescaled KO: 0.79 oct ±0.02 oct, n = 956, p = 0.09, Mann-Whitney U test).We thus conclude that changes in Ca 2+ influx into cortical AC neurons are, if present, not of concern regarding altered event detection thresholds in KO animals.
Next, we analyzed activity correlations and neuronal ensembles within the three AC subfields in response to different sounds.For this purpose, 17 PTs, 34 AM tones with two different modulation frequencies (same carrier frequencies as PT set), and 13 different complex sounds were presented to the animals.Sound-evoked responses were grouped into clusters based on the activity correlations of all neurons within an observed FOV, first concentrating on the set of 17 PTs (Fig. 4A).The number of sound clusters per FOV, the number of sounds per cluster, and the overall fraction of clustered sounds did not differ between genotypes in any of the three subfields (Fig. 4B-D).However, specifically in A1, the mean correlation within clusters and the reliability of neuronal activity patterns in response to repetitions of the same sound (diagonal of correlation matrices for a given cluster) were lower in KO mice.Correlation values for sounds not within the same cluster (a measure of cross talk between different neuronal ensembles) did not differ, indicating that the decrease in correlation strength within neuronal ensembles was not accompanied by a decrease in specificity (Extended Data Table 4-1).When repeating these experiments at different dB levels 1 d later (dataset also analyzed above for tuning, BW, and local heterogeneity analyzes), lower reliability in KO was confirmed for 70 dB, though lower correlation within clusters only in tendency (Fig. 4E, Extended Data Table 4-2).Testing other dB levels did not result in differences.
Analyzing network responses to 13 complex sounds (animal vocalizations and artificial mouse vocalization mimics) revealed similar differences as for the PT set regarding correlation and reliability for A1 (Fig. 5A,B, Extended Data Table 5-1), but interestingly further differences between the genotypes in AAF and A2.In AAF, the fraction of clustered sounds was slightly lower in KO mice, but more importantly, the between-cluster correlation was higher, indicating less specificity of network activity patterns in response to different sounds (Fig. 5C, Extended Data Fig. 5-1).In A2, correlation and reliability were, in contrast to A1, increased in KO mice (Fig. 5D).
To assess the stability of network activity features, experiments were repeated after 1 week (see Materials and Methods for details; Fig. 6, Extended Data Fig. 6-1).Correlation analysis was performed across the complete set of 17 PTs, 34 AM tones, and 13 complex sounds.The between-week similarity of the content of sound clusters for a given FOV did not differ between the genotypes across all subfields (A1: WT: 0.47 ± 0.04, n = 25, KO: 0.40 ± 0.03, n = 25, p = 0.1578, unpaired t test, AAF: WT: 0.40 ± 0.05, n = 21, KO: 0.46 ± 0.04, n = 24, p = 0.3405, unpaired t test, A2: WT: 0.48 ± 0.03, n = 17, KO: 0.49 ± 0.03, n = 11, p = 0.9204, unpaired t test; Fig. 6B left), implying a similar stability of sound categorization for the two genotypes.Furthermore, when limiting the dataset to neurons observed on both experimental days, the correlation between the two was similar for WT and KO as well (A1: WT: 0.22 ± 0.03, n = 25, KO: 0.23 ± 0.01, n = 25, p = 0.6497, unpaired t test, AAF: WT: 0.19 ± 0.03, n = 21, KO: 0.20 ± 0.01, n = 24, p = 0.9237, unpaired t test, A2: WT: 0.22 ± 0.02, n = 17, KO: 0.19 ± 0.02, n = 11, p = 0.3884, unpaired t test; Fig. 6B right).Thus, the basic stability of sound feature analysis appears to be largely unaffected in FMR1 KO mice within the time period observed.It should however be noted that again differences were apparent regarding correlation and reliability within and between clusters (Fig. 6C).For A1, all values were higher for KO animals in the second, but not the first week of experiments.In contrast, in AAF, higher values were observed only in the first week.For A2, correlation within clusters was lower in the second week in KO animals, but correlation between clusters was higher in the first week, overall showing a tendency of lower correlation values for the second week for KO animals (Extended Data Table 6-1, Extended Data Table 6-2).Statistical analysis across the FOVs observed in both weeks revealed that these differences were the results of changes in almost exclusively either WT or KO, depending on the subfields imaged.This was expressed in a decrease in correlations in A1 for WT, an increase for WT in AAF, and a decrease in A2 for KO (Extended Data Fig. 6-2, Extended Data Table 6-3, Extended Data Table 6-4).In summary, while the general stability of sound categorization and network composition appeared to be unaffected in FMR1 KO mice, either activity correlations and response reliability within and between neuronal ensembles underwent alterations within 1 week of observation that did not occur in WT or alterations observed in WT were not present in KO, depending on the subfields.
In summary, frequency-specific alterations of network ensemble activity in KO mice were observed only in A1, while alterations in the processing of complex sounds were present in AAF and A2 as well, though expressed in a subfieldspecific manner.The basic stability of sound feature analysis across 1 week was unaffected, but correlation values within sound clusters appeared to change within this time in both genotypes, depending on the subfield observed.

Discussion
In this study, we performed two-photon imaging of L2/3 neurons in AC of FMR1 KO mice and WT littermate controls to analyze frequency tuning and topography as well as ensemble activity in a mouse model of FXS.We found a reduction in tuning BW and a decreased topographic order of frequency representation in most AC subfields, as well as alterations in ensemble activity, foremost manifested in decreased correlations in A1, increased correlations in A2, and decreased sound specificity in AAF.
Overall, 25% of active neurons were PT-responsive in WT animals, which is in a similar range to a previous report (25-35%; Bowen et al., 2020) but is less compared with Gaucher et al. (2020), who reported 44% of WT neurons being PT-responsive.There are several technical factors that could account for this discrepancy.Gaucher et al. (2020) used GCaMP6m as Ca 2+ sensor, which has a lower sensitivity and smaller peak fluorescence compared with jGCaMP7f used in this study (Chen et al., 2013;Dana et al., 2019).The resulting higher signal-to-noise ratio might have led to more neurons being identified.Typically, neurons with low signal-to-noise ratio exhibit weak tuning (Gaucher et al., 2020;Extended Data Fig. 5-1), shifting the distribution imaged with GCaMP7f toward non-PT-responsive neurons.Furthermore, we used suite2p version 0.10.1 with improved cell detection which was not used by Gaucher et al. (2020), as it was not released at publication date, again increasing the number of cells with low signal-to-noise ratios.These aspects might, together with differences in classification methods used, also account for a lower amount of singlepeak neurons in our dataset (23%) compared with Gaucher et al. (2020), where 60% of PT-responsive neurons display a single-peak tuning pattern.As high signal-to-noise-ratio neurons typically exhibit the strongest responses (Issa et al., 2014;Bowen et al., 2020;Gaucher et al., 2020), the preference of them being detected over neurons with weaker responses shifts their dataset toward highly tuned neurons compared with ours.Interestingly, we found that neurons classified as single-peak actually displayed a reduction in tuning width, which is in contrast to the findings by Rotschafer and Razak (2013), who reported less precise tuning.Several factors might explain this discrepancy.First, Rotschafer and Razak (2013) pooled data across all layers with only 24 and 12 units (20 and 12% of all recordings) in L2/3 for WT and FMR1 KO, respectively, which hampers any conclusions regarding L2/3.Second, they conducted experiments with anesthetized mice, which influences the tuning properties of AC neurons, as somatostatin interneuron activity is vastly reduced during anesthesia (Adesnik et al., 2012), which in turn affects PV cell activity and consequently tuning BW.Whereas deletion of FMR1 in somatostatin interneurons did not result in an ASD-like behavioral phenotype, FMR1 deletion in PV interneurons caused typical behavioral abnormalities (Kalinowska et al., 2022).Furthermore, PV cells are decoupled from pyramidal neurons in early development in the somatosensory cortex in FMR1 KO mice (Kourdougli et al., 2023).Therefore, the observed narrowing of tuning BW in FMR1 KO in this study is possibly caused by an altered PV cell activity, enhancing lateral inhibition despite a reduced number of PV interneurons found in FMR1 KO and FXS patients (Wen et al., 2018;Kourdougli et al., 2023).
One of the major advantages of using two-photon imaging to study neuronal activity is the ability to precisely analyze the topographical order of single neurons within a network.While earlier works in the AC reported a quite unstructured singlecell tonotopy, displaying order only at a large scale (Bandyopadhyay et al., 2010;Rothschild et al., 2010;reviewed in Kanold et al., 2014), recent studies have favored a moderate tonotopy (Romero et al., 2019;Gaucher et al., 2020).Our data revealed a higher local tuning heterogeneity in A1 and AAF of FMR1 KO for single-peak neurons, pointing toward a worse tonotopic organization in these subfields.This is in line with a reduction in tonotopic organization in the primary AC fields observed in a valproic acid-induced rat model of ASD (Anomal et al., 2015), altogether implying that a less precise frequency-related AC topography might be a common occurrence in different forms of ASD-associated disorders.The most likely explanation for this not being observed for A2 in FMR1 KO is a narrower BF distribution in this subfield, which is a factor toward lower heterogeneity.While the direct cause of altered topographic organization and FRA shapes in FMR1 KO remain unclear at this point, they are probably caused by altered critical period plasticity during development, as observed in the AC (Kim et al., 2013) as well as the barrel cortex (Harlow et al., 2010).Alterations in topography might also be caused by a change in local connectivity patterns and integration from a broader frequency range.In V1, neurons with similar orientation tuning exhibit higher synaptic connection probabilities and stronger connections, which might be similar in the AC (Ko et al., 2011;Cossell et al., 2015;Gaucher et al., 2020).
Analyzing ensemble activity revealed PT-related changes in FMR1 KO mice as well.In A1, reliability of network responses as well as correlations between ensembles within a given sound cluster was lower.This was not observed in other subfields, which is in line with A1 coding particularly strong for the spectral component of sounds (Sołyga andBarkat, 2021, 2022), but also makes it difficult to relate these results to changes in BW and topographical order reported above, which also affected AAF.In contrast, ensemble activity in response to complex sounds was altered in AAF and A2 as well.In AAF, which is associated with processing of temporal features of sounds (Sołyga and Barkat, 2019, 2021, 2022), the specificity of ensemble activity between sound clusters was impaired.For A2, which is generally associated with encoding complex sounds, such as vocalizations (Carruthers et al., 2015) and harmonic sounds (Kline et al., 2021), correlations and reliability were actually higher in FMR1 KO mice.This is also interesting because task-relevant sounds modulate activity levels and elicit categorization stronger in A2 than in A1 (Atiani et al., 2014;Yin et al., 2020), which might be related to difficulties in decision-making reported in ASD (Luke et al., 2011;Geurts et al., 2020).These alterations in AC activity patterns might contribute to problems in speech perception and auditory hypersensitivity observed in FXS (Fidler et al., 2007;Finestack et al., 2009;Ethridge et al., 2017;McCullagh et al., 2020) and ASD (Alcántara et al., 2004;Groen et al., 2009;DePape et al., 2012;Schelinski and von Kriegstein, 2020).In line with this, FMR1 KO rats display degraded responses to speech sounds in A1, AAF, and A2 (Engineer et al., 2014).Interestingly, ASD individuals display abnormal auditory temporal and speech processing, but intact spectral processing (Groen et al., 2009), which might lead to the conclusion that slight alterations in activity correlation found in our study are less consequential for auditory processing compared with the decreased sound specificity found in AAF.This view is also supported by another study in FMR1 KO rats, which reported diminished temporal but enhanced spectral integration of sound intensity (Auerbach et al., 2021).However, Schelinski et al. (2017) reported changes in specifically vocal pitch processing in ASD, but not for other forms of pitch.Thus, it might be difficult to directly relate alterations in PT processing in FMR1 KO mice to sound frequency processing in the context of speech.
Cortical circuits are known to undergo representational drift, with neurons changing their response properties to stimuli while keeping overall representation of the stimuli constant to a large degree (Pérez-Ortega et al., 2021;Aschauer et al., 2022;Chambers et al., 2022).ASD is often associated with preservative thinking and repetitive behavior (American Psychiatric Association, 2013), and in children with ASD, learning is associated with more stable rather than plastic neural representation (Liu et al., 2023).Thus, investigating representational drift in FMR1 KO mice is a promising endeavor to uncover the underlying causes on a circuitry level.In fact, developmental plasticity is grossly impaired, as demonstrated by unaltered PT representation in early and late exposed FMR1 KO mice (Kim et al., 2013), and reduced adaptation to stimuli has been observed in the AC (Lovelace et al., 2016) as well as the somatosensory cortex (He et al., 2017).However, both the correlation of neuronal activity of the same set of neurons observed across 1 week in response to various sounds and the similarity of sound clusters between the 2 experimental days did not differ from WT controls, arguing against changes in either representational drift or general cortical sound representation.Nevertheless, changes in plasticity were observed when analyzing sound clusters in more detail.In A1 and AAF, correlations and reliability between sounds and repetitions either decreased or increased during our observation period in WT while staying constant in KO mice.In contrast, these values decreased in A2 in KO mice, but not in WT.These observations should only be regarded as a first assessment of network stability in FMR1 KO AC and need to be investigated in more detail, employing, for example, repeated sound stimulation over an extended time period.This is also important in the context of learning defects reported in the tactile (Arnett et al., 2014), visual (Goel et al., 2018;Kissinger et al., 2020), spatial (Nolan and Lugo, 2018), or auditory domain (Kim et al., 2013;S. Yang et al., 2014) in FMR1 KO.In future experiments, combining large-scale network activity imaging with learning paradigms will be a crucial step in relating the alterations in a passive listening situation reported here with their behavioral consequences.This approach would also help to investigate whether the changes observed in PT processing are related to the less precise topography in FMR1 KO.Such a correlation has been described in the visual cortex in FMR1 KO, relating orientation-tuning deficits and reduced activity of PV interneurons to delayed learning of a visual discrimination task (Goel et al., 2018).
A wide array of network alterations which might underlie the ensemble activity changes observed here have been reported in cortical circuits of FMR1 KO rodents.Studies in the somatosensory cortex revealed deficits in local excitatory drive targeting fast-spiking inhibitory neurons (Gibson et al., 2008), increased excitation-inhibition ratio (Antoine et al., 2019), weak Layer 4 to Layer 3 connectivity (Bureau et al., 2008), broadened receptive fields in L2/3 (Juczewski et al., 2016), absent pruning of connectivity in Layer 5 (Patel et al., 2014), abnormal synaptogenesis (Booker et al., 2019) and spine development (Nimchinsky et al., 2001), weak callosal projections (Zhang et al., 2021), and increased neuronal noise (Bhaskaran et al., 2023).Furthermore, interneurons and pyramidal neurons as well as pyramidal neurons themselves are desynchronized (Paluszkiewicz et al., 2011;Kourdougli et al., 2023), and sensory encoding precision is reduced (Domanski et al., 2019).In the visual cortex, hyperconnectivity and increased dendritic complexity (Haberl et al., 2015;C. Yang et al., 2022) as well as increased spine turnover (Ishii et al., 2018) were reported.Similar alterations have been observed in AC circuits as well, including an increase in spine density, with immature spine morphology (Lee et al., 2019), altered GABAergic signaling (Song et al., 2022), abnormal perineuronal net development around PV interneurons (Lovelace et al., 2016;Wen et al., 2018), and local hyperconnectivity, but long-range hypoconnectivity (Haberl et al., 2015).Interestingly, while forebrain-specific KO of FMR1 leads to similar changes in PV cells and perineuronal net structure, gamma band oscillations, typically associated with synchronized activity upon auditory stimulation, were not altered in contrast to decreased synchronization in the global KO (Lovelace et al., 2019).This points toward an important role of brainstem nuclei in FMR1 KO-related deficits and ASD (Seif et al., 2021;Jure, 2022), which might be the cause of some of the observations reported here, as abnormal activity patterns are transmitted via the thalamus up to the AC.We previously reported altered topography and sound clustering in mice missing the α 2 δ3 calcium channel subunit (Wadle et al., 2022), an ASD-associated mouse model that displays temporal sound-processing defects and synaptic transmission abnormalities in brainstem nuclei (Pirone et al., 2014;Bracic et al., 2022).However, cortical changes observed in this study differed regarding their subfield specificity, displaying decreased correlations in A1 in response to PTs and complex sounds, as well as a reduced heterogeneity of frequency tuning in AAF and A2.These differences highlight the wide array of neuronal alterations associated with ASD, but also point toward common phenotypes concerning auditory processing.Further studies investigating AC activity in multiple ASD mouse models are needed to pinpoint these further.Furthermore, the involvement of neuronal circuits outside of the auditory system needs to be considered as well, as highlighted by a recent study showing a developmental delay in temporal processing in FMR1 KO mice to be located in the frontal cortex rather than in the AC (Croom et al., 2023).
In conclusion, we demonstrate altered sound-evoked activity in AC networks of FMR1 KO mice, expressed in decreased tonotopy and changes in correlations within and between neuronal ensembles, depending on the subfield analyzed.These results provide detailed insight into AC activity in a mouse model of FXS and thus help to understand the causes of soundprocessing defects in ASD-associated disorders.

Figure 1 .
Figure 1.Characterization of PT response patterns.A, Top: Example FRAs for each tuning type in FMR1 KO and littermate control.Greyscale depicts normalized activity level.Dotted rectangle indicates BF.Bottom: Averaged activity across SPLs for each frequency (dots).Unimodal or bimodal Gaussian fits (line) were used to define tuning types according to their R 2 adj value.B, Fraction of PT-responsive neurons displaying single-peak, doublepeak, or irregular tuning properties.Samples are FOVs with at least 100 PT-responsive neurons per subfield.C, Tuning BW of single-peak neurons by subfield.S.L.W. performed experiments and analyzed data.See Extended Data Figure 1-1 for examples of original traces and deconvolution, Extended Data Figure 1-2 for widefield imaging to determine subfield borders, Extended Data Figure 1-3 for BW analysis after trace rescaling, and Extended DataTable 1-1 for statistics for panel B.

Figure 2 .
Figure 2. BF distributions and local BF heterogeneity.A, Subfield-specific BF distribution of single-peak neurons.B, Left: Two-photon maximal intensity projection of GCaMP signal across 1,000 frames in A1 of FMR1 KO with superimposed color-coded dots denoting BFs of single-peak neurons.Local IQR calculation of BFs is shown for two example neurons (white arrows) and the respective area considered for IQR analysis (cyan and purple dotted circles).Right: Subfield-specific IQR analysis of all single-peak neurons.S.L.W. performed experiments.S.L.W. and J.J.H. analyzed data.

Figure 3 .
Figure 3. Single-cell imaging of AP-evoked Ca 2+ amplitudes.A, Left: Electrical recording (bottom) and Ca 2+ fluorescence signals (top) of a single OGB-1-filled AC neuron in WT and KO.Right: Statistics for single AP-evoked Ca 2+ amplitudes.Numbers in bars depict number of recorded neurons.B, Left: Ca 2+ fluorescence signals during bursts of 10 APs at 5 or 10 Hz.Right: Statistics for single AP-evoked Ca 2+ amplitudes.Numbers in bars depict number of recorded neurons.T.C.R. performed experiments.T.C.R., S.L.W., and J.J.H. analyzed data.

Figure 4 .
Figure 4. Activity clustering of neuronal ensembles in response to PTs.A, Correlation matrices of 17 PT-evoked patterns after hierarchical clustering, recorded in A1 and A2 at 70 dB.Color code depicts correlation value.Vertical color bar at the bottom depicts clusters; blank stripes correspond to sounds that are not part of clusters.The diagonal depicts mean correlation across repetitions, thus showing reliability of the network.B, Statistics for sound cluster analysis for 70 dB PT sounds for A1.Numbers in bars depict n-number, either FOVs or sound clusters.C, Same as B, but for data recorded in AAF.D, Same as B, but for data recorded in A2.E, Statistics for sound cluster analysis for PTs played at different SPLs in A1.S.L.W. performed experiments.J.J.H. analyzed data.See Extended Data Tables 4-1 and 4-2 for statistics.

Figure 5 .
Figure 5. Activity clustering of neuronal ensembles in response to complex sounds.A, Correlation matrices of 13 complex sound-evoked patterns after hierarchical clustering, recorded in A1 and A2.Color code depicts correlation value.Vertical color bar at the bottom depicts clusters; blank stripes correspond to sounds that are not part of clusters.The diagonal depicts mean correlation across repetitions, thus showing reliability of the network.B, Statistics for sound cluster analysis for A1.Numbers in bars depict n-number, either FOVs or sound clusters.C, Same as B, but for data recorded in AAF.D, Same as B, but for data recorded in A2.S.L.W. performed experiments.J.J.H. analyzed data.See Extended Data Figure 5-1 for examples of correlation matrices for AAF and Extended Data Table 5-1 for statistics.

Figure 6 .
Figure 6.Activity clustering of neuronal ensembles in response to simple and complex sounds.A, Correlation matrix of 64 sound-evoked patterns (17 patterns for each of PTs, AM tones with 20 Hz modulation, AM tones with 40 Hz modulation, and 13 patterns for complex sounds) after hierarchical clustering, recorded in A1.Color code depicts correlation value.Vertical color bar at the bottom depicts clusters; blank stripes correspond to sounds that are not part of clusters.The diagonal depicts mean correlation across repetitions, thus showing reliability of the network.B, Cluster similarity (left) and neuron correlation (right) between 2 d separated by 1 week.C, Correlation values within clusters, reliability within clusters, and correlation values between clusters.S.L.W. performed experiments.J.J.H. analyzed data.See Extended Data Figure 6-1 for images of recorded neurons 1 week apart, Extended Data Figure 6-2 for statistics between the two recording time points and Extended Data Table 6-1-6-4 for statistics.