Ethics statement

All experiments were approved by the University of Illinois Urbana-Champaign Institutional Animal Care and Use Committee (IACUC) under protocol 20252, in accordance with NIH guidelines.

Subjects

Adult (>3 month old) male Fmr1^tm1sage knockout rats on an outbred Sprague-Dawley background (TGRS5390HTM4 FMR1 -/Y; SAGE Labs, St. Louis, MO) and littermate wild-type controls were used for these studies. Littermate male wildtype (WT) and Fmr1^-/y (KO) rats were generously donated from the laboratory of Dr. Richard Salvi. Male rats were used because FXS occurs more frequently and in greater severity in males due to the X-linked nature of the disorder [31]. Offspring were screened for a 122-base pair (bp) deletion in the Fmr1 gene sequence using published procedures [26,32].

9 WT and 10 Fmr1 KO rats were used for operant conditioning experiments. 6 WT and 5 Fmr1 KO rats were used for auditory brainstem response recordings. 5 WT and 5 Fmr1 KO rats were used for extracellular depth recordings. Rats lived in pair-housed caging in a colony room maintained at 22 °C with a 12-h light–dark cycle. All subjects had free access to food and water except for those undergoing operant conditioning, when rats were food restricted and kept at approximately 90% of their free-feeding weight. Food-restricted animals had unrestricted access to water, except while participating in behavioral testing. Testing sessions lasted approximately 1 h per day, with rats participating in one behavioral testing session per day, 6 days per week.

Operant conditioning

Rats were first trained in a Go/No-Go operant conditioning paradigm to detect tone bursts (300 ms, 5 ms rise/fall time, cosine gated) of varying frequency (4, 8, 16, 32 kHz) using procedures similar to those described in our previous publication [26]. Testing was carried out in operant conditioning chambers (Med-Associates, Model ENV-008-VP, St. Albans, VT) equipped with pellet dispensers (Med-Associates, Model ENV-203 M, St. Albans, VT), illuminated nose-pokes with infrared sensor (Med-Associates, Model ENV-114BM, St. Albans, VT), cue lights for signaling task status (Med-Associates, Model ENV-215M-LED, St. Albans, VT), and speakers (Fostex F17H Horn Super Tweeter, Tokyo, Japan) for sound generation. All components were housed in single-walled sound-attenuating cubicles equipped with ventilation (Med-Associates, Model ENV-018MD, St. Albans, VT). Each behavioral setup was independently controlled by a dedicated microcontroller (Arduino DUE, Italy) that recorded signals from the behavioral devices, responded according to predefined protocols, and communicated with a host computer to transmit control signals and data. The entire system was controlled by custom software written in MATLAB (MathWorks, R2022, USA), which handled system control, stimulus generation, and data collection and visualization. Sound stimuli (192 kHz sampling rate) were generated and delivered using a multi-channel sound card (RME MADIface Pro, Germany), multi-channel digital-to-analog converter (RME M-32 DA Pro, Germany), power amplifiers (Behringer NX1000, Germany), and speakers (Fostex F17H Horn Super Tweeter, Tokyo, Japan) positioned approximately 30 cm above the animal’s head. Stimuli were calibrated using a microphone preamplifier (Larson Davis, Model 2221, Depew, NY) equipped with a ½″ microphone (Larson Davis, Model 2520, Depew, NY) at a location where the animal’s head would be during a trial.

A rat began a trial by placing its nose in the nose-poke hole, which initiated a variable waiting interval ranging from 1 to 4 s. During the waiting interval, the rat had to maintain its position in the nose-poke hole until it heard a tone burst or the trial was aborted. In the Go condition, the target stimulus was the tone burst. If the rat detected this signal, it removed its nose from the nose-poke hole resulting in a food reward (45 mg dustless rodent grain pellets, Bio-Serv, Flemington, NJ); a hit was recorded if the rat correctly responded to the tone within 2 s. A miss was recorded if the rat failed to remove its nose from the nose-poke within the 2 s response interval. No reinforcement was given for a miss. Approximately 30% of all trials were catch trials where tone bursts were not presented. This constituted the No-Go part of the procedure. If the rat removed its nose during a catch trial, a false alarm (FA) was recorded and the rat received a 4–8 s timeout, during which the house light was turned off and the rat could not start another trial. However, if the rat continued to nose-poke, a correct rejection (CR) was recorded. No reinforcement was given for a correct rejection. After initial training using a 70 dB SPL Go stimulus (training criteria: >240 trials, >90% hit rate, <25% FA rate over 5 consecutive days), the range of stimulus intensities was expanded to 30–90 dB SPL, presented in 10-dB steps, and then to 15–45 dB, presented in 5 dB steps, to obtain a detection threshold for each frequency. If necessary, the sound level range was further reduced in 5 dB steps until thresholds were obtained. The tone bursts were presented according to the psychophysical Method of Constant Stimuli. Within each 10-trial block, seven predetermined target intensities were presented randomly along with 3 catch trials without sound stimulus.

Following collection of tone detection thresholds, the rats were moved to the discrimination phase of the Go/No-Go task. In this phase, the rats were trained to respond to a single Go tone frequency (4, 8, 16, or 32 kHz) while the No-Go stimulus was a tone either 1 octave above or below the Go tone. Yoked pairs of WT and Fmr1 KO rats were trained on one of six potential Go/No-Go tone pairings (4/8 kHz, 8/4 kHz, 8/16 kHz, 16/8 kHz, 16/32 kHz, 32/16 kHz). Both the Go and No-Go stimulus were presented at 40 dB above sensation level (SL), based on the individual subject’s detection thresholds acquired above. If the rat responded to the No-Go tone by removing its nose from the nose-poke hole within a 2 s response window, this was recorded as a FA and the rat received a 4–8 s time-out. The percentage of No-Go trials presented was incrementally increased over each session as the animals improved on the discrimination task until sessions were 50:50 Go/No-Go trials. Once rats reached criteria (>200 trials, >90% hit rate, <25% FA rate over 5 consecutive days), they could begin the discrimination testing phase. In the first portion of this testing phase, 5 new No-Go tones were introduced to the Go/No-Go task for 2 consecutive sessions. These tones were equally spaced at 1/6th octave steps between the Go tone and original No-Go tone (e.g., Go: 8 kHz, No-go: 8.9, 10.1, 11.3, 12.7, 14.3, 16 kHz). Rats were then tested in a more fine-grained version of the task using 12 No-Go tones equally spaced in 1/12th octave steps above or below the Go tone. This fine feature phase was split over three sessions, where the rats were tested on No-go tones >2/3 octave from Go (e.g., Go: 8 kHz, No-go: 13.5, 14.3, 15.1, 16 kHz), 1/3–2/3 octave from Go (e.g., Go: 8 kHz, No-go: 10.7, 11.3, 11.9, 12.7 kHz), or <1/3 octave from Go (e.g., Go: 8 kHz, No-go: 8.5, 8.9, 9.5, 10.1 kHz) on any given session in pseudorandom order. Fine-frequency testing sessions were interleaved with 1 octave Go/No-Go training sessions to ensure consistent reinforcement. For all discrimination testing sessions, trials were 50:50 Go/No-Go and if the subject responded to any of the No-Go tones within the 2 s response window, a FA was recorded and the animal received a 4–8 s timeout.

Auditory Brainstem Response (ABR) recordings

Animals were anesthetized with a cocktail containing ketamine (80 mg/kg i.p.) and xylazine (6 mg/kg i.p.). Platinum subdermal needle electrodes (S83018-R9-10, Horizon) were placed behind each ear to record auditory brainstem response (ABR), with ground and reference electrodes placed on the rear of the body and the vertex of the head, respectively. Body temperature was maintained at 37 °C using a homoeothermic heating blanket (Harvard Apparatus, Cambridge, MA). Tone bursts (5 ms, 1 ms rise/fall time, cosine gated) were generated by a RX6-2 multifunction processor (Tucker-Davis Technology, Alachua, FL) at approximately 250 kHz sampling rate and presented at a rate of 21/s. The output of the D/A converter was routed through a programmable attenuator (PA5, Tucker-Davis Technology, Alachua, FL) and amplifier (SA1, Tucker-Davis Technology, Alachua, FL), then delivered through a loudspeaker (MF1, Tucker-Davis Technology, Alachua, FL) located 10 cm from the test ear, while the other ear was plugged. The stimulation intensity varied from 0 to 90 dB SPL with 10-dB steps from 30-90 and 5 dB steps from 0 to 25 dB. Sound levels were calibrated with a ½″ microphone (Larson Davis, Model 2540, Depew, NY), a microphone preamplifier (Larson Davis, Model 2221, Depew, NY) and custom sound calibration software written in MATLAB (MathWorks, R2022, USA). The signals from the electrodes were amplified and digitized using a Medusa4Z amplifier and RZ2 Bioamp processor (Tucker-Davis Technology, Alachua, FL). All experiments were conducted in a double-walled sound-insulated chamber. Signals were band-pass filtered (300–3000 Hz) and averaged over 500 repeats using custom written MATLAB script that allows for the visual inspection of average ABR waves at each intensity-frequency combination. ABR threshold was defined as the lowest level that produced a noticeable ABR waveform, determined by visual inspection by a trained experimenter blind to genotype. Thresholds were determined for both ears and then averaged within each animal, except for two WT rats in which measurements were limited to a single ear due to insufficient anesthesia depth that precluded bilateral testing.

Extracellular depth recordings

Extracellular recordings of spontaneous and tone-evoked neural activity were performed in anesthetized animals as described in previous publications [33,34]. Rats were anesthetized with a cocktail of ketamine (80 mg/kg, i.p.) and xylazine (6 mg/kg, i.p.) and placed in a stereotaxic frame with blunted ear bars (Kopf Instruments, Tujunga, CA). Supplementary doses of ketamine/xylazine (20/2 mg/kg, i.p.) were administered as needed to maintain a stable plane of anesthesia. Body temperature was maintained at 37 °C using a homoeothermic heating blanket (Harvard Apparatus, Cambridge, MA). The dorsal surface of the skull was exposed, the temporal muscle was removed, and craniotomies were made over the left or right (contralateral to the sound source) auditory cortex (ACx) and inferior colliculus (IC) based on stereotaxic coordinates [35]. The skull was carefully opened and the dura was removed from the surface of the cortex. A custom designed head-bar was firmly attached to the skull using dental cement to allow for acoustic stimulation using a free-field loudspeaker.

Two single-shank 16-channel linear silicon microelectrode arrays (A-1 × 16–10 mm 100–177, NeuroNexus Technologies, Ann Arbor, Michigan) were inserted tangential to the surface of the ACx (approximately 40^o) and IC (90^o) to record multi-unit spiking activity. Recordings were made at several different positions along the rostral-caudal axis of each area (ACx: 4.0–6.0 mm caudal from Bregma in 0.5 mm steps,  approximately  7.6 mm lateral from midline; IC: 8.4–9.6 mm caudal from Bregma in 0.4 mm steps, approximately 1.8 mm lateral from midline). Electrodes were slowly advanced using a hydraulic micromanipulator (FHC, Bowdoinham, ME) to a depth of approximately 1.6 mm in the ACx or approximately 3 mm in the IC so that for each penetration, the electrode shank spanned layers of the ACx and iso-frequency lamina of the IC (see Fig 3A). Recordings were targeted to primary ACx and the central nucleus of the IC based on stereotaxic coordinates, response latency, and tuning properties. For each penetration, electrodes were allowed to settle for at least 30 min prior to data collection.

To examine rate level functions (RLFs) and frequency response areas (FRAs), tone bursts (50 ms, 1 ms rise/fall time, cosine gated) were generated by a TDT RX6-2 multifunction processor (approximately 250 kHz sampling rate) and delivered at a rate of 3/s through a loudspeaker (MF1, Tucker-Davis Technologies, Alachua, FL) located 10 cm from the ear contralateral to the recording hemisphere. Sound levels at the position of the ear were calibrated using a microphone preamplifier (Larson Davis, Model 2221, Depew, NY) equipped with a ½″ microphone (Larson Davis, Model 2520, Depew, NY). Tone stimuli spanned from 1 to 64 kHz in 20 log steps, presented from 0 to 90 dB SPL in 10 dB steps. Thirty repeats of each frequency–intensity combination were presented in pseudo-random order. Recorded neural signals were amplified by a RA16PA and sampled at 25 kHz by a RZ5 Bioamp processor (Tucker-Davis Technologies System-3, Alachua, FL). Spike detection was performed online using a manually set voltage threshold (spike signal filtered 300–3,500 Hz). Custom-written MATLAB software (MathWorks, R2022, USA) was used to acquire the neural data as previously described [33,36]. All experiments were conducted in a double-walled sound-insulated chamber.

Data analysis

Multi-unit spiking activity recorded in response to tone bursts was analyzed using a custom-built Python (3.10) pipeline. Spiking data was extracted into a tuning matrix (dB × kHz) for each multi-unit, with each cell corresponding to 30 spike trains from −40 ms pre-stimulus to 100 ms post-stimulus for each presentation of the current dB/kHz combination. Peristimulus time histograms (PTSHs) were then calculated for each cell using 5 ms bins. Cells were determined to have evoked spiking activity if one 5 ms bins from 0 to 40 ms post-stimulus exceed 4 standard deviations above the mean pre-stimulus activity [37].

Rate level functions (RLFs).

After data extraction, RLFs were produced for each frequency column of the dB × kHz matrix. Frequency specific input/output (I/O) functions were calculated as firing rate change over increasing sound intensity for a given frequency (4, 8, 16, and 30 kHz). In order to compare across units with different tuning properties, I/O functions were constructed for each multi-unit cluster using only the frequency column corresponding to the unit’s characteristic frequency (CF) (see CF methods below). This I/O function was then fit with the following 6 parameter Gaussian model [38]:

(1)

Where is the lower asymptote, is the upper asymptote, is the inflection point, is the variance, is the amplitude, and is the center of the Gaussian. Response minimum and maximum were defined as the parameters and , respectively. The threshold was calculated as 20% of the range from the minimum asymptote to the maximum [38]. Gain was taken as the slope of the ascending portion of the RLF [38]. A root mean squared error (RSME) cutoff of <100 was used to limit analysis only to RLFs adequately fit by the model (233/336 ACx units and 166/192 IC units from WT animals; 208/272 ACx units and 153/176 IC units from Fmr1 KO animals)

Spike train discriminability.

In order to determine how well a multi-unit cluster could discriminate between different frequency tones, neural discriminability was quantified using SPIKE-distance, which calculates the dissimilarity between spike train pairs based on differences in spike timing and instantaneous firing rate [39]. For a given spike train, the instantaneous spike timing difference at time t is given as:

(2)

where corresponds to the distance between the spike before time in spike train 1 and the closest spike from train 2 and corresponds to the same distance, but for the spike following time . S₂(t) takes the same form, but from the view of the second spike train in the pair with respect to the first. The two train contributions are then averaged and normalized by the mean to account for relative distances within spike trains. Because mean firing rates were significantly different between WT and Fmr1 KO animals, rate independent SPIKE-distance was used. In this manner, S₁(t) and S₂(t) are left unweighted by the instantaneous ISI:

(3)

Distance calculations were performed by comparing the block of 30 spike trains at a unit’s CF to the spike trains from adjacent frequencies. All comparisons were made for tone-evoked spike trains at 40 dB above minimum threshold (corresponding to the intensity used for behavioral discrimination experiments). In this manner, a 30 × 30 matrix for every pairwise distance frequency combination relative to CF was produced. Due to the large number of spike trains being compared (N > 2), the instantaneous average over all pair combinations of instantaneous spiking disparities was taken:

(4)

Lastly, these instantaneous averages were then integrated across time:

(5)

Population decoder

To characterize the relationship between neural tuning and behavioral discrimination, we constructed a Bayesian population decoder consisting of a population of readout neurons that received frequency tuned input from either ACx or IC recorded neural data [40]. First, we determined the response maximum, minimum, and bandwidth for each multi-unit cluster at 40 dB above threshold (corresponding to the intensity used in our behavioral discrimination experiments) from the ACx or IC recordings to fit asymptotic Gaussians models of the firing data using the following equation:

(6)

Where is the amplitude (peak firing rate), is in the input variable (tone frequency), is the mean center of the Gaussian (best frequency), indicates the variance (tuning width), and is the asymptote (baseline firing rate). Given that the tuning edges were defined as the first and last frequencies that exhibit an evoked response (regardless of the peak response size), the bandwidth value informing the Gaussian width would be equivalent to the first divergence of the Gaussian from the horizontal asymptote. This cannot be analytically solved for but can be approximated as the 99% confidence interval expressing the cumulative density function in terms of the error function ():

(7)

We can then solve for the corresponding sigma by using 99% as the .

Next, two populations of 640 neurons were generated with preferred frequencies equally log-spaced from 1 to 64 kHz. Each neuron was given a fit to the experimentally measured mean tuning curves using either WT or KO Gaussian parameters from the recorded ACx or IC neural data. The simulated neurons were presented with either a Go stimulus (4, 8, 16 or 32 kHz) or No-Go stimuli spaced in 1/12 octave steps up to 1 octave above and below the Go stimulus, paralleling our behavioral paradigm. Neuron activity was defined by the Gaussian tuning curve value for each presented stimulus. This frequency-tuned activity from each neuron was then fed into a read-out layer with the assumption that the activity value for each neuron was the mean for a Poisson process. The output of each neuron in each trial was simulated by drawing a random number from this Poisson distribution with the condition that single trial activity level must surpass a threshold value (90% of the baseline firing rate) to be propagated to the readout layer. This was repeated to produce 10,000 trials for each neuron. Lastly, Bayesian decoding was used to calculate the probability that the current stimulus was a Go or No-Go based on the activity levels from this simulated neuron population. This process was repeated 15 times to illustrate model consistency and allow for statistical testing.

Statistical analyses

Statistical analyses were performed in Python (3.10). For behavior and ABR data, two-way ANOVAs and Kruskal–Wallis Rank Sum tests were used to determine significant main effects of genotype and sound parameter (frequency/intensity), depending on normality as determined by the Shapiro–Wilks test, with Bonferroni-corrected post-hoc comparisons when applicable. Student’s t-tests or Wilcoxon Rank Sum tests were used for single comparisons, depending on the normality. For neural data, a general linear mixed model (GLMM) analysis of variance was used to assess significant main effects of genotype and sound parameter, with individual rat ID, penetration number, and cluster number as random effects. This hierarchical approach allows for statistical comparisons at the subject level while taking into account potential within-subject variability at the unit and penetration level. All statistical values are reported as means ± standard error (SEM), unless otherwise stated. Boxplots include median (gray dash), 25th–75th percentiles (box), full range (whiskers), and outliers (below Q1 − 1.5 * IQR or above Q3 + 1.5 * IQR). Box plot dots in Fig 1 correspond to animal. Box plot dots in Figs 2 and 3 correspond to multi-unit cluster. Box plot dots in Fig 4 correspond to a full 10,000 trial run of the model for the given parameters.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Normal tone detection but impaired discrimination in Fmr1 KO rats.

(A) Schematic of Go/No-Go operant tone detection task. 9 wildtype (black) and 10 Fmr1 KO (red) rats were trained to report the detection of any tone burst (HIT), with failure to do so resulting in a MISS. On 30% of trials no sound was presented (catch trials). Responding on catch trials resulted in a false alarm (FA), while refraining from responding resulted in a correct rejection (CR). (B) Average tone detection performance across animals for all tone frequencies (4, 8, 16, and 32 kHz). Detection performance was comparable between wildtype (black) and Fmr1 KO (red) animals. (C) Behavioral detection thresholds for each tone frequency using a criterion of d′ = 1.5 (dashed line in Fig 1B). No genotype difference was observed at any test frequency. (D) Schematic of auditory brainstem response (ABR) recording setup. (E) Representative ABR waveforms from a wildtype (black) and Fmr1 KO (red) rat. ABR threshold was defined as the lowest intensity that evoked a discernable ABR waveform (black and red arrows) (F) No difference in ABR thresholds between wildtype (black) and Fmr1 KO (red) rats was observed at any sound frequency tested. (G) Schematic of Go/No-Go operant discrimination task. In this task, rats were trained to report the detection of a single Go-tone frequency (4, 8, 16, or 32 kHz) and inhibit their response to a No-Go tone either 1 octave above or below the Go tone. (H) Fine-frequency discrimination task where No-Go tone frequency was varied in 1/12 octave steps from Go tone. Top: Schematic showing the addition of multiple octave steps between the Go and No-Go tones (1/12 of an octave per step). Bottom: FA rate as a function of No-Go tone frequency in wildtype (black) and Fmr1 KO (red) rats. (I) FA rates grouped by 1/3 octaves from Go showing a significant reduction in Fmr1 KO performance only in the optimally difficult middle band, while performance is comparable at the most difficult and easy 1/3 octave bands. Box plots represent the median, 25th, and 75th percentiles. Whiskers represent the minimum and maximum values except for outliers. Boxplots dots represent individual animals. All other values are means ± SEM. *p < 0.05, ***p < 0.001, ns = not significant. Source data for panels C, F, and I are in S1 Data, Fig 1 sheet.

https://doi.org/10.1371/journal.pbio.3003248.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Altered cortical response properties in Fmr1 KO rats.

(A) Schematic of recording set-up. Simultaneous recordings with multichannel electrodes were made from across the tonotopic axis of contralateral auditory cortex (ACx) and inferior colliculus (IC) of 5 wildtype and 5 Fmr1 KO rats. Recordings were made with single shank linear silicon probes that spanned the dorsal-ventral axis of the IC or ACx, with multiple penetrations being made at different positions along the rostral-caudal axis of each area. (B–C) Rate level functions showing relationship between firing rate and sound intensity at characteristic frequency (CF) for each multi-unit cluster in the (B) IC and (C) ACx of wildtype (black) and Fmr1 KO (red) rats. (D–E) Interpolated response minimum (Min) and maximum (Max) from (D) IC and (E) ACx response functions. (F–G) Interpolated response threshold (Thresh) and slope (Gain) from (F) IC and (G) ACx response functions. Box plots represent the median, 25th, and 75th percentiles. Whiskers represent the minimum and maximum values except for outliers. Boxplot dots represent individual multiunit clusters. All other values are means ± SEM. *p < 0.05, **p < 0.01, ***p < 0.0001, ns = not significant. Data and code underlying this figure can be found at http://doi.org/10.5281/zenodo.15559344. Source data for panels D–G are in S1 Data, Fig 2 sheet.

https://doi.org/10.1371/journal.pbio.3003248.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Broader frequency tuning in auditory cortex of Fmr1 KO rats despite unaltered subcortical tuning properties.

(A) Example tuning curves recorded from the inferior colliculus (IC) of a wildtype (left) and Fmr1 KO (right) rat. Each cell represents 30 trials for a given frequency-intensity combination. (B) Distributions of characteristic frequency (CF, left) and minimum threshold (right) for multi-unit clusters from the IC of wildtype (black) and Fmr1 KO (red) rats. (C) Q-value measure of tuning precision at 10, 20, 30, and 40 dB above response threshold for IC units. (D) Neural discriminability of sound frequency in IC as assessed by changes in spike train dissimilarity (Δ Spike-Distance) in response to CF and neighboring tone frequencies. (E) Example tuning curves recorded from the auditory cortex (ACx) of a wildtype (left) and Fmr1 KO (right) rat. Each cell represents 30 trials for a given intensity frequency combination. (F) Distributions of characteristic frequency (CF, left) and minimum threshold (right) for multi-unit clusters from the ACx of WT (black) and Fmr1 KO (red) rats. (G) Q-value measure of tuning precision at 10, 20, 30, and 40 dB above response threshold for ACx units. Lower Q-values in the ACx of Fmr1 KO rats are indicative of broader tuning. (H) Neural discriminability of sound frequency as assessed by changes in spike train dissimilarity (Δ Spike-Distance) in response to CF and neighboring tone frequencies. Decreased Spike-Distance in the ACx of Fmr1 KO rats is indicative of poorer neural discriminability. Box plots represent the median, 25th, and 75th percentiles. Whiskers represent the minimum and maximum values except for outliers. Boxplots dots represent individual multiunit clusters. All other values are means ± SEM. **p < 0.01, ***p < 0.0001, ns = not significant. Data and code underlying this figure can be found at http://doi.org/10.5281/zenodo.15559344. Source data for panels B and F are in S1 Data, Fig 3 sheet.

https://doi.org/10.1371/journal.pbio.3003248.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Modeling Fmr1 KO discrimination deficits using a population decoder.

(A) Schematic of Bayesian decoding from a simulated network of tonotopically organized neurons whose tuning parameters are derived from the population data recorded from the auditory cortex (ACx) or inferior colliculus (IC) of either WT or Fmr1 KO rats. (B–C) Error rates from model readout layer as a function of octave distance using tuning parameters from the (B) ACx or (C) IC of WT (black) or Fmr1 KO (red) rats. (D) Decoder performance for No-Go tones 1/3–2/3 octave from Go tone as a function of model parameters in ACx. Error rate was determined for models using all WT (WT_full) or KO (KO_full) parameters, as well as for each unique combination of individual KO parameters for spontaneous firing rates (spont), peak sound-evoked firing rate (evoked), and tuning width (width). Changing model parameters significantly impacted decoder performance (Kruskal–Wallis Test, ***p < 0.0001), with KO_full (Dunn’s test, ***p < 0.0001), KO_width (Dunn’s test, *p = 0.0144) and KO_{spont + width} (Dunn’s test, ***p < 0.0001) permutations being significantly different from WT_full. Gray and pink dashed lines represent median WT_full and KO_full error rates. Box plots represent the median, 25th, and 75th percentiles. Whiskers represent the minimum and maximum values except for outliers. Boxplots dots represent separate model runs (10,000 repeats each). All other values are means ± SEM. *p < 0.05, **p < 0.01, ***p < 0.0001, ns = not significant. (E) Schematic summarizing results. Cortical gain enhancement in FXS may act to preserve signal-to-noise ratios and signal detection threshold at the expense of sound sensitivity and fine feature discrimination. Data and code underlying this figure can be found at http://doi.org/10.5281/zenodo.15559344. Source data for panel D are in S1 Data, Fig 4 sheet.

https://doi.org/10.1371/journal.pbio.3003248.g004

Tone detection thresholds are unaltered in Fmr1 KO rats

The ability to discriminate between sound frequencies is directly related to a subject’s hearing thresholds, with poorer hearing thresholds associated with a decreased ability to detect small changes in frequency [41,42]. It was therefore essential to first characterize tone detection thresholds in Fmr1 KO rats before examining any potential differences in frequency discrimination. Male Fmr1 KO (n = 10) and littermate WT (n = 9) rats were first trained to detect tone bursts (4, 8, 16, and 32 kHz) using a Go/No-Go operant conditioning paradigm (Fig 1A). Both WT and Fmr1 KO rats learned this task at a comparable rate (WT: 16.9 ± 1.22 days to criteria, KO: 16.6 ± 1.45 days to criteria; Wilcoxon Rank Sum: p = 0.645) and reached similar levels of peak performance (WT d′ = 3.70 ± 0.09, KO d′ = 3.47 ± 0.11; Wilcoxon Rank Sum: p = 0.121), consistent with our previous results [26]. In order to assess tone detection thresholds, tone stimuli were presented at near threshold intensities and psychometrics curves were constructed by plotting d′ as a function of sound intensity (Fig 1B). A conservative cutoff of 1.5 for d′ was used to calculate detection thresholds [43]. There was no significant genotype difference in psychometric functions (Fig 1B; Kruskal–Wallis: df = 1, p = 0.679) or detection thresholds (Fig 1C; Kruskal–Wallis: df = 1, p = 0.974), consistent with our previous results [26].

In addition to behaviorally measured subjective hearing thresholds, we recorded auditory brainstem responses (ABRs) in a separate group of male Fmr1 KO (n = 5 rats, 10 ears) and WT (n = 6 rats, 10 ears) littermates (Fig 1D–1F). ABRs are a clinically used, non-invasive physiological measure of the functional status of the lower auditory pathway that allows for objective assessment of hearing thresholds (Fig 1E) [44]. We found no difference in ABR thresholds between WT and Fmr1 KO animals across any tone frequency tested (Fig 1F; 2-way ANOVA, genotype: df = 1, F = 0.867, p = 0.357; genotype~frequency: df = 40, F = 1.22, p = 0.316), consistent with our behavioral results. These results confirm that Fmr1 KO rats have unaltered hearing thresholds, allowing for the examination of tone discrimination performance without the confound of differences in sensation level.

Fmr1 KO rats exhibit deficits in fine frequency discrimination

After ensuring similar hearing capabilities, we next moved the same group of behaviorally trained 9 WT and 10 Fmr1 KO rats to a Go/No-Go tone discrimination task. In this phase of the task, animals were trained to respond to a single Go tone frequency (4, 8, 16, or 32 kHz, randomly assigned to genotype pairs) and inhibit their response to a No-Go tone whose frequency was 1 octave above or below the target Go tone (Fig 1G). Both genotypes were able to reach and maintain successful criteria (WT: 14.43 ± 2.94 days to criteria, KO: 13.71 ± 2.58 days to criteria; t test: t = 0.183, df = 18, p = 0.858), with no difference in discrimination performance when tones were separated by 1 octave (S1A Fig; WT d′ = 3.39 ± 0.10, KO d′ = 3.46 ± 0.12; t test: t = 0.427, df = 18, p = 0.675). We next tested the limits of their discrimination by keeping the Go tone constant but bringing the frequency of the No-Go tone progressively closer to the Go tone frequency (Fig 1H). In this testing phase of the task, we found that Fmr1 KO animals performed significantly worse than WT littermates (Fig 1H; 2-way ANOVA, genotype: df = 1, F = 11.104, ***p = 0.001). Importantly, other aspects of task performance were not altered in WT or Fmr1 KO rats on discrimination testing days, including hit rate (S1B Fig), trials initiated (S1C Fig), and reaction time (S1D Fig). To further examine discrimination performance, we binned false alarms (FA) by their tonal distance from the Go frequency (Fig 1I). When the No-Go frequency was well separated from the Go tone (>2/3 octave separation), both genotypes performed well, as indicated by FA rates generally below 25%, consistent with octave discrimination performance (t test: t value = 0.410, p = 0.688). Conversely, when the No-Go frequency was very close to the Go tone (<1/3 octave separation), neither genotype performed above chance levels, as indicated by the high FA rates in both WT and Fmr1 KO animals (t test: t = 0.894, p = 0.386). However, for intermediate frequencies (between 1/3 and 2/3 octave separation), WT animals were significantly better at discriminating Go from No-Go tones than Fmr1 KO littermates, as evidenced by significantly lower FA rates (t test: t value = 2.186, *p = 0.044). Together, these results indicate that Fmr1 KO rats have impaired fine feature discrimination despite having normal tone detection thresholds and learning rates.

Divergent cortical and subcortical evoked response properties in Fmr1 KO rats

We next sought to determine the neural mechanisms underlying impaired tone discrimination in Fmr1 KO rats. FMRP is widely expressed in the auditory system and loss of FMRP expression has been shown to alter neuronal function across the auditory neuroaxis, including the brainstem [45–48], midbrain [49–52], and cortex [5,23]. We therefore performed simultaneous extracellular recordings of multi-unit spiking activity in the inferior colliculus (IC) and auditory cortex (ACx) of ketamine/xylazine anesthetized Fmr1 KO (n = 272 ACx and 176 IC multiunit clusters from 5 rats) and WT (n = 336 ACx and 192 IC multiunit clusters from 5 rats) littermates. Multiunit spiking activity was assessed in response to tone bursts (50 ms) presented over a broad range of frequencies (1–64 kHz) and intensities (0–90 dB SPL) in order to examine tone-evoked activity and frequency tuning. The auditory system is tonotopically organized, with neurons tuned to different best frequencies along the dorsal-ventral axis of the IC or the rostral-caudal axis in the ACx. To account for this, multiple penetrations were made along the rostral-caudal axis of the IC and ACx, using 16 ch linear probes spanning the dorsal-ventral axis of each structure (Fig 2A). For each electrode penetration in the IC, activity was sampled across iso-frequency lamina. In the ACx, each penetration sampled activity across cortical layers from neurons roughly tuned to the same best frequency, which varied across penetrations. In this manner, we were able to assess response properties and tuning qualities at multiple levels of the auditory system across the tonotopic axis.

We constructed tone-evoked rate level functions (RLFs) from the IC (Fig 2B) and ACx (Fig 2C) by plotting the mean driven discharge rate at the unit’s characteristic frequency (CF) as a function of tone intensity. No genotype difference was observed for RLFs collected from the IC (Fig 2B; GLLM: p = 0.134). However, we saw that tone-evoked spiking activity was significantly elevated in the ACx of Fmr1 KO animals across intensities (Fig 2C; GLMM; genotype: *p = 0.029; genotype~intensity: ***p < 0.0001). Similar results were found when examining RLFs from all multiunit clusters in response to specific tone frequencies (S2 Fig). To further quantify response properties, RLFs for each multiunit cluster were fit with a 6 parameter gaussian function to interpolate response minimum, maximum, threshold, and gain (see “Materials and methods”). There was no significant genotype difference in any parameter for IC tone-evoked RLFs (Fig 2D and 2F; GLMM: min p = 0.663, max p = 0.762, thresh p = 0.675, gain p = 0.673). In the ACx, both response minimum (GLMM: *p = 0.022) and maximum (GLMM: **p = 0.007) were significantly elevated in Fmr1 KO rats (Fig 2E). Cortical response gain was also significantly elevated in Fmr1 KO rats (GLMM: **p = 0.010) without a change in threshold (GLMM: p = 0.300) (Fig 2G). These results indicate that the ACx is in a hyperexcitable state in Fmr1 KO rats that affects both spontaneous and sound-drive spike rates, resulting in an additive shift in baseline firing rates as well as a multiplicative shift in response gain.

Degraded cortical frequency tuning in Fmr1 KO rats

While the above results highlight a potential cortical locus for sound hypersensitivity in Fmr1 KO rats [26], these changes in evoked response size are not necessarily indicative of differences in frequency tuning or discrimination. To directly assess tuning properties in the IC and ACx of Fmr1 KO rats, frequency response areas (FRAs) were examined by assessing the mean sound-evoked firing rate at each frequency-intensity combination for each multiunit cluster. Tuning properties were quantified in the following ways. First, response bandwidth was characterized by determining the Q-values for each multiunit cluster at multiple intensities (10–40 dB) above threshold. Q-values measure the sharpness of a tuning curve while accounting for differences in characteristic frequency (CF), with higher Q-values indicative of more precise (i.e., narrower) tuning. In addition, we employed a spike-distance-based approach to determine how well a given multiunit cluster could discriminate between different frequency tones based on the dissimilarity of the unit’s spike trains (see “Materials and methods”).

Representative FRAs collected from the IC of WT and Fmr1 KO rats are shown in Fig 3A. We found a similar distribution of CF (GLLM: p = 0.376) and threshold (GLLM: p = 0.245) for multiunit clusters obtained from WT and Fmr1 KO animals, indicating there was no bias in our sampling of responses across tonotopic regions (Fig 3B). Q-values became smaller in both genotypes at higher intensities (GLMM: ***p < 0.0001), indicative of broader tuning at higher sound levels and consistent with classic V-shaped tuning curves (Fig 3C). However, no genotype difference in Q-values was observed at any intensity level in the IC (GLLM: genotype: p = 0.121; genotype~dB: p = 0.094). Likewise, spike distance increased for tones that were further from the CF of each unit in both genotypes (Fig 4D, GLMM: ***p < 0.0001), indicating that neural discrimination improved as a function of frequency separation. However, there was no significant genotype difference in spike distance for IC units (Fig 3D, GLMM: p = 0.365). Together these findings show that frequency tuning is unaltered in the IC of Fmr1 KO rats.

Representative FRAs collected from the ACx of WT and Fmr1 KO animals are shown in Fig 3E. Again, we found a similar distribution of CF (GLLM: p = 0.568) and threshold (GLLM: p = 0.190) in both genotypes, indicating unbiased sampling (Fig 3F). However, contrary to what was found in the IC, tuning curves appeared broader in the ACx of Fmr1 KO animals. Indeed, the Q-values were significantly lower in Fmr1 KO rats across intensity levels, indicative of larger bandwidth (Fig 3G, GLMM: genotype ***p < 0.0001; genotype~intensity ***p < 0.0001). Finally, while spike-distance for ACx units once again increased as a function of frequency separation from CF in both genotypes (GLMM: ***p < 0.0001), the dissimilarity was greater for units from WT compared to Fmr1 KO animals (GLMM: **p = 0.002), indicative of poorer neural discrimination in FXS (Fig 3H). Thus, despite normal subcortical tuning properties, units are more broadly tuned with poorer frequency discriminability in the ACx of Fmr1 KO animals.

Cortical tuning differences can account for frequency discrimination deficits in Fmr1 KO rats

To determine if the above alterations in cortical response properties in Fmr1 KO rats could account for the observed behavioral differences in tone discrimination, we developed a population decoder (see “Materials and methods”) and compared decoder accuracy when using our experimentally measured IC or ACx tone-evoked activity from WT and Fmr1 KO animals as input to the model (Fig 4A). To do this, we first modeled a Gaussian function using the tuning parameters at 40 dB above each unit’s threshold, matching the sound levels used for our behavioral discrimination experiments. We then simulated a population of 640 neurons with characteristic frequencies spanning eight octaves (1–64 kHz), using the Gaussian models for each genotype derived from either the ACx or IC spiking data. Next, we fed this frequency-tuned activity into a layer of readout neurons, with the assumption that activity in the readout neurons follow a Poisson distribution and that only activity above a certain threshold (90% of baseline spiking rate) will be propagated to the readout layer [40]. Finally, we used a Bayesian decoder to classify between Go and No-Go tones of varying frequency difference based on the single-trial activity in the readout layer.

Error rates decreased as a function of frequency distance for both genotypes whether using neural data from the IC (2-way ANOVA: df = 1, ***p < 0.0001) or ACx (2-way ANOVA: df = 1, ***p < 0.0001). However, there was a significant genotype difference in decoder performance when using the experimentally derived mean tuning curves from the ACx as a starting point (2-way ANOVA: df = 1, ***p = 0.0001), with Fmr1 KO ACx neural populations having significantly worse performance for intermediate Go and No-go frequency differences (approximately 1/3–2/3 octave distance), mirroring the behavioral discrimination performance (Fig 4B). This difference in genotype performance was not present in the IC tuning model (Fig 4C, 2-way ANOVA: df = 1, p = 0.613). We can thus recapitulate behavioral discrimination deficits in Fmr1 KO rats from a population model using ACx but not IC tuning parameters, suggesting that cortical response alterations are central to frequency discrimination deficits in Fmr1 KO rats.

We observed differences in multiple response characteristics in the ACx of Fmr1 KO rats that could potentially contributed to impaired frequency discrimination, including elevated spontaneous firing rate, increased sound-evoked responses, and broader tuning width. We therefore systemically varied each of these parameters independently in our decoder model to determine their impact on discrimination performance (Figs 4D, S3, and S4). Tuning width had a profound effect on decoder performance, with broader tuning severely impairing discrimination. Signal-to-noise levels were also critical for frequency discrimination, with increases in baseline firing rate and/or decreases in sound-evoked firing rates degrading decoder performance (Fig S3). However, because both spontaneous and sound-evoked firing rates were increased in Fmr1 KO rats in a similar manner, signal-to-noise ratios were maintained and these changes thus minimally impacted decoder performance (Figs 4D and S3). These results suggest that cortical gain enhancement in Fmr1 KO animals may act to preserve signal-to-noise ratios and signal detection threshold at the expense of tuning precision and fine feature discrimination (Fig 4E).