Abstract
Mutations in MECP2 cause the neurodevelopmental disorder Rett syndrome. MECP2 codes for methyl CpG binding protein 2 (MECP2), a transcriptional regulator that activates genetic programs for experience-dependent plasticity. Many neural and behavioral symptoms of Rett syndrome may result from dysregulated timing and thresholds for plasticity. As a model of adult plasticity, we examine changes to auditory cortex inhibitory circuits in female mice when they are first exposed to pups; this plasticity facilitates behavioral responses to pups emitting distress calls. Brainwide deletion of Mecp2 alters expression of markers associated with GABAergic parvalbumin interneurons (PVins) and impairs the emergence of pup retrieval. We hypothesized that loss of Mecp2 in PVins disproportionately contributes to the phenotype. Here, we find that deletion of Mecp2 from PVins delayed the onset of maternal retrieval behavior and recapitulated the major molecular and neurophysiological features of brainwide deletion of Mecp2. We observed that when PVin-selective mutants were exposed to pups, auditory cortical expression of PVin markers increased relative to that in wild-type littermates. PVin-specific mutants also failed to show the inhibitory auditory cortex plasticity seen in wild-type mice on exposure to pups and their vocalizations. Finally, using an intersectional viral genetic strategy, we demonstrate that postdevelopmental loss of Mecp2 in PVins of the auditory cortex is sufficient to delay onset of maternal retrieval. Our results support a model in which PVins play a central role in adult cortical plasticity and may be particularly impaired by loss of Mecp2.
SIGNIFICANCE STATEMENT Rett syndrome is a neurodevelopmental disorder that includes deficits in both communication and the ability to update brain connections and activity during learning (plasticity). This condition is caused by mutations in the gene MECP2. We use a maternal behavioral test in mice requiring both vocal perception and neural plasticity to probe the role of Mecp2 in social and sensory learning. Mecp2 is normally active in all brain cells, but here we remove it from a specific population (parvalbumin neurons). We find that this is sufficient to delay learned behavioral responses to pups and recreates many deficits seen in whole-brain Mecp2 deletion. Our findings suggest that parvalbumin neurons specifically are central to the consequences of loss of Mecp2 activity and yield clues as to possible mechanisms by which Rett syndrome impairs brain function.
Introduction
Rett syndrome (RTT) is a pervasive neurodevelopmental disorder that results from sporadic, de novo loss-of-function mutations in the MECP2 gene, which codes for the transcriptional regulator methyl CpG binding protein-2 (Amir et al., 1999; Samaco et al., 2008). Because MECP2 is located on the X chromosome, when it possesses a disabling mutation, males (or other individuals with a single X chromosome) lose their sole functioning copy and typically die perinatally; females (or other individuals with two X chromosomes) have heterozygous mosaic expression and frequently survive infancy with impairments in cognition, musculoskeletal structure, metabolism (Van den Veyver and Zoghbi, 2000; Braunschweig et al., 2004), auditory processing, language, and communication (Bashina et al., 2002; Glaze, 2005). MECP2 has been repeatedly implicated in the regulation of neural plasticity (Deng et al., 2010; McGraw et al., 2011; Noutel et al., 2011; Na et al., 2013; Deng et al., 2014; He et al., 2014; Krishnan et al., 2015; Tai et al., 2016; Krishnan et al., 2017; Gulmez Karaca et al., 2018). This observation, combined with the nonlinear developmental course of RTT, has fueled speculation that MECP2 is most essential during periods of elevated neuronal plasticity, for example, early critical periods (He et al., 2014; Krishnan et al., 2015).
Mouse models in which Mecp2 expression has either been disabled (Chen et al., 2001; Guy et al., 2001) or deleted with spatiotemporal selectivity using flanking loxP sites and cell-type-specific expression of Cre recombinase (Gemelli et al., 2006) have been invaluable for understanding the biology of Mecp2. We capitalize on these models with a natural behavior, pup retrieval by female mice, as a readout of cortical plasticity and function (Krishnan et al., 2017; Lau et al., 2020). Our past work was performed in female mice that lacked one functional copy of Mecp2 (Mecp2het). The mosaicism in these mice more closely represents the genetic condition in humans, compared with the more commonly used male null model.
When female mice are exposed to pups, they become primed to exhibit pup retrieval (Rosenblatt, 1967; Sewell, 1970; Ehret et al., 1987), a learned behavioral response to the ultrasonic cries emitted by distressed or wandering pups (Galindo-Leon et al., 2009; Cohen et al., 2011; Cohen and Mizrahi, 2015; Lau et al., 2020; Carcea et al., 2021). Most wild-type females subsequently rapidly increase the speed of retrieval over the first day or two. Emergence of retrieval in adult females is accompanied by changes in the inhibitory circuitry of the auditory cortex (Liu and Schreiner, 2007; Galindo-Leon et al., 2009; Cohen et al., 2011; Lin et al., 2013; Cohen and Mizrahi, 2015; Marlin et al., 2015; Lau et al., 2020). Mecp2 expression is specifically required in the auditory cortex at the time of pup exposure for proper retrieval (Krishnan et al., 2017). Moreover, loss of Mecp2 triggers overexpression of parvalbumin (PV) and the extracellular matrix structures perineuronal nets (PNNs) (Krishnan et al., 2017). These two markers associated with the PVins are thought to act as brakes on cortical plasticity (Krishnan et al., 2017). Restoration of normal levels of PV and PNN expression in the auditory cortex improved behavior and restored physiological plasticity (Krishnan et al., 2017; Lau et al., 2020). Given that changes in PVin-specific markers were correlated with retrieval behavior performance, we speculated that PVins are central to the behavioral phenotype of Mecp2het.
Loss of Mecp2 appears to be more detrimental in certain cell types. For example, inhibitory cells may be particularly impaired by loss of Mecp2. Restriction of Mecp2 mutation to either all GABAergic cells or to selected inhibitory subclasses (e.g., parvalbumin- or somatostatin-positive interneurons) is sufficient for recapitulating the majority of phenotypes in mouse models (Chao et al., 2010; He et al., 2014; Ito-Ishida et al., 2015; Mossner et al., 2020). Other work has demonstrated selected behavioral effects resulting from loss of Mecp2 in excitatory neurons (Chao et al., 2007; Meng et al., 2016). Here, we use cell-type-specific removal of Mecp2 and show that PVins are the only major class of interneurons that significantly affect retrieval when depleted of Mecp2. Mice of the genotype PV-Cre+/Mecp2flox (hereafter, PV-Mecp2 mutants), which lack Mecp2 in all PVin, are delayed in the onset of pup retrieval and recapitulate all the major features of Mecp2het. Specifically, when virgin PV-Mecp2 mutant females were exposed to pups, they exhibited elevated expression of PV and PNNs relative to Mecp2wt. PVin-specific mutants also did not show the experience-dependent disinhibition of auditory cortex seen in Mecp2wt controls. Finally, deleting PVins in the auditory cortex in adulthood was sufficient to delay pup retrieval. Together, these findings are consistent with the conclusion that Mecp2 in auditory cortex PVins is critical for initiating experience-dependent auditory plasticity that facilitates the emergence of maternal retrieval.
Materials and Methods
Animals
All procedures were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the Cold Spring Harbor Laboratory Institutional Animal Care and Use Committee. Animals were maintained on a 12 h light/dark cycle and received food and water ad libitum. Behavioral experiments were conducted during light-cycle hours.
Subjects were adult, female mice 6–12 weeks of age, bred in house from founders obtained from The Jackson Laboratory or the Mutant Mouse Resource and Research Center (MMRRC). The following genotypes were used: CBA/CaJ, B6.129P2(C)-Mecp2tm1.1Bird/J (Mecp2het; catalog #003890, The Jackson Laboratory), B6.129S4-Mecp2tm1Jae/Mmucd (Mecp2flox; catalog #011918, MMRRC), B6.129P2-Pvalbtm1(cre)Arbr/J (PV-Cre; catalog #017320, The Jackson Laboratory), Viptm(cre)zjh/J (VIP-Cre; catalog #010908, The Jackson Laboratory), Ssttm2.1(cre)zjh/J (SST-Cre; catalog #013044, The Jackson Laboratory), B6.129S2tm(emx1)krj/J (Emx1-Cre; catalog #005628, The Jackson Laboratory), and PV-Flp B6.Cg-Pvalbtm4.1(flpo)Hze/J (PV-Flp; catalog #022730, The Jackson Laboratory). All crosses between Cre/Flp recombinase lines were established by pairing carriers of each allele such that all female test subject cagemates (controls and mutants) were homozygous for the Mecp2-flox allele and had 0–2 copies of the relevant recombinase allele. For example, in the PV-Mecp2 line, PV-Mecp2 mutants were either PV-ires-Cre+/− or PV-ires-Cre+/+ and were homozygous for Mepc2flox (Mecp2flox+/flox+). PV-Mecp2 wild-type (WT) controls were negative for the recombinase (PV-ires-Cre−/−) but homozygous for Mecp2flox.
All animals were genotyped at the time of weaning, ∼3 weeks of age, according to standard protocols from the source. In some cases, genotyping was performed by an external service (Transnetyx) using their suggested probes or probes from The Jackson Laboratory. All lines were monitored for the possibility of somatic recombination affecting the Mecp2 gene as per The Jackson Laboratory recommendations.
Behavioral analysis
Pup retrieval behavior was conducted as previously described (Krishnan et al., 2017) and aided by automated tracking with DeepLabCut software (Mathis et al., 2018). In brief, virgin adult female test subjects (surrogates) were cohoused with a WT (CBA) pregnant dam 2–5 d prepartum. Starting at postnatal day (P)0, subjects were tested daily for 3 consecutive days in a pup retrieval assay as follows. Pups were isolated for 2 min and then scattered to set positions in the home cage. Surrogates were allowed to interact with scattered pups for 5 min. Animals not currently performing the retrieval assay, including the dam, were temporarily placed in a group holding cage. Holding cages and home cages were not changed for the duration of retrieval experiments (i.e., from the time of surrogate pairing to P2). A normalized latency score between zero (instantaneous gathering of all pups) and one (failure to gather all pups) was calculated as follows:
Surgeries and injections
All surgeries were performed on a Kopf stereotaxic device. Anesthesia induction was achieved with a bolus intraperitoneal injection of a ketamine (100 mg/kg) and xylazine (5 mg/kg) mixture and maintained for long surgeries (>2 h) with inhaled isoflurane (1–2%) in oxygen (2–4 liters per min), adjusted as needed based on assessment of the depth of anesthesia with a tail/paw pinch every 30 min. At the end of surgery, a nonsteroidal anti-inflammatory (meloxicam, 2 mg/kg, i.p.) and an antibiotic (enrofloxacin, selected for its lack of ototoxicity, 4 mg/kg. i.p.) were administered for analgesia and infection prophylaxis.
All animals used for in vivo physiology or fiber photometry experiments had a custom titanium head bar affixed to the skull at the time of craniotomy. To optimize the stability of the head bar, the surface of the skull was lightly, manually etched with a scalpel, and three different dental cements were applied, Metabond Quick (C&B), Vitrebond Light Cure Glass Ionomer (3M), and Ortho-Jet (Lang Dental). Mice used for fiber photometric recordings were also fitted with optical fiber implants secured with dental cement. Uncleaved fibers (0.39 NA, 200 µm diameter, 1.25 mm length) (catalog #CFMLC12U-20, Thorlabs) were manually cleaved to the desired length using a Ruby Scribe (catalog #S90R, Thorlabs). A guide (catalog #OGL-5, Thorlabs) was also secured (LOCTITE, Henkel Adhesives) into place, and the optical fiber implants were lowered into the auditory cortex to a depth of 700 µm. Implants were allowed to cure >48 h before head fixation or attachment of an optical cable. Braided silk surgical sutures (CP Medical) and/or Vetbond tissue adhesive (3M) were used to close surgical sites around headgear.
Thalamorecipient core auditory cortex (Lin et al., 2013) was targeted for adeno-associated virus (AAV) injections and neurophysiology recordings using the following coordinates relative to bregma: anterior 2.5 mm (±0.4 mm) and lateral 3.9 mm (±0.3 mm). For AAV-injection experiments, 90 nl injections (delivered at 20 nl/min) were administered via glass pipettes (20 µm tip) at six locations per hemisphere, relative to bregma (anterior 2.1 mm, 2.4 mm, and 2.7 mm at lateral 3.8 mm and 4.0 mm). Viruses used were pAAV-EF1a-fDIO-Cre (catalog #121675-AAV9, Addgene) or pAAV.Syn.Flex.GCaMP7s.WPRE.SV (catalog #104491-AAV9, Addgene).
Immunohistochemistry
Subjects were injected intraperitoneally with a lethal dose of Euthasol (pentobarbital sodium and phenytoin sodium cocktail) and transcardially perfused with ice-cold PBS followed by paraformaldehyde (PFA). Brains were extracted and fixed overnight in PFA at 4°C before being transferred to a 30% sucrose solution in PBS again at 4°C overnight or until buoyancy of the tissue was lost. Frozen 45 µm sections were collected using a sliding microtome (Leica SM2010 R). Free-floating sections were preserved for batch immunohistochemical (IHC) staining in cryoprotectant solution at −20°C. Cryoprotectant solution consisted of sucrose (0.3 g/ml), polyvinyl-pyrrolidone (0.01 g/ml), and ethylene glycol (0.5 ml/ml) in 0.1 m PB.
For all staining protocols, free-floating sections were washed three times at room temperature (RT) in PBS and 0.3% Triton X-100, followed with a 30 min wash in 0.3% hydrogen peroxide solution to decrease nonspecific background staining. Subsequently, sections were incubated in 5% normal goat or donkey serum, in accordance with the chosen secondary antibody, and then primary antibody solution at 4°C overnight. The following primary antibodies and dilutions were used to stain for MeCP2, parvalbumin, and perineuronal nets, respectively: rabbit anti-MeCP2 (1:1000; Cell Signaling Technology), mouse anti-parvalbumin (1:1000; Sigma-Aldrich), and lectin from Wisteria floribunda with biotin conjugate (1:1000; Sigma-Aldrich). Rabbit anti-HA-Tag (1:500; Abcam) was used on alternating sections to detect cells expressing pAAV-EF1a-fDIO-Cre in PV-Flp subjects. In a subset of animals used for fiber photometry, staining was used to amplify GCaMP expression with chicken anti-GFP (1:1000; Aves Labs). Sections were washed three times at RT before transfer to secondary antibody solution. Primary anybody staining was visualized with the following Alexa Fluor conjugated secondary antibodies: goat AF 488, goat AF 594, and donkey AF 633 (Invitrogen Technologies). Secondary antibody dilutions matched those for the targeted primary antibody. Sections were exposed to secondary antibodies for 2 h at RT.
Imaging and quantification
A 20× magnification was used to collect z-stack images on a 710 confocal microscope (Ziess) in line scan mode. The spectra for each channel were manually adjusted to optimize signal-to-noise ratio using staining from a naive WT sample. Those settings were as follows: bit depth = 12; laser power = 10%; gain < 700, pinhole size lowest value for each channel; full dynamic range, 1024 × 1024 pixel smoothness, averaging = 4. These settings were used to acquire all images across batches of a given stain. For each brain, six to eight matched sections spanning the rostral–caudal axis of the auditory cortex were selected for imaging. Maximum intensity projection images were generated for each field of view.
To quantify per-cell staining intensity, Fiji (ImageJ) software was used to manually outline cells and apply area-integrated intensity in the set measurements panel. Background intensity readings for each section were subtracted from the cell intensity values. Cells were counted using Fiji (ImageJ) software and normalized to the volume of auditory cortex represented in images based on the size of the image and the depth of the z-stacks (µm × µm × µm). Mean cell density was determined by averaging counts per volume across sections.
In vivo physiology
Loose-patch recordings from the auditory cortex were performed in awake, head-fixed subjects that were allowed to freely run on an axially rotating foam wheel as previously described (Cazakoff et al., 2014; Lau et al., 2020). Animals were habituated to head fixation and the wheel for 15–30 min/d for 1–2 d before recordings were collected. For each subject, a small craniotomy (200 × 200 µm) was made over the auditory cortex. Subjects were head fixed by bolting the titanium bar implant to a frame suspended above a freely rotating foam wheel for the duration of the recording. Recordings were performed over 2–3 consecutive daily sessions (<8 h). Gelfoam (absorbable gelatin sponge, Ethicon) soaked in sterile saline was used to keep the surface of the cortex moist between recording periods during the session, and craniotomy sites were covered with Kwik-Cast between days of recording.
Single-unit recordings were made with a bridge amplifier (BA-03X, NPI). Borosilicate glass micropipettes (15–40 MΩ) were pulled on a horizontal pipette puller (model P-1000, Sutter Instrument). Pipettes were filled with an intracellular solution containing the following (in mm): 125 potassium gluconate, 10 potassium chloride, 2 magnesium chloride, and 10 HEPES. Single neurons were recorded blind by advancing the pipette in 3–5 µm steps using a single-axis stepper motor and controller (Solo, Sutter Instrument) or hydraulic micromanipulator (MX610, Siskiyou) as positive pressure was applied to the tip. Brief, small injected currents (−200 pA, 200 ms) were made at 2 Hz to monitor tip resistance, and the capacitance buzz feature on the amplifier was used to clear debris. Voltage signals were low-pass filtered (3 kHz), digitized (10 kHz), and acquired using Spike2 software and Cambridge Electronic Design hardware (Power1401). All cells were recorded at a depth <1 mm.
Auditory stimuli consisted of seven logarithmically spaced tones from 16 to 64 kHz and a library of eight ultrasonic vocalizations (USVs) recorded from WT CBA/CaJ mouse pups (2–4 d old) inside an anechoic isolation chamber (Industrial Acoustics) using an ultrasound microphone (Avisoft Bioacoustics) suspended 30 cm above the pup. Tone and call stimuli were played separately in a pseudorandom order with a 4 s interstimulus interval. Stimulus files that had been digitally sampled at 195.3 kHz were converted to analog output via Cambridge Electronic Design hardware (Power1401). Stimuli were low-pass filtered (100 kHz) and amplified with custom-built hardware (Kiwa Electronics) before being output through an electrostatic speaker and driver (ED1/ES1, Tucker-Davis Technologies) 4 inches directly in front of the animal. Speaker output was calibrated to 65 dB SPL at the head of the mouse with a sound level meter (model 407736, Extech) using A-weighting by comparing with an 8 kHz reference tone. The speaker had a relatively flat output (±11 dB) at 4–100 kHz.
Fiber photometry
PV-Cre mice (WT, Mecp2flox, and Mecp2het) were prepared by injecting a Cre-dependent AAV-expressing GCaMP7s and optical fiber implants in the auditory cortex as described above. Bulk GCaMP-detected calcium signals from PVins were measured using a custom setup as described previously (Dvorkin and Shea, 2022). Subjects were head fixed by bolting the titanium bar implant to a frame suspended above a freely rotating foam wheel for the duration of the recording. An optical cable (200 µm, 0.39 NA) coupled to the fiber implant was used to deliver 473 nm and 565 nm light from a pair of LEDs (LEDD1B, Thorlabs). Green emitted light was used to measure the activity-dependent fluorescence of GCaMP, whereas red emitted light was used to monitor and correct for potential movement or optical coupling artifacts unrelated to neural activity. No such artifacts were ever detected in our head-fixed recordings. Light from each LED was modulated at 211 Hz but 180° out of phase. Before each recording session, the power of the light emitted at the tip of the patch cable was measured with a power meter (PM100D, Thorlabs) and manually adjusted to 30–33 µW.
Emitted light was split into separate green and red paths, bandpass filtered (Chroma Technologies), and detected by separate photodiodes (Newport). Photodiode signals were digitally sampled at 6100 Hz via a data acquisition board (NI USB-6211, National Instruments). As head-fixed recordings were uncontaminated by movement artifacts, only the green emission signal was used to compute ΔF/F by performing the following steps. For each day of recording, first we measured the peak of each cycle, effectively generating a waveform at 211 Hz sampling rate. We low-pass filtered the data at 15 Hz. Then to account for photobleaching, we fit the trace with a second-order exponential function, which we subtracted from the signal. Finally, we subtracted the mean of the whole trace and divided the result by the same mean. To facilitate direct comparisons between different animals, all fluorescence traces from a given animal were converted to a Z-score using the mean and SD of the entire dataset for that animal.
The same USV stimulus set was presented during fiber photometric recordings as described for electrophysiology recordings with an interstimulus interval of 10 s. Custom MATLAB software was used to present stimuli and acquire data via hardware from National Instruments.
Experimental design and statistical analysis
All data visualization and statistical analysis was performed in MATLAB or Prism (GraphPad) software. Unless otherwise noted, values are reported as mean ± SEM. Behavioral latency data were analyzed with a two-way ANOVA (with factors of time and genotype/treatment) and where warranted, post hoc comparisons were made. All histology was performed in batches wherein one subject from each experimental group was represented, and the scorer was blinded to the group. Per-cell PV intensity was Z-scored within each batch, and for PNN counts, in each batch a threshold was applied at the mean +2 SD for all sections in a batch. Only PNNs visible after thresholding were counted. Significant differences in PV intensity and PNN counts were statistically analyzed with a one-way ANOVA.
Spike2 software (Cambridge Electronic Design) was used to manually threshold and sort single-unit spike shapes based on principal components analysis (PCA) clustering. A total of 287 individual neurons were included in the analysis. Several previous studies, including work from our lab, have identified distinct properties of PVin waveforms, including a narrow spike shape, nearly symmetrical positive and negative peak amplitudes, and elevated firing rates (Cohen and Mizrahi, 2015; Lau et al., 2020). We combined our current dataset with another 26 neuronal recordings previously collected in our lab that used photoidentification of PVins expressing the optogenetic activator ChR2. Each of the total 313 neurons was represented by a vector of 28 points defining the mean spike shape plus the ongoing firing rate of the cell. The 313 × 29 matrix was used as the input to a PCA analysis, and the result was analyzed by k-means clustering (k = 3). All optically identified neurons were contained within a single cluster; therefore, the neurons in that cluster were designated as putatively PVins.
Peristimulus time histograms (PSTHs; 10 ms bin size) were constructed of the mean firing rate of each cell in response to each stimulus (i.e., cell-stimulus pairs), and bin values were transformed to Z-scores for each cell. Significant responses were identified among all cell-stimulus pairs with a bootstrap procedure as follows. If a given stimulus was presented n times, n windows of 150 ms each were randomly chosen from the entire duration of spiking recorded for that neuron, and the mean spike rate for all n windows was calculated. This was repeated 10,000 times to generate a null distribution of randomized spiking rates. Significant cell-stimulus pairs were identified as those for which the actual mean response in the 150 ms after the stimulus onset fell within the upper or lower 2.5% of the spiking rate null distribution. Cells that lacked a significant response to any stimulus were discarded from the analysis. The response for each cell-stimulus pair was computed as the integrated area under the first 200 ms of the PSTH in units of Z-score · s. Mean responses across experimental groups were statistically compared with Mann–Whitney U tests.
For fiber photometry data, to compare fluorescence signals across animals and over time, ΔF/F signals collected from each animal were transformed to a Z-score. Responses to each stimulus were computed as the integrated area under the mean response curve in units of Z-score*s. For each genotype, mean responses to auditory stimuli at the pup-naive time point were compared with mean responses measured at a postnaive time point (P3–P5) with a paired t test.
Results
Acquisition of pup retrieval is delayed by loss of Mecp2 in parvalbumin interneurons
We previously showed that female Mecp2het mice fail to reliably retrieve pups, even after 5 d of cohabitation with a WT dam and her litter (Krishnan et al., 2017). We also found that when mice were crossed between PV-Cre and Mecp2flox, PV-Mecp2 mutants were initially slower to retrieve compared with PV-Mecp2 WT subjects (Krishnan et al., 2017). This raised the possibility that certain cell types within the auditory cortex might be more important than others for the neural plasticity that facilitates retrieval. Here, we replicate that finding, and we compare the results with our observations from knocking out Mecp2 in several other genetically restricted neuronal populations.
We used several mouse lines expressing Cre-recombinase in specific cell types in conjunction with Mecp2flox mice to restrict Mecp2 knockout to three distinct populations of GABAergic inhibitory neurons, namely, parvalbumin-expressing PVins, somatostatin (SST)-expressing SSTins, and vasoactive intestinal peptide (VIP)-expressing VIPins. We also used the Emx1-Cre line to restrict Mecp2 knockout to the majority (∼90%) of excitatory cortical pyramidal neurons (Briata et al., 1996) (Fig. 1A). Female subjects were cohoused with a pregnant WT CBA female, and beginning on P0, were tested daily in a pup retrieval assay (Fig. 1B; see above, Materials and Methods; Krishnan et al., 2017). Cohousing gives the virgin females the opportunity to observe and participate in interactions with pups (Carcea et al., 2021). Therefore, their improvement in performance over time reflects only the influence of experience, not hormonal changes related to pregnancy and parturition.
Cell-type-specific deletion of Mecp2 has varying effects on pup retrieval behavior. A, Schematic of mouse lines crossed to achieve selective Mecp2 deletion in different neuron types. B, Schematic of cohousing and retrieval behavior protocol. C–F, Scatterplots of retrieval latency comparing performance of cell-type-specific Mecp2 mutants to that of their littermate controls for PV-Cre (C), SST-Cre (D), VIP-Cre (E), and Emx1-Cre (F) lines crossed with Mecp2flox mice. Emergence of pup retrieval was delayed in PV-Mecp2 mutants relative to controls (C). A two-way ANOVA revealed significant main effects for time (day of testing; F = 9.21, p < 0.001) and genotype (F = 9.41, p < 0.01) but not an interaction (F = 1.94, p = 0.15). Post hoc tests revealed a significant difference between mutant and WT animals on P0 only (n = 14 controls, latency, 0.17 ± 0.03; n = 25 mutants, latency, 0.32 ± 0.08; Sidak’s test, ***p < 0.001). Timing of emergence of pup retrieval did not differ between SST-Mecp2 mutants and controls (D). A two-way ANOVA revealed a significant main effect for time (day of testing; F = 8.98, p < 0.01), but not for genotype or for an interaction. Post hoc tests revealed no significant difference between mutant and WT animals on any day (Sidak’s test, p > 0.05). Emergence of pup retrieval did not differ between VIP-Mecp2 mutants and controls (E). A two-way ANOVA revealed a significant main effect for time/day (VIP-Mecp2, F = 7.50, p < 0.01), but not for genotype, nor an interaction between those variables. Post hoc tests revealed no significant difference between mutant and WT animals on any day of testing (Sidak’s test, p > 0.05). Emergence of pup retrieval was delayed in Emx1-Mecp2 mutants relative to controls (F). A two-way ANOVA revealed a significant effect only for the interaction between time and genotype (F = 3.86, p < 0.05), but neither a main effect for genotype (F = 2.45, p = 0.13) nor day (F = 2.13, p = 0.14). A post hoc test revealed a significant difference between mutant and WT animals for P0 only (n = 7 controls, latency, 0.26 ± 0.03; n = 18 mutants, latency, 0.39 ± 0.09; Sidak’s test, *p < 0.05). G–J, Bar plots of mean velocity of mice during retrieval sessions on P0, comparing mutants (Mecp2flox) with controls for PV-Cre (G), SST-Cre (H), VIP-Cre (I), and Emx1-Cre (J). No significant difference was found for any of the lines (unpaired t test, PV, p = 0.82; SST, p = 0.27; VIP, p = 0.63; EMX, p = 0.56).
PV-Mecp2 mutants showed significantly longer retrieval latency scores compared with PV-Mecp2 WT on P0. However, these subjects improved over time, matching the performance of PV-Mecp2 WT mice by P1 (Fig. 1C). A two-way mixed effects ANOVA revealed significant effects of time (day; F = 9.21, p < 0.001) and genotype (F = 9.41, p < 0.01), but not an interaction between those variables (F = 1.94, p = 0.15). Post hoc testing revealed a significant difference between the mutant and WT groups for P0 only (Sidak’s test, p < 0.001). Individual unpaired comparisons for each day detected a significant difference between genotypes only on P0 (n = 25 mutant, 15 WT; Mann–Whitney corrected for multiple comparisons, p < 0.01). Therefore, PV-Mecp2 mutants showed a transient disruption in pup retrieval.
To assess the specificity of this result to the PVin population, as opposed to other interneuron types, we ran the same experiment with mice in which Mecp2 was knocked out in one of two other major classes of GABAergic inhibitory neurons (SSTins or VIPins; Fig. 1D,E). Two-way mixed effects ANOVAs revealed significant effects for time (day) in both cohorts (SST-Mecp2, F = 8.98, p < 0.01; VIP-Mecp2, F = 7.50, p < 0.01), but not for genotype or for an interaction between those variables. Post hoc testing revealed no difference between mutant and WT groups for any day of testing (Sidak’s test, p > 0.05). Individual unpaired comparisons for each day also failed to detect significant differences between genotypes on any day (p > 0.05) for either line. Therefore, neither SST-Mecp2 mutants nor VIP-Mecp2 mutants showed a transient disruption in pup retrieval as observed for PV-Mecp2 subjects.
As a comparison to Mecp2 deletion in small interneuron populations, we next crossed Mecp2flox mice with the Emx1-Cre line to restrict Mecp2 knockout to ∼90% of excitatory neurons in the cortex and hippocampus (Briata et al., 1996). Like PV-Mecp2 mutants, Emx1-Mecp2 mutants exhibited a delayed onset of pup retrieval (Fig. 1F). A two-way mixed effects ANOVA revealed a significant effect only for an interaction (F = 3.86, p < 0.05) but neither a main effect for genotype (F = 2.45, p < 0.13) nor for day (F = 2.13, p = 0.14). Post hoc testing revealed a significant difference between mutant and WT groups for P0 only (Sidak’s test, p < 0.01). Individual unpaired comparisons for each day detected a significant difference between genotypes only on P0 (n = 18 mutant, 7 WT; Mann–Whitney corrected for multiple comparisons, p < 0.05). Therefore, like PV-Mecp2 mutants, Emx-Mecp2 mutants showed a transient disruption in pup retrieval. Interestingly, this disruption was comparable in the two groups, despite the disparity in the size of the cell populations. Performance of all mice was unrelated to gross motor deficits, as determined by automated tracking of the animals during retrieval with DeepLabCut software, which showed there was no significant difference in mean velocity between controls and mutants for any of the lines (Fig. 1G–I).
Loss of Mecp2 only in PVins recapitulates changes of molecular expression seen in Mecp2het
High levels of expression of PVin markers (PVs and PNNs) are taken as an indicator of maturity in PVins and are well correlated with reduced capacity for synaptic plasticity and learning in development and adulthood (Pizzorusso et al., 2002; Carulli et al., 2010; de Vivo et al., 2013; Donato et al., 2013; Happel et al., 2014; Hou et al., 2017; Cisneros-Franco and de Villers-Sidani, 2019; reviewed in Rupert and Shea, 2022). Previously, we reported that both markers exhibited overexpression in the auditory cortex of Mecp2het after 5 d of exposure of a virgin female to pups (Krishnan et al., 2017). This experience-dependent overexpression was not observed in Mecp2wt, and genetic and pharmacological approaches that reversed it restored retrieval performance in Mecp2het (Krishnan et al., 2017).
Given that Mecp2 deletion in PVins, a small population of neurons (∼10%), is sufficient to disrupt retrieval behavior (albeit temporarily) and that population is also the locus of key pathologic features in Mecp2het models, we hypothesized that the changes to PV and PNN may reflect a cell-autonomous consequence of Mecp2 deletion from PVins. To test this hypothesis, we compared the level of PV protein and PNN expression by auditory cortex PVins between PV-Mecp2 mutants and PV-Mecp2 WTs. We did this by performing IHC and confocal microscopy of fixed brain sections from naive mice (no pup exposure), and experienced mice (after pup exposure) at the P1 and P5 time points (Fig. 2Ai–iii,Bi–iii). We quantified per-cell intensity of PV staining and, to minimize batch effects, converted intensities from each batch to a Z-score. We compared the distribution of Z-scores for each group, focusing on the changes within each genotype across time points as pup experience increased. A one-way ANOVA revealed significant differences among group means (F = 47.3, p < 0.001). PV-Mecp2 WT virgin mice showed a drop in PV expression; PV staining intensity was significantly lower on P5 compared with intensities of the naive and P1 cohorts (Fig. 2C, Sidak’s test, p < 0.001). In contrast, PV-Mecp2 mutants exhibited an increase in PV expression; PV staining intensity was significantly higher in tissue collected at both P1 and P5 compared with naive animals (Sidak’s test, p < 0.001)
Selective deletion of Mecp2 in PVins recapitulates changes of molecular expression seen in Mecp2het. Ai–iii, Series of photomicrographs of example sections of the auditory cortex taken from a PV-Mecp2 WT naive mouse (i), a mouse after 2 d of cohabitation (P1; ii), and a mouse after 6 d of cohabitation (P5; iii). All sections were stained with IHC using an antibody for parvalbumin (purple) and a biotinylated lectin from Wisteria floribunda for PNNs (green). Scale bar, 200 mm. Bi–iii, Same as Ai–iii but for sections taken from PV-Mecp2 mutants. C, Box plot of the distributions of per-cell intensity of parvalbumin staining. All histology was run in six batches, with one brain from each genotype condition group represented in each batch. All individual neuron intensities were Z-scored per batch. From left to right, The total number of neurons in each group is 1006, 996, 757, 1108, 1084, 1063. A one-way ANOVA detected differences among the groups (F = 47.3, p < 0.001). Post hoc testing revealed a significant decrease of mean PV intensity in PV-Mecp2 WT on P5 compared with naive mice (naive latency, 0.076 ± 0.03 Z-score; P5 latency, 0.25 ± 0.03 Z-score; Tukey’s test, ***p < 0.001). In contrast, PV-Mecp2 mutant mice had higher mean per cell intensity PV staining on P1 (0.31 ± 0.03 Z-score) and P5 (0.047 ± 0.03 Z-score) compared with naive mice (−0.25 ± 0.03 Z-score; Tukey’s test, ***p < 0.001). D, Box plot of mean high intensity PNNs per section, comparing all six groups of mice. One mouse from each group was processed in each batch with total of five batches. Per-section counts were Z-scored for all sections in each batch. Considering all groups, there was a significant difference among them (one-way ANOVA, F = 4.3, p < 0.01). Post hoc comparisons of each experienced time point to the naive time point for each genotype showed that PV-Mecp2 mutant mice on P1 had significantly more high-intensity PNNs per section than naive PV-Mecp2 mutant mice (n = 5 mice/group; naive mutant, −0.11 ± 0.23 Z-score; mutant P1, 0.83 ± 0.08 Z-score; Sidak’s test, *p < 0.05).
To quantify changes in PNN expression, we counted high-intensity PNNs in the auditory cortex in the same set of sections analyzed (see above, Materials and Methods). To minimize batch effects, all images were thresholded and binarized at 2 SDs above the mean pixel value for each staining batch, and the counts of PNNs per section were Z-scored within each batch. An analysis of all groups detected significant differences among the groups (one-way ANOVA; F = 4.30, p < 0.01; Fig. 2D). Post hoc tests comparing PNN counts from experienced mice at P1 and P5 to counts at the naive time point for both genotypes showed PNN counts per section were only significantly higher on P1 in PV-Mecp2 mutants (Sidak’s test, p < 0.05). In light of all these observations, we conclude that deletion of Mecp2 in PVins is sufficient to at least transiently evoke overexpression of molecular markers closely associated with suppression of plasticity on exposure to pups.
PV-Mecp2 mutants lack the auditory cortical disinhibition triggered by pup exposure in WT
Our next goal was to determine whether deletion of Mecp2 only in PVins was sufficient to reproduce the neurophysiological changes we observed in the auditory cortex of pup-experienced Mecp2het mice. We found that pup-experienced Mecp2wt mice exhibited a dramatic decrease in spiking output by auditory cortex PVins relative to that from naive females (Lau et al., 2020). Moreover, we discovered that this disinhibition of the auditory cortex by PVins was absent in Mecp2het (Lau et al., 2020). We hypothesized that deletion of Mecp2 only from PVins may affect their stimulus-evoked firing in a cell-autonomous manner. To test this hypothesis, we made loose-patch, single-unit electrophysiological recordings from auditory cortical neurons in awake head-fixed animals of both genotypes at naive and after pup-experienced time points. We made neuronal recordings from four experimental groups of mice—PV-Mecp2 mutant mice that were naive to pups (PV-Cre/Mecp2 mutant Naive; n = 7 mice), PV-Mecp2 mutant mice that cohabitated with a WT dam and her pups for >5 d (PV-Cre/Mecp2 mutant experienced; n = 10 mice), PV-Cre/Mecp2 WT control littermates without pup experience (PV-Cre/Mecp2 WT Naive; n = 15 mice), and WT littermates that experienced cohabitation (PV-Cre/Mecp2 WT experienced; n = 21 mice).
As previously reported (Wu et al., 2008; Oswald and Reyes, 2011; Cohen and Mizrahi, 2015; Lau et al., 2020), PVins and non-PV neurons had characteristic spike shapes that could be distinguished by their features. We therefore combined the neurons we recorded here with a wild-type dataset from a previous study (Lau et al., 2020) in which we optically identified ChR2-expressing PVins. We identified putative clusters of PVins and non-PV neurons in a PCA with a k-means clustering algorithm (see above, Materials and Methods). Average spike waveforms for our putatively identified populations of cells are plotted as corresponding color traces in Figure 3, B and C. PVins had particularly narrow spike waveforms and were more symmetrical in amplitude around the baseline; non-PV neurons were wider and had more prominent positive peaks (Fig. 3B,C). Despite our use of a novel PCA-based sorting method, our results were very consistent with previous classification results from our group and others (Wu et al., 2008; Oswald and Reyes, 2011; Cohen and Mizrahi, 2015; Lau et al., 2020).
Classification of PVin and non-PVin single-unit recordings by spike waveform and firing rate. A, Scatterplot showing the results of a principal components analysis and k-means clustering analysis of 313 auditory cortex neurons based on the mean spike waveform and baseline firing rate. Red points denote putative PVins, black points denote putative non-PV neurons, points outlined in blue denote a subset of optically tagged PVins, and points outlined in gray denote inconclusive optical identification results. B, Plot of all mean waveforms from putative non-PV neurons (n = 241). C, Plot of all mean waveforms from putatively identified PVins (n = 71).
We first examined the responses of non-PV neurons from each group to a library of eight USVs recorded from pups that were 2-4 d old (Lau et al., 2020). We observed that individual non-PV neurons often exhibited distinct responses to different USVs, responding with either increases or decreases in firing. Therefore, we identified all cell-call pairs (mean responses of one neuron to one stimulus) that exhibited a statistically significant change in firing rate as assessed with a bootstrap procedure (see above, Materials and Methods). A PSTH (bin size, 10 ms) was constructed to visualize the mean response of each cell to each stimulus, and the bins of all PSTHs from each cell were transformed to a Z-score.
Heat maps in Figure 4, A and B, depict 2D PSTHs that each represent the mean responses for all cell-call pairs from one of the four groups of wild types (Fig. 4A) and mutants (Fig. 4B). Rows in each 2D PSTH are sorted from the largest firing decrease to the largest firing increase measured in the 200 ms window after stimulus onset. We separately compared the mean of all excitatory responses (stimulus-driven increase in firing rate) and the mean of all inhibitory responses (stimulus-driven decrease in firing rate) between naive and pup-experienced groups for each genotype. Figure 4, C and D, depict mean ± SEM traces for each sign of response. Gray traces represent recordings collected from naive mice, and purple traces represent recordings collected from pup-experienced mice. We integrated the area under the curve (AUC) for each cell-call pair and compared the distribution of response magnitudes between naive and experienced mice cohorts. Figure 4E summarizes the results of these comparisons for PV-Mecp2 and WT mice. In WT mice, mean inhibitory responses were significantly weaker in mice that had cohousing experience with pups relative to mice that lacked pup exposure (n = 54 naive and 44 experienced cell-call pairs; Mann–Whitney U test, p < 0.01). Mean excitatory responses were unchanged between naive and experienced mice (n = 108 naive and 72 experienced cell-call pairs; Mann–Whitney U test, p = 0.09). Figure 4F shows the corresponding results for PV-Mecp2 mutant mice. In these mice, mean inhibitory responses were significantly stronger in pup-experienced mice than they were in pup-naive mice (n = 31 naive and 30 experienced cell-call pairs; Mann–Whitney U test, p < 0.001). As in WT mice, mean excitatory responses were unchanged between naive and experienced PV-Mecp2 mutant mice (n = 60 naive and 45 experienced cell-call pairs; Mann–Whitney U test, p = 0.61).
Selective deletion of Mecp2 in PVins recapitulates the suppression of non-PV neuron inhibitory plasticity seen in Mecp2het. A, Two-dimensional PSTHs representing the mean responses for all non-PV cell-call pairs with significant responses to USVs from naive (top) and experienced (bottom) PV-Mecp2 WT mice. Each row within the PSTHs represents the Z-scored response of one cell-call pair. Rows are sorted from the greatest stimulus-evoked decrease in firing rate to the greatest stimulus-evoked increase in firing rate. Response window of 200 ms after the stimulus onset is marked by the vertical black line. B, Same as A, but data are taken from auditory cortex recordings in PV-Mecp2 mutant mice. C, Mean ± SEM traces for excitatory responses (top) and inhibitory responses (bottom), comparing data from naive (gray) and experienced (purple) PV-Mecp2 WT mice. D, Same as C, but the data are from PV-Mecp2 mutant mice. E, Box plot of the integrated AUC for each non-PV cell-call pair recorded from PV-Mecp2 WT mice comparing the distribution of inhibitory and excitatory response magnitudes between naive mice and experienced mice. Mean inhibitory responses were significantly weaker in pup-experienced mice relative to pup-naive mice (naive, n = 54 cell-call pairs, 0.132 ± 0.01 Z-score*s; experienced, n = 44 cell-call pairs, 0.101 ± 0.01 Z-score*s; Mann–Whitney U test, **p < 0.01). Mean excitatory responses were unchanged between naive and experienced mice (naive, n = 108 cell-call pairs, 0.195 ± 0.01 Z-score*s; experienced, n = 72 cell-call pairs, 0.161 ± 0.01 Z-score*s; Mann–Whitney U test, p = 0.09). F, Same as E, but data are from non-PV cell-call pairs recorded from PV-Mecp2 mutant mice. Mean auditory cortex inhibitory responses were significantly stronger in experienced mice than they were in pup-naive mice (naive, n = 31 cell-call pairs, 0.134 ± 0.01; experienced, n = 30 cell-call pairs, 0.170 ± 0.01; Mann–Whitney U test, ***p < 0.001). As in WT mice, mean excitatory responses were unchanged between naive and experienced PV-Mecp2 mutants (naive, n = 60 cell-call pairs 0.171 ± 0.01, Z-score*s; experienced, n = 45 cell-call pairs, 0.160 ± 0.01 Z-score*s; Mann–Whitney U test, p = 0.61).
We performed a similar analysis on recordings collected from putative PVins (Fig. 5). In this case, because all USV responses we observed in PVins evoked increased spiking, we included all responses to stimuli for all neurons that had a significant response to at least one USV. As in Figure 4, a PSTH for each cell-call pair was constructed and organized into a 2D PSTH for each group where rows were sorted from the weakest response to the strongest response (Fig. 5A,B). Traces of mean ± SEM firing rate across all PVins cell-call pairs are plotted for naive mice (gray) and experienced mice (purple) of each genotype (Fig. 5C,D). Consistent with our previous report (Lau et al., 2020), auditory cortex PVins in experienced WT mice exhibited dramatically and significantly weaker responses to USVs compared with PVins in naive WT mice (Fig. 5B; n = 80 naive and 104 experienced cell-call pairs; Mann–Whitney U test, p < 0.001). In contrast, mean responses of PVins were unchanged between naive and pup-experienced PV-Mecp2 mutants (n = 64 naive and 32 experienced cell-call pairs; Mann–Whitney U test, p = 0.61). Based on all the data from our electrophysiology experiments, we conclude that selective deletion of Mecp2 in PVins is sufficient for disrupting the auditory cortical disinhibition that is triggered by exposure to pups as observed in WT virgin mice.
Selective deletion of Mecp2 in PVins recapitulates the suppression of PVin neuron inhibitory plasticity seen in Mecp2het. A, Two-dimensional-PSTHs representing the mean responses for all PVin cell-call pairs with significant responses to any USV from naive (top) and experienced (bottom) PV-Mecp2 WT mice. Each row within the PSTHs represents the Z-scored response of one cell-call pair. Rows are sorted from the greatest stimulus-evoked decrease in firing rate to the greatest stimulus-evoked increase in firing rate. Response window of 200 ms after the stimulus onset is marked by the vertical black line. B, Same as A, but the data are from PV-Mecp2 mutant mice. C, Mean ± SEM traces for USV responses of all cell-call pairs, comparing data from pup-naive (gray) and pup-experienced (purple) WT mice. D, Same as C, but the data are from PV-Mecp2 mutant mice. E, Box plot of the integrated AUC for each PVin cell-call pair in PV-Mecp2 WT mice comparing response magnitudes between naive mice and experienced mice. Mean responses were significantly weaker in mice that had experience with pups relative to pup-naive mice (naive, n = 80 cell-call pairs, 0.147 ± 0.03 Z-score*s; experienced, n = 104 cell-call pairs, 0.035 ± 0.01 Z-score*s; Mann–Whitney U test, ***p = 0.001). F, Same as E, but data are from PVin cell-call pairs collected from PV-Mecp2 mutant mice. In contrast to WT mice, PVin responses were unchanged between naive and experienced PV-Mecp2 mutant mice (naive, n = 64 cell-call pairs 0.087 ± 0.01 Z-score*s; experienced, n = 32 cell-call pairs, 0.095 ± 0.01 Z-score*s; Mann–Whitney U test, p = 0.79).
Optical recordings reveal that PVin disinhibition depends on experience and Mecp2 in PVins
One important limitation of our electrophysiology data is that it was not practical to conduct recordings from the same subjects in both naive and pup-experienced states. A more powerful experimental design would be to measure PVin activity in the same animal over the duration of its cohabitation experience with pups. A second limitation is that these recordings yield information about only one PVin at a time. It would be useful to complement those data with recordings of the neuronal population. Therefore, we used fiber photometry to make longitudinal measurements of widespread PVin activity in response to auditory stimuli as mice advanced from the pup-naive state through several days of cohabitation. Subjects were prepared by making injections of Cre-dependent AAV-DIO-GCaMP7s into the auditory cortex and by implanting an optical fiber at the same location (see above, Materials and Methods). Because PV-Cre was necessary in all mice to express GCaMP, subjects were either Mecp2flox+/flox+ (PV-Mecp2 mutant) or Mecp2flox−/flox− (PV-Mecp2 WT). We also included a group of mice that were PV-Cre+ and Mecp2het to compare results between PVin-selective Mecp2 deletion to the nonconditional, mosaic model. We conducted daily recording sessions from head-fixed mice, presenting the same set of USVs used in the neuronal recordings.
Figure 6A shows example data for three different mice. The top row of plots represents data from a PV-Mecp2 WT subject. Each row in the heat maps is a single-trial response to one specific USV. Trials above the green horizontal line were taken from sessions before cohabitation (naive state), and trials below the line were taken from sessions on P3–P5. Each call, denoted by the black vertical tick mark, elicited an abrupt increase in fluorescence that decayed over the course of 2 s. Below each heat map is a trace of the mean ± SEM calcium response from all naive trials (gray) and trials collected on days 3–5 of pup experience (purple).
PVin disinhibition is widespread and depends on experience and presence of Mecp2 in PVin. A, Comparison of longitudinal fiber photometry data from three sample subjects. Fluctuations in bulk fluorescence were measured using GCaMP7s expressed in auditory cortical PVins. Each column shows the responses of each mouse to a different USV call exemplar. Top, Data from a PV-Mecp2 WT mouse. Middle, Data from a PV-Mecp2 WT mouse that was never introduced to or cohoused with pups. Bottom, Data from a PV-Mecp2 mutant mouse. The heat maps depict the response to each USV over many trials gathered over several days. Each heat map row is one trial; those above the green line were taken from sessions before pup exposure (naive time point), and those below the line were taken from sessions on P3–P5. Below each heat map is a plot of mean ± SEM fluorescence traces from naive (gray) and experienced (purple) time points. The onset of call playback is marked with a black tick above the heat map. B, Scatterplot of mean naive and experienced PVin responses to all USVs for all mice in each experimental condition. Responses were quantified as the AUC of the Z-scored fluorescence trace during the first 2 s after stimulus onset. WT mice showed a consistent and significant decrease in response strength to USVs between naive trials and during trials on days 3–5 of pup experience (n = 8 mice; naive, 2.46 ± 0.79 Z-score*s; experienced, 1.03 ± 0.57 Z-score*s; paired t test, p ***< 0.001). No significant differences between the early time point and the late time point responses were found for WT mice that were not exposed to pups but were imaged during USV playback on the same schedule (n = 6 mice; naive, 1.57 ± 1.0 Z-score*s; experienced, 1.24 ± 0.61 Z-score*s; paired t test, p = 0.14), PV-Mecp2 mutant mice (n = 5 mice; naive, 1.61 ± 1.2 Z-score*s; experienced, 1.29 ± 0.68 Z-score*s; paired t test, p = 0.23), or Mecp2het (n = 8; naive, 1.24 ± 0.77 Z-score*s; experienced, 1.11 ± 0.83 Z-score*s; paired t test, p = 0.62). C, Identical data from the same mice, but pure tones were presented instead of calls. Responses to tones in WT mice were significantly decreased during trials on days 3–5 of pup experience compared with naive trials (n = 8 mice; naive, 2.17 ± 0.29 Z-score*s; experienced, 0.92 ± 0.13 Z-score*s; paired t test, p **< 0.01). No significant differences were found for pup-naive WT mice (n = 6 mice; naive, 1.53 ± 0.60 Z-score*s; experienced, 1.09 ± 0.18 Z-score*s; paired t test, p = 0.51), PV-Mecp2 mutant mice (n = 5 mice; naive, 0.89 ± 0.44 Z-score*s; experienced, 0.53 ± 0.20 Z-score*s; paired t test, p = 0.48), or Mecp2het mice (n = 8; naive, 1.20 ± 0.31 Z-score*s; experienced, 1.03 ± 0.34 Z-score*s; paired t test, p = 0.22).
We quantified the responses to calls as the mean AUC across trials and stimuli and compared that measure for all mice before and after pup exposure (Fig. 6B). In agreement with our single-neuron data, we found that the auditory cortex PVin population in PV-Mecp2 WT subjects exhibited consistently weaker mean responses to USVs after several days of pup exposure compared with responses measured in the naive state (Fig. 6B; n = 8 mice; paired t test with Bonferroni correction, p < 0.001). Importantly, this drop in PVin responses required experience; pup-naive control mice that were recorded on the same schedule did not show a significant decrease in PVin activity in response to the same stimulus set presented to the experienced mice (Fig. 6B; n = 6 mice; paired t test, p = 0.40). Neither Mecp2het (n = 8 mice; paired t test, p = 0.16) nor PV-Mecp2 mutants (n = 5 mice; paired t test, p = 0.56) showed a significant decrease in the responses of PVins to USVs. We also presented these same mice with a library of pure tones, and we obtained the same result. Stimulus-driven activity of the PVins in auditory cortex was significantly decreased in response to tones (Fig. 6C; n = 8 mice; paired t test with Bonferroni correction, p < 0.01). This was not true of unexposed control mice (Fig. 6C; n = 6 mice; paired t test, p = 0.51), Mecp2het (n = 8 mice; paired t test, p = 0.48), or PV-Mecp2 mutants (n = 5 mice; paired t test, p = 0.72).
Acquisition of pup retrieval is delayed by adult loss of Mecp2 in auditory cortical PVins
All the above observations indicate that PVins play an important early role in initiating cortical plasticity in response to sensory and social experience with pups. Loss of Mecp2 exclusively in this neuronal subtype, which represents only ∼10% of the neurons in the neocortex, replicates many key features of the maternal behavioral and neural pathology seen in Mecp2het. However, our approach of crossing PV-Cre mice with Mecp2flox does not specifically implicate PVins in the auditory cortex, nor does it distinguish between an acute requirement for Mecp2 in PVins in adulthood (e.g., during the initial exposure of the virgin mouse to pups) and an earlier requirement for Mecp2 in PVins for proper development of the auditory cortex to support later plasticity. We therefore devised an intersectional viral-genetic strategy to address this limitation.
Figure 7A is a schematic depiction of our strategy to target Mecp2 only in PVins in the auditory cortex and only after cortical development (see above, Materials and Methods). Briefly, we crossed Mecp2flox mice with a line that expresses Flp recombinase in PV neurons, and then at 6 weeks of age, bilaterally injected the auditory cortex with either an AAV driving the Flp-dependent expression of Cre recombinase and an HA tag (Flp-Flox) or a control vector expressing GFP (GFP-Flox).
Acquisition of pup retrieval is delayed by adult loss of Mecp2 in auditory cortex PVins. A, Schematic depiction of our experimental strategy. Mice carrying Flp recombinase after a T2A site in PV neurons were crossed with Mecp2flox mice. Offspring mice positive for both alleles were injected with an AAV driving the expression of either Flp-dependent (fDIO) Cre or GFP. The consequence of injecting fDIO-Cre is the deletion of Mecp2 from PV neurons at the time and location of our choosing, in this case, the auditory cortex of young adult mice, thereby deleting Mecp2 in PVins at the injection site. B, Swarm plot comparing retrieval latencies for control subjects that were injected with AAV-GFP (left, black points) to those for subjects that were injected with AAV-fDIO-Cre (orange points). For direct comparison, prior data from PV-Mecp2 WT (right, black points) and PV-Mecp2 mutant (red points) are also provided. A one-way ANOVA of all groups revealed significant differences among them (F = 5.02, p < 0.01). Retrieval latencies were significantly longer in mice injected with fDIO-Cre compared with control mice injected with AAV-GFP (n = 12 controls, latency, 0.230 ± 0.04; n = 12 mutants, latency, 0.381 ± 0.08; Sidak’s test, *p < 0.05).
We compared the pup retrieval performance on P0 Flp-Flox subjects to that of GFP-Flox mice and found that mean retrieval latency was longer for Flp-Flox subjects (Fig. 7B). A one-way ANOVA was used to compare P0 retrieval latencies between those groups and also between PV-Mecp2 mutant and PV-Mecp2 WT. Significant differences were detected among the groups (F = 5.02, p < 0.01), and post hoc testing detected significantly longer latencies for the Flp-Flox group (n = 12 Flp-Flox mice, n = 12 GFP-Flox mice; Sidak’s test, p < 0.05) and the PV-Mecp2 mutant group (n = 25 PV-Mecp2 mutant mice, n = 14 PV-Mecp2 WT mice; Sidak’s test, p < 0.001) when compared with their respective controls.
Discussion
Several lines of evidence from our previous work on Mecp2het mice strongly suggested that dysregulation of PVins in auditory cortex is a critical feature of the neuropathology underlying their failure to learn to perform pup retrieval behavior. Specifically, in the auditory cortices of Mecp2het virgin females cohoused with a dam and her litter, we observed dramatic overexpression of markers associated with PVins (parvalbumin protein and perineuronal nets) that are known to be antagonistic to plasticity (Pizzorusso et al., 2002; Carulli et al., 2010; de Vivo et al., 2013; Donato et al., 2013; Happel et al., 2014; Hou et al., 2017; Cisneros-Franco and de Villers-Sidani, 2019; reviewed in Rupert and Shea, 2022). This was accompanied by a lack of the disinhibitory plasticity found in the auditory cortex WT mice after exposure to pups (Lau et al., 2020). Several manipulations that ameliorated PV and PNN overexpression in Mecp2het subjects led to a resumption of behavior and partial restoration of the neural disinhibitory response (Krishnan et al., 2017; Lau et al., 2020). Here, we present evidence that deletion of Mecp2 selectively in PVins is sufficient to recreate many aspects of the neuropathology linked to the behavioral learning deficits that we observe in nonconditional, mosaic Mecp2het mutants. Importantly, just as in Mecp2het mice, the behavioral impairment could not be explained by gross locomotor deficits because there were no significant differences between wild types and mutants in mean velocity for any of the Cre lines.
Although our data emphasize the central importance of auditory cortical responses for pup retrieval, there are certainly other factors and brain regions that are important, including arousal state, oxytocin, and olfaction (Moreno et al., 2018). For example, pup retrieval is a multisensory behavior that jointly requires sound and smell (Cohen et al., 2011; Wang and Storm, 2011; Weiss et al., 2011; Fraser and Shah, 2014; Nowlan et al., 2022). Interestingly, when pup odor is delivered to the nose of either a dam or a maternally experienced surrogate, auditory cortical responses to sound, including pup calls, are modulated. Previously, we submitted an article to a preprint server in which we propose that this integration is accomplished via a pathway from pup-odor-responsive neurons in the basal amygdala to the auditory cortex (Nowlan et al., 2022). The implication of this is that acquisition and performance of pup retrieval involves multiple brain regions and stimuli. The odor-responsive input to the auditory cortex is especially interesting because it may be a mechanism for exposure to sensory characteristics of pups to trigger maternal experience-induced plasticity.
PVins have a disproportionate role in early establishment of retrieval behavior
Numerically speaking, cortical PVins make up a small population of neurons (accounting for ∼10% of cortical neurons), yet they can powerfully affect neural activity (Cardin, 2018). Indeed, we compared the effects of deleting Mecp2 in PVins only with deleting it in two other major classes of cortical inhibitory neurons, SST and VIP neurons. These populations are slightly less numerous than PVins but are of the same order of magnitude. We found no detectable effect on retrieval performance of selectively deleting Mecp2 in SSTins or VIPins. This points to a specific function during retrieval for PV neurons, among all inhibitory subtypes, that makes the brain especially vulnerable to their loss of Mecp2. As all the interneuron types interact in the cortical circuit, it is somewhat surprising to find such a specific behavioral effect from loss of Mecp2 in only one type. However, this is not unprecedented, as loss of Mecp2 in PV and SST neurons exhibit largely nonoverlapping subsets of the known characteristics of unconditional Mecp2 knockouts (Ito-Ishida et al., 2015). Mice lacking Mecp2 in VIP neurons have their own distinct characteristics such as differences in state-dependent brain activity and certain behaviors (Mossner et al., 2020). In any case, although it is reasonable to expect that different interneuron classes interact, it’s important to note that their synaptic targets and timing of activity relative to behavior may orthogonalize their contributions to network activity in some circumstances.
Moreover, the admittedly short delay in the emergence of retrieval from PVins was no stronger or longer in mice that lacked Mecp2 in homeobox protein box (Emx1) neurons, which constitute ∼88% of cortical neuron. This again suggests that the much smaller PVin population plays a specific and disproportionately large role in auditory cortical plasticity. Notably, in both PVin and Emx1 populations, removing Mecp2 caused only a delayed emergence of retrieval behavior, not the sustained deficit we observed in nonconditional, mosaic Mecp2het. This suggests that Mecp2 in PVin neurons and Emx1 neurons each have an obligatory role in the early initiation of auditory cortical plasticity, but not necessarily in its subsequent maintenance. Yet, complete Mecp2 deletion in either PVin or Emx1 neurons is less potent than mosaic absence of Mecp2 among all cell-type populations. This suggests that compensatory mechanisms involving nontargeted cell types attenuate the effects of deleting Mecp2 in only one cell type. Based on the results of deleting Mecp2 in PVins during early adulthood, such compensatory mechanisms do not involve developmental processes. Moreover, suppression of typical expression patterns of PNNs acutely, just before introduction of pups, was sufficient to improve behavior within 5 d, despite any changes in the preceding developmental trajectory. The four lines also differed considerably in their baseline behavioral variability, but they should not be directly compared with one another. We find that there can be substantial differences in this behavior among wild types of different lines, depending on genetic background. The only fair comparison is between wild types and mutants of the same line, which is why we use littermate controls.
Relationship of behavior to expression patterns and neurophysiology in PVins
PVin-specific deletion of Mecp2 caused a very similar upregulation in PV and PNNs to that seen in mosaic Mecp2het mutants (Krishnan et al., 2017). However, unlike Mecp2het mice, the upregulation was a mix of transient and persistent increases. Specifically, PV-Mecp2 mutants exhibited a persistent increase in staining intensity of PV protein, yet the increase in staining intensity of PNNs present on P1 subsided by P5. It is possible that the failure to sustain high levels of PNN staining limits the duration of the disruption of behavior in PV-Mecp2 mutants. It is worth noting that because the changes in counts of PVins and associated structures are so rapid (within 1–2 d), these changes very likely result from a change in expression intensity relative to our detection threshold, not a change in the absolute number of PVin cells themselves (i.e., cell-type identity is unlikely to change over such a short time course).
Our prior work suggested that PNN expression is closely related to retrieval performance; not only were expression and performance correlated, but the administration of chondroitinase to dissolve PNNs in the auditory cortex actually improved behavior in Mecp2het (Krishnan et al., 2017). We therefore hypothesize that the long-term establishment of well-developed PNNs in the auditory cortex is a crucial barrier to the cortical plasticity underlying pup retrieval learning. Interestingly, deletion of Mecp2 in PVins, although sufficient to establish more mature PNNs, is insufficient to sustain them. This implies that PNNs, despite preferentially surrounding PVins, are influenced by cell-autonomous and non-cell-autonomous processes on distinct timescales. In light of this, it will be interesting in future studies to see whether deletion of Mecp2 in Emx1 neurons also lead to increased PNN expression at PV synapses, that is, through a noncell-autonomous mechanism.
Our past work also revealed that maternal experience triggers disinhibited activity in the auditory cortex of WT females. In WTs this disinhibition was mediated by PVins but was abolished in PV-Mecp2 mutants (Lau et al., 2020). Here, we find that in PV-Mecp2 mutants, PVins also do not decrease their stimulus-driven activity after the female acquires experience caring for pups. Because subjects in electrophysiology experiments were recorded after P6, and subjects in fiber photometry experiments were imaged through P5, the prevention of auditory cortical disinhibition outlasted the transient behavioral and molecular effects of Mecp2 deletion in PVins.
In addition to regulating plasticity, Mecp2 affects other aspects of cortical function. For example, despite the lack of effect here on maternal retrieval, it is important for maintaining normal activity patterns and behavior in other classes of cortical interneurons, including SSTins and VIPins (Ito-Ishida et al., 2015; Mossner et al., 2020). Loss of Mecp2 function in all cortical interneurons, or even only in VIP interneurons specifically, leads to abnormal local field potential oscillations and disrupts the influence of the cortical state on the firing of individual neurons (Mossner et al., 2020). These phenotypes may reflect disruption of the balance between excitation and inhibition at the network level (Calfa et al., 2011; Banerjee et al., 2016; Li, 2022) and may also be a contributing factor to the susceptibility of Mecp2 mutants to seizures (Dolce et al., 2013). Our understanding, as a field, of different contributions of inhibitory cell types to network activity and behavior, even apart from Mecp2, is still incomplete.
Deletion of Mecp2 in PVins likely affects behavior by an acute and cell-autonomous mechanism
Importantly, our experiments with PV-specific Mecp2 knockout removed Mecp2 early in development and from all PV-expressing neurons throughout the brain. To bring greater spatiotemporal specificity to our manipulation, we adopted an intersectional strategy that required both Flp and Cre recombinases to remove Mecp2, allowing us to manipulate only PVins in the auditory cortex and only in adulthood. We found that this also produced a transient delay in pup retrieval, establishing that the plasticity mechanisms that support that behavior also acutely require Mecp2 in adulthood rather than during development alone. This observation argues in favor of the likelihood that the transient behavioral disruption is a cell-autonomous consequence of loss of Mecp2 in PVins of the auditory cortex because those are the same cells that are the effectors of the circuit disruption.
Mecp2 and social behavior
Our study is not the first to show that loss of function of Mecp2 leads to impaired social behavior. Indeed, early studies in mice reported that Mecp2 mutants have altered social interactions, an attribute they share with humans who have Rett syndrome (Zoghbi, 2005; Moretti et al., 2006). However, there are divergent data on whether disabling Mecp2 decreases (Gemelli et al., 2006; De Filippis et al., 2010) or increases (Pearson et al., 2012) sociability. The gene interacts differently with social behavior depending on the affected cell type. For example, loss of Mecp2 in peripheral somatosensory neurons apparently interferes with social interaction by rendering mice hypersensitive and averse to gentle mechanosensory stimulation (Orefice et al., 2016, 2019). Mecp2 expression is also crucial in the medial prefrontal cortex for discrimination of social partners by neuronal ensembles (Xu et al., 2022).
Implications and future directions
A number of questions remain that should be the focus of future work. First, although our optical and electrical recordings from the auditory cortex were performed in awake animals, it is not yet known how the disinhibition we observe interacts with ongoing activity in freely behaving mice that are performing pup retrieval. PVins are important to the phenotype of Mecp2het mice, and they can modulate cortical activity from the single-unit level up to more widespread features of brain state (Cardin, 2018), including gamma oscillations (Cardin et al., 2009; Sohal et al., 2009). Recording from actively retrieving mice may reveal unappreciated dynamic influences such as locomotor activity and arousal that may be mediated by PVins (Nelson et al., 2013; Schneider et al., 2014; Henschke et al., 2021). Second, the relationship between PVin activity and construction of PNNs is not well understood. An interesting goal for future work will be to ascertain the relationship between PNNs and PVin activity and how they are affected by the activity of other cell types and on what time scale. Third, these questions about cell-autonomous and noncell-autonomous influences of Mecp2 will be enlightened by targeted recordings from neurons that are individually identified as Mecp2+ and Mecp2− in mosaic Mecp2het mice.
Footnotes
This work was supported by National Institutes of Health–National Institute of Mental Health Grant R01MH106656 and a Feil Foundation grant to S.D.S. and an Autism Speaks Royal Arch Masons Predoctoral Fellowship (#11001) to D.D.R. We thank A. Zador, H. Hsieh, Z. J. Huang, and A. Banerjee for comments and guidance; C. Nguyen, C. Kelahan, and R. Palaniswamy for technical support; and Rob Eifert and the Cold Spring Harbor Laboratory machine shop staff for equipment support.
The authors declare no competing financial interests.
- Correspondence should be addressed to Stephen D. Shea at sshea{at}cshl.edu
