CDKL5-mediated developmental tuning of neuronal excitability and concomitant regulation of transcriptome

Abstract

Cyclin-dependent kinase-like 5 (CDKL5) is a serine–threonine kinase enriched in the forebrain to regulate neuronal development and function. Patients with CDKL5 deficiency disorder (CDD), a severe neurodevelopmental condition caused by mutations of CDKL5 gene, present early-onset epilepsy as the most prominent feature. However, spontaneous seizures have not been reported in mouse models of CDD, raising vital questions on the human-mouse differences and the roles of CDKL5 in early postnatal brains. Here, we firstly measured electroencephalographic (EEG) activities via a wireless telemetry system coupled with video-recording in neonatal mice. We found that mice lacking CDKL5 exhibited spontaneous epileptic EEG discharges, accompanied with increased burst activities and ictal behaviors, specifically at postnatal day 12 (P12). Intriguingly, those epileptic spikes disappeared after P14. We next performed an unbiased transcriptome profiling in the dorsal hippocampus and motor cortex of Cdkl5 null mice at different developmental timepoints, uncovering a set of age-dependent and brain region-specific alterations of gene expression in parallel with the transient display of epileptic activities. Finally, we validated multiple differentially expressed genes, such as glycine receptor alpha 2 and cholecystokinin, at the transcript or protein levels, supporting the relevance of these genes to CDKL5-regulated excitability. Our findings reveal early-onset neuronal hyperexcitability in mouse model of CDD, providing new insights into CDD etiology and potential molecular targets to ameliorate intractable neonatal epilepsy.

electroencephalography (EEG), RNA sequencing (RNA-seq), CDKL5 deficiency disorder (CDD), neonatal epilepsy, KCC2

Introduction

Cyclin-dependent kinase-like 5 (CDKL5; OMIM#300203) is a serine–threonine kinase highly expressed in the developing mammalian brain [1,2]. Mutations in this gene cause CDKL5 deficiency disorder (CDD; OMIM#300672; ICD-10-CM code: G40.42), a rare developmental encephalopathy characterized by severe early onset seizures, psychomotor dysfunction, autistic features and intellectual disability [3–8]. CDD is known as one of the most common forms of genetic epilepsy in children [9–11]. The epileptic activities accompanied with developmental delays in children with CDD are frequently associated with devastating cognitive, psychiatric and behavioral disabilities later in life. It is likely that amendment of early-onset epilepsy may potentially prevent the consequent neurological and cognitive deficits. However, children with CDD are resistant to most of the anti-epileptic drugs [12–14] and the neuropathogenic mechanisms underlying neonatal seizures in CDD remain unclear.

Several mouse models of CDD have been developed that exhibit behavioral phenotypes mimicking the primary symptoms of CDD [15–18]. However, none of these studies report spontaneous seizures in Cdkl5 null mice. Mice carrying a deletion of exon 6 of Cdkl5 gene exhibit normal electroencephalographic (EEG) activities for up to 240 h when recording was conducted from the hippocampus in male mice at 9–12 weeks of age [15]. Mice lacking exon 4 of Cdkl5 gene also show normal basal EEG in the somatosensory cortex [17]. Moreover, mice carrying Cdkl5 deletion of exon 2 or R59X-knock-in present enhanced seizure susceptibility in response to N-methyl-D-aspartate (NMDA) treatment but no spontaneous seizures are found in untreated mutants [18,19]. Notably, all of these seizure measurements were conducted in mice older than one month of age that is beyond adolescence in humans [20,21]. Concerning the onset of seizures in CDD patients occurs in a median age of 6 weeks after birth [13,14,22], it remains to be determined if early-onset spontaneous seizures occur in neonatal CDD mice, even though epileptic spasms have recently been reported in aged Cdkl5 heterozygous females [23,24].

We previously noticed that Cdkl5 null pups showed increased spasm-like twists during recordings of ultrasound vocalization at the early postnatal period (P4—P10) comparing to their littermate controls [25]. Similar spasm-like behaviors have been described in mice with Arx gene mutations that cause infantile spasm [26,27], suggesting that pups lacking CDKL5 may experience early-onset epilepsy soon after birth as well. Thus, we focused our studies on neonatal mice to determine the role of CDKL5 in early postnatal brains. Using longitudinal EEG recordings through a wireless transmitter system coupled with simultaneous videotaping of ictal behaviors, we found that mice lacking CDKL5 showed spontaneous epileptic EEG discharges at P12 that returned to normal in the preweaning week. Given that previous studies have linked neonatal seizures to dysregulated neuronal excitability triggered by interneuronopathies, channelopathies or imbalanced transition of GABA action from excitation to inhibition during development [28–31], we thus performed an unbiased transcriptome profiling in the dorsal hippocampus (dHC) and motor cortex (mCTX), of Cdkl5 null mice at three different developmental stages, followed by validation of the differentially expressed genes (DEGs) at the transcript and protein levels. We found that the transient epileptic phenotype was coupled with an age- and brain region-specific alteration of gene expression. Importantly, multiple CDKL5-regulated molecules have been identified and validated at different ages closely related to the onset and offset of neonatal epilepsy. These findings reveal new avenues for future assessments of age-specific therapeutics against CDD-related early onset epilepsy.

Results

Loss of CDKL5 triggers seizure-like EEG discharges in neonatal mice

To identify potential EEG abnormalities in Cdkl5 null (Cdkl5^−/y, KO) mice at early postnatal ages, we implanted wireless transmitters in mouse pups at P10, the youngest age feasible for this type of surgery, and recorded EEG activities using a miniature telemetry system (Fig. 1A, B). In cases when the implanted transmitter was occasionally damaged by the dam, those data were excluded from further analysis. Given the battery power limit of 14 days for the wireless transmitters, we collected EEG data from P11 to P24 upon implantation at P10 (Supplementary Material, Fig. S1). The number of spike clusters showed a notable increase in pups lacking CDKL5 during the P11–P12 period (Supplementary Material, Fig. S1A–B’), but this increase disappeared gradually compared to WT pups (Supplementary Material, Fig. S1F, F’).

Figure 1

Loss of CDKL5 increased spontaneous epileptic discharges in neonatal mice. (A) Schematic diagram showing the implantation of a wireless transmitter with two working electrodes targeted the hippocampus (red) and one reference electrode targeted the cerebellum (blue) in a mouse at postnatal day 10 (P10). The gray circle shows a battery. (B) Representative micrographs showing the placement of the working electrodes (upper panel, arrowheads) and reference electrode (lower panel, arrowhead). Scale bar, 500 μm. (C) Representative traces of EEG activities recorded from a pair of pups at P12. The boxed areas are enlarged from C₁ (C₁’, 4 min), C₂ (C₂’, 60 s) and C₃ (C₃’, 10 s) to C₂ (C₂’), C₃ (C₃’) and C₄ (C₄’), sequentially. Scale bars are included at the bottom right corner of C₁-C₄. The recurrent episodes of epileptiform discharges appear in Cdkl5^−/y pups (C₁’–C₄’) but not in wild-type (WT) controls (C₁-C₄). (D-E) Quantification of population spikes shows that the total number of events are significantly increased in Cdkl5^−/y pups at P12 along the 14-days longitudinal study. All data are presented as mean ± SEM. ^**p < 0.01; ^***p < 0.001; mixed effects model followed by multiple comparison with Bonferroni’s test.

Notably, the epileptiform discharges occurred in Cdkl5^−/y pups at P12 are composed of repetitive bursts with each episode lasting for 10 to 20 s, which were absent in WT littermate control pups (Fig. 1C). To quantify the electrographic spikes, we analyzed each EEG trace for the last 90 min (i.e. 10–100 min) using the Clampex software. For event detection, the “threshold” amplitude was set to 0.9 volts based on our pilot analysis (arrow and dash line in Supplementary Material, Fig. S2A). The total number of spike events showed a significant difference between genotypes in an age-dependent manner [interaction of genotype x age: p = 0.0003, F (5, 36) = 6.265], in which the event number was significantly higher in Cdkl5^−/y pups compared to that in WT pups at P12 (3766.0 ± 961.7 in KO vs. 410.3 ± 118.6 in WT, p < 0.01; Fig. 1D). The enhancement appeared even greater when the event number of Cdkl5^−/y pups was normalized to their WT littermate controls (11.69 ± 3.23 folds to WT, p < 0.001; Fig. 1E).

The seizure-like activities were then evaluated through “burst analysis”, in which a burst was defined as a cluster of continuous discharges for more than 20 events whose inter-event-intervals were less than 1 s. The results showed that the “total number of bursts” was increased in Cdkl5^−/y compared to WT pups at P12. However, this enhancement was diminished to WT levels from P14 to P24, demonstrating an age-dependent normalization of EEG activities in Cdkl5^−/y pups (Fig. 2A, B). Notably, the “mean duration of bursts” and the “mean spike number per burst”, both indicating the severity of epileptic discharges, were significantly increased in Cdkl5^−/y pups at P12 (p < 0.001), but normalized to WT levels afterward (Fig. 2C–F). There was no change in the burst event frequency between Cdkl5^−/y pups and WT littermates throughout the recording period of P11 to P24 (Fig. 2G, H).

Figure 2

Burst analysis of EEG spikes. The epileptiform discharges containing repetitive episodes of EEG spikes were quantified using the function of “burst analysis” in Clampex software. A burst was defined as a series of continuous discharges for more than 20 events whose inter-event-intervals are less than 1 s. The total number (A, B) and mean duration (C, D) of bursts, and the mean spike number per burst (E, F) are increased in Cdkl5^−/y mice at the age of P12. The effects of CDKL5 ablation are shown in B, D and F by normalizing the results of mutants with those of their littermate controls. (G-H) The frequency of burst events is comparable between genotypes throughout the recording period. All data are presented as mean ± SEM. ^*p < 0.05; ^***p < 0.001; mixed effects model followed by multiple comparison with Bonferroni’s test.

To compare the amplitude of spike events, we normalized the output spikes of 1 V to 0.25 mV based on the amplification of the transmitters (4000X). The mean event amplitude was indeed comparable between genotypes, although there was a trend of increase at P11 and P12 (Supplementary Material, Fig. S2B, C). In addition, there was no significant alteration in the mean frequency of total events in Cdkl5^−/y pups compared to WT pups with age, except a slight increase found in Cdkl5^−/y pups at P12 (Supplementary Material, Fig. S2D, E). Together, these findings demonstrated that spontaneous epileptiform discharges occurred in CDD mice at a specific neonatal time window.

Loss of CDKL5 elicits ictal behaviors during the prolonged bursts at P12

Given that ictal behaviors have been observed as a primary symptom of infantile epilepsy [26,27,32], moreover, focal motor/hypermotor seizures and massive myoclonic jerks frequently present in CDD patients during the first two phases of disease onset [33], we next examined the time-locked videos captured simultaneously with EEG events of bursts. We assessed the pup behaviors during the prolonged burst events at P12, and found that the burst events were associated with frequent limb twitches, rapid limb stretching, tail flicks and occasional jerky movements of the whole body in Cdkl5^−/y pups (Supplementary Material, Fig. S3A, upper panel; Movie 1). However, WT littermate pups tended to stay quietly without obvious movements during the bursting period (Supplementary Material, Fig. S3A, lower panel; Movie 2). By quantifying these spasm-like behaviors with a score of 0 to 3, with 3 representing spasm-like jerky movements of the body or limbs [27], we found that the accumulative scores were consistently higher in three Cdkl5^−/y pups compared to the scores derived from their WT littermates (Supplementary Material, Fig. S3B). The spasm-like behaviors are diminished in mutant pups at P14 and afterwards. Importantly, upon close examination of the videos and EEGs, we observed that those burst events occurred prior to any pup movements, excluding the possibility that the burst events are the result of muscle contractions.

Previous studies have shown that patients with frontal hyperkinetic seizures frequently display “ictal grasping” after seizure onset [34,35]. We thus examined the motor grasping skills in neonatal mice and found that Cdkl5^−/y pups tended to grasp stronger than their WT littermates in a wire suspension test at two weeks after birth, and the holding time was significantly increased at P18 (46.1 ± 7.2 sec in KO vs. 19.8 ± 2.1 sec in WT, p < 0.001; Supplementary Material, Fig. S3C). Therefore, loss of CDKL5 may cause myoclonic jerky behaviors in the week prior to weaning besides the ictal EEG activities at P12.

Increased firing rate in neonatal CDD mice

In CDD patients, the onset of seizures occurs soon after birth at the median age of 6 weeks [13,14], a timepoint equivalent to the end of the first postnatal week in mice [21]. Interestingly, transient increase of neuronal excitability has been reported in a human iPSC model of CDD during the early stage of neural network formation [36]. It raises a possibility that the ictal activities may appear in Cdkl5 mutant pups in the first postnatal week. However, the technical limitation of wireless EEG recording does not allow us to record EEG in pups younger than P10. We thus performed whole-cell patch clamp to evaluate neuronal excitability by measuring the frequency of evoked action potentials in hippocampal slices of P7 mice. We found that loss of CDKL5 does not change the resting membrane potential of neurons compared to WT littermate controls, consistent with previous findings [37,38]. In contrast, the frequency of evoked action potentials was significantly increased in response to current steps (0–80 pA) compared to WT littermate controls (Supplementary Material, Fig. S4), suggestting that CDKL5 deficiency leads to neuronal hyperexcitability in mice during the first postnatal week, the time window corresponding to that of seizure onset in CDD patients.

Age-dependent alterations of gene expression in the dorsal hippocampus and motor cortex

To decipher potential molecular mechanisms underlying neuronal hyperexcitability in Cdkl5 mutant pups, we took an unbiased approach to analyze transcriptome of Cdkl5^−/y and WT mice at the age of P7, P12, and P17, spanning across the transient manifestation of epileptic activities. The dorsal hippocampus (dHC), the major brain region implicated in epilepsy, was subjected to the transcriptome profiling using whole genome RNA sequencing (RNA-seq). The principal component analysis (PCA) plot showed that the significant age effects were mainly contributed by the principal component 1 (PC1, 37.5%), in which the sample clusters of P7, P12, and P17 were completely separated and aligned along the PC1 axis (Supplementary Material, Fig. S5A, B). Notable differences between genotypes appeared at P7, rather than P12 and P17. The PCA plot showed that the gene expression clusters of Cdkl5^−/y and WT controls at P7 were differently oriented (Supplementary Material, Fig. S5B, Fig. 3A). Analysis of differentially expressed genes (DEGs, by DESeq2) for the sample clusters of P7 (KO = 5, WT = 4) that contained 15 932 expressed genes in the dataset of total 55 536 entries, 104 DEGs were identified (p-value thresholds = 0.001 and log2 fold-change threshold = 1.2), including 21 down-regulated, 77 up-regulated and 6 unknown genes (Fig. 3B). Using the same filtering criteria as that for P7, in contrast, only 9 and 11 DEGs were identified from the transcriptome of P12 and P17, respectively.

Figure 3

Loss of CDKL5 altered transcriptome profile in the neonatal hippocampus. The total RNAs were extracted from the dorsal hippocampus (dHC) of mice at P7 for transcriptome analysis. (A) Principal component analysis (PCA) plot shows transcriptomic distribution of the groups of Cdkl5 null mice and WT controls. Each spot represents the transcriptome data from an individual mouse. (B) Heatmap representation of the 104 differentially expressed genes (DEGs) identified with p-value thresholds of 0.001 and log2 fold-change threshold of 1.2. (C) Top 20 gene ontology (GO) annotations for the 104 DEGs. (D-F) Bar graphs represent the normalized read counts for upregulated (D) and downregulated (E) DEGs, and for the interneuron marker genes (F), as indicated at the bottom of the charts. All data are presented as mean ± SEM. ^*p < 0.05; ^**p < 0.01; ^***p < 0.001; Student’s-t test.

To further understand the developmental changes of gene expression profiles in the cortex, we also analyzed the transcriptome of the motor cortex (mCTX) from mice at the ages of P7, P12, P17, and 3 months (3 m). We focused our transcriptome survey in the mCTX largely because of its relevance to spasm-like motor behaviors (Supplementary Material, Fig. S3). The PCA plot showed that the sample clusters of the mCTX at different ages were mainly different along PC1 (32.68%), similar to that in the dHC, showing an age-dependent gene expression profile. However, the PCA differences between genotypes were primarily exhibited at P12 and P17 in the mCTX (Supplementary Material, Fig. S5C, D).

Altered hippocampal gene expression in neonatal mice lacking CDKL5

We next examined the gene categories of the 104 DEGs identified from the P7 dHC by Gene Ontology (GO) analysis (p. adj. < 0.05), which annotates gene function in terms of three different aspects [i.e., molecular function (MF), biological process (BP) and cellular component (CC)]. The top-20 GO terms (Fig. 3C) showed that the most significantly altered categories were “postsynaptic membrane” (CC, GO:0045211, 19/98, p.adj. < 0.0001), “synapse organization” (BP, GO:0050808, 12/98, p.adj. < 0.001) and “ion gated channel activity” (MF, GO:0005267, 10/98, p.adj. < 0.001). The genes belong to categories of “neuron to neuron synapse” (CC, GO:0098984, 10/98, p.adj. < 0.01), “regulation of membrane potential” (BP, GO:0042391, 10/98, p.adj. < 0.01) and “passive transmembrane transporter activity” (MF, GO:0022803, 10/98, p.adj. < 0.01) were also indicated.

By quantifying the normalized read counts, we found that loss of CDKL5 significantly increased the transcript levels of genes encoding transmembrane proteins regulating ion transport, such as Slc8a3 (solute carrier family 8, member 3, a sodium/calcium exchanger, also called NCX3; p < 0.001) and Kcnn1 (potassium intermediate/small conductance calcium-activated channel, subfamily N, member 1, also called SK channels; p < 0.01). The genes encoding proteins associated with neuronal excitability and excitatory synaptogenesis, such as Pdzd2 (PDZ domain containing 2; p < 0.001) and Lrrtm3 (leucine rich repeat transmembrane neuronal 3; p < 0.01) were also significantly increased in CDKL5 deficient mice (Fig. 3D, Table 1). Intriguingly, loss of CDKL5 significantly increased the transcript levels of GABA receptor, Gabrg1 (gamma-aminobutyric acid A receptor, subunit gamma 1; p < 0.01, Table 2), while unchanged the expression of NMDA or AMPA glutamate receptors (Table 3) in the dHC at P7.

Table 1

The transcript levels of differentially expressed genes in the dHC of Cdkl5 mutants.

Gene code	Gene name	Age	Normalized read count		Statistics
			WT	Cdkl5−/y
Cdkl5	cyclin-dependent kinase-like 5	P7	3137.8 ± 172.1	1177.4 ± 19.0	p = 4.04E-06***
		P12	3377.3 ± 141.7	926.3 ± 15.3	p = 6.71E-05***
		P17	3677.7 ± 26.9	907.8 ± 15.1	p = 9.48E-12***
Slc8a3	solute carrier family 8 (sodium/calcium exchanger), member 3	P7	636.0 ± 23.3	832.9 ± 24.0	p = 0.00067***
		P12	582.5 ± 9.4	584.8 ± 10.6	p = 0.90063
		P17	371.1 ± 9.4	432.7 ± 13.8	p = 0.01739*
Kcnn1	potassium intermediate/small conductance calcium-activated channel, subfamily N, member 1	P7	467.5 ± 10.9	589.7 ± 21.5	p = 0.00232**
		P12	458.1 ± 31.2	473.7 ± 17.2	p = 0.73929
		P17	267.6 ± 11.5	302.7 ± 6.8	p = 0.04591*
Pdzd2	PDZ domain containing 2	P7	873.9 ± 40.9	1221.1 ± 44.4	p = 0.00081***
		P12	959.0 ± 15.2	1020.1 ± 19.2	p = 0.11116
		P17	1102.3 ± 53.1	1129.7 ± 22.4	p = 0.66568
Lrrtm3	leucine rich repeat transmembrane neuronal 3	P7	308.4 ± 9.5	411.4 ± 20.0	p = 0.00383**
		P12	482.0 ± 51.7	468.8 ± 16.5	p = 0.85287
		P17	331.3 ± 19.0	347.9 ± 21.4	p = 0.63308
St18	suppression of tumorigenicity 18	P7	535.8 ± 30.6	326.2 ± 16.2	p = 0.00036***
		P12	369.5 ± 13.6	357.5 ± 30.5	p = 0.78362
		P17	516.8 ± 19.1	560.2 ± 30.6	p = 0.35010
Glra2	glycine receptor, alpha 2 subunit	P7	1120.9 ± 39.3	785.8 ± 48.8	p = 0.00135**
		P12	509.0 ± 28.6	394.2 ± 14.1	p = 0.02280*
		P17	302.5 ± 7.4	210.5 ± 4.3	p = 2.26E-05***
Nrp2	neuropilin 2	P7	4445.4 ± 242.4	3641.7 ± 146.1	p = 0.02055*
		P12	1705.4 ± 228.5	1773.9 ± 35.4	p = 0.78200
		P17	1587.1 ± 42.7	1455.8 ± 33.2	p = 0.06623
Cntn2	contactin 2	P7	3767.8 ± 277.5	2974.6 ± 156.5	p = 0.03386*
		P12	1250.3 ± 46.1	1277.9 ± 23.5	p = 0.62210
		P17	1192.6 ± 4.5	1286.7 ± 26.2	p = 0.02605*
Cck	cholecystokinin	P7	3504.3 ± 208.7	2911.2 ± 136.4	p = 0.03777*
		P12	3266.0 ± 138.5	3113.7 ± 24.7	p = 0.42673
		P17	2682.2 ± 39.5	2433.1 ± 82.1	p = 0.06058
Sst	somatostatin	P7	1497.2 ± 87.7	2021.2 ± 126.4	p = 0.01467*
		P12	2709.5 ± 389.7	2798.7 ± 80.3	p = 0.86365
		P17	1938.5 ± 75.3	2044.1 ± 108.9	p = 0.52466
Npy	neuropeptide Y	P7	476.4 ± 34.9	485.6 ± 13.4	p = 0.79361
		P12	628.6 ± 37.3	629.2 ± 16.7	p = 0.99197
		P17	522.4 ± 16.5	518.1 ± 18.5	p = 0.88423
Nos1	nitric oxide synthase 1, neuronal	P7	1064.3 ± 37.0	998.6 ± 19.9	p = 0.14108
		P12	694.7 ± 31.9	798.9 ± 15.7	p = 0.07481
		P17	797.5 ± 17.9	786.9 ± 18.5	p = 0.73173
Calb1	calbindin 1	P7	1288.7 ± 95.4	1386.2 ± 89.0	p = 0.48134
		P12	1598.9 ± 151.8	1890.7 ± 71.9	p = 0.22892
		P17	1778.7 ± 34.0	1502.2 ± 66.2	p = 0.01867*
Calb2	calbindin 2 (calretinin)	P7	1132.9 ± 40.3	1056.3 ± 57.2	p = 0.33409
		P12	1444.3 ± 515.2	1032.1 ± 138.2	p = 0.56227
		P17	700.1 ± 102.5	561.0 ± 49.7	p = 0.28994
Pvalb	parvalbumin	P7	(none)	(none)	—
		P12	147.1 ± 18.5	159.4 ± 28.0	p = 0.77978
		P17	292.9 ± 17.3	367.8 ± 11.1	p = 0.01256*
Gad1	glutamate decarboxylase 1	P7	2321.3 ± 87.1	2487.5 ± 127.9	p = 0.34451
		P12	3041.1 ± 77.3	3444.2 ± 72.4	p = 0.03597*
		P17	3802.2 ± 29.1	4317.9 ± 95.9	p = 0.00445**
Kcnq3	potassium voltage-gated channel, subfamily Q, member 3	P7	4552.8 ± 273.2	5126.6 ± 173.4	p = 0.10685
		P12	5077.0 ± 124.4	5549.6 ± 48.2	p = 0.04442*
		P17	5403.4 ± 60.5	5852.2 ± 123.0	p = 0.03163*

Notes: dHC, dordal hippocampus; none, not detected.

^*p < 0.05.

^**p < 0.01.

^***p < 0.001.

Student’s t-test. P7: n = 4 (WT) or 5 (KO); P12: n = 3; P17: n = 4 (WT) or 5 (KO).

Table 2

The transcript levels of GABA-related genes in the dHC at P7.

Gene code	Gene name	Normalized read counts		Statistics
		WT	Cdkl5−/y
GABAA receptors
Gabra1	gamma-aminobutyric acid (GABA) A receptor, subunit alpha 1	2675.5 ± 194.1	3176.4 ± 220.2	p = 0.1416
Gabra2	gamma-aminobutyric acid (GABA) A receptor, subunit alpha 2	1999.3 ± 66.7	1894.5 ± 53.0	p = 0.2514
Gabra3	gamma-aminobutyric acid (GABA) A receptor, subunit alpha 3	1283.9 ± 93.6	1533.2 ± 70.3	p = 0.0660
Gabra4	gamma-aminobutyric acid (GABA) A receptor, subunit alpha 4	938.0 ± 32.9	982.2 ± 20.8	p = 0.2751
Gabra5	gamma-aminobutyric acid (GABA) A receptor, subunit alpha 5	4092.7 ± 36.0	4055.3 ± 92.7	p = 0.7436
Gabra6	gamma-aminobutyric acid (GABA) A receptor, subunit alpha 6	(none)	(none)	—
Gabrb1	gamma-aminobutyric acid (GABA) A receptor, subunit beta 1	3059.0 ± 172.5	3103.0 ± 77.1	p = 0.8087
Gabrb2	gamma-aminobutyric acid (GABA) A receptor, subunit beta 2	1740.4 ± 152.5	2030.5 ± 72.9	p = 0.1078
Gabrb3	gamma-aminobutyric acid (GABA) A receptor, subunit beta 3	7316.1 ± 214.8	7284.4 ± 95.5	p = 0.8883
Gabrg1	gamma-aminobutyric acid (GABA) A receptor, subunit gamma 1	466.4 ± 11.0	577.2 ± 26.4	p = 0.0097**
Gabrg2	gamma-aminobutyric acid (GABA) A receptor, subunit gamma 2	2835.1 ± 90.1	2896.9 ± 92.0	p = 0.6512
Gabrg3	gamma-aminobutyric acid (GABA) A receptor, subunit gamma 3	716.6 ± 46.4	864.7 ± 39.1	p = 0.0435*
Gabrd	gamma-aminobutyric acid (GABA) A receptor, subunit delta	114.7 ± 3.2	126.7 ± 11.6	p = 0.4033
Gabrq	gamma-aminobutyric acid (GABA) A receptor, subunit theta	37.3 ± 4.1	53.9 ± 3.6	p = 0.0193  *
GABAB receptors
Gabbr1	gamma-aminobutyric acid (GABA) B receptor, 1	12414.2 ± 306.1	12875.1 ± 242.5	p = 0.2697
Gabbr2	gamma-aminobutyric acid (GABA) B receptor, 2	5721.9 ± 222.8	5900.1 ± 169.8	p = 0.5368
GABA transporters
Slc6a1	solute carrier family 6 (neurotransmitter transporter, GABA), member 1	6595.7 ± 82.7	6933.9 ± 257.7	p = 0.2983
Slc6a11	solute carrier family 6 (neurotransmitter transporter, GABA), member 11	5418.1 ± 117.9	5191.1 ± 75.1	p = 0.1342
Slc6a12	solute carrier family 6 (neurotransmitter transporter, GABA), member 12	27.1 ± 1.2	21.5 ± 4.0	p = 0.2746
Slc6a13	solute carrier family 6 (neurotransmitter transporter, GABA), member 13	170.4 ± 15.9	179.9 ± 24.3	p = 0.7670
Slc32a1	solute carrier family 32 (GABA vesicular transporter), member 1	664.4 ± 80.8	670.3 ± 31.5	p = 0.9426

Notes: dHC, dorsal hippocampus; none, not detected.

^*p < 0.05.

^**p < 0.01.

Student’s t-test. WT, n = 4; KO, n = 5.

Table 3

The transcript levels of glutamate receptors in the dHC at P7.

Gene code	Gene name	Normalized read counts		Statistics
		WT	Cdkl5−/y
NMDA receptors
Grin1	glutamate receptor, ionotropic, NMDA1 (zeta 1)	3497.5 ± 120.3	3640.2 ± 185.0	p = 0.6061
Grin2a	glutamate receptor, ionotropic, NMDA2A (epsilon 1)	2244.1 ± 195.3	2432.7 ± 133.2	p = 0.4923
Grin2b	glutamate receptor, ionotropic, NMDA2B (epsilon 2)	13676.3 ± 1282.6	14078.0 ± 247.7	p = 0.7727
Grin2c	glutamate receptor, ionotropic, NMDA2C (epsilon 3)	165.2 ± 16.6	208.7 ± 22.6	p = 0.2327
Grin2d	glutamate receptor, ionotropic, NMDA2D (epsilon 4)	536.3 ± 9.4	586.0 ± 19.3	p = 0.1007
Grin3a	glutamate receptor, ionotropic, NMDA3A	3164.8 ± 263.6	3621.8 ± 157.5	p = 0.2131
Grin3b	glutamate receptor, ionotropic, NMDA3B	(none)	(none)	—
Grid1	glutamate receptor, ionotropic, delta 1	1253.5 ± 21.2	1314.0 ± 24.3	p = 0.1513
Grid2	glutamate receptor, ionotropic, delta 2	391.5 ± 13.3	419.2 ± 17.6	p = 0.3221
AMPA receptors
Gria1	glutamate receptor, ionotropic, AMPA1 (alpha 1)	12940.3 ± 412.4	12618.8 ± 419.9	p = 0.6492
Gria2	glutamate receptor, ionotropic, AMPA2 (alpha 2)	13640.7 ± 572.3	13602.8 ± 167.0	p = 0.9528
Gria3	glutamate receptor, ionotropic, AMPA3 (alpha 3)	1807.7 ± 54.9	1903.8 ± 67.9	p = 0.3794
Gria4	glutamate receptor, ionotropic, AMPA4 (alpha 4)	1061.4 ± 11.2	1187.4 ± 41.2	p = 0.0506
Kainate receptors
Grik1	glutamate receptor, ionotropic, kainate 1	561.8 ± 26.1	594.7 ± 19.7	p = 0.3976
Grik2	glutamate receptor, ionotropic, kainate 2 (beta 2)	2221.5 ± 51.8	2051.6 ± 61.9	p = 0.1148
Grik3	glutamate receptor, ionotropic, kainate 3	1263.3 ± 89.0	1629.4 ± 46.1	p = 0.0113*
Grik4	glutamate receptor, ionotropic, kainate 4	1141.9 ± 30.6	1029.4 ± 46.3	p = 0.1342
Grik5	glutamate receptor, ionotropic, kainate 5 (gamma 2)	3678.0 ± 63.1	3773.8 ± 101.1	p = 0.5249
Metabotropic glutamate receptors
Grm1	glutamate receptor, metabotropic 1	670.5 ± 30.4	645.2 ± 17.8	p = 0.5284
Grm2	glutamate receptor, metabotropic 2	565.0 ± 63.8	829.9 ± 50.6	p = 0.0227*
Grm3	glutamate receptor, metabotropic 3	1266.4 ± 72.6	1426.4 ± 32.4	p = 0.0994
Grm4	glutamate receptor, metabotropic 4	261.1 ± 7.2	331.2 ± 19.9	p = 0.0320*
Grm5	glutamate receptor, metabotropic 5	6862.1 ± 263.6	6831.3 ± 143.1	p = 0.9267
Grm6	glutamate receptor, metabotropic 6	(none)	(none)	—
Grm7	glutamate receptor, metabotropic 7	1113.0 ± 9.0	1146.5 ± 23.0	p = 0.3083
Grm8	glutamate receptor, metabotropic 8	100.1 ± 1.9	145.6 ± 11.1	p = 0.0152*

Notes: dHC, dorsal hippocampus; none, not detected.

^*p < 0.05.

Student’s t-test. WT, n = 4; KO, n = 5.

Among the down-regulated DEGs, suppression of tumorigenicity 18 (St18; p < 0.001) and glycine receptor, alpha 2 subunit (Glra2; p < 0.01) were the most affected ones in addition to Cdkl5 (p < 0.0001)(Fig. 3E, Table 1). Given that GLRA2 has been implicated not only in glycine-mediated inhibitory transmission, but also in forebrain tangential migration of interneurons [39,40], we noticed that neuropilin 2 (Nrp2) and contactin 2 (Cntn2, also called TAG1), two genes critical for tangential migration [41–44], were also down-regulated in the dHC of Cdkl5 mutants at P7 (p < 0.05, Fig. 3E, Table 1). Moreover, the expression of Glra2 was also significantly reduced in dHC at P12 (p < 0.05) and P17 (p < 0.001), suggesting a persistent regulation of CDKL5 on Glra2 expression in the developing hippocampus.

To further depict the expression of genes related to hippocampal interneurons, we next examined the transcript levels of marker genes for different types of interneurons in the P7 dHC. We found that the normalized read counts of transcripts were significantly reduced for cholecystokinin (Cck), while increased for somatostatin (Sst) in Cdkl5^−/y mice (p < 0.05, Fig. 3F, Table 1). The expression of marker genes for other types of interneurons, such as neuropeptide Y (Npy), vasoactive intestinal polypeptide (Vip), neuronal nitric oxide synthase 1 (Nos1), calbindin (Calb1) and calretinin (Calb2), were comparable between Cdkl5^−/y and WT controls (Fig. 3F, Table 1). No parvalbumin transcript was detected in mouse brain at P7, consistent to previous reports [45]. These data suggest that excessive cation transport and glutamate signaling (bypass NMDA and AMPA receptors), as well as defective inhibitory transmission and tangential migration of GABAergic interneurons, caused by CDKL5 loss, may all contribute to enhanced neuronal excitability in neonatal hippocampus of mice at P7.

Loss of CDKL5 downregulates dopamine D2 signaling-associated genes in the motor cortex

Upon analysis of the transcriptome of the mCTX at P12, we identified 8 DEGs (p-value threshold as 0.001 and log2 fold-change threshold as 2) from 16 655 genes in Cdkl5^−/y compared to WT controls (n = 3). As shown in the MA plot, a scatter plot of log2 fold changes (M, on the y-axis) versus the average expression counts (A, on the x-axis), the 8 DEGs composed of 2 unknown and 6 down-regulated genes, including dopamine d2 receptor (Drd2) and retinoid X receptor gamma (Rxrg), a transcription regulator of Drd2 [46] (Fig. 4B, Supplementary Table 1). Adenosine 2a receptor (Adora2a) and regulator of G-protein signaling 9 (Rgs9) were also significantly down-regulated with a fold-change threshold of 1.2. GO analysis (adjust-p threshold = 0.05) showed that the most significantly affected categories were “synaptic transmission, dopaminergic” (BP, GO:0001963, 3/9, p.adj. < 0.0001). This is further illustrated with a heatmap for those DEGs at P12 (Fig. 4C).

Figure 4

Loss of CDKL5 altered transcriptome profile in the motor cortex. (A) Schematic diagram illustrating the tissues harvested from the motor cortex (mCTX, gray) for transcriptome analysis. MA plot (B), heatmap (C) and normalized read counts (D) shows that specific genes related to dopaminergic D2 signaling are significantly downregulated in the mCTX of Cdkl5^−/y mutants at P12. The 8 DEGs in B (red dots) are numbered and annotated in supplementary Table 1. (E, F) Numerous genes linked to GABAergic transmission are up-regulated in the mCTX of Cdkl5^−/y mutants at P17. All data in D and F are presented as mean ± SEM. ^*p < 0.05; ^**p < 0.01; ^***p < 0.001; Student’s-t test.

By quantification of the normalized read counts, we found that the transcript levels of Rxrg, Drd2, Adora2a, Rgs9 and adenylate cyclase 5 (Adcy5) were significantly reduced in mutant mCTX at P12 (Fig. 4D, Table 4). In contrast, transcripts for glutamate decarboxylase 1 (Gad1, also called GAD67), synaptotagmin II (Syt2), corticotropin releasing hormone binding protein (Crhbp) and solute carrier family 38, member 3 (Slc38a3, a sodium-coupled neutral amino acid transporter 3, also called SNAT3), were significantly increased in the mCTX of Cdkl5^−/y mice comparing to WT controls (Fig. 4C, Table 4).

Table 4

The transcript levels of differentially expressed genes in the mCTX of Cdkl5 mutants.

Gene code	Gene name		Normalized read counts
		Age	WT	Cdkl5−/y	Statistics
Cdkl5	cyclin-dependent kinase-like 5	P7	4478.8 ± 148.3	1647.9 ± 24.6	p = 0.00038***
		P12	5184.1 ± 26.1	1212.9 ± 46.1	p = 1.90E-07***
		P17	4857.1 ± 339.8	1308.7 ± 20.6	p = 0.00048***
		3 m	5174.4 ± 139.9	1132.6 ± 28.4	p = 9.25E-06***
Rxrg	retinoid X receptor gamma	P7	94.6 ± 66.0	31.0 ± 0.9	p = 0.42812
		P12	137.1 ± 14.8	44.3 ± 12.1	p = 0.00833**
		P17	76.1 ± 9.3	50.2 ± 8.7	p = 0.11134
		3 m	43.9 ± 8.3	81.1 ± 29.6	p = 0.29353
Drd2	dopamine d2 receptor	P7	52.7 ± 28.3	25.7 ± 3.4	p = 0.43524
		P12	116.5 ± 14.1	47.4 ± 13.8	p = 0.02486*
		P17	54.3 ± 9.2	40.0 ± 14.0	p = 0.43977
		3 m	50.2 ± 13.9	106.6 ± 44.0	p = 0.28801
Adora2a	Adenosine 2a receptor	P7	125.4 ± 52.2	81.8 ± 6.3	p = 0.48926
		P12	221.7 ± 26.8	122.1 ± 22.9	p = 0.04742*
		P17	94.5 ± 17.6	82.6 ± 16.2	p = 0.64478
		3 m	83.0 ± 17.5	155.7 ± 62.3	p = 0.32400
Rgs9	regulator of G-protein signaling 9	P7	436.0 ± 89.8	398.1 ± 8.1	p = 0.71668
		P12	313.0 ± 8.6	180.0 ± 33.9	p = 0.01911*
		P17	208.6 ± 16.3	195.4 ± 36.2	p = 0.75635
		3 m	228.0 ± 27.2	315.5 ± 69.0	p = 0.30315
Adcy5	adenylate cyclase 5	P7	2113.9 ± 211.2	1890.6 ± 122.0	p = 0.44910
		P12	2426.3 ± 34.8	2026.0 ± 48.1	p = 0.00253**
		P17	2193.4 ± 10.1	1894.4 ± 32.5	p = 0.00092***
		3 m	2246.2 ± 50.8	2409.7 ± 192.8	p = 0.45835
Gad1	glutamate decarboxylase 1	P7	1362.2 ± 49.1	1418.5 ± 6.2	p = 0.35981
		P12	2453.9 ± 51.2	2842.9 ± 53.6	p = 0.00631**
		P17	4067.2 ± 68.9	5095.4 ± 168.5	p = 0.00484**
		3 m	5797.5 ± 83.2	6957.7 ± 137.5	p = 0.00195**
Syt2	synaptotagmin II	P7	149.0 ± 13.3	126.4 ± 4.9	p = 0.22607
		P12	729.8 ± 69.8	980.2 ± 25.4	p = 0.02797*
		P17	975.6 ± 76.2	1222.6 ± 12.6	p = 0.03288*
		3 m	1702.5 ± 73.4	2018.8 ± 95.7	p = 0.05863
Crhbp	corticotropin releasing hormone binding protein	P7	97.6 ± 7.7	131.0 ± 6.4	p = 0.05116
		P12	150.1 ± 19.9	252.3 ± 13.4	p = 0.01300*
		P17	279.1 ± 3.6	364.1 ± 17.2	p = 0.00848**
		3 m	259.6 ± 14.4	410.7 ± 15.3	p = 0.00197**
Slc38a3	solute carrier family 38, member 3 (sodium-coupled neutral amino acid transporter 3, SNAT3)	P7	956.0 ± 23.4	987.2 ± 53.4	p = 0.64652
		P12	1012.0 ± 26.0	1241.0 ± 27.4	p = 0.00372**
		P17	979.7 ± 38.9	915.9 ± 60.2	p = 0.42404
		3 m	880.9 ± 13.0	920.5 ± 31.9	p = 0.31489
Kcnq3	potassium voltage-gated channel, subfamily Q, member 3	P7	5609.1 ± 301.3	5494.4 ± 181.7	p = 0.77745
		P12	6121.1 ± 208.0	6638.8 ± 132.9	p = 0.10398
		P17	5453.1 ± 165.0	6778.3 ± 196.1	p = 0.00665**
		3 m	5205.1 ± 81.4	5645.8 ± 168.6	p = 0.07817
Slc29a4	solute carrier family 29, member 4 (plasma membrane monoamine transporter, PMAT)	P7	1759.8 ± 108.3	1750.2 ± 56.9	p = 0.94556
		P12	537.3 ± 34.9	547.5 ± 5.9	p = 0.78687
		P17	268.3 ± 7.0	168.4 ± 17.4	p = 0.00598**
		3 m	(none)	(none)	—
Aqp4	aquaporin 4	P7	4378.2 ± 270.0	4077.5 ± 64.5	p = 0.37978
		P12	2979.0 ± 460.5	2600.2 ± 198.1	p = 0.49202
		P17	2268.5 ± 63.9	1792.8 ± 112.2	p = 0.02114*
		3 m	1806.1 ± 129.7	1859.7 ± 72.0	p = 0.73608
Dlx6	distal-less homeobox 6	P7	92.9 ± 12.5	86.2 ± 6.4	p = 0.67876
		P12	82.6 ± 1.6	83.5 ± 6.3	p = 0.89582
		P17	69.3 ± 2.0	83.4 ± 2.3	p = 0.01005*
		3 m	73.8 ± 3.5	85.9 ± 5.3	p = 0.12990
Gfap	glial fibrillary acidic protein	P7	2269.2 ± 478.8	1824.2 ± 89.0	p = 0.44969
		P12	2252.1 ± 331.6	1737.3 ± 93.2	p = 0.20929
		P17	1400.0 ± 45.6	1152.7 ± 69.6	p = 0.04096*
		3 m	1237.5 ± 182.5	1229.7 ± 100.9	p = 0.97196
Pvalb	parvalbumin	P7	(none)	(none)	—
		P12	117.1 ± 30.5	180.4 ± 9.4	p = 0.11819
		P17	673.3 ± 47.7	693.3 ± 22.2	p = 0.72258
		3 m	886.6 ± 33.3	1106.0 ± 60.6	p = 0.03370*

Notes: mCTX, motor cortex; none, not detected.

^*p < 0.05.

^**p < 0.01.

^***p < 0.001.

Student’s t-test. The sample numbers for P7, P12, P17 and 3 m are 4 (WT, 2 for KO), 3, 3, 3, respectively.

Loss of CDKL5 increases gene expression for GABAergic transmission in the mCTX at P17

By examining the mCTX transcriptome at P17 followed by DEG analysis with the same thresholding criteria for p-value and log2 fold-change (0.001 and 1.2) as that for P12 mCTX, we identified 21 DEGs in Cdkl5^−/y mice (n = 3). Among these 21 DEGs, several monoaminergic and astrocytic genes were down-regulated in Cdkl5 mutants, while 18 DEGs, including a set of GABAergic marker genes, were upregulated (Fig. 4E). Normalized read counts shows that the transcript levels of potassium voltage-gated channel, subfamily Q, member 3 (Kcnq3,  p < 0.01), Gad1 (p < 0.01), Syt2 (p < 0.05) and Crhbp (p < 0.05) were significantly increased in mutants (Fig. 4F, Table 4). By contrast, the transcripts of Adcy5 (p < 0.001), solute carrier family 29, member 4 (Slc29a4, a plasma membrane monoamine transporter; p < 0.01) and aquaporin 4 (Aqp4, a water channel protein on astrocytes; p < 0.05) were significantly reduced in mutants (Fig. 4F, Table 4). In addition, we found that the transcripts of distal-less homeobox 6 (Dlx6), a developmental regulator for GABAergic cortical interneurons [47], was significant upregulated (p < 0.05), whereas the expression of glial fibrillary acidic protein (Gfap) was reduced (p < 0.05) in mutant cortex at P17. Notably, among these identified DEGs, the transcript levels of Kcnq3, Dlx6, Slc29a4, Aqp4 and Gfap in the cortex of Cdkl5^−/y mice were selectively altered at P17, but not at the other ages.

Validation of the DEGs identified via transcriptome analysis

To confirm the alterations of the DEGs in Cdkl5 mutants, we next carried out validation studies using distinct approaches from transcriptome profiling, such as RNAscope in situ hybridization, real-time quantitative RT-PCR and immunohistochemistry. We reasoned that the transcripts of Glra2 could be colocalized with Cdkl5 if the latter regulates Glra2 expression directly. RNAscope analysis showed that Cdkl5 transcripts (red spots) were greatly co-localized with Glra2 (green spots) in the dentate gyrus (DG) at P7 (C-w/G, ~ 65% of total Cdkl5+ spots) (Fig. 5A–A’). Loss of CDKL5 reduced the proportion of C-w/G beyond 30% (Fig. 5C), suggesting that at least a part of Glra2 is proximate to and regulated by CDKL5 in the DG at P7. Although the Cdkl5 transcripts were prominently reduced in mutant brains, the proportion of C-w/G was not reduced in CA1 and mCTX areas (Fig. 5B, B’, C), suggesting that Glra2 may be region-specifically regulated by CDKL5 in the DG of developing hippocampus.

Figure 5

Validation of reduced expression of Glra2 in Cdkl5 null mice. RNAscope in situ hybridization showing the expression of Cdkl5 (red spots) and Glra2 (green spots) transcripts in the dentate gyrus (DG; A, A’) and the motor cortex (mCTX; B, B′) of wild-type (WT; A, B) and Cdkl5^−/y mutants (A’, B′) at the age of P7. (C) The proportions of colocalized Glra2 and Cdkl5 mRNA are quantified in the DG, CA1 and mCTX. The scale bars represent 50 μm in A and B for A-B′, and 20 μm in A’ and B′ for all insets. All data are presented as mean ± SEM. ^**p < 0.01; Student’s-t test.

Besides our findings that loss of CDKL5 reduced CCK transcripts in P7 dHC (Fig. 3F), reduced expression of CCK have been implicated in pathogenesis of epilepsy [48,49]. We thus examined whether CDKL5 deficiency affects the protein expression of CCK in the dHC at P7. The results showed that CCK-expressing neuropils were hardly detected in the rostral CA1 region of hippocampus in Cdkl5 null mutants at P7 (Fig. 6A–B’). The expression of CCK in mutants was restored to a comparable level of controls at P17 (Fig. 6C, C’). In contrast to CCK, parvalbumin (PV, Pvalb)-expressing interneurons were not appeared until the second postnatal week [45,50]. Through immunostaining, we found that the number of PV-expressing neurons were increased in adult Cdkl5^−/y mice, selectively in the motor area of the neocortex (1.27 ± 0.07 folds of WT, p < 0.05; Fig. 6D, D’) rather in the hippocampal regions (Fig. 6E). Moreover, RNA-seq results showed that loss of CDKL5 does not increase the levels of Pvalb transcripts in the mCTX until the age of three months (p < 0.05; Fig. 6F, Table 4).

Figure 6

CDKL5-mediated developmental regulation of gene expression in the dHC and mCTX. Immunostaining shows that the cholecyctokinin (CCK) expression in the CA1 hippocampal area in WT and Cdkl5^−/y mice at P7 (A-B′) and P17 (C, C′). High magnification micrographs of the CA1 region in A and A’ are shown in B and B′. (D-E) Immunoreactive cells for parvalbumin (PV) proteins in the primary motor cortex (M1) of adult mice. The number of PV-positive cells is increased in the M1, but not in hippocampal areas of CA1 and dentate gyrus (DG), in Cdkl5^−/y mice. (F) Quantification of transcriptomic analysis shows that the normalized read counts of Pvalb transcripts are increased in the mCTX selectively at the age of 3 m. (G) Real-time RT-qPCR validates the alterations of the selected DEGs, noted above the chart, at the indicated ages of P7, P12 and P17 in the mCTX of WT and Cdkl5^−/y mutants. The scale bars represent 200 μm in A’ (for A, A’) and C′ (for C, C′), 50 μm in B′ (for B, B′) and 100 μm in D′ (for D, D′). All data are presented as mean ± SEM. ^*p < 0.05, ^***p < 0.001, Student’s-t test.

We next focused on several DEGs in the mCTX (e.g. Rgs9, Kcnq3, Gad1, and Syt2, see Fig. 4) and validated their transcript levels at different ages. The results of real-time RT-qPCR showed that CDKL5 ablation downregulated the expression of Rgs9 merely at P12 (69.34 ± 8.69% of WT, p < 0.05). In contrast, the expression of Kcnq3 (123.32 ± 3.90% of WT, p < 0.001), Gad1 (128.75 ± 3.58% of WT, p < 0.001) and Syt2 (118.73 ± 2.00% of WT, p < 0.001) was selectively increased at P17 (Fig. 6G). These results are in line with our RNA-seq findings (Fig. 4D, F), suggesting that CDKL5 mediates age-dependent expression changes for those genes related to dopaminergic and GABAergic transmission in the neonatal cortex.

Reduced functional KCC2 in Cdkl5^−/y pups at P12

During the early postnatal stages, neurons contain high intracellular level of chloride (Cl⁻) so that activation of GABA_A receptors triggers Cl⁻ efflux leading to membrane depolarization and firing of neurons [51]. Upon neuronal maturation, a transporter called K⁺/Cl⁻ co-transporter 2 (KCC2) that extrudes intracellular Cl⁻, starts to express in the first week after birth. It renders the GABA_A-mediated Cl⁻ currents to flow inwardly that hyperpolarize the membrane and reduce excitability of neurons [52]. Notably, phosphorylation of KCC2 at the residue of serine 940 (pKCC2-S940), stabilizing KCC2 on the neuronal membrane, is required for transporter activity of KCC2 [53,54]. Dysregulation of KCC2 has been strongly implicated in pathogenesis of neonatal seizures and infantile epilepsy [30, 55–59]. Given that the transcript levels of Kcc2 were not altered in the developing dHC and mCTX of Cdkl5 mutants (Table 5), we thus measured the protein phosphorylation of KCC2 in the mCTX at P12, the age ictal EEG activities appear in Cdkl5^−/y mice. Immunoblotting results showed that loss of CDKL5 significantly reduced the level of pKCC2-S940 as well as the proportion of pKCC2-S940 in total KCC2 (0.58 ± 0.056 of WT, p < 0.01; Fig. 7A, B). In contrast, phosphorylated KCC2 at threonine 1007 (pKCC2-T1007), which has been reported to inhibit Cl⁻ transport of KCC2 [60], was not altered in Cdkl5^−/y mice (0.913 ± 0.071 of WT, p > 0.05; Fig. 7A, B). The reduced proportion of pKCC2-S940 in total KCC2 was recapitulated in P12 dHC of Cdkl5^−/y mice (0.785 ± 0.022 of WT, p < 0.05). No change in the levels of total KCC2 protein was detected (1.005 ± 0.070 of WT, p > 0.05; Fig. 7A, C), consistent to our observation from the RNA-seq study. Moreover, the protein level of Na⁺/K⁺/Cl⁻ co-transporter 1 (NKCC1), a developmentally down-regulated cation-chloride co-transporter [61], was not altered in Cdkl5 null mutants (0.953 ± 0.041 of WT, p > 0.05; Fig. 7A, C).

Table 5

Age- and brain region-dependent alterations of gene expression in Cdkl5 null mice.

Brain Region	dHC				mCTX
Gene\Age (KO vs. WT)	P7	P12	P17	P90 (3 m)	P7†	P12	P17	P90 (3 m)
Cdkl5	Down***	Down***	Down***	—	Down***	Down***	Down***	Down***
St18	Down***	ND	ND	—	ND	ND	ND	ND
Glra2	Down**	Down*	Down***	—	ND	ND	ND	ND
Nrp2	Down*	ND	ND	—	ND	ND	ND	ND
Cntn2	Down*	ND	Up*	—	ND	ND	Up*	ND
Cck	Down*	ND	ND	—	ND	ND	Down*	ND
Sst	Up*	ND	ND	—	ND	ND	ND	Up*
Npy	ND	ND	ND	—	ND	ND	ND	Up*
Nos1	ND	ND	ND	—	ND	ND	ND	ND
Calb1	ND	ND	Down*	—	ND	ND	Down**	ND
Calb2	ND	ND	ND	—	ND	ND	ND	ND
Slc8a3	Up***	ND	Up*	—	ND	ND	ND	ND
Kcnn1	Up**	ND	Up*	—	ND	Up*	ND	ND
Pdzd2	Up***	ND	ND	—	ND	ND	ND	ND
Lrrtm3	Up**	ND	ND	—	ND	ND	ND	ND
Gabrg1	Up**	ND	ND	—	ND	ND	ND	ND
Rxrg	ND	ND	ND	—	ND	Down**	ND	ND
Drd2	ND	ND	ND	—	ND	Down*	ND	ND
Adora2a	ND	ND	ND	—	ND	Down*	ND	ND
Adcy5	ND	ND	ND	—	ND	Down**	Down***	ND
Rgs9	ND	ND	ND	—	ND	Down*	ND	ND
Gad1	ND	Up*	Up**	—	ND	Up**	Up**	UP**
Syt2	ND	ND	Up**	—	ND	Up*	Up*	ND
Kcnq3	ND	Up*	Up*	—	ND	ND	Up**	ND
Pvalb	(none)	ND	UP*	—	(none)	ND	ND	UP*
Dlx6	ND	ND	ND	—	ND	ND	Up*	ND
Crhbp	Up**	ND	ND	—	ND	Up*	Up**	UP**
Slc38a3	ND	ND	ND	—	ND	Up**	ND	ND
Slc29a4	ND	ND	ND	—	ND	ND	Down**	(none)
Aqp4	ND	ND	ND	—	ND	ND	Down*	ND
Gfap	ND	ND	Down*	—	ND	ND	Down*	ND
Slc12a5(Kcc2)	ND	ND	ND	—	ND	ND	ND	ND

Notes: dHC, dorsal hippocampus; mCTX, motor cortex; KO, Cdkl5^−/y; WT, wildtype; “Down” and “Up” indicate decreased and increased expression, respectively, compared to controls. ND, no difference; none, no expression; bolded, validated results; —, not examined.

Student’s t-test.

^*p < 0.05.

^**p < 0.01.

^***p < 0.001.

^†n = 2 for KO.

Figure 7

Loss of CDKL5 impairs phosphorylation of KCC2 at S940 in P12 mouse cortex. (A) Representative immunoblots show protein expression levels in the motor cortex (mCTX) from two pairs of mice at P12. (B-D) Quantification of the total protein levels shows a significant reduction in phosphorylation level of KCC2 at serine 940 (pKCC2-S940) and PSD-95 in Cdkl5^−/y (KO) mice compared to their wild-type (WT) littermate controls. Data are derived from three litters (6 pairs) of animals and presented as mean ± SEM. ^**p < 0.01, Student’s t-test.

Given that CDKL5 ablation altered a set of genes related to excitatory and GABAergic transmission (Figs. 3, 4, Table 4), we next examined if loss of CDKL5 affects protein expression of synaptic molecules in the mCTX at P12. We found that the protein levels of GABA_A receptors (0.991 ± 0.025 of WT, p > 0.05; Fig. 7A, D), GABA synthesizing enzyme glutamate decarboxylase 2 (GAD2; 1.063 ± 0.011 of WT, p > 0.05) and vesicular GABA transporter (vGAT; 1.017 ± 0.026 of WT, p > 0.05) were comparable between Cdkl5^−/y and WT mice. However, the protein levels of PSD-95, a CDKL5-interacting protein in the postsynaptic density [62], appeared to be reduced in pups lacking CDKL5 (0.812 ± 0.049 of WT, p < 0.01; Fig. 7A, D), consistent with previous studies [62–66]. Together, these data suggest that CDKL5 is required to preserve the levels of functional KCC2 and synaptic protein PSD-95 those may critically modulate neuronal excitability at P12.

Discussion

In the present study, we report the discovery of spontaneous epileptic activities in neonatal mice with CDKL5 deficiency that was coupled with developmental time-dependent and brain region-specific alterations of transcriptome. We also uncovered multiple molecular players involved in synaptic transmission, interneuron development and excitation-inhibition transition critical for epileptogenesis in CDD mice. The epileptic EEG discharges were identified in Cdkl5 null mutants at the age of P12, including the increase of total spike events, the number of bursts, and duration of repeated seizure-like activities, in comparison to their wild-type littermate controls (Figs. 1, 2). Intriguingly, the ictal EEG activities disappeared in the third week after birth. Together with the findings of elevated firing rate in Cdkl5^−/y hippocampal neurons at P7, our results support a central theme that CDKL5 deficiency deregulates neuronal excitability age-dependently in the postnatal brains, from week one of developing hyperexcitability, week two of sustained hyperexcitability, to week three of normalized neuronal excitability.

Postnatal week one: Hyperexcitability

To identify the molecular players involved in CDKL5-mediated regulation of neuronal excitability, we analyzed the differentially expressed genes between Cdkl5 null mutants and control mice through unbiased transcriptome profiling across specific time points when the transient spontaneous epileptic activities are present. Despite the expression of NMDA and AMPA subtypes of glutamate receptors was unchanged (Table 3), we found that some upregulated genes encode proteins related to transmembrane cation transport (i.e. Pdzd2) and proteins associated with excitatory synaptogenesis (i.e. Lrrtm3). These changes occur selectively at P7, but not P12 and P17, in the dHC (Table 1). Notably, CDKL5 ablation downregulated Glra2 that has been implicated in forebrain inhibitory neurotransmission and interneuron migration [39,40]. Moreover, the genes encoding regulators for tangential migration (i.e. Nrp2, Cntn2) and a neuropeptide expressed in a subtype of interneurons (i.e. Cck) were also significantly downregulated in the dHC at P7. These lines of evidence suggest that neuronal hyperexcitability observed in Cdkl5 mutant pups in early postnatal age (i.e. P7) may be coupled with elevated excitatory signaling (increased transmembrane cation transport and excessive formation of excitatory synapses) and reduced inhibitory signaling (compromised glycine transmission and interneuron migration) specifically in the dHC, together, leading to neonatal epilepsy.

• (1) Increase of sodium transport and excitatory synaptogenesis: PDZD2 is a multi PDZ-domain protein, largely expressed in the dentate gyrus (DG), interacting with voltage-gated sodium channels that are responsible for action potential initiation and propagation in excitable cells. Introduction of antisense Pdzd2 to cultured neurons causes a great loss of sodium current [67,68]. Our findings that loss of CDKL5 increases the expression of Pdzd2 in neonatal dHC (Fig. 3D) suggest that CDKL5 may be crucial to suppress neuronal excitability by modulating intracellular amount of cation (e.g. Na⁺) in hippocampal neurons. LRRTM3 is a cell-adhesion molecule that functions in promoting the formation of glutamatergic synapses. Loss-of-function of LRRTM3 impairs activity-dependent postsynaptic surface expression of glutamate AMPA receptors in the DG granule neurons and reduces excitatory innervation of mossy fibers from DG granule cells onto the hippocampal CA3 neurons [69,70]. Our findings that CDKL5 ablation increases Lrrtm3 transcripts in the dHC at P7 (Fig. 3D) suggest that excessive glutamatergic synaptogenesis in the DG may partly underlie the neonatal epilepsy in CDD.

• (2) Decrease of inhibitory transmission: Glycine receptors (GlyR) are ligand-gated chloride channels that mediate inhibitory neurotransmission in the spinal cord and developing forebrain. The glycine-activated currents plus GABA_A receptor-mediated tonic inhibition may regulate the excitability of individual neurons cooperatively [40]. In the developing mouse brain, the alpha 2 subunit (GLRA2) is the most abundant GlyR that is highly expressed in the embryos, but gradually downregulated after birth [71,72]. The early and transient expression of Glra2 transcripts suggests that GLRA2-mediated glycinergic transmission may play a critical role in circuit and synapse formation in specific brain areas [73]. Due to high intracellular chloride level in immature neurons [51], activation of GlyR triggers depolarization of neurons in the hippocampus at birth. The GlyR-mediated depolarization is converted into hyperpolarization at around P5 [74], while activation of GABA_AR switches from excitation to inhibition at around P10 [75]. Our findings that CDKL5 ablation reduces Glra2 (Fig. 3E, Fig. 5) but increases Gabrg1 (Fig. 3D) transcripts in P7 dHC suggest a plausible replacement of glycinergic inhibitory inputs by GABA-containing excitatory terminals in the hippocampus. The increased GABAergic transmission in Cdkl5 null mice may enhance anion (Cl⁻) efflux, thus leading to hyperexcitability at P7. Notably, although Cdkl5 deletion also downregulated hippocampal Glra2 at P12 and P17 (Table 1), the responsiveness of the hippocampal neurons to glycine is decreased after P7 and completely disappeared after P14 [74], suggesting a temporally specific role of glycine-mediated inhibitory transmission in developing hippocampus at the first postnatal week.

• (3) Decrease of interneuron migration: In addition to inhibitory transmission, activation of GLRA2 promotes migration of interneurons from the medial ganglionic eminence (MGE) to the cortical plate. Knockdown of Glra2 reduces the number of neurons migrating from ventral to dorsal telencephalon [39]. Although interneuron migration occurs during the mid-gestational stage [43], a time window not included in the present study, migration defects have been observed in cortical neurons derived from CDD patients [36]. It is noteworthy that Cdkl5 deletion reduces the expression of Nrp2 and Cntn2, both encoding regulators for tangential migration [41,43], in P7 dHC (Fig. 3E). Neuropilins (Nrp) are transmembrane receptors expressed in migrating GABAergic neurons, mediating the repulsive actions of class 3 semaphorins on axons [76,77]. Deletion of Nrp2 reduces the MGE-derived migrating neurons toward the neocortex and hippocampus, disrupts hippocampal synaptic function and has been implicated in pathogenesis of autism and epilepsy [44,78,79]. Cntn2 (also called TAG-1) encodes a cell adhesion molecule expressed in the corticofugal fibers, forming the migratory routes to guide the MGE neuronal migration tangentially to the neocortex [41,80]. Deletion of Cntn2 causes guidance defects and miswiring of neurons, increased seizure susceptibility in convulsant-treated mice, and clinically underlies familial cortical myoclonic tremor with epilepsy [42,81,82]. Our study suggests that CDKL5 might be critical to tranquilize neuronal excitability partly by regulating migration of hippocampal interneurons in early postnatal age.

• (4) Decrease of CCK expression: Reduced expression of Nrp2 and Cntn2 in Cdkl5 null brains implied that the development of hippocampal interneurons might be disturbed. Indeed, the CCK-expressing neuropils was drastically reduced in the hippocampal CA1 region at P7 (Fig. 6A–B’) with concomitant downregulation of Cck transcripts in the dHC (Fig. 3F). CCK is the most abundant neuropeptide in the brain, highly expressed in the second prosomere of developing mouse brain after birth [83]. During the first postnatal week, the axons of CCK-expressing interneurons innervate pyramidal neurons to assist the latter’s maturation in the developing hippocampus [84]. Notably, CCK neurons have been shown to modulate behaviors through direct inhibition of neuronal activities [85,86]. Selective reduction of CCK innervations to the CA1 hippocampal region has been found in various mouse models of temporal lobe epilepsy [48,49,87]. Moreover, administration of CCK octapeptides has been demonstrated to attenuate the severity of seizures [88–91]. Our findings that reduced expression of CCK, at both transcript and protein levels, in CDKL5 deficient dHC at P7, suggest that CDKL5 may indispensably regulate CCK availability in the developing hippocampus that may in turn govern neuronal excitability and epileptogenesis during the early postnatal period.

Postnatal week two: Persistent hyperexcitability

• (1)

  Impaired GABA-mediated excitation-to-inhibition transition: As mentioned above, immature neurons are depolarized by GABA because of their high intracellular concentration of chloride [51]. The postnatal induction of KCC2 that extrudes chloride anions is essential for the switch of GABA_A-mediated excitation to inhibition during the second week after birth [52]. Loss of function of KCC2 triggers neuronal hyperexcitability that has been implicated in infantile epilepsy [30, 55–59]. Importantly, we found that the phosphorylation of KCC2 at the residue of serine 940 (pKCC2-S940), which is required for the transporter activity of KCC2 [53,54], was significantly reduced in Cdkl5 null pups at the time window of seizure occurrence (Fig. 7). The reduction of pKCC2-S940/KCC2 in both mCTX and dHC of Cdkl5 null pups suggest that loss of CDKL5 may cause a decline of functional KCC2 at P12 that may lead to diminished chloride extrusion and subsequent GABA_AR-mediated neuronal hyperexcitability. It is noteworthy that upregulated cortical expression of Gad1, Syt2 and Slc38a3 in Cdkl5^−/y mice at P12 (Fig. 4C) may enhance GABA production and release (see below) that in turn exacerbate the GABA-mediated excitation of neurons caused by reduced functional KCC2 (Fig. 7). Given that KCC2 has not been identified as a direct substrate of CDKL5 [92,93], the CDKL5-regulated phosphorylation of KCC2 might be indirectly mediated by the action of protein kinase C, whose kinase activity is known to be affected by loss of CDKL5 [15,53].

• (2)

  Perturbated glutamine homeostasis:  Slc38a3 (also called SNAT3) encodes a Na⁺ and H⁺-coupled glutamine transporter, which is mainly expressed on astrocytes and participates in the glutamate-GABA-glutamine cycle in the brain. SLC38A3 transports L-glutamine from astroglia to the intercellular space for replenishment of both glutamate and GABA production in neurons [94,95]. Recent studies show that individuals with genetic variants of SLC38A3 exhibit early-onset epileptic encephalopathy with features similar to CDD, such as global developmental delay, intellectual disability, hypotonia, absence of speech and visual impairments [96]. Our findings that loss of CDKL5 increases the cortical expression of Slc38a3 selectively at P12 (Fig. 4C, Table 4) suggest that CDKL5 is likely to stabilize neuronal excitability by modulating glutamine/glutamate homeostasis and GABA availability in the neonatal cortex. Considering that CDKL5 ablation downregulates a set of astroglial genes at P17 (i.e. Aqp4, Gfap), the roles of astroglia in neonatal epilepsy of CDD is likely worthy of further investigation.

• (3)

  Decreased dopamine DRD2 signaling: The transcriptome profiling of the mCTX showed that loss of CDKL5 significantly downregulated several dopamine (DA) signaling molecules (Drd2, Rgs9, Adora2a) at P12 (Fig. 4C, D), followed by restoration of their expression to normal levels at P17 (Fig. 4E, Table 4). DA is an endogenous neuromodulator essential for movement control and cognitive function in the mammalian brain. The cerebral cortex, especially the prefrontal and motor areas, receives a substantial DA innervation from the midbrain dopaminergic neurons [97,98]. Activation of DA receptors, D1 or D2 types, modulates G-protein-coupled cAMP signaling resulting in excitatory or inhibitory intracellular responses, respectively [99]. Previous studies have shown that DA increases the spontaneous firing of cortical interneurons but reduces pyramidal neuron excitability in the rat frontal cortex [100]. Application of DRD2 agonists attenuates the excitatory effects of glutamate, while the DRD2 antagonists reverse DA-induced decrease of firing rate in cortical neurons [101,102]. Our findings that loss of CDKL5 reduced Drd2 transcripts in the neonatal mCTX suggest that deficient DRD2 may in part underlie the cortical hyperexcitability in CDD mouse pups. It is notable that RGS9, a regulator of G-protein signaling (RGS) colocalized with DRD2 [103], is also downregulated in Cdkl5 null cortex at P12 (Figs. 4C–D, 6G). RGS9 is known to modulate D2 receptor-mediated calcium channel inhibition in the striatal cholinergic interneurons [104]. Genetic deletion of Rgs9 increases locomotor activities in response to psychostimulants and mimics the behaviors upon DRD2 loss [105,106]. Our findings that the myoclonic ictal movements showed in Cdkl5 null mice at P12 (Supplementary Material, Fig. S3) may possibly reflect the behavioral consequence of the decreased DRD2 signaling in the mCTX.

Postnatal week three: Normalized neuronal excitability

In CDD patients, the onset of seizures occurs soon after birth at the median age of 6 weeks [13,14]. Nearly half of patients (71/163) are reported to experience more than two months of seizure-free period, the so-called “honeymoon period”, typically occurring in the first two years of life and having a median duration of six months (range from 2.5 months to six years) [14]. Some of CDD patients outgrow their epilepsy in childhood [8]. In line with those clinical observations, our findings that the ictal discharges of EEG in Cdkl5 null mice at P12 diminished in the third postnatal week (Figs. 1, 2) may phenocopy the “honeymoon period” in human cases. Based on our transcriptome profiling, here we suggest two putative mechanisms for the normalization of neuronal excitability in CDD mice during the third postnatal week.

• (1) Increased GABAergic inhibitory transmission: In the mCTX at P17, we found that loss of CDKL5 significantly upregulates a set of GABAergic molecules, including Gad1, Syt2, and Crhbp (Figs. 4E–F, 6G). Gad1 (also called Gad67) encodes glutamate decarboxylase 1, an enzyme responsible for GABA synthesis, predominantly expressed in GABAergic neurons in the brain [76]. Syt2 gene encodes synaptotagmine II, a calcium sensor protein located on the presynaptic vesicle membrane to trigger neurotransmitter release [107]. SYT2 is highly expressed in GABAergic synapses, colocalized with vGAT and PV, both in the cortex and hippocampus [108–110]. Crhbp encodes a binding protein for the stress hormone, corticotropin-releasing hormone (CRH), highly expressed in the GABAergic cortical interneurons [111] to attenuate CRH-mediated activation of pyramidal neurons and reduces anxiety-related behavior [112]. Our findings that CDKL5 deletion increased cortical expression of Gad1, Syt2, and Crhbp at P17, the time point with diminished epileptic activity, suggest that increased GABAergic transmission, including GABA synthesis/release and interneuron activation, may play a role in initiation of seizure-free “honeymoon period” in CDD.

• (2) Increased Kcnq3-mediated M-currents: KCNQ3 (also called K_v7.3) is a voltage-gated potassium channel localized at the node of Ranvier or initial segment of axons [113]. Activation of KCNQ3 generates M-currents that serve as a brake of repetitive firing in neurons [114,115]. Loss-of-function of KCNQ3 impairs M-currents and has been implicated in benign familial neonatal epilepsy [116–118]. Interestingly, KCNQ3 can be directly activated by GABA [119], whose levels might be elevated in Cdkl5 null mice as mentioned above. Therefore, increased Kcnq3/Gad1/Syt2 expression in Cdkl5^−/y mCTX at P17 (Fig. 4F, Fig. 6G) may synergistically contribute to the normalization of neuronal excitability in the third postnatal week. The increased expression of these genes might be triggered by compensatory or homeostatic responses to relieve the early postnatal hyperexcitability due to CDKL5 loss, though the exact mechanisms remain to be determined.

Conclusion

In the present study, we demonstrate neuronal hyperexcitability in neonatal Cdkl5 null mice and identify an array of CDKL5-mediated developmentally regulated genes relevant to early-onset epilepsy. In the first postnatal week, a timepoint equivalent to 6–7 week of age in human [21], the median age of seizure onset in CDD children [14], CDKL5-deficiency triggers neuronal hyperexcitability accompanied by enhanced gene expression for excitatory transmission but reduced gene expression for inhibitory transmission in the dorsal hippocampus. The increased excitability sustains to the second postnatal week, which is linked to reduced phosphorylation of KCC2 at S940, increased expression of GABAergic genes, and reduced expression of the dopaminergic DRD2 signaling in the cortex. In addition, loss of CDKL5 enhances cortical expression of genes related to GABAergic transmission and potassium outward transport, which may contribute to the normalization of neuronal excitability in the third postnatal week in mice, mimicking the seizure-free “honeymoon period” in CDD. However, the causal relationships between changed expression of those synaptic genes and alterations in neuronal excitability remain to be determined.

The timeline of neuronal hyperexcitability in neonatal mice lacking CDKL5 corresponds to the timeframe of seizure onset in CDD children, suggesting that the developmentally regulated DEGs identified in Cdkl5^−/y mice during the period of hyperexcitability may represent the pathogenic molecular substrates for neonatal epilepsy in CDD. The unresponsiveness of CDD patients to antiepileptic drugs might result from the complicated age- and brain region-dependency of CDKL5 function in regulating gene expression, protein phosphorylation and neuronal excitability. Therefore, activation or inhibition of single type of receptors, channels or transporters may not be sufficient to prevent or reverse the CDD-related neonatal epilepsy. Given that the reduced expression of Glra2,  Cck, and functional KCC2 are concomitant with neuronal hyperexcitability in neonatal Cdkl5 null mice, a cocktail therapy targeting to these candidate molecules at an age of seizure onset may ameliorate the intractable neonatal epilepsy in CDD.

Materials and methods

Animals

Cdkl5 null pups were generated by crossing heterozygous Cdkl5 females [Cdkl5^+/−; B6.129(FVB)-Cdkl5^tm1.1Joez/J] [15] to C57BL/6 J male mice (National Laboratory Animal Center, Taiwan). Genotyping was performed as previously described [25]. Male hemizygous pups (Cdkl5^−/y) and their littermate controls (Cdkl5^+/y) were used as the experimental cohorts to avoid mosaic expression of Cdkl5 in heterozygous females due to random X-inactivation. All mice were bred and housed in individually ventilated cages (IVC, Alternative Design, USA) at 22 ± 2°C and 60 ± 10% humidity under a 12-h light–dark cycle (light on 08:00 to 20:00). Irradiated diet (#5058, LabDiet, USA) and sterile distilled water were supplied ad libitum. All experiments were approved by the Institutional Animal Care and Use Committee at National Cheng-Chi University and the principles of laboratory animal care in Guide for the Care and Use of Laboratory Animals (Council of Agriculture, Taiwan, 2018) were followed.

Implantation surgery of transmitters

Male pups were tail-marked and genotyped at P9. Only littermate pairs (Cdkl5^+/y, Cdkl5^−/y) with body weight of > 6 g were subjected to the implantation surgery at P10 under anesthesia with 1% vaporized isoflurane (Harvard Apparatus, USA). A transmitter (4 × 6 × 8 mm, 0.5 g, gain = 4000 x; Fig. 1A) was activated prior to use according to the manufacturer’s instructions (Epoch, Epitel, USA) [120]. The working electrodes (2 channels, 2 mm in between) were cut to ~ 2 mm of length and implanted behind the middle level of the cortex targeted to the rostral and dorsal hippocampus (estimated coordinates: AP −2.0 mm, ML ± 1.0 mm to Bregma). A reference electrode (~3.5 mm) was placed to the left hemisphere targeting to the cerebellum (Fig. 1A, B). The transmitter was then secured on the skull with tissue glue (Vetbond, 3 M). All the implanted animals were kept on a pre-warmed plate for recovery, tested for EEG recording to make sure the transmitter is functioning, and then returned to the home cages. Body weight and general health of the implanted animals were monitored on a daily basis.

Pup EEG recording and data analysis

For longitudinal study, EEG were recorded from implanted mice on P11, P12, P14, P17, P21, and P24. We avoided daily recording to minimize the separation stress for both implanted pups and dams. Each implanted mouse was placed on top of a receiver (18 × 18 cm, Epoch, Epital) and recorded for 100 min at the same period of the day (1–6 p.m.). EEG activities were captured with the sampling rate of 1000 Hz (PowerLab, ADInstruments) and processed further by the LabChart 7 software. All raw data were band-pass filtered (3–50 Hz) and then exported to Clampex (Version 10.7, Axon Instruments) to collect the traces of the last 90 min (10 ~ 100 min) of recording. For event detection, the threshold was set as 0.9 V (equivalent to 0.225 mV of input intensity) to measure the number of spike discharges (i.e. events). Based on a criterion that an ictal EEG discharge needs to persist for > 10 s to be considered as a seizure [121], we counted an ictal burst as a continuous train of discharges that contains more than 20 spikes with the inter-spike interval of < 1 s. Four pairs (from four litters) of data were included for statistical analysis with a linear mixed model (Prism 9, GraphPad).

Assessment for seizure-like behaviors

During EEG recordings, the implanted pups were videotaped simultaneously using a webcam to assess potential seizure-like behaviors (Logitech). The behaviors during the prolonged bursts were evaluated in pups at P12 as previously described [27]. The behavioral seizures were scored from 0 to 3 based on the following criteria: 0, sleeping/staying quiet without movement; 1, taking rest with occasional orofacial movement/head bobbing forelimb twitch/tail flick; 2, freezing suddenly during movement; and 3, spasm-like jerky movements of the head, body or limbs. The summed scores for spasm-like behaviors during the prolonged bursts for Cdkl5^−/y pups (see Movie 1) were compared to their WT littermates (see Movie 2).

Wire suspension test

To test grip strength and endurance, the forepaws of male pups were placed on a horizontal wire (3 mm in diameter) suspended at a height of 30 cm above the tabletop (Schneider and Przewlocki, 2005). Soft bedding materials were placed under the wire to protect fallen pups from injury. Each pup was tested for five trials every other day from P10 to P18. The latency to fall was recorded with a cutoff time of 3 min for each trial. The median of daily scores for Cdkl5^−/y pup (n = 11) and their WT littermates (n = 13) were compared by two-way repeated measures ANOVA (Prism 9, GraphPad). Independent cohorts of naïve mice (without transmitter implantation) were subjected to this test.

Nissl staining

Implanted mouse brains were fixed at P12 and sectioned for 30 μm. Dried sections were deparaffinized in xylene and 100% ethanol 3 times for 5 min, followed by rehydration in 95% and 70% ethanol for 5 min each and then in water for 2 min. The sections were socked in 0.1% cresyl violet (C5042, Sigma) solution at 37°C for 10 min, and then removed the dye by a quick rinse in water, followed by a 10–15 min incubation in 95% ethanol containing a few drops of 10% acetic acid solution. The sections were then dehydrated in 95% ethanol once and 100% ethanol three times for 5 min each prior to being coverslipped by mounting media (VectaMount®, Vector Laboratories).

Whole-cell patch-clamp

Neonatal brains from Cdkl5^−/y or WT littermate mice at P7 were sliced coronally at the thickness of 350 μm (DTK-1000 MicroSlicer, Dosaka EM, Japan) in ice-cold oxygenized artificial cerebrospinal fluid (ACSF) containing (in mM) 124 NaCl, 3 KCl, 2 CaCl₂, 1.3 MgSO₄, 25 NaHCO₃, 1.2 NaH₂PO₄, 10 D-glucose and saturated with 95%O₂–5% CO₂ (pH 7.3, 300–310 mOsm). Slices were left in a humid oxygenized chamber at ~ 35°C for at least 30 min for recovery and then kept at room temperature prior to recording [122,123]. For whole-cell patch-clamp recording, CA1 pyramidal neurons in hippocampal slices were identified using infrared/differential interference contrast microscopy (Olympus, Japan). The patch-pipettes were made from borosilicate glass capillaries (outside diameter 1.5 mm, inside diameter 0.86 mm; type GC150F-15, Clark Electromedical Instruments) and had a resistance of 4–6 MΩ when filled with the pipette solution containing (in mM): 131 potassium gluconate, 20 KCl, 8 NaCl, 10 HEPES, 2 EGTA, 2 ATP, and 0.3 GTP (pH 7.2 to 7.3, 300–305 mOsm). Action potentials were recorded using a Multiclamp 700B amplifier (Molecular Devices) in a current-clamp mode by stepping the injected current by 10 pA from 0 to 80 pA. Signals were low pass-filtered at 2 kHz and digitized at 10 kHz with a Micro 1401 interface running Spike2 software (Cambridge Electronic Design, UK).

Library preparation for transcriptome analysis (RNA-seq)

Total RNAs were extracted by TRIzol reagent (Invitrogen) from the dorsal hippocampus (dHC) of mice at the age of postnatal day 7 (P7), P12, P17 and from the motor cortex (mCTX) at P7, P12, P17, and P90 (3 m). RNA quality and integrity were verified by spectrophotometry (Nanodrop) and Qsep 100 DNA/RNA Analyzer (BiOptic Inc., Taiwan), respectively. One microgram of total RNA was used as inputs for library construction following manufacturer’s instructions (KAPA mRNA HyperPrep Kit, KAPA Biosystems, Roche). Briefly, mRNA was purified from total RNA using magnetic oligo-dT beads. Captured mRNA was fragmented and reversely transcribed to the first strand cDNA using random hexamer. The cDNA:RNA hybrids were converted to double-stranded cDNA (dscDNA) followed by fragmentation, which were then purified with KAPA Pure Beads system (KAPA Biosystems, Roche) to select cDNA fragments of preferentially 300 ~ 400 bp in length. The library carrying appropriate adapter sequences at both ends was amplified using KAPA HiFi HotStart ReadyMix (KAPA Biosystems, Roche) with library amplification primers. The PCR products were purified using KAPA Pure Beads system and the library quality was assessed by the Qsep 100 DNA/RNA Analyzer (BiOptic Inc., Taiwan).

RNA-seq data analysis

RNA sequencing was conducted by BioTools INC. (Taiwan). The original data acquired from high-throughput sequencing (Illumina NovaSeq 6000 platform) were transformed into raw sequenced reads by CASAVA base with sequencing depth of 6G. FastQC and MultiQC [124] were used for quality control. The obtained raw paired-end reads were filtered by Trimmomatic (v.0.38) [125] to discard low-quality reads, trim adaptor sequences, and eliminate poor-quality bases. Only the high-quality data (clean reads) was used for subsequent analysis (supplementary Table 2). By using the HISAT2 alignment, 89–93% of total reads were uniquely mapped to the mouse Ensembl GRCm38/mm10 mouse genomic assembly [126,127]. FeatureCounts (v.2.0.0) was used to count the read numbers mapped to individual genes. To compare the read counts between sample groups with biological replicate, we conducted “relative log expression” (RLE) normalization using DESeq2, which normalizes the read counts for sequencing depth and RNA composition by using the median ratio method based on the Poisson distribution model [128–130]. The resulting p-values were adjusted using the Benjamini and Hochberg’s approach for controlling the False Discovery Rate (FDR) correction. Gene ontology (GO) enrichment analysis of DEGs were conducted using clusterProfiler (v.3.14.3) [131,132].

RNAscope in situ hybridization

Cdkl5 null mice and their littermate WT controls at the age of P7 were anesthetized with isoflurane and transcardially perfused with 4% paraformaldehyde (PFA, Merck) in 1 x phosphate-buffered saline (PBS). Fixed brains were kept in 4% PFA for 16 h at 4°C before being cryoprotected in 30% sucrose in PBS for an additional 48 h. Brains were next embedded and frozen in OCT compound (Tissue-Tek®, Sakura Finetek USA, Inc.), cut by cryostat (CM3050S, Leica Biosystems) at 12 μm and mounted onto SuperFrost™ Plus slides (Epredia Inc., USA). RNA in situ hybridization (ISH) was performed manually using the RNAscope™ 2.5 HD duplex assay (Advanced Cell Diagnostics, USA) according to the manufacturer’s instructions. Briefly, tissues were washed in 1X PBS to remove OCT completely and baked at 60°C for 30 min. Pretreatment of sections was performed by incubation of them in H₂O₂ for 10 min, boiled target retrieval solution for 5 min, followed by protease digestion buffer at 40°C for 30 min. After rinsing the slides with distilled water, the probe mixture containing the C1 Probe (Glra2, #510301) and C2 probe (Cdkl5, #500851-C2) in a ratio of 50:1 was applied onto slides and incubated at 40°C for 2 h. Amplification steps were done using the reagents provided in the kit. Signal detection was achieved using two different chromogenic substrates (HRP-C1-Green and AP-C2-Red). Red signals were detected following Amp 1–6 reactions by applying a mixture of Fast RED-B and Fast RED-A (1:60) at room temperature for 10 min. Slides were washed and then incubated in Amp 7–10 sequentially, followed by green signal detection with a mixture of Fast Green-B and Fast Green-A (1: 50) for 10 min. Lastly, slides were counterstained with diluted (20%) hematoxylin (Mayer’s solution, MHS16, Sigma) and coverslipped with mounting medium (VectaMount, Vector Laboratories). To assess quality of the tissues, we detected the expression of a housekeeping gene peptidylprolyl isomerase B (Ppib, #313911) across different brain regions as a positive control.

Real-time quantitative RT-PCR

The cortical tissues from P7, P12 or P17 brains were lysed in Trizol (Invitrogen) followed by RNA extraction with Direct-zol RNA Miniprep Kit (Zymo Research). Primers for regulator of G-protein signaling 9 (Rgs9, NM_011268; FW: 5’-TACGGCGATCAGTCCAAGGTCA, RV: 5’-CTGCATCCAGTACATAGCGGTG), potassium voltage-gated channel subfamily Q member 3 (Kcnq3, NM_152923; FW: 5’-AAGCCTACGCTTTCTGGCAGAG, RV: 5’-ACAGCTCGGATGGCAGCCTTTA), glutamate decarboxylase 1 (Gad1, NM_008077; FW: 5’-CGCTTGGCTTTGGAACCGACA A, RV: 5’-GAATGCTCCGTAAACAGTCGTGC), synaptotagmin II (Syt2, NM_009307; FW: 5′- CATGTTCGCCAAGCTGAAGGAG, RV: 5′- GATGCAGAAGCAACAGGTGAGC) and glyceraldehyde-3-phosphate dehydrogenase (Gapdh, NM_008084; FW: 5′- CATCACTGCCA CCCAGAAGAC TG, RV: 5’-ATGCCAGTGAGCTTCCCGTTCAG) were determined according to online resources (origene.com). The specificity of primers was confirmed by Nucleotide BLAST alignment. RT-qPCR was performed in triplicate with the KAPA SYBR®FAST One-step qRT-PCR Kit (Merck) by the Applied Biosystems 7300 Real-Time PCR System (Applied Biosystems) as described previously [133]. Twenty nanograms RNA and primers at the final concentration of 150 nM were used for each reaction. Following RT reaction at 42°C for 5 min, the PCR reaction was started by 95°C for 5 min, followed by 40 cycles at 95°C for 15 s and 70°C for 60 s, and ended by dissociation at 95°C for 15 s and detection at 60°C for 60 s. The expression levels of Rgs9, Kcnq3, Gad1 and Syt2 genes were normalized with that of Gapdh (as an internal control), and the relative transcript levels of the interested genes in Cdkl5 null mice were presented as “ratio to WT”.

Immunohistochemistry

Immunostaining was performed as described previously [134]. Briefly, brain sections of 12 μm (for P7) or 20 μm (for P17) were pretreated with 0.1 M PBS containing 0.2% Triton X-100, 3% H₂O₂ and 10% methanol for 10 min, and then blocked with 3% normal serum in 0.1 M PBS. Sections were incubated with the rabbit polyclonal primary antibodies against cholecystokinin (CCK-8, 1:5000, ImmunoStar) or mouse monoclonal primary antibodies against parvalbumin (PV, 1:2000, Sigma) at 4°C for 16 h, followed by incubation with biotinylated goat-anti-rabbit or horse-anti-mouse secondary antibody (1:500, Vector Laboratories) in 0.1 M PBS containing 1% normal serum at room temperature for 2 h. Sections were then incubated for 1.5 h with avidin-biotin-complex (Elite ABC kit, Vector Laboratories) and immunoreactivities were detected with 0.02% diaminobenzidine (DAB; Sigma-Aldrich) in the presence of 0.0002% H₂O₂ (Sigma-Aldrich) and 0.08% nickel ammoniosulfate (Sigma-Aldrich). Brain sections from mutant mice were processed in parallel with the sections from their littermate WT controls, and color developed in DAB solution for exactly the same duration as control sections. Photomicrographs were taken by an upright microscopic system (DM2000, Leica) equipped with a CCD camera (DFC450 C, Leica).

Immunoblotting

Immunoblotting was performed as previously described [135]. The cortical and hippocampal tissues were harvested from Cdkl5^−/y pups and their WT littermates (n = 6) at the age of P12, and then homogenized by sonication in lysis buffer containing 1% protease inhibitors and 1% phosphatase inhibitor (Sigma). The protein lysates containing 20 micrograms of protein were subject to electrophoresis in 4–15% polyacrylamide gel (Bio-Rad) with 125 V for 100 min. The proteins were then transferred to PVDF membranes (Millipore) with 350 mA for 2 h (Mini Trans-Blot Cell, Bio-Rad). The membranes were blocked using 5% skim milk for 1 h and then incubated with primary antibodies against CDKL5 (1:1000, Abcam), KCC2 (1:1000, Millipore), pKCC2-S940 (1:5000, PhosphoSolutions), pKCC2-T1007 (1:5000, PhosphoSolutions), NKCC1 (1:10000, Millipore), PSD-95 (1:2000, ThermoFisher), GABA_AR-gamma 2 (GABA_ARγ2; 1:2000, Millipore) and GAPDH (1:50000, Millipore) at 4°C for 16 h. Following incubation with peroxidase-conjugated secondary antibodies at room temperature for 2 h, the target proteins were detected by an enhanced chemiluminescence reagent (ECL, Millipore) with an image acquisition system (Biostep, Germany). After the pKCC2 were detected, the membranes were stripped and re-blocking, followed by incubation of primary antibodies for KCC2. The protein band shown and quantified as pKCC2 was identified based on the predicted molecular weight (~135 kDa) and its sensitivity to a blocking peptide designed against pKCC2. The intensity of target protein expression (with background subtracted for each lane) was quantified with densitometry (Image J, NIH) and normalized with the loading control of GAPDH. The protein expression in mutants was presented as “ratio to WT” and the student’s t-test was used for comparison between genotypes.

Acknowledgements

This work was supported by the Ministry of Science and Technology (National Science and Technology Council), Taiwan [grant number MOST107-2320-B004-001-MY3, MOST110-2320-B-004-001 and MOST111-2320-B-004-002 to W.L.; MOST111-2636-B-110-001 to K-Z.L.) and the International Foundation for CDKL5 Research, USA (2015-2020 to W.L.). We thank Drs. Jin-Chung Chen (Chang Gung University) and Shun-Fen Tzeng (National Cheng Kung University) for critical inputs; Dr Sin-Jhong Cheng (Academia Sinica), Misses San-Hua Su, Yuju Luo and Ting-Han Hu for technical assistance; and Miss Yuan-Ting Lin for graphic illustration.

Conflict of interest statement: None declared.