Heterogeneous subpopulations of GABA_AR-responding neurons coexist across neuronal network scales and developmental stages in health and disease

Summary

Gamma-aminobutyric acid (GABA) is the main inhibitory neurotransmitter in adults. Depolarizing GABA responses have been well characterized at neuronal-population average level during typical neurodevelopment and partially in brain disorders. However, no investigation has specifically assessed whether a mosaicism of cells with either depolarizing or hyperpolarizing/inhibitory GABAergic responses exists in animals in health/disease at diverse developmental stages, including adulthood. Here, we showed that such mosaicism is present in wild-type (WT) and down syndrome (DS) neuronal networks, as assessed at increasing scales of complexity (cultures, brain slices, behaving mice). Nevertheless, WT mice presented a much lower percentage of cells with depolarizing GABA than DS mice. Restoring the mosaicism of hyperpolarizing and depolarizing GABA-responding neurons to WT levels rescued anxiety behavior in DS mice. Moreover, we found heterogeneous GABAergic responses in developed control and trisomic human induced-pluripotent-stem-cells-derived neurons. Thus, a heterogeneous subpopulation of GABA-responding cells exists in physiological/pathological conditions in mouse and human neurons, possibly contributing to disease-associated behaviors.

Graphical abstract

Highlights

• •Subpopulations of GABA_AR-responding neurons exist in mouse and human neuronal networks
• •DS networks exhibit a larger fraction of neurons with depolarizing GABA responses
• •Restoring physiological GABA-mediated inhibition rescues anxiety behavior in DS mice
• •Heterogeneous GABAergic responses coexist in control and DS human iPSC neurons

Behavioral neuroscience; Developmental neuroscience; Cellular neuroscience

Introduction

Gamma-aminobutyric acid (GABA) is the main inhibitory neurotransmitter in the adult mammalian brain by acting on chloride (Cl⁻)-permeable GABA_A receptors (GABA_ARs). In mature neurons, the activity of the Cl⁻ exporter KCC2 predominates over that of the Cl⁻ importer NKCC1, resulting in a low intracellular chloride concentration ([Cl⁻]_i). As a consequence, common knowledge indicates that the chloride reversal potential through GABA_ARs (E_GABA) lies below the cellular resting membrane potential (RMP) in mature neurons. This makes inward the flow of negatively charged chloride ions through the receptor; thus, GABA_AR signaling is hyperpolarizing and inhibitory^1,2,3 in adult animals. Conversely, during neuronal development (until the first or second postnatal week in rodents), GABA exerts a depolarizing and possibly excitatory effect due to low expression of KCC2 in immature neurons, which leads to a high [Cl⁻]_i and an outward flow of Cl⁻ through GABA_ARs.² Depolarizing GABAergic signaling during development is fundamental for organizing early patterns of neuronal activity by enabling neurons to fire and thus form their first synaptic connections.^{4,5,6,7,8,9,10} The excitatory-to-inhibitory developmental switch in the polarity of GABA_AR signaling has been very well characterized at the level of the neuronal population average.^11,12,13 However, there is still a possibility that a small subpopulation of neurons with depolarizing polarity of GABA_AR-mediated responses may persist even after full neuronal maturation. Moreover, some neurons in mature neuronal networks exhibit significant cellular microdomains with diverse intracellular Cl⁻ concentrations, resulting in a mosaicism of GABAergic signaling in various cellular compartments within the same cell (e.g., the axon, axonal initial segment, soma, and dendrites^{14,15,16,17,18}). Although the heterogeneity of GABA_AR-mediated responses at the level of cellular compartments within a same cell is well accepted, the presence of a heterogeneous subpopulation of neurons with depolarizing vs. hyperpolarizing GABA responses in mature neuronal networks is not yet a clear and well-established concept in the neuroscience field. Although this small percentage of depolarizing GABA neurons could be hidden within the population average or may be fairly interpreted as a biological outlier or as standard noise in experimental datasets, these depolarizing-GABA-responding neurons may in fact represent a specific subpopulation of cells coexisting with the large majority of cells with hyperpolarizing-GABA_AR-mediated signaling and have important physiological and pathological roles, even in light of the fact that a depolarizing response to GABA will not be sufficient to directly induce action potential firing in mature neurons. In fact, the relatively small depolarization triggered by GABA could anyhow set the cell in a state of increased excitability (e.g., activation of Ca²⁺ influx from voltage-gated Ca²⁺ channels or partial release of the magnesium block from N-methyl-D-aspartate [NMDA] receptors), and therefore more likely to fire an action potential, as seen during development.

A rapidly increasing evidence reported in the last 20 years of literature has indicated an altered NKCC1/KCC2 expression ratio together with depolarizing-GABA_AR-mediated signaling as a common feature of a very large number of brain pathologies.^13,19,20 These range from neurodevelopmental disorders (e.g., Down syndrome, Rett syndrome, Fragile X syndrome, 22q11.2 microdeletion syndrome, and schizophrenia) to neurodegenerative disorders (e.g., Huntington disease and Parkinson disease) and neurological conditions (e.g., chronic pain and some forms of epilepsy^13,19,20). However, whether the NKCC1/KCC2 expression ratio in these brain disorders reflects generally depolarized GABA_AR-mediated responses in most neurons or simply a specific increase in the numerical extent of a subpopulation of neurons with depolarizing responses is not known, although it may have important functional implications at the neuronal network and behavioral levels.

Here, by performing Ca²⁺ imaging recordings and/or electrophysiological on large populations of neurons at increasing scales of network complexity (in vitro neuronal cultures, ex vivo brain slices, and in vivo freely behaving mice), we found a significant subpopulation of neurons with NKCC1-dependent, network scale depolarizing GABAergic responses both in wild-type (WT) mice and in a widely studied mouse model of DS (Ts65Dn adult mice). Treatment with the NKCC1 inhibitor bumetanide fully rescued the numerical extent of the subpopulation of neurons with inhibitory GABAergic signaling and normalized both anxiety-related behaviors and paradoxical responses to the pharmacological activation of GABA_ARs with a clinically used anxiolytic drug in adult DS animals without altering the behavior of adult WT animals. Finally, we found a heterogeneous subpopulation of neurons with mixed responses to GABA_AR signaling also in human isogenic control and DS neurons derived from induced pluripotent stem cells (iPSCs) obtained from a person with a form of mosaic trisomy 21. Those heterogeneous subpopulations of GABA-responding neurons were still present at a developmental time point when the NKCC1/KCC2 expression levels reached a maturation plateau. In addition, in human developed neurons, trisomic networks showed a significantly larger subpopulation of neurons with depolarizing GABAergic signaling than that in the corresponding isogenic control neuronal networks. Thus, heterogeneous subpopulations of GABA-responding neurons are present in human and murine neuronal networks under physiological and pathological conditions, where they possibly have implications in dysfunctional behaviors.

Results

Mixed subpopulations of neurons with hyperpolarizing or depolarizing GABA signaling are present in WT and Ts65Dn cultures at diverse stages of in vitro development

To investigate whether a heterogeneous population of neurons with mixed (hyperpolarizing and depolarizing) GABA-mediated responses is present in vitro, we performed Ca²⁺ imaging experiments in hippocampal neuronal primary cultures from WT (B6EiC3) and Ts65Dn mice at diverse stages (days in vitro [DIV] after plating) across development, a period characterized by the physiological transition of GABA from excitatory to inhibitory.^21,22,23 Ts65Dn mice are a well-studied animal model in which an average depolarizing GABAergic signaling has already been described by electrophysiological recordings in both neuronal cultures and acute brain slices from adult animals.^24,25 We imaged neurons at diverse stages of maturation (2, 7, and 15 DIV) by evaluating Ca²⁺ responses elicited by bath application of GABA (100 μM) in WT and Ts65Dn neuronal cultures loaded with the Ca²⁺-sensitive dye Fluo-4 (Figure 1A). In fact, the transient increase in intracellular neuronal calcium concentration ([Ca²⁺]_i) due to the opening of voltage-gated Ca²⁺ channels by the application of GABA_AR agonists can be used as a proxy to assess the depolarizing action of GABA, which has been previously described using this approach for neuronal population averages during development.^10,23,26 By calculating the percentage of cells showing GABA-induced Ca²⁺ responses in our neuronal cultures at increasing DIVs, we found that not all neurons underwent the expected switch in the GABA response from excitatory to inhibitory, even at the latest developmental stage analyzed (Figures 1A and 1B). Indeed, at 21 DIV (well beyond the completion of the GABA developmental switch; Figure S1 and^22,23,27), we still observed a subpopulation of WT neurons (14% of cells) presenting depolarizing GABA signaling. This subpopulation was significantly smaller than that in immature WT neurons at 2 DIV (80% of neurons with depolarizing GABA) or at 7 DIV (26%), but similar to that found at 15 DIV (15%; Figure 1B). Interestingly, although the initial developmental drop (2–7 DIV) in the subpopulation of neurons with depolarizing-GABA-mediated responses in Ts65Dn neurons was similar to that of WT neurons, the subpopulation of Ts65Dn neurons with depolarizing GABA responses was significantly greater at later time points (15 and 21 DIV; Figure 1B). These results were corroborated by the significantly higher levels of NKCC1 expression (and no difference in KCC2 levels) that we found in Ts65Dn neurons at 15 and 21 DIV in comparison to WT controls (Figure S1).

Figure 1.

Mixed subpopulations of neurons with depolarizing or hyperpolarizing GABA signaling are present during in vitro development of neuronal primary cultures

(A) Representative calcium traces obtained from live imaging of cultured WT and Ts65Dn hippocampal neurons loaded with the Ca²⁺-sensitive dye Fluo-4 upon bath application of GABA (100 μM) at different time points during in vitro development. ΔF/F0 represents the change in the calcium intensity expressed as variation in fluorescence intensity upon GABA application, relative to the basal fluorescence intensity.

(B) Quantification of the mean percentage (±SEM) of neurons showing GABA-induced Ca²⁺ responses (i.e., depolarizing GABA responses) in experiments as in (A). Numbers in parentheses indicate the number of analyzed coverslips for each time point (obtained from 3 to 4 independent neuronal cultures). ∗∗∗p < 0.001; Tukey post hoc test following two-way ANOVA.

(C) Representative calcium traces of neurons pretreated with vehicle (0.01% DMSO) or the NKCC1 inhibitor bumetanide (10 μM) upon bath application of GABA at 15 DIV.

(D) Quantification of the percentage of neurons showing depolarizing GABA responses in experiments as in (C). In the boxplot, the small square indicates the mean, the central line illustrates the median, the box limits indicate the 25th and 75th percentiles, the whiskers represent the 5th–95th percentiles, and each dot indicates a value obtained from an individual coverslip. The numbers in parentheses indicate the number of analyzed coverslips for each experimental group (obtained from 3 to 5 independent neuronal cultures). ∗p < 0.05, ∗∗∗p < 0.001; Tukey post hoc test following two-way ANOVA.

(E) Representative images of WT and Ts65Dn hippocampal neurons during imaging experiments with the chloride-sensitive dye MQAE upon treatment with vehicle (0.01% DMSO) or bumetanide (10 μM) at 15 DIV. The fluorescence intensity of the dye (color-coded at the bottom) is inversely proportional to intracellular chloride concentration ([Cl⁻]_i). Scale bar: 100 μm.

(F) Beanplot showing the distribution of MQAE raw fluorescence values for all neurons imaged as in E (WT-vehicle: 840 neurons, Ts-vehicle: 818 neurons, WT-bumetanide: 597 neurons, Ts-bumetanide: 840 neurons, from 3 independent experiments).

(G) Quantification of the average [Cl⁻]_i with MQAE in the same dataset described in (F). In the boxplot, the small square indicates the mean, the central line illustrates the median, the box limits indicate the 25th and 75th percentiles, the whiskers represent the 5th to 95th percentiles, and each dot indicates a value obtained from an individual coverslip. The numbers in parentheses indicate the number of analyzed coverslips for each experimental group (obtained from 3 independent neuronal cultures). ∗p < 0.05, ∗∗∗p < 0.001; Tukey post hoc test following two-way ANOVA.

Next, to specifically assess the involvement of NKCC1 in the subpopulation of neurons retaining depolarizing responses to GABA in WT and Ts65Dn cultures, we evaluated the percentage of cells showing GABA-induced Ca²⁺ responses in the presence of the NKCC1 inhibitor bumetanide (10 μM²⁴) in neuronal cultures at 15 DIV. Fifteen DIV is a stage when GABAergic signaling has already switched from excitatory to inhibitory in most WT neurons and it is when we first detected a significant difference in the developmental trajectory of GABA-mediated responses in comparison to Ts65Dn neurons. We found that bumetanide treatment strongly reduced the percentage of neurons showing GABA-induced Ca²⁺ responses in both WT and Ts65Dn cultures (Figures 1C and 1D). Accordingly, when we performed live imaging experiments in primary hippocampal neurons at 15 DIV with the chloride-sensitive dye MQAE, we found that WT neurons displayed a bimodal distribution in the raw MQAE fluorescence values, indicating the presence of neurons with both low and high values of [Cl⁻]_i, (Figures 1E and 1F). Conversely, the distribution of MQAE values of Ts65Dn neurons was broader and shifted toward lower values (reflecting a higher [Cl⁻]_i) compared with that of WT cultures, as indicated by both the beanplot and the boxplot quantifications (Figures 1F and 1G). Interestingly, NKCC1 inhibition by bumetanide (10 μM) abolished the bimodal distribution of the raw MQAE values and significantly decreased [Cl⁻]_i (i.e., higher MQAE fluorescence) in WT neurons (Figures 1E–1G). Similarly, NKCC1 inhibition decreased the higher [Cl⁻]_i found in Ts65Dn neurons to within more physiological levels, similar to those of WT neurons (Figures 1E–1G). Because MQAE fluorescence may be influenced by differences in the dye’s uptake between WT and TS65Dn neurons, we also performed a control experiment by measuring the fluorescent intensity of the non-chloride-sensitive dye calcein AM and found no differences (Figures S2A and S2D).

Next, to confirm that the presence of a subpopulation of neurons with depolarizing GABA signaling in neurons was indeed mediated through GABA_ARs, we performed control experiments in WT and Ts65Dn neuronal cultures at 15 DIV in the presence of the specific GABA_A agonist muscimol (10 μM), the GABA_AR antagonist bicuculline (100 μM), or the GABA_BR antagonist CGP55845 (10 μM²⁴). We found that muscimol elicited GABA_AR-dependent depolarizing Ca²⁺ responses completely similar to those elicited by GABA application (Figures S3A and S3B). This effect was again inhibited by bumetanide application both in WT and Ts65Dn neurons (Figures S3A and S3B). Accordingly, GABA-induced Ca²⁺ responses were largely abrogated by blocking GABA_ARs with bath application of bicuculline (Figures S3C and S3D). Of note, pretreatment of neurons with the L-type voltage-gated calcium channel (CaV1) blocker nifedipine (10 μM) also strongly reduced GABA-induced Ca²⁺ responses, indicating that GABA-induced depolarization generates a Ca²⁺ inflow mainly through the activation of CaV1 channels in both WT and Ts65Dn neurons (Figures S3C and S3D). GABA-induced Ca²⁺ responses were enhanced after blocking GABA_BRs with CGP55845 bath application, consistent with the inhibitory function of GABA_BRs on neuronal networks. Then, we performed additional control experiments in the presence of the voltage-gated sodium channel blocker tetrodotoxin (TTX, 1 μM; Figures S4A and S4B) or with the carbonic anhydrase inhibitor acetazolamide (ACZ, 10 μM; Figures S4C and S4D) to block global neural network activity or the depolarizing component of HCO3⁻ flux through GABA_ARs,²⁸ respectively. We observed that the percentage of cells responding to GABA application in the presence of either TTX or ACZ was nearly identical to that elicited by GABA alone (Figures S4B and S4D), indicating direct Cl⁻ flux through GABA_AR as a first driver of the depolarizing responses to GABA in our experiments.

Finally, we also controlled for the maturation level of the neuronal network along in vitro development in our experimental conditions. By immunostaining experiments with immature (doublecortin; DCX) or mature (NeuN) neuronal markers together with the pan-neuronal marker tubulin β3 (Tubβ3), we observed an expected significant increase in the proportion of NeuN- and Tubβ3-double-positive cells in WT and TS65Dn neurons along development (Figures S2E, S2F, S5A, and S5B). Accordingly, the percentages of DCX- and Tubβ3-double-positive cells, as well as DCX- and NeuN- and Tubβ3-triple-positive cells gradually decreased and reached a minimum at the same time point, indicating the simultaneous loss of immature-neuronal progenitor markers along in vitro development.

Moreover, to exclude possible differences in neuronal density of WT and Ts65Dn cultures affecting the maturation rate of the network, we measured calcein and propidium iodide uptake at DIV 15, to label live and dead cells, respectively. Although Ts65Dn cultures exhibited a slightly lower cell death rate compared with WT cultures, we found no difference in the density of live cells between WT and Ts65Dn cultures (Figures S2A–S2C). Finally, to further assess the maturational stage of the neuronal network in our experimental conditions, we quantified GABAergic and glutamatergic synapse density by immunostaining with antibodies against the vesicular transporters vGAT and vGLUT and found no differences between WT and Ts65Dn cultures at DIV21 (Figures S2G and S2H).

In additional, to assess the level of neuronal maturity of the network at functional level, we evaluated passive membrane properties (i.e., resting membrane potential, input resistance, and rheobase) and functional synapses in WT and Ts65Dn neurons at 15 and 21 DIV (Figure S2I and Table S1). We found that WT and Ts65Dn neurons showed passive membrane properties similar to those recorded in adult WT rodent neurons²⁹ already at DIV 15, reflecting large maturation of the network at this stage. Notably, we found no difference between WT and Ts65Dn neurons in all parameters and time points considered. Thus, because the neuronal maturation process was largely accomplished by 15 DIVs (at least according to the expression of developmental markers and electrophysiological parameters), the presence of neurons exhibiting depolarizing GABAergic responses was not only mirroring the immature stage of the developmental process in our cultures.

Altogether, our results indicated the presence of a small subpopulation of WT neurons with NKCC1-mediated depolarizing GABA_A signaling, which was substantially larger in Ts65Dn cultures.

Subpopulations of neurons with functionally heterogeneous GABA_AR-mediated responses are present in WT or Ts65Dn neuronal networks cultured over MEAs at diverse stages of in vitro development

To assess the functional consequences of the presence of the two disproportional subpopulations of neurons with hyperpolarizing or depolarizing GABAergic responses in WT and Ts65Dn neuronal networks, we recorded their firing activity by microelectrode arrays (MEAs). This technique allows the simultaneous recording of the activity from diverse spots of a large population of neurons without interfering with neuronal [Cl⁻]_i, although MEA recordings by detecting action potentials activity only provided an indirect measure of the polarity of the GABAergic response of neurons. Raw traces were high-pass filtered (>300 Hz) to isolate spike events (multiunit activity) from the low fluctuation of the signal (LFP). In the first set of experiments, we assessed the neuronal mean firing rate (MFR) after blocking endogenous GABAergic signaling by bath application of bicuculline (20 μM²⁴) on WT and Ts65Dn (21 DIV) cultures, which were previously preincubated with vehicle (0.01% DMSO) or bumetanide (10 μM; Figure 2A). For each electrode, we evaluated the changes in activity elicited by bicuculline by calculating the MFR ratio (MFR after the treatment over the MFR during baseline). As expected, the average MFR ratio of WT control cultures (pretreated with vehicle; Figure 2B, dark blue box; raster plots in Figures S6A and S6B) in the presence of bicuculline was (5 times) higher than the MFR of untreated neurons (dotted line). Similarly, WT cultures pretreated with bumetanide (light blue box) also displayed a large average MFR ratio, which was even greater than that seen in the corresponding vehicle-treated WT cultures. Conversely, the Ts65Dn control cultures (pretreated with vehicle; Figure 2B, red box; raster plots in Figure S6B) showed a significantly lower average MFR ratio, indicating a greatly blunted response to GABA_AR inhibition by bicuculline, when compared with that in WT cultures. Interestingly, pretreatment with bumetanide restored the full response to bicuculline application in Ts65Dn neurons, similar to levels observed in WT cultures (Figure 2B, orange box; raster plots in Figure S6B).

Figure 2.

Mixed subpopulations of neurons with hyperpolarizing or depolarizing GABA_AR signaling are present in neuronal cultures grown over MEAs

(A) Left: schematic representation of a primary hippocampal neuronal culture grown over a microelectrode array (MEA) for electrophysiological recordings. Center: representative transmitted-light image showing a well of 60 electrodes MEA seeded with hippocampal neurons at 21 DIV. Right: schematic representation of the experimental protocol; neuronal cultures were preincubated with vehicle (0.01% DMSO) or bumetanide (10 μM) for 45 min followed by 30-min recording of spontaneous activity. Neurons were then recorded for an additional 30 min, 10 min after the addition of bicuculline (BIC, 20 μM) or GABA (100 μM).

(B, E) Quantification of the mean firing rate (MFR) ratio of the WT and Ts65Dn neuronal cultures upon BIC (B) or GABA (E) bath application. The MFR ratio over the baseline firing (dotted line) was calculated for each electrode and then averaged for each MEA. An MFR ratio higher than 1 (representing the baseline level) indicates an increase in activity upon treatment, whereas a ratio below 1 indicates a decrease in activity. In the boxplot, the small square indicates the mean, the central line illustrates the median, the box limits indicate the 25th and 75th percentiles, the whiskers represent the 5th to 95th percentiles, and each dot represents the MFR ratio for each recorded culture. The numbers in parentheses indicate the number of analyzed MEAs for each experimental group (obtained from 5 independent neuronal cultures). ∗∗p < 0.01; ∗∗∗p < 0.001; Tukey’s post hoc test following two-way ANOVA.

(C, F) Scatterplots showing the MFR for each active electrode (plotted as a dot) from all recorded MEAs (in B or E) seeded with WT and Ts65Dn neurons before (x axis) and after (y axis) bath application of BIC (C) or GABA (F). Dark gray dots represent electrodes showing a significant increase in the MFR. Light gray dots represent electrodes showing a significant decrease in the MFR. Black dots represent electrodes showing no significant changes in the MFR. Significant changes in the MFR (numbers with arrows) for each electrode upon BIC or GABA application were evaluated by bootstrap analysis.

(D, G) Quantification of the average percentage number (±SEM) of MEA electrodes (in the same experiments in B, C, E, and F), showing significant changes in the MFR by bootstrap analysis after BIC (D) or GABA (G) administration in comparison to their basal conditions in WT (blue) and Ts65Dn (pink) neuronal cultures. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; Tukey’s post hoc test following two-way ANOVA on ranked transformed data.

As we hypothesized the presence of heterogeneous subpopulations of neurons with hyperpolarizing- or depolarizing-GABA_AR-mediated responses, we took advantage of the large number of MEA observation points (i.e., electrodes) scattered through the network as a proxy to extrapolate information about possibly different subpopulations of neurons within the network.^30,31 For each active (i.e., mean firing rate higher than 0.02 spikes/sec) electrode, we evaluated whether the MFR upon bicuculline application was significantly different from the basal level of activity by using a bootstrap method. Both the scatterplot representation of single-electrode MFR activity before and after bicuculline application (Figure 2C) and the average percentage of electrodes that were significantly changed by bootstrap analysis (Figure 2D) clearly showed an increase in the firing activity for the majority of the electrodes (85% of MFR-increasing electrodes) for vehicle-treated WT cultures. Notably, in the same recordings, we found that 6% of electrodes demonstrated a significant decrease in the firing rate activity, consistent with the presence of a subpopulation of neurons presenting depolarizing rather than hyperpolarizing GABA responses in WT cultures. Bumetanide treatment in these WT cultures caused a small (and nonsignificant) change in the percentage of electrodes, showing either an increase or a decrease in MFR activity (Figures 2C, 2D, and S6B). This likely reflects only a very subtle enhancement of inhibitory GABAergic drive upon bumetanide application due to the prevalence of KCC2 expression over NKCC1 expression at this stage of development.^32,33

Conversely, vehicle-treated Ts65Dn cultures showed far more heterogeneous responses to the blockade of GABA_AR signaling by bicuculline than WT cultures demonstrated. Indeed, we found a higher percentage of electrodes that decreased MFR activity (23% of MFR-decreasing electrodes) and a decrease in electrodes showing increased MFR (59% of MFR-increasing electrodes) in comparison to the WT cultures (Figure 2C), consistent with alterations in GABA_AR-mediated responses toward more depolarizing values (Figure 2D). Notably, pretreatment of Ts65Dn cultures with bumetanide significantly shifted the distribution of the MFR values to WT levels (84% of MFR-increasing electrodes and 7% of MFR-decreasing electrodes; Figure 2C), indicating a full rescue of GABA_AR-mediated inhibition (Figure 2D). We found no statistically significant difference in the percentage of electrodes that did not respond to the treatment of both WT and Ts65Dn cultures under all conditions (Figures S7A and S7B).

Next, we analyzed the average MFR variation upon treatment with exogenously applied GABA (100 μM²⁴; Figure 2E). WT cultures pretreated with either vehicle or bumetanide at 21 DIV showed a strong decrease in the average MFR ratio (well below 1, Figure 2E; raster plots in Figure S6C), consistent with the expected decrease in the neuronal activity level upon GABA application. Conversely, vehicle-treated Ts65Dn cultures displayed a significantly higher average MFR ratio and an overall much larger variability among MEAs than controls upon GABA treatment. Interestingly, pretreatment with bumetanide fully restored inhibitory GABAergic signaling in trisomic cultures to WT levels (Figure 2E; raster plots in Figure S6C).

As described earlier, we used bootstrap methods to evaluate the difference in the firing rate activity at the single-electrode level. Scatterplot representation of single MFR values for each active electrode and the average percentage of electrodes that significantly changed by bootstrap analysis before and after GABA application in vehicle-treated WT cultures showed an abrupt suppression of the firing activity in virtually all (99.7% of MFR-decreasing electrodes) active electrodes (Figures 2F, 2G, and S6C). Bumetanide pretreatment had no significant effect on WT cultures (Figures 2F, 2G, and S6C). In vehicle-treated Ts65Dn cultures with exogenously applied GABA, the distribution of the single MFR values was clearly different from that observed for WT cultures, with a wide dispersion of the data. Only 68% of electrodes were characterized by a decreased MFR, and 8% of electrodes showed an increased MFR, indicating a general reduction in the efficacy of GABA inhibition and even partial GABA-driven excitatory activity (Figures 2F, 2G, and S6C). Notably, treatment with bumetanide fully restored inhibitory GABA signaling in Ts65Dn cultures, as demonstrated by the distribution and average of single MFR values, which was indistinguishable from that of WT cultures (100% of MFR-decreasing electrodes; Figures 2F, 2G, and S6C). In support to the aforementioned data indicating reduced inhibition in Ts65Dn neurons, we found a significant difference in the number of electrodes that did not change significantly upon GABA application between vehicle-treated Ts65Dn and WT cultures (24% vs. 0% of electrodes with no MFR changes, respectively; Figure S7C). This effect was also completely rescued by bumetanide treatment (Figure S7C).

To strengthen our findings, we also performed experiments with the GABA_AR-positive allosteric modulator (PAM) diazepam (a benzodiazepine widely used as an anxiolytic medication) in WT and Ts65Dn neuronal cultures (Figure S8A). We found that diazepam application (1 μM) consistently modulated network activity by decreasing firing in most (but not all) recorded electrodes in WT cultures pretreated with either vehicle (4% of MFR-increasing electrodes, 77% of MFR-decreasing electrodes) or bumetanide (3% of MFR-increasing electrodes, 78% of MFR-decreasing electrodes; Figures S8B–S8E). Conversely, in Ts65Dn cultures, the MFR distribution was more dispersed, with 11% of the electrodes showing increased activity and 63% of the electrodes showing decreased activity. Notably, NKCC1 inhibition by bumetanide fully recovered diazepam-mediated GABAergic inhibition, as indicated by the average MFR ratio, MFR distribution (5% of MFR-increasing electrodes and 78% of MFR-decreasing electrodes), and average percentage of electrodes that were significantly changed by bootstrap analysis of Ts65Dn neurons, which were similar to those of WT cultures (Figures S8B–S8E).

Finally, to assess the influence of the intrinsic excitatory network activity on the subpopulations of GABA-responding cells, we performed control experiments in the presence of the glutamate receptor blockers CNQX (20 μM) and AP5 (50 μM) in neuronal cultures at DIV 21 (Figure S9A). We calculated the average MFR ratio for each culture by dividing the MFR after treatment with CNQX+AP5 or with (control) vehicle by the MFR during baseline for each electrode. The application of CNQX+AP5 led to a significant decrease in activity in both WT and Ts65Dn cultures, as shown by the decrease in the average MFR ratio well below 1 (Figures S9B and S9C). This was consistent with the expected decrease in neuronal activity following the blockade of excitatory neurotransmission. Interestingly, Ts65Dn cultures exhibited a smaller decrease in the average firing rate ratio upon CNQX+AP5 addition when compared with WT cultures. This was consistent with a larger subpopulation of depolarizing GABA neurons (Figure S9C). Next, we more directly assessed the effect of glutamatergic blockers on hyperpolarizing- or depolarizing-GABA-responding subpopulations by calculating the MFR ratio following treatment with diazepam + vehicle or CNQX+AP5 (Figure S9D). We found that glutamate receptor blockers did not significantly change the percentage of electrodes that showed MFR increase vs. decrease upon diazepam (1 μM) in WT (control vehicle: 5% of MFR-increasing electrodes, 65% of MFR-decreasing electrodes; CNQX+AP5: 6% of MFR-increasing electrodes, 56% of MFR-decreasing electrodes) or Ts65Dn cultures (13% of MFR-increasing electrodes, 45% of MFR-decreasing electrodes or CNQX+AP5 28% of MFR-increasing electrodes, 45% of MFR-decreasing electrodes; Figures S9D–S9F). This indicated that glutamatergic activity did not significantly affect the overall proportion of the subpopulations of neurons with hyperpolarizing and depolarizing GABA response in WT or Ts65Dn cultures.

Overall, these results demonstrated the presence of heterogeneous subpopulations of neurons characterized by differences in the polarity of GABA_AR-mediated responses in both WT and Ts65Dn cultures. In particular, Ts65Dn neuronal cultures showed a larger subpopulation of cells characterized by GABA_AR-mediated depolarization, which depended on NKCC1 activity.

Subpopulations of neurons with functionally heterogeneous GABA_AR-mediated responses are present in hippocampal slices from adult WT and Ts65Dn mice recorded by MEAs

Because we found a significant percentage of neurons exhibiting depolarizing GABAergic responses also in neuronal cultures with a maturation process largely accomplished (at least according to the expression of developmental markers and electrophysiological parameters) at DIV15 and 21, we next assessed whether mixed subpopulations of neurons with heterogeneous GABA-mediated responses were also still present in adult animals ex vivo. To this aim, we prepared acute hippocampal brain slices (which preserve at least in part the original architecture of the brain circuits) from adult (10–12 weeks old) WT and Ts65Dn mice and recorded them by MEAs to isolate multiunit activity as mentioned earlier (Figure 3A).

Figure 3.

Mixed subpopulations of neurons with hyperpolarizing or depolarizing GABA signaling are present in acute hippocampal slices from adult WT and Ts65Dn mice

(A) Left: schematic representation of an acute brain slice of the hippocampus-entorhinal cortex (EC) region from an adult animal (postnatal day 70 [P70]). The slice is positioned on an MEA for electrophysiological recordings. Center: representative picture of the experimental setup with a 60-electrode MEA and an overlying hippocampus-EC slice. The gray square highlights the analyzed electrodes, which were positioned in the CA1 region. Right: schematic representation of the experimental protocol. Brain slices were first preincubated with vehicle (0.01% DMSO) or bumetanide (10 μM) for 45 min. Then, we recorded 30 min of spontaneous activity in the presence of vehicle or bumetanide followed by 30 min of recording 10 min after the addition of bicuculline (BIC; 20 μM) or GABA (100 μM).

(B, E) Representative high-pass filtered electrophysiological traces (>300 Hz) for WT and Ts65Dn slices pretreated with vehicle or bumetanide and recorded before and after bath application of BIC (B) or GABA (E). Scale bars: (B) 80 μV, 2 s; (E) 30 μV, 2 s.

(C, F) Quantification of the neuronal mean firing rate (MFR) ratio from WT and Ts65Dn slices upon BIC (C) or GABA (F) bath application. The MFR ratio over the baseline firing (dotted line) was calculated for each electrode and then averaged for each MEA. An MFR ratio higher than 1 (representing the baseline level) indicates an increase in activity upon the treatment, whereas a ratio below 1 indicates a decrease in activity. In the boxplot, the small square indicates the mean, the central line illustrates the median, the box limits indicate the 25th and 75th percentiles, the whiskers represent the 5th to 95th percentiles, and each dot represents the MFR ratio for each recorded slice. The numbers in parentheses indicate the number of analyzed slices for each experimental group. ∗∗p < 0.01; ∗∗∗p < 0.001; Tukey’s post hoc test following two-way ANOVA.

(D, G) Quantification of the average percentage number ± SEM of MEA electrodes (in the same experiments in C and F), showing changes of at least 15% in the MFR after BIC (D) or GABA (G) administration in WT (blue) and Ts65Dn (pink) slices. ∗∗p < 0.01; ∗∗∗p < 0.001; Tukey’s post hoc test following two-way ANOVA on ranked transformed data.

In a first set of experiments, we inhibited endogenous GABAergic signaling by bath application of bicuculline (20 μM²⁴). In particular, after preincubation with vehicle (0.01% DMSO) or bumetanide (10 μM²⁴) for 45 min, we recorded 30 min of baseline data (at the level of the CA1 hippocampal region) followed by an additional 30 min of recording after bicuculline application (Figure 3A). Considering our aim to challenge the neuronal networks under saturating conditions, we chose to utilize prolonged GABAergic treatments (which might also have an impact on intracellular chloride concentration per se), because of the need of a relatively extended time period for the drug to permeate the tissue down to the underlying MEA. As expected, the average MFR ratio of both vehicle-treated and bumetanide-treated WT slices clearly showed a large increase in the firing rate activity upon bicuculline application (Figures 3B and 3C). Conversely, bicuculline administration to vehicle-treated Ts65Dn slices resulted in a clear decrease in the average MFR ratio in comparison to the basal conditions (Figures 3B and 3C). Remarkably, Ts65Dn slices pretreated with bumetanide showed a pronounced increase in the MFR ratio and reached the levels observed in WT slices (Figures 3B and 3C). This outcome is in line with a possibly large subpopulation of NKCC1-dependent GABA depolarizing neurons in Ts65Dn adult mice and with data obtained by cell-attached patch-clamp recordings in the literature.^24,25

Next, to specifically assess whether heterogeneous subpopulations of neurons with mixed responses to endogenous GABA were also present in adult WT and Ts65Dn brain slices, we quantified the percentage of electrodes showing significant changes in spiking activity upon bicuculline application in comparison to their baseline level. We found that the majority of the electrodes detected a significantly increased firing rate upon blocking GABA_ARs with bicuculline in WT slices (80% of MFR-increasing electrodes), irrespective of vehicle or bumetanide pretreatment (Figure 3D). Remarkably, we found a significant percentage of electrodes in WT slices that detected a decrease in the firing rate upon bicuculline treatment, reflecting the presence of a subpopulation of neurons with depolarizing GABA response (12% of MFR-decreasing electrodes). Bumetanide pretreatment caused a small positive shift in the number of electrodes with decreasing spiking activity, but the difference did not reach statistical significance (Figure 3D). Strikingly, in vehicle-treated Ts65Dn slices, almost all the active electrodes showed a decrease in the average MFR level upon bicuculline administration (0% of MFR-increasing electrodes, 97% of MFR-decreasing electrodes), indicating a very profound alteration of inhibitory GABAergic signaling. Notably, Ts65Dn slices pretreated with bumetanide displayed a high MFR at all the active electrodes, with the percentage of electrodes showing an increase in the MFR ratio comparable to that of WT slices (95% of MFR-increasing electrodes, 2% of MFR-decreasing electrodes), indicating a complete rescue of inhibitory GABAergic response (Figure 3D). We found no differences in the percentage of electrodes that did not detect a response to bicuculline among all experimental groups (Figures S10A and S10B).

Next, we also analyzed the firing activity variation upon application of exogenous GABA (100 μM²⁴) to brain slices of WT and Ts65Dn adult mice. As expected, vehicle-treated WT slices clearly showed a decrease of the average MFR (Figures 3E and 3F). Bumetanide pretreatment only led to a nonsignificant shift toward higher MFR ratio values (Figures 3E and 3F). Conversely, vehicle-treated Ts65Dn control slices showed an increased average MFR ratio upon GABA administration in comparison to that in WT slices (Figures 3E and 3F). Notably, pretreatment with bumetanide fully restored inhibitory GABAergic signaling in Ts65Dn slices, as indicated by the decrease in the average MFR ratio to levels similar to those of WT slices (Figures 3E and 3F).

Again, when we evaluated changes in the MFR variation upon GABA application at the single-electrode level, we found mixed subpopulations of GABA-responding neurons in both WT and Ts65Dn slices, with the difference between WT and Ts65Dn slices even more profound than that in neuronal cultures. Indeed, exogenously applied GABA caused a decrease in the firing rate activity at most (but not all) of the electrodes in vehicle-treated WT slices (9% of MFR-increasing electrodes, 82% of MFR-decreasing; Figure 3G). Notably, vehicle-treated Ts65Dn slices displayed a pronounced mixed response to GABA treatment with 32% of electrodes that showed a decrease in the MFR and 57% of those that showed increased MFR activity (Figure 3G). Notably, Ts65Dn slices pretreated with bumetanide showed an evident negative shift in the percentage of electrodes that detected significantly changed responses. Indeed, the MFR at most of the electrodes decreased upon GABA application (20% of MFR-increasing electrodes and 72% of MFR-decreasing electrodes) similar to the WT slices and indicative of a large recovery of physiological proportions between the neuronal subpopulations with hyperpolarizing or depolarizing GABA response. We found no difference in the percentage of electrodes that did not respond to GABA in any of the experimental groups (Figure S10C).

Of note, we also found subpopulations of mature neurons with heterogeneous GABA_AR-mediated responses in another strain of WT animals (C57BL/6J; Figures S11A–S11D). However, when we evaluated a second model of DS (the Dp(16)1Yey/+ mice²⁵) in the same background (C57BL/6J), we found no difference in the subpopulations of mature neurons with heterogeneous GABA_AR-mediated responses in comparison to their corresponding WT littermates. Indeed, slices from adult Dp(16)1Yey/+ mice showed a mixed subpopulation of neurons responding to GABA with increased or decreased MFRs to levels very similar to those of their corresponding WT littermates (Figures S11A–S11D). Therefore, the increased subpopulation of neurons with depolarizing-GABA_A-mediated responses that we identified in adult Ts65Dn animals was not replicated in a second mouse model of DS. Nevertheless, hippocampal NKCC1 levels in adult Dp(16)1Yey/+ animals were similar to those of WT animals, further strengthening the NKCC1 dependency of depolarizing GABAergic responses in Ts65Dn mice (Figures S11F–S11G).

Overall, these results demonstrated the presence of heterogeneous subpopulations of neurons characterized by differences in GABA_AR- and NKCC1-mediated responses in ex vivo slices from adult WT and Ts65Dn mice that appeared even larger than those in mature neuronal cultures.

Subpopulations of neurons with heterogeneous GABA_AR-mediated responses recorded by in vivo Ca²⁺ imaging are present in adult freely moving WT and Ts65Dn mice

Next, to further strengthen our results, we performed Ca²⁺ imaging experiments in a network of neurons in vivo. For these experiments, we chose Ca²⁺ imaging because it allows the activity of a large population of neurons to be recorded with a high degree of spatial resolution in freely moving mice.³⁴ Moreover, we chose to modulate GABAergic activity in vivo with the positive GABA_AR allosteric modulator diazepam (instead of GABA) because it is an FDA-approved drug with translational validity, has recognized behavioral outcomes, and is known to readily pass the blood-brain barrier. We also discarded the usage of GABA_AR blockers (which we used in fact in culture and slice experiments), as we reasoned that they could potentially induce epileptic activity in vivo. This would have strongly complicated the analysis of Ca²⁺ signals and the interpretation of the results especially in the case of DS animals, which have higher susceptibility to seizures.^24,35

To perform Ca²⁺ imaging experiments in the hippocampus in vivo, we implanted a microendoscopic probe coupled to a miniaturized head-mounted microscope in freely moving, adult (3–4 months old) WT and Ts65Dn mice previously injected in the hippocampus with an adeno-associated virus (AAV) to express the Ca²⁺ indicator GCaMP6f in dorsal CA1 pyramidal neurons under the control of the neuronal CamK2a promoter³⁶ (Figure 4A). We set up a crossover study in which the same animals were first imaged after administration of vehicle (2% DMSO in saline, i.p.) followed by diazepam (2 mg/kg, i.p.); after two weeks, imaging was performed again upon administration of bumetanide (0.2 mg/kg, i.p.) followed by diazepam (Figure 4B). This experimental design allowed us to longitudinally follow the same neurons across the two experimental sessions, evaluating the changes in activity induced by the two pharmacological manipulations (diazepam and bumetanide) in each cell and thus comparing the effect of diazepam with and without bumetanide pretreatment at the single-cell level. Finally, to avoid potential photobleaching and/or phototoxicity during imaging, we adopted an intermittent imaging protocol³⁷ (Figure 4B).

Figure 4.

Bumetanide treatment differentially affects neuronal activity in vivo in the hippocampus of freely moving adult WT and Ts65Dn mice and partially rescues aberrant GABA_AR signaling in Ts65Dn animals

(A) Schematic representation of the experimental setup. Left: two- to three-month-old WT and Ts65Dn mice received a stereotaxic injection in the dorsal hippocampal CA1 region with AAV viruses expressing the Ca²⁺-sensor GCaMP6f under the control of the CamK2a promoter. Ts65Dn and WT mice were then implanted with a microendoscopic probe, and neuronal activity was longitudinally assessed by recording in vivo Ca²⁺ events within the CA1 hippocampal region of freely moving mice with a miniaturized head-mounted microscope. Right: examples of projection maps of recorded neurons in a WT and a Ts65Dn adult animal.

(B) Schematic representation of the experimental protocol timeline. Ca²⁺ events were imaged in the same neuron in two consecutive sessions before and after administration of the GABA_AR-positive allosteric modulator diazepam (2 mg/kg), following a subchronic (2 weeks) treatment with either vehicle (2% DMSO in saline) or bumetanide (0.2 mg/kg). Within each imaging session, neuronal activity was recorded in periods of 5 min (“ON”, red) alternated to 5 min of rest (“OFF”, blue) to collect 15 min of neuronal data for each session while avoiding phototoxicity.

(C) Quantification of single neuron mean event rates (MERs) of WT and Ts65Dn mice before and after diazepam administration in experiments as in (B). In the boxplot, the small square indicates the mean, the central line illustrates the median, the box limits indicate the 25th and 75th percentiles, and whiskers represent the 5th to 95th percentiles. Each dot indicates the binned value of the MER (bin size = 0.005 events/sec) obtained from individual neurons (from 5 WT to 8 Ts65Dn adult mice). ∗p < 0.05; ∗∗∗p < 0.001; Tukey’s post hoc test following a linear mixed effects model (with two factors).

(D) Scatterplots showing the MER for each active (MER>0.01 events/s in vehicle) neuron (plotted as a dot) in (C) before (x axis) and after (y axis) administration of diazepam. Dark gray dots represent neurons showing a significant increase in the MER. Light gray dots represent neurons showing a significant decrease in the MER. Black dots represent neurons showing no significant changes in the MER. Significant changes in the MER (number with arrows) for each neuron upon diazepam application were evaluated by bootstrap analysis.

(E) Quantification of the percentage of neurons (in the same experiments in C), showing significant changes in the MER by bootstrap analysis after diazepam administration in comparison to their basal conditions in WT (blue) and Ts65Dn (pink) mice. ∗p < 0.05, ∗∗∗p < 0.001; chi-squared test with Sidak adjustment for multiple comparisons.

As expected, the mean Ca²⁺ event rate (MER) in adult WT mice significantly decreased upon diazepam administration after pretreatment with vehicle. Notably, bumetanide pretreatment led to a further significant decrease in MER upon diazepam treatment in adult WT mice, suggesting the presence of a subpopulation of NKCC1-dependent GABA-depolarizing neurons also in vivo in freely moving animals. In agreement with the cell culture and brain slice data, bumetanide application per se did not cause any significant variation in the MER of WT animals (Figure 4C). Interestingly, Ts65Dn neurons showed a significant mean MER increase upon diazepam treatment. This indicates that decrease of GABA_AR-dependent inhibition and possibly the depolarizing action of GABA also occurs in vivo in adult DS mice, as already reported in vitro.^24,38,39 Remarkably, bumetanide pretreatment reduced the baseline MER and prevented the diazepam-induced MER increase in Ts65Dn animals (Figure 4C), indicating an NKCC1 dependency in the aberrant response to diazepam of adult DS mice.

Next, we specifically investigated the presence of subpopulations of neurons with heterogeneous GABA responses also in freely moving adult animals by bootstrap methods to compute the percentage of neurons that significantly changed their activity by diazepam treatment. The majority of responding neurons of vehicle-treated WT animals showed significantly lowered frequencies of Ca²⁺ transients following diazepam administration (40% MFR-decreasing neurons), indicating the mainly inhibitory action of GABA (Figures S4D and S4E; calcium traces and raster plots in Figures S12A–S12D). Notably, a small neuronal subset (5%) in vehicle-treated WT animals had a significantly increased MER, reflecting the presence of a subpopulation of pyramidal CA1 neurons showing depolarizing responses to GABA_AR signaling also in vivo in WT animals. However, a remarkable 55% of WT neurons did not show a significant change in response to diazepam in vivo (Figure S12E). This indicates a general low inhibitory chloride driving force in neurons in vivo and possibly the existence of a third substantial subpopulation of neurons characterized by no responses to diazepam treatment to be considered in vivo studies. Most importantly, bumetanide pretreatment in WT animals caused a significant enhancement of inhibitory GABA_AR signaling in vivo, as shown by the higher percentage of neurons that had a decreased MER (61% MFR-decreasing neurons; Figures 4D, 4E, and S12A–S12D) and the clear significant decrease in nonresponding neurons (Figure S12E).

Vehicle-treated Ts65Dn animals displayed a much wider dispersion of single MER values in vivo than WT animals (Figures 4D and 4E). Indeed, only 18% of neurons showed a decreased MER, and 30% of neurons showed an increased MER upon diazepam treatment, indicating a strong reduction in the efficacy of GABA_AR-dependent inhibition and even paradoxical GABA_AR-driven depolarizing activity (Figures 4D, 4E, S12B, and S12C). Notably, Ts65Dn mice pretreated with bumetanide showed a significantly lower percentage of neurons that increased the MER upon treatment with diazepam (22% of MER-increasing neurons, Figures 4D, 4E, S12B, and S12C) and an increased number of nonresponding neurons (60% of MER-nonchanging neurons, Figure S12E), indicating that blocking NKCC1 activity rescues GABA_AR-dependent inhibition in a subpopulation of Ts65Dn pyramidal neurons in vivo.

Overall, these results demonstrated the presence of heterogeneous subpopulations of hyperpolarizing- or depolarizing-GABA_AR-responding neurons in adult WT and Ts65Dn animals even in vivo, with proportions comparable to ex vivo data, but with much higher percentages of GABA_AR nonresponding neurons.

Heterogeneous subpopulations of neurons with distinct responses to bumetanide or diazepam are present in WT and Ts65Dn mice in vivo

Because bumetanide pretreatment reduced the average baseline neuronal Ca²⁺ MER and prevented, on average, the diazepam-induced neuronal MER increase in adult Ts65Dn animals, we next investigated these subpopulations of bumetanide-responsive cells in more detail. Conveniently, our longitudinal recordings allowed us to follow the Ca²⁺ activity in the same neuron throughout the bumetanide and diazepam treatments. Therefore, we first investigated whether the neurons that showed a decrease in MER upon bumetanide administration were also resistant to diazepam-induced stimulation, demonstrating that increased NKCC1 activity mediates the paradoxical effects of diazepam in mature Ts65Dn neurons. To this aim, we calculated the variation in neuronal Ca²⁺ activity induced by bumetanide over vehicle treatment (Ratio 1, Figure 5A) and the variation caused by diazepam plus bumetanide treatment over diazepam alone (Ratio 2, Figure 5A) for each Ts65Dn neuron. To evaluate possible relationships between changes in neuronal activity induced by the two treatments, we plotted the value of Ratio 1 vs. Ratio 2 (i.e., the variation in neuronal activity induced by bumetanide vs. diazepam treatments) for each neuron and divided the obtained scatterplot into 4 different quadrants, reflecting the different possible neuronal categories determined by changes in activity induced by the two drugs (Figure 5B). Specifically, cells showing a decrease in MER after both bumetanide and bumetanide plus diazepam treatments were classified into category 1; cells showing an increase in MER after bumetanide and a decrease after bumetanide plus diazepam administrations were classified into category 2; cells showing an increase in MER after both bumetanide and bumetanide plus diazepam administrations were classified into category 3; and cells showing a decrease in MER after bumetanide and an increase after bumetanide plus diazepam administration were classified into category 4 (Figure 5B). We found a significant positive correlation (p < 0.001, Spearman test; Figure 5C) for the activity of Ts65Dn neurons upon bumetanide and diazepam application. Notably, the majority of cells were allocated to category 1, thus implying the presence of a bumetanide-responsive subpopulation of Ts65Dn neurons in which NKCC1 inhibition can ameliorate the paradoxical effects induced by diazepam treatment (Figure 5D).

Figure 5.

Mixed subpopulations of neurons with hyperpolarizing or depolarizing GABA_AR signaling respond differently to bumetanide administration in vivo

(A) Schematic representation of the experimental protocol timeline (top) together with representation of the analysis approach (bottom).

(B) Schematic representation of the possible distribution of neuronal subpopulations based on MER variation in response to bumetanide or diazepam treatment (Ratio 1 vs. Ratio 2, see main text). Cells showing MER ratio changes below 10% were excluded from the analysis (shadowed gray area). The numbers in the black dots indicate the four quadrants representing the possible different responses (increasing or decreasing arrows) of neuronal activity upon diazepam or bumetanide treatment over that elicited after vehicle treatment.

(C) Scatterplot showing the neuronal population distribution based on MER variation in response to bumetanide or diazepam treatment for Ts65Dn neurons in the same experiments described in Figure 4. Pearson correlation showed a significant positive correlation between the responses to bumetanide or diazepam treatment in neurons from Ts65Dn mice.

(D) Quantification of the percentage of neurons for each quadrant on the same dataset described in (C). ∗∗p < 0.01; chi-squared test with Sidak adjustment for multiple comparisons.

(E) Schematic representation of the support vector machine (SVM) classifier used for discrimination between the two classes of variation (i.e., decrease [<1] or increase [>1] of activity after bumetanide plus diazepam when compared with vehicle plus diazepam).

(F) Left: percentage of Ts65Dn neurons used as a training set or a test set for the SVM classification. Right: the upper pie chart reports the percentage of neurons that fell into the two categories (>1 and <1) for the training set. The lower pie chart reports the percentage of neurons that fell in the two categories (>1 and <1) after running the SVM classification on the test set.

(G) Left: confusion matrix showing the performance accuracy (correct green and incorrect white) of the SVM classifier in Ts65Dn neurons based on the two subpopulations of diazepam-responding neurons (<1 decrease upon diazepam plus bumetanide or >1 increase upon bumetanide plus diazepam when compared with diazepam plus vehicle). Right: percentage of neurons in the test set correctly (green boxes) or incorrectly (white boxes) classified into the two categories (<1 or >1).

To further strengthen this finding, we used machine-learning algorithms (support vector machine classifier, SMV; Figure 5E) to try to predict the diazepam response elicited in each Ts65Dn cell based on the knowledge of the previous response caused by bumetanide administration to the same cell (i.e., to understand whether the response to bumetanide could predict the future response to diazepam in terms of variation of Ca²⁺ MER). For simplicity, we used only two different categories that referred to either an increase (>1) or a decrease (<1) in the MER during diazepam plus bumetanide treatment when compared with diazepam alone (i.e., changes in Ratio 2; Figure 5E). First, we split the Ca²⁺ MER in vivo dataset into one training and one test set, with percentages of neurons equal to 54% and 46%, respectively, and with a balanced representation of the categories (i.e., Ratio 2 > 1 and Ratio 2 < 1; Figure 5F). Next, we used the training set to feed the support vector machine (SVM) for the classification of neurons in the test set. When we evaluated the performance accuracy of the predictive model using a confusion matrix⁴⁰ on the test set, we found that 71 out of the 108 total neurons belonging to the <1 category and 47 out of 77 total neurons belonging to the >1 category were correctly classified into the two categories (Figure 5G, left). Conversely, 37 neurons were wrongly discriminated as the >1 category instead of the <1 category, and 30 neurons were assigned to the <1 category instead of the >1 category (Figure 5G, left). By computing the percentage of neurons per category, we successfully discriminated 66% of neurons in the <1 category and 61% of neurons in the >1 category (Figure 5G, right; Table S2).

Taken together, these findings suggested that different levels of NKCC1 activity—probed by diverse responses to bumetanide—are possibly responsible for the paradoxical effect of diazepam on neuronal activity in mature Ts65Dn cells in vivo.

Diazepam treatment exerts abnormal behavioral effects in adult Ts65Dn mice, which are rescued by bumetanide treatment

Next, we investigated the functional implications of our findings at the behavioral level. To this aim, we first treated adult (2–3 months old) WT and Ts65Dn mice with vehicle (2% DMSO in saline, i.p.) or bumetanide (0.2 mg/kg, i.p.)^24,41 followed by saline (0.9% NaCl, i.p.) or diazepam (2 mg/kg, i.p.) administration 30 min later and assessed anxiety behaviors and the anxiolytic effects of diazepam in three different behavioral tests after an additional 30 min (Figure 6A). In the dark-light test, vehicle plus saline-treated Ts65Dn mice showed a reduction in the time spent in the illuminated (i.e., the potentially unsafe) area of the apparatus compared with WT mice, indicative of increased anxiety (Figure 6B). Notably, this behavior was fully rescued by bumetanide treatment, indicating a possible role for NKCC1 in the altered behavior of adult Ts65Dn mice in this test. We also found a trend toward increased time in the light zone in WT animals upon bumetanide treatment, but this did not reach statistical significance. Furthermore, in agreement with the larger subpopulation of cells responding with increased activity to diazepam that we found in imaging experiments in adult Ts65Dn mice, treatment with diazepam actually worsened the anxiety behavior of Ts65Dn mice rather than rescuing it. Interestingly, pretreatment with bumetanide completely rescued anxiety behavior in Ts65Dn animals even after diazepam administration (Figure 6B). We found no differences in the total number of transitions between the dark and light zones for WT and Ts65Dn mice (Figure 6C), indicating that general activity was not altered by drug treatments.

Figure 6.

Bumetanide treatment rescues aberrant behavioral responses to the benzodiazepine diazepam in adult Ts65Dn mice

(A) Schematic representation of the protocol for the diverse experimental groups. Adult WT and Ts65Dn mice were treated with bumetanide (0.2 mg/kg i.p.) or the corresponding vehicle (2% DMSO in saline) for two weeks and then assessed in two different anxiety tests. On the day of testing, mice were pretreated with bumetanide or vehicle and then tested 30 min after diazepam (2 mg/kg i.p.) or saline administration.

(B) Quantification of time spent in the light zone of the dark-light test for WT and Ts65Dn mice.

(C) Quantification of the total number of transitions between the two zones of the dark-light test for WT and Ts65Dn mice.

(D) Quantification of the time spent in the center of the arena during the open-field test for WT and Ts65Dn mice.

(E) Quantification of the total distance traveled in the open field test for WT and Ts65Dn mice. In all boxplots, the small square indicates the mean, the central line illustrates the median, the box limits indicate the 25th and 75th percentiles, whiskers represent the 5th to 95th percentiles, and each dot indicates a value obtained from individual animals. ∗∗∗p < 0.001; Tukey’s post hoc test following two-way ANOVA or two-way ANOVA on ranked transformed data.

Next, we evaluated the effect of diazepam treatment in the open-field test, another behavioral task widely used to assess anxiety behavior in mice. We found that diazepam administration in adult WT animals strongly increased the time spent in the central portion (i.e., the potentially unsafe area) of the open-field arena, indicative of the well-known anxiolytic effect of the drug. Saline-treated adult Ts65Dn mice showed a trend toward increased anxiety compared with WT mice in this test (slight decrease in time spent in the center of the arena). Additionally, diazepam administration showed a trend toward increased anxiety in Ts65Dn mice (Figure 6D). Nevertheless, none of these two trends reached statistical significance (Figure 6D). Pretreatment with bumetanide had a very large effect and fully rescued the anxiolytic effect of diazepam in adult Ts65Dn mice to WT levels, consistent with the results of the dark-light test. We found no differences in the total distance traveled for either WT or Ts65Dn mice (Figure 6E), again indicating no potentially confounding effects of the drugs on locomotor activity in this test.

In line with the recent literature,⁴² we also found a substantial lack of effect of diazepam administration in adult Ts65Dn mice in the elevated plus maze test, as indicated by the similar percentage of time spent in the closed arms after saline or diazepam administration (Figures S13A and S13B). Conversely, diazepam-treated WT mice showed a decrease in the percentage of time spent in the closed arms, reflecting the expected anxiolytic effect of the drug. Notably, in agreement with the results from the open-field test, pretreatment of adult Ts65Dn mice with bumetanide significantly decreased the percentage of time spent in the closed arms and simultaneously increased the percentage of time spent in the open arms (Figure S13B). Thus, bumetanide pretreatment was able to restore the anxiolytic effect of diazepam in Ts65Dn animals also in the elevated plus maze test. Although diazepam administration caused a significant reduction in the average total number of arm entries for both WT and Ts65Dn mice (Figure S13C), this outcome was not paralleled by a decrease in the total distance traveled (Figure S13D), indicating that the other effects observed upon diazepam treatment did not likely depend on decreased locomotor activity.

Together, these data revealed that diazepam treatment induced atypical responses, ranging from a paradoxical anxiogenic effect to a lack of efficacy in Ts65Dn mice compared with that in WT adult mice in diverse behavioral tasks. Most importantly, bumetanide treatment fully reestablished the anxiolytic effect of diazepam in all behavioral tasks in DS animals.

Mixed subpopulations of neurons with depolarizing or hyperpolarizing GABA signaling are present in human iPSC-derived neurons from control and trisomic cell lines from a person with mosaic DS

To further strengthen and increase the translational value of our findings, we investigated whether a heterogeneous population of neurons with mixed responses to GABA_AR signaling was also present in human cells. To this aim, we evaluated iPSC-derived neurons from a previously established DS line (DS4) and its corresponding isogenic control cell line (DS2U), both cell lines derived from the same person with a mosaic form of DS.⁴³ In particular, we aimed at evaluating Ca²⁺ responses elicited by bath application of GABA (100 μM) in human DS and control neurons. Because full maturation of human neurons in vitro is a very lengthy process requiring months, we first referred to the literature to focus our analysis on the earliest time point in which the maturation of GABA_AR reversal potential in iPSC-derived neurons was described to be completed.^22,44 We found that time point to be between 50 and 90 days from final neuronal differentiation in control neurons. To refine our choice of timing for our specific experimental activities on IPSC-derived neurons, we first assessed the expression level of NKCC1 and KCC2 at 30, 45, and 60 days from final neuronal differentiation in our experimental conditions. Our findings indicated that NKCC1 and KCC2 expression levels reached a plateau by 60 days in both control and DS neurons (Figures S14A–S14C). Moreover, we also based our choice on markers of neuronal development (DCX, NeuN, Tubβ3). In particular, we focused on the sharp decrease that we found in the number of DCX- and Tubβ3-double-positive cells in our control and DS neuronal cultures at 60 days after neuronal plating, which reached levels not diverse from 0 (one sample t test vs. zero mean, control p = 0.153 and DS p = 0.360; Figures S14D,S14E, S15A, and S15B). Indeed, in mouse neuronal cultures, only the decrease that we found in DCX- and Tubβ3-double-positive cells correlated with the end of the excitatory-to-inhibitory developmental switch in the polarity of GABA_AR signaling, whereas the increase in the number of DCX- and NeuN- and Tubβ3-triple-positive or NeuN- and Tubβ3-double-positive cells did not correlate (compare Figure 1B with Figure S2F). This indicated levels of DCX- and Tubβ3-double-positive cells close to 0 as the most appropriate maturation marker for the end of the excitatory-to-inhibitory developmental switch in the polarity of GABA_AR signaling. Furthermore, 60 days after final neuronal differentiation also coincided with the timing when the neurons' capacity to generate action potentials in response to current injections (10 pA for 500 msec step) reached levels comparable to those of rodent neurons (Figures S14F and S14G), and a plateau in the development of most passive membrane properties (i.e., resting membrane potential, input resistance, and rheobase) was reached in both control and DS neurons (Table S3). Indeed, our results at 60 days in vitro (DIV) align closely with existing literature.^22,45 Moreover, at 60 days after final neuronal differentiation, neurons exhibited functional synapse with DS neurons showing decreased number of excitatory postsynaptic events as already described in the literature⁴⁶ (Figure S14H). Finally, 60 days after neuronal differentiation was also a time point when the magnitude of the differences in terms of NKCC1 and KCC2 expression levels, neuronal markers, and action potential firing between DS and control neurons reached a plateau.

Thus, we calculated the percentage of cells showing GABA-induced Ca²⁺ responses in iPSC-derived control and DS neurons at 60 days after final neuronal differentiation. We found that a subpopulation of neurons (14% of cells) presented depolarizing GABA responses also in human control cell cultures (Figures 7A and 7B). This minor subpopulation was significantly larger in DS neuronal cultures at the same developmental stage (62% of GABA-responding cells). Interestingly, in agreement with all the experiments in mouse neurons, treatment with bumetanide (10 μM) significantly reduced the percentage of neurons showing GABA-induced Ca²⁺ responses in DS cultures without significantly altering the control neurons (Figures 7A and 7B). These results were corroborated by the significantly higher levels of NKCC1 expression (and no significant difference in KCC2 expression) that we found in DS neurons in comparison to control neurons (Figures 7C–7E).

Figure 7.

Mixed subpopulations of neurons with depolarizing or hyperpolarizing GABA signaling are present in human isogenic control and trisomic iPSC-derived neurons obtained from a person with DS

(A) We analyzed trisomic neurons and corresponding isogenic control (euploid) neurons derived from iPSCs obtained from an individual with a mosaic form of DS. Representative calcium traces of iPSC-derived control and trisomic (DS) neurons pretreated with vehicle (0.01% DMSO) or the NKCC1 inhibitor bumetanide (10 μM) upon bath application of GABA (100 μM) at 60 days following plating for final differentiation. ΔF/F0 represents the change in the calcium intensity expressed as variation in fluorescence intensity upon GABA application, relative to the basal fluorescence intensity.

(B) Quantification of the percentage of neurons showing depolarizing GABA responses in experiments as in (A). In the boxplot, the small square indicates the mean, the central line illustrates the median, the box limits indicate the 25th and 75th percentiles, the whiskers represent the 5th to 95th percentiles, and each dot indicates a value obtained from an individual coverslip. The numbers in parentheses indicate the number of analyzed coverslips for each experimental group (obtained from 5 independent neuronal differentiation experiments). ∗p < 0.05, ∗∗∗p < 0.001; Tukey post hoc test following two-way ANOVA.

(C) Representative immunoblots for NKCC1 and KCC2 protein extracts from control and trisomic (DS) neurons. Actin was used as an internal standard. Full blots are shown in Figure S16A.

(D) Quantification of average NKCC1 protein levels (±SEM; expressed as the percentage of control neurons, dotted line) in same experiments as in (C). Actin was used as an internal standard. Dots indicate values of individual coverslips (obtained from 2 neuronal independent differentiation experiments). ∗∗p < 0.01; Student’s t test.

(E) Quantification of average KCC2 protein levels (±SEM) in the same experiments described in (D). Dots indicate values of individual coverslips (obtained from 2 neuronal independent differentiation experiments).

Overall, these results demonstrated the presence of a subpopulation of depolarizing-GABA_AR-responding neurons in both human control and DS cultures, with DS cultures presenting a larger NKCC1-dependent subpopulation.

Discussion

The current mainstream view regarding the regulation of mature neuronal network activity in the brain suggests that a small number of inhibitory GABAergic neurons dampens the firing of the large majority of excitatory glutamatergic neurons by hyperpolarizing postsynaptic cells.⁴⁷ However, in immature neurons, GABAergic signaling is actually depolarizing and mostly excitatory. The polarity of GABAergic signaling switches to inhibitory during brain development due to the decrease in [Cl⁻]_i by upregulation of the expression of the Cl⁻ exporter KCC2.¹⁹ Interestingly, experimental and computational simulations have nevertheless shown that even in mature neurons with already fully inhibitory GABA_AR-mediated signaling, small changes in E_GABA can potentially shift GABA inhibition into excitation.^8,48 Therefore, we reasoned that it is possible that the GABA polarity switch may never be fully accomplished in the entire population of neurons or that a small percentage of neurons may keep oscillating between depolarizing and hyperpolarizing GABA signaling even when they become fully mature. Here, we experimentally indicated that a small subpopulation of WT neurons showed excitatory GABAergic signaling and that this subpopulation was intermingled with the larger majority of neurons with inhibitory GABA signaling both in vitro and in vivo. This mixed subpopulation of GABA-responding neurons started to appear during development when the proportion of neurons with hyperpolarizing vs. depolarizing GABAergic signaling increased progressively, yet without reaching wholeness. However, only tracking of each individual neuron over time will reveal whether the small subpopulation of mature neurons with depolarizing GABA_AR signaling is derived from neurons that never fully mature or neurons with peculiar proprieties in regulating [Cl⁻]_i. In fact, (at least) some neurons appear to have significant cellular microdomains of diverse intracellular Cl⁻ concentration (i.e., mosaicism of GABAergic signaling in the axon and axonal initial segment, in the soma, and along dendrites). Although this heterogeneity of GABA_AR-mediated responses at the level of cellular compartments is well accepted,^{14,15,16,17,18} the presence of a heterogeneous subpopulation of neurons with depolarizing vs. hyperpolarizing GABA responses in mature neuronal networks has been, at times, hinted by some scant literature (Table 1), but it is not yet a clear and well-established concept in the neuroscience field.¹⁸ This is in spite of the fact that GABA modulation of brain activity in adult neuronal networks is among the most studied topics in neuroscience.

Table 1.

Mixed subpopulations of GABA-responding neurons at an increasing scale of complexity has already been described in the literature

Model	Reference	Structure	DIV/Age	GABAergic agonist/antagonist (concentration)	Technique	% of neurons with depolarizing GABA	Directly reported in the text
NEURONAL CULTURES	Fiumelli et al., Neuron, 200549	Hippocampal neurons	15 DIV	//	Gramicidin-perforated patch clamp	15%	No
	Serratto et al., Molecular Neurobiol., 202050	Hippocampal neurons	18 DIV	GABA (100 μM)	GABAAR single-channel currents	25%	No
	Parrini et al., Molecular Therapy, 202138	Hippocampal neurons	16–20 DIV	BIC (20 μM)	Cell-attached patch clamp	23%	Yes
	Baltz et al., Frontiers in cellular neuroscience, 201051	Lateral cortical neurons	30 DIV	Muscimol (200 μM)	Calcium imaging	45%	Yes
	Sun et al. Molecular Brain 201352	Cortical neurons	12 or 16–21 DIV	GABA (100 μM)		8% (12 DIV) or 2% (16–21 DIV)	Yes
	Lysenko et al., Neurobiology of Disease, 201853	Hippocampal neurons	13 DIV	GABA (100 μM)		10%	No
ACUTE BRAIN SLICES	Riekki et al., Journal of Neurophysiology 200854	CA1 pyramidal neurons	P30	//	Gramicidin-perforated patch clamp	33%	No
	MacKenzie et al., Epilepsy Research, 201555	CA1 pyramidal neurons	8–10 weeks	GABA (5 μM)		33%	Yes
	Deidda et al., Nature Medicine, 201524	CA1 pyramidal neurons	10–16 weeks	GABA (100 μM)		11%	No
	Dargaei et al., PNAS, 201856	CA1 pyramidal neurons	4–5 weeks	GABA (100 μM)		7%	No
	Kim et al., Science Advances, 202157	ILC pyramidal neurons	5 weeks	//		8%	No
	Alfonsa et al., Nature Neuroscience, 20233	Cortical pyramidal neurons (L5)	4–12 weeks	GABA (100 μM)		32%a	Yes
	Deidda et al., Nature Medicine, 201524	CA1 pyramidal neurons	10–16 weeks	GABA (100 μM)	Cell-attached patch clamp	25%	No
IN VIVO	Sulis Sato et al., PNAS, 201758	Cortical neurons	P18–51	//	Chloride imaging	13%	No
	Rahmati et al., J Neuroscience 202159	Layer II/III cortical neurons	P20	//		40%	No
	Berdyyev et al., PLOS ONE, 201437	CA1 hippocampal neurons	8–12 weeks	Zolpidem (10 mg/kg)	Calcium imaging	3%	Yes
iPSC derived-hNEURONS	Tang et al., Science Translational Medicine, 201922	iPSC neurons	2–3 months	GABA (100 μM)	Gramicidin-perforated patch clamp	16%	Excel spread sheet with data in Figure S4B
	Paavilainen et al., Stem Cell Research, 201860	hESC line	4 weeks	GABA (100 μM)	Calcium imaging	24%	Yes
	Mäkinen et al., Frontiers in Neuroscience, 201861	iPSC neurons	4 weeks	GABA (100 μM)		Bimodal distribution	No

RMP, resting membrane potential, reference value of −65 mV; DIV, days in vitro; P, postnatal day; iPSC, induced pluripotent stem cells; hESC, human embryonic stem cells; hneurons, human neurons; ILC, infralimbic cortex.

^aNote: 5% exhibiting action potentials after GABA puffing (100 μM).

Our data and careful analysis of the current literature (Table 1 and Table S4) in fact clearly point to the presence of a subpopulations of neurons with hyperpolarizing and inhibitory or depolarizing and possibly excitatory GABAergic signaling in also mature networks. In particular, in rodent WT primary neuronal cultures, both Ca²⁺ imaging experiments and gramicidin-perforated patch-clamp recordings in the literature reported a percentage of neurons with depolarizing responses to GABA_AR stimulation even at advanced (15–30 DIV) stages of development (Ca²⁺ imaging: 2%–45% of GABA depolarizing neurons^51,52; gramicidin-perforated patch-clamp recordings: 15% of GABA depolarizing neurons⁴⁹; GABA_AR single-channel currents: 25%⁵⁰; cell-attached patch-clamp recordings: 23%³⁸ of GABA depolarizing neurons). Notably, such findings were replicated also in iPSC-derived human neurons at advanced stages of differentiation (4–12 weeks) (Ca²⁺ imaging: 24% of GABA depolarizing neurons^60,61; gramicidin-perforated patch-clamp recordings: 16% of GABA depolarizing neurons²²). Interestingly, such heterogeneity in the polarity of GABA_AR-mediated responses was not limited to neurons in culture, which may indeed never reach a full and complete maturational stage, but also in acute brain slices from adult mice (gramicidin-perforated patch-clamp: 7%–33%^{3; 24; 54,55,56,57}; cell-attached patch-clamp recordings: 25%²⁴). Most notably, in vivo evidence also pointed to a subpopulation of neurons with depolarizing GABAergic responses in juvenile animals after the GABA developmental polarity switch and in fully adult animals (overall age range P18–51 years; chloride imaging: 13%–40% of neurons^58,59; Ca²⁺ imaging: 3% of neurons³⁷). On the other hand, our review of the literature also highlighted that some authors did not report the same heterogeneity in GABA_AR-mediated responses (Table S4). This may simply be due to the different experimental conditions or analysis. It is plausible that in certain experimental designs permissive of a small number of recorded cells, the very low percentage of cells with depolarizing GABA responses may have been missed or may have been interpreted as outliers in the statistical analysis. On the other hand, also in our Ca ²⁺ imaging or MEA experiments, we may have missed some GABA-driven depolarization events subthreshold for opening voltage-gated Ca²⁺ channels or to remove the magnesium block from NMDA receptors.

Our results were strengthened by the fact that we observed the presence of a small percentage of neurons with mixed responses to GABA at three diverse layers of neuronal network complexity ranging from in vitro (primary mouse cultures and human iPSC-derived neurons) to ex vivo (mouse acute brain slices) and in vivo (freely moving mice; summarized in Tables S5 and S6) and used multiple experimental techniques, including Ca²⁺ imaging, chloride imaging, and MEA recordings for our experiments.

Interestingly, the size of the subpopulation of neurons with depolarizing and excitatory GABA responses increased from cell cultures to brain slices. On the other hand, we found a smaller number of GABA-depolarizing neurons in vivo than what we observed at other levels of complexity. This outcome might be explained in part by the fact that to unmask the subpopulation of neurons with excitatory GABA responses, we used diverse pharmacological approaches (GABA and bicuculline application in vitro and diazepam application in vivo) or due to the heavy pharmacokinetic and pharmacodynamic processing in vivo compared with in vitro. Indeed, whereas GABA directly activates GABA_ARs causing complete suppression of firing activity in WT neurons, diazepam is only a positive modulator of GABA_ARs, thus leading to a smaller reduction in neuronal activity,^62,63 as revealed also by our own data in cell cultures. Moreover, in vitro culturing or brain slicing procedures may affect chloride homeostasis per se.⁶⁴ Finally, it is also possible that the much higher number of cells that were nonresponsive to diazepam in vivo (compared with in vitro experiments, summarized in Table 6) are indeed among the cells with high levels of NKCC1 expression, as their responsiveness to bumetanide would indicate in our experiments. Anyhow, our in vivo results are perfectly in line with previous reports indicating a small percentage of neurons with increased [Cl⁻]_i⁵⁸ or depolarizing responses to another GABA_AR-positive allosteric modulator (zolpidem) in freely behaving animals.³⁷

Both experimental and computational studies have shown that in mature neurons, E_GABA value is only a few mV more negative than the RMP; therefore, even small deviations from the set-point in one or both of these measures may cause a switch in the polarity of GABAergic signaling from inhibition to excitation or vice versa.⁸ In this framework, it is not surprising that many brain disorders are characterized by depolarizing GABAergic signaling.^13,20,32,65 In light of our study, it is tempting to speculate that this concept may simply reflect an increase in the percentage number of cells with high [Cl⁻]_i (and thus depolarizing and excitatory GABAergic signaling) and not a generalized small increase in the level of [Cl⁻]_i in all neurons. Accordingly, we found a large subpopulation of neurons with excitatory GABAergic responses in the Ts65Dn mouse model of DS at all levels of investigation and in line with the current literature on altered average Cl⁻ homeostasis in adult trisomic mice.^24,38,41,66

Although we found that the number of depolarizing neurons represents a small minority compared with the number of hyperpolarizing ones in WT and even in Ts65Dn mature brains, this finding may still have important functional implications. However, thus far, manipulations of this subpopulation of neurons with depolarizing GABA in adult WT animals by pharmacological treatment with NKCC1 inhibitors,^{13,24,41,56,66,67,68,69} genetic deletion of NKCC1, or its downregulation by RNA interference^{38,70,71,72,73} have been shown not to significantly affect cognitive (novel object recognition test, Y maze, object location test, conditional fear conditioning test, and operant reversal learning test^{24,38,41,56,57}), social (three chamber and male-female interaction⁴¹), and motor functions (tail suspension, motor coordination, and forced swim test^{74; 72; 75; 76}) or to significantly increase anxiety (open-field test and elevated plus maze⁷⁴). Further experiments addressing more subtle and specific behaviors may reveal a role for the subpopulation of neurons with depolarizing GABA responses in WT animal behavior under physiological conditions. Anyhow, modulating NKCC1 with pharmacological inhibitors or RNA interference in mouse models of a number of pathologies (including DS) as well as in patients has shown positive behavioral outcomes.^13,32,77 This finding indicates that the small subpopulation of neurons with depolarizing GABA responses may have in fact an important role in pathological conditions. Our in vivo experiments on behaving mice and computational analysis add substantial knowledge to this mostly in vitro experiment-based broad discussion in the field aimed at directly addressing whether the subpopulation of neurons with depolarizing GABA signaling have a pathological role in the many mouse models of brain disorders characterized by depolarizing GABA signaling. By unprecedented Ca²⁺ in vivo experiments, we found here that adult DS mice had a larger subpopulation of depolarizing GABA neurons than WT mice. As a consequence, enhancing GABAergic signaling by diazepam treatment in adult DS mice led to a paradoxical anxiety increase in the dark-light behavioral test, which was rescued by bumetanide pretreatment. Because bumetanide treatment significantly reduced the number of mature neurons with depolarizing GABA responses in Ts65Dn neurons in cultures, brain slices, and in vivo, the rescue of paradoxical responses to benzodiazepines in Ts65Dn animals by bumetanide treatment indeed points to a pathological role of the subpopulation of neurons with depolarizing GABA in DS. On the other hand, and in agreement with the current literature,⁴² we found a substantial lack of anxiolytic diazepam activity, but without a clear paradoxical response in Ts65Dn mice in the open field test and elevated plus maze, possibly indicating a low sensitivity of these tests in revealing this phenotype. Despite this, we found that bumetanide treatment strongly restored anxiolytic diazepam action in adult Ts65Dn mice in both tests.

Although benzodiazepines are used as effective anxiolytics and sedatives worldwide for a vast number of therapeutic indications, adverse paradoxical reactions (i.e., anxiogenic vs. anxiolytic effects) have been described in subjects with neurodevelopmental disorders (e.g., autism and schizophrenia).^78,79,80 Moreover, paradoxical reactions to benzodiazepines, including increased emotional release, excitement, excessive movement, panic attacks, and even hostility and rage, were reported in subjects with several predisposing risk factors, such as young age, advanced age, genetic predisposition (i.e., diabetes mellitus and atrial fibrillation), alcoholism, and psychiatric and/or personality disorders (i.e., depressive bipolar disease, bipolar affective disorder, borderline personality disorder, manic depression, and chronic insomnia^79,80). This may indicate that an increased subpopulation of neurons with depolarizing GABA responses may also be present in these pathologies, as we described here in adult DS animals. Furthermore, our results on the modulation of the GABA_AR-dependent effect of diazepam (both on in vivo Ca²⁺ imaging and anxiety behavior) by systemic bumetanide pretreatment indicate that NKCC1 has a role in mediating GABAergic responses in the brain in vivo, also in the light of the cell-type-specific expression patterns of NKCC1 in the brain.⁸¹ Our results on the modulation of diazepam effect by systemic bumetanide pretreatment are thus very relevant, considering the poor blood-brain barrier penetration of bumetanide, which at times has cast doubts on its chloride-dependent (and thus GABA-dependent) mechanism of action.^82,83,84,85

To increase the translational value of our data, we validated our results in human neurons at advanced stage of development (as defined by our experimental condition in Figure S14) derived from a control iPSC line and a trisomic iPSC line, both of which were obtained from the same person with mosaic DS. These data strongly support the design of current (EudraCT 2015-005780-16) and future clinical trials with NKCC1 inhibitors in DS. Moreover, the increased expression of NKCC1 corresponding to an increased subpopulation of trisomic cells with depolarizing GABA (which was inhibited by bumetanide) suggests that iPSC-derived neurons from patients could be used to test NKCC1 levels as a biomarker of depolarizing GABAergic activity or even test responsiveness to pharmacological treatments of particular individuals with DS. Thus, neurons derived from patient iPSC lines could be used for patient stratification in future clinical trials for DS and possibly for many other brain disorders characterized by impaired chloride homeostasis.¹³ This would be very interesting also considering the increasing interest of the academia and pharma companies in drug discovery programs aiming at finding new compounds able to selectively inhibit NKCC1,^86,87,88 activate KCC2,^89,90,91,92 or acting downstream of their signaling pathways.^22,93

In conclusion, we demonstrated the presence of mixed subpopulations of neurons responding to GABA with either hyperpolarizing/inhibitory or depolarizing actions in human and murine neuronal networks across different levels of complexity and developmental stages, including in mature neurons. Although the proportion of neurons exhibiting depolarizing GABA responses was small compared with those with hyperpolarizing responses (especially in WT animals), this finding holds important functional implications. In particular, although the physiological function of the subpopulation of neurons with depolarizing GABA-driven responses in mature WT animals requires further investigation, we showed that the observed increase in this subpopulation in DS animals contributes to determining key aspects of the pathology. As the list of brain disorders associated with depolarizing GABA signaling is already very large and increases every day, our experiments on a mouse model of one of these pathologies (DS) put forward a rather simple concept. Brain networks can normally tolerate (and possibly need) a small heterogeneous subpopulation of neurons with depolarizing GABA signaling, but if disease-associated changes impinge on the subtle balance between the extent of subpopulations of neurons with depolarizing vs. hyperpolarizing GABAergic signaling, very negative consequences can arise.

Limitation of the study

Our study provides valuable insights into the existence of heterogeneous subpopulations of GABA_AR-responding neurons in both physiological and pathological neuronal networks at increasing scale of complexity and developmental stages, including neurons in adult animals. However, we have not established whether these neurons constitute a minority that never complete full maturation, represent a stable and distinct subset of cells with unique mechanisms governing intracellular chloride levels, or simply reflect the fact that [Cl⁻]_i may physiologically fluctuate in individual neurons and promote the oscillation of GABA-mediated signaling from hyperpolarizing to depolarizing even in fully mature neurons. To define the “maturity” of cultured neurons and thus refine our choice of timing for our specific experimental activities in cell cultures, we relied on assessing the expression levels of the chloride transporters NKCC1 and KCC2, along with markers indicative of neuronal development such as DCX for immature neurons, NeuN for mature neurons, and Tubβ3 acting as a pan-neuronal marker along the development. Additionally, we evaluated electrophysiological parameters throughout the developmental stages. Anyhow, to ultimately address the questions highlighted earlier on the nature of neurons with depolarizing GABA responses, long-term tracking of individual neurons from early development to complete maturation would be necessary.

Moreover, our study primarily aimed to highlight the existence of a subset of neurons exhibiting depolarizing responses to GABA in neuronal networks. However, our focus did not extend to single-neuron properties such as membrane potential (Vm) changes, which may be addressed in future studies.

Furthermore, the diverse pharmacological approaches employed (GABA and bicuculline application in vitro and diazepam application in vivo), as well as the substantial differences in pharmacokinetics and pharmacodynamics between in vivo and in vitro settings may introduce some variability in detecting depolarizing GABAergic responses.

Moreover, in vitro culturing and brain slicing procedures may impact chloride homeostasis per se. Importantly, the Ts65Dn and Dp(16)1Yey/+ models used here differ in both the kind of genetic triplication (freely segregating extra chromosome vs. tandem duplication) and the genetic background (B6EiC3 vs. C57BL6/J), potentially explaining the differences in GABA-mediated signaling found here.

Additionally, exploring more subtle and specific behaviors in further experiments may unveil the role of the subpopulation of neurons with depolarizing GABA responses in the behavior of WT animals under physiological conditions.

Finally, although our findings shed light on the potential pathological relevance of this subpopulation, further research is needed to fully elucidate their role in the various brain disorders characterized by altered Cl homeostasis and GABAergic signaling.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit anti-Actin	Sigma-Aldrich	Cat# A2066, RRID:AB_476693
Mouse anti-NKCC1	Developmental Studies Hybridoma Bank (DSHB)	Cat# T4 RRID:AB_528406
Rabbit anti-KCC2	Millipore	Cat# 07-432 RRID:AB_310611
Rabbit anti-NeuN	CST	Cat#24307 RRID:AB_2651140
Goat anti-DCX	Santa Cruz	Cat#sc-8066 RRID:AB_2088494
Mouse anti-Tubulin beta 3 (clone Tuj1)	BioLegend	Cat#801201 RRID:AB_2728521
Rabbit anti-vGAT	Synaptic Systems	Cat#131-013 RRID:AB_2189938
Guinea pig anti-VGlut1	Synaptic Systems	Cat# 135 304, RRID:AB_887878
HRP-conjugated goat anti-mouse	Thermofisher scientific	Cat# 31430 RRID:AB_228307
HRP-conjugated goat anti-rabbit	Thermofisher scientific	Cat# 31460 RRID:AB_228341
Cy5-conjugated goat anti rabbit	GE Healthcare	Cat# PA45012, RRID:AB_772204
Donkey anti-Goat IgG Secondary Antibody, Alexa Fluor™ 488	Thermofisher scientific	Cat# A-11055 RRID:AB_2534102
Donkey anti-Mouse IgG Secondary Antibody, Alexa Fluor™ 568	Thermofisher scientific	Cat# A-10037 RRID:AB_2534013
Donkey anti-Rabbit IgG Secondary Antibody, Alexa Fluor™ 647	Thermofisher scientific	Cat# A-31573 RRID:AB_2536183
Goat anti-Guinea Pig IgG Secondary Antibody, Alexa Fluor™ 488	Thermofisher scientific	Cat# A-11073 RRID:AB_2534117
Goat anti-Rabbit IgG Secondary Antibody, Alexa Fluor™ 568	Thermofisher scientific	Cat# A-11036 RRID:AB_10563566
Goat anti-Mouse IgG Secondary Antibody, Alexa Fluor™ 647	Thermofisher scientific	Cat# A-21236 RRID:AB_2535805
Bacterial and virus strains
AAV1.Camk2a.GCaMP6f.WPRE.bGHpA (Ready-to-Image Virus Mouse Dorsal CA1 Hippocampus)	Inscopix	Cat# 1000-002244
Chemicals, peptides, and recombinant proteins
HBSS	Gibco™	Cat# 14170088
Dimethyl Sulfoxide (DMSO)	Sigma-Aldrich	Cat# 276855
Paraformaldehyde	Sigma-Aldrich	Cat# 158127
NaCl	Sigma-Aldrich	Cat#71376
HEPES Sigma-Aldrich	Sigma-Aldrich	Cat# H3375
Nonfat dry milk	Euroclone	Cat# EMR180500
Sucrose	Sigma-Aldrich	Cat# S7903
Magnesium chloride hexadrydate	Sigma-Aldrich	Cat# M9272
Potassium chloride	Sigma-Aldrich	Cat# P9333
Sodium phosphate monobasic dihydrate	Sigma-Aldrich	Cat# 71505
Calcium chloride dihydrate	Sigma-Aldrich	Cat# C3306
Sodium bicarbonate	Sigma-Aldrich	Cat# S5761
D(+) Glucose	Sigma-Aldrich	Cat# G7021
Bicuculline methiodide	Sigma-Aldrich	Cat# 14343
Tetradotoxin	Hello Bio	Cat# HB1035
Bovine Serum Albumin	Sigma-Aldrich	Cat# A9647
Gentamicin	Gibco™	Cat# 15710049
Poly-D-Lysine	Sigma-Adrich	Cat# P7886
Sodium chloride	Sigma-Adrich	Cat# S9888
Magnesium sulfate hexadrydate	Sigma-Adrich	Cat# M5921
Pyruvic Acid, 98%	Sigma-Adrich	Cat# 107360
Tergazyme	Sigma-Adrich	Cat#Z273287-1EA
Trypsin	Gibco™	Cat# 25050014
Deoxyribonuclease I (DNAse)	Sigma-Adrich	Cat# D5025
Soybean trypsin inhibitor	Sigma-Adrich	Cat# T6522
Poly-L-lysine	Sigma-Adrich	Cat# P2636
Neurobasal™-A Medium	Gibco™	Cat# 10888022
GlutaMax	Gibco™	Cat# 35050038
mTeSR™1	Stem Cell Technologies	Cat# 85850
Corning® Matrigel®	Corning	Cat# 354277
ReLeSR™	Stem Cell Technologies	Cat#100-0484
STEMdiff™ Neural Induction Medium	Stem Cell Technologies	Cat#05839
ROCK inhibitor Y-27632	Stem Cell Technologies	Cat# 72302
StemPro™ Accutase™	Gibco™	Cat# A1110501
DMEM/F-12	Gibco™	Cat# 31330038
B-27™ Supplement	Gibco™	Cat# 17504044
N-2 Supplement	Gibco™	Cat# 17502048
Penicillin-Streptomycin	Gibco™	Cat# 15140122
Human FGF-basic	Peprotech	Cat# 100-18B
Human BDNF Recombinant Protein	Peprotech	Cat# 450-02
Human GDNF Recombinant Protein	Peprotech	Cat# 450-10
Dibutyryl-cAMP	Sigma-Adrich	Cat# D0627
Ascorbate	Sigma-Adrich	Cat# A4403
Laminin	Sigma-Adrich	Cat# L 2020
Poly-L-ornithine	Sigma-Adrich	Cat# P4957
Potassium D-gluconate	Sigma-Aldrich	Cat# G4500
Adenosine 5-triphosphate magnesium salt	Sigma-Aldrich	Cat# A9178
Ethylene glycol-bis(2-aminoethylether)-N,N,N,N-tetraacetic acid	Sigma-Aldrich	Cat# E43785
Guanosine 5-triphosphate sodium salt hydrate	Sigma-Aldrich	Cat# G8877
Phosphocreatine disodium salt hydrate	Sigma-Aldrich	Cat# P7936
Fluo4-AM	Molecular Probes	Cat# F14201
γ-Aminobutyric acid	Sigma-Aldrich	Cat# A2129
Muscimol	Sigma-Aldrich	Cat# M1523
CGP 55845	Tocris	Cat#1248
Acetazolamide	Tocris	Cat# 6742
Nifedipine	Tocris	Cat# 1075
Tetrodoxine	Tocris	Cat# 1078
Bumetanide	Sigma-Adrich	Cat# B3023
Diazepam	Tocris	Cat# 2805
N-(Ethoxycarbonylmethyl)-6-Methoxyquinolinium-Bromide	Molecular Probes	Cat# E3101
Calcein-AM	Thermo-Fisher Scientific	Cat# C1430
Propidium Iodide	Sigma-Adrich	Cat# P4864
NP40	Sigma-Aldrich	Cat# 74385
Sodiumdeoxycholate	Sigma-Aldrich	Cat# D6750
Sodium Dodecyl Sulfate (SDS)	Thermo-Fisher Scientific	Cat# 28312
NuPAGE™ MOPS SDS Running Buffer	Thermo-Fisher Scientific	Cat# NP0001
PMSF solution	Sigma-Aldrich	Cat # 93482
Sodium fluoride	Sigma-Adrich	Cat# S6776
Sodium orthovanadate	Sigma-Adrich	Cat# S6508
Protease Inhibitor Cocktail	Sigma-Adrich	Cat #P8340
Phosphatase Inhibitor Cocktail 2	Sigma-Adrich	Cat #P5726
Phosphatase Inhibitor Cocktail 3	Sigma-Adrich	Cat #P0044
Lithium-dodecyl-sulfate (LDS) sample buffer	Thermo-Fisher Scientific	Cat# NP0007
Dithiothreitol (DTT)	Sigma-Adrich	Cat# D0632
NuPAGE™ 4–12% Bis-Tris Midi Protein gels	Thermo-Fisher Scientific	Cat# WG1402A
Nitrocellulose membrane	Sigma-Aldrich	Cat# GE10600001
CNQX	Tocris	Cat# 1045
AP5	Tocris	Cat# 0106
ProView™ Lens Probe 1.0mm diameter	Inscopix	Cat# 1050-002202
Baseplate with cap screw	Inscopix	Cat# 1050-004201
Baseplate cover	Inscopix	Cat# 1050-002193
Critical commercial assays
Pierce BCA assay	Thermo-Fisher Scientific	Cat# 23225
SuperSignal West Pico PLUS Chemiluminescent Substrate detection system	Thermo-Fisher Scientific	Cat# 34578
Experimental models: Cell lines
Human iPS Down Syndrome: UWWC1-DS4	WiCell	Cat# uwwc1-ds4, RRID:CVCL_EJ83
Human iPS isogenic control: UWWC1-DS2U	WiCell	Cat# uwwc1-ds2u, RRID:CVCL_EJ82
Experimental models: Organisms/strains
Mouse: B6EiC3Sn a/A-Ts(1716)65Dn/J	The Jackson Laboratory	Cat# JAX:001924, RRID:IMSR_JAX:001924
Mouse: B6.129S7-Dp(16Lipi-Zbtb21)1Yey/JB6.129S7-Dp(16Lipi-Zbtb21)1Yey/J	The Jackson Laboratory	Cat # JAX:013530; RRID:IMSR_JAX:013530
Mouse: C57BL/6J	The Jackson Laboratory	Cat# JAX:000664, RRID:IMSR_JAX:000664
Mouse: (C57BL/6JEiJ x C3H/HeSnJ)F1/J	The Jackson Laboratory	Cat# JAX:001875, RRID:IMSR_JAX:001875
Software and algorithms
GraphPad PRISM	GraphPad Software Inc.	https://www.graphpad.com/scientific-software/prism/
Anymaze	Anymaze	http://www.anymaze.co.uk/index.htm
MATLAB	Mathworks	https://it.mathworks.com/products/matlab.html
NIS-Elements software	Nikon Instruments Inc.	https://www.microscope.healthcare.nikon.com/it_EU/products/software/nis-elements
ImageJ software	ImageJ	http://rsbweb.nih.gov/ij/
Beanplots R package	Kamstra P. et al. 200894	http://cran.r-project.org/web/packages/beanplot/index.html
ImageQuant software	GE Healthcare	https://pdf.medicalexpo.com/pdf/ge-healthcare-life-sciences/imagequant-tl-81/80554-233594.html
MC-Rack software	Multichannel System	https://www.multichannelsystems.com/software/mc-rack
SPYCODE MATLAB Graphical user interface	Bologna et al. 200895	https://pubmed.ncbi.nlm.nih.gov/20554151/
mapMEA MATLAB Graphical user interface	Panuccio et al., 201896	https://www.jove.com/it/t/57548/recording-modulation-epileptiform-activity-rodent-brain-slices?language=Italian
Boostrap MATLAB algorithm	Slomowitz E et al. 201531	https://elifesciences.org/articles/04378#info
Inscopix data processing software (IDPS)	Inscopix	https://inscopix.com/software-analysis-miniscope-imaging/
SigmaPlot	Systat	https://sigmaplot.it.softonic.com/
nlme R package	RCoreTeam	https://cran.r-project.org/web/packages/nlme/index.html
Other
60MEA200/30iR-Ti w/o ring	Multichannel System	https://www.multichannelsystems.com/products/60mea20030ir-ti-wo
MEA Custom Chamber	Crisel Instrument	Cat: SKE-chamber MEA
Ag/AgCl electrode, pellet, 1.0 mm	Crisel Instrument	Cat #64-1309
nVista 3.0	Inscopix	https://inscopix.com/nvista-system/

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, laura.cancedda@iit.it.

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •The datasets of this work are publicity available on Mendeley Data: https://data.mendeley.com/datasets/yrzcdyzbwh/1
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Experimental model and subject details

Mice

The Ts65Dn colony⁹⁷ was maintained by crossing Ts65Dn females to C57BL/6JEi x C3SnHeSnJ (B6EiC3) F1 males (Jackson Laboratories). Dp(16)1Yey/+⁹⁸ males were purchased from The Jackson Laboratory and used to create a colony by mating with C57BL6/J females (Charles River). All animals were genotyped by PCR as previously described.^98,99 A veterinarian was employed to monitor health and comfort of the animals. Mice were housed in a temperature-controlled room with a 12:12 h dark/light cycle and ad libitum access to water and food. All animal procedures were approved by IIT licensing, the Italian Ministry of Health (D.Lgs 26/2014) and EU guidelines (Directive 2010/63/EU). All the electrophysiological and behavioral experiments were performed on adult (2–3 months old) mice.

Ethical approval declaration

All animal experiments were performed in accordance with the guidelines established by the European Community Council Directive 2010/63/EU of September 22nd 2010 and were approved by the Italian Ministry of Health (authorizations no: 829/2015-PR and 658/2016-PR to A.C.).

Primary neuronal culture

Primary neuronal cultures were prepared from WT and Ts65Dn mouse pups at postnatal day 2 (P2), as previously described.^25,41,100 In brief, brains were dissected under a stereomicroscope in ice-cold buffer (DB) composed of Hank’s balanced salt solution (HBSS; Gibco) supplemented with 6 mg/mL glucose, 3 mg/mL bovine serum albumin (BSA), 5.5 mM MgSO4, 5 μg/mL gentamycin and 10 mM HEPES, pH 7.4 (all from Sigma). Cortical or hippocampal tissue was minced and then enzymatically digested with 0.25% trypsin in HBSS containing 0.6 mg/mL deoxyribonuclease I (DNAse; Sigma) for 5 min at 37°C. Tissue chunks were washed in DB, incubated for 5 min in DB supplemented with 1 mg/mL of Soybean trypsin inhibitor (Sigma) and mechanically dissociated in DB supplemented with 0.6 mg/mL DNAse. Cells were passed through a 40 μm strainer and then centrifuged (110 x g for 7 min at 4°C) to remove cellular debris. Cells were plated on glass coverslips (Menzel Gläser), 6-well plates (BD Falcon) or Microelectrodes Arrays (MEAs, Multichannel System, Germany) all coated with poly-L-lysine (Sigma; 0.1 mg/mL in 100 mM borate buffer, pH 8.5) at a density of 250–500 cells/mm2. Neurons were maintained in a culture medium consisting of Neurobasal-A supplemented with 2% B-27, 1% GlutaMax and 5 μg/mL gentamycin (all from Gibco) at 37°C in humidified atmosphere (95% air, 5% CO2). Primary neurons were recorded between 15 and 21 days in vitro (DIVs).

Human iPSC-derived neurons

The previously generated Down syndrome cell line DS4 of human induced pluripotent stem cells (iPSCs) and the corresponding isogenic control line DS2U (Weick et al., 2013) were obtained from WiCell (Madison, Wi). Both iPSC lines were obtained from an individual with a mosaic form of DS. Human DS and control iPSCs were cultured and maintained in mTeSR1 medium (Stem Cell Technologies) in tissue culture dishes coated with hESC-qualified Matrigel (Corning). Cells were routinely passaged as aggregates with ReLeSR (Stem Cell Technologies) dissociation reagent.

For neuronal differentiation, a commercially available differentiation method (Stem Cell Technologies) based on the generation of embryoid bodies (EBs) was employed to first generate neural progenitor cells (NPCs). In brief, EBs were generated by seeding single-cell suspension of iPSC in AggreWell-800 plates (Stem Cell Technologies) in STEMdiff Neural Induction Medium (NIM; both from Stem Cell Technologies) in the presence of the ROCK inhibitor Y-27632 (10 μM). After 5 days, EBs were removed and plated onto Matrigel-coated dishes (BD Falcon) in NIM, where they attached and formed neural rosettes. After further 5 days, the rosettes were isolated with Neural Rosette Selection Reagent (Stem Cell Technologies) and re-plated onto Matrigel-coated dishes in NIM. After additional 5–7 days, NPCs outgrew from rosette and formed a monolayer of cells. NPCs were then detached with StemPro Accutase (Gibco) and re-plated on Matrigel-coated dishes in NPC Medium (DMEM/F12, 2% B27, 1% N2, 1% Pen-strep; all from Gibco) supplemented with 20 ng/mL bFGF (Peprotech). Cells were passaged using StemPro Accutase and re-plated at 72,000 cells/cm2 in complete NPC Medium for maintenance.

For neuronal differentiation, NPCs were first pre-differentiated by plating into 10 cm plastic dishes coated with Matrigel in complete NPC Medium as above. The next day, a complete media change with NPC Medium was performed. On the second day, half volume of Neuronal Differentiation Medium (NDM; Neurobasal A, 2% B27, 1% N2, 1% Glutamax, 1% Pen-strep, 20 ng/mL BDNF and GDNF, 100 μM dibutyryl-cAMP and 200 nM Ascorbate) was added to the culture. On the fourth day, half volume of the media was changed with fresh NDM media. On the fifth day, cells were detached with StemPro Accutase, counted in a haemocytometer chamber, and re-plated for final differentiation at the density of 18,000 cells/cm2 on PLO/Laminin coated glass coverslips (Menzel Gläser). Coverslips were first coated overnight with poly-L-ornithine (Sigma; 0.05 mg/mL in 100 mM borate buffer, pH 8.5) in the incubator at 37°C. On the day of plating, coverslips were washed with autoclaved milliQ water and left to dry before coating with 1 μg/mL of Laminin (Sigma) diluted in DMEM/F12 for 4 h in the incubator at 37°C. Laminin was removed just before plating the cells. For culture maintenance, half of the medium was changed three times a week with the complete NDM medium supplemented with 1 μg/mL of Laminin. Cells were used for Ca²⁺ imaging, patch clamp recording or lysed for biochemistry between 30 and 60 days after plating for the final differentiation.

Method details

Patch-clamp recordings of human iPSC-derived neurons

IPSC-derived neurons were recorded at 30 or 60 days after final plating, at room temperature (22°C–24°C) in an extracellular solution containing in mM: 145 NaCl, 5 KCl, 10 HEPES, 5.55 Glucose, 1 MgCl2, 2 CaCl2. Cells were visualized with an upright microscope (Olympus BX51WI) with infrared differential interference contrast optics. Patch pipettes (3–5 MΩ) were filled with an intracellular solution containing in mM: 130 K-Gluconate, 7 KCl, 10 HEPES, 4 MgATP, 0.6 EGTA, 0.3 NaGTP, 10 phospho-creatine. Access resistance was monitored during voltage-clamp recordings, and cells with values more than 25 MΩ were excluded. The resting membrane potential (RMP) was measured immediately after reaching the whole-cell configuration, while injecting no current (I = 0). The input resistance (RIN) was calculated from the linear voltage deflection in response to small current steps (from −40 to +30 pA, 10 pA increments, 500 ms for each sweep). Rheobase recordings were performed by injecting small depolarizing currents (5 pA increaments, 500 ms for each sweep).

Patch-clamp recordings of primary neurons

Primary neurons were recorded at 15 and 21 DIVs, at room temperature (22°C–24°C) in an extracellular solution containing in mM: 145 NaCl, 5 KCl, 10 HEPES, 5.55 Glucose, 1 MgCl2, 2 CaCl2. Cells were visualized with an upright microscope (Olympus BX51WI) with infrared differential interference contrast optics. Patch pipettes (3–5 MΩ) were filled with an intracellular solution containing in mM: 130 K-Gluconate, 7 KCl, 10 HEPES, 4 MgATP, 0.6 EGTA, 0.3 NaGTP, 10 phospho-creatine. Access resistance was monitored during voltage-clamp recordings, and cells with values more than 25 MΩ were excluded. The resting membrane potential (RMP) was measured immediately after reaching the whole-cell configuration, while injecting no current (I = 0). The input resistance (RIN) was calculated from the linear voltage deflection in response to small current steps (from −40 to +30 pA, 10 pA increments, 500 ms for each sweep). Rheobase recordings were performed by injecting small depolarizing currents (5 pA increaments, 500 ms for each sweep).

In vitro calcium imaging

Mouse primary neuronal cultures

Calcium imaging experiments on WT or Ts65Dn hippocampal primary neuronal cultures grown on glass coverslips were performed with the Ca²⁺-sensitive dye Fluo4-AM (Molecular Probes). Coverslips at different days in vitro (2, 7, 15 or 21) were loaded with 2.5 μM Fluo4-AM in extracellular solution (NaCl 145 mM, KCl 5 mM, CaCl2 2 mM, MgCl2 1 mM, HEPES 10 mM, D-glucose 5.5 mM, pH 7.4) for 20 min in the dark at room temperature. Coverslips were then transferred to a holding chamber and perfused (2 mL/min) with the same extracellular solution for 2–3 min before imaging. Live imaging was performed with a Swept Field Confocal (SFC) microscope (Nikon) equipped with a 20× air-objective (NA 0.75) under constant perfusion. After 10–15 min of baseline imaging, neurons were bath stimulated with 100 μM GABA or 10 μM muscimol (both from Sigma) for additional 8–10 min. Following GABA or muscimol washout, neurons were stimulated with 30 mM KCl to depolarize cells and evaluate the total number of viable neurons present in the field of view. In some experiments, we used the GABA_AR antagonist bicuculline (Sigma; 100 μM), the GABA_BR antagonist CGP55845 (Tocris; 10 μM), the carbonic anhydrase blocker acetazolamide (Tocris; 10 μM), the L-type voltage-gated calcium channel blocker nifedipine (Tocris; 10 μM), or the voltage-gated sodium channel blocker TTX (Tocris, 1 μM) in combination to GABA stimulation (100 μM). For these experiments, we first recorded 10–15 min of baseline (as above), then we performed 10–15 min of imaging in the presence of the tested drug, followed by 8–10 min of imaging in the presence of GABA (100 μM) plus the tested drug. Finally, after washout of the drugs, neurons were depolarized with 30 mM KCl (as above). In other experiments, neurons were pretreated with 10 μM of bumetanide (Sigma) or the corresponding vehicle (0.01% DMSO) during Fluo4-AM loading and subsequently imaged (as above), but under constant perfusion with extracellular solution containing either bumetanide or DMSO. One field of view for each coverslip was acquired.

iPSC-derived human neurons

For calcium imaging experiments on control and DS iPSC-derived human neurons, cultures were grown on glass coverslips for 60 days after final plating. Cells were loaded with 2.5 μM of Fluo4-AM directly in their culture medium for 20 min in the incubator. Coverslips were then transferred to a 35 mm dish containing an imaging solution (NaCl 95 mM, KCl 5 mM, CaCl2 1.8 mM, MgCl2 0.8 mM, NaH2PO4 1 mM, NaHCO3 26 mM, HEPES 10 mM, D-glucose 10 mM, pH 7.4) before being transferred to a holding chamber and perfused (2 mL/min) with the same solution for additional 2–3 min. Similar to WT and Ts65Dn mouse primary neurons, human iPSC-derived neurons were pretreated with 10 μM bumetanide (Sigma) or the corresponding vehicle (0.01% DMSO) during Fluo4-AM loading and subsequently imaged with constant perfusion with imaging solution containing either bumetanide or DMSO. Live imaging was performed with an inverted microscope (Nikon) equipped with a 20× air-objective (NA 0.75), a pE-300 LED illumination system (CoolLED) and an iXON EMCCD camera (Andor Technology). After 10 min of baseline imaging under constant perfusion, neurons were bath stimulated with 100 μM GABA (Sigma) for additional 8 min. Following GABA washout for 4 min, neurons were stimulated with 30 mM KCl to depolarize cells and evaluate the total number of viable neurons present in the field of view. One field of view for each coverslip was acquired.

For all cells, image analysis was performed by NIS-Elements software (Nikon). First, regions of interest (ROIs) were placed on the cell body of individual neurons identified in video frames, following KCl depolarization of cells (representing the maximum fluorescence of viable neurons). Next, average Fluo4 fluorescent intensity for each ROI was measured by NIS-Elements software on every frame of the recording to generate the corresponding calcium traces for each cell. The raw calcium traces obtained for each neuron were then transformed to relative changes in fluorescence (ΔF/F0) over time. The value of ΔF/F0 for each cell was calculated as (F-F0)/F0, where F0 is the minimal fluorescence signal for a given cell. An operator blinded to the experimental sample code analyzed the calcium traces and scored neurons showing GABA-dependent depolarizing responses. Neurons were scored positive if GABA or muscimol application elicited a calcium transient that was at least 50% higher than the basal fluorescence. The percentage of neurons showing GABA-dependent depolarizing responses was calculated for each field of view over the total number of viable neurons (i.e., neurons depolarized by KCl application) in the same field of view.

In vitro chloride imaging

Imaging of intracellular Cl⁻ in WT or Ts65Dn primary hippocampal neuronal cultures grown on coverslips was performed with the fluorescent chloride-sensitive indicator MQAE [N-(Ethoxycarbonylmethyl)-6-Methoxyquinolinium-Bromide; Molecular Probes], as previously described.^25,41 The MQAE dye detects Cl⁻ ions via diffusion-limited collisional quenching, resulting in a concentration-dependent decrease of fluorescence emission following an increase in Cl⁻ concentration.¹⁰¹ Therefore, a decrease in MQAE fluorescence is indicative of a higher [Cl⁻]i and vice versa. Hippocampal neurons at 15 DIV were loaded with 5 mM MQAE in their culture medium for 30 min at 37°C in the presence of bumetanide (10 μM) or the corresponding vehicle (0.01% DMSO). Coverslips were then transferred to a holding chamber and perfused (2 mL/min) with extracellular solution (as above) containing either bumetanide or vehicle for 5 min before imaging. Images were acquired with a Nikon A1 scanning confocal microscope equipped with a 20× air-objective (NA 0.75). MQAE was excited with a 405 nm diode laser and fluorescence collected with a 525/50 nm band-pass emission filter. All excitation and acquisition parameters (laser intensity, PMT offset and gain) were keep constant throughout experiments. Image analysis was performed with NIS-Elements software (Nikon) by measuring the mean fluorescent intensity of ROIs placed on the cell body of individual neurons from 6 randomly selected fields for each coverslip. For each experiment, the average fluorescent intensity of all ROIs from a coverslip was normalized to the average fluorescent intensity of control samples (WT neurons treated with vehicle) in the same experiment. Pseudo-color images were generated with ImageJ software (http://rsbweb.nih.gov/ij/). The beanplots were generated with R package⁹⁴ (http://cran.r-project.org/web/packages/beanplot/index.html).

Live/dead assay

WT and Ts65Dn neurons grown on coverslips were stained at 15 DIV for 10 min in the dark with extracellular solution (as above) containing 0.5 μM Calcein-AM (ThermoFisher Scientific) and 5 μg/mL of propidium iodide (PI; Sigma) to stain live and dead cells respectively, as previously described.¹⁰² Coverslips were then washed twice with extracellular solution, and fluorescent images immediately taken with a Nikon A1 confocal microscope. The number of Calcein- and PI-positive cells was counted with NIS-Elements software (Nikon).

Immunofluorescence and cell counting

Cells grown on glass coverslips were fixed (mouse: 2, 7, 15 and 21 DIV, human: 30 and 60 days after final neuronal differentiation) with 4% paraformaldehyde in 100 mM phosphate buffer (pH 7.4) for 20 min and then extensively washed with PBS. For immunofluorescence, mouse and human neurons were permeabilized in PBS containing 0.1% Triton X-100 (PBST), incubated for 1 h with blocking-buffer (5% normal Goat or Donkey serum in PBST) and incubated overnight at 4°C with the following primary antibodies: rabbit anti-NeuN (CST, catalog no: 24307; 1:400); goat anti-DCX (Santa Cruz catalog no: sc-8066; 1:200); mouse anti-Tubulin β3 (clone Tuj1; BioLegend catalog no: 801201; 1:400), guinea pig anti-Map2 (SySy, catalog no:188-004; 1:500); rabbit anti-vGAT (SYSY, catalog n°: 131-013; 1:500); guinea pig anti-vGLUT1 (SYSY, catalog n°: 135–304; 1:500). Fluorophore conjugated (Alexa Fluor 488, Alexa Fluor 568 and Alexa Fluor 647), goat or donkey secondary antibodies (1:500; ThermoFisher Scientific) were used for detection. Fluorescent images were acquired with a Leica SP5 confocal scanning microscope or a Nikon A1 confocal scanning microscope. Percentage of DCX-, NeuN- or Tubβ3-positive cells were counted with NIS-Elements software (Nikon). Puncta immunostained for vGAT and vGLUT1 were automatically quantified using the particle-analysis plugin available in ImageJ software (http://imagej.nih.gov/ij/), as previously described.¹⁰³ On the same images, the area occupied by Tubβ3 staining was measured after applying an identical threshold on fluorescent intensity for all images. The area occupied by staining of the neuronal marker Tubβ3 was used to normalize the number of vGAT and vGLUT1 puncta.

Biochemistry

Primary neuronal cultures, iPSC-derived neurons or brain samples from WT, Ts65Dn and Dp(16) mice were lysed in ice-cold RIPA buffer (1% NP40, 0.5% Deoxycholic acid, 0.1% SDS, 150 mM NaCl, 1 mM EDTA, 50 mM Tris, pH 7.4) containing 1 mM PMSF, 10 mM NaF, 2 mM sodium orthovanadate and 1% (v/v) protease and phosphatase inhibitor cocktails (Sigma). Samples were clarified through centrifugation at 20,000 x g at 4°C, and the protein concentration was determined using the BCA kit (Pierce). For immunoblot analysis, protein extracts were prepared in lithium-dodecyl-sulfate (LDS) sample buffer (ThermoFisher Scientific) containing 50 mM dithiothreitol (DTT). To avoid NKCC1 or KCC2 protein aggregation/precipitation, samples were not heat-treated before loading.^104,105 Equal amounts of proteins were run on 4–12% Bis-Tris, NuPAGE gels (ThermoFisher Scientific) with MOPS buffer and transferred overnight at 4°C onto nitrocellulose membranes (GE Healthcare) with Tris-Glycine transfer buffer (25 mM Tris-base, 192 mM glycine, 20% methanol). Membranes were probed with mouse anti-NKCC1 (clone T4c, Developmental Studies Hybridoma Bank; 1:4,000), rabbit anti-KCC2 (Millipore, catalog no. 07–432; 1:4,000), and rabbit anti-Actin (Sigma, catalog no: A2066; 1:10,000), followed by HRP-conjugated goat anti-mouse, goat anti-rabbit secondary antibodies (ThermoFisher Scientific; 1:10,000). Chemiluminescent signals were revealed with SuperSignal West Pico substrate (Pierce) and digitally acquired on a LAS 4000 Mini imaging system (GE Healthcare). Alternatively, a fluorescent Cy5-conjugated goat anti rabbit secondary antibody (GE Healthcare; 1:5,000) was used and the fluorescent signal acquired with a Typhoon fluorescent scanner (GE Healthcare). Band intensities were quantified using ImageQuant software (GE Healthcare). For some experiments, membranes were first probed with anti-NKCC1 antibody, stripped and then re-probed with anti-KCC2 antibody. Specificity of the anti-NKCC1 or anti-KCC2 antibodies was previously verified on brain samples from NKCC1 and KCC2 deficient mice.²⁴

MEA recordings

Primary neuronal cultures

WT and Ts65Dn primary hippocampal neuronal cultures were grown on MEAs consisting of 60 TiN/SiN planar, round electrodes (30 μm diameter; 200 μm inter-electrode distance) divided into six separated wells (Multichannel Systems). Each well contained nine recording electrodes, arranged in a 3 × 3-square grid, and one big ground electrode. The network activity of all neuronal cultures was recorded by means of the MEA60 System (Multichannel Systems). After 1200 × amplification, signals were sampled at 10 kHz and recorded using a commercial software (MC-Rack software, Multichannel Systems). To reduce thermal stress of the neurons during each experiment, cultures were recorded in their original medium, kept at 37 °C by means of a controlled thermostat (Multichannel Systems) and covered by a Polydimethylsiloxane (PDMS) cap to avoid evaporation and prevent changes in osmolarity.^38,106 Additionally, a custom incubation chamber was used to maintain a controlled humidified atmosphere consisting of 95% air and 5% CO2 during the entire recording, as previously reported.^38,106,107 In experiments where the neuronal network activity was pharmacologically reduced by application of glutamate receptor blockers (20 μM CNQX plus 50 μM AP5; both from Tocris) we used six-well MEAs with higher electrode density (42 electrodes/well arranged in a 7 × 6-square grid) to achieve better spatial coverage of the network (Multichannel Systems).

Cultures at 20–21 DIVs were pretreated with bumetanide (10 μM) or with the corresponding vehicle (0.01% DMSO) for 30 min in the incubator prior to each recording session, and then accommodated on the MEA set-up for 15 min, to let the culture adapt to the new environment and reach a stable level of activity.^38,106 Next, basal spontaneous network activity was recorded for 30 min. After recording basal activity, bicuculline (20 μM; Sigma), GABA (100 μM; Sigma) or diazepam (1 μM; Tocris) were added by directly pipetting in the medium, and activity was recorded for an additional 40 min. Since we previously noticed that mechanical perturbation from the pipette injection in the medium could cause a temporary instability in the firing rate,^38,107 we discarded the first 10 min at the beginning of the 40-min recording phase. Recordings were performed at DIV 21, since the firing rate activity of cultures on MEA increases along development with a maximum at DIV 21 (Chiappalone et al. 2006).

Data analysis was performed offline with MATLAB software (MathWorks) using the custom software package SPYCODE.⁹⁵ The analysis was performed as previously described.^38,106 Briefly, raw traces were high-pass filtered (>300 Hz) to isolate spike events from the low fluctuation of the signal (LFP). Spike detection was computed offline using a custom software, to discriminate spike events from the noise (Bologna et al. 2010). To characterize the level of the activity of the recorded cultures, the mean firing rate (MFR) for each MEA was defined as the mean number of spikes per second (spikes/sec) over the total recording time and computed. Active electrodes were defined as those showing a firing rate greater than 0.02 spikes/s. This low threshold guaranteed that only electrodes that were not covered by cells or those electrodes recording very few spikes were excluded by the analysis, and that all other electrodes were retained for analysis. To evaluate the effect of the drug administration on the firing-rate activity, the MFR ratio, defined as the MFR after drug treatment divided by the MFR during baseline (MFRdrug/MFRbaseline) was computed for each active electrode, and then averaged for each culture.

To assess the presence of mixed subpopulations of neurons, statistically significant changes induced by drug treatments (either GABA, bicuculline or diazepam) in comparison to the basal condition were analyzed at the level of the single electrode with the bootstrap method, as previously described.^31,38 Briefly, the peak train for each time experimental segment (basal or drug) was divided into 1-min bins. The MFRs recorded during each 1-min bin for the two time segments for each well were randomly shuffled into two groups for 10,000 times. The differences between the means of the randomly shuffled groups produced a null distribution. For each electrode, the real difference between the basal and the drug values was computed and assessed whether it fell outside the 95% confidence interval of the null distribution.

Hippocampal brain slices

Adult (2–3 months old) WT and Ts65Dn male mice were perfused intracardially with ice-cold oxygenated (95% O2 and 5% CO2) artificial cerebrospinal fluid (ACSF; composed of 115 mM NaCl, 1.25 mM NaH2PO4, 26 mM NaHCO3, 3.5 mM KCl, 1.3 mM MgCl2, 2 mM CaCl2, 25 mM glucose and 1 mM L-Ascorbic Acid, pH 7.4) under deep isoflurane anesthesia and then decapitated. Brains were quickly removed from the skull into a bowl containing ice-cold oxygenated ACSF and then placed into a small beaker containing oxygenated slushy cutting solution (composed of 230 mM sucrose, 3.5 mM KCl, 1.25 mM KH2PO4, 26 mM NaHCO3, 2 mM MgSO4, 0.5 mM CaCl2, 25 mM D-glucose, 1 mM L-Ascorbic Acid, 3 mM Pyruvic Acid, pH 7.4) for 90–120 s. Hippocampus-entorhinal cortex slices (400 μm thick) were horizontally cut with a VT1000S vibratome (Leica Microsystem) in the same solution. After separating the two brain hemispheres with a scalpel blade, slices were gently transferred to a beaker by using an inverted Pasteur pipette and rinsed twice with standard oxygenated ACSF. The slices were then transferred to a commercial holding chamber (KF Technologies) containing oxygenated ACSF and incubated for approximately 20 min at 32°C and then for at least 1 h at room temperature for recovery.

For brain slice recordings on MEA, we substituted the standard recording ring of the MEA with a customized recording chamber inspired by the design of patch-clamp recording chambers; this provided stable and reliable laminar flow that we kept to a rate of 2 mL/min⁹⁶

Slices were pre-incubated with oxygenated recording ACSF (composed of 115 mM NaCl, 1.25 mM NaH2PO4, 26 mM NaHCO3, 3.5 mM KCl, 1 mM MgCl2, 2.4 mM CaCl2, 25 mM glucose and 1 mM L-Ascorbic Acid, pH 7.4), containing either vehicle (0.01% DMSO) or bumetanide (10 μM) for 30 min, and then transferred to the MEA recording chamber while continuously perfusing with the same solution. Then, slices were let to acclimatize into the MEA setup for 15 min to adapt to the new environment and reach a stable level of activity. Next, basal spontaneous network activity was recorded for 30 min. Upon bath application of either bicuculline (20 μM) or GABA (100 μM), additional 15 min were waited to ensure complete exchange of the solution in the chamber before recording further 30 min of activity.

Data analysis was performed offline with the MATLAB software (MathWorks) using the custom software package SPYCODE.⁹⁵ Only electrodes positioned in the CA1 region were mapped and analyzed using a custom-made software that allowed the user to select the electrodes that correspond to specific structures of interest of the brain slice.⁹⁶ Raw signals were high-pass filtered (>300 Hz) to isolate the spiking activity. Once spikes were detected using the same algorithms used for cultures, the MFR (spikes/sec) for each slice were computed. Active electrodes were considered those showing a firing rate greater than 0.02 spikes/s, in line with the threshold used for cultures.

To evaluate the effect of the drug administration on the firing activity the MFR ratio, defined as the MFR after drug treatment divided by the MFR during baseline (MFRdrug/MFRbaseline), was computed for each active electrode and then averaged for each slice. In line with the analysis for primary neuronal cultures, differences in the spiking frequency induced by drug treatments at the level of the single electrodes were evaluated to assess the presence of mixed subpopulations of neurons. However, to obtain a reliable null distribution after the bootstrapping procedure, a consistent and well-distributed activity during the recording is required. Therefore, given the more scattered distribution of the firing activity recorded in hippocampal slices compared to primary cultures, a hard threshold of at least 15% change in firing activity was adopted.

Animal treatment and behavioral testing

For behavioral experiments, 3–4 month old male Ts65Dn and WT control littermates were randomly assigned to the diverse experimental groups. Animals were treated daily by intraperitoneal (i.p.) injection for two weeks with either bumetanide (Sigma; 0.2 mg/kg body weight) or vehicle (2% DMSO in saline). On the day of behavioral testing, bumetanide or vehicle injections were given about 1 h before the beginning of the behavioral task followed by diazepam (2 mg/kg in saline) or saline IP injections 30 min before the begging of the test. For these behavioral experiments, injectable diazepam (Valium 5 mg/mL from Roche) was diluted to 0.2 mg/mL with saline prior to administration, as previously described.⁴²

Animal behavior was video-recorded throughout the experimental sessions by ANY-maze software.¹⁰⁸ The operator was blind to both genotype and treatment groups. The open field and elevated plus maze tests were performed in dim light illumination (12–15 Lux), while the dark-light test was conducted under strong illumination (180 Lux). WT and Ts65Dn mice were always evaluated in parallel and within the same time schedule. To minimize olfactory cues from previous trials, all used apparatuses were thoroughly cleaned with 70% ethanol and dried at the beginning of each animal task. Three days before starting the behavioral experiments, mice where handled once a day for 5 min each.

Open field test

This test measures the preference of mice for the central or peripheral area of a square arena. Mice were habituated to the test room for 30 min and then placed in the center of a gray acrylic arena (44 × 44 cm) for 10 min of spontaneous exploration. The distance traveled and the time spent in the central or peripheral area of the arena by mice were automatically measured with ANY-maze tracking software. The central area of the arena was defined as a square of 24 × 24 cm (10 cm away from each wall).

Dark-light test

The apparatus used for the dark-light test consisted of a box divided in two compartments by a panel with a sliding door. The light and dark compartments were made of transparent or black opaque acrylic, respectively. Animals were transferred with a black cage (to avoid exposure to strong illumination before the test) into the behavior room 30 min prior to the beginning of the test. Mice were then individually placed into the dark comportment of the box under dim-light illumination of the room and after 5 s (during which the halogen lamp providing the strong illumination was switched on) the sliding door was opened. Mice were allowed to freely explore the two compartments for 10 min. The total number of transitions to and the time spent in the light chamber was analyzed.

Elevated plus maze test

The Elevated Plus Maze apparatus consisted of four arms (30 × 5 cm) perpendicularly linked to a central platform (5 × 5 cm). Two opposing arms were open and the other two enclosed by black walls (15 cm high). The platform height from the floor was 40 cm. The platform was placed in the center of a circular tank to prevent mouse escaping in case of fall from the open arms. Mice were habituated to the test room for 30 min and then placed in the center of the apparatus with the head directed toward one of the closed arms. Mice were allowed to freely explore the apparatus for 10 min. The distance traveled, the number of the entries into each arm, and the time spent in each arm were automatically calculated with ANY-maze tracking software. In this behavioral experiment, the following exclusion criteria was adopted independently of the genotype or treatment (before the blind code was broken): mice that entered directly one of the closed arms from the starting central platform and spent the entire 10 min of the test in this closed arm were excluded from the analysis (4 out of 110 mice).

In vivo viral injection and micro-optic implantation

All procedures were performed similarly to published protocols.^{37,109,110,111} For stereotaxic surgery, animals were anesthetized with 2% isoflurane (Iso-Vet; Piramal Critical Care) and positioned into a digital stereotaxic frame (Stoelting) over a heating pad (∼37°C). After surgeries, the animals were maintained under a heating lamp until recovery. To reduce pain, inflammation and avoid infections, animals were administered with ketorolac (5 mg/kg), dexamethasone (5 mg/kg) and enrofloxacin (5 mg/kg). During the first surgery, the viral vector expressing GCamP6f¹¹² was injected into the dorsal CA1 hippocampal region. In brief, after exposing the skull, a small hole was drilled at the following coordinates relative to bregma¹¹³: antero-posterior: −1.94 mm; lateral: +1.20 mm. Animals were then injected with the GCaMP6f-expressing vector at −1.3 mm from the cortical surface using pulled glass capillaries connected to a Hamilton syringe mounted on a digital infusion micro-pump (Harvard Apparatus). Mice received 0.8 μL (flow rate: 0.1 μL/min) of Ready-to-Image virus AAV1.Camk2a.GCaMP6f.WPRE.bGHpA (Inscopix). After completing the infusion, the capillary was maintained in place for 10 min, and the skin was sutured upon the capillary removal. Upon recovery of the animals (3–4 weeks), a ProView Lens Probe (Inscopix) micro optic (1.0 mm diameter, 4.0 mm length) was implanted over the dorsal CA1 for imaging. To this aim, a 2.8 mm hole was drilled into the skull with a trephine drill over the stereotaxic coordinates where the GCaMP6f-expressing AAV was previously injected. To avoid compression of the CA1 by the lens, a cylindrical column of cortical tissue (∼2 mm diameter) and the upper part of the corpus callosum above the CA1 were aspirated with a 30G blunt-end needle, taking care of not damaging the tissue below. During surgery, the exposed brain tissue was constantly rinsed with ice-cold sterile PBS. Next, the nVista 3.0 miniaturized microscope (Inscopix) was used to place the lens probe over a suitable imaging field by assessing the overall fluorescence and the presence of blood vessels to be used as landmarks. Once completing the positioning, the exposed brain tissue around the base of the lens probe and the craniotomy were protected with Kwik-sil silicone (WPI). Then, the lens probe was permanently fixed to the skull with super-bond dental cement (Sun Medical). After an additional 4 weeks of animal recovery, a microscope magnetic baseplate was permanently fixed to the animal skull with the same dental cement. For this procedure, animals were lightly sedated with isoflurane and the miniaturized microscope was used again to guide the positioning of the baseplate over the lens probe to visualize the GCaMP6f-expressing neurons in the field of view. The magnetic baseplate attached to the skull allowed for repeated attachment and detachment of the microscope and thus the longitudinal imaging from the same CA1 field of view over multiple sessions. Between each imaging sessions, the microscope was removed and the top of the lens probe was protected with a magnetic cover sticking to the baseplate.

In vivo imaging sessions in the open field arena

Five days before starting the in vivo imaging experiments, mice were handled once a day for 5 min each day. Before the imaging experiments, mice were then habituated for 2-3 weeks to the microscope, wire, and arena. During the first week of familiarization, mice were first habituated for one day to an empty arena made of gray opaque acrylic (44 × 44 cm) for 30 min. On the next two days of habituation, a ‘‘dummy’’ plastic scope (Inscopix) mimicking the shape and weight of the actual microscope was attached to the baseplate and again the animals were left exploring the arena for 30 min. During the following two weeks, the animals were daily treated with i.p. injections of vehicle (2% DMSO in saline) and also subjected to four additional habituation session of 1 h each with the real microscope in order to familiarize to the microscope’s wire needed for data transfer. All habituation and imaging sessions were performed with dim light illumination (15–20 Lux).

To assess neuronal activity in the same neurons upon diazepam administration following vehicle or bumetanide treatment, we setup a longitudinal experiment consisting in two separate imaging sessions two weeks apart. During both imaging sessions, an ‘‘interrupted’’ imaging regime was used³⁷ to avoid potential photobleaching and/or phototoxicity deriving from exposure to the blue light needed to excite GCaMP6f fluorescence. The ‘‘interrupted’’ imaging regime consisted of active imaging periods (i.e., with blue light illumination) of 5 min alternated with 5 min of no light illumination.

On the first imaging session, animals were injected with vehicle (2% DMSO in saline) and placed after 10 min in the arena for 30 min during which neuronal calcium activity was recorded for a total of 15 min. Next, animals were injected with diazepam (Tocris; 2 mg/kg in saline) and again placed after 10 min in the arena for 30 min to record neuronal imaging data for further 15 min. Following this first imaging section, animals were daily treated with i.p. injections of bumetanide (0.2 mg/kg in saline) for two weeks. During this period, animals were also subjected to four additional habituation session of 1 h each with the real microscope in order to maintain the experimental routine and habit to the microscope and wire. On the second imaging session, animals were injected with bumetanide (0.2 mg/kg in saline) and placed after 10 min in the arena for 30 min during which neuronal calcium activity was recorded for 15 min. Finally, animals were injected with diazepam (2 mg/kg in saline) and placed after 10 min in the arena for 30 min to record neuronal imaging data for further 15 min.

Processing and analysis of in vivo calcium imaging videos

Videos of calcium activity acquired by the miniature microscopes were processed with Inscopix data processing software (IDPS) following the manufacture recommendations. First, videos of each imaging sessions were spatially down-sampled by a factor of two and spatially band-passed (High cutoff: 0.5; Low cutoff: 0.005). Videos were also corrected for motion artifacts using an image registration method¹¹⁴ and transformed to relative changes in fluorescence (ΔF/F0). The value of ΔF/F0 for each pixel in the movie was calculated as (F-F0)/F0, where F0 is the mean fluorescence signal of the pixel during the entire recording. Cells were identified by principal component analysis (PCA) and independent component analysis (ICA), as previously described,¹¹⁵ and the corresponding calcium traces were extracted. Next, the calcium traces from the two imaging sessions (vehicle plus diazepam, and bumetanide plus diazepam) were longitudinally concatenated in time. Traces were manually inspected to remove cells showing more than one independent component. Finally, calcium event features (i.e., starting time and amplitude) were automatically detected using changes in ΔF/F0 of at least 8 units of median absolute deviation (MAD) and a minimal decay time constant (τ) of 100 ms.¹¹²

The calcium event data were then analyzed with custom scripts in MATLAB (Mathworks). Cells presenting a signal-to-noise-ratio lower than 6 were excluded from the analysis. This threshold allowed the removal of neurons with low quality in terms of signal-to-noise-ratio or background fluctuations. To determine the cut-off parameter for exclusion, the overall distribution of the SNR for all recorded neurons was plotted and the threshold values selected to remove values less than the 10th percentile of the distribution. To quantify the level of activity, the mean events rate (MER, events/sec), defined as the mean number of calcium events in the diverse recording time windows, was computed. Neurons presenting a MER lower than 0.01 events/sec were excluded from the analysis to eliminate silent cells or cells showing very low activity. The same parameters were used for all animals. As previously described for cultures and slices, the MER ratio calculated as the MER during diazepam treatment divided by the MER during vehicle or bumetanide treatment was calculated. To quantify the mixed subpopulations of neurons the bootstrap methods used for cultures was adapted to the calcium imaging data. To this aim, four null distributions reflecting the overall changes in MER for all neurons in the four different experimental conditions (i.e., WT and Ts65Dn mice pretreated with vehicle plus diazepam or bumetanide plus diazepam) were computed as it follows. For each neuron belonging to the four experimental conditions, the event trains for each time segment (vehicle plus diazepam and bumetanide plus diazepam) were divided into 1-min bins. The MERs recorded during each 1-min bin for the two time segments for each neuron were randomly shuffled into two groups for 10,000 times. The differences between the MER of the randomly shuffled groups belonging to the same experimental conditions were used to generate the four corresponding null distributions. Then, for each neuron, the real difference between the MER during vehicle or bumetanide and the MER during vehicle plus diazepam or bumetanide plus diazepam was computed, and assessed whether it fell outside the 85% confidence interval of the corresponding null distribution. Thus, the overall percentage of neurons showing significant changes in the MER and belonging to the three different categories (i.e., increase, decrease or not change) was computed for each experimental condition.

To identify the subpopulations of neurons with different responses to bumetanide and/or diazepam treatment, two different ratios for each neuron were computed: i) ratio-1: the MER during bumetanide divided by the MER during vehicle; ii) ratio-2: the MER during diazepam plus bumetanide divided by the MER during diazepam plus vehicle. The first ratio reflected the variation in the neuronal activity elicited by bumetanide treatment in comparison to vehicle, while the second ratio returned the variation in the neuronal activity upon bumetanide plus diazepam administration in comparison to vehicle plus diazepam treatment. The two ratios were correlated in a scatterplot (i.e., ratio-1 on x axis, ratio-2 on y axis). The plot was divided into four quadrants based on the possible different neuronal activity in response to diazepam or bumetanide treatment over vehicle. Cells showing a decrease in the MER after both bumetanide and bumetanide plus diazepam treatments fell in quadrant 1 (decrease in both ratios). Cells showing an increase in MER after bumetanide and a decrease after bumetanide plus diazepam administrations fell in quadrant 2 (increase in ratio-1, decrease in ratio-2). Cells showing an increase in MER after both bumetanide and bumetanide plus diazepam administrations fell in quadrant 3 (increase in ratio-1, increase in ratio-2). Cells showing a decrease in MER after bumetanide and an increase after bumetanide plus diazepam administration fell in quadrant 4 (decrease in ratio-1, increase in ratio-2). Cells showing MER ratios changes below 10% were considered non-responsive to treatments and were excluded from further analysis. To predict the effect of diazepam plus bumetanide elicited in each Ts65Dn neuron based on MER variations during the previous sessions, a machine learning approach was implemented. For each neuron, six different features including MER, mean, minimum and maximum amplitude of the Ca²⁺ transients, mean and maximum distance between Ca²⁺ events were calculated. These features were computed in the first three temporal concatenated experimental phases (i.e., vehicle, diazepam and bumetanide) while the features in bumetanide plus diazepam phase were hidden to the classification algorithm. Then, the Ts65Dn dataset was randomly divided in the training and test sets, representing respectively the 54% and 46% of the total dataset. For simplicity, the machine learning classificatory was asked to predict only two possible output labels (categories: an increase (>1) or a decrease (<1) of the MER during diazepam plus bumetanide treatment, when compared to the diazepam plus vehicle. Both datasets contained a comparable percentage of neurons belonging in the two categories (training set: 65% of <1 category and 35% of >1 category; test set: 58% < 1 category and 45% > 1 category). Supporting Vector Machine classificator (SVM,¹¹⁶) was trained to predict the categorical label (<1 or >1) of the neuron from the feature vectors. The accuracy of the classificator in the test set was tested using the MATLAB function “classperf”, which computed the following parameters: the correct rate (i.e., number of correctly classified samples divided by the number of classified samples), the error rate (i.e., number of incorrectly classified samples divided by the number of classified samples), sensitivity (i.e., number of correctly classified positive samples divided by the number of true positive samples), specificity (i.e., number of correctly classified negative samples divided by the number of true negative samples), positive predictive value (i.e., number of correctly classified positive samples divided by the number of positive classified samples), negative predictive value (i.e., number of correctly classified negative samples divided by the number of negative classified samples), predicted likelihood (1 - Sensitivity/Specificity) and prevalence (number of true positive samples divided by the total number of samples; Table S2).

Analysis of the literature

To classify cells into neurons characterized by hyperpolarizing or depolarizing GABA-mediated responses, we relied on explicit reports from the literature. Those included (Ca²⁺ or Cl⁻) imaging and electrophysiological (gramicidin-perforated, GABA_AR single-channel currents, or cell-attached patch-clamp) datasets obtained from rodent neuronal primary cultures, brain slices, and in vivo experiments, as well as human neurons derived from control iPSC lines. When the classification of cells showing depolarizing and hyperpolarizing GABA responses was not explicitly reported by the authors, we examined the single-neuron values directly in the figures and calculated the approximate percentage of neurons that fell into the two different categories (i.e., cells with hyperpolarizing vs. depolarizing GABA responses) ourselves. For Ca²⁺ imaging experiments, we considered an increase in intracellular Ca²⁺ following GABA_AR agonist treatments as indicative of a depolarizing GABA response. For Cl⁻ imaging, we classified neurons with predicted depolarizing GABA responses as those presenting a high intracellular Cl⁻ concentration (>40 mM, as seen during neurodevelopment), which would likely lead to an outward (depolarizing) flow of Cl⁻ through GABA_ARs. Regarding electrophysiological gramicidin-perforated or GABA_AR single channel currents, we classified neurons with depolarizing GABA responses those displaying an E_GABA higher than the cell resting membrane potential (RMP). For cell-attached patch-clamp recordings, we considered neurons with depolarizing GABA responses those showing an increase in the frequency of action potentials in response to a GABA_AR agonist and those showing a decrease in the frequency of action potentials in response to a GABA_AR antagonist.

Quantification and statistical analysis

All behavioral and biochemical analyses were performed blinded by the operator. Except were otherwise stated, the results were presented as the means ± SEM. Statistical analysis was performed using SigmaPlot (Systat) or GraphPad (Prism) software. Where appropriate, the statistical significance was assessed using the following parametric test: Student’s t test or two-way ANOVA followed by all pairwise Turkey’s post hoc test. In case normal distribution or equal variance assumptions were not valid, statistical significance was evaluated using the following nonparametric tests: Mann-Whitney Rank-Sum Test or two-way ANOVA on ranked transformed data. Post-hoc tests were performed when at least one of the three sources of variation (genotype, treatment, genotype x treatment) was significant. For statistical evaluation of in vivo calcium imaging longitudinal data, we used the linear mixed model (LME) from nlme package in R (http://CRAN.R-project.org), as recently suggested.¹¹⁷ Specifically, we fitted the LME using genotype and treatment as fixed effects, and cell number and mouse ID as nested random effects. Statistical difference in the percentage of neurons (Figures 4; 5) were analyzed by Chi-squared test with Sidak adjustment for multiple comparisons. For all tests, p values <0.05 were considered significant. All the statistical information is summarized in Tables S7 and S8. Statistical details of experiments can be found in figure legend for each figure.

Published: March 8, 2024