Abstract
During sleep and anesthesia, neocortical neurons exhibit rhythmic UP/DOWN membrane potential states. Although UP states are maintained by synaptic activity, the mechanisms that underlie the initiation and robust rhythmicity of UP states are unknown. Using a physiologically validated model of UP/DOWN state generation in mouse neocortical slices whereby the cholinergic tone present in vivo is reinstated, we show that the regular initiation of UP states is driven by an electrophysiologically distinct subset of morphologically identified layer 5 neurons, which exhibit intrinsic rhythmic low-frequency burst firing at ∼0.2–2 Hz. This low-frequency bursting is resistant to block of glutamatergic and GABAergic transmission but is absent when slices are maintained in a low Ca2+ medium (an alternative, widely used model of cortical UP/DOWN states), thus explaining the lack of rhythmic UP states and abnormally prolonged DOWN states in this condition. We also characterized the activity of various other pyramidal and nonpyramidal neurons during UP/DOWN states and found that an electrophysiologically distinct subset of layer 5 regular spiking pyramidal neurons fires earlier during the onset of network oscillations compared with all other types of neurons recorded. This study, therefore, identifies an important role for cell-type-specific neuronal activity in driving neocortical UP states.
Introduction
In the absence of sensory input, the mammalian brain exhibits a wide array of structured, spontaneous activity. An example of this is the alternation of neocortical UP and DOWN states, which are periods of persistent, widespread network activity and collective neuronal disfacilitation, respectively (Crunelli and Hughes, 2010; Timofeev and Chauvette, 2011). Such UP and DOWN states are evident during non-REM sleep (Steriade et al., 2001; Timofeev et al., 2001; David et al., 2013) and under anesthesia where they occur rhythmically at ∼0.2–1 Hz to form a slow oscillation (Steriade et al., 1993c; David et al., 2013). Up until now, synchronous UP and DOWN states have not been observed in neocortical slices maintained using a traditional recording medium (i.e., containing 2 mm Ca2+) but have required a medium containing 1–1.2 mm Ca2+ that is purported to more closely reflect the in vivo ionic composition (Sanchez-Vives and McCormick, 2000; Cossart et al., 2003; Shu et al., 2003; see also Cunningham et al., 2006). Although this approach reproduces several properties of the in vivo UP/DOWN states, DOWN states are typically much more prolonged than in vivo (Sanchez-Vives and McCormick, 2000); and when applied in rodent neocortical slices, UP states tend to be sporadic and lack rhythmicity (e.g., Cossart et al., 2003; Fanselow and Connors, 2010). Thus, the mechanisms underlying the initiation and robust rhythmicity of UP states have remained elusive.
In anesthetized cats, UP and DOWN state duration is dramatically reduced by mAChR antagonists, suggesting a role for the cholinergic system in UP state generation (Steriade et al., 1993b). Although neurons in both brainstem cholinergic nuclei (Steriade et al., 1990) and basal forebrain (Szymusiak et al., 2000; Lee et al., 2004, 2005; Hassani et al., 2009) generally fire substantially less during sleep and drowsiness, their firing does not entirely cease. Indeed, although cortical ACh levels are diminished during slow-wave sleep, they nonetheless remain at significant levels (Marrosu et al., 1995; Vazquez and Baghdoyan, 2001). Furthermore, functional imaging in humans reveals a strong activation of brainstem regions that encompass cholinergic nuclei during slow waves (Dang-Vu et al., 2008).
Here we show that pharmacological reactivation of the cholinergic input in mouse neocortical slices brings about rhythmic UP and DOWN states with properties that are equivalent to those observed during the slow oscillation in vivo. We further demonstrate that UP states are driven by low-frequency (∼0.2–2 Hz) rhythmic burst firing in a small subset of layer 5 neurons, which in these, but not in the remaining intrinsically bursting (IB) pyramidal neurons, is resistant to GABA and ionotropic glutamate receptor antagonists. Indeed, engagement of these putative “network driver” neurons by cholinergic activation is essential for bringing about the strong rhythmicity of UP states that is present in vivo. In contrast, a compromise of the “network driver” bursting in the low Ca2+ model explains the inability to consistently reproduce this key characteristic of the natural slow oscillation in previous in vitro studies using this model.
Materials and Methods
All procedures were performed in accordance with local ethical committee guidelines and the United Kingdom Animals (Scientific Procedure) Act, 1986. All efforts were made to minimize the suffering and number of animals used in each experiment.
Surgical procedures and in vivo electrophysiology.
Male C57BL/6J mice (P30-P60, Harlan) were anesthetized with urethane (1.0 g/kg) and supplemented with ketamine and xylazine (20 and 2 mg/kg, respectively) to maintain anesthesia. Body temperature was maintained at constant levels using a heating plate (Supertech). After exposing the skull, a small hole was drilled above the target area at stereotaxic coordinates (anteroposterior: −3.1; lateral: 3.5–3.9 mm). Stainless steel screws (0.8 mm diameter) were placed into the skull over the cerebellum and the frontal cortex, which served as a ground and reference, respectively. Local field potentials (LFPs, filtered between 0.1 and 200 Hz) and extracellular single units (filtered between 0.5 and 5 kHz) were recorded using glass pipettes filled with ACSF (resistance: 500–600 kΩ for LFP, 4–6 mΩ for single units) connected to a Multiclamp 700B amplifier (Molecular Devices). Extracellular action potentials were recorded in loose-cell mode (seal resistance: 20–50 mΩ). Intracellular recordings were performed using glass electrodes filled with 1 m potassium acetate (resistance: 15–40 mΩ), and in some cases 2% biocytin or neurobiotin, and connected to an Axoclamp-2A amplifier (Molecular Devices) operating in bridge mode. Voltage records were digitally acquired and processed using pClamp 9 (Molecular Devices). All in vivo intracellular and extracellular recordings were obtained from the primary auditory cortex (Crunelli et al., 2012). Scopolamine (1 mm) was added to the ACSF present in the recording chamber (constructed from dental acrylic cement) that was fixed to the mouse skull. At the end of the experiments, the animals were given a lethal dose of urethane.
In vitro slice preparation and maintenance.
Male C57BL/6J mice (P21-P60, Harlan) were deeply anesthetized with isoflurane, decapitated, and the brain quickly removed into continuously oxygenated (95% O2/5% CO2) 4°C cutting solution containing the following (in mm): 60 sucrose, 85 NaCl, 2.5 KCl, 1.25 NaH2PO4, 2 MgCl2, 25 NaHCO3, 1 CaCl2, 25 NaHCO3, 10 glucose, 0.045 indomethacin, and 3 kynurenic acid. Coronal slices (400 μm thick) containing the auditory, somatosensory, visual, temporal association, or medial prefrontal cortex were cut on a vibrotome (Leica). Slices were allowed to recover for 60 min at room temperature. For recording, slices were perfused in an interface chamber with a warmed (35 ± 1°C) continuously oxygenated (95% O2, 5% CO2) ACSF containing the following (in mm): 134 NaCl, 2 KCl, 1.25 KH2PO4, 1 MgSO4, 2 CaCl2, 26 NaHCO3, and 10 glucose. Drugs were dissolved directly in either ACSF or in DMSO, in which case the concentration of DMSO did not exceed 0.1% v/v. The modified ACSF contained (mm) the following: 125 NaCl, 3.5 KCl, 1.25 NaH2PO4, 1 MgCl2, 1.0–1.2 CaCl2, 25 NaHCO3, and 10 glucose.
In vitro electrophysiology.
Extracellular recordings were performed using glass pipettes filled with ACSF (resistance: 1–5 mΩ) connected to a Neurolog 104 differential amplifier (Digitimer). Field and unit activities were simultaneously recorded through the same electrode by bandpass filtering at 0–200 Hz and 0.2–5 kHz, respectively. Multisite extracellular recordings were performed with linear arrays (FHC) connected to a multichannel differential amplifier (Plexon or Neuralynx). In all cases, the interelectrode distance was 190 μm. Independently mounted intracellular recordings, using the current-clamp technique, were performed with standard-wall glass microelectrodes filled with 1 m potassium acetate (resistance: 80–120 mΩ), and in some cases 2% biocytin or neurobiotin, and connected to an Axoclamp-2A amplifier (Molecular Devices) operating in bridge mode. Voltage and current records were digitally acquired and processed using pClamp 9 (Molecular Devices). Unless explicitly stated otherwise, all in vitro recordings were obtained from slices of the primary auditory cortex.
Data analysis.
UP states from the LFP signal were detected from the high pass filtered and trace (Mukovski et al., 2007), and a minimum of 100 cycles for each recorded neuron were used to perform statistical analysis. Phase values of neuronal firing relative to the EEG or LFP, expressed as the angle of the mean vector (μ), and strength of phase coupling, expressed as the normalized mean vector length (r), were computed with circular statistical methods using Oriana 2.0 software (Kovach Computing Services). The range of r is 0 to1, with a value of 1 representing perfect phase coupling, whereas when events are random r → 0. Absolute spike times were established using a straightforward visually determined threshold approach. To construct a spike-timing histogram, the times of >500 consecutive spikes were determined relative to the nearest negative peaks of the LFO using custom-written transform routines in SigmaPlot 9 (Systat). These times were subsequently assigned a given phase between these peaks (i.e., between 0° and 360°), binned at 20°. The values in each bin were then divided by the total number of oscillation cycles to give a value in units of spikes per bin. For clarity, and to afford a sense of rhythmicity, these data were plotted over two full cycles of the oscillation (−360° to 360°) to produce the final plot (Lőrincz et al., 2009). For measuring the membrane potential of UP and DOWN states, we generated a membrane potential distribution plot and used the peaks (or peak for nonbimodal UP/DOWN states) from these plots to produce these values. All auto-correlation, cross-correlation, and power spectra plots were produced with OriginPro 8.0 (OriginLab). All quantitative data are expressed as mean ± SEM, and statistical significance was assessed using Student's t test.
Histology.
Intracellularly recorded cells were labeled by applying depolarizing current steps (0.3–0.6 nA, 500 ms, 1 s duty cycle) for 10 min through the bridge circuitry. Slices were fixed in 4% PFA overnight. Cells were visualized using the avidin-biotin-HRP reaction as previously described (Lőrincz et al., 2009) and were reconstructed using Neurolucida (MicroBrightField Europe).
Sources of drugs.
dl-APV (NMDA receptor antagonist), CNQX (AMPA/kainate receptor antagonist), CGP 54626 hydrochloride (GABAB antagonist) 1,1-dimethyl-4-diphenylacetoxypiperidinium iodide (4-DAMP) (M3 receptor antagonist), 8-methyl-8azabicyclo-3-endo[3,2,1]oct-3-yl-1,4-dihydro-2-oxo-3(2H)-quinazolinecarboxylic acid ester hydrochloride (DAU-5884) (M3 receptor antagonist), 2,3-dihydroxy-6-nitro-7-sulfamoyl-benzo[f]quinoxaline-2,3-dione (NBQX) (AMPA/kainate receptor antagonist), 6-imino-3-(4-methoxyphenyl)-1(6H)-pyridazinebutanoic acid hydrobromide (SR95531, gabazine) (GABAA receptor antagonist), and 4,9-dihydro-3-methyl-4-[(4-methyl-1-piperazinyl)acetyl]-10H-thieno[3,4-b][1,5]benzodiazepin-10-one dihydrochloride (telenzepine) (M1 receptor antagonist) were from Tocris Cookson; carbamylcholine chloride (carbachol, CCH) (nonselective cholinergic agonist) and 5,11-dihydro-11-([4-methyl-piperazin]acetyl)-6H-pyriso(2,3-b) (1,4) benzodiazepine-6-one (pirenzipine) (M1/M3 receptor antagonist) were obtained from Sigma.
Results
UP and DOWN states in neocortical neurons of anesthetized mice
Intracellular recordings from neocortical neurons in layer 5 obtained under anesthesia exhibited prominent, rhythmic UP/DOWN states (0.85 ± 0.04 Hz; n = 20) (Fig. 1A,C). UP states in neocortical neurons were characterized by brief barrages of synaptic activity that were accompanied by prominent action potential firing (Fig. 1A,C). During the course of our experiments, we encountered both regular spiking (RS) (n = 15) (Fig. 1A) and IB (n = 19) (Fig. 1C) neocortical neurons. In the majority of RS cells (n = 11 of 15; 73%), UP states arose from a relatively hyperpolarized baseline and led to clear membrane potential bimodality (DOWN state: −73.7 ± 2.5 mV; UP state: −61.2 ± 1.4 mV; n = 8) (Fig. 1A1). In other RS cells, UP states could also emerge from a more depolarized baseline leading to an undulating appearance and a lack of membrane potential bimodality (n = 3 of 15; 20%) (Fig. 1A2), as seen previously in cats (Steriade et al., 1993a). Indeed, in some RS cells, UP states essentially appeared on top of a baseline (i.e., DOWN state) that already exhibited some tonic action potential output (Fig. 1A2). Such a pattern of activity is not due to damage done to the cell by intracellular impalement because it has also been observed in single-unit extracellular recordings (Crunelli et al., 2012). When RS cells were considered together, action potential output occurred primarily close to the negative peaks of the LFP (μ = −4.3 ± 6.6°; n = 15), with cells showing a normalized vector length (r) of 0.61 ± 0.09 (n = 15) (Fig. 1B).
In contrast to RS cells, UP states in IB cells always emerged from a relatively hyperpolarized baseline (Fig. 1C). As such, action potential activity was rarely observed in these cells during the DOWN state, and they were typically associated with membrane potential bimodality (DOWN state: −74.3 ± 2.2 mV; UP state: −60.1 ± 1.9 mV; n = 19) (e.g., Fig. 1C1). A robust burst was almost always present at the beginning of an UP state in IB neurons (Fig. 1C1,C2, insets) (Steriade et al., 1993a). In these cells, UP states were also often preceded by a buildup of spontaneous EPSPs (Fig. 1C2, inset, arrows). Firing in IB cells was found to occur slightly earlier than in RS cells (μ = −15.8 ± 6.9°; n = 6, p = 0.03) (Fig. 1D). There was no difference in the strength of phase coupling for IB cells compared with RS cells (r = 0.58 ± 0.07; n = 19; p > 0.5).
Neocortical UP states in anesthetized mice are dependent on mAChRs
In anesthetized cats, systemic administration of the mAchR antagonist scopolamine leads to a substantial reduction in the duration of both UP states and DOWN states, indicating an important role for mAChRs in controlling UP states in this species (Steriade et al., 1993b). To test whether mAChRs also influence neocortical UP states in mice, we locally administered scopolamine to the auditory cortex of anesthetized animals (1 mm; see Materials and Methods) while monitoring UP states using extracellular LFP and single-unit recordings (Fig. 2). The effects of local scopolamine administration were clearly evident in the LFP where they were reflected as a reduction in the duration of both UP (control: 1.2 ± 0.1 s; scopolamine: 0.7 ± 0.1 s; n = 5; p < 0.001) and DOWN states (control: 1.5 ± 0.1 s, scopolamine: 1.1 ± 0.1 s; n = 5; p < 0.05) (Fig. 2). The associated action potential output during UP states also decreased (control: 0.1 ± 0.05 Hz; scopolamine: 0.04 ± 0.02 Hz; n = 6 neurons; p < 0.001).
Activation of cholinergic receptors brings about a slow (<1 Hz) oscillation and associated rhythmic UP/DOWN states in mouse neocortical slices
Given that neocortical UP states in anesthetized mice are suppressed by scopolamine administration, we examined the effects of the nonselective AChR agonist carbachol (CCH, 30–50 μm) on network excitability in slices of the primary auditory cortex from adult mice. When assessed with extracellular multielectrode array recordings in control conditions, there was a lack of spontaneous activity in all slices (n = 140 slices) (Fig. 3A–C). Correspondingly, intracellular recordings from neocortical neurons revealed a hyperpolarized membrane potential (<−60 mV), a lack of spontaneous action potential firing, and only sporadic, uncoordinated synaptic activity (Fig. 3C). Immediately following CCH application, extracellular recordings revealed the appearance of LFP oscillations at ∼2–4 Hz in layer 5 (Fig. 3A1). Over the subsequent course of a few minutes, these LFP oscillations increased in amplitude, slowed to ∼0.6 Hz (0.66 ± 0.17 Hz; n = 10), and became evident in other cortical layers (Fig. 3A2).
When assessed with extracellular recordings, CCH application was associated with a rapid and pronounced increase in spontaneous action potential firing that almost exclusively consisted of single spike or tonic activity (9.77 ± 6.77 Hz, n = 32 neurons of 150, 21%) that was present even before the emergence of LFP oscillations (Fig. 3B, top right). Neurons that generated this firing were termed “early firing” cells. Interestingly, upon the emergence of LFP oscillations at 2–4 Hz (Fig. 3B, middle), activity in “early firing” cells was usually not immediately entrained to the LFP fluctuations (Fig. 3B, middle) and often only became fully engaged in network activity later on when slow (<1 Hz) oscillations dominated the LFP (Fig. 3B, bottom). In contrast to “early firing” cells, in a separate group of neurons, a distinct type of firing consisting primarily of brief bursts was initiated simultaneously with the onset of LFP oscillations (n = 12 neurons of 150, 8%), with this firing being robustly synchronized with the negative LFP peaks (Fig. 3C, middle). These neurons were termed putative “network drivers.” Apart from “early firing” cells and “network drivers,” all other cells (n = 106) commenced firing only after oscillation onset had occurred, at which point it was immediately synchronized with the LFP (e.g., Fig. 3C, intracellular). For such cases, simultaneous intracellular recordings revealed the appearance of rhythmic EPSPs phase-locked to 2–4 Hz oscillations, followed by firing and full-blown UP states when slow (<1 Hz) oscillations were fully developed (Fig. 3C, intracellular). The transformation of 2–4 Hz oscillation to slow (<1 Hz) oscillations was gradual (Fig. 3D).
Importantly, CCH-elicited network oscillations and correlated cellular firing, observed in the auditory cortex, could also be recorded in medial prefrontal, somatosensory, visual, and temporal association cortex (data not illustrated). In agreement with our in vivo results, CCH-elicited network oscillations could be blocked by the nonselective mAchR antagonist scopolamine (5 μm, n = 5), the M1 antagonists pirenzipine (1 μm, n = 6), telenzepine (2 μm, n = 5), and by the M3 receptor-preferring antagonists DAU-5884 (2 μm, n = 5) and 4-DAMP (1 μm, n = 4) (data not illustrated) indicating a reliance of UP states on both types of muscarinic receptors. Thus, pharmacologically reinstating the cholinergic drive with CCH is sufficient to bring about slow network oscillations and rhythmic UP/DOWN state dynamics in slices of the mouse neocortex.
Further analysis of network activity using extracellular multielectrode recordings, obtained with 8, 16, or 32 channel arrays, revealed that CCH-induced UP states showed widespread synchronization across the cortical slice being apparent at all recording sites (Figs. 3A and 4). Overall, the most pronounced activity was recorded in deep layers (i.e., layer 5 and 6; Fig. 4A, top), with current source density analysis revealing that the initiation of each oscillatory cycle was associated with a prominent current sink around layer 5 (Fig. 4A, bottom left). To estimate the relative timing of synchronized activity in regional neuronal populations, we examined cross-correlograms of the LFP recorded in different layers. In accordance with other in vivo and in vitro studies (Sanchez-Vives and McCormick, 2000; Sakata and Harris, 2009; Beltramo et al., 2013; Hughes and Crunelli, 2013), these revealed that, in the majority of cases, the oscillation starts in layer 5 and then spreads to layer 6 before propagating to more superficial layers (Fig. 4A, bottom right).
The propagation of oscillatory activity from layer 5 to layer 6 and subsequently to superficial layers observed for LFPs (Fig. 4A, bottom right) also held true when the timing of multiple single-unit extracellular action potentials recorded from different cortical layers was compared using multielectrode arrays: in the majority of cases (n = 14 of 17), layer 5 neurons discharged before neurons in deeper and more superficial layers (Fig. 4A, bar graph). However, in a few cases (n = 3 of 17), the layer of initiation could alternate in a seemingly random manner between layer 5 and layer 2/3 (Fig. 4B). Interestingly, in those cases where UP states appeared to commence in layer 2/3, the duration of UP states in this layer was significantly longer than when UP states commenced in layer 5 (524 ± 96 ms vs 310 ± 28 ms; n = 30 UP states from 3 slices; p < 0.05) (Fig. 4C, top right plot). Overall, the mean firing rate during UP states was significantly higher in layer 5 than in layer 2/3 (layer 5: 22.6 ± 4.9 Hz; layer 2/3 12.3 ± 3.3 Hz, n = 50 UP states from 5 slices; p < 0.001) (Fig. 4C, bottom plots).
As is the case in vivo (Steriade et al., 1996; Hasenstaub et al., 2005; Mukovski et al., 2007; Ruiz-Mejias et al., 2011; Crunelli et al., 2012), UP states induced by CCH in neocortical slices were associated with prominent high-frequency (∼20–100 Hz) oscillations in the LFP (Fig. 5A). These oscillations occurred at a mean frequency of 52.4 ± 1.5 Hz (n = 57) (Fig. 5B) and were tightly synchronized within local neocortical areas (mean time lag in cross-correlation over 190 μm: 0.45 ± 0.08 ms; n = 20 pairs of recordings from 4 slices) (Fig. 5C). Interestingly, when assessed with linear electrode arrays placed approximately parallel to the cortical layers, epochs of high-frequency oscillations tended to randomly switch between different directions of propagation (Fig. 5C, top and middle traces) or appear to not propagate at all (Fig. 5C, bottom traces). This was not related to the propagation of the slow oscillation (i.e., underlying UP states), which typically remained relatively constant in this orientation (Fig. 5C, red dots, D). In addition, regardless of the manner in which epochs appeared to propagate, the tight local synchrony of high-frequency oscillations, the duration of oscillation epochs and the mean frequency within epochs remained unchanged (Fig. 5C, far right column, D). Simultaneous intracellular and extracellular recordings (Fig. 5E) showed that high-frequency oscillations in the LFP were clearly correlated with synaptic fluctuations (as shown for an RS cell in Fig. 5E1,E2) and therefore also action potential output in individual pyramidal neurons (Fig. 5E, bottom right). All types of cells recorded in this study (i.e., RS, IB, and nonpyramidal neurons) showed evidence of phase-coupling of their spikes to high-frequency oscillations (Fig. 5F).
In addition to layer 5 IB and RS pyramidal neurons, we also obtained intracellular recordings in neocortical slices from a variety of nonpyramidal neurons in infragranular cortical layers (i.e., 5/6; Fig. 6). Recordings of nonpyramidal neurons were subdivided into three groups: FS (fast-spiking) neurons, LTS (low threshold-spiking) neurons, and UC (unclassified) cells, some of which had late-spiking characteristics but did not consistently possess the morphology that identified them as a well-characterized neocortical cell class (Fig. 6C, top). All FS neurons were tightly phase-coupled to the LFP (r = 0.58 ± 0.09; n = 4) with action potential output being maximal before the negative LFP peak (μ = −23.4 ± 29.4°; n = 4) (Fig. 6A). In contrast, LTS neurons showed much weaker phase-coupling to the LFP (r = 0.26 ± 0.06; n = 9; p < 0.05) and exhibited no obvious peak in firing relative to the LFP (Fig. 6B). Phase-coupling for UC neurons was not significantly different from FS neurons (r = 0.76 ± 0.11; n = 5; p = 0.23) but was significantly stronger than for LTS cells (p < 0.005) (Fig. 6C). Peak firing in UC cells occurred close to the negative LFP peak (μ = −17.4 ± 29.3°; n = 5) (Fig. 6C). Finally, in all nonpyramidal neurons recorded (n = 16; see Fig. 6), development of UP states and engagement in network oscillations occurred via the gradual increase in the amplitude of LFP-related postsynaptic potentials (data not illustrated).
Putative “network drivers” are a subset of layer 5 IB pyramidal neurons
To understand the mechanisms that lead to network oscillations and UP/DOWN states in neocortical slices, we obtained intracellular recordings of neocortical neurons, both during the onset of LFP oscillations and subsequently for a prolonged period when the slow (<1 Hz) oscillation has been fully established. Of our total sample of 86 intracellularly recorded neurons, 6 (i.e., 7%) were identified as putative “network driver” neurons (i.e., cells where burst firing commenced simultaneously), and in synchrony, with 2–4 Hz LFP oscillations (delay from LFP onset: 3.3 ± 1.8 s, n = 6) (Figs. 3C and 7A,C; see 12D). All “network driver” cells were layer 5 IB pyramidal neurons that in control conditions were able to exhibit intrinsic rhythmic bursting at ∼0.2–2 Hz in response to positive d.c. current injection (Fig. 8A). Following application of CCH, these cells were depolarized (control: −76.8 ± 4.4 mV; CCH: −62.4 ± 3.5 mV; n = 6) and became able to generate these rhythmic bursts spontaneously, with bursting occurring immediately in close association with negative peaks in the LFP (Fig. 8A, right; see also Fig. 12D). After a short period (∼30–60 s), these bursts were occasionally closely followed by an additional depolarization indicative of recurrent excitation (Fig. 7A, middle, arrows). Over the course of the subsequent few minutes, these additional depolarizations became increasingly common and prolonged, eventually developing into full-blown, bimodal UP states that were present on every cycle of the oscillation (DOWN state: −70.4 ± 0.7 mV; UP state: −52.4 ± 1.5 mV; n = 6) (Fig. 7A, right). “Network driver” neurons never changed category over the entire duration of the recordings.
Whereas all “network driver” cells were found to be layer 5 IB pyramidal cells, not all layer 5 IB pyramidal neurons commenced firing simultaneously with the onset of LFP oscillations, with some of these cells being recruited to fire only after (delay from LFP onset: 30.5 ± 10.4 s; Fig. 7C), the LFP oscillation had been initiated (8 of 86; 9%; Fig. 7D). Following a similar CCH-induced depolarization to that which occurs in “network driver” cells (control: −73.6 ± 2.7 mV; CCH: −67.7 ± 3.5 mV; n = 8), this recruitment occurred via the gradual emergence, and increase in the amplitude, of LFP-associated rhythmic EPSP complexes in these cells (Fig. 7D, left). However, once fully recruited into slow (<1 Hz) oscillations, UP states in these neurons displayed a similar form to “network driver” cells in that they were always initiated by a burst of action potentials (Fig. 7D, middle and right) and were typically associated with membrane potential bimodality (DOWN state: −69.9 ± 0.7 mV; UP state: −53.3 ± 1.1 mV; n = 8) (Fig. 7B,E). As such, there was no difference in the timing of action potential output relative to the LFP between “network driver” and non-“network driver” IB cells with firing occurring at a mean phase (μ) of −18.9 ± 7.3° (n = 14; 6 “network driver” neurons, 8 conventional IB cells) and showing a normalized vector length (r) of 0.91 ± 0.02 (n = 14) (Fig. 7F). An additional shared feature was the common presence of a buildup of EPSPs just before the generation of an UP state (Fig. 7D, middle and right) as is also present in vivo (Fig. 1C, bottom). We also observed no obvious differences in cell morphology between “network driver” cells and other IB neurons, with all cells showing morphology typical of layer 5 IB pyramidal cells reported previously (Chagnac-Amitai et al., 1990; Franceschetti et al., 1993), comprising a relatively thick apical dendrite, prominent apical dendritic tuft, and an extensively arborizing axon in deep cortical layers (Fig. 8A,B, right).
Thus, the most apparent difference between “network driver” cells and the remainder of layer 5 IB pyramidal neurons is that the former respond more readily to CCH with low-frequency (∼0.2–2 Hz) rhythmic bursting. Importantly, this is due solely to a direct depolarization of these neurons rather than a modulation of their intrinsic properties because, in the absence of CCH, intrinsic low-frequency rhythmic bursting could be readily brought about by injecting a small amount of steady depolarizing current (Fig. 8A), with larger amounts of current leading to single spike activity (compare Wang and McCormick, 1993; Schwindt et al., 1997). In contrast, in IB cells that are not “network driver” cells, burst firing occurred at notably higher frequencies than in “network driver” cells (Fig. 8B,C) (compare Schwindt et al., 1997).
“Early firing” cells are a subset of layer 5 RS pyramidal neurons
We next turned our attention to “early firing” cells: neurons in which preestablished (delay of LFP onset from start of firing: 123.45 ± 19.96 s, n = 11) (Figs. 3B and 9A,C) spontaneous single spike firing is gradually entrained to network activity. Overall, 26% (i.e., 22 of 86) of intracellularly recorded neurons were of this type. All “early firing” cells were layer 5 RS pyramidal neurons (Fig. 9A, left). Following application of CCH, these cells were rapidly and strongly depolarized (control: −66.6 ± 5.3 mV; CCH: −52.5 ± 3.1 mV; n = 11), leading to continuous firing (9.7 ± 2.29 Hz; n = 11) (Fig. 9A, left). As stated above, the subsequent onset of LFP oscillations was not always immediately associated with an appreciable change in this firing. However, as the slow (<1 Hz) oscillation in the LFP developed, the firing of these neurons became entrained to network oscillations (Fig. 9A, middle and right). On each oscillation cycle, this was characterized by a progressive buildup of firing followed by a pronounced hyperpolarization that occurred after the negative LFP peak (Figs. 9A, right, and 10A, top). UP states in these cells often emerged from an already relatively depolarized baseline and were not usually associated with overt membrane potential bimodality (Fig. 10A, top right), as is also sometimes the case for RS cells recorded in vivo (e.g., Fig. 1A2) (see also Crunelli et al., 2012). However, steady hyperpolarization of these neurons via negative d.c. current injection brought about conventional-looking UP states that were associated with membrane potential bimodality (Fig. 10A, middle). For “early firing” neurons, action potential output occurred at a mean phase (μ) of −43.1 ± 15.5° (n = 6) relative to LFP oscillations with cells showing a normalized vector length (r) of 0.51 ± 0.08 (n = 6) (Fig. 9B).
Although all “early firing” cells were found to be layer 5 RS pyramidal cells, not all layer 5 RS pyramidal neurons were “early firing” cells with a substantial proportion of RS neurons (32 of 86; 37%) being entrained into the oscillation through the progressive development of rhythmic EPSP complexes (Fig. 9D, left) that appeared following a small CCH-induced depolarization (control: −68.6 ± 5.0 mV; CCH: −64.5 ± 5.7 mV; n = 10). Firing in these neurons commenced after the LFP slow (<1 Hz) oscillation has been developed (delay of firing from LFP onset: 40.3 ± 7.73 s; Fig. 9C). In these cells once UP states were fully developed, they exhibited a conventional appearance and were associated with membrane potential bimodality (DOWN state: −67.1 ± 0.8 mV; UP state: −55.3 ± 0.6 mV; n = 10; Figs. 9D, right, and 10B, middle). However, steady depolarization of these neurons via positive d.c. current injection brought about UP states that were indistinguishable from those exhibited by “early firing” neurons (Fig. 10B, top). In non-“early firing” RS cells, action potentials occurred at a mean phase (μ) of 11.1 ± 10.9° (n = 6) and cells showed a normalized vector length (r) of 0.93 ± 0.02 (n = 6) (Fig. 9E). Thus, the main apparent difference between “early firing” cells and the remainder of layer 5 RS pyramidal neurons is the degree to which CCH causes a direct depolarization and spontaneous firing (Fig. 9A,D).
Blocking excitatory and inhibitory synaptic activity differentially affects UP/DOWN states in layer 5 RS and IB pyramidal neurons
In all intracellularly recorded cells in neocortical slices with added CCH, the frequency of UP states was unaltered by changes in steady injected current, further illustrating that they are dependent on network mechanisms (e.g., Fig. 10A,B). Indeed, in all neurons, hyperpolarization with steady current to prevent action potential generation revealed rhythmic barrages of synaptic activity (Fig. 10A,B, bottom traces). Consistent with these observations and further substantiating that oscillatory activity is fundamentally driven by synaptic interactions between neocortical neurons, blocking ionotropic glutamate and GABA receptors by applying a combination of 10 μm NBQX, 50 μm APV, 10 μm gabazine, and 10 μm CGP abolished the generation of rhythmic fluctuations in the LFP (n = 16 slices) (Fig. 11A,B).
Although oscillatory activity in the LFP was blocked in all cases by blocking fast excitatory and inhibitory synaptic transmission, this manipulation affected individual layer 5 pyramidal neurons in distinct ways. First, in IB neurons that were “network driver” cells, blocking synaptic transmission caused a progressive shortening of UP states until these neurons exhibited low-frequency burst firing only (n = 4) (Fig. 11A1). This confirmed that the direct postsynaptic effect of CCH in these cells was to bring about intrinsic rhythmic bursting. Second, in all other IB neurons (n = 4) and in RS neurons, which were not “early firing” cells (n = 4), blocking synaptic transmission caused a gradual reduction in the amplitude of rhythmic synaptic barrages and, consequently, an abolition of action potential firing (Fig. 11A2,B2). Third, in “early firing” RS neurons, the pharmacological blockade of excitatory synaptic transmission led to the progressive loss of DOWN states and the appearance of continuous firing (7.6 ± 3.6 Hz; n = 4; Fig. 11B1), thereby removing network input to these cells while leaving the strong, direct muscarinic depolarization intact.
Absence of rhythmic burst firing in “network driver” neurons recorded in slices maintained in a reduced extracellular Ca2+ concentration
In earlier studies, UP and DOWN states have been brought about in neocortical slices by modifying the ionic composition of the recording medium to supposedly more closely mimic that present in vivo (Sanchez-Vives and McCormick, 2000; Cossart et al., 2003; Shu et al., 2003; Cunningham et al., 2006; Rigas and Castro-Alamancos, 2007; Compte et al., 2008; Rigas and Castro-Alamancos, 2009; Fanselow and Connors, 2010). We therefore compared the effect of such a manipulation with that achieved by applying CCH in the same slices of the mouse primary auditory cortex. Following the wash-in of modified medium (i.e., 1.2 mm, Ca2+, 3.5 mm K+, 1 Mg2+), we observed synchronized UP states that occurred sporadically (UP states per minute: 1.7 ± 0.4; n = 10), were therefore separated by very long DOWN states (27.3 ± 6.3 s; n = 5), and were nonrhythmic (Fig. 12A,B,D, left traces). On the other hand, in the same slices, a 30 min washout of this modified medium with a conventional medium (i.e., 2 mm Ca2+, 3.15 mm K+, 1 Mg2+) followed by the addition of CCH brought about robust, rhythmic UP states (0.75 ± 0.12 Hz; n = 5) separated by DOWN states with a significantly shorter duration (0.84 ± 0.14 s; n = 5; p < 0.001) (Fig. 12A,B,D, right traces). Apart from an obvious difference in frequency and rhythmicity, UP states brought about by CCH application were of shorter duration (modified medium: 0.375 ± 0.07 s, CCH: 0.251 ± 0.02 s; n = 5; p < 0.001) and exhibited less firing (action potentials/UP state: modified medium: 5.4 ± 5.6 s, CCH: 3.8 ± 3 s; n = 5; p < 0.05) than those elicited by a modified recording medium (Fig. 12A,C). Otherwise, there were no obvious differences between individual UP states in the two conditions as assessed with intracellular recordings (Fig. 12B,C, enlarged traces). CCH increased the frequency of UP states following its application directly in the modified medium (UP states per minute: modified medium: 4.0 ± 2.2; modified medium + CCH: 21.3 ± 5.7, p < 0.001, n = 5, data not shown).
While performing this particular series of experiments, we obtained intracellular recordings from 3 “network driver” cells (Fig. 12D). As already described above, the onset of LFP oscillations following CCH application coincided exactly with rhythmic burst firing in these cells (Fig. 12D, right). Interestingly, however, such rhythmic burst firing never occurred in these cells in the presence of a modified recording medium (Fig. 12D, left).
Discussion
By using a pharmacological reinstatement of the cholinergic drive in mouse neocortical slices, which brings about rhythmic UP/DOWN states with characteristics indistinguishable from those observed in vivo, the main finding of this study is that neocortical UP states are fundamentally reliant on a small (∼7%) subset of layer 5 IB pyramidal neurons (which we call putative “network drivers”) that exhibit intrinsic burst firing at low frequencies (∼0.2–2 Hz). These bursts are resistant to the block of excitatory and inhibitory transmission and occur rhythmically, but are compromised when recording in a low Ca2+ medium, thus explaining why UP states are sporadic and nonrhythmic in the latter experimental condition and thus different from the dynamics of the slow oscillation recorded in vivo.
Role of ACh in slow waves
Both the current study in mice and previous work in cats (Steriade et al., 1993b) support a role for mAChRs in shaping UP states because the systemic application of scopolamine in both species causes a pronounced reduction in UP state duration and intensity accompanied by an increase in frequency (from ∼0.3 to 0.6–0.7 Hz). A recent study has shown that ACh can depolarize layer 5 pyramidal neurons and affect the UP and DOWN states present in vitro in a dose-dependent manner (Wester and Contreras, 2013). Although an involvement of the cholinergic system in UP state generation may seem counterintuitive, it is fully consistent with previous studies showing that during anesthesia cholinergic neurons fire prominently and preferentially during the UP state of the slow oscillation (Nunez, 1996; Détári et al., 1997; Manns et al., 2000; Mena-Segovia et al., 2008). This fits well with our finding that persistent activation of mAChRs in neocortical slices is sufficient to bring about rhythmic UP states and suggests that, at least during anesthesia, cortical UP states may be largely sustained by direct cholinergic input. In our slice experiments, a lower level of mAChR activation (i.e., during CCH wash-on) also brought about faster, lower amplitude activity at ∼1–2 Hz, which closely matches the in vivo effects of scopolamine (Steriade et al., 1993b; but see also Carracedo et al., 2013). This increase in oscillation frequency also parallels the change that occurs in slow waves as sleep and anesthesia are deepened (Amzica and Steriade, 1998), a scenario that likely involves a lessening of cholinergic influence. Indeed, as the firing of cholinergic neurons is diminished but not abolished during natural NREM sleep (Steriade et al., 1990; Marrosu et al., 1995; Szymusiak et al., 2000; Vazquez and Baghdoyan, 2001; Lee et al., 2004, Lee et al., 2005; Hassani et al., 2009) and in humans there is a strong activation of cholinergic nuclei during NREM slow waves (Dang-Vu et al., 2008), the basic neocortical network mechanisms of UP state generation described here have also relevance to the UP/DOWN state dynamics of natural sleep (Steriade et al., 2001; Timofeev et al., 2001; Luczak et al., 2007).
Importance of intrinsic rhythmic burst firing of layer 5 “network driver” IB pyramidal neurons for initiating UP states
The initiation of network oscillations and associated UP states in this study was unequivocally due to the activity of the “network driver” cells for the following two reasons. First, at the point of inception of oscillations in the LFP, these neurons are the only group that fire coherently with this signal. Viewed another way, the fact that we could not detect any other neocortical cell type that fires in synchrony with the LFP fluctuations at the time point of slow-wave initiation, or even fires at all (with the exception of few “early firing” layer 5 RS pyramidal neurons, see below), means that only “network driver” cells, and the subsequent downstream activation of synaptic currents in neurons to which they project, can explain the LFP signal. Second, in a modified recording medium containing a reduced Ca2+ concentration, the intrinsic low-frequency bursting activity of “network driver” neurons that characterizes the initial response of these cells to CCH does not occur and, as a consequence, UP states are irregular and sporadic. Thus, by defining a subset of layer 5 “network driver” neurons as having a key role in slow-wave generation and UP state initiation, our work provides a firm mechanistic basis for understanding the centrality of layer 5 in generating slow waves in the neocortex as previously proposed (Amzica and Steriade, 1998; Sanchez-Vives and McCormick, 2000; Sakata and Harris, 2009; Chauvette et al., 2010; Wester and Contreras, 2012; Beltramo et al., 2013).
Additional significance of cell type-specific features
Once the slow (<1 Hz) oscillation is fully established, it is likely that additional mechanisms also contribute to shaping and sustaining UP states. First, given the similarity in UP state manifestations to “network driver” neurons, it is probable that conventional IB cells (i.e., those later entrained to network activity) also play a role in driving UP states. Indeed, once network wide synchrony is established, a combined assessment of “network driver” neurons and conventional IB cells indicates that these cells (1) exhibit the strongest phase coupling to the LFP of any cell class, (2) show a peak in firing that occurs before other excitatory neurons, and (3) demonstrate UP states that nearly always commence with a burst that is subsequently followed by activity that is indicative of recurrent excitation.
Second, given that action potential output in “early firing” cells commences considerably before that in IB cells on each oscillation cycle, we suggest that resultant synaptic input to IB cells, and of course to other RS cells, will assist the generation of a burst, thereby facilitating full network activation. Indeed, examination of the subthreshold activity of IB neurons clearly shows the presence of EPSPs before burst generation and UP states, a feature that is also present in vivo (Chauvette et al., 2010). The strong tonic drive to “early firing” cells and their related activity is also consistent with several modeling studies showing that a group of neurons that are relatively depolarized, and therefore exhibit spontaneous activity, and/or the presence of spontaneous EPSPs is required to successfully simulate slow-wave activity (Bazhenov et al., 2002; Compte et al., 2003; Hill and Tononi, 2005).
Functional implications
As previously argued (Crunelli and Hughes, 2010) and recently demonstrated (David et al., 2013; Lemieux et al., 2014), the full expression of slow waves in the mammalian EEG, and in particular the robust and regular initiation of neocortical UP states, requires rhythmic input from intrinsically oscillating thalamocortical (TC) neurons. We also suggested that intrinsically rhythmic cells in the neocortex (Le Bon-Jego and Yuste, 2007) may also aid in the generation of UP states (Crunelli and Hughes, 2010). Regardless of the actual sources of pacemaker activity, the current study verifies the general concept that regularly occurring, rhythmic neocortical UP states require pacemaker input, with this pacemaker input being most effectively served by neurons, which exhibit intrinsic, rhythmic burst firing, a common feature of both TC neurons and “network driver” cells. Interestingly, that the expression of rhythmic burst firing in “network driver” cells occurs at relatively hyperpolarized membrane potentials and is replaced by tonic firing when these cells are depolarized further aligns their electrical phenotype with that of TC neurons while distinguishing them from other IB pyramidal cells in layer 5, which tend to exhibit bursting in response to stronger depolarization (Schwindt et al., 1997).
Last, given the close similarity between rhythmic network UP states that are observed during periods of reduced vigilance, such as non-REM sleep and anesthesia, and the discrete responses of the neocortex to certain types of external stimuli (Sakata and Harris, 2009), it is likely that these responses also have relevance to the processing of sensory input and for understanding fundamental neocortical circuit operations (Haider and McCormick, 2009). As such, the capacity to easily instate rhythmic UP/DOWN states in neocortical slices, as described here, may prove to be a useful tool for interrogating not only the basic mechanisms of sleep- and anesthesia-related oscillations but also the elemental structure of neocortical network dynamics during attentive states.
Footnotes
This work was supported by the Wellcome Trust 78403 and 91882 to V.C. and 78311 to S.W.H., Ely Lilly & Co. Lilly Research Award Program to V.C., Hungarian Brain Research Program KTIA_NAP_13-2-2014-0014 to M.L.L., Hungarian Scientific Research Fund OTKA NF 105083 to M.L.L., and Human Frontier Science Program fellowship LT001009/2010L to M.L.L. We thank Mr. Timothy Gould for technical assistance.
D.G., J.T.R.I., and S.W.H. were full-time employees of Eli Lilly & Co. when this research was conducted. The remaining authors declare no competing financial interests.
- Correspondence should be addressed to Dr. Magor L. Lőrincz, Research Group for Cellular and Network Neurophysiology of the Hungarian Academy of Sciences, Department of Physiology, Anatomy and Neuroscience, University of Szeged, Közép fasor 52, Szeged, 6726, Hungary. mlorincz{at}bio.u-szeged.hu
This article is freely available online through the J Neurosci Author Open Choice option.