Layer‐specific excitation/inhibition balances during neuronal synchronization in the visual cortex

Abstract

Key points

• Understanding the balance between synaptic excitation and inhibition in cortical circuits in the brain, and how this contributes to cortical rhythms, is fundamental to explaining information processing in the cortex.

• This study used cortical layer‐specific optogenetic activation in mouse cortex to show that excitatory neurons in any cortical layer can drive powerful gamma rhythms, while inhibition balances excitation.

• The net impact of this is to keep activity within each layer in check, but simultaneously to promote the propagation of activity to downstream layers.

• The data show that rhythm‐generating circuits exist in all principle layers of the cortex, and provide layer‐specific balances of excitation and inhibition that affect the flow of information across the layers.

Abstract

Rhythmic activity can synchronize neural ensembles within and across cortical layers. While gamma band rhythmicity has been observed in all layers, the laminar sources and functional impacts of neuronal synchronization in the cortex remain incompletely understood. Here, layer‐specific optogenetic stimulation demonstrates that populations of excitatory neurons in any cortical layer of the mouse's primary visual cortex are sufficient to powerfully entrain neuronal oscillations in the gamma band. Within each layer, inhibition balances excitation and keeps activity in check. Across layers, translaminar output overcomes inhibition and drives downstream firing. These data establish that rhythm‐generating circuits exist in all principle layers of the cortex, but provide layer‐specific balances of excitation and inhibition that may dynamically shape the flow of information through cortical circuits. These data might help explain how excitation/inhibition (E/I) balances across cortical layers shape information processing, and shed light on the diverse nature and functional impacts of cortical gamma rhythms.

Key points

• Understanding the balance between synaptic excitation and inhibition in cortical circuits in the brain, and how this contributes to cortical rhythms, is fundamental to explaining information processing in the cortex.

• This study used cortical layer‐specific optogenetic activation in mouse cortex to show that excitatory neurons in any cortical layer can drive powerful gamma rhythms, while inhibition balances excitation.

• The net impact of this is to keep activity within each layer in check, but simultaneously to promote the propagation of activity to downstream layers.

• The data show that rhythm‐generating circuits exist in all principle layers of the cortex, and provide layer‐specific balances of excitation and inhibition that affect the flow of information across the layers.

Introduction

The dynamic relationship between excitation and inhibition, and its contribution to cortical rhythms, particularly in the gamma band, is fundamental to the propagation of neural signals in the brain. Rhythmic activity, mediated by rhythmic excitation and inhibition (Cardin et al. 2005; Hasenstaub et al. 2005; Atallah & Scanziani, 2009), has been observed in a wide array of brain circuits and has been shown to correlate with perception, attention, and cognition in various conditions (Gray & Singer, 1989; Fries, 2009). In the mammalian forebrain, gamma oscillations (∼20–80 Hz) are thought to depend on the reciprocal interaction between recurrently connected excitatory and inhibitory neurons (Jefferys et al. 1996; Mann et al. 2005; Buzsaki & Wang, 2012; Cardin, 2016). In the hippocampus and in the cortex, multiple rhythm generators are likely to exist that synchronize neurons in different frequency bands across different subfields or layers (Colgin et al. 2009; Oke et al. 2010; Ainsworth et al. 2011; Whittington et al. 2011; Welle & Contreras, 2016), but it remains unclear which subpopulations of cortical neurons are sufficient to drive oscillatory activity.

Regardless of their source, many experimental and theoretical models for rhythm generation rest on a dynamic interaction between synaptic excitation and inhibition (Cardin et al. 2005; Hasenstaub et al. 2005, 2016; Atallah & Scanziani, 2009; de Almeida et al. 2009; Tiesinga & Sejnowski, 2009; Buzsaki & Wang, 2012; Jadi & Sejnowski, 2014a; Salkoff et al. 2015; Neske & Connors, 2016). The balance, or ratio, between excitation and inhibition, critically determines the timing, amplitude, and population dynamics underlying oscillations (Shadlen & Newsome, 1994; van Vreeswijk & Sompolinsky, 1998; Haider & McCormick, 2009; Isaacson & Scanziani, 2011). Indeed, defects in this balance or in gamma oscillations per se are thought to underlie a variety of neurological disorders (Uhlhaas & Singer, 2010; Yizhar et al. 2011; Cho et al. 2015; Nelson & Valakh, 2015).

During gamma rhythms the amplitude of excitation and inhibition co‐fluctuates across each oscillation cycle, such that their ratio in time remains approximately fixed (Atallah & Scanziani, 2009). Likewise, during gamma oscillations within layers L2/3, the E/I ratio also remains constant over cortical space, i.e. over the sensory map (Adesnik & Scanziani, 2010). Tight E/I matching across layers has also been suggested with glutamate uncaging experiments (Xu et al. 2016), but the net E/I ratio each gamma source might generate in its targets remains essentially unknown.

Anatomical and physiological studies have outlined the distinct connectivity patterns of each cortical layer, and demonstrated that cortical excitatory neurons make connections with both excitatory and inhibitory neurons within and across layers (Binzegger et al. 2004; Thomson & Lamy, 2007; Lefort et al. 2009; Jiang et al. 2015; Narayanan et al. 2015; Markram et al. 2015). Therefore, in principle, each layer could drive gamma band synchronization, but this has yet to be tested. Furthermore, how individual cortical layers might contribute to E/I balances during gamma band activity is still unclear. Therefore, the goal of this study was to answer three questions: first, which cortical layers are sufficient to induce gamma oscillations? Second, what is the balance of excitation and inhibition each layer generates in its targets during gamma activity? Third, what is the net impact of these E/I balances on downstream activity?

To address these questions an optogenetic approach was employed in the primary visual cortex of mice, and layer‐specific populations were photo‐stimulated one at a time, as has been done with L2/3 in the rodent somatosensory cortex (Adesnik & Scanziani, 2010; Shao et al. 2013; Takada et al. 2014). Such an approach can demonstrate whether a specific population of excitatory neurons can generate gamma oscillations. Moreover, simultaneous intracellular recordings can be made to track the excitation, inhibition, and the net functional impact of gamma activity across the cortex.

The data indicate that robust gamma rhythms can be generated by groups of excitatory neurons in any of the principal cortical layers in vivo and in vitro. During gamma activity, inhibition dominates excitation within the layer of origin, but shifts towards excitation in downstream target layers. The net impact on neuronal firing is suppression within the layer of origin (in non‐opsin expressing neurons in these experiments), but facilitation in downstream layers. These data collectively support the notion that gamma activity may be generated by any cortical layer, but that the net impact of these oscillations depends on layer‐specific E/I balances. These data help provide a mechanistic framework for the origin and functional impact of gamma oscillations across the layers of the visual cortex.

Methods

Transgenic mice

All experiments were performed in accordance with the guidelines and regulations of the ACUC of the University of California, Berkeley and the IACUC of the University of California, San Diego. Both female and male mice were used in approximately equal proportions. Experiments in vivo were performed on animals aged between 3.5 and 8 weeks. In vitro experiments were performed on animals aged 3.5–5 weeks. Mice strains used in this study were scnn1‐tg3‐Cre (Allen Institute, JAX Stock number 009613; Madisen et al. 2010), Rbp4‐Cre (GENSAT, line number KL100; Gong et al. 2003), Ntsr1‐Cre (GENSAT, line number GN220) or wild‐type ICR mice (Charles River, Wilmington, MA, USA) subjected to in utero electroporation at E15–E16 as previously described. Cre reporter lines were Rosa‐LSL‐tdTomato (Ai9, JAX stock number 007909) and Rosa‐LSL‐H2B‐mCherry (JAX stock number 023139).

Ntsr1‐Cre have been well characterized previously (Olsen et al. 2012; Kim et al. 2014; Denman & Contreras, 2015). A similar analysis was performed on scnn1‐tg3‐Cre and Rbp4‐Cre mice. In representative brain sections from these two Cre lines no overlap was observed between Cre‐reporter expression and GFP expression when these lines were crossed to the GAD67‐GFP mouse line (L4‐Cre;GAD67‐GFP: 111 tdTomato^+, 38 GFP⁺, 0 co‐labelled; L5‐Cre;GAD67‐GFP: 155 tdTomato⁺, 75 GFP⁺, 0 co‐labelled, example images in Fig. 4). To target L2/3 pyramidal cells specifically, timed in utero electroporation was used (Saito & Nakatsuji, 2001). Post hoc histology from electroporated brains demonstrated that there was little to no overlap between ChR2 expression and the inhibitory marker GAD67‐GFP, and that about 25% of L2/3 pyramidal cells expressed the opsin. Therefore, under all these conditions, ChR2 expression was restricted to excitatory cells, and thus all the optogenetically evoked inhibition was driven through the network, rather than being of monosynaptic origin. This represents a key advantage over photo‐stimulation via glutamate uncaging (Xu et al. 2016), or direct photo‐stimulation of inhibitory neurons (Katzel et al. 2011), which will necessarily recruit monosynaptic inhibition when mapping E/I circuits across layers (Xu et al. 2016). It is important to note that the cells expressing ChR2 in this study do not necessarily represent all the myriad subtypes of cortical excitatory neurons within each layer, but within each experiment ChR2 was restricted to subsets of excitatory cells within one of the principal cortical layers at a time.

Figure 4. Histological characterization of scnn1‐tg3‐ and Rpb4‐Cre lines.

A, left: confocal image of a vertical strip of V1 from an scnn1‐tg3‐Cre mouse crossed to a Rosa‐LSL‐H2B‐mCherry reporter, and immunostained for the neuron‐specific marker, NeuN (blue). Right: quantification of mCherry‐expressing neurons versus cortical depth (normalized across brain sections to correct for variations in cortical thickness). Horizontal dashed lines indicate the approximate L4 boundaries (grey area is SEM) B, example image from a triple transgenic mouse labelling Cre‐expressing cells with tdTomato and inhibitory neurons with GFP; NeuN is blue. C, quantification of the fraction of layer 4 neurons expressing GAD67‐GFP (GFP⁺/NeuN⁺, left bar), Cre (mCherry⁺/NeuN⁺, middle bar), and the total estimated fraction of excitatory neurons in layer 4 expressing Cre (mCherry⁺/NeuN⁺‐GFP⁺ right bar). No overlap was seen between GFP and tdTomato. D–F, as in A–C but for the Rbp4‐Cre line in layer 5. All error bars are SEM.

Viral infection

Neonatal scnn1‐tg3‐Cre, Rbp4‐Cre, Ntsr1‐Cre or SOM‐Cre mice (P1‐3) were briefly cryo‐anaesthetized and placed in a head mould. Transcranial injection of ∼45 nl of undiluted AAV9‐CAGGS‐flex‐ChR2‐tdTomato.SV40 or AAV9‐DIO‐Ef1a‐eNpHR3.0‐YFP (UPenn Vector Core) was performed using a Drummond Nanoject injector at three locations in the mouse primary visual cortex (V1). With respect to the lambda suture coordinates for V1 were 0.0 mm AP, 2.2 mm L and injection was 200–500 μm under the skull.

In utero electroporation

Timed pregnant mice were obtained from Charles River. Pregnant mice at E15–16 were anaesthetized with 2.0% isoflurane, the abdomen was cleaned with 70% ethanol and swabbed with iodine, and a small vertical incision was made in the skin and abdominal wall and 8–12 embryos gently exposed. Each embryo was injected with 0.5–1 μl of DNA solution and 0.05% Fast Green. pCAG‐ChR2‐Venus plasmid DNA was mixed with pCAG‐GFP for a total of 1–2 μg ChR2 DNA and 0.5–1 μg of fluorophore DNA per injection. A pressure‐controlled bevelled glass pipette (Drummond, Custom Microbeveller) was used for injection. After each injection, the embryos were moistened with saline and voltage steps via tweezertrodes (BTX, 5 mm round, platinum, BTX electroporator) were applied with the positive electrode placed over the visual cortex and the negative electrode placed under the head of the embryo. Voltage was 40 V for 5 pulses at 1 Hz, each pulse lasting 50 ms. The embryos were returned to the abdomen, which was sutured, followed by suturing of the skin. The procedure typically lasted under 30 min. On the day of birth, animals were screened for location and strength of transfection by trans‐cranial epifluorescence under an Olympus MVX10 fluorescence stereoscope.

In vitro recording

Mice were deeply anaesthetized with isoflurane and quickly decapitated. Slices 400 μm thick were cut on a microslicer (DTK‐1000) and incubated at 34°C for 30–45 min, and then at room temperature in sucrose cutting solution (in mM: NaCl, 83; KCl, 2.5; MgSO₄, 3.3; NaH₂PO₄, 1; NaHCO₃, 26.2; D‐glucose, 22; sucrose, 72; and CaCl₂, 0.5, bubbled with 95% O₂ and 5% CO₂). A slice was transferred to a submerged chamber perfused with warmed (∼32°C) ACSF (in mM: NaCl, 119; KCl, 2.5; NaH₂PO₄, 1.3; NaHCO₃, 26; D‐glucose, 20; MgCl₂, 1.3; CaCl₂, 2.5; 305 mOsm, bubbled with 95% O₂ and 5% CO₂) and held down with nylon threads on a platinum harp. Excitatory and inhibitory currents were recorded in the voltage clamp mode with a caesium based internal (in mM: CsMeSO4, 115; NaCl, 4; Hepes, 10; Na₃GTP, 0.3; MgATP, 4; EGTA, 0.3; QX‐314‐Cl, 2.5; BAPTA‐5Cs), and action potentials were recorded in a solution in which caesium was exchanged with potassium and QX‐314 and BAPTA were omitted. Patch pipettes had resistances of 2‐3 MΩ. Signals were amplified with two Multiclamp 700B amplifiers (Molecular Devices), filtered at 2 kHz and digitized via a National Instruments A/D card at 20 kHz. Custom software in Igor Pro (Wavemetrics) or Matlab (Mathworks) controlled all aspects of the experiment. For Figs 5, 6, 7, simultaneous whole cell recordings were made from a L2/3, a L4, and a L5 pyramidal shaped neuron along a vertical axis perpendicular to the pia. In some cases, only two cells were recorded, with one cell always being from the layer expressing ChR2 for normalization. Cells that exhibited direct photocurrents were discarded, and in any experiment if the recorded cell in the layer expressing ChR2 was lost, the experiment was terminated, as this cell was used for comparison of all evoked currents. Series resistance was less than 20 MΩ (uncompensated), and if this value changed in any cell by more than 15% the experiment was discarded. For Figs 4, 5, 6, photo‐stimulation was via a collimated blue LED (Thorlabs) mounted to the epifluorescent pathway of the microscope (Olympus BX51) through a 40× objective, illuminating an area of the brain slice ∼200 μm wide controlled with the field aperture. To generate network activity 1–2 s ramps of light were used, as in vivo, with a final intensity of 0.1–2 mW cm⁻². Prior experiments demonstrated similar oscillations in the barrel cortex are blocked by both glutamatergic and GABAergic antagonists (Adesnik & Scanziani, 2010). The slope of the ramp was adjusted in each recording to obtain network activity (synaptic currents) with a largely stable power for the duration of the light stimulus. The light intensity needed to accomplish this in any given slice (or mouse) likely correlated with variations of opsin expression. Higher intensity stimulation (exceeding 2 mW cm⁻²) invariably caused potent adaptation of network response, making quantification more difficult. A dose–response curve on L4 for increasing light intensity is shown in Fig. 3 H, and published previously for L2/3 in the barrel cortex (Adesnik & Scanziani, 2010). The light stimulus was always centred on the ChR2‐expressing layer.

Figure 5. Layer 4 induced gamma rhythms suppress L4 but facilitate L2/3 and L5.

A, left: recording schematic diagram. A simultaneous triple whole cell recording is obtained from a L4, a L2/3, and a L5 excitatory neuron along the same vertical axis of V1. ChR2 is expressed specifically in L4 excitatory neurons (indicated by red shading) in an scnn1‐tg3‐Cre mouse injected with a Cre‐dependent ChR2‐tdTomato virus. Right: example image from an scnn1‐tg3‐Cre mouse crossed to a Rosa‐LSL‐tdTomato reporter strain. B, example synaptic excitatory (red) and inhibitory (blue) currents recorded simultaneously in a L4 (left), a L2/3 (middle), and a L5 excitatory neuron. C, left: average E/I ratio of the synaptic charge in L4, L2/3 and L5 excitatory neurons across 10 triple and an additional 10 paired whole cell recordings. Middle: average net excitatory charge, normalized to the charge measured in the L4 neuron. Right: average normalized net inhibitory charge. D, average coherence spectra (± 95% confidence intervals) of the EPSCs (E‐E, red) and IPSCs (I‐I, blue) recorded across the pair of cells indicated at the top of each plot. E, example membrane potential traces from a ChR2‐negative L4 excitatory neuron (left), a L2/3 pyramidal cell (middle), and a L5 pyramidal cell (right) driven to spike with a square current pulse through the patch electrode (left) and with the addition of the optogenetic activation of L4 (right). Red dashed lines are to aid in the visual comparison of the V _m between the presence and absence of photo‐stimulation. Left inset: scatter plot of mean membrane potential changes in the L4 cell during photo‐stimulation (P < 0.005, signed rank test). F, left: scatter plot of the change in spike rate for all recorded neurons during optogenetic activation of L4. Right: plot of the mean change in firing rate during photo‐induced gamma activity across the three layers. All error bars are SEM.

Figure 6. L5 induced gamma rhythms suppress L5 but facilitate L2/3.

A, left: recording schematic diagram. A simultaneous triple whole cell recording is obtained from a L4, a L2/3, and a L5 excitatory neuron along the same vertical axis of V1. ChR2 is expressed specifically in L5 excitatory neurons (indicated by red shading) in an Rbp4‐Cre mouse injected with a Cre‐dependent ChR2‐tdTomato virus. Right: example image from an Rbp4‐Cre mouse crossed to a Rosa‐LSL‐tdTomato reporter strain. B, example synaptic excitatory (red) and inhibitory (blue) currents recorded simultaneously in a L4 (left), a L2/3 (middle), and a L5 (right) excitatory neuron. C, left: average E/I ratio of the synaptic charge in L4, L2/3 and L5 excitatory neurons across 11 multiple whole cell recordings. Middle: average net excitatory charge, normalized to the charge measured in the L5 neuron. Right: average normalized net inhibitory charge. D, average coherence spectra (± 95% confidence intervals) of the EPSCs (E‐E, red) and IPSCs (I‐I, blue) recorded across the pair of cells indicated at the top of each plot. E, example membrane potential traces from a L4 cell (left), a L2/3 cell (middle) and a ChR2‐negative L5 cell (right) driven to spike with a square current pulse through the patch electrode (left) and with the addition of the optogenetic activation of L5 (right). Red dashed lines are to aid in the visual comparison of the V _m between the presence and absence of photo‐stimulation. F, left: scatter plot of the change in spike rate for all recorded neurons during optogenetic activation of L5. Right: plot of the mean change in firing rate during photo‐induced gamma activity across the three layers. All error bars are SEM.

Figure 7. L2/3 induced gamma oscillations suppress L2/3 but facilitate L5.

A, left: recording schematic diagram. A simultaneous triple whole cell recording is obtained from a L4, a L2/3, and a L5 excitatory neuron along the same vertical axis of V1. ChR2 is expressed specifically in L2/3 pyramidal neurons (indicated by red shading) via timed in utero electroporation of a ChR2‐expressing plasmid. Right: example image from a mouse electroporated in utero with ChR2. B, example synaptic excitatory (red) and inhibitory (blue) currents recorded simultaneously in a L4 (left), a L2/3 (middle), and a L5 (right) excitatory neuron. C, left: average excitatory/inhibitory ratio of the synaptic charge in L4, L2/3 and L5 excitatory neurons across 11 multiple whole cell recordings. Middle: average net excitatory charge, normalized to the charge measured in the L2/3 neuron. Right: average normalized net inhibitory charge. D, average coherence spectra (± 95% confidence intervals) of the EPSCs (E‐E, red) and IPSCs (I‐I, blue) recorded across the pair of cells indicated at the top of each plot. E, example membrane potential traces from a ChR2‐negative L2/3 cell (left), and a L5 cell (right) driven to spike with a square current pulse through the patch electrode (left) and with the addition of the optogenetic activation of L2/3 (right). Red dashed lines are to aid in the visual comparison of the V _m between the presence and absence of photo‐stimulation. F, left: scatter plot of the change in spike rate for all recorded neurons during optogenetic activation of L2/3. Right: plot of the mean change in firing rate during photo‐induced gamma activity across L2/3 and L5. All error bars are SEM.

Figure 3. Gamma rhythms can be generated by excitatory cells in any cortical layer in vitro .

A, schematic diagram of a brain slice expressing ChR2 in L4 excitatory neurons (red shading) prepared for optogenetic activation. B, example excitatory (red) and inhibitory (blue) synaptic currents recorded in a L2/3 pyramidal cell during optogenetic activation of L4. C, example average power spectrum (±SEM) of the inhibitory currents for a L4 cell. D, left: example excitatory (red) and inhibitory (blue) synaptic current traces recorded in ChR2‐negative neurons in brain slices from mice expressing ChR2 in L2/3 (left), in L5 (middle), or in L6 (right). E, example average power spectra (±SEM) for the inhibitory currents in mice where L23, L5, or L6 was photo‐stimulated. Spectra come from a ChR2⁻ cell in the expressing layer. F, average peak frequencies (20–80 Hz) for the power spectra of the LFP from mice expressing ChR2 in L4, L2/3, L5, and L6 (P < 10⁻⁸, one‐way ANOVA). G, contrast–response functions of synaptic excitation and inhibition in L2/3 neurons recorded in vivo (n = 6 cells). Top: experimental schematic diagram and example traces. H, light dose–response functions from L2/3 neurons recorded in brain slices as the slope of the light ramp was systematically increased (n = 7 cells). Top: experimental schematic diagram and example traces. Error bars are SEM.

In vivo recording

During surgery, mice were initially anaesthetized isoflurane (1–2%) and injected with a combination of urethane (1 g kg⁻¹) and chlorpithixene (5 mg kg⁻¹). During electrophysiological recording isoflurane was switched off. The scalp was removed, the fascia retracted, and the skull lightly etched with a 27 gauge needle. Following application of Vetbond to the skull surface, a custom stainless steel headplate was fixed to the skull with dental cement (Orthojet). A craniotomy (∼1–2 mm) was opened, the dura removed, and the brain stabilized with 1–2% agarose. Extracellular and intracellular recordings (Margrie et al. 2002) were made with conventional glass patch pipettes. Extracellular electrodes were filled with ACSF. Intracellular solution for voltage clamp was the same as for in vitro recordings and were targeted to L2/3 neurons 100–350 μm below the pia. Whole cell recordings were made with the blind patch technique (Margrie et al. 2002), and patch pipettes were 3–5 MΩ resistance. Series resistance was typically under 25 MΩ.

Multi‐electrode recording in vivo

Mice were prepared as above, but a smaller craniotomy (<0.5 mm) was made, and a 16 channel multi‐electrode array (Neuronexus model a1x16‐3mm50‐177) was inserted into the primary visual cortex at a ∼30 deg angle from vertical, and the craniotomy covered with ACSF. The depth of the electrode was calibrated by the reading off a sub‐micron precise micro‐manipulator (Luigs and Neumann) and correct for the insertion angle. Neural data was filtered (0.3 Hz–10 kHz) and digitized at 20–30 kHz (National Instruments) with custom routines written in Igor Pro. Data were subsequently ported to and analysed in Matlab using custom software and signal processing routines from the Chronux tool box (Bokil et al. 2010).

Visual stimulation

Visual stimuli were generated with Psychophysics toolbox (Brainard, 1997) and were displayed on a gamma corrected LCD monitor (60 Hz refresh rate), positioned 20–25 cm from the contralateral eye. All visual stimuli were full screen (∼60 deg of visual angle) square wave drifting gratings with a temporal frequency of 2 Hz and a spatial frequency of 0.04 cycles deg⁻¹ at either 100% contrast (Fig. 1), or varying contrast (Fig. 2).

Figure 1. Visually and optogenetically evoked gamma rhythms in the primary visual cortex in vivo .

A, top: experimental schematic diagram. Bottom: example local field potential (LFP) response recorded in layer 2/3 of V1 to a full field drifting grating. B, top: experimental schematic diagram. Bottom: LFP response in the same mouse in a subsequent trial to optogenetic stimulation of excitatory neurons. The blue ramp indicates the time and shape of the optogenetic stimulus. C, average power spectra (±SEM) computed for the visual stimulus and optogenetic stimulus conditions. D, left: average peak frequency (20–80 Hz) of the LFP power spectrum during visual and optogenetic stimulation. Right: mean power in the same spectra. Error bars are SEM. E, scatter plot of the peak power in the gamma band between visually induced and optogenetically induced gamma oscillations on interleaved trials (n = 8 mice; P < 0.05, signed rank test). F, example synaptic current traces recorded in a L2/3 neuron in voltage clamp in vivo during visual stimulation. G, as in F, but during optogenetic activation of L4 in the same cell. The blue ramp indicates the time and shape of the optogenetic stimulus.

Figure 2. Characterization of gamma rhythms induced by excitatory neurons in different cortical layers.

A, left: example LFP traces recorded in vivo in response to a slow ramp of blue light in three different mice expressing ChR2 in L2/3, L5 or L6. B, example average power spectra (± SEM) for the LFP when photo‐stimulating neurons in the indicated layers. C, scatter plot of the frequency of the gamma band peak (20‐80 Hz) measured for in vivo photo‐stimulation of excitatory neurons in each layer (P < 0.05, one‐way ANOVA). D, scatter plot of peak gamma band power (power at the peak frequency) for the same experiments (P = 0.51, one‐way ANOVA). E, example polar histogram of the computed phased of detected spikes in an example L4‐ChR2 unit recording. F, top: low‐pass filtered (0–200 Hz, black) and high‐pass filtered (500–5000 Hz, grey) trace from an example trial during photo‐stimulation of L4 (light is on during the entire trace). Bottom: extracted phase vector (green) and detected spike times (asterisks) from the traces above. G, plot of the average phase of detected spikes during photo‐stimulation versus electrode depth in V1 for L4 photo‐stimulation (left) and L5 photo‐stimulation (right) (n = 5 L4‐ChR2 mice, n = 6 L5‐ChR2 mice, P < 0.005, rank sum test). H, scatter plot of computed pairwise phase consistency (PPC) for detected spikes in L4‐ChR2 experiments (left), and those for L5‐ChR2 experiments (right).

Optogenetic stimulation in vivo

For optogenetic stimulation of ChR2 in vivo, a blue (∼473 nm) LED (Thorlabs) coupled to 1 mm multimode fibre and positioned <3 mm from the craniotomy was used. The optical fibre was set to illuminate a wide area including all of V1. This light was prevented from hitting the eyes by covering the skull with black acrylic (iron oxide powder in dental acrylic) and a small light shield made of black aluminum foil. Ramps of blue light (1–2 s; final intensity 0.1–2 mW cm⁻²) were found to produce rhythmic activity similar to what was observed with visual stimulation and what was observed in vitro. As in brain slices, the slope of the ramp was adjusted for each animal to obtain network activity (LFP response) with a largely stable power for the duration of the light stimulus. Although gamma oscillations could be evoked for a broad range of power, higher intensities of light led to strongly adapting responses both extracellularly (LFP) and intracellularly (synaptic currents) and were thus not used. Similar adaptation could also be seen in the light driven multi‐unit activity, although this was not used during the experiment to aid in the choice of light intensity. The mean light intensity did not vary significantly across the different transgenic lines.

Extracellular data analysis

Analysis was performed in Matlab (Mathworks). Power spectra were computed using multitaper estimation in Matlab with the Chronux package. All spectral plots show means ± SEM. Spiking activity from multielectrode arrays was extracted by filtering the raw signal between 800 and 7000 Hz. Spike detection was performed using the UltraMega Sort package (Hill et al. 2011). To quantify locking of spiking activity to the gamma band, we bandpass‐filtered the LFP (20–80 Hz) extracted the oscillation's instantaneous phase by using the imaginary part of the analytical signal using the Hilbert transform. Each spike was thus assigned an exact phase in the gamma oscillation. Phase‐locking magnitudes were determined for each unit by the pairwise phase consistency (PPC), a measure of synchrony that is not biased by the number of spikes (Vinck et al. 2010). The significance of locking was determined by the Rayleigh test for non‐uniformity on the distribution of spike‐phases. All units with P < 0.05 were considered to be significantly locked.

Intracellular data analysis

Synaptic charge during gamma activity was computed using trapezoidal integration (Igor Pro ‘area’ function) of the voltage clamped currents at the corresponding reversal potentials for excitation (0 mV) and inhibition (−70 mV) after correction for the liquid junction potential (∼7–10 mV) and subtraction of the baseline, and expressed as pC s⁻¹ to reflect the fluctuations in charge transfer per second during rhythmic activity. Synaptic charge was normalized to the charge measured in the cell in the layer expressing ChR2 (but this cell itself was negative for ChR2 expression). For example, in scnn1‐tg3‐Cre mice in which ChR2 was expressed in L4, synaptic charge in the L2/3, L4 and L5 cells were normalized to that measured in the L4 cell. The optogenetic modulation index (OMI) is a normalized measured (between −1 and 1) that represents whether optogenetic activation of a group of neurons increases of decreases the firing rate of the cell under consideration, and is calculated as the difference over the sum of the mean firing rates in photo‐stimulated and control conditions. For all intracellular experiments involving triple or paired recording, the reported sample size value corresponds to the number of triples or pairs. For in vivo recordings, the sample size corresponds to the number of animals unless otherwise stated. Coherence between the synaptic currents in cell pairs was determined using the Chronux package with the same number of tapers as the power analysis. The coherence spectra show 95% confidence intervals.

Statistical analysis

All statistical analysis was performed in the Matlab environment. For unpaired analysis the Wilcoxon rank sum test was use, for paired the signed rank test was used, and for multiple comparisons one‐way ANOVA was used where appropriate.

Histology

Animals were anaesthetized with a combination of ketamine (100 mg kg⁻¹) and xylazine (10 mg kg⁻¹) and perfused with cold PBS followed by cold 4% PFA. Brains were dissected and post‐fixed for 2 h at 4°C, rinsed 3 × 15 min in PBS, and cryopreserved for 24 h in 30% sucrose in PBS at 4°C. Brains were then sectioned using a frozen microtome at a thickness of 40 um. Floating sections were immediately blocked (0.6% Triton X‐100, 0.2% Tween‐20, 3% normal goat serum, and 3% BSA in PBS) for 1 h at 4°C, and then incubated in mouse anti‐NeuN primary antibody (EMD Millipore) overnight in block at 4°C. The next day, sections were washed 3 × 15 min in PBS‐T (PBS with 0.25% Triton X‐100), incubated in Alexa Fluor 405 goat anti‐mouse secondary antibody (Life Technologies) for 2 h in block at room temperature, and washed 3 × 15 min in PBS‐T. Sections were then mounted with Vectashield Hardset (Vector Laboratories). Confocal images of visual cortex were stitched and cells were counted using Fiji (ImageJ). Counts of neurons were done in a 300 μm wide box drawn across all cortical layers and averaged over 3 sections. Layer boundaries were identified using cytological differences of neurons stained with NeuN.

Results

Excitatory neurons in all cortical layers can induce gamma rhythms

In order to address the laminar origins and impacts of rhythmic activity in the neocortex, ChR2 was expressed selectively in excitatory neurons in individual layers of mouse primary visual cortex (V1) using appropriate layer‐specific Cre lines (for L4, L5, and L6) or in utero electroporation for L2/3. First, extracellular and intracellular recordings were made in V1 from anaesthetized mice expressing ChR2 in layer 4 (L4, scnn1‐tg3‐Cre line) (Madisen et al. 2010) and the cortex was either visually stimulated with drifting gratings or stimulated directly by illuminating the cortex itself with blue light to optogenetically activate L4 excitatory neurons (see Methods for selection of illumination intensity). Consistent with prior reports (Nase et al. 2003; Niell & Stryker, 2008; Welle & Contreras, 2016), high contrast visual gratings generated strong oscillatory activity, recorded extracellularly via the local field potential (LFP), and a mixture of excitation and inhibition, recorded intracellularly with whole cell patch clamp from neurons in L2/3 (Fig. 1 A–F). In the same cells, this sensory response was compared to the dynamics of the response to optogenetic stimulation of layer 4. Rather than impose a specific temporal pattern of activity with trains of light pulses, layer 4 was optogenetically stimulated with low intensity ramps of blue light (ramps rather than step pulses, to compensate for the intrinsic desensitization of ChR2; Adesnik & Scanziani, 2010). The slope of the ramp was adjusted for each animal to obtain network activity (LFP response) with a largely stable power for the duration of the light stimulus. Interestingly, minimally structured optogenetic stimulation of L4 in V1 drove an oscillatory pattern of activity that was similar in its power spectrum in the LFP from visual stimulation (Fig. 1 B–D; mean peak frequency visual stimulation: 38 ± 3 Hz, mean peak frequency L4 optogenetic stimulation: 44 ± 2 Hz, P = 0.25, rank sum test, n = 6 mice). The same optogenetic stimulus also drove a mixture of synaptic excitation and inhibition in L2/3 cortical neurons (Fig. 1 F; mean excitation during visual stimulation: 90 ± 20 pC s⁻¹, mean excitation L4 optogenetic stimulation: 50 ± 10 pC s⁻¹, P < 0.05; mean inhibition visual stimulation: 130 ± 40 pC s⁻¹; mean inhibition L4 optogenetic stimulation: 200 ± 40 pC s⁻¹, P < 0.05, n = 5 cells, rank sum test). It is important to note that while the visual stimulus drives activity throughout the entire cortical circuit, engaging the thalamus and all cortical layers, optogenetic stimulation in this L4‐Cre line drives activity that exclusively originates from layer 4 excitatory cells. Thus, the two evoked responses exhibited similarities in the temporal pattern of their V1 response with respect to their mean centre frequency, but also differences in the evoked excitation and inhibition.

Next, whether other cortical layers could generate gamma rhythms was addressed. Indeed, ramp‐like optogenetic stimulation of excitatory neurons in L2/3, L5, and L6 generated gamma rhythms in vivo (Fig. 2 A and B; L2/3: n = 4 mice; L4: n = 9 mice; L5: n = 8 mice; L6: n = 5 mice), and extracellularly recorded units across the depth of the cortex showed pronounced phase locking to the ongoing gamma oscillation in the LFP as evidenced by the narrow distribution of phases at which they occur and their pairwise phase consistency (Vinck et al. 2010; Fig. 2 E–H). These data establish that populations of excitatory neurons in any of the four main cortical layers can induce or initiate cortical oscillations that potently synchronizes and entrains neural activity across the cortical layers.

As in vivo, photo‐stimulation of the same neuronal sub‐classes in vitro also generated potent gamma oscillations (Fig. 3 A–F; peak frequency, L4: 36 ± 1 Hz, L2/3: 28 ± 1 Hz, L5: 38 ± 2 Hz; L6: 51 ± 3 Hz, n = 8 cells each). Notably, the optogenetic stimulation generated oscillations with statistically different centre frequencies both in vivo and in vitro (Figs 2 C and 3 F, P < 0.05, one‐way ANOVA), with photo‐stimulation of L6 evoking the highest frequency gamma oscillation in both conditions. This difference could be due to distinctions in the intrinsic properties of cells in different layers, the GABAergic circuits that might be recruited in each condition, or other sources. To assess how varying stimulus strength drove different amounts of excitation and inhibition, L2/3 neurons were recorded in vivo while varying stimulus contrast, or in vitro, in L4‐Cre mice while the slope of the blue light ramp was increased. In both cases, excitation and inhibition increased monotonically with stimulus strength (Fig. 3 G and H), consistent with the monotonic increase in membrane potential and synaptic currents with stimulus contrast previously observed (Contreras & Palmer, 2003; Adesnik, 2017).

The cortical impact of L4 during photo‐induced gamma rhythms

Although excitatory neurons in different cortical layers could induce gamma oscillations, they might generate different patterns of excitation and inhibition across their target layers. While the well‐known pattern of monosynaptic excitatory connectivity in the cortex can provide simple predictions for the pattern of excitation that might be observed, it is far more difficult to predict what the net inhibition in the network would be during layer‐specific gamma rhythm generation. Therefore, a set of experiments was performed in which one cortical layer was photo‐stimulated and low resistance voltage clamp recordings were used to measure synaptic excitation (E), inhibition (I), and their balance.

First, the cortical E/I balance generated by L4 during gamma activity was examined. ChR2 was expressed selectively in L4 excitatory neurons (Fig. 4 A–C) and simultaneous triple whole cell voltage clamp recordings were made from a single neuron in L2/3, L4 and L5 in acute brain slices (Fig. 5 A). All recordings were from neurons that did not express ChR2 themselves to avoid contamination by its photo‐conductance. Additionally, neurons with pyramidal shaped somata were specifically targeted, which typically correspond to excitatory neurons, though some recordings may have been from inhibitory neurons.

The illumination area was set to photo‐stimulate an approximately 200–300 μm wide region of the layer. All voltage clamp recordings were conducted in cells loaded with Cs⁺ and the ion channel blocker QX‐314, which provides strongly improved space clamp (Adesnik, 2017), although our measurements are still likely to be underestimates of the true conductances. Using the same optical stimulus that could mimic visually evoked activity in vivo, L4 cells in scnn1‐tg3‐Cre mice were activated and the resulting excitatory and inhibitory conductances were measured in the three postsynaptic cells by voltage clamping at the corresponding reversal potentials for excitation (EPSCs) and inhibition (IPSCs). It was essential that data were collected from simultaneous whole cell recordings to control for variability in ChR2 expression and in the slice preparation itself.

Excitatory neurons in L4 are known to synapse on multiple subtypes of inhibitory neurons across the cortical layers, and thus may provide network inhibition to all of their downstream layers (Helmstaedter et al. 2008; Xu et al. 2013; Pluta et al. 2015). Conversely, each target layer might receive a specific E/I balance that would differentially influence it firing. It was found that ramp‐like photo‐stimulation of L4 excitatory neurons in scnn1a‐tg3‐Cre mice drove substantial rhythmic excitatory input to postsynaptic excitatory neurons in L4, L2/3 and L5, consistent with their known monosynaptic excitatory connectivity (Burkhalter, 1989; Feldmeyer et al. 2002; Lefort et al. 2009; Fig. 5 B and C; mean excitatory charge L4: 56 ± 6 pC s⁻¹; L2/3: 64 ± 7 pC s⁻¹; L5: 53 ± 5 pC s⁻¹; n = 10 triple recordings, 10 paired recordings always including a L4 neuron, P = 0.48, Kruskal Wallis ANOVA). To measure inhibitory currents, the three recorded neurons were then voltage clamped at the excitatory reversal potential. The recordings showed that inhibition was substantially stronger in the L4 neuron, compared to the neurons in L2/3 or L5 (Fig. 5 B and C; mean inhibitory charge L4: 220 ± 20 pC s⁻¹; L2/3: 140 ± 20 pC s⁻¹; L5: 130 ± 20 pC s⁻¹, P < 0.005, Kruskal Wallis ANOVA). All three postsynaptic cells received a mixture of E and I, but the net E/I balance in L4 was far more tilted towards inhibition than in the other two layers (E/I ratio L4: 0.27 ± .04; L2/3: 0.67 ± 0.12; L5: 0.58 ± 0.15, P < 0.05, Kruskal Wallis ANOVA). Inhibitory currents showed strong coherence in the gamma band across all pairs of recorded cells (Fig. 5 D), providing a potential mechanism for the strong synchronization of neural activity observed in vivo under similar photo‐stimulation conditions.

To test how these E/I balances might influence spiking activity across L4's target layers, a subsequent set of triple recordings were made in current clamp. Each neuron was made to spike through direct current injection in order to monitor bidirectional changes in firing rates. Consistent with the measured E/I balances, photo‐stimulating L4 powerfully suppressed other L4 neurons (optogenetic modulation index (OMI): −0.58 ± 0.09, n = 8 cells, P < 0.005, signed rank test), but facilitated the activity of neurons in L2/3 (OMI: 0.20 ± 0.07, n = 8, P < 0.05, signed rank test) and L5 (OMI: 0.12 ± 0.04, n = 9 cells, P < 0.05, signed rank test; Fig. 5 E and F). The suppression of L4 cell spiking could be explained by a net hyperpolarization of the mean membrane potential in L4 cells (mean V _m change: −4.6 ± 1.3 mV; range: 0.6–14.9 mV change, Fig. 5 E, inset, P < 0.005 signed rank test). This suggests that L4‐induced gamma oscillations have the net impact of restraining activity within L4, but promoting activity downstream in layers 2/3 and 5, consistent with the significant differences in the E/I balance L4 generates.

The cortical impact of L5 during photo‐induced gamma rhythms

Would a similar pattern of strong intralaminar inhibition but translaminar excitation be found when driving other cortical layers? To address this question, the impact of photo‐stimulating layer 5 was addressed next. L5 pyramidal cells are well known to recruit multiple subtypes of nearby inhibitory neurons, including those that target their soma and dendrites, within L5 (Silberberg & Markram, 2007; Jiang et al. 2015). These inhibitory neurons, in turn, send axons that ramify not only within L5, but also across nearly the entire extent of the vertical axis of the cortex (Jiang et al. 2015; Munoz et al. 2017). Thus, activation of L5 excitatory neurons might also provide potent inhibition to most, if not all, cortical layers. L5 pyramidal cells, particularly of the subtype that sends a dense intracortical axon, also connect to excitatory neurons across the cortical layers. Despite what is known about L5's synaptic targets, the net E/I balance L5 generates across cortical layers is essentially unknown. To empirically address these questions, a L5‐specific Cre line (Rbp4‐Cre) was used that drives Cre expression in subsets of L5 pyramidal neurons but not in inhibitory neurons (Fig. 4 D–F). Again, light ramp photo‐stimulation was found to drive an oscillatory pattern of network activity (Fig. 6 A and B). As above, simultaneous triple whole cell recordings across layers 2/3, 4 and 5 were next used to measure synaptic output from L5. Layer 5 photo‐stimulation drove prominent synaptic excitation in other L5 cells, but much less excitation in L4 and L2/3 neurons (Fig. 6 B and C; mean excitatory charge L4: 12 ± 1 pC s⁻¹; L2/3: 23 ± 2 pC s⁻¹; L5: 60 ± 7 pC s⁻¹; n = 11 triple recordings and 13 paired recordings always including a L5 cell, P < 0.0005, Kruskal Wallis ANOVA). Upon isolating synaptic inhibition, it was found that L5 photo‐stimulation drove powerful intralaminar inhibition within L5, but much weaker inhibition in the upper cortical layers (Fig. 6 B and C; mean inhibitory charge L4: 26 ± 5 pC s⁻¹; L2/3: 60 ± 10 pC s⁻¹; L5: 370 ± 40 pC s⁻¹; P < 0.005, Kruskal Wallis ANOVA), resulting in a substantially lower E/I ratio in L5 than in the other two layers (E/I ratio L4: 0.8 ± 0.2; L2/3: 1.1 ± 0.2; L5: 0.2 ± 0.2). Therefore, despite the unique intracortical connectivity patterns of L5 compared to L4, the same principle held true: an intralaminar E/I balance strongly tilted towards I, but a translaminar balance tilted more towards E. As for L4 photo‐stimulation, photo‐stimulating L5 cells generated strong coherence between IPSCs recorded between pairs of neurons (Fig. 6 D).

In L4, this pattern of layer‐specific E/I balances resulted in intralaminar suppression, but translaminar facilitation. Would the same hold true during photo‐stimulation of L5? A subsequent set of intracellular current clamp recordings demonstrated that driving L5 suppressed activity in ChR2⁻ L5 cells (OMI: −0.7 ± 0.1, n = 13 cells, P < 0.0005, signed rank test), but facilitated activity in neurons in L2/3 (OMI: 0.15 ± 0.09, n = 19 cells, P < 0.05, signed rank test), although no significant impact was observed in L4 (OMI: −0.02 ± 0.04, n = 10 cells, P = 0.47, signed rank test; Fig. 6 E and F). Thus, the same pattern of within layer suppression combined with simultaneous translaminar facilitation (in this case L2/3) holds true for L5 activity as well. The lack of impact on L4 may be a consequence of the relatively small amount of synaptic drive L5 provides to L4.

The cortical impact of L2/3 during photo‐induced gamma rhythms

Finally, the impact of L2/3 neuronal activity on other cortical layers was addressed. Despite prior work in the barrel cortex (Adesnik & Scanziani, 2010), the impact of photo‐stimulating L2/3 on other cortical layers in the visual cortex is not known. If L2/3 does generate a balance tilted towards I within L2/3, and one tilted towards excitation in L5, its primary translaminar target, it might suggest that the pattern of layer specific E/I balances seen for L4 and L5 may be generalizable both across cortical layers, and even across sensory cortical areas.

To test this idea, ChR2 was expressed specifically in L2/3 pyramidal cells via timed in utero electroporation and ChR2‐expressing L2/3 neurons were photo‐stimulated with ramps of blue light. Again, triple whole cell recordings were used to quantify the excitation and inhibition ChR2‐expressing neurons generated in their targets. Photo‐stimulation of L2/3 drove excitation primarily within L2/3 and in L5 pyramidal cells, but only limited excitatory input to L4, consistent with prior work in the barrel cortex (Lefort et al. 2009; Adesnik & Scanziani, 2010; Fig. 7 B and C; mean excitatory charge L4: 17 ± 5 pC s⁻¹; L2/3: 50 ± 10 pC s⁻¹; L5: 60 ± 10 pC s⁻¹, n = 11 triple recordings, P < 0.005, Kruskal Wallis ANOVA). This is in contrast to photo‐stimulations of L4, which drove excitation non‐specifically in all three upper cortical layers. Next, it was observed that stimulation of L2/3 drove powerful intralaminar inhibition in other L2/3 pyramidal cells, significantly less inhibition in L5 neurons, and minimal inhibition in L4 (Fig. 7 B and C; mean inhibitory charge L4: 40 ± 10 pC s⁻¹; L2/3: 330 ± 60 pC s⁻¹; L5: 190 ± 60 pC s⁻¹; P < 0.005, Kruskal Wallis ANOVA). These results confirm that L2/3's primary targets in V1 are neurons within its own layer and in L5, but not in L4. Furthermore, as was found for L4 photo‐stimulation, the strongest network inhibition was evoked in the same layer in which the activity originated, driving a significantly lower E/I ratio in L2/3 compared to the other two layers (E/I ratio L4: 0.67 ± 013; L2/3: 018±.01; L5: 039±.05, P < 0.005, Kruskal Wallis ANOVA), consistent with prior data in S1. This supports the notion that strong intralaminar inhibitory feedback may be a hallmark of intracortical E/I balance during rhythmic activity. Coherence spectra for IPSCs recorded in pairs of cells showed prominent peaks in the gamma band (Fig. 7 D). Interestingly, peak coherence occurred at a lower frequency for L2/3 photo‐stimulation, compared to that of L4 and L5. This is in line with the significantly lower peak frequency of the gamma oscillation when stimulating L2/3 (Fig. 3 F), and may be due to the intrinsic physiological properties of L2/3 pyramidal neurons, or perhaps to the inhibitory source of these IPSCs (for instance, slower dendritic inhibition from somatostatin cells, compared to fast somatic inhibition from parvalbumin cells; Adesnik et al. 2012; Veit et al. 2017).

What is the net impact of these layer‐specific E/I balances from L2/3? Subsequent experiments using intracellular current clamp recordings showed that photo‐stimulating L2/3 almost completely suppressed activity in other L2/3 neurons (those not expressing ChR2, OMI: −0.90 ± 0.06, n = 8 cells, P < 0.005), but facilitated activity in downstream L5 cells (OMI: 0.28 ± 0.06, n = 7, P < 0.05; Fig. 7 E and F). Thus, as with L4 and L5, the induction of rhythmic activity within L2/3 drives potent intralaminar suppression but translaminar activation. Taken together, these data demonstrate that layer‐specific induction of rhythmic activity by excitatory neurons in any of the upper three cortical layers generates powerful, intralaminar inhibition that suppresses activity within the layer of origin, but drives a net weaker inhibition across layers that permits the propagation of activity through downstream targets. Since it was not practical to record or photo‐stimulate all cortical subtypes, it remains to be seen if this principle generalizes to all cortical excitatory neuron subtypes in any cortical layer. Very similar experiments have been previously conducted in brain slices layer 6 (Olsen et al. 2012), and these data are put into context below.

Discussion

Using layer and cell‐type specific optogenetic manipulations, this study explored the cells and circuits that are sufficient to induce neuronal synchronization in the sensory neocortex, and the excitation and inhibition they generate in their targets. The first key finding is that optogenetically stimulating subsets of excitatory neurons in any cortical layer, with a nearly unstructured light stimulus, generated gamma rhythms that propagated across the vertical axis of the cortex, and potently synchronized and entrained neuronal spiking across cortical layers. It is important to note that the gamma frequency response (20–80 Hz) emerged intrinsically from the cortical network, and was not imposed by the external optogenetic stimulus, consistent with extensive prior work that various sensory, behavioural, or pharmacological manipulations also drive gamma rhythms, although it should be noted that the slope of the light ramp was adjusted to maintain a gamma oscillation with largely stable power. The optogenetic data thus demonstrate that activity originating in any cortical layer is sufficient to give rise to gamma rhythms in the visual cortex, measured through the LFP, spiking activity, or synaptic currents. Stimulating cells in different layers drove gamma rhythms with significantly different mean centre frequencies, an interesting topic for explanation in future work, possibly due to intrinsic physiological differences in the neurons that induce the rhythms, or the nature of the inhibitory circuits that may be critical for the entrainment. In vivo, when multiple cortical layers may be simultaneously contributing to gamma rhythms as during sensory stimulation, the absolute frequency of the net resulting rhythm and the degree of its translaminar coherence may represent a mean of these different layer specific gamma frequencies, or the frequency of one layer may dominate and ensure coherence across the vertical axis of the cortex. These questions might be addressed by optogenetically silencing different layers during sensory induced gamma rhythms, or experiments with two different colour opsins expressed in two different cortical layers.

Physiological recordings in vivo have demonstrated that cortical rhythms occur in different cortical layers (Ainsworth et al. 2011; Whittington et al. 2011), and their power and coherence can vary with stimulus parameters, brain state, and behavioural context (Singer & Gray, 1995; Buzsaki & Draguhn, 2004; Fries et al. 2007; Fries, 2009). Thus, as in the hippocampus, it appears likely that these different gamma rhythms are mechanistically distinct, both in the cell types that induce them, and in their functional impact across the cortex. Although this study used optogenetic rather than sensory or behavioural stimulation to induce gamma rhythms, the data demonstrate that subsets of excitatory neurons in all principal cortical layers are sufficient to drive robust gamma rhythms. Exactly how these optogenetically induced gamma rhythms relate to those induced pharmacologically or to sensory stimulation in prior studies remains to be determined, since pharmacology and sensory stimulation cannot be layer specific. Nevertheless, if these optogenetically‐induced gamma rhythms reflect circuits that naturally generate gamma rhythms in the cortex, it seems possible that different layers might be particularly important for entraining rhythmicity in different behavioural or sensory states.

The second key finding in this study, perhaps more interesting than the first, is that each layer generates target‐specific differences in the net E/I ratio in different cortical layers. All layers induced gamma rhythms in cells in their own layer that were characterized by a relatively low E/I balance – that is one tilted more strongly towards inhibition and net suppression of activity. This powerful recurrent inhibition is likely to be crucial for stabilizing feedback excitation (Tsodyks et al. 1997; Pinto et al. 2003; Ozeki et al. 2009) that leads to gamma band activity (Jadi & Sejnowski, 2014a,b) by providing the basis for the cortex to enter an inhibitory stabilized network (Ozeki et al. 2009). This inhibition would then contribute to fundamental cortical computations such a normalization and contextual modulation (Carandini et al. 1997; Ozeki et al. 2009; Nassi et al. 2015).

At the same time, the data from this study also demonstrate that across layers, gamma rhythms induced by the upper three cortical layers drive an E/I balance tilted more towards excitation. This translaminar bias towards excitation may promote the effective propagation of activity through the core cortical microcircuit from L4 to L2/3 to L5 during gamma rhythms. This difference in the E/I balance within versus across layers helps explain how within a layer, inhibition acts to stabilize excitation and prevent runaway activity, while across layers it can promote downstream spiking.

The Cre lines or the method of in utero electroporation used in this study only labelled subsets of excitatory neurons in each layer. It is not yet certain whether these are random subsets that draw from each subtype in each layer, or target specifically certain subtypes. Large‐scale single cell sequencing is currently being used to address this question (J. Ngai, H. Adesnik, unpublished). Similarly, the postsynaptic recordings in this study were from pyramidal shaped neurons across the cortical layers. While it is not clear whether the results of this study will generalize to all types of cortical excitatory neurons, extending these experiments and combining with post hoc single cell sequencing can address this question.

Notably, under nearly identical experimental conditions as in this study, photo‐stimulation of L6 corticothalamic cells in brain slices (Olsen et al. 2012; Bortone et al. 2014) also revealed strong intralaminar inhibition. However, it was found that L6 photo‐stimulation generated potent suppression of neurons in all layers tested (but see Kim et al. 2014). However, in contrast to L6 corticothalamic neurons, which primarily drive inhibition across all cortical layers (Olsen et al. 2012; Bortone et al. 2014), most of the translaminar circuits from layers 2/3, 4, and 5 drive a more ‘balanced’ ratio of excitation and inhibition (∼1:1–1:3). Thus while the impact of L6 intracortical output is primarily suppressive and reduces cortical gain (Olsen et al. 2012), the translaminar effects of the other three layers during rhythmic activity are more likely to facilitate the propagation of activity, and perhaps contribute to feature tuning per se (Womelsdorf et al. 2012).

It is important to note that optogenetic activation of cortical circuits carries both strengths and weaknesses. As a gain‐of‐function manipulation, its main strength is its ability to reveal the synaptic and network dynamics that a genetically defined pool of cortical neurons is sufficient to initiate on its own. A sensory stimulus, in contrast, engages the entire circuit from sensory periphery to deep in the cortex, making it difficult to assign a contribution to a specific neuronal subtype embedded within this network. A critical caveat of any optogenetic stimulation experiment is that although the activation is targeted to a genetically defined population of neurons, it cannot recreate the exact pattern of stimulus‐driven activity since even genetically similar neurons exhibit a wide heterogeneity of functional properties, such as orientation tuning. Nevertheless, the results presented here outline the possible contributions of each layer, on its own, to the excitatory/inhibitory balance that might naturally underlie gamma activity in vivo. It should also be emphasized that the gamma oscillations generated in this study did not require the photo‐stimulus to have any temporal structure. Rather, these data demonstrate that simple depolarization of excitatory neurons in each cortical layer inherently engages gamma band activity, plausibly according to the commonly employed PING model (Pyramidal‐Inhibitory neuron‐Network‐Gamma) (Tiesinga & Sejnowski, 2009; Buzsaki & Wang, 2012). Future work with much more sophisticated optogenetic methods, such as multiphoton holography (Papagiakoumou et al. 2010; Packer et al. 2015; Carrillo‐Reid et al. 2016; Pegard et al. 2017), might be able to better recreate even more physiological patterns of activity to ask how different ensembles of neurons with functionally identified properties (such as common orientation tuning) differentially contribute to gamma activity.

Taken together, the data in this study empirically assess the laminar origins of the E/I balance in the neocortex during layer‐specific optogenetically induced gamma oscillations. These are measurements that could not be readily predicted based on prior connectivity mapping, since the net excitation and inhibition a given layer will drive in its targets is a sophisticated function of a large number of variables that are difficult to measure. Thus these experiments both shed light on the potential origins of the E/I balance in cortical neurons during gamma rhythms, and also provide a powerful data set to guide, validate and revise sophisticated models of neocortical dynamics (Markram et al. 2015). It remains a matter of debate whether gamma oscillations in the neocortex per se are important for various aspects of neural computation or perception. The data in this study do not readily address this debate. However, on the one hand, they demonstrate that cortical circuits can be driven to oscillate with a minimally structured activation of subsets of excitatory cells in any given layer, potentially helping to explain why cortical circuits so readily oscillate under a wide array of ‘natural’ conditions. Conversely, the data from this study also demonstrate that different circuits (i.e. different cortical layers) generate similar gamma band activity and E/I ratios regardless of their many underlying differences. This might imply that the gamma oscillation generated by different circuits may be a simple consequence of net activity in the circuit, and not necessarily be utilized by the system for any particular purpose. However, the optogenetic approach used in this study could still be used as a platform to address how the cortical E/I balance may be perturbed in disease models, as impairments in the E/I balance and gamma rhythms have been linked to a host of neuropsychiatric disorders (Uhlhaas & Singer, 2010; Yizhar et al. 2011; Cho et al. 2015).

Additional information

Competing interests

None declared.

Funding

This work was supported by NEI grant R01EY023756‐01 and an HHMI‐Helen Hay Whitney Postdoctoral Fellowship to H.A. H.A. is a New York Stem Cell Foundation Robertson Investigator. This work was supported by the New York Stem Cell Foundation.

Biography

Hillel Adesnik is a professor of neurobiology at University of California, Berkeley where his lab studies the neural basis of perception and sensory coding. He did his training studying hippocampal and cortical circuits with Drs Roger Nicoll and Massimo Scanziani. This manuscript highlights an organizing principle for excitation and inhibition generated between cortical layers, particularly during gamma band activity. Ultimately, the goal of his research is to understand how patterns of excitation and inhibition in the cortex mediate neural computation, helping us perceive and identify objects in the world.