1. Neuroscience
Download icon

Corticothalamic gating of population auditory thalamocortical transmission in mouse

  1. Baher A Ibrahim
  2. Caitlin A Murphy
  3. Georgiy Yudintsev
  4. Yoshitaka Shinagawa
  5. Matthew I Banks
  6. Daniel A Llano  Is a corresponding author
  1. Department of Molecular and Integrative Physiology, University of Illinois, United States
  2. Beckman Institute for Advanced Science and Technology, University of Illinois, United States
  3. Department of Neuroscience, School of Medicine and Public Health, University of Wisconsin-Madison, United States
  4. Department of Anesthesiology, School of Medicine and Public Health, University of Wisconsin-Madison, United States
  5. Neuroscience Program, University of Illinois, United States
  6. College of Medicine, University of Illinois, United States
Research Article
  • Cited 0
  • Views 562
  • Annotations
Cite this article as: eLife 2021;10:e56645 doi: 10.7554/eLife.56645

Abstract

The mechanisms that govern thalamocortical transmission are poorly understood. Recent data have shown that sensory stimuli elicit activity in ensembles of cortical neurons that recapitulate stereotyped spontaneous activity patterns. Here, we elucidate a possible mechanism by which gating of patterned population cortical activity occurs. In this study, sensory-evoked all-or-none cortical population responses were observed in the mouse auditory cortex in vivo and similar stochastic cortical responses were observed in a colliculo-thalamocortical brain slice preparation. Cortical responses were associated with decreases in auditory thalamic synaptic inhibition and increases in thalamic synchrony. Silencing of corticothalamic neurons in layer 6 (but not layer 5) or the thalamic reticular nucleus linearized the cortical responses, suggesting that layer 6 corticothalamic feedback via the thalamic reticular nucleus was responsible for gating stochastic cortical population responses. These data implicate a corticothalamic-thalamic reticular nucleus circuit that modifies thalamic neuronal synchronization to recruit populations of cortical neurons for sensory representations.

Introduction

Our experience of the world relies on how sensory information is processed. Thalamocortical projections are critical for activation of the cerebral cortex, which is thought to contain the neural circuits that drive our conscious awareness of sensory stimuli. A classical view of thalamocortical function is that cortical activity during sensory perception is predictable based on activity in the thalamus and patterns of synaptic convergence of thalamocortical axons onto cortical neurons. Thus, sensory processing in this classical scheme behaves as a series of hierarchical linear filters, whose tuning is modified by feedback, creating more complex response properties that reach the level of conscious perception by activating the cortex (Riesenhuber and Poggio, 1999; Vidyasagar and Eysel, 2015). However, this view does not comport with findings that population activity in sensory cortices is often stereotyped and recapitulates patterns of cortical spontaneous activity (MacLean et al., 2005; Kenet et al., 2003; Miller et al., 2014; Sakata and Harris, 2009). As such, a hypothesis has emerged that sensory representations are developed by early exposure to sensory stimuli and stored in the cortex within intracortical networks, and that the thalamus activates these pre-wired sensory representations upon sensory stimulation (MacLean et al., 2005; Petersen, 2005). Some observations that support this hypothesis are that spontaneous cortical activity is highly determined by the internal cortical connectivity (Kenet et al., 2003; Sanchez-Vives and McCormick, 2000), and is not easily predictable from activity in thalamocortical afferents (MacLean et al., 2005; Cossart et al., 2003). In addition, spontaneous cortical activity is the main substrate for the generation of the internal percepts during states without external sensory input, such as memory recall and dreaming (Takeuchi et al., 2011; Nir and Tononi, 2010). Further, despite substantial differences in the form and organization in the initial processing stages of different modalities of sensation, at the level of the thalamus and cortex, neural circuits across modalities are relatively homogeneous (Shepherd, 2011; Douglas et al., 1989; Phillips et al., 2019). These findings suggest that there is a common function of thalamocortical circuits that is not tied to specific modalities of perception.

Given that connected neuronal ensembles are likely the main functional unit for behavior and cognition (Uhlhaas et al., 2009; Buzsáki, 2010), a major control point for the activation of cortical ensembles, and therefore cognition more generally, may lie in the thalamus. Here, we examine the mechanisms by which all-or-none population responses (hereafter called ‘Population ON’ or ‘Population OFF’ responses) in the auditory cortex (AC) are gated. We find that gating of cortical activity occurs via corticothalamic projection to the thalamic reticular nucleus (TRN), which is a long-enigmatic structure that partially surrounds and sends GABAergic projections to thalamocortical neurons. Further, we observed that population OFF responses in the AC are associated with increases in synaptic inhibition and neuronal desynchronization in the auditory thalamus. These findings suggest that one mode of thalamocortical function is to select groups of cortical neurons for activation based on feedback from cortical layer 6.

Results

Stochastic AC responses to sound presentations in vivo

Transcranial calcium imaging of the AC of an anesthetized GCaMP6s mouse following repeated presentations of a 5 kHz-37 dB SPL pure tone (Figure 1A) revealed cortical activity in three distinct areas, the primary AC (A1), secondary AC (A2), and anterior auditory field (AAF), consistent with previous work (Issa et al., 2014Figure 1B, MATLAB code in Figure 1—source code 1). Such sound-evoked cortical activity represented the average of the cortical responses to 40 presentations of the same tone. Across the 40 trials of the same sound presentation, we observed that A1 showed either a full population response (here called population ON cortical responses) or no response (referred to here as population OFF cortical responses), with some variability within each class (Figure 1C, MATLAB code in Figure 1—source code 1). Inspection of the individual cortical responses over time revealed the responses to be independent of stimulus presentation order (Figure 1D, Figure 1—source data 1). Collectively, a histogram of Δf/f of sound-evoked calcium signals from A1 across trials showed two classes of A1 responses (‘Population ON’ vs. ‘Population OFF,’ Figure 1E, Figure 1—source data 1). A greater frequency of the population ON cortical responses was observed with increasing sound pressure level (Figure 1F, Figure 1—source data 1). Similar stochastic population sensory cortical responses have been observed previously (MacLean et al., 2005; Kenet et al., 2003; Miller et al., 2014; Sakata and Harris, 2009), but the mechanisms that determine whether a population ON or OFF response occurs have not yet been elucidated.

Stochastic auditory cortical population responses to repeated sound presentations in vivo.

(A) A cartoon image showing the experimental design of transcranial calcium imaging of the left AC of GCaMP6s mouse during a 500 ms pure tone (5 kHz) exposure to the right ear. (B) Pseudocolor image representing the average map of AC activation indicated by Δf/f of sound-evoked calcium signals following 40 trials of 5 kHz-37 dB SPL pure tone stimulation. (C) Two pseudocolor images representing an individual trial for population ON cortical response (top) versus population OFF cortical response (bottom). (D) A line graph of Δf/f of the sound-evoked Ca signals of A1 across 40 trials of playing 5 kHz, 37 dB SPL over time. (E) A histogram of the Δf/f of sound-evoked calcium signals following 40 trials of 5 kHz-37 dB SPL stimulation (bin size = 1%). (F) A bar graph showing the percentage of population ON cortical responses across different sound levels at 5 kHz pure tone; A1: Primary auditory cortex, A2: Secondary auditory cortex, AAF: Anterior auditory field, C: Caudal, D: Dorsal.

Stochastic AC responses to electrical stimuli to IC in vitro

To examine the circuit mechanisms underlying the stochastic nature of the responses (Figure 2), brain slices that retain connectivity between the inferior colliculus (IC), medial geniculate body (MGB), thalamic reticular nucleus (TRN) and AC were constructed (Figure 2—figure supplement 1Llano et al., 2014; Slater et al., 2015). The auditory colliculo-thalamocortical (aCTC) mouse brain slice retains the synaptic connections between these structures: hence, electrical stimulation of the IC evoked neuronal activity in all of these brain structures as indicated by stimulus-evoked flavoprotein autofluorescence (FA) and calcium-dependent fluorescence signals (Figure 2A–E). Figure 2B and D, (MATLAB codes in Figure 2—source code 1 and 2), show the average of the stimulus-evoked activity of the connected brain structures in the aCTC slice after 10 trials of IC stimulation. However, review of the individual time series of Δf/f of the stimulus-evoked FA and calcium signals across trials of IC stimulation (Figure 2C and E, Figure 2—source data 1 and 2) reveals stochastic responses in the AC; a pattern that was similar to the in vivo data, despite the fact that MGB, TRN, and IC were always responsive to IC stimulation. The FA and calcium signals were determined based on predetermined criteria (Figure 2—figure supplement 2, Figure 2—figure supplement 2—source data 1, see Materials and methods). This finding was consistent with whole-cell recording from MGB and TRN cells (Figure 2—figure supplement 3), where spikes were seen in response to each stimulus irrespective of the presence of a population AC response. We note that similar to previous studies, FA and calcium signals from the MGB are smaller than those from AC (Dana et al., 2014; Llano et al., 2009), which may be related to limited alignment of mitochondria-containing neuropil processes or low baseline expression of GCaMP in the MGB, respectively.

Figure 2 with 6 supplements see all
Stochastic auditory cortical responses to repeated electrical stimuli in vitro.

(A) A cartoon image showing the experimental design of simultaneous FA or calcium imaging and IC stimulation of the aCTC slice, which were also associated with LFP or whole-cell recording from cortical 3/4 layers in other experimental setups. (B and D) Pseudocolor images showing the neuronal activation indicated by evoked FA or calcium signals, respectively, in the IC, MGB, TRN, and AC following IC stimulation (nine animals). (C and E) The time series of Δf/f of evoked FA or calcium signals, respectively, in the IC, MGB, TRN, and AC following IC stimulation. (F) The time series of the L3/4 LFP signals (top panel) and Δf/f of the evoked cortical FA signals (bottom panel) following IC stimulation. (G) The time series of evoked calcium signals of a small population of layer 4 cells following IC stimulation, the two rose ribbons indicate the absence of neuronal activation following the IC stimulation during trial # 9 and 11. (H) The time series of layer four whole-cell recording (top panel) and the Δf/f of the evoked cortical FA signals (bottom panel) following IC stimulation, green box: The magnification of the post-stimulus activity of layer 4 cells showing evoked UP state associated with action potentials; Black arrows refer to the occurrence of population OFF cortical responses indicated by the absence of cortical FA or calcium signals as well as post-stimulus cortical LFP signals, gray vertical lines indicate the onset of IC stimulation, Yellow circle indicates the position of the electrical stimulation of the IC; aCTC: auditory colliculo-thalamocortical mouse brain slice, C: Caudal, L: Lateral.

To ensure that metabolic or imaging artifacts did not drive the observation of these stochastic population cortical responses, local field potentials (LFPs) were recorded in the AC while simultaneously conducting FA imaging, and the LFPs also showed all-or-none response patterns that strongly correlated with the FA signals (Figure 2F, Figure 2—source data 3). We also observed stochastic population AC responses in brain slices prepared using a different anesthetic (isoflurane) without transcardiac perfusion in a different laboratory using a biphasic stimulator vs. monophasic stimulator (Figure 2—figure supplement 4), suggesting that the stochastic responses were not specific to a particular slice preparation or stimulation technique.

Looking more closely at the activity of cortical layer 4 cells, the stimulus-evoked calcium signals of a small population of layer 4 cells following IC stimulation showed that a similar population layer 4 cells was activated with each population ON response and that population OFF responses were associated with an absence of layer four responses. For instance, all neurons ceased to respond during stimuli # 9 and 11 (Figure 2G, rose ribbons, Figure 2—source data 4) and (Figure 2—figure supplement 5). Whole-cell recordings of layer 4 cells during population ON population responses revealed a plateau of depolarization upon which rode several action potentials (Figure 2H, green box, Figure 2—source data 5), resembling previously described UP states (Sanchez-Vives and McCormick, 2000; Rigas and Castro-Alamancos, 2007; Fanselow and Connors, 2010), which was supported by the bimodal distribution shown by the post stimulus membrane potential of such layer 4 cells during population ON cortical responses (Figure 2—figure supplement 6, Figure 2—figure supplement 6—source data 1, see, Materials and methods). Collectively, these data suggest that IC stimulation produces stochastic population responses in the AC, and that these population responses are stereotyped and represent stimulus-induced UP states.

To determine the locus of control for whether a population ON or population OFF response would be elicited, we moved the stimulating electrode from the IC to the MGB and then to the subcortical white matter (Figure 3A, MATLAB codes in Figure 3—source code 1 and 2). Population ON/OFF cortical responses occurred only after electrical stimulation of IC, and not by the direct stimulation of MGB or the subcortical white matter (Figure 3B and D, Figure 3—source data 1 and 2, Kruskal-Wallis: p=0.001, pairwise comparison: p=0.023 for IC vs. MGB or subcortical white matter and p=1 for MGB vs. subcortical white matter, Dunn's post hoc test, n = 5). To determine the impact of stimulating current amplitude on the likelihood of eliciting a population ON-cortical response, different amplitudes of stimulating current were delivered to the IC (Figure 3C, Figure 3—source data 3). Similar to the in vivo data, the percentage of population ON cortical responses increased with increasing amplitude of the stimulating current (Figure 3E, Figure 3—source data 4, RM-ANOVA: *F(3,9)=9.7, p=0.003, pairwise comparison: *p=0.009 for 50 vs. 150–200 μA, *p=0.008 for 50 vs. 250–300 μA, Bonferroni post hoc test, n = 4 slices from four animals), which was consistent with the in vivo results (Figure 1F). Moreover, while there was a log-linear relationship between the FA signals of either IC or MGB and the stimulating current amplitude (Figure 3F, Figure 3—source data 5, R2 = 0.68, p=2.2×10−19 for IC and R2 = 0.53, p=3.3×10−13 for MGB), this relationship was lost between the cortical responses and stimulating current amplitude (Figure 3F, Figure 3—source data 5, R2 = 0.004, p=0.6), suggesting that the magnitude of the cortical response is not predictable from stimulus amplitude.

The effect of stimulation location and amplitude on the occurrence of OFF-cortical responses.

(A) Pseudocolor images of the evoked FA signals in aCTC slice after electrical stimulation of the IC, the MGB, or the subcortical white matter, respectively. (B) The time series of Δf/f of the evoked FA signals in AC, MGB, and IC following electrical stimulation of IC, MGB, or the subcortical white matter, respectively. (C) The time series of Δf/f of the evoked FA signals in AC, MGB, and IC following the stimulation of the IC with different stimulating current amplitudes. (D and E) Bar graphs showing the percentage of population ON cortical responses at different loci of stimulation and across different stimulating current amplitudes at the IC, respectively. (F) Line graphs showing the relationship between stimulating current amplitude and Δf/f of FA signals from IC (top), MGB (middle), and AC (bottom), respectively; Black arrows refer to the missing cortical responses (‘population OFF’ responses) indicated by the absence of cortical FA, vertical gray lines indicate the onset of IC stimulation, and yellow circles indicate the position of the electrical stimulation; sWM: subcortical white matter.

Population OFF cortical responses are driven by inhibition in the MGB

Because population OFF cortical responses represented a full absence of stimulus-evoked cortical activity, we reasoned that population OFF cortical responses could be driven by inhibition. To investigate this idea, global disinhibition in the aCTC slice by bath application of gabazine (SR-95531, 200 nM), the GABAΑ-R blocker, was conducted. Under simultaneous FA imaging and IC stimulation, gabazine perfusion was able to retrieve all missing cortical responses compared to control (Figure 4B and C, Figure 4—source data 1 and 2, Kruskal-Wallis: p=2.5×10−7, pairwise comparison: p=5.08×10−4 for gabazine vs. control, p=4.1×10−7 for gabazine vs. wash, and p=0.34 for control vs. wash, Dunn's post hoc test, n = 14 trials obtained from five animals), which suggested that the population OFF cortical responses were driven by inhibitory inputs. We further investigated the AC and MGB to search for the site of inhibition that drove population OFF cortical responses. Whole-cell recording of cortical layer 4 or MGB cells voltage-clamped at +10 mV was conducted simultaneously with FA imaging following the stimulation of the IC to track the IPSCs during population ON vs. population OFF cortical responses. Consistent with the previous finding that stimulation of the IC evoked a post-stimulus UP-states activity in layer 4 cells during the population ON cortical responses only (Figure 2HFanselow and Connors, 2010; Haider et al., 2006), layer 4 cells demonstrated a surge of evoked post-stimulus IPSCs during population ON cortical responses only (Figure 4D). In contrast, MGB cells showed evoked post-stimulus IPSCs following every trial of IC stimulation during population ON and population OFF cortical responses (Figure 4E). Quantitatively, while these evoked cortical post-stimulus IPSCs and EPSCS in the layer 4 cells were significantly smaller during population OFF compared to population ON cortical responses (Figure 4F, Figure 4—source data 3, paired sample Wilcoxon Signed Rank Test: p=0.006 for IPSCs, n = 10 cells from four animals and p=0.036 for EPSCs, n = 6 cells from four animals), the evoked post-stimulus IPSCs in the MGB cells were larger during population OFF compared to population ON cortical responses, with no difference in the net excitatory transferred charges (Figure 4G, Figure 4—source data 4, paired sample Wilcoxon Signed Ranks Test: p=0.011 for IPSCs, n = 20 cells from seven animals and p=0.56 for EPSCs, n = 14 cells from four animals). These findings suggest that MGB cells receive more inhibition during population OFF cortical responses with no change in excitation, which led us to hypothesize that MGB activity could be modulated by inhibitory inputs during the population OFF cortical responses.

Figure 4 with 1 supplement see all
The population OFF cortical responses are driven by MGB inhibition.

(A) A cartoon image showing the experimental design of simultaneous FA imaging and IC stimulation as well as gabazine perfusion. (B) The time series of Δf/f of the evoked cortical FA signals with ACSF (control, top trace), with gabazine (middle trace), or wash by ACSF (bottom trace). (C) A plot of the results showing that the percentage of population ON cortical responses was significantly higher than that of control and wash. (D) Evoked post-stimulus IPSCs recorded from layer 4 cells (top panel), and the Δf/f of the evoked cortical FA responses (bottom panel) following the IC stimulation. (E) Evoked post-stimulus IPSCs recorded from MGB cells (top panel), and the Δf/f of the evoked cortical FA responses (bottom panel) following the IC stimulation. (F and G) Scatterplots of the area under the curve (AUC) of the evoked post-stimulus IPSCs and EPSCs recorded from AC or MGB, respectively, during population OFF cortical responses and normalized against those recorded during population ON cortical events. (H) A cartoon image showing the experimental design of simultaneous FA imaging, IC stimulation, and selective gabazine injection into MGB or AC using a picospritzer. (I and J) The time series of Δf/f of the evoked cortical FA signals during a counterbalanced gabazine injection starting into MGB then AC or AC then MGB, respectively. (K) A plot of the results showing that the percentage of population ON cortical responses were significantly higher after the injection of gabazine into MGB; Black arrows refer to the occurrence of population OFF cortical responses indicated by the absence of the cortical FA signals; Vertical gray lines indicate the onset of IC stimulation; IPSCs: Inhibitory postsynaptic currents, and EPSCs: Excitatory postsynaptic currents.

To test this hypothesis, we determined whether disinhibition of the MGB could yield population ON cortical responses to all stimuli. Under simultaneous FA imaging and IC stimulation, the specific injection of gabazine (Figure 4—figure supplement 1) into the MGB in the aCTC slice was able to significantly increase the percentage of population ON cortical responses (Figure 4I and J, Figure 4—source data 5 and 6), indicated by the Δf/f of the evoked cortical FA signals (Figure 4K, Figure 4—source data 7, RM-ANOVA: F(3,21) = 13.9, p=3.17×10−5, pairwise comparison: *p=5.9×10−5, 1.5 × 10−4, 0.0026 for gabazine in MGB vs. control, wash, and gabazine in AC, respectively, Bonferroni post hoc test, n = 8 slices from seven animals). In contrast, the same effect was not observed after selective gabazine injection into the AC (Figure 4K, Figure 4—source data 7, RM-ANOVA: F(3,21) = 13.9, pairwise comparison: p=0.73 and 1.0 for gabazine in AC vs. control and wash, respectively, Bonferroni post hoc test, n = 8 slices from seven animals), consistent with data obtained from the whole-cell recording of layer 4 cells (Figure 4D and F). Accordingly, these data confirmed that the population OFF cortical responses could be driven by thalamic inhibition.

Synchronized MGB cells are associated with population ON-cortical responses

Although the mean peak latencies of IPSCs and EPSCs received by MGB cells showed no difference during population ON and population OFF cortical responses (Figure 5A and B, Figure 5—source data 1, paired sample Wilcoxon Signed Ranks Test: p=0.28, n = 20 cells from seven animals for IPSCs and paired t-test: t(12) = −0.72, p=0.48, n = 14 cells from four animals for EPSCs), the cumulative distribution function showed that some MGB cells (~50% of the recorded cells) received earlier IPSCs during population OFF cortical responses (Figure 5C, Figure 5—source data 1). These earlier inhibitory signals received by some MGB cells could negatively impact the synchronization between the MGB cells required to pass the sensory information to the cortex. To test this hypothesis, the time courses of calcium signals of MGB cells, which were found to match the voltage signals that were simultaneously recorded in a separate experiment (Figure 5—figure supplement 1), were imaged following the stimulation of the IC, and were compared during population ON vs. population OFF cortical responses (Figure 5D and E, Figure 5—source data 2). We observed that the variance of the peak latencies of the evoked calcium signals from all thalamic cells was larger during population OFF cortical responses than population ON responses (Figure 5F, Figure 5—source data 3, paired sample t-test: t(13) = −2.37, *p=0.033, n = 14 paired responses (population ON vs. population OFF) including 339 cells from five animals). We note that the absolute latencies of the calcium imaging responses are considerably longer than those measured electrophysiologically. However, despite the relatively slow time course of the calcium response, the peak times remained relatively stable during population ON responses. These data suggest that synchronous thalamic relay cell activity is required to evoke a cortical population ON response, consistent with previous work (Bruno and Sakmann, 2006).

Figure 5 with 1 supplement see all
Desynchronized MGB cellular activity is associated with population OFF-cortical responses.

(A and B) Plots of peak latencies of IPSCs (A) and EPSCs (B) measured in the MGB after IC stimulation. No difference was seen in the mean latency during population ON vs. OFF responses. (C) Cumulative distribution functions of the peak latencies of IPSCs (dotted line) and EPSCs (solid line) during population ON (blue line) or population OFF-cortical responses (red line) showing that ~50% of MGB cells received earlier IPSCs during population OFF cortical responses (red dotted line). (D) A cartoon image showing the experimental design of simultaneous calcium imaging of MGB, LFP recording form the cortex, and the IC stimulation. (E) The sweeps of the evoked calcium signals of all activated cells during population ON (top) vs. population OFF (bottom) cortical responses indicated by the post-stimulus cortical LFP signals recorded from L3/4. (F) A scatterplot showing a higher variance of the peak latencies of MGB cells activated during population OFF cortical response compared to those activated during population ON cortical responses. Vertical gray line indicates the onset of IC stimulation.

Corticothalamic layer 6 cells are the main driver of the missing cortical response via TRN

The data described above suggest that MGB cells receive more inhibition during population OFF cortical responses. Therefore, we investigated the source of these potential inhibitory inputs. As previously reported, the MGB can be inhibited by the cortex through the feedback inhibition by TRN (Lam and Sherman, 2010) or by the IC through the feedforward inhibition by IC GABAergic cells (Peruzzi et al., 1997; Winer et al., 1996). Therefore, we determined if inhibition of either of these pathways (to disinhibit the MGB) could retrieve the missing cortical responses. Corticothalamic layer 6 cells indirectly inhibit thalamic relay cells by exciting GABAergic neurons in the TRN, a shell-like structure of GABAergic neurons that surrounds the most of dorsolateral part of the thalamus (Crandall et al., 2015; Guo et al., 2017). Therefore, the feedback inhibition of MGB via corticothalamic layer 6-TRN pathway was examined. Injection of the AC of NTSR1-Cre neonates with halorhodopsin-AAV resulted in a successful Cre-dependent expression of eNpHR3.0 receptors in the corticothalamic layer 6 as well as their projections to TRN and MGB indicated by YFP tag (Figure 6B). Photoinhibition of corticothalamic layer 6 cells by illumination with 565 nm light resulted in a significant increase in the probability of population ON cortical responses as indicated by the recovery of the post-stimulus LFP signals from L3/4 compared to the control (Figure 6C and D, Figure 6—source data 1, paired t-test: t(15) = −7.06, **p=3.8×10−6, n = 16 trials from four animals). A control experiment was done by injecting NTSR1-Cre mice with a virus expressing YFP without the inhibitory opsin. The same illumination of the aCTC slice expressing YFP in layer 6 cells (Figure 6F) with a 565 nm light of could not retrieve the population ON cortical responses (Figure 6G and H, Figure 6—source data 2, paired t-test: t(24) = −1.3, p=0.2, n = 25 trials from five animals). To examine the impact of silencing corticothalamic layer 6 neurons using FA imaging, the AC of a separate group of NTSR1-Cre neonates was injected with DREADDs-AAV virus, which resulted in a successful Cre-dependent expression of the inhibitory chemogenetic receptors, hM4Di, as indicated by mCherry tag specifically in corticothalamic layer 6 cells as well as their projections to TRN and MGB (Figure 6—figure supplement 1B). Consistent with the previous data, the chemical inhibition of corticothalamic layer 6 cells expressing hM4Di receptors by their chemical actuator, CNO, significantly increased the frequency of population ON cortical responses as indicated by Δf/f of evoked cortical FA signals compared to the control (Figure 6—figure supplement 1C and D, Figure 6—figure supplement 1—source data 1 and 2, paired test: t(3) = −3.66, p=0.035, n = 4 trials from four animals). The control experiment was done by injecting other NTSR1-Cre mice with a virus expressing mCherry without the DREADDs, and the perfusion of CNO to the aCTC slices taken from these animals could not retrieve the population ON cortical responses (Figure 6—figure supplement 1E, Figure 6—figure supplement 1—source data 3, paired test: t(4) = 0.9, p=0.4, n = 5 trials from five animals).

Figure 6 with 2 supplements see all
The population OFF cortical responses were driven by the feedback inhibition of MGB by corticothalamic layer six via TRN.

(A and E) Cartoon images showing the experimental design of simultaneous IC stimulation, LFP recording, and photoinhibition for aCTC slice with and without eNpHR3.0 receptors, respectively. (B and F) Images of aCTC slice of NTSR1-Cre mouse showing the expression of eNpHR3.0 receptors as indicated by YFP tag and YFP only, respectively, in NTSR1-positive corticothalamic layer 6 cells as well as their projections to TRN and MGB. (C, G, K and O) The time series of the post-stimulus cortical LFP signals from L3/4 following IC stimulation without (top panel) and with 565 nm light (bottom panel). (D) A scatterplot of percent population ON responses pre- and post-light application showing that the percentage of population ON cortical events was higher during the photoinhibition of corticothalamic layer 6 cells. (H) A scatterplot of percent population ON responses pre- and post-light application showing that the illumination of the aCTC slice expressing no inhibitory opsin with 565 nm light could not retrieve the population ON cortical responses. (I) A cartoon image showing the experimental design of simultaneous IC stimulation, LFP recording, and photoinhibition of layer 5 cells (J) Image of aCTC slice of RBP4-Cre mouse showing the expression of eNpHR3.0 receptors as indicated by YFP tag in RBP4-positive layer 5 cells. (L) A scatterplot of percent population ON responses pre- and post-light application showing that the photoinhibition of layer 5 cells could not retrieve the population ON cortical responses. (M) A cartoon image showing the experimental design of simultaneous IC stimulation, LFP recording, and photoinhibition of IC GABAergic cells (N) Image of aCTC slice from GAD2-Cre mouse showing the Cre-dependent expression of halorhodopsin indicated by YFP tag in GABAergic cells of the IC as well as their projections to MGB. (P) A scatterplot of the percentage of cortical population ON responses showing no change after the photoinhibition of IC GABAergic cells by light; Black arrows refer to the occurrence of population OFF cortical responses indicated by the absence of post-stimulus cortical LFP signals from L3/4; Vertical gray lines indicate the onset of IC stimulation; Orange lines indicate the time period of illumination (3 s).

Given that corticothalamic layer 6 cells project to MGB through a direct excitatory synapse, further examination was required to test if the TRN is the main driver of corticothalamic layer six effect. Blocking of the TRN activity by NBQX, the AMPA-R blocker (Sun et al., 2013; Lee et al., 2010a), which was specifically injected into TRN, significantly increased the probability of population ON cortical events indicated by Δf/f of the evoked cortical FA signals (Figure 6—figure supplement 1G and H, Figure 6—figure supplement 1—source data 4 and 5, paired test: t(5) = −6.03, **p=5.2×10−4, n = 8 trials from seven animals). To test the specificity of corticothalamic layer 6-TRN pathway to modulate MGB activity, we performed a control experiment by inhibiting layer 5 corticothalamic cells, which do not have significant projections to the TRN (Crabtree, 2018). The injection of the AC of RBP4-Cre neonatal mice with halorhodopsin virus resulted in a successful Cre-dependent expression of eNpHR3.0 receptors in layer 5 cells as indicated by YFP tag (Figure 6J). In contrast to layer 6 cells, the projections from layer five to MGB in the aCTC slice are sparse, as has been previously described (Llano and Sherman, 2008), so they are not well seen in the low-magnification image (Figure 6J). However, with high magnification, the layer 5 projections to MGB were observed (Figure 6—figure supplement 2). Consistent with the previous data, the photoinhibition of layer 5 cells expressing eNpHR3.0 receptors by illumination with 565 nm light could not retrieve the missing cortical responses, indicated by no recovery of the post-stimulus LFP signals from L3/4 compared to control (no light) (Figure 6K and L, Figure 6—source data 3, paired t-test: t(36) = 0.46, p=0.18, n = 37 trials from four animals), which suggests that the population OFF cortical responses were specifically driven by MGB inhibition through corticothalamic layer 6-TRN pathway. To suppress feedforward inhibition, the IC of neonatal GAD2-Cre mice was injected with halorhodopsin-AAV virus (see Materials and methods) to induce the expression of halorhodopsin specifically in the GABAergic cells of the IC in a Cre-dependent manner. As expected, GABAergic cells of the IC as well as their projections to MGB expressed halorhodopsin indicated by the presence of YFP (Figure 6N). Photoinhibition of GABAergic cells of the IC by illumination with 565 nm light was not able to retrieve the missing cortical responses as indicated by no recovery of the post-stimulus LFP signals recorded from L3/4 (Figure 6O and P, Figure 6—source data 4, Paired t-test: t(6) = −1.03, p=0.34, n = 7 trials from five animals). Accordingly, based on the data shown in Figure 6 and Figure 6—figure supplement 1, we conclude that the population OFF cortical responses were most likely driven by the feedback inhibition of MGB by corticothalamic layer 6 cells via TRN.

Discussion

We observed stochastic population cortical responses in the mouse AC following the presentations of pure tones in vivo or electrical stimulation of the IC in vitro. Population ON responses were associated with synchronized responses among MGB cells, whereas population OFF responses were associated with TRN-mediated inhibition at the level of the MGB, under the control of layer 6 corticothalamic projections. Other inhibitory projections to the MGB from the IC had no impact on the probability of eliciting a population ON response. It is unlikely that population OFF cortical responses were a sign of cortical adaptation, because adaptation responses are generally characterized by gradual decrease of the response amplitude, rather than the all-or-none responses observed here (Chung et al., 2002; Abolafia et al., 2011). We conclude from these findings that layer 6 corticothalamic neurons gate population activity in the AC via their projections to TRN, which desynchronize MGB neurons. Previous work has suggested that corticothalamic axons, via the strong inhibition to thalamocortical cells by way of the TRN (Crandall et al., 2015; Olsen et al., 2012; Paz and Huguenard, 2015; Steriade et al., 1996; Guillery and Harting, 2003; McAlonan et al., 2008), control thalamocortical information flow (Destexhe, 2000; Yu et al., 2004). In the current study, the gating mechanism appears to involve TRN-based desynchronization of thalamocortical neurons rather than diminishing the overall thalamic response, which is consistent with previous findings showing that the TRN can desynchronize thalamocortical cells via multiple mechanisms (Pita-Almenar et al., 2014). This explanation is also consistent with our finding that direct electrical stimulation of the thalamus did not lead to stochastic AC response because direct electrical stimulation is likely to elicit highly synchronous responses among MGB neurons. It is also possible that silencing layer six corticothalamic neurons led to changes in likelihood of activating the AC via intracortical connections of NTSR1-positive neurons. Future work involving selective silencing of corticoreticular terminals vs. corticocortical terminals will be needed to distinguish between these possibilities. It will also be important to determine if the mechanisms observed in the current slice data are also seen in adult animals. Juveniles were used to maximize slice connectivity, but now that specific hypotheses may be proposed, elements of the proposed thalamocortical gating mechanisms can be examined in adult preparations. We also note that population OFF cortical responses in the current study were never initiated by the first stimulation of the IC. Rather they were always elicited by the subsequent IC stimulation, which suggests that population OFF cortical responses occurred only after evoked cortical activity that could change the internal dynamics of the cortical cells. Moreover, the strong association between the high probability of the OFF cortical responses and the low level of stimulation suggests that cortico-reticulothalamic control of thalamocortical transmission is most likely to be effective when signal-to-noise ratio is low. This finding is consistent with the notion that top-down modulation is mostly required for the attentional modulation of weak signals and that highly salient signals rely on bottom-up mechanisms to activate perceptual representations (reviewed in Asilador and Llano, 2020). Classical models have viewed sensory processing stations, including the thalamus and cortex, as a series of feedforward, hierarchically organized filters whereby combinations of receptive fields produce increasingly selective feature detectors, culminating in uniquely selective neurons (Riesenhuber and Poggio, 1999; Vidyasagar and Eysel, 2015). This type of organization implies that moment-to-moment perception is a reflection of detailed streams of information coursing through ascending sensory systems, to be consciously perceived when those streams engage highly selective cells in the cortex. An alternative view is that ascending information is used to create and modify a bank of sensory representations that are recruited depending on behavioral needs, and that conscious perception reflects activation of these pre-wired circuits (Ringach, 2009; Llinás, 1994; Mishkin, 1982). Thus, conscious perception may involve the release of stereotyped patterns of cortical activity. One potential role of thalamocortical transmission in this model is to select cortical ensembles rather than impress sensory information upon them. Our data suggest that at the thalamocortical level, sensory inputs activate stereotyped patterns of population cortical activity. Future work in awake preparations will help to establish whether population ON responses correspond to perception of sensory stimuli.

Critical for any such mechanism of control of cortical ensembles is a means by which those ensembles are selected. The current data suggest that the TRN, under the control of layer 6 projections, gates populations of thalamic neurons by desynchronizing their responses, decreasing the likelihood of engendering a population cortical response. A mechanism of TRN-based modification of thalamic synchrony to activate the cortex has been proposed previously (Saalmann and Kastner, 2011) and is consistent with the finding that synchronized populations of thalamic neurons are required to optimally activate the cortex (Bruno and Sakmann, 2006) and that the TRN is at the heart of a prefrontal cortex-based mechanism to shape cortical activation under changing cognitive demands (Zikopoulos and Barbas, 2006; Wimmer et al., 2015; Yingling and Skinner, 1975). Further, the TRN receives inputs from basal forebrain, amygdala and non-reciprocally linked regions of the thalamus (Kimura et al., 2007; Lam and Sherman, 2005; Lam and Sherman, 2015; Crabtree et al., 1998; Crabtree and Isaac, 2002; Lee et al., 2010b; Desîlets-Roy et al., 2002; Zikopoulos and Barbas, 2012; Aizenberg et al., 2019), forming an assortment of inputs to potentially modulate TRN, and ultimately select cortical circuits for activation. Given the putative role of the TRN in the selection of thalamic, and therefore cortical circuits during sensory perception, one would predict that disruption of TRN activity could lead to uncontrolled release of patterns of cortical activity. Consistent with this idea, ample evidence has accumulated to suggest that schizophrenia, a disease characterized by the presence of auditory hallucinations, involves disruption of the TRN (Pratt and Morris, 2015; Ferrarelli and Tononi, 2011; Young and Wimmer, 2017; Bencherif et al., 2012; Light and Braff, 1999; Patterson et al., 2008; Ferrarelli et al., 2007; Ferrarelli et al., 2010; Wamsley et al., 2012; Ferrarelli and Tononi, 2017).

Conclusion

Here, we describe a unique stimulus-evoked population cortical all-or-none response, which suggests that thalamus recruits cortical ensembles of pre-wired sensory representations upon external stimulation in conjunction with internal cortical dynamics of corticothalamic neurons. These data also suggest that corticothalamic modulators control the temporal coordination between the thalamic cells to gate the activation of the intracortical network. It will be important in future studies to more fully understand how other regulators of the TRN, such as the basal forebrain, prefrontal cortex and amygdala, influence the selection of cortical circuits during active behavior.

Materials and methods

Animals

C57BL/6J (Jackson Laboratory, stock # 000664), C57BL/6J-Tg (Thy1-GCaMP6s) GP4.3Dkim/J a.k.a. GCaMP6s mice (Jackson Laboratory, stock # 024275), BALB/c (Jackson Laboratory, stock # 000651), Gad2-IRES-Cre (Jackson Laboratory, stock # 010802), NTSR1-Cre (MMRRC, 017266-UCD) and RBP4-Cre (MMRRC, 031125-UCD). Mice of both sexes were used. All applicable guidelines for the care and use of animals were followed. All surgical procedures were approved by the Institutional Animal Care and Use Committee (IACUC). Animals were housed in animal care facilities approved by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC).

In vivo imaging

Request a detailed protocol

The detailed procedures have been described before (Yudintsev et al., 2020). In brief, adult GCaMP6s mice were used for transcranial in vivo imaging of evoked calcium signals from the left AC. For each experiment, the mouse was anesthetized with a mixture of ketamine and xylazine (100 mg/kg and 3 mg/kg, respectively) delivered intraperitoneally. The animal’s body temperature was maintained within the range of 35.5°C and 37°C using a DC temperature controller (FHC, ME, USA). Mid-sagittal and mid-lateral incisions were made to expose the dorsal and lateral aspects of the skull along with the temporalis muscle. The temporalis muscle was separated from the skull to expose the ventral parts of the underlying AC. The site was cleaned with sterile saline, and the surface of the skull was thinned by a surgical drill. A small amount of dental cement (3M ESPE KETAC) was mixed to a medium level of viscosity and added to the head of the bolt just enough to cover it. A head-bolt was bonded to the top of the skull, and the dental cement was allowed to set. An Imager 3001 integrated data acquisition and analysis system (Optical Imaging Ltd., Israel) was used to image the cortical responses to sound in mice. A macroscope consisting of 85 mm f/1.4 and 50 mm f/1.2 Nikon lenses was mounted to an Adimec 1000 m high-end CCD camera (7.4 × 7.4 µm pixel size, 1004 × 1004 resolution), and centered above the left AC, focused approximately 0.5 mm below the surface of the exposed skull. Acoustic stimuli were generated using a TDT system three with an RP 2.1 enhanced real-time processor and delivered via an ES1 free field electrostatic speaker (Tucker-Davis Technologies, FL, USA), located approximately 8 cm away from the contralateral ear. All imaging experiments were conducted in a sound-proof chamber and images were obtained at 10 frames per second. Each trial of sound presentation was composed of two conditions (10 s each); Condition 0 (C0), where there is no sound and condition 1 (C1), where there is a 500 ms sound presentation that comes after 5 s from the onset of the C1. For imaging, the blue light exposure was only on during the 10 s of each trial, and there was a 5 s interval between the two conditions during which the blue light was off. This intermittent schedule of the blue light exposure was done to lower the likelihood of photobleaching. 500 ms pure tones of 5 kHz, at 37, 44, 50, 55, 60, 65, 70, 75, or 80 dB SPL were used for acoustic stimulation. The response window was set as one second starting from the 5th second for C0 or from the onset of sound for C1. The Δf/f of the response window was computed as the difference between Δf/f0 (C0) and Δf/f1 (C1) using a custom MATLAB code (Figure 1—source code 1).

Virus injection

Request a detailed protocol

To modulate specific cell types, Cre recombinase-expressing mice were used to provide an expression of opto- or chemo-genetic probes in those specific cells 11 days after viral injection at P4. The detailed procedures have been described before (Huynh et al., 2020). For all neonates, cryoanesthesia was induced after five to ten minutes. A toe pinch was done to confirm that the mice were fully anesthetized. A small animal stereotaxic instrument (David Kopf Instruments, Tujunga, CA) was used with a universal syringe holder (David Kopf Instruments, Tujunga, CA) and standard ear bars with rubber tips (Stoelting, Wood Dale, IL) were used to stabilize the head. The adaptor stage was cooled by adding ethanol and dry ice to an attached well. A temperature label (RLC-60-26/56, Omega, Norwalk, CT) was attached to the stage to provide the temperature of the stage during cooling. The temperature was kept above 2°C to prevent hypothermia or cold-induced skin damage of the neonatal mice and below 8°C to sustain cryoanesthesia. Glass micropipettes (3.5-inches, World Precision Instruments, Sarasota, FL) were pulled using a micropipette puller (P-97, Sutter Instruments, Novato, CA) and broken back to a tip diameter between 35 and 50 μm.

The micropipette was filled with mineral oil (Thermo Fisher Scientific Inc, Waltham, MA) and attached to a pressure injector (Nanoliter 2010, World Precision Instruments, Sarasota, FL) connected to a pump controller (Micro4 Controller, World Precision Instruments, Sarasota, FL). The AC of NTSR1-Cre (Olsen et al., 2012; Bortone et al., 2014; Kim et al., 2014; Mease et al., 2014) or RBP4-Cre (Jeong et al., 2016; Grant et al., 2016) neonates was injected with eNpHR3.0 AAV1 (AAV-EF1a-DIO-eNpHR3.0-YFP with titer equal to 4.7–5.7 × 1012, referred to hereafter as halorhodopsin-AAV) constructs from UNC Vector Core (Chapel Hill, NC), AAV1-Ef1a-DIO EYFP (control for halorhodopsin), Gi-coupled hM4Di DREADDs AAV8 (AAV8-DIO-hSyn-hM4Di-mCherry, referred to here as DREADDs-AAV), AAV8-hSyn-DIO-mCherry (control for DREADDs), or AAV9-FLEX-EGFP (for the histology done on RBP4-Cre mouse) constructs from Addgene (Cambridge, MA). The micropipette carrying the viral particles was first located above the AC in the left hemisphere at 1.5 mm anterior to lambda and just at the edge of the skull’s flat horizon. The tip was lowered to 1.2 mm from the brain surface and was then pulled back to 1.0 mm for the first injection where 200 nL of halorhodopsin-AAV or DREADDs-AAV was injected at 200 nL/min. After the injection was finished, the micropipette was left in the brain for 1 min before removing to allow the injectate to settle into the brain. Following the first injection, the tip was pulled back stepwise in 0.1 mm increments, and 200 nL of the injectate was injected at every step until the tip reached 0.3 mm from the surface. In total, 1600 nL of AAV was injected into the AC. The incision was sutured using 5/0 thread size, nylon sutures (CP Medical, Norcross, GA). To target the GABAergic cells of the inferior colliculus (IC), the IC of GAD2-Cre (Taniguchi et al., 2011; Lammel et al., 2015; Villalobos et al., 2018) neonatal mice was injected with halorhodopsin-AAV following the same procedures shown above, but the micropipette loaded by halorhodopsin-AAV was located over the IC at the left hemisphere at 2.0 mm posterior to lambda and 1.0 mm laterally from the midline. The neonates were transferred back onto a warming pad to recover. After 5–7 min, their skin color was returned to normal, and they started moving. After recovery, all neonates were returned to their nest with the parents.

Brain slicing

Request a detailed protocol

For all in vitro experiments, 15- to 18-day-old mice were initially anesthetized with ketamine (100 mg/ kg) and xylazine (3 mg/kg) intraperitoneally and transcardially perfused with chilled (4°C) sucrose-based slicing solution containing the following (in mM): 234 sucrose, 11 glucose, 26 NaHCO3, 2.5 KCl, 1.25 NaH2PO4, 10 MgCl2, 0.5 CaCl2. After the brain was removed from the skull, it was cut to obtain auditory colliculo-thalamocortical brain slice (aCTC) as shown (Figure 2—figure supplement 1) and as described before (Llano et al., 2014; Slater et al., 2015). 600 μm thick horizontal brain slices were obtained to retain the connectivity between IC, MGB, TRN and AC. All slices were incubated for 30 min at 33°C in a solution composed of (in mM: 26 NaHCO3, 2.5 KCl, 10 glucose, 126 NaCl, 1.25 NaH2PO4, 3 MgCl2, and 1 CaCl2). After incubation, all slices were transferred to a perfusion chamber coupled to an upright Olympus BX51 microscope, perfused with artificial cerebrospinal fluid (ACSF) containing (in mM) 26 NaHCO3, 2.5 KCl, 10 glucose, 126 NaCl, 1.25 NaH2PO4, 2 MgCl2, and 2 CaCl2. Another set of experiments was done in a different laboratory to exclude any experimental factors related to our laboratory environment, chemicals, or anesthesia. As reported previously (Krause et al., 2014), following full anesthesia by isoflurane, a C57BL/6J mouse was immediately decapitated without cardiac perfusion, the animal’s brain was extracted and immersed in cutting artificial CSF [cACSF; composed of (in mM) 111 NaCl, 35 NaHCO3, 20 HEPES, 1.8 KCl, 1.05 CaCl2, 2.8 MgSO4, 1.2 KH2PO4, and 10 glucose] at 0–4°C. Slices were maintained in cACSF at 24°C for >1 hr before transfer to the recording chamber, which was perfused at 3–6 ml/min with ACSF [composed of (in mM) 111 NaCl, 35 NaHCO3, 20 HEPES, 1.8 KCl, 2.1 CaCl2, 1.4 MgSO4, 1.2 KH2PO4, and 10 glucose]. All of the solutions were bubbled with 95% oxygen/5% carbon dioxide and all experiments were done at room temperature.

Electrical stimulation

Request a detailed protocol

All the electrical stimulation protocols in the IC evoked similar cortical response patterns. One second electrical train pulses of (250 µA, 40 Hz, 1 ms pulse width) to IC was the main stimulating protocol as described before (Ibrahim et al., 2017). However, one second durations of electrical stimulation were not suitable for electrophysiology experiments because responses were intermingled with the stimulus artifact. Therefore, for whole-cell and LFP recording, the stimulation of IC was done with a single 3 ms pulse (300–500 μΑ) or 100 ms long trains of pulses (250 µA, 40 Hz, 1 ms pulse width) were used. The electrical stimulation was done using a concentric bipolar electrode (Cat#30201, FHC) every 10–20 s. The parameters of the electrical pulses were adjusted by a B and K precision wave generator (model # 4063) and World Precision Instruments stimulation isolator (A-360). In another stimulation set up (Our collaborator’s laboratory), a biphasic current pulse (200 µA, 5 ms; STG4002 stimulator, Multichannel Systems, Reutlingen, Germany) was delivered to IC at 0.05 Hz using bipolar tungsten electrodes (100 KΩ, FHC Inc, Bowdoin, ME).

Imaging

Request a detailed protocol

For calcium imaging in vitro, slices from GCaMP6s mice or wild-type mice loaded with CAL-520, AM (Abcam, ab171868) calcium dye were used. For CAL-520, AM calcium dye loading, the aCTC slices were incubated in 48 µl of DMSO dye solution +2 µl of Pluronic F-127 (Cat# P6866, Invitrogen) in 2.5 ml of the incubating solution at 35–36°C for 25–28 min according to MacLean and Yuste, 2009; Yuste et al., 2011. The slices then were incubated in normal incubating solution (shown above) for 30 min to wash the extra dye. Imaging was done under ACSF perfusion as described before. Depending on the experiment, the evoked calcium or flavoprotein autofluorescence (FA) signals following IC stimulation (Ibrahim et al., 2017; Shibuki et al., 2003; Shibuki et al., 2009; Husson et al., 2007) were measured using a stable DC fluorescence illuminator (Prior Lumen 200) and a U-M49002Xl E-GFP Olympus filter cube set (excitation: 470–490 nm, dichroic 505 nm, emission 515 nm long pass, 100 ms exposure time for FA and 5 ms for calcium signals). All data were collected using Retiga EXi camera at a frame rate of 4 Hz for FA and 10 Hz. The time trace of the FA or calcium population signals were obtained by placing regions of interest (ROI) over brain regions (IC, MGB, TRN or AC). The collected time traces were used to compute the Δf/f. For the calcium signals obtained from cortical or thalamic cells, the ROIs were manually made around the cell body. A strong correlation between voltage and calcium signals validated the calcium signals obtained from thalamic cells containing CAL-520 dye after simultaneous whole-cell recording and calcium imaging following the injection of a positive current to the cell (Figure 5—figure supplement 1). The average background value was calculated by drawing an ellipse with radii 2.25 times that of the original ROI and subtracting all cell ROI from that ellipse to eliminate overlap. Finally, the Δf/f was computed after 40% of the background value was subtracted from the cell's average value for neuropil correction. Δf/f was computed using the average signal of a 2 s period before the stimulus onset. As illustrated in Figure 2—figure supplement 2, the ON-cortical responses were determined following two criteria. Each signal required a rising and falling phase as well as a z-score greater than 3.

Pharmacological intervention

Request a detailed protocol

To inhibit GΑΒΑΑ receptors globally, GABAA-receptor antagonist (Yaron-Jakoubovitch et al., 2013), gabazine (Cat# 1262, Tocris) was added to the bath ACSF solution 200 nM, which is near the synaptic IC50 for this compound (Lindquist et al., 2005). To specifically inhibit the GABAA receptors in either MGB or AC, a continuous flow of gabazine (400 nM) was injected into MGB or AC in a counterbalanced manner through a glass pipette (broken tip, 35 µm) which was connected to a picospritzer (Toohey Company, New Jersey, USA). The pipette was filled by a solution composed of 1 ml ACSF +10 µL Alexa Fluor 594 hydrazide, sodium salt dye (Cat#A10438, Invitrogen) to visualize the flow of the solution and to ensure that it was restricted to the site of injection (Figure 4—figure supplement 1). The injection was done under 10 psi pressure for 5 min and continuously during imaging. As reported before, to block TRN activity (Sun et al., 2013; Lee et al., 2010a), the AMPA receptor blocker NBQX (20 μM, Cat# 0373, Tocris) was injected into TRN of the aCTC slice following the same described procedures. The chemical inhibition of corticothalamic layer 6 cells was conducted by bath perfusion of clozapine-n-oxide (CNO, 5 µM, Cat# 4936, Tocris), the chemical actuator of the chemogenetic probe, hM4Di (Roth, 2016) that was solely expressed in corticothalamic layer 6 of NTSR1-Cre mouse after viral injection. CNO was also perfused to the imaging chamber containing a slice with a layer 6 expressing m-cherry only as a control for DREADDs.

Electrophysiology and photoinhibition

Request a detailed protocol

Whole-cell recording of cortical layer 4, TRN, or MGB cells was performed using a visualized slice setup outfitted with infrared-differential interference contrast optics. Recording pipettes were pulled from borosilicate glass capillary tubes and had tip resistances of 2–5 MΩ when filled with potassium gluconate based intracellular solution (in mM: 117 K-gluconate, 13 KCl, 1.0 MgCl2, 0.07 CaCl2, 0.1 ethyleneglycol-bis(2-aminoethylether)- N,N,N′,N′-tetra acetic acid, 10.0 4-(2-hydroxyethyl)−1- piperazineethanesulfonic acid, 2.0 Na-ATP, 0.4 Na-GTP, and 0.5% biocytin, pH 7.3, 290 mOsm) for current-clamp mode. Voltage was clamped at −60 mV or +10 mV to measure either the excitatory or inhibitory currents, respectively, using cesium-based intracellular solution (in mM: 117.0 CsOH, 117.0 gluconic acid, 11.0 CsCl, 1.0 MgCl2*6H2O, 0.07 CaCl2, 11.0 EGTA, 10.0 HEPES, pH 7.3, 290 mOsm). Local field potential (LFP) recordings were performed using glass pipette with a broken tip 5–10 µm. LFP signals were filtered offline using Clampfit 10.7 software under Gaussian low pass frequency at 300 Hz as well as a notch filter at 60 Hz. A Multiclamp 700B amplifier and pClamp software (Molecular Devices) were used for data acquisition (20 kHz sampling). To analyze the distribution of the membrane potential of layer 4 cells during the UP state, a time period starting from the stimulation offset and including a similar time of both the cortical upstate event and the basal membrane potential was used. The sampling rate was reduced from 20 kHz to 2 kHz, and spikes and stimulus artifacts were excluded from the analysis.

For photoinhibition, the halorhodopsin eNpHR3.0 probe expressed selectively in either corticothalamic layer 6, layer 5, or IC-GABAergic cells, was activated by illuminating a yellow light (565 nm) obtained from DC fluorescence illuminator (Prior Lumen 200) and Olympus filter cube (U-MF2, Olympus, Japan). The same light was used to illuminate the brain slice with a layer six expressing eYFP only as a control for halorhodopsin. The light was set to illuminate the whole field of the LFP recording chamber using a 4X objective for three seconds extending from one second pre-stimulus and two seconds after the onset of IC stimulation. Based on the initial results related to the peri-stimulus dynamics of corticothalamic layer 6 cells and the post-stimulus cortical activity, 3 s of illumination was chosen to cover the time period one second before the onset of the stimulus as well as the post-stimulus period.

Histology and confocal imaging

Request a detailed protocol

After collecting the aCTC slice from RBP4-Cre mouse injected with AAV9-FLEX-EGFP into the AC, the aCTC slice was placed in a 4% paraformaldehyde (PFA) solution for fixation. After one day, the aCTC slice was removed from the PFA and moved to graded sucrose solution (10, 20, and 30%). The aCTC slice was sectioned using a Leica cryostat as 50 μm sections. The sections were imaged using Leica SP8 confocal microscope (excitation: 488 nm and emissions: 515–550 nm).

Imaging analysis and statistics

Request a detailed protocol

Using customized MATLAB codes, all the pseudocolor images (jet colormap) were produced showing the tonotopic map of AC in vivo and the activated brain regions in the aCTC slice. Origin-Pro 2017 software was used to run the statistical tests and generate the graphs. The normality of the distributions of the data was examined using a Kolmogorov-Smirnov test. Accordingly, the suitable parametric (paired t-test or RM-One Way ANOVA test followed by Bonferroni post hoc test) or non-parametric tests (Paired Wilcoxon signed-rank test or Kruskal-Wallis test followed by Dunn's Test) were used. Differences were deemed significant when p-value<0.05. Power analyses were run in G*Power (http://www.gpower.hhu.de). Using an effect size of 50% and alpha <0.05, sample sizes in all experiments provided power to detect significant differences of at least 80%. The animals were randomly allocated in each experimental group.

Work art

Request a detailed protocol

All figures were designed and made using Adobe Illustrator (Adobe, San Jose, CA). To keep working within Adobe environment to avoid losing the resolution of the figures, Adobe Photoshop (Adobe, San Jose, CA) was used to crop the borders of some images to save space, draw scale bars, and adjust brightness and gamma balance of grayscale images showing the electrophysiological traces. All manipulations in brightness/contrast/gamma were uniform across the entire image.

Data availability

All data generated or analyzed during this study are included in the manuscript and supporting files. Source data files have been provided for Figures 1, Figure 2, Figure 2-figure supplement 2, Figure 2-figure supplement 6, Figure 3, Figure 4, Figure 5, Figure 6, Figure 6-figure supplement 1. The datasets are available at Dryad under a DOI (https://doi.org/10.5061/dryad.qrfj6q5c4).

The following data sets were generated
    1. Ibrahim BA
    2. Murphy CA
    3. Yudintsev G
    4. Shinagawa Y
    5. Banks MI
    6. Llano DA
    (2020) Dryad Digital Repository
    Corticothalamic gating of population auditory thalamocortical transmission in mouse.
    https://doi.org/10.5061/dryad.qrfj6q5c4

References

  1. Book
    1. Llinás UR
    (1994)
    Perception as an Oneiric-Like State Modulated by the Senses, in Workshop, Large-Scale Neuronal Theories of the Brain
    MIT Press.
    1. Mishkin M
    (1982) A memory system in the monkey
    Philosophical Transactions of the Royal Society B 298:85–95.
    https://doi.org/10.1098/rstb.1982.0074
  2. Book
    1. Shibuki HR
    2. Tohmi M
    3. Takahashi K
    4. Kitaura H
    5. Kubota Y
    (2009)
    Flavoprotein Fluorescence Imaging of Experience-Dependent Cortical Plasticity in Rodents
    In: ShibukiK H. R, editors. In Vivo Optical Imaging of Brain Function. CRC Press/Taylor & Francis. pp. 1–7.

Decision letter

  1. Andrew J King
    Senior and Reviewing Editor; University of Oxford, United Kingdom
  2. Daniel B Polley
    Reviewer; Massachusetts Eye and Ear Infirmary, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This study employs an impressive combination of in vivo and in vitro methods to investigate the neural circuitry that determines whether neurons in auditory cortex respond or not to relatively quiet sounds. The results suggest that layer 6 corticothalamic feedback via the thalamic reticular nucleus is responsible for gating cortical population responses, proving new insight into the role of descending projections from the auditory cortex to the thalamus.

Decision letter after peer review:

Thank you for submitting your article "Corticothalamic gating of population auditory thalamocortical transmission in mouse" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by Andrew King as the Senior Editor and Reviewing Editor. The following individual involved in review of your submission has agreed to reveal their identity: Daniel B Polley (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, but as described below that additional experiments are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)

Summary:

This is an impressive paper, which utilizes an array of state-of-the-art tools to dissect the contribution of cortical feedback to the thalamic reticular nucleus in the omission of responses to sound in the auditory cortex. This is an important and surprising finding, which provides new insight into the role of corticothalamic circuits. These findings are based on a combination of in vivo and in vitro population recordings and make excellent use of a powerful slice preparation that allows the connectivity of the midbrain, thalamus and cortex to be explored. Although the reviewers were enthusiastic about much of the study, they thought that the findings were not well presented and disagreed with some of your interpretations of the data. In particular, they were unanimous in their view that the sections dealing with gamma oscillations, thalamic synchrony and deep learning are weak and incomplete, and should be removed from the paper. They were also concerned that critical control groups (the use of a control fluorophore in the case of the halorhodopsin experiments and a CNO only group for the DREADDS study) are not reported. This is particularly important where such large injection volumes were used in very young mice. Consequently, it was felt that additional data as well as re-analysis and re-writing would be needed before this paper could be considered acceptable for publication.

Reviewer #1:

This is a nice paper that proposes a mechanism that allows for gating of population cortical activity. The paper uses a combination of in vivo and in vitro population recordings that are backed up with some excellent slice physiology from a courageous colliculo-thalamocortical slice preparation.

Line 69. Why must these hierarchical filters be linear?

Line 73. This formed complex hallucinations argument isn't really clear. Are you just arguing that a hierarchical linear filter model cannot explain anything that causes elevated activity in primary sensory cortices, that isn't present in subcortical stations?

Line 85. I'm not sure I agree with the statement that the thalamus is strikingly similar across modalities. The two references that are cited only discuss canonical cortical circuits.

Line 91. The statement "all-or-none" to describe the population response seems inadequate, given the variability in "ON" responses in Figure 1D.

Figure 1. Where do these ON/OFF responses fall within the trial presentations – are there any time dependence that could reflect photobleaching? There is also a large fraction of OFF responses that seem to be driven below baseline (in Figure 1D) – is this an artifact of the dF/F computation? On that note, the methods do not explain how you computed dF/F (there are different ways that one can do this).

Line 105. 37dB is quite a specific intensity – why did you choose this?

Line 108 and Line 110. You say "to 10 presentations of the same tone", and "across the 40 trials of the same sound presentation". It's not quite clear what is meant.

Line 123. Activating the IC and observing AC response is due to di-synaptic circuitry. Therefore, does the AC "failure rate" change with the intensity of the stimulation? Can you stimulate hard enough to overcome the thalamic inhibition that you claim mediates the OFF responses?

Figure 2E. Why is the dF/F in the MGB orders of magnitude lower than IC or ACtx?

Line 139. It's not immediately clear to me how are you getting single-cell resolution using this one-photon preparation? The images shown throughout are "wide field" population images. Are you able to show any cellular examples? Are you using some kind of NMF algorithm to achieve cellular resolution? How are you controlling for issues like neuropil contamination (that are ubiquitous in 2P data).

Line 146. On the issue of up states vs down states. If you histogram the membrane potential, is it bimodal to reflect the different states?

Line 155-160. This isn't particularly well described. What exactly is it being modeled here? It seems like you are performing a binary classification, e.g. using either MGB activity or striatum activity to predict whether the corresponding cortical response was ON or OFF? Why does this imply that network-level activity may determine whether an on or off response occurred? Why is the striatum region irrelevant (i.e. it is still (indirectly) connected the cortex via the thalamus, no?). Why the need to use such an advanced machine learning method? Did simpler binary classification tools fail? Does this method also give the same result when used on the GCaMP or whole cell recordings?

Figure 3B. Why are there are more OFF responses that appear in the wash condition? Can you add a third variable to Figure 3C to show this – shouldn't the wash be statistically indistinguishable from control?

Line 195. This is actually a question regarding choices throughout the paper – what dictated the choice of either flavoprotein or calcium? The deep learning results referenced here were established with flavoprotein, but these follow-up experiments use calcium.

Figure 4B. It's not quite clear what is going in this figure. Would some kind of population signal help? Or an alternative way to plot the data? Do these individual traces all have well-defined peak latencies (it's hard to tell from the figure). You're showing the variance of the peak latencies – but couldn't you just compare the peak latencies directly? Does it give the same result? Also, you talk about three categories in the text, but are only showing two categories in the figure. A latency analysis is also a little bit difficult to interpret with calcium data (both from the calcium timescale and the acquisition frame rate) – it might be a good idea to either discuss these limitations, since you are arguing for synchronicity based on a small reduction in variance within this data.

Line 195. "Calcium signals of MGB cells were imaged following the stimulation of the IC during ON vs OFF cortical responses" – is it not the IC stimulation that is causing the ON or OFF cortical responses?

Line 373 – "The detailed procedures have been described before (77)". This seems like it may be the wrong citation? (77) is a theoretical conference paper with no in vivo imaging details.

Line 220 – Is there a particular reason to use neonates for the experiments in this section?

Figure S7. I'm slightly confused by the MGB of the RBP4-Cre mouse – it looks like there is no TRN labelling but also no MGB labelling?

Line 270. It's not possible to understand what you did here without jumping to the methods.

Line 274. Is there an intuitive reason why a shorter pre-stimulus period (which should contain more oscillatory information) be less predictive?

Line 276. "A classifier?" Can you justify your classifier choice here, compared to the more complicated deep learning approach used for classification earlier in the paper?

Figure S8. Based on the text, the gamma prediction is important to the circuit story. It may be a better idea to include it as a main figure?

Reviewer #2:

There is much to admire in the paper by Ibrahim and colleagues. They are the only lab in the world (to the best of my knowledge) that has worked out the geometry of a brain slab that leaves the synaptic connections intact from the midbrain-thalamus-MGB-TRN-cortex-back to TRN/MGB. Here, they use this slab to great effect by revitalizing the old cortical up/down state phenomenon in the context of all-or-nothing responses in A1 to stimulation of midbrain afferents. They have convinced me that the basic phenomenon is real and that it is interesting. Further, I am convinced that they can reduce the probability of OFF response trials by: (1) blocking GABAA receptors, (2) blocking AMPA receptors in TRN, (3) inactivating Ntsr1-Cre L6 CT neurons. They also have data to show that (4) inactivating feedforward inhibition from the IC or glutamatergic L5 projection neurons does not affect OFF response probability.

While there is a lot of good in this manuscript, there is plenty of bad and ugly as well. In their efforts to weave a complete theory of the phenomenon, they are forced to rely on a few very weak/poorly developed pieces of data. Emphasizing these more tenuous findings (Figure 4, Figure S4, Figure S8), ends up doing them a disservice because it takes away from their more solid observations. Apart from eliminating the dubious findings, the figures and the analysis are poorly organized, confusing and do not convey the process of processing the data from example cases to summary plots that relate to their statistics. They look more like lab meeting figures than manuscript figures.

In the final analysis, I think the work offers a significant conceptual advance based on an innovative technical approach. The manuscript can be substantially improved through eliminating the under-developed analyses, reorganizing the remaining figures and adding some additional analysis. The authors needn't feel compelled to oversell the reader on a complete comprehensive story when some of holes are not yet filled in. It's fine that some of the mechanisms are unknown so long as the shortcomings and caveats are clearly identified.

1. Figure 1 is potentially important but is not too convincing in its current form. One reason that it's important is that the rest of the manuscript is a developmental neuroscience study that is never acknowledges this fact. The intro and discussion with the talk of schizophrenia and hallucinations are too far-ranging for my taste, but particularly in the context that their study was performed in mice that had only been connected to the auditory world for less than 1 week. It would be very easy to write off the phenomenon they describe here as an oddity o development that would not apply to the mature cortex.

For me, this underscores the potential importance of Figure 1, where they demonstrate the phenomenon in vivo using older animals. The biggest impediments in its current form are (i) the age of the animals used for transcranial imaging is never mentioned. I presume they are adult? (ii) The entire analysis is based on a single tone frequency presented a single sound intensity that is very near threshold (5k at 37 dB SPL). This makes me question the generality of the principle described here. Having done this trial-by-trial analysis myself on this type of data, I don't think it really works for strong, suprathreshold stimulus intensities. Is this entire phenomenon restricted to a very narrow range of near-threshold intensities? In that case…is it really a phenomenon or just what happens near minimally effective stimulation levels (which is already very well known). Obviously, the same critique applies to the current levels they use for the rest of the study. It would have been nice to see a more systematic and quantitative examination of the stimulus parameters where this phenomenon is strong versus weak.

2. Imaging from "single cells". Single cell imaging is not possible with epifluorescence illumination in a slice this thick, especially without neuropil removal. At best they are ROIs but even then, what is the value when there is no basis for the argument that their ROIs are independent response units? I suggest eliminating 2G, Figure 4 and Figure S4. The issues with Figure 4 (and S4) go way beyond the cellular resolution issues, they are uninterpretable for a number of reasons.

3. In Figure 2D, I'm not convinced that they are looking at signals from TRN cell bodies as opposed to thalamocortical/corticothalamic axons. The thalamic radiation forms a dense fascicle right at the point of the TRN in this plane so it could be very hard to distinguish between TRN cell bodies and axon signals with the optics they are using. Too bad they didn't express GCaMP in the GAD-Cre mice they use for inhibiting the feedforward inhibitory IC projection neurons.

4. Calcium imaging in the MGB and IC. Looking at other publications of the Thy1-GCaMP6 mouse, I see lots of labeling in pyramidal neurons in neocortex and hippocampus, but no labeling in the MGB or IC. I'm confused about what is being measured in Figure 2E. Maybe there is expression in younger mice or is just very faint?

5. Figure 3E-H (graphics could be greatly improved here) show that EPSCs in MGB are unchanged in cortical ON vs OFF trials but IPSCs are elevated in OFF trials. This is one of the mysteries of their study that they try to address with a bunch of uninterpretable analysis methods (Figure 4 and Figure S4). The conundrum they face is how does enhanced thalamic inhibition create a cortical OFF response without making the MGB excitatory response weaker? Figure 4 and S4 attempt to explain this but are contrived and unconvincing. I don't disagree but I would suggest that they are thinking about it too simplistically. It doesn't matter what the MGB cell bodies are doing, what matters is what they are transmitting to the cortex during OFF trials. It would be helpful if they quantified these data more carefully by looking at the relative timing of the IPSC and EPSC responses. Looking at the study by Reinhold, Lien and Scanziani 2015, the TRN can regulate sustained spiking and thalamocortical synaptic depression in ways that cannot be dismissed just by measuring the excitatory response. They don't have to fully answer this here. They can do the best with whatever straightforward analysis they can manage and then devote some space in the discussion to actually discussing alternative explanations for their experiments instead of rhapsodizing about how these data solve the greatest mysteries of the brain.

6. Figure 3K – It would be important to see the individual data points. The legend says the data are an n=5, reflecting 5 slices from 4 animals. How do you get two of these slices from one animal? Regardless, this is a very small sample to make such a strong conclusion as blocking cortical GABAA receptors does not affect the OFF response probability. Was the AC Gabazine always performed after the MGB wash? The order should be counterbalanced and really I'm not even sure what the justification is for using a t-test on a sample this small. Unclear how the assumptions of the test can really be determined.

7. I suggest putting all three of their halorhodopsin experiments into one figure (IC inhibitory feedforward neurons, L6 CT and L5 projection neurons). I'm also bothered that there isn't a control condition where mice have undergone a perinatal injection to express a control fluorophore. This seems particularly warranted because they injected an enormous volume of virus solution (1600 nL) into very small area. Apart from this control, while I think the L6 inactivation result is very interesting, I strongly disagree with their often repeated interpretation of the data, that "Population OFF responses were associated with TRN-mediated inhibition at the level of the MGB, under the control of layer 6 corticothalamic projections".

Their data do not support this conclusion. Yes, blocking AMPA receptors in TRN reduced the probability and Yes, silencing L6 CT also reduced the probability but it is a logical fallacy to assume those are causally related. Ntsr1 neurons make synaptic contacts on TRN neurons, MGB neurons, local AC inhibitory neurons and local AC excitatory neurons. It is entirely possible that the Ntsr1 neurons are mediating this effect through local connections in the cortex and not through the descending projection. In that regard, it is a bit frustrating that the main point of the Guo 2017 study was the Ntsr1 neurons can strongly bias AC neurons to become less responsive to afferent activity via intracortical connections, yet these findings are not mentioned in this context.

Don't get me wrong, they might be right. But providing that the L6 CTs modify MGB via TRN would require a different type of experiment to isolate the influence of L6 CT axons onto TRN neurons and not the other three types of neurons that it communicates with. Again, good topic for a revamped Discussion section.

8. Figure S3 should be upgraded to a main figure and quantified.

9. Figure S6 is interesting but preliminary. There are only four data points and DREADD experiments are usually considered uninterpretable without a control condition in which CNO is applied with the expression of a control fluorophore (i.e., without the designer receptor) to account for known off-target effects of CNO.

10. The issues with Figure S7 are that Rbp4-Cre does not selectively label "PT-like" L5 cells that project sub-cerebrally. In fact, most of their axons remain within the cortex. To this point, the expression they show in no way resembles the cartoon in Panel A but more problematically does not compare to the expression in the Ntsr1 neurons. As such, it is hard to interpret much from the negative result because it may be that they are hyperpolarizing many fewer neurons overall and very often not the L5 neurons that project to the MGB.

11. Figure S8 is another example of them trying to squeeze too much out of their data. First of all, I was unclear about how they were confident that they patched Ntsr1 neurons and not other L6 cell types. But mainly the analysis methods were a little strange to me. Why not just filter in the Gamma band (or other bands)? Also there is a huge literature showing the phase of ongoing 2-6Hz oscillations gate the response probability of AC neurons (e.g., in Guo 2017). It would have been reasonable calculate the phase and amplitude of the pre-stimulus membrane currents from its analytical signal using the Hilbert transform. In its current form, the quantification in this figure is not convincing at all.

Apart from that, the Pre-stimulus L6 neuronal activity could only accurately predict 58.1 {plus minus} 3.5 SD% ON and OFF cortical responses, which is subtle compared to the theoretical random accuracy 55.1%. Moreover, should the random accuracy be calculated by training and testing the model with shuffled ON and OFF responses?

Reviewer #3:

This is an interesting paper that finds that the cortical feedback to the thalamic reticular contributes to the omission of responses in the cortex. This is an important and surprising finding, and the authors bring a whole arsenal of state-of-the-art tools to dissect the contribution of this pathway. The variety of the methods, that include calcium imaging, flavoprotein imaging, electrical stimulation, whole cell recording, pharmacologic/optogenetic/chemogenetic manipulations, all point towards the same conclusions. The data that are presented support most of the claims of the paper. Most of my concerns reflect the interpretation of the data, the focus of the introduction and discussion, some missing controls and the lack of report of statistics in the main body. I believe that the authors should be able to address these concerns with the data in hand, through additional analysis and re-interpretation.

1. The emphasis in the introduction and discussion on the "classical models" of purely feedforward hierarchical processing seems somewhat forced. Numerous studies, especially in the auditory system, have addressed the importance of feedback in anatomy, auditory processing, learning and behavior. Rather than skipping over this important work, the paper would benefit from the discussion of previous work on the function of feedback in sensory processing. This would furthermore allow the authors to better define what role their work plays in this context.

2. Similarly, the emphasis on the role of this feedback in neuronal oscillations seems to be overstated. In the introduction, the discussion of hallucinations seems to come out of nowhere. In the paper, only one supplementary figure is devoted to analysis of oscillations, whereas in the abstract, introduction and discussion, these is an extensive focus on the gamma oscillations. The gamma activity -based prediction of ON vs OFF responses seems to be weak. The classifier only had an accuracy of 58%, which is not very high, and is only barely above random accuracy of 55. This focus and confusing result detracts from the main message of the paper.

3. The terminology of ON and OFF responses seems to be at odds with the accepted terminology in the auditory field. Typically, in the auditory field, ON responses refer to the activity at the onset of a stimulus, and OFF responses refer to the activity after stimulus offset. OFF responses, where there is no response at all, is confusing. The authors allude to the use of UP and DOWN states in the literature, but there is only indirect evidence that this activity indeed corresponds to those states. Perhaps using "Response" and "Omission"; yes/no; 1/0 or +/- could work.

4. The claim that the thalamus selects co-activated cortical ensembles, referred to in the abstract, last sentence of the introduction as well as the discussion, is not supported by data. Perhaps it is known that the auditory cortex uses partially overlapping ensembles of coactivated neurons to represent auditory stimuli, but that data is not represented in this paper. In fact, the data in this paper hinges on an "all or nothing response," which undermines this idea of coactivated ensembles in the AC. These contrasting ideas deserve reconciliation.

5. Statistics for all claims should be reported in the main text. Some statistics are missing, for example, in Figure 5D, the distribution of the histograms.

6. It is great that the paper starts with results in vivo. However, those are under anesthesia. Are OFF responses also present in awake animals, perhaps influenced by attention or arousal levels? Perhaps worth speculating in the discussion on the functional importance of this corticothalamic gating in awake behaving animals.

7. The authors use a deep learning algorithm to establish the predictive power of the response, training on MGB data vs irrelevant striatum. How would the algorithm work for IC data? Furthermore, it is unclear whether deep learning is required for this analysis. Would more simple statistical measures, such as correlation or Granger causality measures work here as well? Can the predictive ability of layer 6 activity be improved with just using pre-stimulus layer 6 firing rate, as opposed to LFP power spectra?

8. The authors observed cortical IPSCs during ON responses but not OFF responses. The importance of this finding would be better demonstrated by comparing it with cortical EPSCs. Presumably, ON responses are characterized by large EPSPs and medium IPSCs, whereas OFF responses are characterized by no EPSCs or IPSCs. That way, it's clearer that OFF responses are not simply caused by increase in IPSC magnitude. It could then be contrasted that with the increase in IPSC magnitude observed in MGB, as they show in figure 3G. The importance of figure 3E/3F is unclear. It would be great to explain why IPSCs are observed in cortical ON responses and not OFF responses.

9. How well do gabazine and NBQX stay localized to the area of injection? Is there any chance of diffusion of the agent confounding results? How was localization confirmed?

10. What are the controls for chemogenetics?

11. In Figure 4, when performing calcium imaging in MGB, which trials had cortical ON vs OFF responses? Was that determined through simultaneous LFP recording in AC (diagram in figure 4A)?

12. How well can latency of response and response timing variability (Figure 4) be assessed using GCaMP6s, which has a response time scale on the order of seconds? Is the variability in response timing really on the order of hundreds of milliseconds (Figure 4D)? Perhaps this would be better evaluated using whole cell recording or electrophysiology.

13. Is it MGB synchrony, or MGB inhibition, that leads to cortical ON vs OFF responses? The importance of MGB synchrony, or at least its interaction with TRN-MGB inhibition, is unclear. Furthermore, the synchrony analysis should be conducted with more appropriate statistical tools, including correlation measures. It remains unclear how a desynchronized MGB leads to an OFF response (aka no firing) in the cortex. Do layer 4 cortical cells not reach firing threshold when the inputs are asynchronous? Or do cortical neurons not receive inputs from MGB when the neurons fire asynchronously (due to further gating by third party cells)?

14. Is the difference in power at ~36Hz significant when controlling for multiple comparisons across 11 different LFP power spectra frequencies? Perhaps the authors could try running a 2-way ANOVA with frequency and ON/OFF status as factors.

15. I recommend tightening the writing style throughout: e.g. in abstract: "Here, we elucidate the mechanism for gating of population activity." Some paragraphs are multiple pages long. There are run-on sentences, and the authors use passive tense extensively. I would recommend splitting up the paragraphs, so that each paragraph contains one result/analysis, as well as two sentences for motivation and conclusion.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Corticothalamic gating of population auditory thalamocortical transmission in mouse" for further consideration by eLife. Your revised article has been evaluated by Andrew King as the Senior Editor and Reviewing Editor.

The authors are to be commended for carrying out a number of new experiments to provide essential control data and to provide additional information that has strengthened the study. The majority of the other concerns raised by the reviewers have been dealt with adequately, but there are a few remaining issues that need to be addressed.

Both the in vivo and slice data now clearly illustrate the dependence of the cortical responses on the stimulus magnitude, with the probability of an OFF response occurring declining with increasing stimulus strength. This reinforces one of the points made by reviewer 2, namely that these effects appear be present only for near minimally effective stimulation levels. That raises the question of what function they serve and how these responses contribute to our understanding of cortical processing under more natural conditions (including in awake animals at higher sound levels). These are important issues that need to be addressed in the Discussion.

The readability of the text would benefit further from greater use of paragraphs (as pointed out by reviewer 3). For example, one paragraph runs from lines 304-380. This does include some deleted text, but is still far too long.

Line 364: 5J should be 6J.

Line 423: "exposure to pure tones" might be construed as a form of passive acoustic environment. "Presentation of pure tones" would be less ambiguous.

Figure 1 legend: there are several full stops missing.

Figure 2 – Supplementary Figure 3: the legend refers to red and green boxes, but there are none in the figure (at least in the composite pdf of the manuscript).

Line 943: the sentence ends "showing".

Figure 6 – Supplementary Figure 1G: all traces are black, whereas the legend claims that some are brown or green.

https://doi.org/10.7554/eLife.56645.sa1

Author response

Reviewer #1:

This is a nice paper that proposes a mechanism that allows for gating of population cortical activity. The paper uses a combination of in vivo and in vitro population recordings that are backed up with some excellent slice physiology from a courageous colliculo-thalamocortical slice preparation.

Line 69. Why must these hierarchical filters be linear?

We apologize for the lack of clarity. Indeed, we do not believe that hierarchical filters need to be linear. However, we argue that classical models of sensory processing have relied on combinations of linear filters that are arranged hierarchically. In the current case, we in fact find a strongly nonlinear response in the cortex in response to midbrain stimulation (revised Figure 3). We have added the qualifier “classical” to our statement about traditional views of sensory processing.

Line 73. This formed complex hallucinations argument isn't really clear. Are you just arguing that a hierarchical linear filter model cannot explain anything that causes elevated activity in primary sensory cortices, that isn't present in subcortical stations?

We argue that classical models of sensory processing cannot explain the presence of formed patterns of cortical activity in the absence of subcortical activity. In retrospect, although our current data may lead to a greater understanding of how patterns of responses occur in the auditory cortex, our data do not address how hallucinations are formed. What we speculate here is that the bottom-up flow of sensory information could activate stereotyped cortical populations of neurons. We have thus toned down our discussion of the potential implications of the current work with respect to hallucinations.

Line 85. I'm not sure I agree with the statement that the thalamus is strikingly similar across modalities. The two references that are cited only discuss canonical cortical circuits.

We have changed “strikingly similar” to “relatively homogeneous” and have added another more recent reference to support this statement.

Line 91. The statement "all-or-none" to describe the population response seems inadequate, given the variability in "ON" responses in Figure 1D.

Although the ON cortical events showed some variability, the responses fell into two discrete categories. We have added the text “, with some variability within each class (Figure 1C)” to indicate that there is variability within the classes.

Figure 1. Where do these ON/OFF responses fall within the trial presentations – are there any time dependence that could reflect photobleaching? There is also a large fraction of OFF responses that seem to be driven below baseline (in Figure 1D) – is this an artifact of the dF/F computation? On that note, the methods do not explain how you computed dF/F (there are different ways that one can do this).

This is an excellent question. We note that the blue excitation light was not on during the whole presentation sequence. It was turned off between presentations and turned on 5 seconds prior to the stimulus onset, precisely to avoid bleaching. In addition, we interleave stimulus trials with blank trials and subtract the blank traces to eliminate any residual photobleaching effects, which can also produce negative values for dF/F. We have now added a figure to show that there are no apparent order effects in our data and added text to clarify the method for computing ∆F/F.

Line 105. 37dB is quite a specific intensity – why did you choose this?

We have found that the population ON/OFF phenomenon is seen near response threshold. Below threshold, no responses are found and above threshold, consistent full responses are found in the AC. In the case of 37 dB SPL, we did use higher intensities and found consistent ON responses. We have modified Figure 1 and the text to indicate that the presence of ON and OFF responses appears to be restricted to near-threshold stimuli.

Line 108 and Line 110. You say "to 10 presentations of the same tone", and "across the 40 trials of the same sound presentation". It's not quite clear what is meant.

We apologize for this error. We modified the figure to have an average image for the 40 trials.

Line 123. Activating the IC and observing AC response is due to di-synaptic circuitry. Therefore, does the AC "failure rate" change with the intensity of the stimulation? Can you stimulate hard enough to overcome the thalamic inhibition that you claim mediates the OFF responses?

Thank you for raising this issue. As stated above, the presence of ON/OFF responses does indeed depend on stimulus amplitude. The ON/OFF phenomenon is seen just above threshold and at high stimulus amplitudes, only ON responses are seen. We have added Figure 3 and corresponding text to address this issue.

Figure 2E. Why is the dF/F in the MGB orders of magnitude lower than IC or ACtx?

Excellent observation. We have previously shown that flavoprotein autofluorescence responses in the thalamus are significantly smaller than corresponding responses in the cortex (Llano et al. 2009, PMID: 19321634). We do not know the reason for the smaller responses, but speculate that because the flavoprotein signal is derived primarily from neuropil (where the majority of signal-generating mitochondria are), that brain regions with strongly aligned neuropil processes (like apical dendrites in pyramidal cells) produce stronger signals. We do note that the flavoprotein responses in the MGB, though small, are consistent and align well with the electrophysiological results presented in Figure 4. We have added clarifying text to the revised manuscript.

Line 139. It's not immediately clear to me how are you getting single-cell resolution using this one-photon preparation? The images shown throughout are "wide field" population images. Are you able to show any cellular examples? Are you using some kind of NMF algorithm to achieve cellular resolution? How are you controlling for issues like neuropil contamination (that are ubiquitous in 2P data).

Respectfully, we submit that single-cell analysis of 1-photon calcium imaging has been routinely done for some time (Maclean et al. 2005 PMID: 16337918, Berger et al. 2007 PMID: 17360827, Cameron et al. 2016 PMID: 27183102). We have now re-analyzed the data using neuropil correction and find similar results, and present modified data in Figures 2G and 5.

Line 146. On the issue of up states vs down states. If you histogram the membrane potential, is it bimodal to reflect the different states?

Thank you for the suggestion. We have made histograms of the membrane potential of layer 4 cells, and it clearly showed a bimodal distribution that implicates the different states of the cellular activity. The histograms are now in a supplementary figure (Figure 2—figure supplement 6).

Line 155-160. This isn't particularly well described. What exactly is it being modeled here? It seems like you are performing a binary classification, e.g. using either MGB activity or striatum activity to predict whether the corresponding cortical response was ON or OFF? Why does this imply that network-level activity may determine whether an on or off response occurred? Why is the striatum region irrelevant (i.e. it is still (indirectly) connected the cortex via the thalamus, no?). Why the need to use such an advanced machine learning method? Did simpler binary classification tools fail? Does this method also give the same result when used on the GCaMP or whole cell recordings?

Complying with the other reviewers’ and editor’s suggestions, we have removed this section from the manuscript.

Figure 3B. Why are there are more OFF responses that appear in the wash condition? Can you add a third variable to Figure 3C to show this – shouldn't the wash be statistically indistinguishable from control?

We apologize for the confusion. We have now compared the wash vs. baseline responses, and the differences are not significant. In revised figure 3C (now Figure 4C), we now use one-way ANOVA instead of paired t-test, and the test showed a nonsignificant difference between the control and wash groups.

Line 195. This is actually a question regarding choices throughout the paper – what dictated the choice of either flavoprotein or calcium? The deep learning results referenced here were established with flavoprotein, but these follow-up experiments use calcium.

We apologize for the confusion. The main imaging technique used in the paper is flavoprotein autofluorescence imaging because it does not require any modifications to the slice or to the animal. Calcium imaging was used to demonstrate that the ON/OFF phenomenon was not an artifact of the use of a metabolic signal, as well as to examine the cellular activity in layer 4 cells and MGB cells. The deep learning component of the manuscript has been removed, as suggested by the other reviewers and editor. We have revised the text accordingly.

Figure 4B. It's not quite clear what is going in this figure. Would some kind of population signal help? Or an alternative way to plot the data? Do these individual traces all have well-defined peak latencies (it's hard to tell from the figure). You're showing the variance of the peak latencies – but couldn't you just compare the peak latencies directly? Does it give the same result? Also, you talk about three categories in the text, but are only showing two categories in the figure. A latency analysis is also a little bit difficult to interpret with calcium data (both from the calcium timescale and the acquisition frame rate) – it might be a good idea to either discuss these limitations, since you are arguing for synchronicity based on a small reduction in variance within this data.

We apologize for the lack of clarity. Figure 4B was meant to show an overlay of all of the calcium traces during ON vs. OFF response. Figures 4B-4D are alternative ways of showing the same data. We respectfully would like to keep Figure 4B (now Figure 5E) because it shows the reader the raw traces and illustrates the increased synchrony during ON responses. Revised Figure 5A-C is taken from the latency of the responses as suggested. We agree that extracting latency data from calcium traces has significant limitations, which we now discuss in the revised manuscript.

Line 195. "Calcium signals of MGB cells were imaged following the stimulation of the IC during ON vs OFF cortical responses" – is it not the IC stimulation that is causing the ON or OFF cortical responses?

Thank you for pointing this out. The reviewer is correct and we have re-written this sentence for clarity as “To test this hypothesis, the time courses of calcium signals of MGB cells were imaged following the stimulation of the IC, and were compared during population ON vs. population OFF cortical responses.”

Line 373 – "The detailed procedures have been described before (77)". This seems like it may be the wrong citation? (77) is a theoretical conference paper with no in vivo imaging details.

Reference 77 (now reference 71) refers to a book chapter that we have written that describes the details of the imaging procedure.

Line 220 – Is there a particular reason to use neonates for the experiments in this section?

To obtain an aCTC slice with intact connection between IC, MGB, and AC, the slice should be taken from P15-18 mouse pups. As such, the viruses should have been injected into P4 neonates to allow a good time for gene expression the virus carries. We have added clarifying text to the revised manuscript and added more discussion about the issue that the slice work was all done in neonates.

Figure S7. I'm slightly confused by the MGB of the RBP4-Cre mouse – it looks like there is no TRN labelling but also no MGB labelling?

Layer 5-labeled terminals in the thalamus are sparse (Llano and Sherman 2008, PMID: 18181153), so at this magnification one cannot see them. We have added clarifying language, and, based on suggestions from Reviewer 2, an additional control which is to use a virus that does not express halorhodopsin and have added a figure that shows a high-magnification view of thalamic terminals present in the slice after layer 5-RBP4 labeling. The new data are in Figure 6—figure supplement 2.

Line 270. It's not possible to understand what you did here without jumping to the methods.

Based on the suggestion of the other reviewers and editor, this part of the manuscript was removed.

Line 274. Is there an intuitive reason why a shorter pre-stimulus period (which should contain more oscillatory information) be less predictive?

Based on the suggestion made by the other reviewers, this section was removed.

Line 276. "A classifier?" Can you justify your classifier choice here, compared to the more complicated deep learning approach used for classification earlier in the paper?

Based on the suggestion made by the other reviewers, this section was removed.

Figure S8. Based on the text, the gamma prediction is important to the circuit story. It may be a better idea to include it as a main figure?

Based on the suggestion made by the other reviewers, this section was removed.

Reviewer #2:

There is much to admire in the paper by Ibrahim and colleagues. They are the only lab in the world (to the best of my knowledge) that has worked out the geometry of a brain slab that leaves the synaptic connections intact from the midbrain-thalamus-MGB-TRN-cortex-back to TRN/MGB. Here, they use this slab to great effect by revitalizing the old cortical up/down state phenomenon in the context of all-or-nothing responses in A1 to stimulation of midbrain afferents. They have convinced me that the basic phenomenon is real and that it is interesting. Further, I am convinced that they can reduce the probability of OFF response trials by: (1) blocking GABAA receptors, (2) blocking AMPA receptors in TRN, (3) inactivating Ntsr1-Cre L6 CT neurons. They also have data to show that (4) inactivating feedforward inhibition from the IC or glutamatergic L5 projection neurons does not affect OFF response probability.

While there is a lot of good in this manuscript, there is plenty of bad and ugly as well. In their efforts to weave a complete theory of the phenomenon, they are forced to rely on a few very weak/poorly developed pieces of data. Emphasizing these more tenuous findings (Figure 4, Figure S4, Figure S8), ends up doing them a disservice because it takes away from their more solid observations. Apart from eliminating the dubious findings, the figures and the analysis are poorly organized, confusing and do not convey the process of processing the data from example cases to summary plots that relate to their statistics. They look more like lab meeting figures than manuscript figures.

In the final analysis, I think the work offers a significant conceptual advance based on an innovative technical approach. The manuscript can be substantially improved through eliminating the under-developed analyses, reorganizing the remaining figures and adding some additional analysis. The authors needn't feel compelled to oversell the reader on a complete comprehensive story when some of holes are not yet filled in. It's fine that some of the mechanisms are unknown so long as the shortcomings and caveats are clearly identified.

1. Figure 1 is potentially important but is not too convincing in its current form. One reason that it's important is that the rest of the manuscript is a developmental neuroscience study that is never acknowledges this fact. The intro and discussion with the talk of schizophrenia and hallucinations are too far-ranging for my taste, but particularly in the context that their study was performed in mice that had only been connected to the auditory world for less than 1 week. It would be very easy to write off the phenomenon they describe here as an oddity o development that would not apply to the mature cortex.

For me, this underscores the potential importance of Figure 1, where they demonstrate the phenomenon in vivo using older animals. The biggest impediments in its current form are (i) the age of the animals used for transcranial imaging is never mentioned. I presume they are adult?

Thank you for making these points. Yes, the animal shown in Figure 1 is an adult animal, which has now been described in the revised text. We have also added text to describe the limitations on the interpretation of the study, given the finding that the animals in the slice part of the study were young.

(ii) The entire analysis is based on a single tone frequency presented a single sound intensity that is very near threshold (5k at 37 dB SPL). This makes me question the generality of the principle described here. Having done this trial-by-trial analysis myself on this type of data, I don't think it really works for strong, suprathreshold stimulus intensities. Is this entire phenomenon restricted to a very narrow range of near-threshold intensities? In that case…is it really a phenomenon or just what happens near minimally effective stimulation levels (which is already very well known). Obviously, the same critique applies to the current levels they use for the rest of the study. It would have been nice to see a more systematic and quantitative examination of the stimulus parameters where this phenomenon is strong versus weak.

We now show that the likelihood of observing an OFF response decreases with increasing stimulus amplitude in modified Figure 1. This dependence on stimulus amplitude is also seen in the slice data and is now shown in revised Figure 3.

2. Imaging from "single cells". Single cell imaging is not possible with epifluorescence illumination in a slice this thick, especially without neuropil removal. At best they are ROIs but even then, what is the value when there is no basis for the argument that their ROIs are independent response units? I suggest eliminating 2G, Figure 4 and Figure S4. The issues with Figure 4 (and S4) go way beyond the cellular resolution issues, they are uninterpretable for a number of reasons.

We have removed the section on deep learning, and therefore removed previous Figure S4. We respectfully suggest that single-photon imaging of calcium activity in slices has been done for some time (Maclean et al. 2005 PMID: 16337918, Berger et al. 2007 PMID: 17360827, Cameron et al. 2016 PMID: 27183102). We have now re-analyzed the data using neuropil correction as suggested by the reviewer and have found similar results and present them in modified Figures 2 and 5.

3. In Figure 2D, I'm not convinced that they are looking at signals from TRN cell bodies as opposed to thalamocortical/corticothalamic axons. The thalamic radiation forms a dense fascicle right at the point of the TRN in this plane so it could be very hard to distinguish between TRN cell bodies and axon signals with the optics they are using. Too bad they didn't express GCaMP in the GAD-Cre mice they use for inhibiting the feedforward inhibitory IC projection neurons.

While it is correct that corticothalamic axons pass through this area, we think that these are likely signals from the TRN for four reasons. First, flavoprotein imaging produces very weak signals in axons. Second, this region is the precise location of the auditory region of the TRN. Third, in the current studies, we show that application of a glutamate blocker to this region strongly facilitates thalamocortical transmission – a finding that is inconsistent with this region being reflective of axons of passage. Fourth, to definitively establish that there are TRN cell bodies in this region, we were able to record from several TRN cells, and found that their firing was synchronized with IC stimulation and that these neurons had current/voltage profiles consistent with TRN neurons. We added this result as Figure 2—figure supplement 3.

4. Calcium imaging in the MGB and IC. Looking at other publications of the Thy1-GCaMP6 mouse, I see lots of labeling in pyramidal neurons in neocortex and hippocampus, but no labeling in the MGB or IC. I'm confused about what is being measured in Figure 2E. Maybe there is expression in younger mice or is just very faint?

This is a good point. GCaMP signal is smaller (but present) in the IC and fairly weak in the MGB in the original description of these mice by Dana et al. (PMID: 25250714). Our data are consistent with their findings, with the smallest signals deriving from the MGB. It is also certainly possible that part of the signal that is observed in IC and MGB is contaminated by flavoprotein signals, which are obtained with the same filter set, but are generally about an order of magnitude smaller than GCaMP signals. Either way, it does not impact the main conclusion from the use of GCaMP mice which is that the cortical ON/OFF responses are not an artifact of metabolic imaging. We have modified the text accordingly.

5. Figure 3E-H (graphics could be greatly improved here) show that EPSCs in MGB are unchanged in cortical ON vs OFF trials but IPSCs are elevated in OFF trials. This is one of the mysteries of their study that they try to address with a bunch of uninterpretable analysis methods (Figure 4 and Figure S4). The conundrum they face is how does enhanced thalamic inhibition create a cortical OFF response without making the MGB excitatory response weaker? Figure 4 and S4 attempt to explain this but are contrived and unconvincing. I don't disagree but I would suggest that they are thinking about it too simplistically. It doesn't matter what the MGB cell bodies are doing, what matters is what they are transmitting to the cortex during OFF trials. It would be helpful if they quantified these data more carefully by looking at the relative timing of the IPSC and EPSC responses. Looking at the study by Reinhold, Lien and Scanziani 2015, the TRN can regulate sustained spiking and thalamocortical synaptic depression in ways that cannot be dismissed just by measuring the excitatory response. They don't have to fully answer this here. They can do the best with whatever straightforward analysis they can manage and then devote some space in the discussion to actually discussing alternative explanations for their experiments instead of rhapsodizing about how these data solve the greatest mysteries of the brain.

Thank you for these suggestions. We have modified the graphics on this figure. We have quantified the timing of the IPSCs and EPSCs in the modified version of figure 3 (now Figure 5). Also, we have discussed these issues in the revised Discussion section.

6. Figure 3K – It would be important to see the individual data points. The legend says the data are an n=5, reflecting 5 slices from 4 animals. How do you get two of these slices from one animal? Regardless, this is a very small sample to make such a strong conclusion as blocking cortical GABAA receptors does not affect the OFF response probability. Was the AC Gabazine always performed after the MGB wash? The order should be counterbalanced and really I'm not even sure what the justification is for using a t-test on a sample this small. Unclear how the assumptions of the test can really be determined.

We are occasionally able to obtain two viable brain slices from a single animal retaining IC-MGB-AC connectivity. We have tested whether the data were normally distributed, and based on this result, ran a suitable non-parametric test. We have also re-run the pharmacology experiments in a counterbalanced fashion as suggested and obtained similar results (see modified Figure 4I-K).

7. I suggest putting all three of their halorhodopsin experiments into one figure (IC inhibitory feedforward neurons, L6 CT and L5 projection neurons). I'm also bothered that there isn't a control condition where mice have undergone a perinatal injection to express a control fluorophore. This seems particularly warranted because they injected an enormous volume of virus solution (1600 nL) into very small area. Apart from this control, while I think the L6 inactivation result is very interesting, I strongly disagree with their often repeated interpretation of the data, that "Population OFF responses were associated with TRN-mediated inhibition at the level of the MGB, under the control of layer 6 corticothalamic projections".

We appreciate the reviewer’s suggestion and we have now placed all of the halorhodopsin experiments into one figure. Also, we conducted a control experiment in which we injected 4 day old mice with the same serotype viral particles that do not express halorhodopsin receptor. We put the results of this experiment in revised Figure 6.

Their data do not support this conclusion. Yes, blocking AMPA receptors in TRN reduced the probability and Yes, silencing L6 CT also reduced the probability but it is a logical fallacy to assume those are causally related. Ntsr1 neurons make synaptic contacts on TRN neurons, MGB neurons, local AC inhibitory neurons and local AC excitatory neurons. It is entirely possible that the Ntsr1 neurons are mediating this effect through local connections in the cortex and not through the descending projection. In that regard, it is a bit frustrating that the main point of the Guo 2017 study was the Ntsr1 neurons can strongly bias AC neurons to become less responsive to afferent activity via intracortical connections, yet these findings are not mentioned in this context.

Don't get me wrong, they might be right. But providing that the L6 CTs modify MGB via TRN would require a different type of experiment to isolate the influence of L6 CT axons onto TRN neurons and not the other three types of neurons that it communicates with. Again, good topic for a revamped Discussion section.

This is an excellent point that we had not considered previously. The reviewer is absolutely correct that silencing NTSR1+ neurons may have multiple effects outside of their projections to the TRN, particularly based on their intracortical projections. We have modified the Discussion accordingly.

8. Figure S3 should be upgraded to a main figure and quantified.

We appreciate the reviewer’s suggestion, and we upgraded the figure to a main figure 3 and quantified the results.

9. Figure S6 is interesting but preliminary. There are only four data points and DREADD experiments are usually considered uninterpretable without a control condition in which CNO is applied with the expression of a control fluorophore (i.e., without the designer receptor) to account for known off-target effects of CNO.

We appreciate the reviewer’s suggestion, and we have done the control experiment by injecting 4 days old mice with the same stereotype viral particles that do not express the DREADD receptor. We put the results of this experiment in Figure 6—figure supplement 1.

10. The issues with Figure S7 are that Rbp4-Cre does not selectively label "PT-like" L5 cells that project sub-cerebrally. In fact, most of their axons remain within the cortex. To this point, the expression they show in no way resembles the cartoon in Panel A but more problematically does not compare to the expression in the Ntsr1 neurons. As such, it is hard to interpret much from the negative result because it may be that they are hyperpolarizing many fewer neurons overall and very often not the L5 neurons that project to the MGB.

We agree with the reviewer’s comment. Initially, we conducted this experiment as a negative control thinking that inhibiting another cell type in the cortex, not layer 6 cells, will not be able to retrieve the OFF cortical response. We respectfully suggest keeping the RBP4 data in the paper to help differentiate layer 5 vs. layer 6 effects. Clearly if an effect was seen after RBP4 silencing, one could not know if the effect was due to subcortical vs. cortical effects.

11. Figure S8 is another example of them trying to squeeze too much out of their data. First of all, I was unclear about how they were confident that they patched Ntsr1 neurons and not other L6 cell types. But mainly the analysis methods were a little strange to me. Why not just filter in the Gamma band (or other bands)? Also there is a huge literature showing the phase of ongoing 2-6Hz oscillations gate the response probability of AC neurons (e.g., in Guo 2017). It would have been reasonable calculate the phase and amplitude of the pre-stimulus membrane currents from its analytical signal using the Hilbert transform. In it's current form, the quantification in this figure is not convincing at all.

Apart from that, the Pre-stimulus L6 neuronal activity could only accurately predict 58.1 {plus minus} 3.5 SD% ON and OFF cortical responses, which is subtle compared to the theoretical random accuracy 55.1%. Moreover, should the random accuracy be calculated by training and testing the model with shuffled ON and OFF responses?

Based on the suggestion made by the reviewers and the editor, this part was removed.

Reviewer #3:

This is an interesting paper that finds that the cortical feedback to the thalamic reticular contributes to the omission of responses in the cortex. This is an important and surprising finding, and the authors bring a whole arsenal of state-of-the-art tools to dissect the contribution of this pathway. The variety of the methods, that include calcium imaging, flavoprotein imaging, electrical stimulation, whole cell recording, pharmacologic/optogenetic/chemogenetic manipulations, all point towards the same conclusions. The data that are presented support most of the claims of the paper. Most of my concerns reflect the interpretation of the data, the focus of the introduction and discussion, some missing controls and the lack of report of statistics in the main body. I believe that the authors should be able to address these concerns with the data in hand, through additional analysis and re-interpretation.

1. The emphasis in the introduction and discussion on the "classical models" of purely feedforward hierarchical processing seems somewhat forced. Numerous studies, especially in the auditory system, have addressed the importance of feedback in anatomy, auditory processing, learning and behavior. Rather than skipping over this important work, the paper would benefit from the discussion of previous work on the function of feedback in sensory processing. This would furthermore allow the authors to better define what role their work plays in this context.

Thank you for the suggestion. We have made substantial modifications to the Introduction in accordance with this suggestion.

2. Similarly, the emphasis on the role of this feedback in neuronal oscillations seems to be overstated. In the introduction, the discussion of hallucinations seems to come out of nowhere. In the paper, only one supplementary figure is devoted to analysis of oscillations, whereas in the abstract, introduction and discussion, these is an extensive focus on the gamma oscillations. The gamma activity-based prediction of ON vs OFF responses seems to be weak. The classifier only had an accuracy of 58%, which is not very high, and is only barely above random accuracy of 55. This focus and confusing result detracts from the main message of the paper.

Thank you for these suggestions. We have modified the Discussion with respect to the potential connection to hallucinations. Based on the suggestion made by the reviewers and the editor, the gamma oscillations has been removed.

3. The terminology of ON and OFF responses seems to be at odds with the accepted terminology in the auditory field. Typically, in the auditory field, ON responses refer to the activity at the onset of a stimulus, and OFF responses refer to the activity after stimulus offset. OFF responses, where there is no response at all, is confusing. The authors allude to the use of UP and DOWN states in the literature, but there is only indirect evidence that this activity indeed corresponds to those states. Perhaps using "Response" and "Omission"; yes/no; 1/0 or +/- could work.

Thank you and we have struggled with this terminology ourselves. Reviewing the classical auditory literature, most authors use “onset” and “offset” to refer to response at the beginning vs. the end of a tone rather than ON/OFF (Qin et al. 2007, PMID: 17360820, Anderson and Linden 2016, PMID: 26865621, Kopp-Scheinpflug et al. 2018, PMID: 30274606). We have been reluctant to call these UP/DOWN responses because we do not yet know how completely UP/DOWN aligns with ON/OFF. We thank the reviewer for the other suggestions, but also find them to be less descriptive. We respectfully suggest using the terms “population ON” and “population OFF” responses and have made the corresponding changes in the revised manuscript.

4. The claim that the thalamus selects co-activated cortical ensembles, referred to in the abstract, last sentence of the introduction as well as the discussion, is not supported by data. Perhaps it is known that the auditory cortex uses partially overlapping ensembles of coactivated neurons to represent auditory stimuli, but that data is not represented in this paper. In fact, the data in this paper hinges on an "all or nothing response," which undermines this idea of coactivated ensembles in the AC. These contrasting ideas deserve reconciliation.

We agree that our data do not establish that specific ensembles of neurons are activated by particular stimuli. We have replaced the word “ensemble” with “population” or “group” of neurons in the Abstract and Introduction, respectively, which we believe has less specific implications than “ensemble.”

5. Statistics for all claims should be reported in the main text. Some statistics are missing, for example, in Figure 5D, the distribution of the histograms.

Thank you, we have moved the statistics from the figure legend to the main text.

6. It is great that the paper starts with results in vivo. However, those are under anesthesia. Are OFF responses also present in awake animals, perhaps influenced by attention or arousal levels? Perhaps worth speculating in the discussion on the functional importance of this corticothalamic gating in awake behaving animals.

Thank you. We have added text to highlight this point in the revised Discussion.

7. The authors use a deep learning algorithm to establish the predictive power of the response, training on MGB data vs irrelevant striatum. How would the algorithm work for IC data? Furthermore, it is unclear whether deep learning is required for this analysis. Would more simple statistical measures, such as correlation or Granger causality measures work here as well? Can the predictive ability of layer 6 activity be improved with just using pre-stimulus layer 6 firing rate, as opposed to LFP power spectra?

We appreciate the reviewer’s suggestion but based on the suggestion made by the reviewers and the editor, this part was removed.

8. The authors observed cortical IPSCs during ON responses but not OFF responses. The importance of this finding would be better demonstrated by comparing it with cortical EPSCs. Presumably, ON responses are characterized by large EPSPs and medium IPSCs, whereas OFF responses are characterized by no EPSCs or IPSCs. That way, it's clearer that OFF responses are not simply caused by increase in IPSC magnitude. It could then be contrasted that with the increase in IPSC magnitude observed in MGB, as they show in figure 3G. The importance of figure 3E/3F is unclear. It would be great to explain why IPSCs are observed in cortical ON responses and not OFF responses.

We apologize for the lack of clarity. The important outcome of these experiments is the absence of cortical IPSCs during OFF responses. The presence of barrages of IPSCs during ON responses was initially surprising to us, but is consistent with ON responses being found during UP states. UP states are associated with barrages of both EPSCs and IPSCs (Hasenstaub et al. 2005, PMID: 16055065, Tahvildari et al. 2012, PMID: 22933799, Salkoff et al. 2015, PMID: 26180200). We have now quantified the magnitude of the IPSCs during ON vs. OFF responses (revised Figure 4F) to make this point more clear, and to provide more contrast with the MGB IPSC magnitudes, which were larger during the OFF responses.

9. How well do gabazine and NBQX stay localized to the area of injection? Is there any chance of diffusion of the agent confounding results? How was localization confirmed?

We mixed the drug and the ACSF as control with Alexa Fluor-594 dye, so we could visualize the injection of the solution injected. We provided the images or the videos of these injections. Figure 4—figure supplement 1 shows the localized injection of the drugs.

10. What are the controls for chemogenetics?

Thank you for this suggestion. We have now done the control for chemogenetic probe and we put the results in Figure 6—figure supplement 1.

11. In Figure 4, when performing calcium imaging in MGB, which trials had cortical ON vs OFF responses? Was that determined through simultaneous LFP recording in AC (diagram in figure 4A)?

Yes, there was a simultaneous LFP recoding and MGB calcium imaging. We have modified the text to make this point more clear.

12. How well can latency of response and response timing variability (Figure 4) be assessed using GCaMP6s, which has a response time scale on the order of seconds? Is the variability in response timing really on the order of hundreds of milliseconds (Figure 4D)? Perhaps this would be better evaluated using whole cell recording or electrophysiology.

We agree that there are significant limitations in the interpretability of response timing using calcium signals. However, gold-standard approaches to measuring timing, such as electrophysiology in the slice, limit the number of neurons simultaneously sampled. We have now added text describing the limitations to this approach.

13. Is it MGB synchrony, or MGB inhibition, that leads to cortical ON vs OFF responses? The importance of MGB synchrony, or at least its interaction with TRN-MGB inhibition, is unclear. Furthermore, the synchrony analysis should be conducted with more appropriate statistical tools, including correlation measures. It remains unclear how a desynchronized MGB leads to an OFF response (aka no firing) in the cortex. Do layer 4 cortical cells not reach firing threshold when the inputs are asynchronous? Or do cortical neurons not receive inputs from MGB when the neurons fire asynchronously (due to further gating by third party cells)?

We believe that the data suggest that inhibition of the MGB leads to desynchronized MGB output, leading to OFF responses. Synchrony in this case was measured as variance in latency across the population which is a direct measurement of whether MGB neurons are responding at the same time. The idea that thalamic synchrony is needed to elicit a cortical response and that TRN may disrupt thalamic synchrony is established in the literature (Bruno and Sakmann et al. 2006, Pita-Almenar et al. 2014 PMID: 25339757). We have made this point more clearly in the revised Discussion.

14. Is the difference in power at ~36Hz significant when controlling for multiple comparisons across 11 different LFP power spectra frequencies? Perhaps the authors could try running a 2-way ANOVA with frequency and ON/OFF status as factors.

Based on the suggestion made by the reviewers and the editor, the gamma oscillations part was removed.

15. I recommend tightening the writing style throughout: e.g. in abstract: "Here, we elucidate the mechanism for gating of population activity." Some paragraphs are multiple pages long. There are run-on sentences, and the authors use passive tense extensively. I would recommend splitting up the paragraphs, so that each paragraph contains one result/analysis, as well as two sentences for motivation and conclusion.

We appreciate the reviewer’s suggestion and have revised the manuscript extensively.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The authors are to be commended for carrying out a number of new experiments to provide essential control data and to provide additional information that has strengthened the study. The majority of the other concerns raised by the reviewers have been dealt with adequately, but there are a few remaining issues that need to be addressed.

Both the in vivo and slice data now clearly illustrate the dependence of the cortical responses on the stimulus magnitude, with the probability of an OFF response occurring declining with increasing stimulus strength. This reinforces one of the points made by reviewer 2, namely that these effects appear be present only for near minimally effective stimulation levels. That raises the question of what function they serve and how these responses contribute to our understanding of cortical processing under more natural conditions (including in awake animals at higher sound levels). These are important issues that need to be addressed in the Discussion.

We agree and have added the following text to the Discussion:

“Moreover, the strong association between the high probability of the OFF cortical responses and the low level of stimulation suggests that cortico-reticulothalamic control of thalamocortical transmission is most likely to be effective when signal-to-noise ratio is low. This finding is consistent with the notion that top-down modulation is mostly required for the attentional modulation of weak signals and that highly salient signals rely on bottom-up mechanisms to activate perceptual representations (reviewed in [45]).”

The readability of the text would benefit further from greater use of paragraphs (as pointed out by reviewer 3). For example, one paragraph runs from lines 304-380. This does include some deleted text, but is still far too long.

Thank you for the suggestion. We have broken this paragraph up into two paragraphs. The second paragraph focuses on the TRN experiments only.

Line 364: 5J should be 6J.

We have corrected this error.

Line 423: "exposure to pure tones" might be construed as a form of passive acoustic environment. "Presentation of pure tones" would be less ambiguous.

We have changed this phrase as suggested

Figure 1 legend: there are several full stops missing.

We have added the full stops.

Figure 2 – Supplementary Figure 3: the legend refers to red and green boxes, but there are none in the figure (at least in the composite pdf of the manuscript).

Thank you for finding this error. These colors refer to the previous version of the manuscript. We have changed the legend.

Line 943: the sentence ends "showing".

Thank you. We deleted the word “showing.”

Figure 6 – Supplementary Figure 1G: all traces are black, whereas the legend claims that some are brown or green.

Thank you. We removed references to brown/green in the legend.

https://doi.org/10.7554/eLife.56645.sa2

Article and author information

Author details

  1. Baher A Ibrahim

    1. Department of Molecular and Integrative Physiology, University of Illinois, Urbana-Champaign, United States
    2. Beckman Institute for Advanced Science and Technology, University of Illinois, Urbana-Champaign, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing, Performing most of the experiments
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0062-7589
  2. Caitlin A Murphy

    1. Department of Neuroscience, School of Medicine and Public Health, University of Wisconsin-Madison, Wisconsin-Madison, United States
    2. Department of Anesthesiology, School of Medicine and Public Health, University of Wisconsin-Madison, Wisconsin-Madison, United States
    Contribution
    Validation, Investigation, Methodology, Performing some experiments
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6319-9470
  3. Georgiy Yudintsev

    Neuroscience Program, University of Illinois, Urbana-Champaign, United States
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
  4. Yoshitaka Shinagawa

    Department of Molecular and Integrative Physiology, University of Illinois, Urbana-Champaign, United States
    Contribution
    Software, Formal analysis, Writing - review and editing
    Competing interests
    No competing interests declared
  5. Matthew I Banks

    1. Department of Neuroscience, School of Medicine and Public Health, University of Wisconsin-Madison, Wisconsin-Madison, United States
    2. Department of Anesthesiology, School of Medicine and Public Health, University of Wisconsin-Madison, Wisconsin-Madison, United States
    Contribution
    Validation, Investigation, Methodology
    Competing interests
    No competing interests declared
  6. Daniel A Llano

    1. Department of Molecular and Integrative Physiology, University of Illinois, Urbana-Champaign, United States
    2. Beckman Institute for Advanced Science and Technology, University of Illinois, Urbana-Champaign, United States
    3. Neuroscience Program, University of Illinois, Urbana-Champaign, United States
    4. College of Medicine, University of Illinois, Urbana-Champaign, United States
    Contribution
    Conceptualization, Resources, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Project administration, Writing - review and editing
    For correspondence
    d-llano@illinois.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0933-1837

Funding

NIDCD (R01DC013073)

  • Daniel A Llano

NIDCD (R21DC014765)

  • Daniel A Llano

National Science Foundation (1515587)

  • Daniel A Llano

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Dr. Murray Sherman (The Department of Neurobiology, University of Chicago), Dr. Shane Crandall (The Department of Physiology, Michigan State University), and Dr. Brian Theyel (The Department of Neuroscience, Brown University) for their valuable comments on the manuscript.

Ethics

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols (#18236) of the University of Illinois at Urbana-Champaign. All surgery was performed under anesthesia, and every effort was made to minimize suffering.

Senior and Reviewing Editor

  1. Andrew J King, University of Oxford, United Kingdom

Reviewer

  1. Daniel B Polley, Massachusetts Eye and Ear Infirmary, United States

Publication history

  1. Received: March 5, 2020
  2. Accepted: May 23, 2021
  3. Accepted Manuscript published: May 24, 2021 (version 1)
  4. Version of Record published: June 8, 2021 (version 2)

Copyright

© 2021, Ibrahim et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 562
    Page views
  • 85
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Computational and Systems Biology
    2. Neuroscience
    Jack Goffinet et al.
    Research Article Updated

    Increases in the scale and complexity of behavioral data pose an increasing challenge for data analysis. A common strategy involves replacing entire behaviors with small numbers of handpicked, domain-specific features, but this approach suffers from several crucial limitations. For example, handpicked features may miss important dimensions of variability, and correlations among them complicate statistical testing. Here, by contrast, we apply the variational autoencoder (VAE), an unsupervised learning method, to learn features directly from data and quantify the vocal behavior of two model species: the laboratory mouse and the zebra finch. The VAE converges on a parsimonious representation that outperforms handpicked features on a variety of common analysis tasks, enables the measurement of moment-by-moment vocal variability on the timescale of tens of milliseconds in the zebra finch, provides strong evidence that mouse ultrasonic vocalizations do not cluster as is commonly believed, and captures the similarity of tutor and pupil birdsong with qualitatively higher fidelity than previous approaches. In all, we demonstrate the utility of modern unsupervised learning approaches to the quantification of complex and high-dimensional vocal behavior.

    1. Neuroscience
    Iris Bachmutsky et al.
    Short Report Updated

    Opioids are perhaps the most effective analgesics in medicine. However, between 1999 and 2018, over 400,000 people in the United States died from opioid overdose. Excessive opioids make breathing lethally slow and shallow, a side-effect called opioid-induced respiratory depression. This doubled-edged sword has sparked the desire to develop novel therapeutics that provide opioid-like analgesia without depressing breathing. One such approach has been the design of so-called ‘biased agonists’ that signal through some, but not all pathways downstream of the µ-opioid receptor (MOR), the target of morphine and other opioid analgesics. This rationale stems from a study suggesting that MOR-induced ß-arrestin 2 dependent signaling is responsible for opioid respiratory depression, whereas adenylyl cyclase inhibition produces analgesia. To verify this important result that motivated the ‘biased agonist’ approach, we re-examined breathing in ß-arrestin 2-deficient mice and instead find no connection between ß-arrestin 2 and opioid respiratory depression. This result suggests that any attenuated effect of ‘biased agonists’ on breathing is through an as-yet defined mechanism.