Interhemispheric callosal projections enforce response fidelity and frequency tuning in auditory cortex

Sensory cortical areas receive glutamatergic callosal projections that link information processing between brain hemispheres. However, the role of interhemispheric projections in sensory processing is unclear. Here we use single unit recordings and optogenetic manipulations in awake mice to probe how callosal inputs modulate spontaneous and tone-evoked activity in primary auditory cortex (A1). Although activation of callosal fibers increased firing of some pyramidal cells, the majority of responsive cells were suppressed. In contrast, callosal stimulation consistently increased fast spiking (FS) cell activity and brain slice recordings indicated that parvalbumin (PV)-expressing cells receive stronger callosal input than pyramidal cells or other interneuron subtypes. In vivo silencing of the contralateral cortex revealed that callosal inputs linearly modulate tone-evoked pyramidal cell activity via both multiplicative and subtractive operations. These results suggest that callosal input regulates both the salience and tuning sharpness of tone responses in A1 via PV cell-mediated feedforward inhibition.


Introduction
Cortical sensory representations driven by thalamic inputs are strongly influenced by local intracortical circuits and long range projections including interhemispheric inputs (Carrasco et al., 2013a(Carrasco et al., , 2015Cerri et al., 2010;Lee et al., 2019;Li et al., 2013;Lien and Scanziani, 2013;Schmidt et al., 2010;Wunderle et al., 2015). In most sensory systems there is an early decussation such that each hemifield of a sensory modality is primarily represented in the contralateral hemisphere of the brain. However, sensory areas for a particular modality in both cortices are linked to each other via interhemispheric projections from axons within the corpus callosum. These long range, cortico-cortical projections contact a majority of neurons in both 3 supra-and infragranular layers (Carr and Sesack, 1998;Petreanu et al., 2007;Wise and Jones, 1976), but their postsynaptic targets and degree of connectivity vary for different sensory cortical areas (Harris et al., 2019). The differences in callosal connectivity with pyramidal cells and local interneurons is reflected in previous studies indicating that activation of callosal inputs can drive excitation and/or inhibition in cortical circuits (Anastasiades et al., 2018;Karayannis et al., 2007;Lee et al., 2014;Rock and Apicella, 2015). Although recent studies have begun to characterize the functional properties of interhemispheric cortical projections, how these pathways contribute to sensory coding in vivo is not well understood.
Although callosal inputs arise from the axons of pyramidal cells in the opposite cortex, this pathway may not simply lead to cortical excitation. Indeed, in anesthetized ferrets, electrical stimulation of callosal inputs caused a variety of effects on sound-evoked firing rates including enhancement, suppression, or a mixture of the two (Kitzes and Doherty, 1994). Furthermore, intracellular recordings in A1 of anesthetized cats found that electrical stimulation in contralateral A1 elicited excitatory postsynaptic potentials that were often followed by inhibitory postsynaptic potentials (Mitani and Shimokouchi, 1985). These findings are consistent with a 4 recent brain slice study indicating that A1 callosal inputs drive strong activation of layer 5 (L5) PV cells that mediate feedforward inhibition of pyramidal cells (Rock and Apicella, 2015).
Despite these results suggesting a potential inhibitory influence of callosal inputs in auditory processing, removing interhemispheric input in anesthetized cats using cortical cooling was found to reduce sound-evoked activity in contralateral primary cortex (Carrasco et al., 2013a).
However, anesthesia itself strongly influences spontaneous and sensory-evoked activity in sensory cortex (Harris and Thiele, 2011;Kato et al., 2015) and it is unclear how callosal input modulates A1 sensory processing in the awake state.
In this study, we use linear silicon probes spanning cortical layers to record spontaneous and tone-evoked single unit activity in A1 of awake, head-fixed mice. We express channelrhodopsin-2 (ChR2) in callosal fibers to study how their local activation modulates activity in vivo and identify the local circuits driven by callosal input in brain slice recordings.
Finally, we use ChR2 in GABAergic interneurons to acutely suppress activity in one hemisphere while recording tone-evoked responses in contralateral A1 to show how the callosal pathway modulates cortical sensory processing.

Results
We first studied how local activation of callosal projections modulates cortical excitability by targeting injection of adeno-associated virus (AAV) expressing ChR2 to A1 of the left hemisphere (Fig. 1A). Dense expression of ChR2 in fibers within the left medial geniculate body (MGB) confirmed that injections targeted primary auditory cortex (Fig. 1A2). We inserted linear silicon electrodes in A1 of the right hemisphere to monitor single unit activity in the awake state. Post-hoc analysis of probe recording sites revealed callosal ChR2-expressing fibers 5 distributed across all layers of A1 (Fig. 1A2). Trough to peak time and full width at half maximum of spike waveforms (Fig. 1B) were used to classify single units as regular spiking (principal cells) or fast spiking (presumptive PV-expressing interneurons).
We used brief (5 ms) LED illumination (470 nm) of the recording site to activate callosal inputs. On average, callosal stimulation caused a biphasic response in both RS (n = 264) and FS (n = 33, n = 7 mice) cells: a rapid increase in firing rate followed by a decrease in firing that returned to baseline over 50-100 ms (Fig. 1C). However, individual RS cells in the same experiments responded quite differently from each other: some cells were transiently excited by callosal stimulation, while others were exclusively inhibited ( Fig. 1D1,2). We used a modulation index (Methods) to quantify early changes in firing (within 10 ms of callosal LED stimulation).
We found that RS cells were more likely to be significantly inhibited than excited (Fig. 1D3, p < 0.05, sign test) in layers 2/3 (L2/3), 4 (L4) and 5, while cells were equally likely to be excited or We next used voltage clamp recordings in brain slices to better understand the layer and cell type specificity of callosal input. We first examined the relative strength of callosal input onto PV and pyramidal cells. PV-Cre mice were crossed to a td-Tomato reporter line (Ai14) to 6 target whole-cell recordings of visually identified PV cells and AAV-ChR2 was injected into A1 of the left cortex. We measured responses using simultaneously recorded pairs of PV and pyramidal cells (Pyr) from L2/3 of the cortex contralateral to the injection (Fig. 2A1). At -70 mV (near the reversal potential for GABAergic inhibition), brief LED illumination (470 nm, 2-4 ms) elicited excitatory postsynaptic currents (EPSCs) that were much larger in PV than pyramidal cells (peak EPSC amplitude PV=628±80 pA, Pyr=168±50 pA, n=6 pairs, p=0.003, paired t-test).
Depolarization to +10 mV (near the reversal potential for glutamatergic excitation), revealed callosal input-evoked inhibitory postsynaptic currents (IPSCs) in both cell types. IPSCs always followed EPSCs with a brief delay in pyramidal and PV cells (average latency 2.13±0.51 ms, n = 8, and 1.81 ±0.2 ms, n = 10, respectively) indicating that inhibition was evoked indirectly by callosal input in a feedforward fashion (Isaacson and Scanziani, 2011). The ratio of excitation to inhibition (E/I ratio) was also markedly smaller in pyramidal than PV cells in L2/3 (0.11±0.01 and 0.33±0.06, respectively, n = 5 pairs, p = 0.01, paired t-test). Similarly, recordings in pairs of To directly examine the functional role of interhemispheric input in vivo, we recorded from A1 in awake mice while optogenetically suppressing activity in the contralateral auditory cortex. We injected AAV-FLEX-ChR2 (Atasoy et al., 2008) in the left cortex of Gad2-Cre mice to express ChR2 in GABAergic interneurons (Fig. 3A1, Kato et al., 2015). Recordings in the injected cortex confirmed that LED illumination (20 Hz train of 10 ms pulses) drove firing of FS cells (Fig. 3A2) while RS cell activity was largely abolished (Fig. 3A3,4). We next monitored spontaneous activity in A1 of the right hemisphere while silencing contralateral A1 (Fig. 3B1).
Although it has been suggested that GABAergic interneurons in auditory cortex can make 8 interhemispheric projections (Rock et al., 2018), we did not observe ChR2-expressing fibers in A1 contralateral to the AAV-injected cortex (Fig. 3B2 We next examined how silencing contralateral cortex modulates tone-evoked activity of RS cells in A1. The right ear was occluded and pure tones (9 log-spaced frequencies, 4-60 kHz, 250 ms, 60 dB) were delivered to the left ear during optogenetic silencing of the left hemisphere on interleaved trials (Fig. 4A, tone onset 250 ms following start of LED illumination). RS cells recorded from right A1 were frequency-tuned (Fig. 4B) such that particular frequencies drove strong firing ("preferred tones") while others evoked weak responses ("non-preferred tones").
Interestingly, the effects of cortical silencing on RS cell activity were dependent on the strength of tone-evoked responses. Firing rates during non-preferred tones were enhanced by contralateral silencing, while firing evoked by preferred tones were largely unaffected or reduced (Fig. 4B1,  4D2). This effect could be described by a simple linear transformation: firing rates during tones with vs. without LED-induced silencing could be fit by a line with a slope < 1 and y-intercept > 0 (Fig. 4B2). In other words, removing callosal input had both an additive and divisive action on A1 tone responses. The effects of contralateral cortical silencing were uniformly divisive across all cortical layers (Fig. 4C1) while additive effects were prominent in all but L6 (Fig 4C2).
Together, these results suggest that callosal input normally regulates sound-evoked responses via multiplicative and subtractive effects.
Divisive/multiplicative operations exert gain control of neural responses while subtractive/additive operations modulate response fidelity via changes in variability associated with stimulus-independent ("background") activity (Isaacson and Scanziani, 2011;Silver, 2010).
Both the increase in spontaneous activity and additive effects on tone responses during contralateral cortical silencing suggest that callosal inputs enforce response fidelity. To address this possibility, we computed the discriminability index (d', Methods), a measure of response reliability from signal detection theory (Duguid et al., 2012;Sturgill and Isaacson, 2015;Tolhurst et al., 1983) with and without contralateral cortical silencing. Optogenetic cortical inactivation significantly reduced the discriminability of tone-evoked activity (Fig. 4D1 d′LED-off = 7.37 ± 0.45, d′LED-on = 5.58 ± 0.41, n = 124, P < 0.001, t-test) indicating that callosal input normally serves to enhance the representation of tone responses relative to spontaneous activity in A1.
We examined how callosal input modulates the shape of frequency tuning curves by normalizing cell responses to their best frequency (BF, tone eliciting strongest increase in firing) under control conditions. Silencing contralateral cortex caused a small decrease in the amplitude of responses at BF (Fig. 4D2, p = 0.01, t-test), consistent with the divisive effect we observed on 10 input-output relationships (Fig. 4C). However, due to its additive action, cortical silencing also increased responses to non-preferred frequencies. The net effect is thus a "flattening" of the population frequency tuning curve (Fig. 4D2). Thus, in addition to regulating response fidelity, callosal inputs normally play an important role in enforcing the sharpness of frequency tuning in A1.

Discussion
We show that activating interhemispheric callosal projections can inhibit pyramidal cells in all layers of A1 in awake mice. These findings are consistent with slice recordings indicating that callosal inputs evoke strong feedforward inhibition of pyramidal cells in supra-and infragranular layers. This feedforward inhibition likely reflects the recruitment of PV cells, which receive stronger callosal excitation than SOM or VIP cells in upper and lower cortical layers. In loss-of-function experiments, acute in vivo silencing of contralateral cortex increased pyramidal cell spontaneous activity in all but L6. Finally, we used tone-evoked activity to show that cortical silencing linearly transforms A1 input-output relationships via subtractive and divisive operations. This indicates that interhemispheric projections normally enhance the salience of tone representations (by regulating signal to noise ratio) and sharpen frequency tuning in primary auditory cortex.
It is well established that callosal inputs make direct excitatory connections onto cortical pyramidal cells (Anastasiades et al., 2018;Karayannis et al., 2007;Lee et al., 2014Lee et al., , 2019Petreanu et al., 2007;Rock and Apicella, 2015) and drive disynaptic feedforward inhibition via contacts onto local GABAergic interneurons (Anastasiades et al., 2018;Karayannis et al., 2007;Rock and Apicella, 2015). Indeed, we found that brief activation of callosal fibers drives a 11 biphasic increase and decrease in the firing of RS and FS cells in awake mice. Surprisingly, individual RS cells across all cortical layers were more likely to be inhibited than excited by callosal stimulation. In contrast, FS cells were more routinely activated, suggesting that the suppressive effects of callosal stimulation on RS cell firing are due to widespread PV cellmediated feedforward inhibition. Consistent with this idea, brain slice recordings revealed that PV cells receive more callosal input than neighboring pyramidal cells or other interneuron subtypes and deep layer PV cells received ~2X stronger input than L2/3 PV cells.
Previous studies in sensory cortical areas have used callosal sectioning (Engel et al., 1991;Payne et al., 1980) or reversible cortical cooling to probe the functional role of callosal inputs in anesthetized animals (Carrasco et al., 2013(Carrasco et al., , 2015Cerri et al., 2010;Schmidt et al., 2010;Wunderle et al., 2015). We show in awake mice that acute optogenetic silencing has heterogeneous effects on spontaneous activity: although a subset of RS cells shows a rapid and sustained decrease in activity, the majority of cells responded with a slow sustained increase in firing. The simplest interpretation of these results is that decreases in activity reflect the withdrawal of direct excitatory callosal input onto particular cells, while paradoxical increases in firing reflect indirect network effects. Increases in firing are most likely due to a reduction in inhibition provided by PV cells. Indeed, we observed that the spontaneous firing of deep layer PV cells was strongly suppressed during contralateral cortical silencing. This suggests that much of the tonic activity of deep layer PV cells is driven by interhemispheric input. Deep layer interneurons have recently been shown to project axons through all cortical layers towards the pia (Bortone et al., 2014;Frandolig et al., 2019). It is possible that interlaminar projections from deep layer PV interneurons mediate the indirect network effects underlying principal cell excitation following withdrawal of callosal input.
In contrast to previous work in auditory cortex of anesthetized animals (Carrasco et al., 2013b(Carrasco et al., , 2015, we did not observe a simple reduction in the strength of tone-evoked responses during contralateral silencing in the awake state. Rather, input-output plots of tone-evoked firing were linearly transformed in a divisive and additive fashion, reflecting both the withdrawal of direct callosal excitatory input on pyramidal cells and reduction in feedforward inhibition. Higher spontaneous activity and stronger inhibition in the awake state are likely to underlie these differences (Haider et al., 2013;Kato et al., 2015). In terms of frequency tuning, this led to a small reduction in responses to BF while responses to flanking non-preferred frequencies were enhanced. Thus, in addition to enhancing the discriminability of sound-evoked responses by maintaining a high signal to noise ratio, callosal inputs sharpen frequency tuning in primary auditory cortex. These findings are in agreement with previous studies indicating that interhemispheric connections modulate the specificity of sensory-evoked activity in visual (Hubel and Wiesel, 1967;Schmidt et al., 2010;Wunderle et al., 2015) and somatosensory cortex (Clarey et al., 1996). In future, it will be useful to determine how callosal input contributes to binaural cortical sound representations and auditory-directed behaviors such as sound localization.

Extracellular recordings
A 32-(Neuronexus) or 64-(Cambridge Neurotech) channel silicon probe was used for extracellular recordings. Signals were recorded using an Intan RHD2000 and digitized at 20 kHz using Open Ephys (Siegle et al., 2017). Spikes were sorted using Kilosort (Pachitariu et al., 2016), followed by manual curation in phy (Rossant et al., 2016) to obtain single units used for analyses. Cells were excluded from analysis if they did not maintain consistent firing and amplitude throughout recording, and a firing rate of at least 1Hz. The probe was coated in DiI to verify probe track for depth of recording as well as recording location. Current source density (Pettersen et al., 2006) coupled with anatomical verification of probe track was used to identify laminar single unit locations. For all recordings spike waveforms were obtained from the lead with the largest amplitude template, these were then averaged to obtain an average spike waveform. Units were classified as fast spiking if their average spike waveform had a trough to peak time of less than 300 µs and a full-width at half max of less than 125 µs.
A fiber-coupled LED (470 nm, 20 mW, 0.4 mm fiber, 0.48 N.A., Thorlabs) was positioned within 1-2 mm of the exposed cortical surface for activating ChR2-expressing callosal fibers or ipsilateral cortical silencing. For experiments using contralateral silencing, the skull over right auditory cortex was exposed and covered with cyanoacrylate glue before the LED fiber was positioned at the skull surface. Callosal fiber activation was achieved using a single 5 15 ms flash. For cortical silencing in Gad2-cre mice expressing ChR2 we used a train of 10 ms light pulses (510 ms, 20 Hz) to activate inhibitory interneurons.
Immediately prior to recording, mice were anesthetized and the ear canal contralateral to the recording was filled with cyanoacrylate glue to occlude the ear. To prepare the recording site, a well filled with artificial cerebrospinal fluid (aCSF, in millimoles: 142 NaCl, 5 KCl, 10 glucose, 10 HEPES, 3.1 CaCl2, 1.3 MgCl2, pH 7.4, 310 mOsm) was constructed around the recording site and a small (<0.3 mm) craniotomy was performed through thinned skull. Mice recovered for >1hr before the start of recording. Pure tones logarithmically spaced between 4 kHz and 60 kHz were delivered via a calibrated free-field speaker (ES1, TDT) directed to the left ear. Tones were generated by software (BControl; http://brodylab.org) running on MATLAB (MathWorks) communicating with a real-time system (RTLinux). Tone frequencies (250 ms duration) were presented in a pseudo-random fashion and LED illumination was delivered on interleaved trials.

Analysis of in vivo data
For presentation of pooled neuronal responses, firing rates were normalized to the average baseline firing rate of each neuron 250 ms before the LED period. The analysis window for callosal terminal excitation was 10 ms from LED onset to capture both the initial excitation and recurrent inhibition. In contralateral A1 silencing experiments, the window for analysis was a 250 ms time period that started 250 ms after LED onset. All statistical tests were two sided and used a significance level of 0.05 (corrected for multiple comparisons where noted). Units were considered significantly modulated by the LED if the mean firing rate during the analysis window was different than that of the baseline period as determined by a Wilcoxon sign-rank test α = 0.05. Modulation index was calculated as [(mean firing rate in analysis window) -(mean firing rate during baseline period)]/[(mean firing rate in analysis window) + (mean firing rate during baseline period)]. Average modulation of units was tested for significance using a one sample t-test.
Sound responses were determined as significant at a given frequency if p<0.05 for a Wilcoxon rank sum test of firing rate over 250ms starting 10 ms after sound onset as compared to the same time period during interleaved trials with no tones (blank trials). A Holm-Bonferroni correction was used for multiple comparisons. Units were considered sound responsive if they responded to at least one tone frequency. Unit responses to a given frequency were averaged and these average responses were fit with a linear polynomial. RS units were included in analysis if they were sound responsive and had a linear fit with r 2 > 0.25. Slope significance was determined using a 95% confidence interval for the linear fit, slopes were considered significantly modulated either divisively or multiplicatively if the upper bound was <1 or the lower bound was >1 respectively. Intercept significance was determined using a 95% confidence interval for the linear fit, intercepts were considered significantly modulated in either an additive or subtractive fashion where lower bound was >0 or the upper bound was <0 respectively . The discriminability index, d′, was calculated for the average of every LED modulated tone response as (mean Spikessound − mean Spikesspontaneous)/√ [0.5 × (σ 2 sound + σ 2 spontaneous)]. Tone responses for a given unit were excluded if their tone response versus spontaneous firing rate z-score was <2. The d′ values are presented as the mean of d′ values for a given unit. To generate a frequency tuning curves, individual unit responses were average at each frequency. The responses were then centered to the best frequency (BF) chosen as the frequency which had the strongest tone response in the control condition for each unit. Significant modulation at each frequency by cortical inactivation was determined using a paired t-test followed by a Holm-Bonferroni correction for multiple comparisons.