Synaptic Mechanisms of Top-Down Control in the Non-Lemniscal Inferior Colliculus

Corticofugal projections to evolutionarily ancient, sub-cortical structures are ubiquitous across mammalian sensory systems. These “descending” pathways enable the neocortex to control ascending sensory representations in a predictive or feedback manner, but the underlying cellular mechanisms are poorly understood. Here we combine optogenetic approaches with in vivo and in vitro patch-clamp electrophysiology to study the projection from auditory cortex to the inferior colliculus (IC), a major descending auditory pathway that controls IC neuron feature selectivity, plasticity and auditory perceptual learning. Although individual auditory cortico-collicular synapses were generally weak, IC neurons often integrated inputs from multiple corticofugal axons that generated reliable, tonic depolarizations even during prolonged presynaptic activity. Latency measurements in vivo showed that descending signals reach the IC within 30 ms of sound onset, which in IC neurons corresponded to the peak of synaptic depolarizations evoked by short sounds. Activating ascending and descending pathways at latencies expected in vivo caused a NMDA receptor dependent, supra-linear EPSP summation, indicating that descending signals can non-linearly amplify IC neurons’ moment-to-moment acoustic responses. Our results shed light upon the synaptic bases of descending sensory control, and imply that heterosynaptic cooperativity contributes to the auditory cortico-collicular pathway’s role in plasticity and perceptual learning.


Introduction 29
The auditory system is organized as a network of feedback loops, such that most central auditory 30 nuclei receive descending projections from higher levels of the processing hierarchy ( Suthakar and Ryugo, 2017). The auditory cortex is a major 33 source of excitatory (glutamatergic) descending projections, with the density of descending fibers 34 often rivaling that of ascending fiber tracts (Winer et al., 2001;Winer, 2006;Stebbings et al., 2014). 35 These corticofugal projections likely play a major role in hearing by providing an anatomical 36 substrate for "top-down" information to control early acoustic processing. Indeed, stimulating or 37 silencing the auditory cortex in vivo changes spontaneous and sound-evoked activity throughout 38 the central auditory system (Massopust and Ordy, 1962;Ryugo and Weinberger, 1976 2002; Stuart and Spruston, 2015), addressing these knowledge gaps is necessary to understand 47 how the auditory cortex exerts control over early auditory processing.

49
Of particular interest is the descending projection from auditory cortex to the inferior colliculus (IC), 50 a midbrain hub important for sound localization, speech perception, and an early site of divergence 51 for primary and higher-order auditory pathways (Masterton et al., 1968 Saldaña et al., 1996). Thus, auditory cortico-collicular 63 synapses seem uniquely positioned to modulate acoustic signals destined for limbic circuits 64 supporting learned valence and habit formation; this prediction is further supported by the fact that 65 chemical ablation of auditory cortico-collicular neurons selectively impairs certain forms of auditory 66 perceptual learning while sparing the performance of previously learned task associations (Bajo et 67 al., 2010). Nevertheless, little is known regarding how auditory cortico-collicular synapses control 68 activity in single IC neurons. Intriguingly, auditory cortex inactivation typically does not abolish IC 69 neuron sound responses, but rather causes divisive, non-monotonic changes in receptive field 70 properties and feature selectivity (Yan and Suga, 1999;Nakamoto et al., 2008Nakamoto et al., , 2010; Anderson and 71 4 ± 1.5 ms ( Figure 1F). Interestingly, EPSP amplitudes varied over 2 orders of magnitude across 116 different cells and were occasionally large enough to drive IC neurons beyond spike threshold 117 (Figure 1 -Figure Supplement 3A). These data indicate that bulk activation of corticofugal neurons 118 triggers potent EPSPs in superficial IC neurons. However, under behaviorally relevant conditions, 119 the extent of synaptic depolarization provided by descending inputs will depend on the rate and 120 synchrony of corticofugal neuron firing. 121 122 Figure 1: Biophysical properties of auditory cortico-collicular synapses A) Cartoon of experiment. In vivo whole-cell recordings are obtained from IC neurons 2-4 weeks following Chronos injections; an optic fiber is positioned above the auditory cortex. B) Example EPSPs following in vivo optogenetic stimulation. Gray traces are single trials; magenta is average. Inset is the EPSP rising phase at a faster timebase, arrow denotes light onset. C) Dorsal-ventral locations (relative to dura) for IC neurons where auditory cortical stimulation did (magenta) and did not (gray) evoke an EPSP. D-F) Summary of EPSP onset (D) amplitude (E), and half-width (F). G) whole-cell recordings obtained from dorso-medial shell IC neurons in vitro. H) EPSPs evoked by in vitro optogenetic stimulation (2 ms light flash). Inset: EPSP rising phase. I-K) Histograms of EPSP onset (I), amplitudes (J), and half-widths (K) in vitro.
Because the auditory cortex projects to many sub-cortical targets besides the IC, in vivo stimulation 123 could drive polysynaptic excitation onto IC neurons that would complicate estimates of 124 monosynaptic connectivity. We thus prepared acute IC brain slices from mice injected with Chronos 125 in auditory cortex to quantify the functional properties of descending synapses in a more controlled 126 setting. We targeted whole-cell current-clamp recordings specifically to neurons in the dorso-medial 127 shell IC, as this region shows the highest density of corticofugal axons (Song et al., 2018; our 128 Figure 1 - Figure Supplement 1,2). Stimulating auditory cortico-collicular axons via single blue light 129 flashes delivered through the microscope objective (1-10 ms duration) drove EPSPs in n = 78 130 neurons from N = 40 mice ( Figure 1G,H). EPSPs had short-latency onsets following photo-131 stimulation ( Figure 1I; 3.0 ± 0.1 ms), indicating a monosynaptic rather than polysynaptic origin. 132 EPSPs in vitro had a similar range and mean peak amplitude as those recorded in vivo ( Figure 1J; 133 2.97 ± 0.35 mV, p=0.3, Mann-Whitney test), and similarly could drive spikes in a subset of 134 recordings (Figure 1 -Figure Supplement 3B). Although the EPSP half-width was significantly 135 slower in vitro compared to in vivo ( Figure 1K; 39.6±2.3 ms, p<0.001, rank-sum test), this result is 136 not surprising: The constant barrage of synaptic inputs in vivo is expected to generate a "high 137 conductance state" that accelerates the membrane time constant (Destexhe et al., 2003).

139
In a separate set of experiments (n = 24 neurons from N = 10 mice), we quantified the kinetics of 140 auditory cortical excitatory postsynaptic currents (EPSCs) as they appear at the soma using  EPSP amplitudes spanned two orders of magnitude under our conditions ( Figure 1E,J). Does this 151 variability reflect a differential potency of individual synapses, or alternatively, differences in the 152 number of presynaptic auditory cortical axons impinging onto individual shell IC neurons? We first 153 estimated unitary EPSP amplitudes using a minimal stimulation paradigm designed to activate one 154 (or very few) auditory cortico-collicular fibers. In these experiments, the LED intensity was titrated to 155 the minimum power required for optogenetic responses to fluctuate between successful EPSPs and 156 failures on a trial-by-trial basis (Figure 2A,B; mean failure rate across experiments: 44 ± 3%, n = 18 157 cells from N = 14 mice). The mean amplitude of successful EPSPs was generally small (0.84 ± 0.09 158 mV; Figure 2C) and similar to previous reports of unitary synapses between layer 5 pyramidal 159 neurons in sensory cortex (Brown and Hestrin, 2009;Lefort et al., 2009). By contrast, progressively 160 stronger LED flashes increased the amplitude of successful EPSPs compared to the minimal 161 stimulation condition in many cells tested (n=15 cells from N = 11 mice; Figure 2D, Figure 2F), indicating that the EPSP amplitude variability across individual 165 6 neurons likely reflects the number of presynaptic auditory cortical fibers recruited during stimulation 166 rather than differences in unitary strength. Importantly, the EPSP half-width was constant across 167 the range of stimulus intensities ( Figure 2D, inset. EPSP halfwidth ratios at maximal and threshold 168 LED intensities: 1.04 ±0.07). Similar results were obtained in voltage clamp recordings, with EPSC 169 rise times and half-widths being similar at threshold and maximal LED intensities (Figure 2 - Figure  170 Supplement 1). Together, these results indicate that increased LED intensities recruit more axons 171 rather than prolonging Chronos activation and temporally dispersing vesicle release from single 172 presynaptic boutons. We thus conclude that although the strength of individual auditory cortico-173 collicular synapses is weak, the convergence of multiple presynaptic fibers ensures that 174 descending signals will substantially increase shell IC neuron excitability.   ( Figure 3A,B) and qualitatively similar firing patterns as the shell IC neurons recorded in rat slices 186 by Smith (1992). However, 18 of these "adapting" neurons responded to negative current with a 187 sustained hyperpolarization ( Figure 3A), whereas 20 displayed a prominent Ih-like "sag" likely 188 mediated by HCN channels ( Figure 3B, middle trace). These data suggest a minimum of two shell 189 IC neuron sub-types with adapting firing patterns, which can be differentiated based on the extent 190 of Ih sag. By contrast, other neurons had delayed first spikes (9/64, Figure 3C) or showed non-191 adapting discharge patterns (15/64, Figure 3D). Finally, 2/64 cells displayed a strikingly distinct 192 phenotype, with a burst of spikes riding atop a "hump"-like depolarization similar to neurons 193 expressing T-type Ca 2+ channels ( Figure 3E). EPSP amplitude and half-width of were similar 194

Temporal integration of auditory cortical inputs is moderately sub-linear 199
Auditory cortico-collicular neurons in awake mice respond to acoustic stimuli with ~10-50 Hz spike 200 trains . Given the kinetics of auditory cortico-collicular EPSPs ( Figure  201 1), these spike rates are expected to result in significant temporal summation of descending 202 signals. However, the use-dependent dynamics neurotransmitter release, as well as postsynaptic 203 ion channels, can enforce sub-or supra-linear summation that effectively dictates the temporal sustained cortical activity? We addressed this question by repetitively stimulating auditory cortico-206 collicular axons (10 light pulses at 20 or 50 Hz, 2 ms pulse width; Figure 4A,B, black traces; n = 12 207 neurons from N = 7 mice). We quantified temporal integration by comparing the peak EPSP 208 amplitudes observed after each light flash in the train to the amplitudes expected from the linear 209 summation of a single auditory cortico-collicular EPSP recorded in the same neuron ( Figure 4A,B; 210 magenta traces). The observed peak amplitude of the 10th EPSP in the train reached 92 ± 13 % 211 and 77 ± 16% of that expected from linear summation at 20 and 50 Hz, respectively ( Figure 4C,D).

212
These results argue that at the population-level, auditory cortical firing rates are read out as 213 moderately sub-linear shifts of the membrane potential towards threshold. These sub-linear effects 214 were likely due to frequency-dependent synaptic depression, as the trough-to-peak amplitude of 215 individual EPSPs during train stimulation showed greater reduction at faster rates ( respectively. p<0.001, sign-rank test).

219
Auditory cortico-collicular neurons display elevated firing rates at ~20 Hz for the entirety of long- value of 0. n = 10 cells from n = 7 mice). Importantly, these effects were not an artifact of directly 227 stimulating auditory cortico-collicular nerve terminals in brain slices: A similar tonic depolarization 228 was observed in superficial IC neurons recorded in vivo using an optic fiber positioned over 229 auditory cortex ( Figure 4G,H; mean amplitude of DC component: 1.5 ± 0.7 mV, n=19 cells from n = 230 6 mice, p=0.53 compared to in vitro data, rank-sum test).

232
In addition, cessation of cortical stimulation in vivo also caused a long-lasting after-233 hyperpolarization (AHP) in 14/19 cells: The membrane potential rapidly fell below baseline after the 234 last stimulus (t = 76.5 ± 1.5 ms) and recovered over several seconds (t = 2.6 ± 0.6 s; Figure 4 -235 Figure Supplement 2). Interestingly, this AHP was independent of postsynaptic spiking and thus 236 may reflect buildup of feed-forward inhibition from local and long-range sources, or alternatively, a 237 transient cessation of tonic descending excitation. Altogether, these experiments show that auditory 238 cortico-collicular synapses can sustain transmission on seconds time scales via tonic and phasic 239 excitation. In addition, the profound AHP following in vivo stimulation suggests that IC neuron 240 . G,H) Same as E and F but during in vivo recordings. Of note is the large AHP. excitability is bi-directionally yoked to auditory cortical firing patterns. Thus, increases as well as 241 pauses in auditory cortico-collicular neuron activity may be comparably significant to IC neurons.

243
NMDA receptors contribute to temporal integration of descending signals 244 Central excitatory transmission is predominantly mediated by AMPA and NMDA type glutamate 245 receptors, with NMDA receptors being particularly crucial for dendritic integration, associative 246 plasticity, and learning. The auditory cortico-collicular pathway is involved in perceptual learning 247 following monaural hearing loss (Bajo et al., 2010), and NMDA receptors in the avian IC shell 248 homologue preferentially contribute to receptive fields generated by experience-dependent 249 plasticity (Feldman et al., 1996). We thus asked to what extent auditory cortico-collicular synapses 250 activate NMDA receptors in shell IC neurons. Interestingly, bath application of the AMPA/kainate depolarizations during sustained cortical activity. 285 Figure 5. Pharmacology of descending transmission. A) EPSPs (mean ± SEM) from a single neuron before (black) and after (magenta) bath application of NBQX (10 µM). Of note is the absence of residual synaptic depolarization after NBQX. B) Summary data. For the effect of NBQX on descending EPSPs. C) EPSPs before and after R-CPP (black and magenta traces, respectively). D, E) Group data for the effect of R-CPP on EPSP peak amplitude (D) and half-width (E). Asterisks denote statistical significance. F) Upper panel: EPSPs (average ± SEM) evoked by 10 light flashes at 50 Hz. Black and magenta are in control and R-CPP. Lower panel is the cumulative integral of the waveforms. G) Group data showing amplitude of each EPSP in the train (mean ± SEM) before and after R-CPP (black and magenta, respectively). H) Group data for the effect of R-CPP on voltage integral as in F. Asterisks denote significance. Differential contribution of NMDA receptors at ascending and descending synapses 286 The modest contribution of NMDA receptors at descending synapses is surprising, as previous 287 studies suggest a rather prominent NMDA component at excitatory synapses in the IC (Smith,288 1992; Wu, 2004). We thus wondered if our results reflected a relative paucity of NMDA receptors at 289 all excitatory synapses onto shell IC neurons, or rather a unique feature of auditory cortico-collicular 290 synapses. Shell IC neurons receive a prominent intra-collicular projection from the central IC which 291 likely transmits a significant amount of ascending acoustic information (Saldaña et al., 1996;Sun 292 and Wu, 2009). We tested if the NMDA component differed between ascending and descending 293 EPSPs in the same neuron using a dual pathway stimulation approach in vitro: A bipolar stimulating 294 electrode was positioned in the central IC and Chronos was expressed in auditory cortex to activate 295 ascending and descending synapses, respectively ( Figure 6A). 2.5-5 µM gabazine was present in 296 all experiments to isolate excitatory transmission. Repetitive stimulation of either pathway (5x, 50 297 Hz) led to summating EPSPs which were differentially sensitive to NMDA receptor blockade ( Figure  298 6B-C): 5 µM R-CPP reduced the cumulative integral of central IC and auditory cortical EPSPs to 299 44±4% and 78±4% of baseline, respectively ( Figure 6D, n=9 cells from n = 8 mice, p=0.0004, 300 paired t-test), indicating that NMDA receptors contributed less at descending compared to 301 ascending synapses. These results were not solely due to greater contribution of NMDARs during 302 train stimuli at ascending compared to descending synapses. Indeed, qualitatively similar results 303 were also observed in a between cell comparison of R-CPP effects on ascending and descending 304 EPSPs evoked with single stimuli: R-CPP caused a significantly greater reduction in the peak 305 amplitude of ascending EPSPs evoked with single shocks compared to the auditory cortical EPSPs Fraction halfwidth remaining in R-CPP: 64 ± 5 vs. 82 ± 5% for central IC and auditory cortical 312 inputs, respectively. p = 0.024, paired t-test). Thus, the extent of NMDA receptor contribution to 313 13 descending transmission reflects a synapse-specific property of auditory cortico-collicular inputs 314 rather than the global distribution of NMDARs in shell IC neurons.

316
Auditory cortical inputs are predicted to arrive at the peak of EPSPs evoked by transient sounds 317 Layer 5 auditory cortico-collicular neurons in awake mice respond to sound with a mean first-spike 318 latency of ~21 ms . This value is surprisingly shorter than reported 319 first spike latencies in shell IC neurons (~35 ms; Lumani and Zhang, 2010). Whether auditory 320 cortical excitation arrives before, during, or after the onset of sound-evoked EPSPs in shell IC 321 neurons effectively determines how ascending and descending signals integrate at the single cell 322 level, but the relative timing of distinct inputs onto IC neurons is unknown. We thus quantified the 323 relative timing of ascending sound-evoked and descending cortical EPSPs using in vivo whole-cell 324 recordings from superficial IC neurons of anesthetized mice. We first determined the onset latency 325 of descending EPSPs in IC neurons using electrical stimulation of the auditory cortex ( Figure 7A). 326 We employed electrical, rather than optogenetic stimulation for these experiments because spike 327 onset following optogenetic stimulation is limited by the cell's membrane time constant and effective 328 spike threshold, whereas electrical stimulation bypasses somato-dendritic depolarization by directly 329 triggering axonal spikes. Single shocks delivered to the auditory cortex evoked EPSPs with an 330 onset latency of 5.4 ± 0.6 ms (n = 7 cells from n = 4 mice; mean depth of recorded neurons 188 ± 331 14 14 µm, Figure 7B,C), indicating that descending information reaches IC neurons within a few ms of 332 AP initiation in auditory cortex. Since the mean first-spike latency of auditory cortico-collicular 333 neurons is ~21 ms  and the synaptic latency is 5-8 ms ( Figure 7B,C), 334 these data collectively argue that that cortical feedback begins to excite IC neurons <30 ms after 335 sound onset. Furthermore, assuming an axon path length of ~8 mm from auditory cortex to shell IC 336 ( Our latency measurements (Figure 7) suggest that ascending information is rapidly followed by 377 descending cortical excitation. This temporal overlap is intriguing because ascending and 378 descending synapses express NMDA receptors (Figure 6), which in other cell-types, enable 379 cooperative interactions between co-active pathways onto the same neuron (Takahashi and  380 Magee, 2009). We thus hypothesized that appropriately timed cortical feedback might integrate 381 non-linearly with ascending inputs from the central IC, thereby generating a synaptic depolarization 382 larger than expected from the sum of either pathway active in isolation. We tested this idea in vitro 383 using our dual pathway stimulation approach (Figure 6) while recording from shell IC neurons. We 384 first recorded the synaptic depolarization following stimulation of ascending and descending 385 pathways in isolation (5 stimuli at 50 Hz; Figure 8A, upper traces). We next simultaneously 386 activated the two pathways such that the onset of cortical EPSPs collided with the peak of 387 ascending EPSPs, as predicted from our in vivo latency measurements (Figure 7). The observed 388 depolarization during synchronous pathway activation was on average significantly larger than 389 expected from the arithmetic sum of each pathway stimulated alone ( Figure 8A; n=12 cells from 390 N=9 mice. Cumulative integral observed: 1.90 ± 0.35 mV*ms, expected: 1.45±0.22 mV*ms, 391 p=0.007, paired t-test test), indicating that coincident activity of ascending and descending 392 pathways summates supra-linearly. Importantly, R-CPP mostly abolished this supra-linear 393 summation, such that the depolarization during coincident activation in R-CPP now equaled the 394 expected sum of each pathway activated alone ( Figure 8B,C, lower traces; n=7 cells from N = 7 395 mice. Observed/Expected control: 1.39±0.05; in R-CPP: 1.02±0.04, p=0.0156, sign-rank test). 396 Thus, synaptic NMDA receptors provide a supra-linear boost when descending excitation follows 397 Magenta is the depolarization expected from the arithmetic sum of the waveforms following stimulation of either pathway alone (e.g., left and middle traces). Of note, the observed depolarization under control conditions is larger than expected from linear summation; blocking NMDA receptors linearizes pathway integration (compare black and magenta traces in R-CPP). B) Group data plotting the ratio of observed and expected peak amplitude for each of 5 EPSPs in a 50 Hz train during synchronous activation of central IC and auditory cortical synapses. Asterisks denote statistical significance of Bonferroni post-hoc test for the fourth and fifth stimuli following a main effect of drug condition (p=0.038, F(1,9), Two-way repeated measures ANOVA). C) Group data, observed over expected ratio of the cumulative integral during combined pathway activation in control conditions and in the presence of R-CPP. Asterisks denote statistical significance (sign-rank test). ascending activity, thereby promoting the non-linear mixing of distinct pathways in single IC 398 neurons. 399 400 Discussion 401 402 We have shown that the majority of neurons in the superficial (shell) IC layers receive reasonably 403 strong excitation from auditory cortex that is predicted to arrive within 30 ms following sound onset. 404 Given the relatively long latency of sound-evoked PSPs in shell IC neurons (our Figure  However, these newly identified corticofugal GABAergic synapses have low release probability, 445 signal mainly via "spillover" transmission (Isaacson et al., 1993;Szabadics et al., 2007) neuron's multiple dendrites, thereby enabling cortical signals to non-linearly control ascending 507 information irrespective of the spatial relationship of co-active inputs. However, further studies are 508 necessary to identify the precise anatomical relationship between ascending and descending 509 synapses in single IC neurons. Finally, we did not observe any overt correlation between the 510 strength of descending EPSPs and the diverse biophysical properties of shell IC neurons. However, 511 an important consideration is that our metrics of neuronal diversity perhaps do not reflect explicit Implications for predictive control of tectal activity, synaptic plasticity, and perceptual learning. 518 The properties of corticofugal synapses could enable a context-dependent modulation of IC 519 neurons across multiple timescales. Indeed, the rapid onset of descending EPSPs following cortical 520 spikes (~5-8 ms) is more than twice as fast as sensory-evoked cortical gamma rhythms (30-50 Hz). 521 Thus, descending signals could effectively synchronize neural ensembles across the ascending 522 auditory hierarchy either to the temporal envelope of sound (Weible et al., 2020) or to internally 523 generated rhythms. Alternatively, rapid auditory cortico-collicular transmission may be particularly 524 advantageous in driving innate behaviors in response to sound. Indeed, auditory cortico-collicular 525 neuron activity, either via optogenetic stimulation or loud sounds, directly triggers escape and flight 526 behaviors in mice; these effects likely occur due via a descending activation of shell IC neurons 527 projecting to the PAG (Xiong et al., 2015). As such, cortical signals could potentially trigger 528 evolutionarily conserved motor programs to benefit survival. 529 19 530 We also found that descending synapses sustained transmission and drove tonic depolarizations 531 even during seconds-long activity patterns, such that IC neurons may also integrate slower cortical 532 state fluctuations. Intriguingly, auditory cortical neurons in behaving animals show enhanced firing 533 rates during the delay period of auditory working memory tasks (Gottlieb et al., 1989) which 534 apparently, precedes similar activity patterns in prefrontal cortex (Huang et al., 2016). If these 535 working memory related neuronal ensembles include auditory cortico-collicular neurons, sustained 536 transmission from descending synapses could cause seconds-long increases in IC neuron 537 excitability based on working memory content. Accordingly, persistent delay period activity is 538 observed in ~10% of IC neurons when rats engaged in an auditory working memory task (Sakurai, 539 1990), although future studies are necessary to determine the extent to which this activity is 540 inherited from descending auditory cortical pathways.

542
Several studies now show that layer 5 corticofugal pyramidal neurons are necessary for perceptual 543 learning in multiple different sensory tasks. Optogenetic inhibition of layer 5 pyramidal neurons in 544 somatosensory cortex prevents behavioral adaptation following cue-related changes in a tactile 545 detection task, although the same manipulation had no effect on touch perception (Ranganathan et 546 al., 2018). Similarly, lesioning visual corticostriatal neurons prevents acquisition, but not 547 performance of a visual detection task (Ruediger and Scanziani, 2020). In the auditory system, 548 chemical lesions of auditory cortico-collicular neurons prevent the experience-dependent recovery 549 of sound localization following monaural hearing loss (Bajo et al., 2010), although auditory cortex 550 becomes dispensable once animals have learned to localize sounds using monaural cues (Bajo et  551 al., 2019). Thus, although necessary for perceptual learning, corticofugal synapses may not be the 552 primary locus of experience-dependent plasticity. Indeed, classic studies in barn owls suggest that 553 ascending central IC -> external (shell) IC synapses are the first site of experience-dependent, 554 spatial map plasticity in the auditory system (Brainard and Knudsen, 1993). In tandem with our 555 current study, these results suggest that auditory cortico-collicular synapses' contributions to 556 perceptual learning may not lie in their explicit ability to undergo classical Hebbian associative 557 plasticity, but rather as permissive forces of heterosynaptic plasticity at ascending synapses. 558 559 560 Methods 561 562 Surgery for viral injections: All experiments were approved by the University of Michigan's IACUC 563 and performed in accordance with NIH's Guide for the care and use of laboratory animals. All 564 surgical procedures were performed under aseptic conditions. Surgeries were performed on 4-7 565 week old male or female C57BL6/J mice purchased from Jackson Labs or offspring of CBA x 566 C57BL6/J matings bred in house for electrophysiology experiments. was subsequently lowered to 1-2% to maintain a deep anesthetic plane, as assessed by the 571 absence of paw withdrawal reflex and stable respiration (1-1.5 breaths/s). Body temperature was 572 maintained near 37-38° C using a feedback controlled, homeothermic heating blanket (Harvard 573 20 Apparatus). Mice were administered 5 mg/kg carprofen after induction as a pre-surgical analgesic. 574 The scalp was clear of hair, swabbed with betadine, and a small incision was made in the skin 575 overlying the left hemisphere. Topical 2% lidocaine was then applied to the wound margins. The 576 stereotaxic frame was rotated ~50 degrees, allowing a vertical approach perpendicular to the layers 577 of auditory cortex. A 200-400 µm craniotomy was carefully opened over the left auditory cortex (-578 2.75 mm from Bregma, centered on the lateral ridge) using a 0.5 mm diameter dental burr  05, Fine Science Tools) and Foredom microdrill. The skull was frequently irrigated with chilled 580 phosphate buffered saline (PBS) to prevent overheating during drilling. Following the craniotomy, a 581 glass pipette (0.1-0.2 mm diameter at the tip) containing the pAAV-Syn-Chronos-GFP (Addgene 582 #59170-AAV1) or AAV1-hSyn-Cre (Addgene #105553-AAV1) virus penetrated the auditory cortex 583 at a rate of <10 µm/s using a motorized micromanipulator. A total of 100-200 nL virus was injected 584 at 2-4 sites 810 and 710 µm below the pial surface (25-50 nL per site). Following injections, the 585 pipette was maintained in place for an additional 5 min before slowly retracting at a rate of <10 586 µm/s. At the end of the surgery, the craniotomy was filled with bone wax, the skin was sutured, and 587 the mouse was removed from the stereotax. Immediately following surgery, mice were given an 588 analgesic injection of buprenorphine (0.03 mg/kg, s.c.) and allowed to recover on a heating pad 589 before returning to their home cage. An additional post-operative dose of carprofen was 590 administered 24 hours following surgery.

592
In vivo electrophysiology: 2-4 weeks following viral injections, mice were deeply anesthetized with 593 isoflurane and mounted in a stereotax as described above. The skin overlying the skull was 594 removed, the left temporal muscle was retracted, the stereotax was rotated ~50 degrees, and a 2-595 2.5 mm craniotomy was carefully opened over the left auditory cortex. For optogenetic stimulation 596 in Figure 1, the dura overlying the auditory cortex was left intact and a cranial window was 597 implanted over the exposed brain with cyanoacrylate glue and dental cement. For the electrical 598 stimulation experiments in Figure 7, a small slit was carefully made in the dura and the craniotomy 599 was subsequently sealed with silicone elastomer. The stereotaxic frame was returned to the 600 horizontal position and a custom titanium headbar was affixed to the skull with dental cement. A 601 300-500 µm craniotomy was opened over the left IC and filled with a silicone elastomer plug. The 602 mouse was then removed from the stereotax, anesthetized with urethane (1.5 g/kg, i.p.), and head-603 fixed in a custom-made sound attenuation chamber. Body temperature during the experiment was 604 maintained at 37-38° C with a custom designed, feedback-controlled heating blanket. For 605 optogenetic stimulation, a 0.5 NA, 400 µm core optic fiber (Thorlabs M45L02) coupled to a 470 nm 606 LED (Thorlabs M470F3) was mounted on a micromanipulator and positioned <1 mm away from the 607 auditory cortex cranial window. For electrical stimulation experiments, the silicone plug over 608 auditory cortex was removed and a bipolar platinum-iridium electrode (FHC 30210) was carefully 609 inserted ~800 µm into auditory cortex at an angle roughly perpendicular to the cortical layers. 610 Electrical stimuli were delivered via a custom stimulus isolator designed in house. Sound clicks (0.2 611 ms duration) were presented at ~91 dB peak equivalent SPL via a free-field speaker (Peerless Industries). The hot side of the peltier chip was mounted to a 6.35 mm diameter copper rod (length: 627 ~76 mm) using thermally conductive adhesive. Fin type heatsinks were mounted to the copper rod.

628
A machined blunt copper pin (3 mm at the base, 2 mm at the tip) was mounted on the cold side 629 using the same adhesive and made contact with the dura mater over the auditory cortices. We 630 performed control experiments in anesthetized mice (n = 5 cooling attempts in N = 4 mice) to verify 631 that our devices effectively cooled the auditory cortices while minimally affecting the IC 632 (Dtemperature in IC during cortical cooling = -1.9 ± 0.6° C). The settings used during our recordings 633 bilaterally reduced the temperature of deep cortical layers to 14-17° C, which suffices to largely 634 abolish auditory cortical activity (Lomber et al., 1999;Coomber et al., 2011;Anderson and 635 Malmierca, 2013).

637
We used 8-12 week old C57Bl6/J mice for these experiments. Awake mice were handled for 3-5 638 days prior to recording and acclimated to head fixation while sitting comfortably in a Plexiglas tube. 639 Following acclimation, mice were anesthetized, a ~0.5 mm craniotomy was opened over the left IC, 640 and 2 mm craniotomies were opened over the left and right auditory cortices. The craniotomies 641 were sealed with silicone elastomer and the mouse was allowed to recover for ~2 hours prior to 642 recording. For recording, the silicone plugs were removed, the copper pin of the peltier devices 643 were positioned in contact with the left and right auditory cortices, and the craniotomies were 644 covered with 3-4% agar in saline. A saline filled glass electrode (~1-2 MOhm open tip resistance) 645 was lowered into the superficial IC (~200 µm from surface) to record click-evoked field potentials 646 and multi-unit clusters before, during, and after cooling of auditory cortices. After the recording, the 647 craniotomies were sealed with silicone and the mouse was returned to its homecage. were then incubated at 34° C in a holding chamber filled ACSF for 25-30 min and subsequently 657 stored at room temperature. Experiments were generally performed within 3-4 hours following slice 658 preparation. Following incubation, a slice was transferred to a recording chamber and held in place 659 with single strands of unwaxed dental floss tightly strung around a platinum "harp". The slice was 660 continuously perfused with oxygenated ACSF heated to 32-34° C (2-4 mL/min; chamber volume: ~ 661 22 1 mL). 2-5 µM SR95531 was added to the ACSF to block GABAA receptors in most experiments of 662 Figure 5C-H, Figure 6, Figure 6 - Figure Supplement 1, Figure 8, all voltage-clamp experiments, 663 and some experiments of Figure 1 vitro. In these cases, APs were digitally removed prior to averaging the traces by linearly 702 interpolating 0.1-0.2 ms of datapoints after the membrane potential crossed spike threshold (~20 703 mV/ms). Shock artifacts during electrical stimulation experiments were similarly removed via linear 704 23 interpolation. In summary plots, black symbols are individual cells, magenta is mean ± SEM, and 705 lines connect data from the same recording unless otherwise stated.

707
The expected linear waveforms for temporal summation experiments were calculated as follows. 708 The average waveform of a single optogenetically evoked EPSP was peak normalized to the first 709 EPSP in the recorded 20 or 50 Hz train from the same cell. We subsequently convolved the single 710 EPSP waveform with a 20 or 50 Hz binary pulse train using the Matlab function convr(). We then 711 calculated the peak amplitude ratios for each EPSP in observed and expected trains. 712 713 For the free-field sound presentation experiments of Figure 7D-E, we limited our analyses to 714 superficial IC neurons that showed onset EPSPs in response to clicks. Other IC neurons 715 encountered during these experiments showed either sound-evoked IPSPs (n=7 cells from N=5 716 mice) or IPSPs followed by rebound depolarizations (n=6 cells from N=5 mice); analyses of these 717 data will be presented in a separate report. Click-evoked field potentials in Figure 7 - Figure  718 Supplement 1 and 2 were analyzed after baseline subtraction and low-pass filtering the records at 719 500 Hz. Multi-unit activity was detected as threshold crossings after applying a bandpass filter 720 (between 300 Hz and 5 kHz) to the recordings. Peristimulus time histograms (PSTHs) of spike 721 activity were generated with 100 µs bin-widths and smoothed with a 2 ms sliding window. Onset 722 latencies of field potentials and multi-unit PSTHs were defined as the time following click onset at 723 which the data reached 20% of its peak. Average LFP waveforms and PSTHs were composed of 724 300-302 trials per condition.

726
In dual pathway experiments of Figure 8, the onset latency of ascending and descending EPSPs 727 varied across cells. Thus, the relative timing of electrical and optogenetic stimulation during 728 combined pathway activation was calculated online and on a cell-by-cell basis, such that the onset 729 of descending auditory cortical EPSPs collided with the peak of ascending EPSPs from central IC 730 as predicted from our in vivo latency measurements (Figure 7; range of Dt between stimulation of 731 descending and ascending synapses: -1.3 to 16.4 ms). These stimulation parameters were held 732 constant across control and R-CPP conditions for each cell. The expected linear summation was 733 calculated by digitally summing the average synaptic waveforms following stimulation of either 734 pathway alone, accounting for the temporal offset employed during synchronous pathway 735 activation.

737
Histology and Confocal Imaging: Mice were deeply anesthetized in a glass induction chamber 738 circulated with 4.2 mL isoflurane and transcardially perfused with ~80-100 mL of PBS followed by 739 ~80-100 mL of 10% buffered formalin (Fisher Scientific catalog # 23-245684). Brains were carefully 740 removed, stored in 10% formalin and protected from light for 24 hours. Subsequently, brains were 741 stored in PBS for up to 72 hours and 100 µm thick coronal slices were cut using a ceramic blade 742 (Cadence Endurium) and a Leica VT1000s vibratome, mounted onto slides and coverslipped using 743 Fluoromount, then protected from light and allowed to dry at room temperature for ~12-24 hours. 744 Slides were then stored at 4 C until ready for use. Images were collected using a Leica TCS SP8 745 laser scanning confocal microscope equipped with a 10x objective. 746 747  Top trace shows the raw extracellular waveform obtained following a click sound (arrow). Middle and lower traces are the LFP and multi-unit spike activity obtained following digital filtering (see Methods). C) Example average LFP traces before (black), during (magenta), and after (gray) auditory cortical cooling. Inset shows the rising phase of the LFP at a faster time base to highlight the decrease in slope during cooling. Data are from a different experiment as the one shown in B. Arrow denotes click onset. D) Multi-unit PSTHs following click onset (arrow) before, during and after auditory cortical cooling. Data are from the same recording as that of panel C. Of note is that the increase in multi-unit spiking is reversibly delayed by cooling, suggesting that auditory cortex indeed influences first spike latencies in the IC. E,F) Group data quantifying the effect of auditory cortical cooling on LFP slope (E) and onset of multi-unit spiking (F). Asterisks denote statistical significance (Dunnett's test following one-way repeated measures ANOVA, p=0.006 and 0.0016 for E and F, respectively).