Corticocortical signaling drives activity in a downstream area rapidly and scalably

How effectively does activity in an upstream cortical area drive activity in a downstream area? To address this, we combined optogenetic photostimulation with multi-unit electrophysiology to study a parietofrontal corticocortical pathway from retrosplenial cortex to posterior secondary motor cortex in mice. Photostimulation in the upstream area produced local activity that decayed quickly. This activity in turn drove downstream activity that arrived rapidly (5-10 ms latencies), and scaled in amplitude across a wide range of stimulus parameters as an approximately constant fraction (~0.2) of the upstream activity. A model-based analysis could explain the corticocortically driven activity with exponentially decaying kernels (~20 ms time constant) and small delay. Reverse (antidromic) driving was similarly robust. The results show that corticocortical signaling in this pathway drives downstream activity in a mostly linear manner. The regular and predictable responses further suggest that precise stimulation driven control of cortical population activity should be possible.

predictable responses further suggest that precise stimulation driven control of cortical 24 population activity should be possible. 25

INTRODUCTION 27
Corticocortical pathways support inter-areal communication, which is central to behavior 28 (Felleman and Van Essen, 1991;Misic and Sporns, 2016). Connectomics studies in both humans 29 and animal models have identified a structural basis for many corticocortical pathways (Oh et  optogenetic mapping studies in rodents have begun to characterize dynamic signaling at the 32 mesoscopic scale (Lim et al., 2012). However, the functional properties of inter-areal signaling 33 in these pathways have been challenging to resolve, particularly for higher-order pathways that 34 are many synapses removed from the sensory periphery and thus difficult to activate in a 35 spatiotemporally precise manner with natural stimuli. Extracellular electrical stimulation has 36 been used in efforts to artificially generate focal activity, but is inherently limited due to its 37 nonspecificity, antidromic activation, and other issues (Nowak and Bullier, 1998;Histed et al., 38 2009). More work is needed to understand the dynamics of corticocortical signaling. 39 modulate activity driven by other inputs and have very little impact on its own (Sherman and 69 Guillery, 2011). These questions again speak to ways of theorizing about neural computation. 70 Here we sought to answer these questions by developing an approach for assessing and 71 manipulating corticocortical circuit dynamics in the intact brain. We used stereotaxic viral 72 injections to express ChR2 in presynaptic RSC neurons (Yamawaki et al., 2016), and developed 73 in vivo methods in the anesthetized mouse for sampling photo-evoked multi-unit activity in M2 74 driven by RSC photostimulation. Duplication of the setup to permit both stimulation and 75 recording at both ends of the RSC→M2 projection allowed a detailed parametric characterization 76 of both local (upstream) and downstream activity evoked both ortho-and antidromically. This 77 allowed us to carefully measure the influence inter-areal signaling as a function of stimulation 78 amplitude, duration, and the area being stimulated. 79

81
To investigate corticocortical signaling in the RSC→M2 pathway, we used viral methods to label 82 neurons with ChR2, optical fibers to photostimulate them, and linear arrays to record the evoked 83 activity. Similar to previous studies of this pathway (Yamawaki et al., 2016), we infected 84 neurons in RSC with an AAV encoding ChR2 and a fluorescent protein (Fig. 1A,B). After a 85 recovery period of several weeks, animals were anesthetized with ketamine and underwent 86 placement of photostimulation fibers and silicon probes in the RSC and M2 (Fig. 1C). 87 88

RSC photostimulation drives downstream M2 activity 89
To understand how RSC affects M2 activity, we photostimulated in RSC and measured multi-90 unit activity in M2 ( Fig. 2A). In single trials, activity was typically detected on multiple channels 91 ( Fig. 2B). Over repeated trials, on channels showing responses, photostimulation reliably evoked 92 spiking activity (Fig. 2C). The peristimulus time histogram shows clear stimulus triggered 93 activity (Fig. 2D, top). These histograms of M2 activity showed robust, transient increases in 94 activity starting with a short delay after the onset of photostimulation in RSC. 95 It is important to understand how the virus and the construct might affect the responses. 96 We therefore performed parallel experiments with two different AAV serotypes carrying 97 different variants of ChR2 driven by different promoters: AAV1-ChR2-Venus, carrying wild-98 type ChR2 driven by the CAG promoter, and AAV9-ChR2-eYFP, carrying ChR2 with the 99 H134R mutation driven by the CaMKII promoter (see Methods). The two viruses gave similar 100 responses (Fig. 2D), an impression that was borne out in further detailed comparisons that will 101 be presented in later sections. Our findings suggest that our strategy is not overly affected by 102 details of the virus or construct. 103 We need to be sure that the M2 responses reflect synaptically driven spikes of 104 postsynaptic M2 neurons, rather than spikes in presynaptic axons. We therefore sampled M2 105 responses before and after injecting M2 with muscimol, a GABA agonist, which suppresses 106 spiking in cortical neurons while preserving presynaptic spiking (Chapman et al., 1991; 107 Chatterjee and Callaway, 2003). As expected, muscimol injection abolished most of the activity 108 in M2 (3 of 3 animals) (Fig. 2E, top), whereas injection of saline had no effect (2 of 2 animals). 109 Thus, M2 responses are, indeed, driven by corticocortical synaptic activity. 110 We also want to be sure that our results cannot be overly affected by probe placement. In 111 earlier pilot experiments the probe was sometimes inadvertently placed slightly lateral by ~0.5-1 112 mm, resulting in recordings in M1 instead of M2. In this case we observed little or no photo-113 evoked activity (Fig. 2E, bottom), consistent with the anatomy and electroanatomy of the 114 RSC→M2 projection, which provides little or no direct input to M1 neurons (Yamawaki et al., 115 2016). Mistaken probe placement would thus simply decrease the observed activity. 116 From the results of these initial characterizations we conclude that (i) optogenetically 117 stimulating RSC drives a delayed, brief wave of spiking activity in M2; (ii) the evoked activity 118 appears to reflect mostly the properties of the corticocortical circuit itself rather than the those of 119 the viruses and/or constructs; and (iii) the M2 activity appears to arise from orthodromically 120 driven signaling along the RSC→M2 corticocortical pathway, rather than non-specific (e.g. 121 cortex-wide) activation. Next, we turned to a more in-depth characterization of the technique by 122 recording in both areas. 123 124

Comparison of local RSC and downstream M2 activity evoked by RSC photostimulation 125
To better understand signaling in the RSC→M2 circuit, we recorded from both the RSC and M2 126 during RSC photostimulation, allowing us to assess both the locally driven activity in upstream 127 RSC and the orthodromically driven activity in downstream M2 (Fig. 3A). As observed with 128 both AAV9-ChR2 ( Fig. 3B-F) and AAV1-ChR2 ( Fig. 3G-K), with RSC photostimulation the 129 activity recorded in RSC rose rapidly at the onset of photostimulation and declined rapidly as 130 well, whereas activity recorded in M2 followed with a brief latency (in ms after the RSC peak: 131 7.5 for AAV9, and 6.5 for AAV1) and rose to lower levels than observed in RSC (RSC/M2 132 amplitude ratio: 3.8 for AAV9, 4.1 for AAV1). The results of this two-probe characterization of 133 RSC photostimulation thus reveal two important aspects of corticocortical driving. First, at the 134 upstream end there is a rapid and strong decay of the local activity in the directly 135 photostimulated RSC (Fig. 3B,G). This decay is generally consistent with ChR2 desensitization 136 (Lin et al., 2009), and the greater decay observed with AAV1 is consistent with the reduced 137 desensitization of ChR2-H134R mutation (in AAV9) compared to wild-type ChR2 (in AAV1) 138 (Nagel et al., 2003;Nagel et al., 2005). Second, at the downstream end the corticocortically 139 driven activity in M2 was reduced in amplitude and slightly delayed relative to the RSC activity. 140 A caveat is that these properties might not be generalizable, reflecting instead the particular 141 photostimulus parameters used in these experiments. Therefore, we next investigated in detail the 142 stimulus dependence of the responses by exploring a wide range of stimulus intensities and 143 durations. 144

Parametric characterization of orthodromic (forward) driving 146
Key parameters for the dynamics of a circuit are the dependency on stimulus amplitude (light 147 intensity) and duration (pulse width). Stimulus trials were delivered at five different intensities 148 (20, 40, 60, 80, and 100% relative to maximum) and durations (1, 5, 10, 20, and 50 ms), with 149 random interleaving and many repetitions (typically 30 trials per experiment) for each of the 25 150 unique intensity-duration combinations (Fig. 4A). Responses on the local RSC probe and the 151 downstream M2 probe were averaged across trials as before, and the median responses were 152 determined across animals (AAV9 data shown in To better understand the responses we want to fit a simple model to the data. Visual inspection of 159 the waveforms of both the RSC and M2 responses (Fig. 4) shows roughly linear increases with 160 intensity. Clearly, activity in the photostimulated RSC decays rapidly and extensively, consistent 161 with ChR2 densensitization (as discussed above). However, in the downstream M2, it is unclear  162   how responses scale directly with upstream RSC activity; for example, do they scale linearly, or  163 show signs of adaptation? We would like a simple model to allow us to both describe and 164 interpret the data. 165 Explorative data analysis revealed that we could fit the directly stimulated (upstream) 166 area well with briefly delayed activation followed by a large and rapid decay (Fig. 5A). Hence, 167 we modeled stimulation as a time-shifted delta function divided by a linear function of the 168 integral of the stimulus history. So this first-stage model has 3 parameters for gain, delay, and the 169 steady state adaptation. These parameters seem intuitively necessary: the gain describes basic 170 physiology; the delay is needed due to the ~3 ms blanking of the stimulus artifact (see Methods), 171 but can also account for ChR2 activation kinetics; adaptation is expected from ChR2 172 inactivation/desensitization kinetics, and allows for additional factors contributing to a temporal 173 decline in activity (e.g. GABA release, synaptic depression). 174 Indeed, we found this model to produce good fits when we analyzed activity in the 175 stimulated (RSC) area. We find that the model qualitatively describes the data, describing both 176 its initial rise, and its decay over time (AAV9 data shown in Next, we assessed whether the reduced amplitude of M2 responses (compared to upstream RSC 204 activity, discussed above) was a consistent property across stimulus parameters. Plotting the 205 response amplitudes in RSC and M2 for all 25 stimulus combinations (Fig. 6A) showed that 206 these ranged widely but with a consistent relationship, substantially greater in RSC than in M2. 207 The same pattern was observed for both viruses (factor of 4.7 for AAV9 and 6.8 for AAV1 208 experiments), even though absolute response amplitudes were generally stronger for AAV9 209 compared to AAV1 (1.5-fold for RSC responses and 2.1-fold for M2 responses; p < 10 -3 , sign 210 test). Overall, the 'driving ratio', the ratio of the remotely driven activity in M2 relative to the 211 locally driven activity in RSC, was ~0.2 (Fig. 6B). In other words, activity in the downstream 212 area, M2, was generally about a fifth of that in RSC, across a wide range of stimulus parameters. 213 Of further importance to the interaction are latencies. These also showed a consistent 214 relationship, with M2 responses peaking with a short delay after RSC responses (Fig. 6C). The 215 same pattern was observed for both viruses (median latency of M2 response relative to RSC 216 response of 8 ms for AAV9 and 7 ms for AAV1 experiments). In this case, unlike the absolute 217 response amplitudes, the latencies of the responses in RSC and M2 did not differ significantly for 218 AAV9 vs AAV1 (p > 0.05, sign test). In contrast to the amplitudes, the latencies were largely 219 stimulus-independent. the same experiments we also delivered photostimuli to M2 (via a second optical fiber) as a way 243 to activate ChR2-expressing axons there (i.e., projecting from RSC) and thereby gain insight into 244 the properties of antidromic signaling in the same RSC→M2 pathways (Fig. 7A). 245 In particular, we wondered if antidromic activation would result in similar or different 246 effects compared to orthodromic activation. Photostimulation in M2 resulted in a short-latency, 247 short-duration wave of antidromically generated activity in both RSC and a similar but smaller-248 amplitude wave of locally generated activity in M2. Similar results were found for experiments 249 with AAV9 ( Fig. 7B-F) and AAV1 ( Fig. 7G-K). Neither amplitudes nor latencies differed with 250 antidromic activation for the 10-ms, 100% stimulus combination. However, across all stimulus 251 combinations the response amplitudes were overall ~2-fold greater in RSC relative to M2 (Fig.  252 7L), contrasting with the reduced amplitude in the downstream area observed with orthodromic 253 stimulation. Similar to orthodromic stimulation, absolute response amplitudes were generally 254 stronger for AAV9 compared to AAV1 (2.6-fold for RSC responses and 3.8-fold for M2 255 responses; p < 10 -3 , sign test). Latencies in the two areas were indistinguishable with AAV1 and 256 slightly delayed (by 3 ms) in M2 with AAV9 (Fig. 7M). Latencies in RSC were slightly shorter 257 with AAV9 than AAV1 (by 2.5 ms; p < 10 -4 , sign test), but those in M2 were the same with the 258 two viruses (p > 0.05, sign test). These results indicate that RSC axons forming this 259 corticocortical projection can be robustly activated in M2, generating activity both locally in M2 260 and antidromically in RSC -which is in effect the 'downstream' area in this experimental 261 configuration. 262 263

Laminar analysis 264
Lastly, we considered the laminar profile of M2 activity generated by activation of the 265 RSC→M2 pathway. As in the previous experiments involving orthodromic activation, we 266 injected virus into the RSC, and subsequently inserted the silicon probe (32 channels and 50 µm 267 spacing) to record downstream activity in M2. The probe was inserted leaving several contacts 268 out of the cortex; the depth of penetration was estimated both by viewing the site of entry with a 269 high-power stereoscope, and by assessing channel noise variance, which was low for contacts 270 outside cortex (see Methods) ( Fig. 8A,B). Group analysis (n = 9 mice injected with AAV1-271 ChR2) of activity across channels indicated a bias towards deeper layers (Fig. 8C,D). Previous Comparison of layer 2/3 and layer 5 neurons showed significantly greater tendency of photo-280 activated RSC axons to generate spikes in layer 5 neurons (Fig. 8E), consistent with the laminar 281 profile recorded in vivo (Fig. 8C,D) We analyzed corticocortical signaling in the RSC→M2 pathway in vivo using optogenetic 289 photostimulation and electrophysiology. Across a wide range of stimulus parameters, the 290 downstream responses arrived rapidly and scaled systematically with the photo-evoked activity 291 in the upstream area. We found that a simple model involving linear integration, delay, and 292 thresholding could describe much of the data. 293 In using optogenetic photostimulation to analyze this circuit we did not attempt to mimic 294 naturalistic activity patterns of the RSC but rather used this as a tool to perturb the circuit 295 (Miesenbock, 2009). This approach allowed us to systematically vary the stimulus intensity and 296 duration and assess whether and how response properties depended on input parameters. Another 297 artificial aspect of these experiments was the use of anesthesia, without which extensive 298 parametric testing would have been challenging with head-fixed animals. Our approach is aimed 299 at understanding computational aspects of corticocortical population signaling, rather than how 300 detailed corticocortical signals relate to the high-dimensional aspects of behavior (Carandini, 301 2012). 302 We found that a simple two-stage model captured the broad features of the data. At the 303 upstream end, the conversion of light energy into local spiking activity in the upstream area (the 304 RSC) could be described as a simple transfer function dominated by strong and rapid decay. short-term synaptic depression). One potential application of this first-stage model of the local 310 photoactivation process is that it could be used to design photostimuli that precisely compensate 311 for the decay. 312 At the downstream end, the conversion of upstream activity (in RSC) into downstream 313 activity (in M2) could be described by a simple exponential process with a brief delay, and no 314 adaptation mechanism. Although a small non-linearity was included in the form of a threshold, 315 the efficacy of the model suggests that corticocortical signaling is mostly linear. The efficacy of 316 the second-stage model implies that corticocortical driving of downstream activity is highly 317 scalable. It also implies that adaptation is not a major factor in shaping the downstream response, 318 at least on the short time scales (tens of milliseconds) studied here. However, some contribution 319 of an adaptation process may be reflected in the early component of the responses, which tend to 320 be larger than the fitted traces. Whether this simple model can describe corticocortical signaling 321 in other inter-areal pathways remains to be determined, but similarities between our findings 322 using optogenetic activation and related work in the visual system (e.g. (Carandini et al., 1997)) 323 suggest this is plausible. 324 The scalability of corticocortical signaling observed here may be particular to the 325 RSC→M2 pathway, or may represent a more general computational principle of cortical 326 operation (Miller, 2016;Rolls, 2016). Although cortical circuit organization appears basically 327 conserved, areas can also differ substantially in their quantitative properties (Harris and 328 Shepherd, 2015). Corticocortical signaling in other pathways might therefore be expected to 329 exhibit broadly similar scalability, but with pathway-specific differences in the details of In vivo circuit analysis: general procedures. Mice were anesthetized with ketamine-xylazine 405 (ketamine 80-100 mg/kg and xylazine 5-15 mg/kg, injected intraperitoneally), placed in the 406 recording apparatus, and head-fixed using the set screw as described above. Body temperature 407 was monitored with a rectal probe and maintained at ~37.0 °C via feedback-controlled heating 408 pad (FHC, Bowdoin, ME). Craniotomies were opened over the RSC and M2 using a dental drill, 409 just large enough (~1 mm) to allow passage of the linear arrays and the tips of the optical fibers. 410 During the subsequent recordings, ACSF was frequently applied to the exposed brain area to 411 prevent damage from dehydration. The level of anesthesia was continuously monitored based on 412 paw pinching, whisker movement, and eye-blinking reflex. Additional doses of anesthesia were 413 given (50% of induction dose) when required. in the RSC (Fig. 2A). The tip of the fiber was ~0.5 mm away from the surface of the brain, 419 immersed in ACSF. In most experiments, a second fiber was similarly positioned directly over 420 the M2 (Fig. 2A). The probe was then slowly inserted into the cortex at a rate of 2 µm/s (controlled by software), 446 until it reached a depth of 1600 µm from the pia. In most experiments, a second array was 447 similarly inserted into the RSC (same stereotaxic coordinates as given above for the viral 448 injections), except that in this case the array was inserted perpendicular to the horizontal plane, 449 and the fiber was slightly tilted (Fig. 2A). 450 Signals were amplified using a RHD2132 amplifier board based on a RHD2132 digital 451 electrophysiology interface chip (Intan Technologies, Los Angeles, CA). The RHD2132 chip is 452 an AFE (analog front end) which integrates the analog instrument amplifiers, filters, analog-to-453 digital converters, and microcontrollers in one chip. The SPI (serial peripheral interface) port is 454 used to configure the chip and to stream the bio-signal data to the DAQ board. The gain of the 455 amplifier was fixed at 96  2 = 192 (2-stage amplifier). The filter was set to an analog bandpass 456 of 0.1~ 7.5K Hz with a digital filter cutoff of 1Hz. Because the 32 channels of the bio-signal 457 inputs share the same 16 bit ADC with a multiplexer, and the maximum sample rate of the ADC 458 is 1.05M SPS, the single channel sample rate was set to 30K SPS. 459 For hardware control, we used a RHD2000 USB Interface Evaluation Board (Intan) or 460 DAQ board based on a breakout board with a XEM6010 USB/FPGA module (Opal Kelly, 461 Portland, OR), a field-programmable gate array (FPGA) with many digital I/O channels for 462 communicating with other digital devices and streaming in all the bio-signal data from the 463 RHD2000 amplifiers. The USB port of the module was linked with a USB cable to pipe the data 464 stream in or out the PC. The RHD2000 amplifier boards were connected to a DAQ board using 465 SPI interface cables in low-voltage differential signal mode, which is well suited for 466 communication via longer cables. In this experiment, the digital ports included in the DAQ board 467 were only used for acquiring the LED photostimulation parameters from the LED controller (see 468