Context-dependent cross-modal interaction in the medial prefrontal cortex of rats

Cross-modal interaction (CMI) could significantly influence the perceptional or decision-making 36 process in many circumstances. However, it remains poorly understood what integrative strategies are 37 employed by the brain to deal with different task contexts. To explore it, we examined neural activities 38 of the medial prefrontal cortex (mPFC) of rats performing cue-guided two-alternative forced-choice 39 tasks. In a task requiring rats to discriminate stimuli based on auditory cue, the simultaneous 40 presentation of an uninformative visual cue substantially strengthened mPFC neurons' capability of 41 auditory discrimination mainly through enhancing the response to the preferred cue. Doing this also 42 increased the number of neurons revealing a cue preference. If the task was changed slightly and a 43 visual cue, like the auditory, denoted a specific behavioral direction, mPFC neurons frequently showed 44 a different CMI pattern with an effect of cross-modal enhancement best evoked in information- 45 congruent multisensory trials. In a choice free task, however, the majority of neurons failed to show a 46 cross-modal enhancement effect and cue preference. These results indicate that CMI at the neuronal 47 level is context-dependent in a way that differs from what has been shown in previous studies. direction: auditory choice selectivity, 77%, 142/184; multisensory choice selectivity, 71%, 130/184).


Introduction 50
In real life, we often receive multiple sensory cues simultaneously (with most being visual and 51 auditory). The brain must combine them properly and form an effective decision in response to 52 whatever the combination represents accurately. During this process, the brain must decide what 53 sensory inputs are related and what integrative strategy is appropriate. In the past three decades, this 54 process of cross-modal interaction (CMI) or multisensory integration has been widely examined in many 55 brain areas such as superior colliculus, and both primary sensory and association cortices [1-6]. A series 56 of integrative principles that govern this process have been derived (i.e., spatial, temporal, and inverse 57 effectiveness), and testing has shown them to be operant in many brain areas [2]. In the classic example, 58 multisensory neurons in superior colliculus can show greatly enhanced responses to spatiotemporally 59 congruent multisensory cues [7]. Similarly, in monkeys performing a directional task, neurons in several 60 cortical regions such as the dorsal medial superior temporal area have shown enhanced heading 61 selectivity when matched visual and vestibular cues are given simultaneously [8]. In like manner, 62 effectively integrating cross-modal cues was also found to improve perceptual performance [9-11] and 63 shorten reaction times [12][13][14]. regardless of whether the preferred was A3k (51/54) or A10k (27/33). Such biased enhancement further 156 strengthened neurons' choice selectivity (Fig. 5C&D). 157 The influence of information congruence/incongruence between visual and auditory cues on mPFC 158

neurons' cross-modal interaction 159
In behavioral Task 1, animals made their behavioral choice based on the auditory cue alone. We 160 then wondered how mPFC neurons would change their integrative strategy if the behavioral choice 161 became dependent on both auditory and visual cues. To examine this, we trained 7 rats to perform a 162 new behavioral task (Task 2). In this task, the only difference from Task 1 is that an individual visual 163 stimulus (V) was introduced into the stimulus pool as an informative cue. If the triggered stimulus is A10K, 164 VA10k, or V, animals should go to the left port to get the reward (Fig. 6A). Otherwise, they should move 165 to the right port to be rewarded. This task took animals about two months of training to surpass 75% 166 correct performance for five consecutive sessions. Although there was no difference in behavioral 167 performance between two auditory alone conditions (A3k vs. A10k: 83.7% vs. 85.7%, t(6)=0.888, p=0.41, 168 paired t-test, Fig. 6B), the task showed a difference between two multisensory conditions. This 169 performance increased when the cues themselves had congruent information content and declined 170 when they indicated a cued directional mismatch (VA3k vs. VA10k: 77.1% vs. 91.1%, t(6)=5.214, p=0.002, 171 paired t-test, Fig. 6B). The mean reaction time in multisensory trials across animals was still significantly 172 shorter than that in corresponding auditory trials regardless of whether the auditory is A3k or A10k  showing cross-modal enhancement in the information-congruent VA10k trials, 20 of them (36%) favored 187 A10k (see the example in Fig. 6D) and 28 of them (51%) showed no overt preference of auditory choice 188 (see the example in Fig. 6E). In several cases, like the neuron shown in Fig. 6F, the visual stimulus 189 appeared to reverse selectivity, and for auditory, they showed a preference for A3k, but for multisensory, 190 favored VA10k. 191 The mean MI in information-incongruent VA3k condition across populations (n=247) is nearly zero 192 (0.01±0.16), which was significantly lower than 0.07±0.15 in the congruent VA10k condition (p<0.00001, 193 Mann-Whitney Rank Sum Test; see the comparison of an individual case in Fig. 7E). Also, one would 194 expect that the information match should induce more substantial effects of cross-modal enhancement. 195 It was not the case, however. In examining all neurons exhibiting cross-modal enhancement in VA10k 196 condition in Task 1 and Task 2, we found no difference between them (mean MI: 0.21±0.18 in Task 2 vs. 197 0.20±0.11 in Task 1, p=0.489, Mann-Whitney Rank Sum Test). Summarily, these results indicate that the 198 activities of mPFC neurons reflected the context of the task and maintained their ability to discriminate, 199 and, ostensibly, aid in successful task completion. 200

Cross-modal interaction in a choice-free task 201
Tasks 1&2 required animals to discriminate sensory cues. The next intriguing question to us was how 202 then mPFC neurons would treat different combinations of sensory cues and CMI when cue 203 discrimination is not required? To investigate this, we trained another group of rats (n=9) to perform a 204 choice-free task (Task 3). In this task, animals would get a water reward in either the left or right port 205 regardless of which stimulus was presented, rendering the cueing discrimination irrelevant. We carefully 206 examined 184 mPFC neurons recorded during the performance of Task 3. For consistency with the 207 earlier analyses, neuron's response in A3k_right_choice trials was compared with the response in 208 . This result, taken together with those given above, 226 reveals that the differential neural activities in mPFC likely reflect the context of the given task. When 227 stimulus discrimination is not required, the neuronal activity exhibits no selectivity. When demanded 228 by an appropriate task, mPFC neurons are quite capable of sensory discrimination. 229 230

Discussion 231
We used cue discrimination tasks to understand context-dependent CMI in rat mPFC, an area that is 232 believed to be essential both for perception and decision-making. The result showed that, in a task 233 requiring auditory discrimination, the presence of an uninformative visual stimulus mostly served only 234 to heighten the preferred auditory choice signal. As a result, the neurons exhibited better perceptual 235 decision capability for multisensory conditions than for auditory alone conditions. However, if a visual 236 cue, like the auditory, was made informative, mPFC neurons frequently showed a different CMI pattern 237 with an enhanced multisensory perceptual signal when both auditory and visual cues indicated the 238 same behavioral instruction. When no cue discrimination was required in the task, the majority of 239 neurons failed to show the same pattern of CMI and a similar choice strategy. This result greatly expands 240 our understanding of the role that CMI can play in the brain. 241 Most of our understandings regarding CMI were developed using anesthetized or passively 242 sensing animals. In these studies, the spatiotemporal arrangement and intensities of stimuli were 243 found to be critical for CMI. We believe that more factors should influence CMI when humans and 244 9 animals perform tasks. Also, as we know now, the levels of neural activity found in an alert, active 245 brain are dramatically different from an anesthetized or passive preparation. provided new evidence for backing this conclusion. However, it remains to be discovered whether this 257 context-dependent CMI is unique to mPFC or if it exists in other brain areas, which we intend to examine 258 in future studies. Also, the brain state should be a critical factor for influencing CMI, and a recent study 259 showed that cross-modal inhibition dominated in mPFC in anesthetized rats [39]. 260 When performing the task, rats showed shorter reaction times in multisensory conditions. This In contrast, the prefrontal cortex is known to be involved in higher-order cognitive functions, including 297 decision making. It is, therefore, reasonable to conclude that different brain regions would need to apply 298 different strategies of CMI to process multisensory inputs in line with their overall processing goals. 299 300 polyurethane foam for sound attenuation (outer size: 120×100×120 cm). Three snout ports, each 320 monitored by a photoelectric switch, are located on one sidewall of the operant chamber (see Fig. 1A). 321

Rat subjects 302
The signals from the photoelectric switches were first fed to an analog-digital multifunction card and 322 digitized (DAQ NI 6363, National Instruments, Austin, TX, USA) and sent via USB to a PC running the 323 training program. 324 Rats initiated a trial by poking their nose into the center port. Following a short variable delay (500-325 700ms), a stimulus (two auditory, two auditory-visual, or one visual, randomly selected) was presented. 326 After presentation of this cue, rats could immediately initiate their behavioral choice, moving to the left 327 or right port (Fig. 1A). If rats made a correct choice (hit trial), they could obtain a water reward, and a 328 new trial could immediately follow. If animals made wrong or no behavioral choice within 3 seconds 329 after cue onset, the punishment of a 5-6s timeout was applied. 330 The auditory cue was delivered via a speaker (FS Audio, Zhejiang, China), using a 300ms-long 3kHz 331 (low) or 10kHz (high) pure tone with 25ms attack/decay ramps given at 60 dB sound pressure level (SPL) 332 against an ambient background of 35-45 dB SPL. SPLs were measured at the position of the central port 12 (the starting position). The visual cue was a 300ms-long flash of white light given at 5~7cd/m 2 intensity, 334 delivered by a light-emitting diode. The auditory-visual cue (multisensory cue) was the simultaneous 335 presentation of both auditory and visual cues. 336

Assembly of tetrodes 337
Formvar-Insulated Nichrome Wire (bare diameter: 17.78 μm, A-M systems, WA, USA) was twisted in 338 groups of four as tetrodes (impedance: 0.5-0.8 MΩ at 1 kHz). Two 20 cm-long wires were folded in half 339 over a horizontal bar for twisting. The ends were clamped together and manually twisted clockwise. 340 Finally, their insulation coating was fused with a heat gun at the desired level of twist and cut in the 341 middle to produce two tetrodes. To reinforce each tetrode longitudinally, each tetrode was then 342 inserted into Polymide tubing (inner diameter: 0.045 inches; wall: 0.005 inches; A-M systems, WA, USA) 343 and fixed in place by cyanoacrylate glue. An array of 2×4 tetrodes were then assembled using an inter-344 tetrode gap of 0.4-0.5 mm. After assembly, the insulation coating of each wire was gently removed at 345 the tip, and then the wire was soldered to a connector pin. The reference electrode used was a tip-346 exposed Ni-Chrome wire of diameter 50.8μm (A-M systems, WA, USA), and a ground electrode was a 347 piece of copper wire of the diameter of 0.1mm. Both of these were also soldered to a connector pin. 348 The tetrodes and reference were then carefully cemented by silicon gel and trimmed to an appropriate 349 length immediately before implantation. 350

Electrode Implantation 351
The animal was administered a subcutaneous injection of atropine sulfate (0.01 mg/kg b.w.) before 352 surgery and then was anesthetized with an initial intraperitoneal (i.p.) injection of sodium pentobarbital 353 (40-50 mg/kg b.w.). After anesthesia, the animal was fixed on the stereotaxic apparatus (RWD, 354 Shenzhen, China). The tetrode array was then implanted in the left mPFC (AP 2.5-4.5 mm, ML 0.3-0.8 355 mm, 2.0-3.5 mm ventral to the brain surface) by slowly advancing a micromanipulator (RWD, Shenzhen, 356 China). Neuronal signals were monitored throughout implantation to ensure appropriate placement. 357 Tissue gel (3M, Maplewood, MN, US) was used to seal the craniotomy. The tetrode array was then 358 secured to the skull with stainless steel screws and dental acrylic. After surgery, animals were given a 4-359 day course of antibiotics (Baytril, 5mg/Kg b.w., Bayer, Whippany, NJ, US). They had a recovery period of 360 at least 7 days (usually 9-12 days with free access to food and water). 361

Neural recordings 362
When recovered from the surgery, animals resumed performing the behavioral task in the same 363 13 training chamber but now situated inside a larger acoustically and electrically shielded room (size 2.5 × 364 2 × 2.5m, length X width X height). Recording sessions began after the animal's behavioral performance 365 recovered to the level attained before surgery (typically 2-3 days). Wideband neural signals (250-6000 366 Hz) were recorded using a head-stage amplifier (RHD2132, Intantech, CA, USA). Amplified (×20) and 367 digitized (at 20 kHz) neural signals were combined with trace signals representing both the stimuli and 368 session performance information and sent to a USB interface board (RHD2000 Intan technology, CA, 369 USA), and then to a PC for on-line observation and data storage. Correct performance rate (%) = 100*hit trials/ total number of trials. 385 Raw neural signals were recorded and stored for offline analysis. Spike sorting was later performed 386 using Spike 2 software (CED version 8, Cambridge, UK). Recorded raw neural signals were band-pass 387 filtered in 300-6000 Hz to remove field potentials. A threshold criterion of no less than 3-fold standard 388 deviations (SD) above background noise were used for identifying spike peaks. The detected spike 389 waveforms were then clustered by principal component analysis and a template-matching algorithm. 390 Waveforms with inter-spike intervals of <2.0 ms were excluded. Relative spike timing data for a single 391 unit were then obtained for different trials of different cued conditions and used to construct both raster 392 plots and prestimulus time histograms (PSTHs) using custom Matlab scripts. Only neurons for which the 393 14 overall meaning firing rate within the session was at least 2Hz were included for analysis. As generally 394 observed, behavioral and neuronal results were similar across all relevant animals for a particular testing 395 paradigm. Thus, the data across sessions were combined to study population effects. 396 To render PSTHs, all spike trains were first binned at 10 ms and convolved with a smoothing Gaussian 397 Kernel (δ=100ms) to minimize the impact of random spike-time jitter at the borders between bins. The 398 mean spontaneous firing rate was calculated from a 500-ms window immediately preceding stimulus 399 onset. Decision-making-related neural activity was quantified as mean firing rates in the 500-ms after 400 cue onset after subtracting the mean spontaneous firing rate. 401 We quantified the choice selectivity between two different cue conditions used during a task (for 402 example, low tone trials vs. high tone trials) by using a receiver operating characteristic (ROC) based 403 analysis [73]. Firstly, we set 12 threshold levels of activity covering the range of firing rates obtained in 404 cue_A and cue_B trials. Following that, a ROC curve is generated, for each threshold criterion, by 405 plotting the proportion of cue_A trials on which the response exceeded criterion against the proportion 406 of cue_B trials on which the response exceeded criterion. The value of choice selectivity is defined as 407 2*((area under the ROC curve)-0.5). Therefore, a value of 0 indicates no difference in the distribution 408 of responses between cue_A and cue_B. A value of 1/-1 represents the highest selectivity, that is, 409 responses triggered by cue_A were always higher or lower than those evoked by cue_B. 410 To test the significance of each choice selectivity value, we ran a permutation test. This was 411 accomplished by randomly distributing all trials from a neuron into two groups, independent of the 412 actual cue conditions. These groups were nominally called cue_A trials and cue_B trials and contained 413 the same number of trials as the experimentally obtained groups. The choice selectivity value was then 414 calculated from the redistributed data, and the procedure was repeated 5000 times, thereby giving a 415 distribution of values from which to calculate the probability of the result we obtained. When our actual 416 value was found in the top 5%, it was defined as significant (i.e., p<0.05). 417 To quantify the difference between responses in visual-auditory (multisensory) and auditory trials, 418 we calculate the index of cross-modal interaction (MI) using the following function： MI = (VA-A) / (VA+A), 419 where VA and A represent firing rates in multisensory and auditory trials, respectively. MI has a range 420 of -1 to 1, with more positive values indicating the response in multisensory trials was much stronger 421 and more negative values meaning the response in auditory trials was more robust. 422

Statistical analysis 423 15
All statistical analyses were conducted in Matlab 2015b with statistical significance assigned for 424 findings attaining a p-value of < 0.05. All behavioral data (for example, mean reaction time differences 425 between auditory and multisensory trials) were compared using the paired t-test. We performed the 426 Chi-square test to analyze the difference in proportions of neurons (recorded in different Tasks) showing 427 choice selectivity. To compare MIs between different cue conditions within the same group of neurons, 428 we performed a paired t-test or Mann-Whitney Rank Sum Test where appropriate. Unless stated 429 otherwise, all data group results are presented as mean ± SD. 430