The synaptic dialogue of cortical representation

3 downloads 0 Views 1MB Size Report
Teki S, Kumar S, von Kriegstein K, Stewart L, Lyness CR, Moore BC, Capleton B,. 895. Griffiths TD (2012) Navigating the auditory scene: an expert role for the.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Comodulation enhances signal detection via priming of auditory cortical circuits

Joseph Sollini and Paul Chadderton Department of Bioengineering, Imperial College London, London SW7 2AZ, United Kingdom.

21 22 23 24 25 26 27 28

ACKNOWLEDGMENTS

29 30

We thank Andrei Kozlov for feedback on the manuscript, Matthew Brown and

31

Alexander Morris for discussions and support. This research was supported

32

by a MRC Career Development Award to PC (G1000512).

33 34

The authors declare no competing interests.

1

35

ABSTRACT

36 37

Acoustic environments are composed of complex overlapping

38

sounds that the auditory system is required to segregate into

39

discrete perceptual objects. The functions of distinct auditory

40

processing stations in this challenging task are poorly understood.

41

Here we show a direct role for mouse auditory cortex in detection

42

and segregation of acoustic information. We measured the

43

sensitivity of auditory cortical neurons to brief tones embedded in

44

masking noise. By altering spectrotemporal characteristics of the

45

masker, we reveal that sensitivity to pure tone stimuli is strongly

46

enhanced

47

corresponding to the psychoacoustic phenomenon comodulation

48

masking release. Improvements in detection were largest following

49

priming periods of noise alone, indicating that cortical segregation

50

is enhanced over time. Transient opsin-mediated silencing of

51

auditory cortex during the priming period almost completely

52

abolished these improvements, suggesting that cortical processing

53

may play a direct and significant role in detection of quiet sounds

54

in noisy environments.

in

coherently

modulated

55 56 57 58 59 60 61 62 63 64 65 66 67

2

broadband

noise,

68

SIGNIFICANCE STATEMENT

69 70

Auditory systems are adept at detecting and segregating competing sound

71

sources, but there is little direct evidence of how this process occurs in the

72

mammalian auditory pathway. We demonstrate that coherent broadband noise

73

enhances signal representation in auditory cortex, and that prolonged exposure

74

to noise is necessary to produce this enhancement. Using optogenetic

75

perturbation to selectively silence auditory cortex during early noise processing,

76

we show that cortical processing plays a crucial role in the segregation of

77

competing sounds.

78

3

79

INTRODUCTION

80 81

In the acoustic world, animals are challenged to detect salient sounds in noisy

82

backgrounds, a process of critical importance in communication, hunting and threat-

83

detection. The auditory system is well suited to the task as it demonstrates a

84

remarkable spectral (Rosenblith and Stevens, 1953) and temporal resolution (Klumpp

85

and Eady, 1956; Plomp, 1964), and this acuity is valuable for detecting changes in

86

natural soundscapes (McDermott et al., 2013; Andreou et al., 2015). Predictable, or

87

coherent, non-random features of soundscapes may be exploited by the auditory

88

system to improve sound processing (Ulanovsky et al., 2003; Taaseh et al., 2011;

89

Yaron et al., 2012; Teki et al., 2013; Krishnan et al., 2014; Nelken, 2014). A prevalent

90

feature of natural sound is coherent amplitude fluctuations across frequency, termed

91

‘comodulation’. Comodulation is present in both environmental sounds and

92

vocalizations (Nelken et al., 1999). Given its pervasiveness in nature, comodulation

93

may be a critical cue for grouping and segregating overlapping sounds (Nelken et al.,

94

1999; Krishnan et al., 2014).

95 96

Comodulation masking release (CMR) is a psychoacoustic phenomenon whereby

97

adding coherently modulated noise to an existing masker makes signals easier to

98

perceive (Hall et al., 1984). This effect is striking as additional noise energy normally

99

doesn’t change, or reduces, the detectability of signals (Fletcher, 1940). CMR

100

encompasses two separate processes, dependent on the relative frequencies of the

101

noise and the signal: within-channel CMR (signal and noise similar in frequency) and

102

across-channel CMR (signal and noise dissimilar in frequency). Within-channel CMR

103

can be implemented in the auditory periphery, but across-channel or “true” CMR

104

(Verhey et al., 2003) cannot be explained by mechanical processes in the ear and is

105

sensitive to cues of auditory grouping (Buss et al., 2009; Dau et al., 2009; Verhey et

106

al., 2012). Across-channel CMR is therefore a result of brain processing, but the

107

mechanism and location of such processing is not well understood.

108 109

Only a small number of studies have directly sought to understand the representation

110

and underlying mechanism(s) of co-modulation masking release at the cellular level.

111

In the peripheral auditory system, neuronal responses to pure tones are enhanced by

112

across-channel comodulation in a way consistent with human behaviour (Pressnitzer

113

et al., 2001). However, it is not clear whether this information is inherited or influenced

114

by processing at later stages of the auditory system. A CMR correlate has been shown

115

to develop progressively between the inferior colliculus (IC), medial geniculate body

4

116

(MGB) and auditory cortex (Nelken et al., 1999; Las et al., 2005), although, this work

117

explored both within and across-channel cues simultaneously. As such, it remains

118

unclear how much of the observed CMR is attributable to across-channel processes

119

(Verhey et al., 2003; Grose et al., 2004). Neuronal correlates of within-channel CMR

120

have been observed in the avian auditory forebrain area L2a, however, when

121

measured in an across-channel configuration (comparing narrowband and broadband

122

comodulated noise) no significant CMR was found (Nieder and Klump, 2001; Hofer

123

and Klump, 2003).

124 125

In this study, we set out to quantify the influence of across-channel comodulation on

126

signal detectability in the neuronal activity of primary auditory cortex (A1), a critical site

127

in auditory perception (Bizley and Cohen, 2013). We then sought to establish whether

128

processing by auditory cortical circuits plays a functional role in the formation of across-

129

channel CMR. We provide the first quantification of across-channel CMR in A1.

130

Transient inactivation during early sound processing significantly reduces subsequent

131

masking release, suggesting that auditory cortex may play a causal role in enhancing

132

signal representation in the presence of comodulated noise.

133 134 135

5

136 137 138 139

MATERIALS AND METHODS

Animals and preparation

140 141

The care and experimental manipulation of animals was performed in accordance with

142

institutional and United Kingdom Home Office guidelines. Homozygous Pvalb-IRES-

143

Cre mice (JAX Stock No. 008069) of both genders were used in this study. At 6 weeks,

144

auditory cortex was injected unilaterally with AAV-EF1a-DIO-hChR2(H134R)-EYFP

145

virus in order to express selectively ChR2 and EYFP proteins in PV+ interneurons.

146

Mice were anesthetized with 1-2% isoflurane under aseptic conditions and held using

147

ear bars on a stereotaxic mount (Angle 2, Leica). A small burr hole was made with a

148

dental drill ~1.m lateral from midline and ~2.7mm caudal to bregma. A small glass

149

pipette holding the virus was advanced to reside in auditory cortex. 0.5 µl was then

150

slowly injected into cortex over a period of 15 minutes. The pipette was then removed

151

and the burr hole sealed with Kwik-Cast (World Precision Instruments, USA), once dry

152

acrylic dental cement was layered over the top forming a hard seal over the site. The

153

area was then cleaned (Iodine) and the tissue sealed (Histoacryl, Braun Corporation,

154

USA). Analgesia was administered during the procedure via intra-peritoneal injection

155

(Carprofen; 5mg/kg). The animal was then recovered and the virus left to express for

156

2 weeks, during which time Buprenorphine (0.8mg/kg) jelly was used for postoperative

157

analgesia.

158 159

In vivo electrophysiology

160 161

For electrophysiological recordings mice were anaesthetized with a fentanyl/

162

midazolam/ medetomidine/mixture (0.05, 5.0 and 0.5 mg/kg). A midline incision was

163

made and all tissue was cleared from the cranium. Hatched score lines were made on

164

the cranial surface with a dental drill to increase surface area and a customized head

165

plate was fixed using adhesive (Histoacryl, Braun Corporation, USA). A small

166

craniotomy was made directly above auditory cortex and dura was removed using fine

167

forceps. Mice were fixed by clamping the headplate to a customized clamp stand. A

168

silicon microelectrode (A32, 4x2Tet Neuronexus) was advanced via micromanipulator

169

(IVM, Scientifica, UK) into auditory cortex at an angle perpendicular to the cortical

170

surface. An optical fiber, attached to DPSS laser (473nm, 100 mW; SLOC, China),

171

was positioned between the middle two probe shanks ~1mm from the cortical surface.

172

The laser was controlled via TTL signals generated via digital processor (RZ6, Tucker

173

Davis Technology, USA). A number of laser durations were trialed, with the intention

6

174

of disrupting cortical activity for the period of the pre-cursor (0-400ms) but not

175

disrupting cortical activity to the later masker (500-1000ms). We found that laser

176

duration of 150ms (from 0 to 150ms) was sufficient to produce transient suppression

177

(Figure 6C).

178 179

Data was acquired via Digital Lynx 16SX system (Neuralynx, USA) and stored on a

180

PC. Recordings were made in primary auditory cortex from a total of 21 sites at

181

different penetrations, in 6 animals.

182 183

Auditory stimulation

184 185

Auditory stimuli were generated and calibrated using Matlab (Mathworks, USA) and

186

stored as files to be used when required. All signal levels were calibrated (5-100kHz

187

flat spectrum ±1.5dB SPL). Stimuli were sinusoid amplitude modulated (SAM) tone

188

maskers presented in the presence of a pure tone signal. Three bandwidth

189

configurations were used: a narrowband condition (NB: 10Hz SAM 20kHz pure tone),

190

and two broadband conditions (IM and CM). The broadband conditions consisted of

191

an on-frequency band (10Hz SAM 20kHz pure tone), low off-frequency bands (3 x

192

0.125 octave spaced 10Hz SAM pure tones centered at 12.9kHz) and high off-

193

frequency bands (3 x 0.125 octave spaced 10Hz SAM pure tones centered at

194

30.8kHz). Off-frequency bands were designed so that all components fell outside a

195

mouse auditory filter (May et al., 2006). For all conditions the phase of the envelope of

196

the on-frequency band remained identical. In the IM condition, the envelope of the off-

197

frequency bands was selected at random between 0-180°, and in the CM condition the

198

off-frequency bands had identical envelope phases with the on frequency band. The

199

NB, IM and (Long-) CM conditions were all composed of a pre-cursor (400ms) and

200

masker portion (500ms) with a short gap in between (100ms). A short-CM condition

201

was also used this was identical to the Long-CM condition but without the pre-cursor.

202

An inter-stimulus interval of 2 seconds was used. The signal was comprised of three

203

50 ms SAM tone pips (10Hz modulator), occurring in the 3-5th troughs of the masker.

204

8 different signal conditions were used for each masker condition, a noise alone (i.e.

205

no signal) condition and seven noise + signal conditions (-10 to 20dB SNR in 5 dB

206

steps, where the masker level was held at 65dB SPL and the signal level varied).

207 208

Frequency response areas (FRAs) were measured using pure tones, at 25 different

209

frequencies (500ms duration, 7-57kHz, with 0.25 octave spacing), 8 different sound

210

levels (10-80 dB SPL at 10dB steps), with a 1 second inter-stimulus interval. Stimuli

7

211

for both FRA and masking experiments were presented in blocks, a single repeat of

212

each stimulus combination was presented in each block (frequency/level or masker

213

signal combination, respectively). The order of the stimuli within a block was randomly

214

selected. Blocks of masking condition and FRA stimuli were interleaved such that five

215

blocks of masking condition were followed by one block of the FRA stimuli. In total

216

there were 30 repeats of each masker/signal combination and 6 repeats of each

217

frequency/SPL combination. All stimuli were saved in the above structure into a single

218

file to be used for neuronal recordings. Sound files were presented using Matlab to

219

interface with RPvdsEX (TDT) running on an RZ6 (TDT) driving a free-field speaker

220

(ES1, TDT).

221 222

Data Analysis

223 224

Spikes

225

(http://sourceforge.net/projects/spikedetekt/) and initially clustered using KlustaKwik.

226

Clusters were manually inspected using Klusters and reclustered when necessary,

227

clusters that contained >1% of spikes within a 1ms interspike interval were rejected

228

(http://neurosuite.sourceforge.net). Spike time data was extracted and exported to

229

Matlab for further processing.

were

extracted

from

the

raw

data

files

using

SpikeDetekt

230 247

Raw FRAs were initially smoothed using a 3x3 pyramidal window, iso-response curves

248

(FRA edges) were then determined by finding a 30% change from baseline firing rate

249

(Sutter and Schreiner, 1991; Scholes et al., 2011). Both excitatory (positive criterion)

250

and inhibitory (negative criterion) curves were defined but processed separately. Each

251

curve circumscribed a defined region. Inspection of the resulting defined regions

252

revealed that using a single criterion meant that two strong responsive regions,

253

separated by a weak response in between, could be defined as two independent

254

regions (because the weak region did not cross criterion). To prevent this an additional

255

stage was added where a 15% criterion was used and regions connected by this lower

256

criterion were grouped into single regions (regions were defined using bwboundaries

257

command in Matlab, excitatory regions were only grouped with other excitatory

258

regions). The largest defined region was always used for further analysis.

259

Characteristic frequency was taken as the frequency yielding a defined response at

260

the lowest signal level.

261 262

Significant changes in firing rate of two conditions were detected by pooling the single

263

presentation repeats of those conditions and bootstrapping to produce a distribution of

8

264

mean firing rate (500 samples, sample with replacement). This distribution was then

265

used to define significance increases in firing rate (mean firing rate p>0.95). In order

266

to compare responsiveness to the signal and ignore the plethora of responses to the

267

noise we calculated the signal response:

268 269 270 271

Where SR is the signal response PSTH, Sig the noise and signal PSTH and N the

272

noise alone PSTH (5ms bins were used for all PSTHs) and where i is the cell number,

273

j the background noise condition index and k the SNR index. σ was calculated as the

274

standard deviation of all possible firing rates for that cell (i.e. all bins from all PSTHs

275

for that cell were included in this measure). The SR PSTH for each condition was then

276

averaged between 1.175-1.275s (i.e. the period of the first tone pip). This produced a

277

SR/SNR function for each cell, which was up-sampled (linear interpolation) in the SNR

278

dimension and the individual cell threshold estimated (0.5 STD criterion crossing).

279 280

The population signal response was the mean of the individual signal responses. As

281

before this was then averaged between 1.175-1.275s to produce the mean SR/SNR

282

function. To compare signal sensitivity and tuning properties we grouped cells, based

283

on their CF, and data were bootstrapped to allow comparison across populations (50

284

selections per subpopulation with replacement, 500 repeats). This gave 500 SR/SNR

285

functions, the mean of which was used to estimate the SR/SNR function for each

286

subpopulation. An arbitrary criterion of 0.25 STD was selected, this value was

287

intentionally low to allow subpopulations with only weak responses to be included in

288

this analysis. To quantify thresholds, firing rates were converted into a neurometric

289

function for each individual condition, by comparing the mean firing rate to the noise

290

alone versus each SNR measured. Each cell was given a “vote” to respond whether

291

that cell detected a signal. A relative increase in firing rate (≥25%) was equivalent to a

292

“yes” (and given a value of 1), a decrease (≥25%) was equivalent to a “no” (a value of

293

0) and anything in-between was set to chance (a value of 0.5). The population

294

proportion correct was the mean value across the population. This neurometric

295

function (proportion correct vs SNR) was up-sampled (linear interpolation) and

296

binomial probability used to estimate threshold (above chance performance, p