Sound localization and timing computations in the auditory brain stem. J Rinzel, NYU

with G Svirskis, R Dodla, V Kotak, D Sanes, M Day, B Doiron, P Jercog, N GoldingFunded by NIMH, NIDCD and NSF.

• Computational and experimental study – coincidence detection and ITD coding (gerbil MSO, in vitro)

• Subthreshold dynamic negative feedback: GKLT activ’n; phasic firing; brief temporal integration window; integration of noisy inputs (STA)

•The definitive feedforword neuron: bipolar dendrites and distrib’n of Iion

• Coding: •population coding (slope or place code?); role of inhibition; role of EPSP asymmetries + IKLT;• stimulus dependent filter/selectivity.

In vivo data from the barn owl shows NL neurons encode ITD

A

B

C

D

E

PLACE CODEOUTPUTS

DE

LA

Y L

INE

INPU

TS

DE

LA

Y L

INE

INPU

TS

C

ITD sensitivity arises from a coincidence detection mechanism, as in the Jeffress model5

left ear leads right ear leadsINTERAURAL TIME DIFFERENCE (µsec)

100

50

0

4409 Hz

0-300

-150 150 300

-30 µsec

% M

AX

IMU

M R

ESPON

SE

… place code or slope code?

• in vivo gerbil: ITD-tuning peak is outside physiol range.

• Inhibition shapes ITD-tuning.

Brand et al. Nature, 2002

MSO neurons fire phasically, not to slow inputs. Blocking I KLT may convert to tonic.

J Neurosci, 2002

Even after reducing I KLT, some neurons (older) remained phasic.

INa fairly inactivated near rest.

J Neurosci, 2002

HH-type model with currents: INa IKHT

and subthreshold IKLT

INa

IKHT

IKLT

mV

msec

mV

Phasic firing properties

0

30

0 80

KLT

ThKLTKLTm

nVnn

tIVVnGVGVC

VThKLT

VTh

I

V

Idealized model: integrate and fire with “IKLT”

Network, 2003.

Slow ramp: no spike

Fast ramp:one spike

Schematic of circuit for low frequency coincidence detectionin mammals. (D Sanes w/ focus on gerbil.)

IKLT narrows temporal integration window.

Notice “dip”: IKLT is partially active at rest;transient hyperpolarization promotes spiking by deactivating IKLT.

-8 -6 -4 -2 0

0.0

0.1

0.2

time before spike, ms

I, nA

leaky I&F + IKLT leaky I&F leaky I&F + IKLT below RMP

Spike-generating current by reverse correlation.

Network, 2003.

+-

++

--

“MSO”cell

+-

++

--

+-

++

--

++-

++

--

“MSO”cell

Poisson PSGs from Nex + Ninh input fibers

spont rate

Some expts: Detection of subthreshold signal amidst noisy background

+-

++

--

“MSO”cell

+-

++

--

+-

++

--

++-

++

--

“MSO”cell

Poisson PSGs from Nex + Ninh input fibers

DTX (IKLT blocker) ==> -- widening of integration window -- reduction of “dip”

J Neurosci, 2002

Control

After DTX

Response of MSO cell to brief “signal” in noise.

Spike triggered average “Isyn”, expeimental

Selectivity endowed by IKLT depends on spectralprofile of the input. w/ Day, Doiron J Neurophys, 2008.

• Rothman-Manis (HH-type) 2003 model: Dynamic vs Frozen IKLT

• Noisy input I(t); STEs {IST(tj)} discrete time ti 2 clouds in vector space• Discriminant analysis (feature extraction) finds “direction”

that maximizes “distance” between clouds (Fisher criterion) projections of {IST(tj)}• For white noise input: no difference in STAs.

150 Hz

650 Hz

Stim selec’n diff’ce (SSD)= 1-misclassification error

Coincidence detection – a role for dendrites

Gradient of length alongtonotopic axis.

Agmon-Snir, Carr, Rinzel: Nature ‘98

Reduction of“false positives”

Compartmental model; 2-variable minimal phasic model

“HH-type” cable model, based on I,V-clamp data (in vitro, gerbil, Golding, 2006).

gex(t), τex=0.2 ms

spike generation

gKLT in S, IS and weak in D;active or “frozen” (passive);

gNa only in Axon.

Biophysical model: gerbil MSO -- dendritesw/ P Jercog and Golding lab … ongoing

l/λ ≈ 0.6-0.8τm ≈ 0.6-1 ms

EPSP attenuation and temporal sharpening - subthresholdExperiment Golding lab + V-clamp

Theory Jercog, Rinzel

If gKLT is “frozen”.

0.5 ms 5 mV

73 µm

Dendrite 35 µm

55 µm

Soma 80

60

40

20

0

EPSP

ampl.

(mV)

100806040200Propagation distance (µm)

Dend Soma Dual Somatic

Attenuation and sharpening grow with propagation distance in model.

Experiment

Theory

Time difference sensitivity, enhanced for inputs to dendrite – subthreshold case.

Motion direction sensitivity. Passive cable, Rall (1964).

“directionselectivity”

Proximal to distal sequence: rapid rise, broad EPSP at soma.

Distal to proximal sequence: latency, buildup to higher peak EPSP.

Response to “near then far” input is disadvantaged by wake of (dendritic) gKLT along path to Soma.

τex=0.2 … spike τex=0.5 no spike

Include axonalspike generation

Synaptic input must be fast for spike generation.

Coincidence detection in model… with spikes in axon the definitive feedforward neuron.

“ITD” = 0.1 ms “ITD” = 0.15 ms

No back-propagating action potential.

Grothe, New roles for synaptic inhibition in sound localization, Nat. Rev. (2003)

• ITD peak is outside physiol range

• Blocking inhibition shifts the ITD-tuning curve to “0”.

Tuning for Interaural Time Difference (ITD), shaped by transient inhibition

• Contralateral excitation is preceded by inhibition.• Ipsilateral excitation precedes inhibition.

in vivo, gerbil Brand et al, 2002

Place code or slope code?

ITD

ipsi contraΔ

ITD tuning in small mammals issensitive to timed inhibition slope code

Brand et al, Nature, 2002

Results with MSO cell model. Rothman et al ’93

Key parameters:τinh= 0.1 ms, Δ = 0.2 ms

Asymmetry in EPSPs shapes ITD tuning

In vitro thick slice ITD in dish.

w/ Jercog, Sanes, Svirksis, Kotak - ongoing

If contra-EPSP is slower-rising, it recruits more IKLT before fast rise to threshold – lowering probability to fire.

Ipsi leading Contra leading

Contra EPSPs slower than Ipsi EPSPs

Asymmetry in EPSPs shapes ITD tuningw/ Jercog, Sanes, Svirksis, Kotak - ongoing

In vitro thick slice ITD in dish.

Contra pathway is longer greater latency for EPSPs

Contra inputs are slower rising.

Effect of inhibition -- counteracts the advantage of faster-rising ipsi inputs...

τinh = 2 ms

With inhibition

Inhibition blocked

Sound localization and timing computations in the auditory brain stem. J Rinzel, NYU

with G Svirskis, R Dodla, V Kotak, D Sanes, M Day, B Doiron, P Jercog, N GoldingFunded by NIMH, NIDCD and NSF.

• Computational and experimental study – coincidence detection and ITD coding (gerbil MSO, in vitro)

• Subthreshold dynamic negative feedback: GKLT activ’n; phasic firing; brief temporal integration window; integration of noisy inputs (STA)

•The definitive feedforword neuron: bipolar dendrites and distrib’n of Iion

• Coding: •population coding (slope or place code?); role of inhibition; role of EPSP asymmetries + IKLT;• stimulus dependent filter/selectivity.