Spatial hearing evolved independently in mammals and birds and is thought to adapt to altered developmental input in different ways. We found, however, that ferrets possess multiple forms of plasticity that are expressed according to which spatial cues are available, suggesting that the basis for adaptation may be similar across species. Our results also provide insight into the way sound source location is represented by populations of cortical neurons.
Subscribe to Journal
Get full journal access for 1 year
only $17.42 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Rent or Buy article
Get time limited or full article access on ReadCube.
All prices are NET prices.
Polley, D.B., Thompson, J.H. & Guo, W. Nat. Commun. 4, 2547 (2013).
Popescu, M.V. & Polley, D.B. Neuron 65, 718–731 (2010).
Knudsen, E.I. & Knudsen, P.F. Science 230, 545–548 (1985).
Knudsen, E.I., Knudsen, P.F. & Esterly, S.D. J. Neurosci. 4, 1012–1020 (1984).
Gold, J.I. & Knudsen, E.I. J. Neurosci. 20, 862–877 (2000).
Keating, P., Dahmen, J.C. & King, A.J. Curr. Biol. 23, 1291–1299 (2013).
Keating, P. & King, A.J. Front. Syst. Neurosci. 7, 123 (2013).
Grothe, B., Pecka, M. & McAlpine, D. Physiol. Rev. 90, 983–1012 (2010).
McAlpine, D. & Grothe, B. Trends Neurosci. 26, 347–350 (2003).
Wightman, F.L. & Kistler, D.J. J. Acoust. Soc. Am. 101, 1050–1063 (1997).
Middlebrooks, J.C. & Green, D.M. Annu. Rev. Psychol. 42, 135–159 (1991).
Heffner, H.E. & Heffner, R.S. in Handbook of Research Methods in Experimental Psychology (ed. Davis, S.F.) 413–440 (Blackwell, 2003).
Keating, P., Nodal, F.R. & King, A.J. Eur. J. Neurosci. 39, 197–206 (2014).
Lesica, N.A., Lingner, A. & Grothe, B. J. Neurosci. 30, 11696–11702 (2010).
McAlpine, D., Jiang, D. & Palmer, A.D. Nat. Neurosci. 4, 396–401 (2001).
Stecker, G.C., Harrington, I.A. & Middlebrooks, J.C. PLoS Biol. 3, e78 (2005).
Day, M.L. & Delgutte, B. J. Neurosci. 33, 15837–15847 (2013).
Goodman, D.F., Benichoux, V. & Brette, R. eLife 2, e01312 (2013).
Miller, L.M. & Recanzone, G.H. Proc. Natl. Acad. Sci. USA 106, 5931–5935 (2009).
Kacelnik, O., Nodal, F.R., Parsons, C.H. & King, A.J. PLoS Biol. 4, e71 (2006).
Nodal, F.R., Bajo, V.M., Parsons, C.H., Schnupp, J.W. & King, A.J. Neuroscience 154, 397–408 (2008).
Moore, D.R. & Hine, J.E. Brain Res. Dev. Brain Res. 66, 229–235 (1992).
Silverman, M.S. & Clopton, B.M. J. Neurophysiol. 40, 1266–1274 (1977).
Hogan, S.C., Stratford, K.J. & Moore, D.R. Br. Med. J. 314, 350–353 (1997).
Bizley, J.K., Nodal, F.R., Nelken, I. & King, A.J. Cereb. Cortex 15, 1637–1653 (2005).
Moore, D.R., Hutchings, M.E., King, A.J. & Kowalchuk, N.E. J. Neurosci. 9, 1213–1222 (1989).
Mogdans, J. & Knudsen, E.I. Hear. Res. 74, 148–164 (1994).
Dahmen, J.C., Keating, P., Nodal, F.R., Schulz, A.L. & King, A.J. Neuron 66, 937–948 (2010).
Zohary, E., Shadlen, M.N. & Newsome, W.T. Nature 370, 140–143 (1994).
Bizley, J.K., Walker, K.M., King, A.J. & Schnupp, J.W. J. Neurosci. 30, 5078–5091 (2010).
Rothschild, G., Nelken, I. & Mizrahi, A. Nat. Neurosci. 13, 353–360 (2010).
Davison, A.C. & Hinkley, D.V. Bootstrap methods and their application (Cambridge University Press, 1997).
We are grateful to S. Spires, D. Kumpik, F. Nodal and V. Bajo for their contributions to behavioral testing, R. Campbell for helpful discussion and B. Willmore and G. Douaud for commenting on the manuscript. This work was supported by the Wellcome Trust through a Principal Research Fellowship (WT076508AIA) to A.J.K. and a Newton Abraham Studentship to P.K.
The authors declare no competing financial interests.
Integrated supplementary information
(a) Simulated binaural interaction functions are shown for an individual unit. Data are shown for normal hearing (blue) and plugged (magenta) conditions either pre- (top) or post- (bottom) adaptation. Top: Immediately after plugging, the stimulus needs to be more intense in the plugged ear in order to match the neural response seen prior to plugging. Because the left ear was plugged, this occurs for more negative ILDs (i.e. left-ear greater; convention illustrated by cartoon). The earplug therefore immediately shifts neural tuning to more negative ILDs. Consequently, the peak of the binaural interaction function (best ILD; arrows) becomes more negative. Bottom: To compensate for the earplug, the binaural interaction function should shift so that, under plugged conditions, it resembles that of the pre-adapted neuron under normal hearing conditions. This requires a shift toward more positive ILDs (i.e. intensity greater in the right ear). This is evident as a positive shift in best ILD when the earplug is subsequently removed. (b) We measured binaural interaction functions under normal hearing conditions, which correspond to the blue curves in (a). The filled circles therefore show the best ILDs that we would expect to see for control animals and in juvenile-plugged ferrets that had adapted to the altered ILDs produced by a left earplug.
Supplementary Figure 2 Decoding ILDs by comparing the mean activity between neuronal populations on single trials.
(a) Within each hemisphere, there exists a mixture of units with opposing ILD preferences. Each symbol represents a unit; symbol color denotes ILD preference; symbol shape indicates left or right A1. (b) Decoders classified stimuli by comparing activity between two neuronal populations (P1, P2). Neurons were assigned to populations on the basis of either the hemisphere in which they were recorded (hemispheric decoder) or their ILD preference (opponent-process decoder). (c) (1) On trial i, the firing rate for each neuron j belonging to a particular population k (ri,j,k, gray dots) was averaged across neurons (black). This value represents the mean activity of population k on trial i (rPopi,k). In the test phase, the difference in population activity on a single trial (Δi) was calculated by subtracting the values of rPopi,k corresponding to each population. (2) Δi was then compared with a lookup table to decode ILD, which was constructed by first averaging, for each ILD, the value of rPopi,k across trials. This produced a trial-averaged rate-ILD function for each population (fk); these were then subtracted from one another to produce the lookup table. This lookup table therefore recorded the trial-averaged (TA) difference in population activity as a function of ILD: ΔTA(ILD). (d) Test ILDs were decoded for a single trial i by computing Δi, and then identifying the ILD associated with the most similar value of ΔTA.
(a-c) Population-averaged rate-ILD functions (a) and performance of the hemispheric (b) and opponent-process decoders (c) under normal hearing conditions. Plotting conventions are identical to those in Fig. 3a-c. (d) Population-averaged rate-ILD functions (plotted as in a) of control ferrets with a virtual earplug in the left ear.
Supplementary Figure 4 Intersection between right- and left-hemisphere binaural interaction curves does not fully shift into the ILD range experienced with an earplug.
Binaural interaction curves for juvenile-plugged animals are replotted from Fig. 2e. Shaded regions indicate the expected physiological ILD range experienced by ferrets at 15 kHz under left-ear plugged or normal hearing conditions. Data shown in Fig. 3 were obtained using a virtual earplug and therefore correspond to the plugged ILD range. For example, an ILD of 0 dB in Fig. 3 corresponds to an ILD of ~45 dB both here and in Fig. 2d,e. Arrows indicate the expected intersection between the left- and right-hemisphere binaural interaction curves for varying amounts of adaptation, either with no adaptation (yellow), following the partial adaptation seen in juvenile-plugged ferrets (light green), or following complete adaptation in which units shift their tuning to fully compensate for the effects of the earplug (dark green). In juvenile-plugged ferrets, the intersection between the binaural interaction curves does not fully shift into the range of ILDs experienced under plugged conditions, which may contribute to the failure of the hemispheric model (Fig. 3). Consequently, the hemispheric model may prove more successful in situations where neurons are able to shift their tuning to fully compensate for the effects of hearing loss.
Supplementary Figure 5 Effect of population size and noise correlations on the performance of population decoders.
(a-c) Mean normalized error is shown for the hemispheric (a) and opponent-process (b,c) decoders as a function of population size (i.e. the number of units used by the decoder to judge stimulus ILD). Results are shown for the opponent-process decoder using data recorded either from both hemispheres (b) or from a single hemisphere, with performance averaged across both hemispheres (c). Exponential functions (lines) have been fitted to the raw data (symbols) in a and b; for comparison purposes, the exponential fits in b are replotted in c. Results are shown for decoders applied to data that either contain (filled symbols, solid lines) or do not contain (unfilled symbols, dotted lines) noise correlations (i.e. stimulus-independent interneuronal correlation). Color denotes group type (controls with normal hearing, blue; controls with a virtual earplug, red; juvenile-plugged ferrets with a virtual earplug, black). Maximum population size is constrained separately for each group by the number of units recorded for different sub-population types. Overall, errors are reduced as population size is increased, but an asymptotic level of performance is reached when population size reaches approximately ~200 units. Decoder performance is also improved when noise correlations are absent. The hemispheric decoder performs very poorly for virtual earplug conditions, but can decode ILD accurately for controls with normal hearing (a). Better performance is achieved by the opponent-process decoder (b,c). This is particularly the case in both the control and juvenile-plugged animals with a virtual earplug, but smaller errors were also obtained in the controls under both hearing conditions. This pattern of results is robust to both noise correlations as well as changes in population size. Similar results are also observed when the opponent-process decoder is applied to data obtained from a single hemisphere, although restricting analyses to a single hemisphere necessarily reduces the maximum population size (c).
(a) In Supplementary Figure 2, the opponent-process decoder required neurons to be assigned labels based on their ILD preference. We therefore built a very similar model that did not need to be given these labels. The structure of this new model differed primarily in the way population activity was averaged and involved replacing step (1) in Supplementary Figure 2. In particular, rPopi,k can be thought of as the output of a decision unit Pk that computes the weighted mean of activity across all neurons included in the model (referred to as the input layer) on a single trial i. rPopi,k is therefore determined by the weight it gives to each neuron j in the input layer (wj,k). We created the model with two decision units (P1, P2), each of which initially gave equal weight to all neurons in the input layer. In the training phase, each decision unit received a 'teaching signal' (Ri,k) that varied as a function of space, but with opposite spatial preferences assigned to P1 and P2. On each simulated trial i, ri,j was determined by randomly sampling from the set of firing rates recorded for neuron j in response to a particular stimulus. Ri,k for each decision unit was set by (i) identifying the target location that corresponded to the ILD used on that trial and (ii) reading off the associated value of Rk (dashed lines). With the top layer fixed, each weight wj,k was adjusted by a learning rule: , where R̄ indicates the mean teaching signal across trials. This is equivalent to a simple Hebbian mechanism when the decision unit activity rPopi,k is dominated by the teaching signal (i.e. rPopi,k ≍ Ri,k) (Dayan, P. & Abbott, L.F. Theoretical neuroscience: computational and mathematical modeling of neural systems (MIT Press, Cambridge, Mass., 2001). (b) During the training phase, the weights between each input layer neuron and each decision unit (dots; top, middle) became increasingly correlated with the ILD slopes of input-layer neurons, but opposite in sign, i.e. corresponding to either left-ear greater or right-ear greater ILDs (bottom).
(a) Following the learning phase, each decision unit Pk tends to preferentially receive inputs with non-zero weights from neurons in the input layer that have similar ILD selectivity (denoted by color; only connections with non-zero weight are shown). The two decision units (P1, P2) show opposite preferences. (b) Cross-validated performance of the opponent-process model provided a good fit to behavioral data when using learned weights (magenta), and resembles that of our original opponent-process model in which the population labels for each neuron were simply given to the model (black). For comparison purposes, data are replotted from Fig. 3d. The critical components of this weight-learning model are (i) the teaching signal, which could be provided by the visual system during development, and (ii) the learning rule, which is consistent with a simple Hebbian mechanism that enables high pre-synaptic firing rates to produce strengthening of connections when post-synaptic firing rates are also high, and weakening of connections when post-synaptic firing rates are low.
About this article
Cite this article
Keating, P., Dahmen, J. & King, A. Complementary adaptive processes contribute to the developmental plasticity of spatial hearing. Nat Neurosci 18, 185–187 (2015). https://doi.org/10.1038/nn.3914
Microsecond interaural time difference discrimination restored by cochlear implants after neonatal deafness
Children With Congenital Unilateral Sensorineural Hearing Loss: Effects of Late Hearing Aid Amplification—A Pilot Study
Ear & Hearing (2020)
PLOS Computational Biology (2020)
Physics of Life Reviews (2019)
Frontiers in Neuroscience (2019)