Over-representation of speech in older adults originates from early response in higher order auditory cortex

Transcription

1 Over-representation of speech in older adults originates from early response in higher order auditory cortex Christian Brodbeck, Alessandro Presacco, Samira Anderson & Jonathan Z. Simon

2 Overview 2 Puzzle Compared to young adults, older adults (60+) exhibit - Impaired auditory temporal processing - More difficulty comprehending speech, especially in challenging circumstances Yet, the speech envelope can be reconstructed more accurately from their cortical responses, recorded with MEG (Presacco et al., 2016)

3 Overview 2 Puzzle Compared to young adults, older adults (60+) exhibit - Impaired auditory temporal processing - More difficulty comprehending speech, especially in challenging circumstances Yet, the speech envelope can be reconstructed more accurately from their cortical responses, recorded with MEG (Presacco et al., 2016) Different possible explanations, for example Low level deficit, e.g. excitation-inhibition imbalance Recruitment of additional top-down resources Increased attention

4 Overview 2 Puzzle Compared to young adults, older adults (60+) exhibit - Impaired auditory temporal processing - More difficulty comprehending speech, especially in challenging circumstances Yet, the speech envelope can be reconstructed more accurately from their cortical responses, recorded with MEG (Presacco et al., 2016) Different possible explanations, for example Low level deficit, e.g. excitation-inhibition imbalance Recruitment of additional top-down resources Increased attention This talk Localize cortical responses to speech of young and older adults - Anatomy: localization in cortex - Time: latency at which information is contained

5 MagnetoEncephaloGraphy (MEG) 3 (a) (b) V I II B III IV V VI white matter (Hari & Parkkonen, 2015)

6 MagnetoEncephaloGraphy (MEG) 3 (a) (b) B V I II III IV Single trial (pure tone) V VI white matter (Hari & Parkkonen, 2015)

7 MagnetoEncephaloGraphy (MEG) 3 (a) (b) B V I II III IV Single trial (pure tone) V VI white matter Average (~100 trials) (Hari & Parkkonen, 2015)

8 Neural source localization 4

9 Neural source localization 4

10 Continuous Speech 5 Sound pressure Frequency (Hz) Time (s)

11 Linear convolution model 6 Convolve ( ) aresponsefunction withasparsesignal Response Stimulus

12 Linear convolution model 6 Convolve ( ) aresponsefunction withasparsesignal Resp. Stim. Response Stimulus The same response function with a more dense signal:...

13 Linear convolution model 7 Stimulus and response are known; find the best response function to produce the response from the stimulus: Response Stimulus

14 Linear convolution model 7 Stimulus and response are known; find the best response function to produce the response from the stimulus: Response Stimulus Estimated response function

15 Linear convolution model Stimulus Stimulus and response are known; find the best response function to produce the response from the stimulus: Response Estimated response function Actual response Predicted response 7

16 Linear convolution model 7 Stimulus and response are known; find the best response function to produce the response from the stimulus: Response Stimulus Actual stimulus Reconstructed stimulus "Decoder"

17 Methods 8 Design 60 s long audiobook excerpts, 3 repetitions each 2 excerpts were clean speech 4 excerpts with second speaker at different signal to noise ratios (SNRs; +3, 0, -3, -6) Participants 17 young adults (aged years) 15 older adults (aged years) - Clinically normal audiogram

18 Reconstructing speech envelope 9

19 Reconstructing speech envelope 9 "Decoder"

20 Decoding speech envelope 10 "Decoder" Older Younger Cortex: older > younger (Presacco, Simon, & Anderson, 2016)

21 Midbrain 11 Midbrain Older listeners have reduced frequency following response (FFR) Increased cortical responses not due to stronger input from midbrain Younger Older Midbrain: younger > older (Presacco, Simon, & Anderson, 2016)

22 Possible explanations 12 Low level change, e.g., excitation/inhibition imbalance Decrease in cortical inhibition could lead to stronger evoked responses - Reduction in inhibitory neurons in A1 (de Villers-Sidani et al., 2010) - Increased firing rates in A1 (Overton & Recanzone, 2016) Prediction: - Even low latency responses show enhancement in older listeners, e.g., 30 ms

23 Possible explanations 12 Low level change, e.g., excitation/inhibition imbalance Decrease in cortical inhibition could lead to stronger evoked responses - Reduction in inhibitory neurons in A1 (de Villers-Sidani et al., 2010) - Increased firing rates in A1 (Overton & Recanzone, 2016) Prediction: - Even low latency responses show enhancement in older listeners, e.g., 30 ms Top-down/strategic processing Higher level processes recruited to compensate for degraded input from the periphery - Recruitment of additional frontal and temporal regions for complex sentences (Peelle et al., 2010) Prediction: - Response enhancement delayed until longer latency responses, e.g., ms

24 Possible explanations 12 Low level change, e.g., excitation/inhibition imbalance Decrease in cortical inhibition could lead to stronger evoked responses - Reduction in inhibitory neurons in A1 (de Villers-Sidani et al., 2010) - Increased firing rates in A1 (Overton & Recanzone, 2016) Prediction: - Even low latency responses show enhancement in older listeners, e.g., 30 ms Top-down/strategic processing Higher level processes recruited to compensate for degraded input from the periphery - Recruitment of additional frontal and temporal regions for complex sentences (Peelle et al., 2010) Prediction: Attention - Response enhancement delayed until longer latency responses, e.g., ms Increased attention associated with stronger responses - Attention increases MEG response amplitudes with similar field topography (Woldorff et al., 1993) Prediction: - Similar spatial distributions of responses, but overall enhancement in older adults - More activity in core auditory areas

25 Encoding model 13

26 Encoding model 13

27 Statistics 14 Evaluate model predictions: Pearson correlation: r(predicted response, measured response) Bias-correction: Compute r of a temporally shuffled model Test for better r of the true model Significance test: Mass-univariate t-test Threshold-free cluster enhancement Max statistic distribution with 10,000 permutations

28 Localization 15 Heschl's gyrus (core auditory cortex) Superior temporal gyrus 0.12 old z(r) 0.00 young

29 Localization 15 Heschl's gyrus (core auditory cortex) Superior temporal gyrus < old.01 p z(r).05 old > young 0.00 young

30 Encoding model 16

31 Temporal response function 17 Temporal response function (TRF) Brain response to an elementary temporal feature in the stimulus Time axis: latency between acoustic feature and brain response TRF amplitude [normalized] * old young Time [ms]

32 Temporal response function 17 Temporal response function (TRF) Brain response to an elementary temporal feature in the stimulus Time axis: latency between acoustic feature and brain response TRF amplitude [normalized] * old young < Time [ms] p.05 old > young, ms

33 Summary 18 Over-representation of speech in older adults originates from early response in higher order auditory cortex Temporal lobe, outside of core auditory cortex Primarily affecting earliest cortical responses ~30 ms response Strategic/top-down processing - Latency too short Attention - Localized outside of core auditory cortex Low level change, e.g., excitation/inhibition imbalance - Short latency - Fast spread to areas outside core auditory cortex

34 Funding 19 NIH Grant P01-AG055365

35 Appendix

36 Mix of 2 speakers 21 ANOVA: Age (2) SNR (4) Main effect of Age old young < p 0.06 z(r) TRF amplitude [normalized] * old young SNR Time [ms]

37 MEG 22 Properties of MEG: Excellent temporal resolution (magnetic fields) Mediocre spatial resolution (inverse problem) - Relatively good estimation of center - Spatial dispersion Simulation (Minimum Norm Estimates): Current estimate