advances.sciencemag.org/cgi/content/full/5/10/eaax2211/DC1

Supplementary Materials for

Descending pathways mediate adaptive optimized coding of

natural stimuli in weakly electric fish

Chengjie G. Huang, Michael G. Metzen, Maurice J. Chacron*

*Corresponding author. Email: maurice.chacron@mcgill.ca

Published 30 October 2019, Sci. Adv. 5, eaax2211 (2019)

DOI: 10.1126/sciadv.aax2211

This PDF file includes:

Fig. S1. Measures of neuronal activity are not significantly altered during presentation of adaptation stimuli characterized by a power law exponent of –2. Fig. S2. Measures of neuronal activity are not significantly altered during presentation of adaptation stimuli characterized by a power law exponent of 0. Fig. S3. Adaptive optimized coding cannot be predicted from changes in neuronal tuning to sinusoidal stimuli alone. Fig. S4. Adaptive optimized coding cannot be predicted from changes in neuronal tuning to adaptation stimuli alone. Fig. S5. Adaptive optimized coding can be predicted from changes in both neuronal tuning and variability during adaptation stimulus presentation. Fig. S6. Behavioral gains as a function of frequency. Fig. S7. Best-fit power law exponents for stimulus statistics are largely independent of the threshold used to separate low and high velocities. Fig. S8. Neuronal response statistics are not changing after lesioning forebrain. Fig. S9. Neuronal response statistics are not altered by KET injection.

Supplementary Materials

Fig. S1. Measures of neuronal activity are not significantly altered during

presentation of adaptation stimuli characterized by a power law exponent of –2. (A)

Population-averaged firing rate (FR, left), coefficient of variation (CV, middle), and burst

fraction (BF, right) early (black) and late (green) during stimulus presentation. No

significant changes were observed (FR: P = 0.12; CV: P = 0.43; CV: P = 0.10; n = 14,

Wilcoxon signed rank tests). (B) Population-averaged ISI distribution early in stimulus

presentation. (C) Population-averaged ISI distribution late in stimulus presentation. The

population-averaged ISI distributions were not significantly different (P = 0.78;

Kolmogorov-Smirnov test).

Fig. S2. Measures of neuronal activity are not significantly altered during

presentation of adaptation stimuli characterized by a power law exponent of 0. (A)

Population-averaged firing rate (FR, left), coefficient of variation (CV, middle), and burst

fraction (BF, right) early (black) and late (orange) during stimulus presentation. No

significant changes were observed (FR: P = 1.00; CV: P = 0.58; CV: P = 0.37; n = 11,

Wilcoxon signed rank tests). (B) Population-averaged ISI distribution early in stimulus

presentation. (C) Population-averaged ISI distribution late in stimulus presentation. The

population-averaged ISI distributions were not significantly different (P = 0.48;

Kolmogorov-Smirnov test).

Fig. S3. Adaptive optimized coding cannot be predicted from changes in neuronal

tuning to sinusoidal stimuli alone. (A) Schematic representation illustrating the stimulus

(left) and response (right) relation in order to give a tuning function (middle). (B)

Schematic predicting that optimized coding occurs because neuronal tuning (middle) is

matched to the stimulus statistics (left) to give a response whose power is independent of

frequency (i.e., white, right). (C) Left: Population-averaged neuronal sensitivity (i.e.,

tuning curve) obtained to sinusoidal stimuli as a function of frequency before (black) and

after adaptation with the stimulus characterized by a power law exponent of -2 (green).

Inset: Box-plots showing the population-averaged best-fit power-law exponents for before

(black) and after adaptation (green). Middle: Predicted (dashed) and actual (solid) neural

response power spectral density of an example PCell as a function of envelope frequency

according to the theory shown in (B). Right: Plot of predicted against actual whiteness

indices of our dataset. Inset: Box-plots showing population-averaged values of actual

(black) and predicted (green) whiteness indices according to theory. (D, E) Same as (B,

C), but for the adaptation stimulus characterized by a power law exponent of 0.

Fig. S4. Adaptive optimized coding cannot be predicted from changes in neuronal

tuning to adaptation stimuli alone. (A) Schematic representation illustrating the relation

between the stimulus (left) and the neuronal response (right) in order to give rise to a

specific transfer function (middle). (B) Schematic predicting that optimized coding occurs

because neuronal tuning (middle) is matched to stimulus statistics (left) to give a response

whose power is independent of frequency (right). (C) Left: Actual (solid green) and

predicted (dashed green) neural response power spectral density from the tuning to the

adaptation stimulus characterized by a power law exponent of -2. Right: Plot of predicted

against actual whiteness indices of our dataset. Inset: Box-plots showing population-

averaged values of actual (black) and predicted (green) white whiteness according to

theory. (D, E) Same as (B, C), but for the adaptation stimulus characterized by a power

law exponent of 0.

Fig. S5. Adaptive optimized coding can be predicted from changes in both neuronal

tuning and variability during adaptation stimulus presentation. (A) Schematic

predicting that optimized coding (right) occurs because variability (left) and neuronal

tuning (middle right) are matched to stimulus statistics (middle left), thereby leading to a

neural response power that is independent of frequency (right). (B) Left: Actual (solid

green) and predicted (dashed green) neural response power spectral density for an example

PCell as a function of envelope frequency after adaptation to a stimulus characterized by a

power law exponent of -2. Right: Plot of predicted against actual whiteness indices. Inset:

Box-plots showing the population-averaged whiteness index values of actual (black) vs.

predicted (green) according to theory with variability added. (C, D) Same as (A, B), but

for the adaptation stimulus characterized by a power law exponent of 0.

Fig. S6. Behavioral gains as a function of frequency. (A) Population-averaged

behavioral gain as a function of frequency before and after adaptation for stimuli

characterized by power law exponents of -2. The gain after adaptation was significantly

higher for frequencies >= 0.5 Hz (0.05 Hz: χ2: 0.01; P = 0.94; 0.1 Hz: χ

2: 1.00; P = 0.34;

0.25 Hz: χ2: 1.42; P = 0.23; 0.5 Hz: χ

2: 6.89; P = 8.7 * 10

-3; 0.75 Hz: χ

2: 9.02; P = 2.7 *

10-3

; 1 Hz: χ2: 9.02; P = 2.7 * 10

-3; Kruskal-Wallis ANOVA). (B) Population-averaged

behavioral gain as a function of frequency before and after adaptation for stimuli

characterized by power law exponents of 0. The gain was significantly higher for

frequencies <= 0.5 Hz (0.05 Hz: χ2: 8.80; P = 0.03; 0.1 Hz: χ

2: 5.11; P = 0.02; 0.25 Hz: χ

2:

4.05; P = 0.04; 0.5 Hz: χ2: 5.30; P = 0.02; 0.75 Hz: χ

2: 1.82; P = 0.18; 1 Hz: χ

2: 1.30;

P = 0.25; Kruskal-Wallis ANOVA).

Fig. S7. Best-fit power law exponents for stimulus statistics are largely independent

of the threshold used to separate low and high velocities. (A) Left: Power spectral

density of the as a function of stimulus frequency for the full (black), low (dark blue), and

high (light blue) velocity stimuli with shuffled segments. Right: Whisker box-plots

showing the best-fit power-law exponents for the corresponding envelopes (αlow = -

0.72 ± 0.13; P = 0.10; αhigh = -0.85 ± 0.10; P = 0.32; t-tests). (B) Box-plots showing the Δ

power-law exponent as a function of velocity threshold. “*” indicates significance to the

exponent of the full envelope at the P = 0.05 level (t-test).

Fig. S8. Neuronal response statistics are not changing after lesioning forebrain. (A)

Population-averaged firing rate (FR, left), coefficient of variation (CV, middle), and burst

fraction (BF, right) before lesioning forebrain (gray), as well as early (black) and late

(green) during stimulus presentation. Lesioning forebrain did not alter neuronal responses

(FR: P = 0.53; CV: P = 0.84; BF: P = 0.55; n = 8, Kruskal-Wallis tests) and no significant

changes were observed between early and late conditions (FR: P = 5.5*10-2

; CV: P =

0.55; BF: P = 0.20; n = 8, Wilcoxon signed rank tests). (B) Population-averaged ISI

distribution early in stimulus presentation. (C) Population-averaged ISI distribution late in

stimulus presentation. The population-averaged ISI distributions were not significantly

different (P = 0.48; Kolmogorov-Smirnov test). (D) Left: Actual (solid green) and

predicted (dashed green) neural response power spectral density for an example PCell as a

function of envelope frequency. Right: Plot of predicted against actual whiteness indices

obtained from data for all three conditions. Inset: Box-plots showing the population-

averaged whiteness index values of actual (black) vs. predicted (green) according to

theory with variability. (E) Population-averaged behavioral gain as a function of

frequency under control conditions before (gray) and after lesion (black), and after

adaptation (green). No significant changes in gain were observed across frequencies

between the control measurements and after adaptation (0.05 Hz: χ2: 3.76; P = 0.15;

0.1 Hz: χ2: 3.02; P = 0.22; 0.25 Hz: χ

2: 2.39; P = 0.30; 0.5 Hz: χ

2: 1.50; P = 0.47; 0.75 Hz:

χ2: 1.98; P = 0.37; 1 Hz: χ

2: 2.05; P = 0.36; Kruskal-Wallis ANOVA).

Fig. S9. Neuronal response statistics are not altered by KET injection. (A) Population-

averaged firing rate (FR, left), coefficient of variation (CV, middle), and burst fraction

(BF, right) before injecting KET (gray), as well as early (black) and late (green) during

stimulus presentation after KET injection. Injecting KET did not alter neuronal responses

(FR: P = 0.83; CV: P = 0.85; BF: P = 0.46; n = 8, Kruskal-Wallis tests) and no significant

changes were observed between early and late conditions (FR: P = 1.00; CV: P = 0.38;

BF: P = 0.11; n = 8, Wilcoxon signed rank tests). (B) Population-averaged ISI distribution

early in stimulus presentation. (C) Population-averaged ISI distribution late in stimulus

presentation. The population-averaged ISI distributions were not significantly different (P

= 0.16; Kolmogorov-Smirnov test). (D) Left: Actual (solid green) and predicted (dashed

green) neural response power spectral density for an example PCell as a function of

envelope frequency. Right: Plot of predicted against actual whiteness indices obtained

from data for all three conditions. Inset: Box-plots showing the population-averaged

whiteness index values of actual (black) vs. predicted (green) according to theory with

variability added. (E) Population-averaged behavioral gain as a function of frequency

under control conditions before ketanserin injection (gray), after ketanserin injection but

before adaptation (black), and after adaptation (green). No significant changes in gain

were observed across frequencies between the control measurements and after adaptation

(0.05 Hz: χ2: 0.36; P = 0.85; 0.1 Hz: χ

2: 0.32; P = 0.85; 0.25 Hz: χ

2: 0.04; P = 0.98; 0.5

Hz: χ2: 0.97; P = 0.62; 0.75 Hz: χ

2: 1.00; P = 0.61; 1 Hz: χ

2: 0.52; P = 0.77; Kruskal-

Wallis ANOVA).