An emergent population code in primary auditory cortex supports selective attention to spectral and temporal sound features

An emergent population code in primary auditory cortex supports selective attention to spectral and temporal sound features Downer Verhein Rapone O’Connor Sutter

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

An emergent population code in primary auditory cortex supports selective attention to spectral and temporal sound features 2020 bioRxiv 10.1101/2020.03.09.984773 Downer2020-an-emergent-population-code-in-primary-audit Downer2020-an-emergent-population-code-in-primary-audit.txt, the project's staged copy b1f28d3ab23346e44c9eda973a2ea93629df2c78 not-redistributable neuroscience

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them mechanism Quotation withheld because this record is not cleared for redistribution. feature-selective auditory attention Quotation withheld because this record is not cleared for redistribution. electrophysiology electrophysiology Quotation withheld because this record is not cleared for redistribution. mechanisms Because feature tuning co-varies within single neurons, a gain applied to individual cells cannot separate the two features; only re-weighting neurons at the population level recovers a feature-specific signal. Quotation withheld because this record is not cleared for redistribution. mechanisms Mixed selectivity cancels the average effect of feature-specific gain, which is why single-neuron measures show nothing while the population does. Quotation withheld because this record is not cleared for redistribution. mechanisms The attentional benefit is carried by higher-order correlations among neurons, i.e. it is an emergent property of the ensemble rather than a sum of single-cell effects. Quotation withheld because this record is not cleared for redistribution. mechanisms Population coding unfolds in stages over the stimulus: a fast feature-general change detection followed by finer feature discrimination. Quotation withheld because this record is not cleared for redistribution. The analysis uses only firing rate, not spike timing. Quotation withheld because this record is not cleared for redistribution. Because many bandwidth targets were also amplitude-modulated, phase locking could signal a target in both attention conditions, so the task may not have been demanding enough to reveal attentional effects on spike timing. Quotation withheld because this record is not cleared for redistribution. Neurons were pooled across sessions and trials resampled, so real correlated and intrinsic variability had to be reintroduced by modelling rather than measured simultaneously. Quotation withheld because this record is not cleared for redistribution. The task may be too easy to force attention-related changes in A1 temporal dynamics. Quotation withheld because this record is not cleared for redistribution. The two levels of analysis disagree on the same data: attention does not enhance the single-neuron representation, while the population analysis shows enhanced encoding. what this limits: a single-neuron measurement used to decide whether attention enhances sensory encoding: in these data it does not, while the population-level analysis of the same recordings shows that it does Quotation withheld because this record is not cleared for redistribution. The physical stimulus differs between animals, because the values were tailored to each subject's own psychophysical thresholds rather than fixed. what this limits: pooling or comparing the two subjects at 'the same' stimulus level: the AM and ∆BW values were tailored per subject to that subject's own psychophysical thresholds Quotation withheld because this record is not cleared for redistribution. The condition labels are defined by a sensitivity criterion, not by a physical quantity: 'near threshold' means the level at which d-prime equals 1. what this limits: the near/above-threshold conditions as physical stimulus levels: they are defined by the subject's measured sensitivity, at d-prime = 1, ~1.2 and >1.5 Quotation withheld because this record is not cleared for redistribution. An inclusion decision was made and checked, and the check is reported rather than the criterion being justified in advance. what this limits: the unit count in this study: 58 single units held for only two blocks are included, and the authors report that removing them does not qualitatively change the results Quotation withheld because this record is not cleared for redistribution. per-subject psychophysical detection thresholds measured beforehand and used to set the stimulus values for single-unit recording, so neural responses are read against each animal's own perceptual scale Quotation withheld because this record is not cleared for redistribution. the AM frequency presented on each trial chosen AT RANDOM rather than adaptively eliminates: SAMPLING BIAS TOWARDS THE CELLS THE EXPERIMENT IS ABOUT - searching for a frequency that drives the neuron finds AM-sensitive cells and quietly leaves the bandwidth-sensitive ones out of the population being described · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. population code population code Quotation withheld because this record is not cleared for redistribution. single-neuron representation single-neuron representation Quotation withheld because this record is not cleared for redistribution. psychophysical threshold psychophysical threshold Quotation withheld because this record is not cleared for redistribution. conditions conditions Quotation withheld because this record is not cleared for redistribution. variables variables Quotation withheld because this record is not cleared for redistribution. sensory representation - AT THE SINGLE-NEURON LEVEL: not enhanced by attention in these data, which is the paper's own negative result Quotation withheld because this record is not cleared for redistribution. sensory representation - AT THE POPULATION LEVEL: enhanced, tracking behaviour, and described as an emergent code NOT MEASURABLE at the single-neuron level Quotation withheld because this record is not cleared for redistribution. psychophysical threshold - defined as the modulation level at which the subject's measured d-prime equals 1, so the condition is fixed by sensitivity rather than by the stimulus Quotation withheld because this record is not cleared for redistribution. `conditions` / `variables` - THE TWO MOST GENERIC WORDS IN EXPERIMENTAL SCIENCE, BOTH STIPULATED LOCALLY AND IN ONE BREATH. `in the present study we refer to “conditions” as distinct combinations of AM, ∆BW and context ... We refer to “variables” as the three dimensions along which all conditions vary.` Neither word is being used oddly - both are being pinned, because the analysis needs 26 of one and 3 of the other and the ordinary senses would collapse them. `in the present study` is the paper marking the scope of its own stipulation, which is exactly the marker a cross-paper corpus needs and almost never gets. A tag joining this record's `variables` to another paper's `variables` would join 3 dimensions to a table of measurements. Quotation withheld because this record is not cleared for redistribution. contradicts → Niwa et al. (2012b), Schwartz and David (2017) and von Trapp et al. (2016), which reported attention-related changes in A1 single-neuron tuning Quotation withheld because this record is not cleared for redistribution. builds-on → Mante et al. (2013) and Rigotti et al. (2013) population-level analyses of prefrontal cortex Quotation withheld because this record is not cleared for redistribution. single-neuron representation —contrasted-with→ population code single-neuron representation population code strength of the relation: 1 Quotation withheld because this record is not cleared for redistribution. variables —distinct-from→ conditions variables conditions the relation itself is separately evidenced · strength of the relation: 1 · note: The paper separates them by construction - conditions are combinations, variables are the axes those combinations range over, 26 against 3. Cue: 'distinct' appears in the conditions clause and the whole span is one stipulation. · the relation stated: We refer to “variables” as the three dimensions along which all conditions vary Quotation withheld because this record is not cleared for redistribution. logistic regression logistic regression Quotation withheld because this record is not cleared for redistribution. multiple linear regression multiple linear regression Quotation withheld because this record is not cleared for redistribution. Wilcoxon signed rank test Wilcoxon signed rank test Quotation withheld because this record is not cleared for redistribution. ROC / signal detection analysis ROC / signal detection analysis Quotation withheld because this record is not cleared for redistribution. bootstrap bootstrap Quotation withheld because this record is not cleared for redistribution. Spearman correlation Spearman correlation Quotation withheld because this record is not cleared for redistribution. t-test t-test Quotation withheld because this record is not cleared for redistribution. support vector machine with held-out test set support vector machine with held-out test set Quotation withheld because this record is not cleared for redistribution. Fano factor (variance-to-mean ratio) Fano factor (variance-to-mean ratio) Quotation withheld because this record is not cleared for redistribution. principal component analysis principal component analysis Quotation withheld because this record is not cleared for redistribution. z-score standardisation z-score standardisation Quotation withheld because this record is not cleared for redistribution. linear regression linear regression Quotation withheld because this record is not cleared for redistribution. Individual A1 neurons carried mixed selectivity, encoding both the spectral and temporal sound features and the attention condition. Quotation withheld because this record is not cleared for redistribution. Attention produced no average enhancement of feature encoding in single A1 neurons. Quotation withheld because this record is not cleared for redistribution. Neurons were as likely to increase as to decrease their AM ROC area across attention conditions (p = 0.74). Quotation withheld because this record is not cleared for redistribution. Context had a significant main effect in more neurons than either acoustic feature. Quotation withheld because this record is not cleared for redistribution. After targeted dimensionality reduction, the population code decoded each sound feature better when that feature was attended. Quotation withheld because this record is not cleared for redistribution. The attentional advantage in the population code was large and highly reliable across simulated populations. Quotation withheld because this record is not cleared for redistribution. The observed attentional modulation exceeded that of every surrogate population in which higher-order correlations had been destroyed. Quotation withheld because this record is not cleared for redistribution. AM-axis population projections were significantly related to task performance (mean choice probability 0.573). Quotation withheld because this record is not cleared for redistribution. Both monkeys weighted the attended feature more heavily than the unattended one in their yes/no reports. Quotation withheld because this record is not cleared for redistribution. Attention did not alter spike timing: phase locking to the temporal envelope was unaffected by the attention condition. Quotation withheld because this record is not cleared for redistribution. Population trajectories began with an early non-selective course in which each stimulus projected to about the same region of the subspace. Quotation withheld because this record is not cleared for redistribution. targeted dimensionality reduction (TDR) targeted dimensionality reduction (TDR) Quotation withheld because this record is not cleared for redistribution. de-mixing de-mixing Quotation withheld because this record is not cleared for redistribution. feature-specific gain (FSG) feature-specific gain (FSG) Quotation withheld because this record is not cleared for redistribution. ROC area ROC area Quotation withheld because this record is not cleared for redistribution. choice probability (CP) choice probability (CP) Quotation withheld because this record is not cleared for redistribution. surrogate population surrogate population Quotation withheld because this record is not cleared for redistribution. task context task context Quotation withheld because this record is not cleared for redistribution. AM depth AM depth Quotation withheld because this record is not cleared for redistribution. spectral bandwidth spectral bandwidth Quotation withheld because this record is not cleared for redistribution. trial-by-trial spike count trial-by-trial spike count Quotation withheld because this record is not cleared for redistribution. behavioral report behavioral report Quotation withheld because this record is not cleared for redistribution. AM frequency AM frequency Quotation withheld because this record is not cleared for redistribution. noise correlations noise correlations Quotation withheld because this record is not cleared for redistribution. neuron-intrinsic variability neuron-intrinsic variability Quotation withheld because this record is not cleared for redistribution. 92 — single A1 neurons recorded, from two rhesus macaques Quotation withheld because this record is not cleared for redistribution. 0.151 — average increase in population ROC area for AM when AM was the attended feature Quotation withheld because this record is not cleared for redistribution. p = 0.73 and 0.74 — sign rank p-values showing no attentional change in single-neuron BW and AM sensitivity Quotation withheld because this record is not cleared for redistribution. 0.633 (p=1.12e-61) — mean choice probability for projections into the full three-dimensional task-variable subspace Quotation withheld because this record is not cleared for redistribution. 1000 — surrogate populations built with intact single-neuron tuning but destroyed higher-order correlations Quotation withheld because this record is not cleared for redistribution. 26 (13 x 2) — total experimental conditions, being 13 sound combinations crossed with two attention blocks Quotation withheld because this record is not cleared for redistribution. 47 total sessions, p = 8.98e-8 — behavioural sessions and the sign rank p-value showing attention shifted the animals' feature weighting Quotation withheld because this record is not cleared for redistribution. n = 434 pairs — simultaneously recorded neuron pairs used to estimate the noise correlations imposed on simulated populations Quotation withheld because this record is not cleared for redistribution. 65 dB SPL — calibrated presentation level of every sound stimulus at the outer ear Quotation withheld because this record is not cleared for redistribution. 400 ms — duration of each of the two sounds in a trial Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1101/2020.03.09.984773.