Electrophysiology and imaging during olfactory Pavlovian conditioning.

(A) Trial structure in Pavlovian conditioning task.

(B) Timeline for mouse training.

(C) Mean (+/− standard error of the mean (SEM)) lick rate across mice (n = 5) on each trial type for each odor set during electrophysiology sessions. CS50(r) and CS50(u) are rewarded and unrewarded trials, respectively. Inset: mean anticipatory licks (change from baseline) for the CS+ and CS50 cues for every session, color-coded by mouse. F (1, 66) = 36.6 for a main effect of cue in a two-way ANOVA including an effect of subject.

(D) Same as (C), for Day 3 imaging sessions (n = 5 mice). t(4) = −5.4 for a t-test comparing anticipatory licks on CS+ and CS50 trials.

(E) Neuropixels probe tracks labeled with fluorescent dye (red) in cleared brain (autofluorescence, green). AP, anterior/posterior; ML, medial/lateral; DV, dorsal/ventral. Allen CCF regions delineated in gray. Outline of prelimbic area in purple.

(F) Reconstructed recording sites from all tracked probe insertions (n = 44 insertions, n = 5 mice), colored by mouse.

(G) Sample histology image of lens placement. Visualization includes DAPI (blue) and GCaMP (green) signal with lines indicating cortical regions from Allen Mouse Brain Common Coordinate Framework.

(H) Location of all lenses from experimental animals registered to Allen Mouse Brain Common Coordinate Framework. Blue line indicates location of lens in (A). The dotted black line represents approximate location of tissue that was too damaged to reconstruct an accurate lens track. The white dotted line indicates PL borders.

(I) ML and DV coordinates of all neurons recorded in one example session, colored by region, and spike raster from example PL neurons.

(J) ROI masks for identified neurons and fluorescence traces from 5 example neurons.

Graded cue and lick coding across the recorded regions.

(A) Location of each recorded neuron relative to bregma, projected onto 1 hemisphere. Each neuron is colored by CCF region. Numbers indicate total neurons passing quality control from each region.

(B) Mean normalized activity of all neurons from each region, aligned to odor onset, grouped by whether peak cue activity (0 - 2.5 s) was above (top) or below (bottom) baseline in held out trials. Number of neurons noted for each plot.

(C) Example kernel regression prediction of an individual neuron’s normalized activity on an example trial.

(D) CS+ trial activity from an example neuron and predictions with full model and with cues, licks, and reward removed. Numbers in parentheses are model performance (fraction of variance explained).

(E) Coordinates relative to bregma of every neuron encoding only cues or only licks, projected onto one hemisphere.

(F) Fraction of neurons in each region and region group classified as coding cues, licks, reward, or all combinations of the three.

(G) Additional cue (left) or lick (right) neurons in region on Y-axis compared to region on X-axis as a fraction of all neurons, for regions with non-overlapping 95% confidence intervals (see Methods).

Widespread cue value coding.

(A) Normalized activity of an example value cell with increasing modulation for cues of higher value.

(B) For the same neuron, model-fit cue kernel for the original value model and with one of the 89 alternatively-permuted value models.

(C) Distribution of best model fits across all cue neurons. Light blue is value model, purple is trial type models, gray is untuned model, and the remaining models are dark blue.

(D) First principal component of all neurons best fit by the original value model or other trial type and untuned models.

(E) Fraction of neurons in each region and region group classified as value cells (blue) and other cue neurons (gray), as well as fraction (+/− 95% CI) estimated from a linear mixed effects model with random effect of session (see Methods). n.s. indicates overlapping 95% CI for all three region groups.

(F) Additional value cells in region on Y-axis compared to region on X-axis as a fraction of all neurons, for regions with non-overlapping 95% confidence intervals. * indicates non-overlapping 95% CI for all three region groups.

(G) Principal component most related to value for value cells from all regions (2nd component in ACA, 1st component in all others).

(H) Same as (F), for region groups.

(I) PL population activity 0 to 2.5 s from odor onset projected onto first 3 principal components, defined on odor set 1 activity.

(J) Normalized distance (+/− SD) from baseline firing in odor set 1 PCA space (first 3 components) for odor set 2 trial types for value cells (top) and other cue cells (bottom) for 5000 selections of neurons (see Methods).

A subset of value cells incorporate reward history.

(A) Coefficient weight (+/− standard error from model fit) for reward outcome on the previous 10 trials of any type (left) and on the previous 10 trials of the same cue type (right) for a linear model predicting the number of anticipatory licks on every trial of every session. Lick rates were normalized so that the maximum lick rate for each session was equal to 1. Colored lines are models fit to each individual mouse.

(B) Mean (+/− SEM) lick rate across mice (n = 5 mice) on trials binned according to value estimated from the lick model, incorporating recent reward history.

(C) Normalized activity of an example history value cell with increasing modulation for cues of higher value.

(D) For the same neuron, model-predicted activity with the original value model (left) and with trial-by-trial value estimates from the lick model (right).

(E) The activity of all cells in each category projected onto the coding dimension maximally separating CS− and CS+ for CS50 trials binned by value estimated from the lick model.

(F) The mean (+/− std across 5000 bootstrapped selections of neurons) activity (1 - 2.5 s from odor onset) along the coding dimension maximally separating CS− and CS+ for CS50 trials binned by value estimated from the lick model. * = p < 10−7 comparing highest and lowest value CS50 trials (other categories p > 0.23).

(G) Fraction of neurons in each region and region group classified as history value (light blue), non-history value (blue), and other cue neurons (gray), as well as estimated fraction (+/−95% CI) with random effect of session (see Methods). * indicates non-overlapping 95% CI for PFC and olfactory regions.

(H) Fraction of value neurons in each region group with history effect and estimated fraction (+/− 95% CI) with random effect of session. * indicates non-overlapping 95% CI for olfactory compared to PFC and motor regions.

Acquisition of conditioned behavior and cue encoding in PFC.

(A) Training schedule for 5 of the mice in the imaging experiment. An additional 3 were trained only on odor set 1.

(B) Mean (+/− SEM) licking on early (first 60) and late (last 60) trials from day 1 of odor set 1 (n = 8 mice).

(C) Mean (+/− SEM) baseline-subtracted anticipatory licks for early and late trials from each day of odor set 1. Thin lines are individual mice (n = 8 mice).

(D) Standard deviation of fluorescence from example imaging plane.

(E) Normalized activity of each pixel following CS+ presentation in early and late day 1 trials.

(F) Normalized deconvolved spike rate of all individual neurons for early and late trials on day 1.

(G) Proportion of neurons classified as coding cues, licks, rewards, and all combinations for each third of day 1.

(H) Mean(+/− SEM) unique variance explained by cues, licks, and rewards for neurons from each mouse. Thin lines are individual mice. Unique variance was significantly different across session thirds for cues (F (2, 21) = 3.71, p = 0.04) but not licks (F (2, 21) = 0.37, p = 0.69) or reward (F (2, 21) = 0.65, p = 0.53, n = 8 mice, one-way ANOVA).

(I) Mean (+/− SEM) normalized deconvolved spike rate for cells coding cues, licks, both, or neither on early and late trials, sorted by whether peak cue activity (0 - 2.5 s) was above (top) or below (bottom) baseline for late trials.

Cue and lick coding is stable across days.

(A) Standard deviation fluorescence from example imaging plane.

(B) Masks (randomly colored) for all tracked neurons from this imaging plane.

(C) Deconvolved spike rate on every CS+ trial from all three days of odor set 1 for an example neuron. Vertical dashed line is reward delivery. Color axis as in (D).

(D) Normalized deconvolved spike rate for all tracked neurons on all three days of odor set 1.

(E) Fraction of tracked neurons coding cues, licks, rewards, and their combinations on day 3.

(F) Model performance when using day 3 models to predict the activity of individual neurons across odor set 1 training, plotted as mean (+/− SEM) correlation between true and predicted activity across mice. Thin lines are individual mice. Performance was greater than shuffled data at all time points (p < 0.0001) except early day 1 (p = 0.21, Bonferroni-corrected, n = 8 mice).

(G) Mean (+/− SEM) unique cue, lick, and reward variance for cells classified as coding cues, licks, both, or neither on day 3. Day 3 cue cells had increased cue variance on day 2 (p < 10−7, see Methods) and 1 (p < 0.03) relative to lick and reward variance. Same pattern for lick cells on day 2 (p < 0.0001) and day 1 (p < 0.01).

Cue and lick coding in separately trained odor sets.

(A) Normalized activity of all pixels in the imaging plane following CS+ presentation on day 3 of each odor set.

(B) Fraction of neurons coding for cues, licks, rewards, and their combinations for day 3 of each odor set.

(C) Mean(+/− SEM, across mice) correlation between activity predicted by odor set 1 models and true data, for real (black) and trial shuffled (gray) activity. Thin lines are individual mice. F (1, 16) = 3.2, p = 0.09 for main effect of odor set, F (1, 16) = 135, p < 10−8 for main effect of shuffle, F (1, 16) = 2.2, p = 0.16 for interaction, n = 5 mice, two-way ANOVA.

(D) Mean(+/− SEM, across mice) unique cue, lick, and reward variance for cells classified as coding cues, licks, both, or neither for odor set 1. For each category, odor set 1 unique variance preference was maintained for odor set 2 (p < 0.04) except for both cells, for which lick and reward variance were not different in odor set 2 (p = 0.22, Bonferroni-corrected, n = 5 mice).

(E) Distribution of best model fits across all cue neurons, and the first principal component of the activity of all neurons best fit by the top 3 models.

Anticipatory licking during the electrophysiology sessions.

(A) Mean anticipatory licks (change from baseline) for the CS+ and CS50 from odor set 1 (left) and 2 (right) for every session, color-coded by mouse. F (1, 66) = 32.07 and F (1, 66) = 26.93 in each odor set for a main effect of cue in a two-way ANOVA including an effect of subject.

(B) As above, for the CS+ and CS− from odor set 1 (left) and 2 (right). F (1, 66) = 433.1 and F (1, 66) = 574.6 in each odor set for a main effect of cue in a two-way ANOVA including an effect of subject.

(C) As above, for the CS50 and CS− from odor set 1 (left) and 2 (right). F (1, 66) = 252.3 and F (1, 66) = 450.1 in each odor set for a main effect of cue in a two-way ANOVA including an effect of subject.

Similar neural activity in prelimbic area using electrophysiology and imaging.

(A) Heatmap of the normalized activity of each neuron recorded with electrophysiology in PL, aligned to each of the 6 odors. All columns sorted by mean firing 0 - 1.5 s following odor onset for odor set 1 CS+ trials.

(B) As in (A), for all neurons imaged in PL on day 3 of each odor set.

(C) The score from the first 4 principal components of the normalized activity presented in (A), with variance explained in parentheses.

(D) As in (C), for the activity in (B).

Task-related neural activity across brain regions.

(A) For each of the 5 mice in the electrophysiology experiment, the number of neurons recorded in each region.

(B) Heatmap of the normalized activity of each neuron (n = 51 trials per cue). All columns sorted by region and then by mean firing 0 - 1.5 s following odor onset for odor set 1 CS+ trials.

(C) Mean (+/− SEM) activity of neurons from 4 regions aligned to each cue type, grouped by whether peak cue activity (0 - 2.5 s) was above (top) or below (bottom) baseline in held out trials.

Identification of cue and lick cells with GLM.

(A) Mean variance explained (fraction) by linear models in each region for each session (x) and the mean (+/− SEM) across those sessions.

(B) Mean (+/− SEM) activity of neurons encoding cues, licks, both, or neither aligned to each cue type, grouped by whether peak cue activity (0 - 2.5 s) was above (top) or below (bottom) baseline in held out trials.

(C) Normalized activity of every neuron encoding cues, licks, or both, aligned to CS+ onset, sorted by mean firing 0 - 1.5 s following odor onset.

(D) Mean (+/− SEM) activity of neurons encoding cues or licks, grouped as in (B), on CS50 trials, divided into rewarded (lighter colors) or unrewarded (darker colors) trials.

Comparing proportions of cue and lick neurons across regions.

(A) Fraction of neurons in each region classified as coding cues (left), licks (middle), or both (right), as well as estimated fraction(+/−95% CI) with random effect of session (see Methods).

(B) Additional cue/lick/both cells in region on Y-axis compared to region on X-axis as a fraction of all neurons, for regions with non-overlapping 95% confidence intervals.

(C) As in (A), for region groups.

(D) As in (B), for region groups.

Schematic of value model shuffles.

(A) For each of the 90 permutations of the Value model, the value taken on by the variable cue kernel on trials corresponding to each of the 6 cue types. Value is determined in units of predicted anticipatory licks, from 0 to 1 (maximum number of anticipatory licks made). Additionally, there is an Untuned model, where all cues take on the same value.

Additional analysis of odor coding schemes.

(A) Normalized activity of every value neuron, sorted by mean firing 0 - 1.5 s following odor set 1 CS+ onset.

(B) Normalized activity of every trial type neuron, sorted by model and then by mean firing 0 - 1.5 s following odor set 1 CS+ onset.

(C) Projecting the activity of all trial type and value cells onto the coding dimensions maximally separating CS− and CS+ (x-axis) and CS− and CS50 (y-axis). Solid line is activity during cue, dashed line is activity following reward delivery. X marks baseline activity.

(D) For the odor set 1 projection, distribution of 5000 bootstrapped distances between baseline activity (prior to odor onset) and the CS− representation (0, 0). Value cells were closer to CS− at baseline than trial type cells.

(E) For the odor set 1 projection, distribution of 5000 bootstrapped angles between CS+ and CS50 vectors (baseline to peak). Value cells had a smaller angle than trial type cells.

Relative proportions of value and trial type cells across regions.

(A) Additional cue value (left) or trial type (right) neurons in region on Y-axis compared to region on X-axis as a fraction of all neurons, for regions with non-overlapping 95% confidence intervals.

(B) As in (A), for region groups.

Value coding as a proportion of cue cells.

(A) Fraction of cue neurons in each region classified as coding value (left) or trial type (right), as well as estimated fraction(+/−95% CI) with random effect of session (see Methods).

(B) Additional value/trial type cue neurons in region on Y-axis compared to region on X-axis as a fraction of all cue neurons, for regions with non-overlapping 95% confidence intervals.

(C) As in (A), for region groups.

(D) As in (B), for region groups.

Comparing PFC and striatum.

(A) Fraction of neurons in each region and region group classified as coding cues (left), licks (middle), or both (right), as well as estimated fraction(+/−95% CI) with random effect of session (see Methods).

(B) Fraction of neurons in each region and region group classified as coding value (left) or trial type (right), as well as estimated fraction(+/−95% CI) with random effect of session. Light gray bars are remaining cue neurons not in that category.