Research Article

Recurrent cortical circuits implement concentration-invariant odor coding

See allHide authors and affiliations

Science  14 Sep 2018:
Vol. 361, Issue 6407, eaat6904
DOI: 10.1126/science.aat6904
  • Whenever a mouse inhales, volatile molecules activate odorant receptors in the nose, evoking sequences of activity in the olfactory bulb.

    Bulb cells driven by the most specific receptors, which therefore best represent the odor stimulus (cheese), will always respond earliest. When this information is relayed to piriform cortex, activated principal neurons (red cells) recruit inhibitory neurons (green cells) that then suppress cortical responses to subsequent, less-specific olfactory bulb input (such as garlic, shoe, or flower), preserving the identity of the stimulus.

    ILLUSTRATION BY JULIA KUHL
  • Fig. 1 Concentration-invariant odor representations emerge in PCx.

    (A) Experimental schematic. Odor panel included four odors at a single concentration and two odors at four concentrations. (B) Example responses from simultaneously recorded pairs of (left) OB or (right) PCx cells to two odors at different concentrations. Responses are aligned to start of inhalation. (C) Percent of cells significantly activated by odors of increasing concentration (P < 0.05 rank-sum test, odor vs mineral oil) in the OB (red) or PCx (black, n = 5 simultaneous OB-PCx recordings, two odors, four concentrations). (D) (Left) PCA representation of OB pseudopopulation (n = 94 cells) response in a 330-ms window after inhalation to ethyl butyrate (blue) and hexanal (magenta) at different concentrations (0.03 to 1%, different shades). Dots represent responses on individual trials; ellipsoids are mean ± 1 SD. (Right) Relative population response distances in neural activity space projected onto the first three principal components. Distances were computed for each stimulus between trials of the same odor and concentration (repeat, n = 12 stimuli), different odors (Δ odor, n = 12 stimuli), or same odor and different concentration (Δ conc., n = 8 stimuli), and normalized to the average Δ odor distance. OB responses to different concentrations were as dissimilar as responses to different odors (one-sample t test versus mean of 1, P = 0.851). (E) As in (D), but for PCx pseudopopulation (n = 330 cells). PCx responses to different concentrations were more clustered than responses to different odors (P = 0.001).

  • Fig. 2 PCx predominantly responds to early OB inputs.

    (A) Example single-trial response to isoamyl acetate (0.3% v/v) in populations of simultaneously recorded OB and PCx cells. Negative-going respiration signal (top) indicates inhalation. Bold blue line marks start of first inhalation after odor onset. Thin blue line marks second inhalation. Cells in each population are sorted by trial-averaged response peak latency. (B) Example trial-averaged peristimulus time histograms (PSTHs) for populations in (A). Blue lines indicate inhalation times on all 15 trials. (C) Average PSTHs for same OB and PCx populations responding to three odors. Shading is SEM across cells. (D) PSTHs for all OB cell-odor pairs sorted by latency to peak show uniform tiling of sniff cycle. (E) Same as (D) but for PCx. Majority of PCx responses occur within 60 ms after inhalation. (F) Average PSTHs for all cell-odor responses at different concentrations (OB, n = 188; PCx, n = 664 cell-odor pairs; mean ± SEM). Gray shading indicates initial (0 to 60 ms) and sustained (100 to 300 ms) analysis windows. Dashed line indicates inhalation onset. (G) Normalized multiunit activity (MUA) rates during initial phase (n = 5 experiments, two odors, four concentrations) in OB versus PCx. MUA is determined by recombining individual cell responses. (Top) Average OB (red) and PCx (black) response across recordings and odors. MUA was normalized to baseline activity 1 s before odor. (Bottom) Each point is the average response of one simultaneously recorded OB-PCx population response pair. Shading indicates concentration. Cyan lines are linear fits across concentrations for each OB-PCx population response pair. Black line is the linear fit to all data. (H) As in (G) but for the sustained phase.

  • Fig. 3 Feedback inhibition shapes cortical odor responses.

    (A) Schematic of PCx circuit: FFIs in layer 1 receive OB input; principal cells in layer 2 provide recurrent excitatory input to other principal cells and to FBIs in layer 3. (B) Recording schematic. Light-responsive FFIs and FBIs in VGAT-ChR2 mice are differentiated by their depths relative to VGAT principal cells in layer 2, which are suppressed. (C) FFIs (magenta, n = 13), FBIs (teal, n = 46), and VGAT (black, n = 855) are classified by light-responsiveness and depth (dashed line). Dots with error bars are mean ± SEM. Light gray indicates unclassified light-responsive cells. (Top) Average waveform of each cell type (mean ± SEM). Scale bars, 0.5 ms, 0.1 mV. (D) Example light responses for one PC (black) and four cells classified as FFIs or FBIs (blue). (E) Example odor responses for cells in (D). (F) Average population PSTHs (mean ± SEM) for each cell type. (G) Normalized PSTHs for FBIs and VGAT cells. (H) Average population PSTHs for (left) VGAT, (middle) FFIs, and (right) FBIs responding to odors at increasing concentrations. (I) Normalized firing rates in response to increasing odor concentrations for each cell type (mean ± SEM).

  • Fig. 4 TeLC expression selectively abolishes recurrent excitation.

    (A) Schematic of circuit changes after TeLC expression in PCx principal cells. (B) Focal coinfection in PCx with ChR2 and either GFP or TeLC-GFP, followed by whole-cell recordings from uninfected cells. (C) Light-evoked synaptic responses are abolished by TeLC. Example light-evoked response from non-ChR2–expressing neurons in (top) GFP- or (bottom) TeLC-GFP–infected PCx. i Light-evoked EPSC amplitudes in control and TeLC-expressing PCx (control: 239 ± 68 pA, n = 11 cells from two mice; TeLC 35 ± 10 pA, n = 12 cells from three mice; unpaired t test, P = 0.0133). (D) Example recordings from an (left) uninfected and (right) TeLC-infected neuron in the same slice in response to 50 pA current steps. (i) Resting membrane potentials (TeLC, 73.4 ± 2.03 mV, n = 14 cells from three mice; TeLC+, 70.7 ± 2.01 mV, n = 11 cells from two mice; unpaired t test, P = 0.335) and (ii) input resistances (TeLC, 162 ± 13.3 megohm; TeLC+, 188 ± 13.8 megohm; P = 0.188) were equivalent. (E) Synaptic inputs from OB are unaffected. Example recordings of EPSCs [membrane voltage (Vm), –70 mV] and disynaptic feedforward IPSCs (Vm, +5 mV) evoked by means of electrical stimulation of the lateral olfactory tract (LOT) in (top) an uninfected control slice or (bottom) a TeLC-infected neuron. Both EPSCs and IPSCs were blocked by 2,3-dihydroxy-6-nitro-7-sulfamoylbenzo[f]quinoxaline (NBQX) (10 μM) and d,l-2-amino-5-phosphonovaleric acid (APV) (50 μM, not shown). (F) Summary of LOT-evoked EPSC and IPSC amplitudes from (i) uninfected control slices, (ii) TeLC+ neurons and (iii) TeLC neurons in TeLC-infected slices. (iv) EPSC/IPSC ratios were equivalent in all conditions; P > 0.05, unpaired t tests. (G) LOT EPSC paired-pulse ratios were not significantly altered after TeLC expression. n.s., not significant. (H) Example recordings showing recruitment of FBI is impaired, whereas FBI is unaffected. EPSCs and IPSCs were evoked by electrical stimulation of layer 2/3 226 ± 17 μm from recorded cell. EPSCs and IPSCs were attenuated in TeLC-infected slices. Blocking glutamate receptors with NBQX and APV eliminates the disynaptic component of IPSCs, with the residual IPSC evoked through direct stimulation of FBIs. The residual IPSC was fully blocked by gabazine (GBZ) (10 μM). (I) Summary of residual IPSC amplitudes. (i) The fractional size of residual IPSCs after NBQX/APV was substantially smaller in TeLC-infected slices (control, n = 6 cells from three mice; TeLC, n = 6 cells from three mice; unpaired t test, P = 0.0055), but (ii) the amplitudes of residual IPSCs were equivalent (P = 0.957).

  • Fig. 5 Recurrent circuitry truncates and normalizes cortical output.

    (A) Extensive infection of layer 2 principal cells across PCx in an example mouse. GFP, green; NeuroTrace, magenta. Numbers indicate distance from bregma. Bottom row are the square sections from the top row. Scale bars, 500 μm (top) and 50 μm (bottom). (B) Percent cells expressing TeLC-GFP in six of seven mice used. Sections from one mouse were damaged, and infection could not be quantified. (Top) TeLC infection across rostral-caudal PCx. (Bottom) Low variation in TeLC expression across mice. (C) Experimental schematic. Simultaneous bilateral recordings from TeLC-infected and contralateral control hemisphere with odor stimuli. (D and E) Example responses (D) and average population PSTHs (E) (mean ± SEM; control, n = 450 cell-odor pairs; TeLC, n = 388 cell-odor pairs) (F) Normalized peaks in MUA rates (n = 4 experiments, two odors, four concentrations). (Left) Peak responses across recordings and odorant concentrations. (Right) Each point is average response of one simultaneously recorded TeLC-Control PCx pair normalized to mineral oil response. Shading indicates concentration. Cyan lines are linear fits for each experiment through all concentrations. (G) As in (F) but for average rate over the first 330 ms after inhalation.

  • Fig. 6 Centrifugal inputs from PCx control gain but not time course of OB responses.

    (A) Schematic of circuit changes in OB after TeLC expression in ipsilateral PCx principal cells. (B) Centrifugal PCx fibers expressing TeLC in OB ipsilateral to PCx infection. GFP, green; NeuroTrace, magenta. (C) Experimental schematic. Simultaneous bilateral recordings from OB ipsilateral and contralateral to TeLC-infected PCx with odor stimuli. (D) Peaks in OB MUA rates averaged across population-odor pairs (n = 3 experiments, two odors, four concentrations). Odor responses are normalized to mineral oil responses. (E) As in (D) but for average rate over the first 330 ms after inhalation. (F) Average PSTH of all OB cell-odor pairs in control (black, n = 406) or TeLC (green, n = 384) side responding to odor (mean ± SEM). Thick lines are exponential fits to decay from peak to minimum. (Inset) Rescaled control OB response (dotted line) overlaid on TeLC-OB response. Response dynamics are similar in control and TeLC hemisphere despite change in response gain. (G) Same as (F) but for PCx (control, n = 1660; TeLC, n = 1532). Here, decay constants differ by an order of magnitude between control and TeLC hemisphere.

  • Fig. 7 PCx truncates sustained input from OB.

    (A) Simultaneous OB-PCx recordings with direct optical OB activation. (Top) Experimental schematic. (Bottom) ChR2 expression in mitral cells. Scale bar, 100 μm. (B) Responses from example (top) OB and (bottom) PCx cells to 1-s light pulses over OB. (C) Average population PSTHs for responses from experiment in (B). Gray shading indicates initial and sustained analysis windows. (PCx time constants for 20 mW light pulses; decay from peak, 18.9 ± 2.0 ms; recovery from post-stimulus trough, 87.4 ± 46.3 ms; n = 5 population recordings.) (D) Normalized MUA rates during initial phase (n = 5 experiments) in OB versus PCx. (Left) Average OB (red) and PCx (black) responses across recordings. MUA was normalized to baseline activity 1 s before stimulation. (Right) Each point is the average response of one simultaneously recorded OB-PCx response pair. Shading indicates light intensity. Light gray lines are linear fits for each OB-PCx population pair. Black line is the linear fit to all data. (E) As in (D) but for the sustained phase. (F) Experimental schematic. Simultaneous OB-PCx recordings from TeLC-infected or contralateral control hemisphere with optical OB activation. (G) Example responses from cells to 1-s light pulses over OB. (H) Average population PSTHs for responses from experiments in (G). (I) Normalized peak MUA rates during initial phase (n = 13 TeLC and 8 control experiments) in OB versus PCx. (Left) Average TeLC-PCx (green) and contralateral control PCx (black) MUA rates. (Right) Each point is the average response of one simultaneously recorded OB-PCx pair at one intensity. Light lines are linear fits for each TeLC (green) or control (gray) OB-PCx population pair. Solid lines are linear fits for all TeLC (green) and control (black) data. (J) As in (I) but for sustained rate.

  • Fig. 8 Recurrent circuits implement concentration-invariant decoding.

    (A) (Left) PCA representation of pseudopopulation responses for contralateral control PCx hemispheres. (Right) Mean distance between population responses in PCA space normalized to Δ odor responses. Δ conc. responses were more similar than Δ odor responses in the control PCx (one-sample t test versus mean of 1, P = 2.03 × 10–5). (B) As in (A), but for TeLC-PCx. Δ conc. responses were no more similar than Δ odor responses (P = 0.985). (C) Linear classifier performance for odorant decoding (choose 1 of 6 odors) using TeLC-infected (green) or contralateral control (black) PCx pseudopopulations. Classifier was trained and tested on spike counts in 20-ms bins in an expanding time window starting at odor inhalation. Pseudopopulation size in both conditions was held at 180 cells. Mean ± 95% confidence intervals from 200 permutations. Dashed line is chance accuracy. (D) Same as (C) for classification of different concentrations of the same odorant (choose 1 of 4 dilutions). (E) Accuracy for generalization task in which classifier is trained and tested on different concentrations of odors. Loss of recurrent circuits severely impairs odor identity recognition across concentrations.

  • Recurrent cortical circuits implement concentration-invariant odor coding

    Kevin A. Bolding and Kevin M. Franks

    Materials/Methods, Supplementary Text, Tables, Figures, and/or References

    Download Supplement
    • Figs. S1 to S6

Stay Connected to Science

Navigate This Article