Spiking network model of adult-neurogenesis.

(A1). Schematic of early olfactory circuit. M/T: mitral/tufted cells, GC: granule cells, abGCs: adult-born GCs, PCs: piriform cortical cell, FFIN: feedforward inhibitory neuron, FBIN: feedback inhibitory neuron. Example plus signs indicate excitatory synapses and minus signs indicate inhibitory synapses. (A2). Schematic of synaptic reshuffling of GCs due to adult-neurogenesis. On each day, 10% of GCs have their weights reshuffled, including the synaptic weights from M/T cells, other GCs or short-axon cells (SAC), and PCs to GC (feedback). (B1). Partial weight matrix on two example days between GCs and M/T cells (top), other GCs or SACs (middle), and feedback from PCs (bottom). (B2). Histograms of all synaptic weights across days. (B3). Weight matrix dissimilarity between Day-0 and each other day. (C1). An example model odor generated by stimulating different combinations of glomerular identity and timing of activation. (C2). Pair-wise correlations between model odors show the relative similarities and differences across all model odors.

Adult-neurogenesis modulates individual M/T cell responses but preserves population representation.

(A1). Trail-averaged firing rate of two example M/T cells responding to the same odor (within-odor) on three different days (mean ± SD, n = 10 trials). (A2). Firing rate patterns of odor-activated M/T cells to the same odor on three days. M/T cells driven by different glomeruli are separated by white dashed line. The two arrows correspond to the two example cells in (A1). (A3). Pairwise within-odor full-ensemble correlation of single-trial M/T cell responses between each day averaged across all odors (n = 100). Each small box separated by the dashed lines is a 10x10 matrix (n = 10 trials) corresponding to autocorrelation (same day on diagonal) and cross-correlation (different days off diagonal). (A4). Full-ensemble correlation of trial-averaged M/T cell responses for within-odor (black solid line with error bar: mean ± SD, n = 100 odors).) and across-odor (black dashed line with error bar: mean ± SD, n = 10 pairs) between Day-0 and each other day. Grey dashed line with no error bar: same as the black solid line in (B4). (B1). Low-dimensional trajectories of M/T responses to two example odors (color) on two different days (darkness of color). Thin curves: single-trial trajectories, thick curves: trial-averaged trajectories, points: maximal-distance points on trajectory to the origin. (B2). Only maximal-distance points are shown for different odors on different days. (B3). Similar to (A3) but for reduced-space correlation computed using the low-dimensional M/T trajectories (single-trial). (B4). Similar to (A4) but for reduced-space correlation computed using the low-dimensional M/T trajectories (trial-averaged). Grey dashed line with no error bar: same as the black solid line in (A4).

Adult-neurogenesis modulates both individual PCs responses and population representations.

(A1). Trail-averaged firing rate of two example PCs responding to the same odor on three different days (mean ± SD, n = 10 trials). (A2). Firing rate patterns of partial PCs responding to the same odor on three days. The two arrows correspond to the two example cells in (A1). (A3). Pairwise within-odor full-ensemble correlation of single-trial PCs responses between each day averaged across all odors (n = 100). Each small box separated by the dashed lines is a 10x10 matrix (n = 10 trials) corresponding to autocorrelation (same day on diagonal) and cross-correlation (different days off diagonal). (A4). Full-ensemble correlation of trial-averaged cell (PCs: purple; M/T: black) responses for within-odor (solid line with error bar: mean ± SD, n = 100 odors).) and across-odor (dashed line with error bar: mean ± SD, n = 10 pairs) between Day-0 and each other day. The curves for M/T cells are the same lines as in Fig.2. (B1). Low-dimensional trajectories of PCs responses to two example odors (color) on two different days (darkness of color). Thin curves: single-trial trajectories, thick curves: trial-averaged trajectories, points: maximal-distance points on trajectory to the origin. (B2). Only maximal-distance points are shown for different odors on different days. (B3). Similar to (A3) but for reduced-space correlation computed using the single-trial PCA trajectories of PCs. (B4). Similar to (A4) but for reduced-space correlation computed using the trial-averaged PCA trajectories of PCs.

Experience-dependent plasticity enhances representational stability in PCx.

(A1). Different types of plastic synapses (blue) related to an abGC. Circle: inhibitory; triangle: excitatory. (A2). Dependence of synaptic modification on pre/post inter-spike interval used for the spike-timing-dependent plasticity (STDP). S1 and S2 are two example synapses with different modification – S1: facilitated; S2: suppressed. (B). Weight history of the two example synapses in A2 when an odor is repeatedly presented. Adult-neurogenesis happens on day-4 which randomly resets the weights. The synaptic weights stay constant without plasticity (black) while change trial by trial with plasticity (blue). (C1). Left: reduced-space correlation of trial-averaged M/T trajectories between Day-0 and each day. Blue solid line with error bar: within-odor and with STDP (mean ± SD, n = 10 odors); black dashed line without error bar: within-odor and without STDP. Black dashed line with error bar: across-odors, mean ± SD, n = 10 pairs. Right: drift rate for M/T trajectories with and without STDP. (C2). Same as C1 but for PCs trajectories.

Parameters of Izhikevich neuron model for different cell types

Network parameters controlling the connectivity between cell types.

Odor definition and synaptic weight changes

(A). Glomerular activation latency for all 100 odors we defined. Each column corresponds to one odor. (B). Histogram of the number of activated glomeruli by all odors (𝑛 = 100 odors). (C). Histogram of correlation coefficients between each pair of odors, measuring the similarity between odors. (D). Sparsely sampled synaptic weights on day-0 (x axis) and day-10 (y axis). Each dot is a synapse. Left: M/T to GCs; middle: GC/SAC to GCs; right: PCs to GCs.

Example M/T cell responses on day-0 and day-10

Three example M/T cells (column) in response to the five example odors (row). Black: response on day-0; red: response on day-10. Some cells increase their responses while some cells decrease their responses from day-0 to day-10 (mean ± SD, 𝑛 = 10 trials).

Example Piriform cell responses on day-0 and day-10

Three example piriform cells (column) in response to the five example odors (row). Black: response on day-0; red: response on day-10 (mean ± SD, 𝑛 = 10 trials).

Example piriform cell responses (trial-averaged) on all 10 days

(A). Another three example piriform cells in response to four odors from day-0 and day=10. Each curve is the response averaged across 10 trials (smoothed by a gaussian kernel with 20ms width). Note some piriform cells increase or decrease their responses, while some cells change their temporal profile (cell-3).

Cell response changes between day-0 and day-10

(A). Histogram showing the differences between the trial-averaged response on day-10 and the trial-averaged response on day-0 for each odor-activated M/T cell (n=100 odors). Positive values indicate that the responses on day-10 are larger than day-0. (B). Same as (A) but for piriform cells.

Odor manifold and geometric reshaping of odor representations in PCx.

Odor manifolds of M/T cells (top) and PCs (bottom) on two example days (red and blue) are plotted by clusters of transparent round dots in the PCA space. Each column contains the trial-averaged trajectories on two days (red and blue) evoked by three example odors. M/T trajectories stay close while PCs trajectories are reshaped in different ways: shifted (A), rotated (B), and warped (C).

Quantifying geometric reshaping and decoding analysis on M/T and PCs

(A1). Cosine similarity to quantify the geometric reshaping using population firing rate as a function of intervals (number of days separated). (A2). Decoding accuracy by K-nearest neighbors algorithm trained on M/T population firing rate of each sample day and tested on M/T population firing rate of each reference day. (A3). Same as (A2) but using PCs population firing rate. (B). Same as (A) but using PCA trajectories.

Synaptic weight history of STDP

(A). Traces of five example synapses from M/T cells to GCs. STDP was applied when adult-neurogenesis happened at the start of the day. STDP drove the synaptic weight (y axis) to certain values through trial-by-trial exposure to certain odors. For some synapses, adult-neurogenesis only happened on one day while for other it happened on multiple days. (B). With STDP applied, three example piriform cells (column) in response to the three example odors (row). Black: response on day-0; red: response on day-10 (mean ± SD, n = 10 trials). (C). Slope of the synaptic weight traces over the last five trials of odor exposure with STPD applied. The slopes were all close to zero indicating the convergence of synaptic weights.