Single cell plasticity and population coding stability in auditory thalamus upon associative learning

Abstract

Cortical and limbic brain areas are regarded as centres for learning. However, how thalamic sensory relays participate in plasticity upon associative learning, yet support stable long-term sensory coding remains unknown. Using a miniature microscope imaging approach, we monitor the activity of populations of auditory thalamus (medial geniculate body) neurons in freely moving mice upon fear conditioning. We find that single cells exhibit mixed selectivity and heterogeneous plasticity patterns to auditory and aversive stimuli upon learning, which is conserved in amygdala-projecting medial geniculate body neurons. Activity in auditory thalamus to amygdala-projecting neurons stabilizes single cell plasticity in the total medial geniculate body population and is necessary for fear memory consolidation. In contrast to individual cells, population level encoding of auditory stimuli remained stable across days. Our data identifies auditory thalamus as a site for complex neuronal plasticity in fear learning upstream of the amygdala that is in an ideal position to drive plasticity in cortical and limbic brain areas. These findings suggest that medial geniculate body's role goes beyond a sole relay function by balancing experience-dependent, diverse single cell plasticity with consistent ensemble level representations of the sensory environment to support stable auditory perception with minimal affective bias.

Fig. 1. Imaging neuronal activity of auditory thalamus in freely moving mice.

a Mouse with a head-mounted miniaturized microscope (left). Location of gradient refractive index (GRIN) lens in the medial geniculate body (MGB, right). b Example GCaMP6f expression in MGB. Similarly replicated expression patterns for all animals where GCaMP6f was injected in MGB (N = 24 mice). c High magnification of GCaMP6f-expressing MGB neurons from B. d Individual motion corrected fields of view (maximum intensity projection) of one example animal across a four-day fear conditioning paradigm (Hab, FC, Ext. 1, Ext. 2) as well as the maximum intensity projection across all days. Red circles indicate selected individual components. Replicated for all mice (N = 24) that underwent calcium imaging. e Average number of individual ICs/animal (93 ± 4 neurons, N = 24 mice). Boxplots show median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. f Changes in Ca²⁺ fluorescence of five individual neurons during the habituation session. Lines indicate CS tone presentations (red: 12 kHz, blue: 6 kHz). g Tone responses on habituation day 1 of all recorded MGB neurons in fear conditioning experiments (N = 855 neurons, N = 9 mice).

Fig. 2. Mixed selectivity tone CS+ and shock US coding of MGB neurons upon fear conditioning.

a Details of the 4-day fear conditioning paradigm. b Conditioned stimulus (CS) CS+ and CS− freezing (mean ± s.e.m.) during the habituation, fear conditioning as well as extinction days (Ext. 1, Ext. 2. e and l indicate early and late phases of extinction, i.e., the first four or last four CS+ of the session. Friedman test, p < 0.001, followed by Dunn’s multiple comparisons test, Ext.1e vs Ext.1 l p = 0.0069, Ext.1e vs Ext.2 l p = 0.0002, Ext.2e vs Ext.2 l p = 0.0281, N = 15 mice). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. c Population response of one example animal to the CS+ and US (unconditioned stimulus, mean ± s.e.m.). Blue dots indicate CS+ tone pips. Green bar indicates shock US. Example cell response to the CS+ (d) and US (e). Mean ± s.e.m. of five trials. Dots indicate CS+ tone pips. Inset represents average response to single pips. f Proportion of CS+, CS− and US responsive neurons. Friedman test, p < 0.001, followed by Dunn’s multiple comparisons test: CS+ vs. US, p = 0,029; CS− vs. US, p = 0.0005 (N = 9 mice). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. g Proportion of mixed selectivity CS± and US coding neurons. Red line indicates chance overlap level (N = 9 mice). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. h Example spatial map of unisensory and multisensory mixed selectivity CS and US coding neurons in MGB. i Relationship between within response group and across response group pairwise spatial distance between neurons (N = 855 cells, N = 9 mice). j Cumulative distribution function of pairwise distances between all, US-responsive, CS+ and CS− responsive neurons (N = 855 cells, N = 9 mice). k Mean Ca²⁺ activity (± s.e.m) of sound-correlated neurons during shock evoked sound events e.g., mouse escape sounds and low frequency harmonic vocalizations (LFH, orange), the first CS+ pip (blue) and the US (green) from N = 550 CS+/US or N = 4956 sound event trials from 110 cells out of N = 3 mice. *, **, *** indicate p values smaller than 0.05, 0.01 and 0.001, respectively.

Fig. 3. Single cell response plasticity in MGB upon fear learning.

a Heat map of single cell CS+ responses on the habituation, extinction 1 and extinction 2 days. Cells were clustered into groups depending on their CS+ response pattern (N = 386 cells, N = 9 mice). b Average traces ± s.e.m of neuronal clusters in (a). c Proportion of individual plasticity groups within CS+ responsive cells/animal (Kruskal-Wallis test, p < 0.05, followed by Dunn’s multiple comparisons test; Stable vs cs down, p = 0.013, cs down vs fear down, p = 0.0003, cs down vs extinction down, p = 0. 035, fear vs fear down p = 0.0041; N = 9 mice). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. d Proportion of CS+ and CS− stable and plastic neurons (2-way ANOVA followed by Sidak multiple comparisons test, p < 0.001; stable CS− vs. stable CS+, p = 0.002; stable CS− vs. plastic CS+, p < 0.001; stable CS+ vs. plastic CS−, p < 0.001; stable CS+ vs. plastic CS+, p < 0.001; plastic CS− vs. plastic CS+, p = 0.002, N = 9 mice, data presented as mean values ± s.e.m.). Circles represent individual animals. e Proportion of CS+ stable and plastic neurons in a pseudoconditioning paradigm compared to fear conditioned animals (2-way ANOVA followed by Sidak multiple comparisons test, p < 0.001; stable unpaired vs plastic unpaired, p = 0.019; stable paired vs. plastic paired, p < 0.001; stable unpaired vs. stable paired, p < 0.001; plastic unpaired vs. plastic paired, p < 0.001, Unpaired group N = 5 mice, Paired group N = 9 mice, data presented as mean values ± s.e.m.). Circles represent individual animals. f Experimental paradigm. Auditory tuning was tested before and after fear conditioning. g Population response to individual 200 ms pips before (black) and after (orange) fear conditioning (mean ± s.e.m, N = 681 cells, N = 7 mice). h Proportion of tone-responsive cells before and after fear conditioning (Two-tailed, Wilcoxon signed-rank test, p = 0.0156, N = 7 mice, data presented as mean values ± s.e.m.). Lines represent individual animals. i Mean auditory responses ± s.e.m of one example neuron before (black) and after (orange) fear conditioning to different tone frequencies (numbers indicate kHz). CS+ : 8 kHz, CS−: 20 kHz. j Population statistics for BF tuning towards the CS+ (two-tailed Wilcoxon signed-rank test, p < 0.001, n = 284 neurons, 7 mice). Horizontal lines represent median. k Population statistics for BF tuning towards the CS− (two-tailed Wilcoxon signed-rank test, p > 0.05, N = 284 neurons, 7 mice). Horizontal lines represent median. l Heat maps of single cell US responses to the five US stimulations during the fear conditioning day (N = 634 cells, N = 9 mice). m Average response ± s.e.m of plasticity subtypes of US-responsive MGB neurons (N = 634 cells, N = 9 mice, see also Supplementary Fig. 3). n Proportion of individual plasticity groups within US responsive cells/animal (Kruskal-Wallis test, p < 0.01 followed by Dunn’s multiple comparisons test; Stable vs down, p = 0.043; down vs off-up, p = 0.0006; down vs. off-down, p < 0.001; down vs. inh, p = 0.0055; off-down vs inh type 1, p = 0.018; N = 9 mice). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. o Proportion of US (N = 5 mice, p > 0.05, Wilcoxon signed-rank test) and CS plastic cells (N = 5 mice, p > 0.05, two-tailed Wilcoxon signed-rank test) in the ventral (MGBv) and medial (MGBm) subdivisions of MGB. Insert: Schematic of GRIN lens location above the different MGB subdivisions. *, **, *** indicate P values smaller than 0.05, 0.01 and 0.001, respectively.

Fig. 4. Functional subclasses of CS and US coding neurons are not enriched in amygdala projecting MGB neurons.

a Injection of AAV2-retro.hSyn1.mCherry.WPRE.hGHp(A) and latex beads in the basolateral amygdala (BLA). MGB was counterstained for calretinin (cyan) and NeuN (yellow) to quantify the BLA projectors (red). b Distribution of BLA-projecting neurons within MGB (N = 6 mice, Friedman test p < 0.001, followed by Dunn’s multiple comparisons test %MGBm vs. %MGBv, p = 0.0016). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. c Region-specific proportion of BLA-projecting neurons within MGB subdivisions (N = 4 mice, Friedman test, p < 0.01, followed by Dunn’s multiple comparisons test, %MGBm vs. %MGBv, p = 0.014). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. d Proportion of calretinin-positive BLA-projecting neurons (N = 4 mice). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. e Schematic of viral strategy and location of GRIN lens in MGB to image neuronal activity of MGB → BLA-projecting neurons. f MGB field of view with MGB → BLA-projecting neurons. Replicated in all animals that underwent calcium imaging (N = 6 mice). g Number of identified individual components per animal (69 ± 9, N = 6 mice). h Mean ± s.e.m population response of one example animal to the CS+ and CS−. Black dots indicate CS + tone pips. Bar indicates shock US. Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. i Proportion of CS + , CS− and US responsive neurons for the total MGB population and amygdala-projecting neurons (2-way ANOVA, main effect group, F_(1,13) = 3.3, p > 0.05, N = 9 total MGB population mice and N = 6 MGB → BLA projection neurons mice, see also Fig. 2f). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. j Proportion of mixed selectivity CS± and US coding neurons for the total MGB population and amygdala projecting neurons 2-way ANOVA, F_(1,13) = 3.9, p > 0.05, N = 9 mice for the total MGB population and N = 6 mice for the population of MGB → BLA projection neurons, see also Fig. 2. Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. g Dotted lines indicate chance overlap level. k Examples traces of groups of stable, onset down, fear and extinction neurons. l Proportion of individual plasticity groups within CS + responsive cells / animal (2-way ANOVA, F_(1,13) = 1.2, p > 0.05, N = 9 mice for the total MGB population and N = 6 mice for the population of MGB → BLA projection neurons). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. m Proportion of individual plasticity groups within US responsive cells / animal (2-way ANOVA, F_(1,13) = 0.5, p > 0.05, N = 9 mice for the total MGB population and N = 6 mice for the population of MGB → BLA projection neurons). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. ** indicate p values smaller than 0.01.

Fig. 5. Inhibition of amygdala projecting MGB neurons prevents memory consolidation and enhances plasticity in MGB.

a Optogenetic approach to inhibit MGB → BLA projection neurons. b Example ArchT expression in MGB → BLA projection neurons. Replicated for all animals that underwent optogenetic inhibition of MGB → BLA neurons (N = 9 mice). c Experimental paradigm: MGB → BLA neurons are manipulated during the CS+ and US on the fear conditioning day. d Freezing of GFP and MGB → BLA ArchT-expressing animals at the end of the fear conditioning paradigm (mean freezing levels to the last two CS+, GFP: N = 13 mice, MGB → BLA ArchT: N = 9 mice, p > 0.05, two-tailed Mann–Whitney test). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. e Freezing of GFP and ArchT-expressing animals upon fear recall during early extinction 1 (Ext. 1, mean freezing during the first four CS+, GFP: N = 13 mice, MGB → BLA ArchT: N = 9 mice, two-tailed Mann–Whitney test, p = 0.0056). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. f All-optical approach to inhibit MGB → BLA projection neurons during simultaneous recording of total MGB population activity. g Example GCaMP6f expression in MGB (top) and ArchT expression in MGB → BLA projection neurons (bottom). Replicated for all animals that underwent the all-optical paradigm (N = 6 mice). h Mean MGB population activity ± s.e.m in response to CS+ and US stimuli upon optogenetic light presentation of tdTomato (tdTom, black, N = 5 mice) and ArchT animals (orange, N = 6 mice). i Proportion of MGB neurons with stable and plastic CS+ responses after fear conditioning in tdTom (N = 5) and ArchT (N = 6) mice (2-way ANOVA, F_{(1, 9)} = 10.09, p < 0.05, Sidak’s multiple comparisons test, p = 0.0104 each). Boxplots represent median, 2^nd, 3^rd quartile, minimum and maximum. Cross indicates mean. *, ** indicate p values smaller than 0.05 and 0.01, respectively.

Fig. 6. MGB population dynamics are stable across days.

a Intraday three-way decoder of CS+, CS− and baseline population responses in CaMKII-positive (black) and identified amygdala-projecting MGB neurons (turquoise) reached a minimum mean accuracy of 81% across animals. Decoder accuracy dropped to chance levels for decoders trained on randomly label training sets (data presented as mean values ± s.e.m.). b Intra- and across day accuracy of decoders trained on CS+ or CS− vs. baseline responses, respectively. 1: Hab, 2: FC, 3: Ext. 1, 4: Ext. 2. c Quantification of intra and across day decoder accuracy for decoders trained on habituation day data. Mean decoder accuracy across days is >70% for CS+ and CS− population responses in CaMKII-positive and identified amygdala-projecting MGB neurons (MGB→BLA N = 6 mice, MGB N = 9 mice, data represent mean ± s.e.m.). Relative change in Euclidean population vector distance between the CS+ (d, e) or CS− (f, g) and the US within the fear conditioning session (d, f) or across the individual days of the behavioural paradigm (e, g). Statistics: d: Friedman test across the relative change in CS+ to US PVD of MGB → BLA-projectors (p < 0.01), Dunn-Sidak multiple comparisons test 1^st and 3^rd vs. 5^th CS/US pairing, p < 0.05. G: Friedman test across the relative change in CS− to US PVD of the total BLA-population (p < 0.05), Dunn-Sidak multiple comparisons test FC vs. Ext.2: p < 0.05. All other data sets in d-g: p > 0.05. MGB population: N = 9 mice, MGB → BLA-projectors: N = 6 mice (data represent mean ± s.e.m.). h Relative change in PVD between the CS and US for Control and ArchT (2-way ANOVA followed by Sidak’s post hoc test, p < 0.005, 5^th CS+ presentation Control vs. ArchT, p = 0.0438, Control N = 5 mice, ArchT N = 6 mice, data represent mean ± s.e.m.).