SST+ INs regulate memory strength and distribution

We next tested whether SST+ IN activity control experience- dependent ensemble size and memory formation by testing for an effect in contextual fear memory and cellular engram activation. Chemogenetic modulation of SST+ IN activity during the training

population of cFos+ GCs while increased SST+ IN activity at the time of training reduced the number of cFos+ GCs when compared to control mice (F(2,30) = 19.93, p < 0.001, Figures 6C and S6E).

We performed similar experiments in PV-Cre mice to test the role of PV+ IN activity in memory formation. Chemogenetic acti-vation of hM3D expressing PV+ INs during SE increased the frac-tion of these neurons expressing cFos (Control 25 ± 10, Active 69% ± 6% of PV+ neurons, p < 0.01,Figure S5E) and caused sig-nificant membrane depolarization and spiking activity (Fig-ure S5D), suggesting effective modulation of neuronal activity in this IN population. However, increased PV+ IN activity during training in the contextual fear memory test resulted in no differ-ence in the time spent in freezing behavior in any of the recall sessions (1 day, 1 week, F(1,9) = 1.40, p = 0.27; Figure 6E).

Similar number of cFos+ neurons was counted in the GC layer after the last recall session in both groups (p = 0.27, Figure 6F).

In line with these results, optogenetic activation of PV+ neurons during training did not alter freezing levels during test sessions (F(1,12)<0.01 p = 0.99, Figure S6A) or the number of cFos+

GCs after SE (Figurse S6B and S6C). These results suggest that SST+ but not PV+ IN activity during training bidirectionally modulates long-term memory and the size of cellular engram.

The effect of SST+ IN activity in freezing behavior after condi-tioning could reflect an influence of this IN type on fear generalization. We addressed this question by testing whether hM4D-mediated inactivation of this IN type during training affects freezing behavior in a different context than the one used for training. Although inactivation of SST+ INs increased freezing levels during subsequent exposition to fear conditioning chamber, it did not affect fear response when mice were exposed to a different unrelated context (Figure S6D). Chemoge-netic modulation of SST+ IN activity did not alter performance of mice in the elevated plus maze, excluding a role of anxiety like

behavior in the increased freezing response observed in mice with decreased SST activity (Figure S6D). These results show that enhanced memory performance upon SST+ INs inactivation during training is specific for the training context and suggest that SST+ INs do not influence fear generalization.

DISCUSSION

Our study identifies lateral inhibition between GCs as a mecha-nism that constrains the ensemble of activated neurons during SE of a new environment. In contextual fear paradigm, the lateral inhibition determines the size of the cellular engram and the sta-bility of the memory trace.

Several factors may limit the size of the active neuronal en-sembles that emerge in the DG during SE. First, the number of GCs is several fold higher than the number of upstream excit-atory neurons in the entorhinal cortex (Schmidt et al., 2012). Sec-ond, GCs’ dendrites strongly attenuate synaptic inputs so that the cell fires only when many inputs are concomitantly active (Krueppel et al., 2011). In addition, our results now reveal the ex-istence of lateral inhibition among GCs. This inhibitory interaction maintains the low level of neuronal excitation in cells not imme-diately active during SE that is typically required for DG function (Treves and Rolls, 1994).

As memory formation progresses, naturally formed neuronal ensembles acquire engram properties (i.e., they become neces-sary and sufficient to evoke specific memories) (Tanaka et al., 2014; Denny et al., 2014). Previous studies have manipulated the memory trace associated with naturally formed neuronal en-sembles (Han et al., 2009; Ramirez et al., 2013; Cowansage et al., 2014). Here, we succeed to assign fear memory to a neuronal ensemble that was artificially generated in the hippo-campus without specific context representation. Optogenetic stimulation of this subset of neurons at the time of training Figure 5. Sparse Population of DG Excit-atory Cells Activate Strong Dendritic Target-ing Lateral Inhibition

(A) Brief light stimulation of a sparse population of ChR2+ GCs induced a large amplitude GABAergic current (GABA) and little glutamatergic response (Glu) in ChR2! GCs. Quantification of the ampli-tude of GABAergic and glutamatergic currents is shown in C (black bars). Scale bars: 20 pA, 50 ms.

(B) GABAergic response before (black trace) or after pharmacological treatment with SR95531 (Gabazine, [GBZ], 10mM), tetrotodotoxin ([TTX], 1mM), and kynurenate ([Kyn], 6 mM) (gray traces).

Scale bars: 20 pA, 50 ms.

(C) Quantification of GBZ, TTX, and Kyn treatment on GABAergic current amplitude (white bars).

***p < 0.001, n = 17, 5, 4, 4, and 7 for GABA, Glu, GBZ, TTX, and Kyn, respectively; one-way ANOVA, Bonferroni post hoc test.

(D) Representative normalized traces of GABAergic response induced by optogenetic activation of PV (light gray), SST (dark gray), and CamKII population (Black). Blue arrow shows the onset of light stimulation. Graphs show the average peak amplitude, delay, rise time, and slope of responses obtained upon stimulation of three different neuronal populations. *p < 0.05, **p < 0.01, ***p < 0.001, n = 7, 7, and 17 for PV, SST, and CamKII, respectively; one-way ANOVA, Bonferroni (peak amplitude and delay) or Dunn (rise time and slope) post hoc tests. Scale bar: 5 ms. Error bars, SEM.

See alsoFigure S5.

Please cite this article in press as: Stefanelli et al., Hippocampal Somatostatin Interneurons Control the Size of Neuronal Memory Ensembles, Neuron (2016), http://dx.doi.org/10.1016/j.neuron.2016.01.024

74 session had no effect on freezing behavior (Figures 6A and 6B, p > 0.99). In contrast, mice where SST+ INs were silenced during training showed enhanced freezing behavior during recall 1 day and 1 week later (Figure 24B, p < 0.01, black bars). Conversely, when activity in SST+ INs was increased during training, freezing during recall was lower than in control mice during the 1 week test session (F(2,35) = 19.63, p < 0.001, Figure 24B, gray bars).

Bidirectional modulation of behavior by SST+ IN activity was reflected by the number of c-fos+ GCs detected after the last recall session. Mice with silenced SST+ INs showed a larger population of c-fos+ GCs while increased SST+ IN activity at the time of training reduced the number of c-fos+ GCs when compared to control mice (F(2,30) = 19.93, p <

0.001, Figures 24C and S6E).

determines the allocation of an artificial memory trace to the active fraction of neurons. However, changes in reactivation of the ensembles due to rapid uncoupling of the context and the fear response by the non-naturalistic stimulation pattern may cause memory extinction and reduced optogenetic memory retrieval (Figure S2F).

Figure 6. Dendritic but Not Somatic Target-ing INs Control the Size of the Active Neuronal Population and Memory For-mation

(A) Mice expressing DREADD receptors in SST+

INs received saline (Control) or CNO injections in order to decrease (Silenced, hM4D) or increase (Stimulated, hM3D) SST+ IN activity during training. Mice were tested 1 day and 1 week later.

1 hr after the last test, mice were perfused for cFos analysis.

(B) Chemogenetic inactivation of SST INs during training (Silenced, hM4D) increased while activa-tion (hM3D) decreased fear responses. Training hM4D p = 0.99, non significant (ns); hM3D p = 0.99, ns; 1 Day: hM4D **p < 0.01; hM3D p = 0.99, ns;

1 Week, hM4D, hM3D, ***p < 0.001, one-way ANOVA, Bonferroni post hoc test; 18,11, 9 mice per group.

(C) The number of cFos+ GCs after the 1 week recall session was larger in mice with silenced SST+ INs (hM4D) but was reduced in mice with increased SST+ IN activity during training (hM3D),

*p < 0.05, ***p < 0.001, one-way ANOVA, Bonfer-roni post hoc test, 16, 8, 9 mice per group. Shaded area is the number of cFos+ GCs (mean ± SEM) in mice perfused 1 hr after the training with unaltered neuronal activity.

(D) PV-Cre mice and wild-type litter mates with bilateral infection of AAV-DIO-hM3D in the DG were treated CNO (2 mg/kg) 1 hr before fear con-ditioning.

(E) No difference was observed in freezing re-sponses during 1 day and 1 week recall sessions between control wild-type (Control) and PV-Cre mice with increased PV+ IN activity (hM3D).

Training, 1 Day and 1 Week, p = 0.99, non signifi-cant (ns); two-way ANOVA, Bonferroni post hoc test, 5, 6 mice per group.

(F) The number of cFos+ GCs observed after the last test session did not differ between control (Ctrl) mice and mice with enhanced neuronal ac-tivity in PV+ INs (hM3D) at the time of training,t(9) = 1.18, p = 0.27, 5–6 mice per group. Shaded area is the number of cFos+ (mean ± SEM) in WT unin-fected mice perfused 1 hr after the training ses-sion. Error bars, SEM.

See alsoFigures S5andS6.

At the same time, the active ensemble inhibits the recruitment of neighboring cells avoiding retrieval of memory by nat-ural cues and providing specificity to the artificial cellular engram. How is this achieved? Since the sizes of both the ChR2-tagged population of GCs (!13%) and the fraction of GCs receiving strong synaptic input from the EC (!8% of GC express cFos+ after SE) are small, the chances for a neuron to belong simultaneously to both pools (that would allow association of sensory information with the ChR2 induced activity) are negligible. Optogenetic activation of GCs abolishes the formation of an ensemble that could guide Please cite this article in press as: Stefanelli et al., Hippocampal Somatostatin Interneurons Control the Size of Neuronal Memory Ensembles, Neuron (2016), http://dx.doi.org/10.1016/j.neuron.2016.01.024

Figure 24. Dendritic but not somatic targeting INs control the size of the active neuronal population and memory formation

A. Mice expressing DREADD receptors in SST+ INs received saline (Control) or CNO injections in order to decrease (Silenced, hM4D) or increase (Stimulated, hM3D) SST+ IN activity during training. Mice were tested 1 day and 1 week later. 1 h after the last test, mice were perfused for c-fos analysis. B. Chemogenetic inactivation of SST INs during training (Silenced, hM4D) increased while activation (hM3D) decreased fear responses. Training hM4D p = 0.99, non significant (ns); hM3D p = 0.99, ns; 1 Day: hM4D **p < 0.01; hM3D p = 0.99, ns; 1 Week, hM4D, hM3D, ***p < 0.001, one-way ANOVA, Bonferroni post hoc test; 18,11, 9 mice per group. C. The number of c-fos+ GCs after the 1 week recall session was larger in mice with silenced SST+ INs (hM4D) but was reduced in mice with increased SST+ IN activity during training (hM3D), *p < 0.05,

***p < 0.001, one-way ANOVA, Bonferroni post hoc test, 16, 8, 9 mice per group. Shaded area is the number of c-fos+ GCs (mean ± SEM) in mice perfused 1 h after the training with unaltered neuronal activity. D. PV-Cre mice and wild-type litter mates with bilateral infection of AAV-DIO-hM3D in the DG were treated CNO (2 mg/kg) 1 h before fear conditioning. E. No difference was observed in freezing responses during 1 day and 1 week recall sessions between control wild-type (Control) and PV-Cre mice with increased PV+ IN activity (hM3D). Training, 1 Day and 1 Week, p = 0.99, non significant (ns); two-way ANOVA, Bonferroni post hoc test, 5, 6 mice per group. F. The number of c-fos+ GCs observed after the last test session did not differ between control (Ctrl) mice and mice with enhanced neuronal activity in PV+ INs (hM3D) at the time of training, t(9) = 1.18, p = 0.27, 5–6 mice per group. Shaded area is the number of c-fos+ (mean ± SEM) in WT uninfected mice perfused 1 h after the training session. Error bars, SEM. See also Figures S5 and S6.

We performed similar experiments in PV-Cre mice to test the role of PV+ IN activity in memory formation. Chemogenetic activation of hM3D expressing PV+ INs during SE increased the fraction of these neurons expressing c-fos (Control 25 ± 10, Active 69% ± 6% of PV+ neurons, p < 0.01, Figure S5E) and caused significant membrane depolarization and spiking activity (Figure S5D), suggesting effective modulation of neuronal activity in this IN population. However, increased PV+ IN activity during training in the contextual fear memory test resulted in no difference in the time spent in freezing behavior in any of the recall sessions (1 day, 1 week, F(1,9) = 1.40, p = 0.27; Figure 24E).

Similar number of c-fos+ neurons was counted in the GC layer after the last recall session in both groups (p = 0.27, Figure 24F). In line with these results, optogenetic activation of PV+ neurons during training did not alter freezing levels during test sessions (F(1,12)<0.01 p = 0.99, Figure S6A) or the number of c-fos+ GCs after SE (Figures S6B and S6C).

These results suggest that SST+ but not PV+ IN activity during training bidirectionally modulates long-term memory and the size of cellular engram.

The effect of SST+ IN activity in freezing behavior after conditioning could reflect an influence of this IN type on fear generalization. We addressed this question by testing whether hM4D-mediated inactivation of this IN type during training affects freezing behavior in a different context than the one used for training. Although inactivation of SST+

INs increased freezing levels during subsequent exposition to fear conditioning chamber, it did not affect fear response when mice were exposed to a different unrelated context (Figure S6D). Chemogenetic modulation of SST+ INs activity did not alter performance of mice in the elevated plus maze, excluding a role of anxiety like behavior in the increased freezing response observed in mice with decreased SST activity (Figure S6D). These

results show that enhanced memory performance upon SST+ INs inactivation during training is specific for the training context and suggest that SST+ INs do not influence fear generalization.