High-Speed Imaging Reveals Neurophysiological Links to Behavior in an Animal Model of Depression

See allHide authors and affiliations

Science  10 Aug 2007:
Vol. 317, Issue 5839, pp. 819-823
DOI: 10.1126/science.1144400


The hippocampus is one of several brain areas thought to play a central role in affective behaviors, but the underlying local network dynamics are not understood. We used quantitative voltage-sensitive dye imaging to probe hippocampal dynamics with millisecond resolution in brain slices after bidirectional modulation of affective state in rat models of depression. We found that a simple measure of real-time activity—stimulus-evoked percolation of activity through the dentate gyrus relative to the hippocampal output subfield—accounted for induced changes in animal behavior independent of the underlying mechanism of action of the treatments. Our results define a circuit-level neurophysiological endophenotype for affective behavior and suggest an approach to understanding circuit-level substrates underlying psychiatric disease symptoms.

The hippocampus, as an integral component of the limbic system, is a focus of depression research (1), drives other brain regions implicated in depression, and appears to serve as a primary site of action for antidepressants that inhibit pathological hyperactivity (2, 3). Complicating this picture, however, is evidence suggesting that antidepressants can stimulate hippocampal activity. Antidepressant-induced hippocampal neurogenesis is linked to behavioral responses (4, 5); moreover, excitatory hippocampal neurons are injured by chronic stress (6, 7). Animal models have proven useful in identifying molecular and cellular markers relevant to depression (810) but have not identified neurophysiological final common pathways relevant to behavior. Voltage-sensitive dye imaging (VSDI) could allow analysis of disease-related neural activity on millisecond time scales, with micrometer spatial resolution and a scope spanning entire brain networks (11). We applied VSDI to hippocampal physiology in the chronic mild stress (CMS) model, a well-validated rodent model of core depressive behavioral symptoms (12).

Evoked activity was recorded with VSDI in acute horizontal slices prepared from the ventral hippocampus of adult rats (Fig. 1A) (13, 14). Signals reflecting local neuronal network activity were extinguished by 2,3-dihydroxy-6-nitro-7-sulfamoylbenzo[f]quinoxaline-2,3-dione (NBQX) and d-2-amino-5-phosphonopentanoic acid (D-AP5) and therefore required excitatory transmission (Fig. 1B). We used cross-correlation to extract reliable, quantitative features of the network response (Fig. 1, C and D, and figs. S1 and S2), computing the maximal response amplitude of each pixel (Fig. 1D, top right, and fig. S3, right) and the time at which this maximal amplitude occurred (“phase”; Fig. 1D, top left, and fig. S3, left). In our experiments, phase distributions were independent of treatment group and coherent in the region responding to stimulation, which was isolated computationally in blinded analysis (Fig. 1D, bottom, and figs. S4 to S6). A simple metric of circuit response (“total activity,” defined as the mean amplitude multiplied by the area of the calculated region of interest) (13) was found to be linear in this stimulus range and reliable across slices (Fig. 2B).

Fig. 1.

Voltage-sensitive dye imaging (VSDI) of hippocampal network activity. (A) Representative filmstrip acquired using VSDI. Elapsed times are relative to a single stimulus pulse applied to DG; warmer colors indicate greater activity (relative change in VSD fluorescence, ΔF/F). Data represent the average of four individual acquisitions. (B) VSDI signal is abolished by blockers of excitatory synaptic transmission (10 μMNBQX and 25 μM D-AP5). GABAzine (20 μM) and tetrodotoxin (TTX, 1 μM) application subsequently confirmed signal extinction. (C) Single-pixel response (ΔF/F versus time; top) from the indicated region to the given stimulus train (bottom). (D) Phase (top left) and amplitude (top right) of maximal correlation between the stimulus and response at each pixel. The synchronously responding region was extracted computationally, with the same phase criteria applied to all treatment groups, on the basis of similar phase values of responding pixels (bottom); au, arbitrary units.

Fig. 2.

Hippocampal network dynamics in depression-related behavioral states. (A) Left: Chronic mild stress (CMS)–treated animals displayed increased immobility on a 5-min forced swim test (FST) relative to controls (Student's t test, n = 6 animals per group). Right: Fluoxetine (Flx) and imipramine (Imi), but not haloperidol (Hal), decreased immobility [analysis of variance (ANOVA), F3,22 = 29.46, n = 5 or 6 animals per group; Ctrl, control]. (B) VSDI total activity response to applied stimulus in DG (top left and center, n = 7 slices, r2 = 0.9855) and CA1 (bottom left and right, n = 5 slices, r2 = 0.9926). Left: Sample frames of DG and CA1 responses. (C) Activity of DG relative to CA1, calculated for each slice and averaged for each animal, was reduced by CMS (Student's t test, n = 6 animals per group). (D) Activity of DG relative to CA1 was specifically increased by antidepressants (ANOVA, F3,22 = 12.74, n = 5 or 6 animals per group). Data are means ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001.

To quantify behaviorally relevant hippocampal dynamics in a rodent model, we applied CMS (Fig. 2A, left) or delivered one of two chronic antidepressant treatments: the selective serotonin reuptake inhibitor fluoxetine or the tricyclic antidepressant imipramine (the typical antipsychotic haloperidol served as a pharmacological control; Fig. 2A, right). Depression-like behavior was quantified, blind to treatment condition, in terms of immobility on the forced swim test (FST); in this test, immobility is increased by CMS and decreased by antidepressants (15). In all drug experiments, a 48-hour delay between the last dose and behavioral assessment excluded acute drug effects on behavior that do not have relevant clinical correlates (16). Relative to controls, CMS animals were more immobile over a 5-min FST, indicating a depressed-like state (Fig. 2A, left), whereas antidepressant-treated but not antipsychotic-treated animals showed less immobility (Fig. 2A, right).

We then conducted VSDI of evoked activity in ventral hippocampal slices (14) from these animals; blind to treatment condition, we probed both the dentate gyrus (DG) and CA1, hypothesizing different effects in the input and output hippocampal subfields (1720). We found that activity propagation in DG relative to CA1 provided the most reliable predictor of FST performance on an animal-by-animal basis (Fig. 2C). DG activity was reduced in CMS-treated animals (fig. S7A), whereas CA1 activity was increased (fig. S7B); the CA1 contribution is compatible with results linking depression to elevated hippocampal output (2, 3, 21), and the DG contribution is consistent with data suggesting reduced hippocampal activity in depression (4, 6).

We found the opposite pattern in antidepressanttreated animals (Fig. 2D and fig. S7), with increased activity propagation in DG (fig. S7C) and reduced activity in CA1 (fig. S7D). These effects were specific to antidepressants; antipsychotic treatment showed no effect on either subfield (fig. S7, C and D). Again, the activity propagation in DG relative to CA1 provided the most reliable (across-individual) indicator of the behavioral phenotype (Fig. 2D; r2 = 0.5251, P <10–6, across CMS and drug groups).

To model clinical use of antidepressants, we next concomitantly administered fluoxetine during the last 2 weeks of 5-week CMS (Fig. 3). FST blinded to treatment condition confirmed that fluoxetine treatment reversed the behavioral effects of CMS (Fig. 3A), and VSDI demonstrated that the activity propagation in DG relative to CA1 significantly accounted for this effect (Fig. 3, B and C). On an individual-animal basis, this measure of activity propagation on the millisecond time scale regressed linearly with the FST scores and explained more than half of the bidirectional behavioral variation (Fig. 3C; r2 = 0.5545, P <10–6) across all four independent treatment arms. Open-field tests (OFTs) from the same animals provided a test of the specificity of the network dynamics phenotype. We observed no significant differences between groups in locomotion or anxiety-related behavior on the OFT (Fig. 3, D and E), and there was no correlation between VSDI physiology and OFT scores (Fig. 3F; r2 = 0.0306, P > 0.4), indicating specificity for depression-relevant behavior.

Fig. 3.

Hippocampal network dynamics predict antidepressant treatment of depressed-like states. Fluoxetine was concomitantly administered during the last 2 weeks of 5-week CMS. (A) CMS increased immobility and fluoxetine decreased immobility in both control and CMS groups (ANOVA, F3,34 = 19.24, n = 8 to 12 animals per group). (B) Activity of DG relative to CA1 was decreased in CMS animals and was increased by fluoxetine (ANOVA, F3,34 = 16.17, n = 6 to 12 animals per group). (C) Linear regression of activity of DG relative to CA1 against FST scores for each individual animal (r2 = 0.5545, P <10–6, n = 35 individual animals). (D and E) On the open-field test (OFT), no differences were observed in percent time in center [(D); ANOVA, F3,20 = 1.021, P > 0.05, n = 4 to 9 animals per group] or total distance [(E); ANOVA, F3,20 = 1.776, P >0.05, n = 4 to 9 animals per group]. (F) Linear regression of activity of DG relative to CA1 against percent time in center for each individual animal (r2 = 0.0306, P >0.4, n = 25 individual animals).

To address the cellular mechanism, we next probed for changes in hippocampal neurogenesis—hypothesized to be relevant to depression [(4, 5), but see (22)]—in the same animals represented in Fig. 3. In accord with previous observations (4, 5), fluoxetine increased both the number and density of newborn neurons [as assessed by blinded, unbiased stereology of cells positive for 5-bromo-2′-deoxyuridine (BrdU) and Double-cortin (Dcx; immature neuronal marker)] (13) in the ventral DG, both in the presence and absence of CMS (Fig. 4, A and B), whereas CMS alone did not significantly alter the production (Fig. 4, A and B) of new neurons despite behavioral and circuit dynamics effects in the same animals (Fig. 3, A to C). Similarly effective CMS did not affect the survival of newborn neurons (Fig. 2 and fig. S12). These data indicate that circuit dynamics changes can account for bidirectional affective state modulation despite fundamental differences in cellular processes occurring during depressed-like state induction and treatment.

Fig. 4.

VSDI resolution of neurogenesis-dependent circuit dynamics changes underlying antidepressant response. (A) Representative confocal DG images labeled for BrdU (green), NeuN (mature neuronal marker, red), and Dcx (immature neuronal marker, cyan). Arrowheads indicate BrdU+ neurons. Scale bar, 50 μm. (B) New neuron (BrdU+/Dcx+) counts (top) and density (bottom) in ventral hippocampus (same animals as in Fig. 3) were increased with fluoxetine treatment but unchanged with CMS (counts: ANOVA, F3,27 = 9.670; density: F3,27 = 20.68; n = 6 to 8 animals per group). (C) One month after irradiation designed to ablate hippocampal neurogenesis, fluoxetine or vehicle was administered for 1 week, followed by a 3-week delay for newborn neuron incorporation. (D) Top:BrdU+ cell density was increased after fluoxetine treatment and substantially decreased with irradiation (ANOVA, F3,24 = 29.72, n = 4 to 8 animals per group). Bottom: Fluoxetine treatment specifically increased the density of newborn neurons (BrdU+/NeuN+) in DG [glial fibrillary acidic protein (GFAP), astrocytic marker; Student's t test, n = 6 animalsper group]. (E) Top: Fluoxetine-treated animals showed decreased FST immobility; no effects were observed with irradiation (ANOVA, F3,23 = 7.757, n = 6 animals per group). Bottom: Irradiation blocked increased activity of DG relative to CA1 after fluoxetine treatment (ANOVA, F3,22 = 3.997, n = 5 or 6 animals per group).

To test the capability of a temporally defined cohort of new neurons to modulate behavior and circuit dynamics, we treated animals for 1 week with fluoxetine to up-regulate neurogenesis, followed by a 3-week delay to permit functional integration of neurons born during treatment (Fig. 4C). In some animals, we ablated hippocampal neurogenesis via irradiation (10 Gy/day for 2 days) 1 month before drug exposure; control experiments revealed no effect of irradiation alone on excitability, network dynamics, or behavior on this time scale (Fig. 4E and figs. S8 and S11). The fluoxetine pulse gave rise to a temporally defined cohort of new neurons (Fig. 4D and figs. S13 and S14) and reduced FST immobility in a manner blocked by irradiation (Fig. 4E, top), indicating that increased neurogenesis indeed is required for these antidepressant behavioral effects (5). However, irradiation alone did not affect behavior (Fig. 4E, top); therefore, inhibition of neurogenesis is neither sufficient (Fig. 4E, top) nor necessary (Fig. 2A, left; Fig. 3A; Fig. 4, A and B; and fig. S12) to induce a depressed-like state.

To quantitatively explore circuit dynamics modulation by the temporally defined cohort of new neurons, we conducted VSDI in the ventral hippocampus from these animals. The activity propagation in DG relative to CA1 was indeed increased (Fig. 4E, bottom), and only the DG effect was neurogenesis-dependent (fig. S8, A and B). Although it may be counterintuitive that a small number of new neurons (23) could affect circuit dynamics, simple modeling predicted that rare new neurons can increase the recruited active network area (fig. S9). We therefore analyzed VSDI signal components (area and amplitude) to determine their contribution to the observed changes in DG physiology, and found that the circuit-level effect of a temporally defined cohort of fluoxetine-induced newborn neurons on DG activity is indeed due primarily to increased active DG area (fig. S8), a parameter readily detectable by high-speed VSDI as demonstrated here.

These data suggest that behavioral changes can be linked to a common network dynamics phenotype without requiring a common etiology or mechanism such as neurogenesis. Indeed, we propose that genetic or environmental factors with diverse cellular mechanisms (47, 17, 18, 24) that are operative in different individuals may exert behavioral effects through a common activity-percolation phenotype. Although many antidepressants are associated with increased seizure risk and therefore could involve increased activity propagation through the DG, other antidepressant treatments clearly do not directly target the hippocampus, such as deep brain stimulation (DBS), which typically targets Cg25 or the nucleus accumbens. However, DBS reduces activity in Cg25 (25), which receives excitatory drive from the hippocampus (2, 3, 21), suggesting that Cg25 DBS can intervene downstream of an overactive CA1. There had been no obvious way to unify into a single model the hippocampal atrophy seen in depression (7, 24) with the likely increased excitatory drive from hippocampus to cortex associated with depression (2, 25). Our results suggest that the increased subgenual cingulate activity in depression could result in part from increased CA1 activity, whereas the reduced intrinsic hippocampal function observed in depression is consistent with decreased DG activity.

Hippocampal dysfunction related to mood may be experienced cognitively [e.g., as hopelessness (26)], which can manifest clinically as patients' inability to foresee or navigate a reasonable and hopeful plan within the environment. Theoretical models of the dorsal hippocampus have described comparative interactions between DG and CA1 (19, 20) in which CA1 activity indicates discrepancies between predictive information from DG and sensory information from the cortex. Depression therefore could be associated with the failure to predict, navigate through, or adapt to environmental changes (experienced as hopelessness) resulting from failed ventral DG associative/predictive activity or increased error signals from CA1. If that is the case, the intensity of the resulting dysphoria may be modulated by anxiety or reward pathways (amygdala, nucleus accumbens, and mesolimbic dopamine projections) or the prefrontal and cingulate cortices (27). Indeed, identification of this hippocampal neurophysiological endophenotype may serve as a starting point in mapping the network-level changes in other brain regions implicated in depression. High-speed, circuit-level optical methods are better suited than single-cell physiology to detect and quantitatively describe spatiotemporal dynamics (such as areal spread of activity) that may be altered in psychiatric disease. These circuit dynamics measures relate to how information propagates rather than to a specific neural code. We propose that depression may depend on changes in the ability of information representations to organize and percolate through sparsely active networks.

Supporting Online Material

Materials and Methods

Figs. S1 to S14


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article