Temporal binding function of dorsal CA1 is critical for declarative memory formation

Edited by Larry R. Squire, Veterans Affairs San Diego Healthcare System, San Diego, CA, and approved July 31, 2017 (received for review November 29, 2016)
September 5, 2017
114 (38) 10262-10267

Significance

Our ability to form declarative memories depends on the hippocampus, and this capacity degrades with age. To identify the critical determinants of this age-associated memory loss, we explored the relationships between two functions of the hippocampus known to be age-sensitive, temporal binding and relational organization. We found that (i) temporal binding relies on the activity of the dorsal (d)CA1 subfield across temporal gaps between events; and (ii) loss in this dCA1-dependent function, through ensuing disruption of relational organization, is the primary cause of declarative memory loss occurring in aging.

Abstract

Temporal binding, the process that enables association between discontiguous stimuli in memory, and relational organization, a process that enables the flexibility of declarative memories, are both hippocampus-dependent and decline in aging. However, how these two processes are related in supporting declarative memory formation and how they are compromised in age-related memory loss remain hypothetical. We here identify a causal link between these two features of declarative memory: Temporal binding is a necessary condition for the relational organization of discontiguous events. We demonstrate that the formation of a relational memory is limited by the capability of temporal binding, which depends on dorsal (d)CA1 activity over time intervals and diminishes in aging. Conversely, relational representation is successful even in aged individuals when the demand on temporal binding is minimized, showing that relational/declarative memory per se is not impaired in aging. Thus, bridging temporal intervals by dCA1 activity is a critical foundation of relational representation, and a deterioration of this mechanism is responsible for the age-associated memory impairment.
Our ability to form declarative memories depends on the hippocampus, and this capacity degrades with age (1). To identify the critical determinants of this age-associated memory loss, we explored the relationships between two fundamental functions of the hippocampus known to be age-sensitive.
First, the hippocampus plays an essential role in forming a “relational organization” that links independently acquired memories via common elements and consequently supports cardinal flexibility of declarative memory, exemplified in the capability to make inferences from memory (2) or to compare separately acquired information to guide a choice decision in a novel situation (3, 4). This capacity is compromised in aging (58). Second, as suggested three decades ago (9), the hippocampus supports “temporal binding,” by which discrete stimuli can be associated in memory despite their temporal separation. A critical role for the hippocampus in temporal binding was demonstrated in “trace” conditioning tasks, where a brief temporal gap separates the conditioned stimulus (CS) and unconditioned stimulus (US) presentations (1013). This temporal binding function is disrupted in aging (14).Thus, relational organization and temporal binding are well-known functions of the hippocampus and are sensitive to aging, but potential links between these functions in declarative memory and its age-related decline remain hypothetical. Here, we test the hypothesis that the bridging of temporal gaps sustained by the hippocampus is a necessary condition for relational organization of memories (15, 16).
To unveil the link between hippocampal function in temporal binding and relational organization and their critical role in aging, we combined behavioral, cellular imaging, and spatially targeted interventional approaches following a two-step strategy. First, we used a trace conditioning procedure to identify the limit of temporal binding capability in young and old mice, and demonstrated with optogenetic tools that temporal binding relies on dorsal (d)CA1 activity over temporal gaps. Then, we demonstrated that dCA1-dependent temporal binding is necessary for the development of a relational organization of memories, and that loss of this activity plays a critical role in the aging-associated decline in relational memory. To examine the development of relational organization, we used a two-phase radial-maze task in mice and its virtual analog in humans. Cumulative evidence dissociates the performance when the hippocampus is compromised between the two phases of our task. Young mice with hippocampal lesions or inactivation (3, 4), like old mice (57), normally acquire reward associations of individual arms presented successively in the initial phase of our task but fail in choosing the rewarded arm when subsequently challenged to choose between a pair of the same arms in the test phase. Similarly, in a second version of our radial-maze task and its virtual analog for humans, aged mice (6), like aged humans (17), can learn individual pairs of arms (with one arm rewarded and one arm not rewarded in each pair) but fail in choosing the rewarded arm when presented within a recombined pair of the same arms in the test phase. To interpret the dissociation, we propose that a relational organization of associations among individual arm experiences made during the initial learning phase is needed for flexible memory expression as assessed in the test phase. In contrast, the learning of adaptive responses to individual arms or pairs would rely on simple stimulus–reward or stimulus–response associations acquired by repetition in the initial phase. By manipulating the temporal separation between individual arm experiences in the initial phase, we here confirmed our relational interpretation and found that the formation of a relational representation necessary for flexible memory expression is restricted by the limits of dCA1-dependent temporal binding.

Results

Activity of dCA1 Cells Across Temporal Gaps: A Necessary Condition for Temporal Binding and a Sufficient Condition for Reversing the Aging-Related Loss of Temporal Binding Capacity.

All experiments were conducted in accordance with European Directive 2010-63-EU and with approval from the Bordeaux University Animal Care and Use Committee CCEA50 (agreement number A33-063-098; authorization N°5012035A-N°1377). Young and old mice were trained in a trace fear conditioning procedure (Fig. 1A and SI Methods). In young mice, retention of the CS–US association is limited to less than 60-s trace intervals and requires activation of dCA1 neurons during acquisition (Fig. 1 B and C). Acquisition of tone conditioning was comparable among all trace conditions (as was the retention of context conditioning; Fig. S1A) but was retained only with trace intervals of less than 60 s. Thus, 24-h retention of the tone–shock association was similar in magnitude among groups trained with 0-, 5-, or 20-s trace intervals but was diminished in mice trained with a 40-s trace interval and null in mice trained with a 60-s trace (Fig. 1B, tone test). Correspondingly, conditioning with a trace interval of less than 60 s induced a specific activation of dCA1 neurons. In dCA1 (but not in other areas studied), postconditioning Fos level was higher in mice trained with a 20-s trace, compared with those trained with either a 0-s or a 60-s trace (Fig. 1B, Fos for dCA1 and Fig. S1B, other structures). Thus, dCA1 activation is associated with the combination of a demand for and success in temporal binding. Conversely, optogenetic inhibition of dCA1 neurons during the trace interval blocks otherwise successful retention of the CS–US association. In freely moving mice expressing the inhibitory ArchT in dCA1 cells and chronically implanted with optic fibers in this hippocampal subfield, transitory inhibition of dCA1 neuronal activity was performed during the conditioning using a 20-s trace interval. ArchT mice with light on in each trace interval that normally acquired the 20-s trace tone conditioning (Fig. S1C, Left) were significantly impaired in the 24-h retention test of the tone–shock association, compared with both GFP controls and ArchT mice that were submitted to conditioning with light on during equivalent periods outside the trace interval (Fig. 1C, tone test). In contrast, tone conditioning acquired under the 0-s trace condition was not affected by light on in the 20-s tone-CS period immediately preceding the shock (Fig. S1C). Also, retention of context conditioning was remarkably unaffected under all inactivation conditions (Fig. 1C, context test). Finally, additional experiments showed that inhibition of the dCA2/CA3 subfield failed to produce any impairment (Fig. S1D), hence confirming that dCA1 was the critical area involved in the impairment of temporal binding, even though marginal contribution of extra CA1 cells cannot be ruled out. Thus, dCA1 neuronal activity during the trace interval is a necessary condition for bridging the temporal gap and enabling CS and US events to be bound together in memory. In contrast, dCA1 neuronal activity does not seem to be critically needed for linking the environmental cues into a (relational) memory representation of the context. Since environmental cues are temporally contiguous, the present dissociation between trace and contextual memories indicates a selective role of dCA1 cells in temporal binding in memory.
Fig. 1.
In trace tone–fear conditioning, the formation of long-term memory of the trace CS–US association is limited to less than 60-s intervals between the CS and US, sustained by CA1 cell activity across the trace interval, and disrupted in aging. (A) Protocol: Young (3- to 4-mo-old) and aged (21- to 23-mo-old) mice were submitted to the acquisition of conditioning: three pairings of a tone (CS) and mild electric foot shock (US) with a time interval between the two (trace) of either 0, 5, 20, 40, or 60 s depending on the group. Young mice of the 0-, 20-, and 60-s groups were prepared for Fos immunostaining. The remaining mice were submitted to the “tone” and “context” retention tests the day after conditioning: % time spent freezing during 2 min of exposure to the tone (in a neutral context) and to the conditioning context, respectively, was compared with % freezing during 2 min before the tone in a neutral context (cf. SI Methods). (B) Behavior and Fos imaging in young mice. (B, Left) Tone test: The 24-h retention of the tone–shock association is dependent on the trace condition during conditioning {two-way ANOVA: significant trace x tone [repeated (rep.) measures: no tone vs. tone] interaction (F4,38 = 15.457; P < 0.0001); tone effect was significant for all trace conditions except for the 60-s trace [rep. measures: P < 0.001 for 0-, 5-, and 20-s trace, P < 0.01 for 40-s trace, and P = 0.201, not significant (ns) for 60-s trace]}, showing that successful temporal binding in long-term memory is limited to less than 60-s distant stimuli. In contrast, neither the acquisition of conditioning nor the retention of context conditioning was dependent on the trace condition (cf. Fig. S1A). N = 8, 8, 8, 7, and 12 for the 0-, 5-, 20-, 40-, and 60-s trace group, respectively. (B, Right) CA1 Fos+ cells measured after the conditioning phase are also dependent on the trace condition [one-way ANOVA: significant effect of group (F3,46 = 10.113; P < 0.0001); post hoc: P = 0.0156, P < 0.0001, and P = 0.0031, respectively, for 0-, 20-, and 60-s vs. naive; P = 0.0102 and P = 0.0299, respectively, for 0 and 60 s vs. 20 s; and P = 0.6219, ns for 0 vs. 60 s], showing that conditioning leading to maximal temporal binding (i.e., with a 20-s trace) is associated with a specific CA1 activation (cf. Fig. S1B). N = 16, 11, 11, and 12 for the naive, 0-, 20-, and 60-s trace group, respectively. (C) Retention effects of optogenetic inactivation of CA1 during the acquisition of 20-s trace conditioning in young mice. The 24-h retention of tone trace conditioning is altered by in-trace inactivation compared with both control conditions [significant group x tone interaction (F2,52 = 6.812; P = 0.0024); significant interaction for in trace vs. out of trace (F1,29 = 9.622; P = 0.0043) and vs. GFP (F1,40 = 12.951; P = 0.0009) but not for out of trace vs. GFP (P = 0.4538, ns)]. In contrast, the retention of context assessed by freezing difference between the neutral and conditioning context is similar among the groups [group x context (rep. measures: neutral vs. conditioning): F2,52 = 2.278; P = 0.1126, ns], and in-trace inhibition of CA2/CA3 instead of CA1 has no effect on tone retention (Figs. S1D and S4, histology). Thus, CA1 activity across the trace interval during conditioning is a necessary condition for successful temporal binding of the CS and US in memory. n = 18, n = 13, and n = 24 for in trace, out of trace, and GFP, respectively. (D) Retention of conditioning in old mice. The retention of tone conditioning is dependent on the trace, indicating that memory of the CS–US association is only retained when the temporal separation of the CS and US was less than 20 s [significant trace x tone interaction (F2,31 = 5.341; P = 0.0102); tone effect is significant for 0- and 5-s trace conditions (rep. measures: P = 0.0034 and P = 0.0024, respectively) but not for the 20-s trace (P = 0.923, ns)]. In contrast, the retention of context is largely similar among the groups (trace x context interaction: F2,31 = 2.246; P = 0.1228, ns), just as was the acquisition of conditioning (cf. Fig. S1E). Thus, temporal binding capability is limited to less than 20-s gaps in old mice, and diminished in comparison with young animals. N = 12, 8, and 14 for 0-, 5-, and 20-s trace group, respectively. (E) Retention effects of optogenetic activation of CA1 during acquisition of 40-s trace conditioning in old mice. Without affecting the retention of context conditioning, in-trace (but not out-of-trace) activation enables the retention of tone conditioning, which is normally not retained in old mice [significant group x tone interaction (F2,15 = 5.17; P = 0.0196); the tone effect is significant for the in-trace group (P = 0.0002) but not in the other two groups (P = 0.114, ns and P = 0.335, ns, respectively, for out of trace and no light)]. Thus, CA1 (but not CA2/CA3; cf. Fig. S1F) activity across temporal gaps is sufficient to restore the age-related defect of temporal binding in memory. Note that the age-related deficit is associated with overall increased levels of freezing, suggesting fear generalization that was normalized by both in-trace and out-of-trace activation. n = 8, 5, and 5 for the in-trace, out-of-trace, and no-light groups, respectively. *P < 0.05; **P < 0.01; ***P < 0.001. Data are presented as mean ± SEM.
Fig. S1.
Trace tone–fear conditioning in mice. See protocol and main results in Fig. 1. (A) Behavioral results in young mice. (A, Left) The acquisition of tone conditioning was similar among the trace conditions {two-way ANOVA: significant effect of tone rep. (F2,76 = 66.67; P < 0.0001); trace (P = 0.1121), and trace x tone rep. [P = 0.2811, not significant (ns); in each ITI condition: significant tone rep. effect (all P < 0.008)]}. (A, Right) The 24-h retention of the context–shock association is also similar among the trace conditions [rep. measures (neutral vs. conditioning): F1,38 = 90.534; P < 0.0001; for each ITI condition: P < 0.005; trace and trace x rep. measures: P = 0.405 and P = 0.2811, ns, respectively], hence contrasting to the retention of the tone–shock association that diminishes progressively with trace intervals increasing above 20 s (Fig. 1B, Left). (B) Fos results in young mice. (B) There is significant Fos activation induced by acquisition of conditioning in all structures studied compared with the naive condition [significant group effect in all structures (all P < 0.001); post hoc: naive vs. each trace condition: all P < 0.01] but, in contrast to what is seen in dCA1 (Fig. 1B, Right), the activation is not higher under the 20-s trace condition than with the other two traces: 0 s, which requires no temporal binding, and 60 s, which requires temporal binding above young mouse capability. These findings indicate that activation of these brain structures during conditioning is unrelated to successful temporal binding of the CS and US events in memory. BLA, basolateral amygdala; LA, lateral amygdala; PrL and IL: prelimbic and infralimbic part of the prefrontal cortex, respectively. * and **: P < 0.05 and P < 0.01. (C) Optogenetic inactivation of dCA1 during the acquisition of conditioning in young mice. (C, Left) Acquisition of tone conditioning is similar among the groups whether under the 20-s trace condition {two-way ANOVA: significant effect of tone rep. (F2,102 = 94,054; P < 0.0001); group (P = 0.5187) and group x tone rep. [P = 0.5918, ns; in each group condition: significant tone rep. effect (all P < 0.0001)]} or the 0-s trace condition [two-way ANOVA: significant effect of tone rep. (F2,28 = 35.235; P < 0.0001); group (P = 0.079) and group x tone rep. (P = 0.6247, ns); in each group condition: significant tone rep. effect (all P < 0.008)]. (C, Middle and Right) In groups trained under delay conditioning (0-s trace), 24-h retention of both tone and context conditioning is also similar among the groups [Left, tone retention: two-way ANOVA: significant effect of tone (F1,14 = 112.577; P < 0.0001); group (P = 0.2876), group x tone rep. (P = 0.2514, ns); Right, context retention: two-way ANOVA: significant effect of context (F1,14 = 96.786; P < 0.0001); group (P = 0.1524), group x context rep. (P = 0.1544, ns)]. (D) Optogenetic inhibition of dCA2 and dCA3 during the 20-s trace interval in the acquisition of tone–fear conditioning in young mice. (D, Left) In contrast to what was seen following optogenetic inhibition of dCA1 (Fig. 1C), inhibition of either dCA2 or dCA3 during the 20-s trace interval failed to affect the 24-h retention of tone conditioning [for dCA2 (Left): two-way ANOVA: significant effect of tone (F1,9 = 33,892; P = 0.0003), group (P = 0.091) and group x tone rep. (P = 0.9641, ns); for dCA3 (Right): two-way ANOVA: significant effect of tone (F1,9 = 122,33; P < 0.0001), group (P = 0.1196) and group x tone rep. (P = 0.5803, ns); in each group condition: significant tone effect (all P < 0.02)]. These observations show that the impairment of temporal binding seen with the manipulation of dCA1 is due to selective silencing of dCA1 cells. n = 5 to 7 per group. (D, Right) Graphs showing individual performance in each group. (E) Aging effect on the acquisition and 24-h retention of tone–fear conditioning. Aged mice acquired the tone conditioning at least as well as young mice in all trace conditions [acquisition: two-way ANOVA for each trace condition: significant effect of tone rep. (all P < 0.0001); age x tone rep. (all P > 0.09, ns) and age (P > 0.23, ns), except for the 20-s trace condition in which the old group displays overall more freezing than the young one (age effect: F1,28 = 6,169; P < 0.0193)]. The retention of this conditioning is also similar among the ages for mice trained with no or a short (5-s) trace, but there is an age-dependent impairment of the retention in the 20-s trace condition, where old mice display no retention of the tone–shock association [24-h retention: two-way ANOVA: significant effect of tone for each trace condition (all P < 0.0001); age and age x tone: ns for the 0- and 5-s trace conditions (all P > 0.10) but significant interaction in the 20-s trace condition (age x tone: F1,28 = 22,499; P < 0.0001)]. (F) Optogenetic activation of dCA1, dCA2, or dCA3 during the 40-s trace interval in the acquisition of tone–fear conditioning in old mice. In contrast to what is seen in the group subjected to optogenetic activation of dCA1 (replication of the findings reported in Fig. 1E), aged mice in which dCA2 or dCA3 was maintained activated during the 40-s trace interval in the acquisition of tone conditioning show no improvement of the 24-h retention of tone conditioning relative to the control aged group with laser off during acquisition. A two-way ANOVA reveals a significant effect of tone (F1,12 = 6,845; P = 0.0225) and group x tone interaction (F3,12 = 4,812; P = 0.02) but nonsignificant group effect (P = 0.4195, ns). The tone effect is significant in the group submitted to dCA1 activation (P = 0.0211) but not in the other three conditions (all P > 0.26, ns in each case). These observations show that the restoration of the temporal binding defect seen with the activation of dCA1 is due to selective manipulation of the dCA1 cells, supporting the conclusion that the age-related impairment comes from a lack of dCA1 activity during stimulus-free temporal intervals. n = 3 to 5 per group. *P < 0.05; **P < 0.01; ***P < 0.001. Data are presented as mean ± SEM.
Activation of dCA1 cells during the trace interval ameliorates the age-related impairment in the retention of the CS–US association (Fig. 1 D and E). Comparison of young and old normal mice trained with different trace intervals revealed an age-related reduction in temporal binding capacity. Both young and aged mice acquired the conditioning task successfully, and displayed significant retention of context conditioning (Fig. 1D, context test and Fig. S1E). By contrast (Fig. 1D, tone test), while the old mice trained with 0- or 5-s trace intervals exhibited significant 24-h retention of the tone–shock association, those trained with a 20-s trace interval exhibited no retention of the CS–US association. Thus, neither the ability to acquire an association across time nor the ability to form durable associative memories (like contextual memory) in general is altered in aged mice. Aging results in a selective deficit in long-term retention of an association between events separated by 20 s, corresponding to a reduction in the capacity for temporal binding compared with young mice. We next examined whether this selective impairment was due to a reduction in dCA1 neuronal activity bridging the trace interval. We thus tested whether optogenetic activation of dCA1 neurons during the trace interval could reverse the age-related impairment. In old mice expressing the activating channel rhodopsin 2 (ChR2) in dCA1 neurons, we compared the effects of activation “in trace” and “out of trace” and no activation of dCA1 (5 Hz, 40 s, three times) performed during conditioning using a 40-s trace, the longest trace interval associated with successful retention in young mice (Fig. 1B). The group trained with in-trace dCA1 activation only exhibited significant retention of the tone–shock association (Fig. 1E). Thus, maintaining dCA1 but not dCA2/CA3 (Fig. S1F) neuronal activity during the trace interval between the tone and the shock is sufficient to restore retention of the association between these events in old animals. Taken together, these findings reveal that dCA1 neuronal activity bridging the trace interval between the CS and the US is a necessary condition for subsequent storage of the CS–US association. In young mice, dCA1 activity bridges temporal gaps of up to 40 s, enabling the separate CS and US events to be bound in memory. This temporal binding capacity by dCA1 cells is compromised in aged mice, resulting in an inability to form associative memories of events separated by more than a few seconds.

Temporal Binding Supported by dCA1 Is Critical to the Development of a Relational Memory Organization, and Lack of Sustained dCA1 Activity During Periods of Temporal Binding Is the Cause of Age-Associated Loss of Relational Memory in Mice.

To study the role of temporal binding in the formation of relational organization allowing flexible memory expression, we used a radial-maze task in which aged mice and mice with hippocampal damage or hypofunction succeed in learning the individual arms in the acquisition phase but fail on the subsequent flexibility test (47). This selective deficit in flexible memory expression is believed to come from an impairment of relational organization of the maze arms and reward associations in memory.

Temporal binding capacity is a limiting factor in the formation of relational/flexible memories.

We here manipulated the temporal separation between the individual learning events during the acquisition phase by varying the intertrial interval (ITI) among different groups of young and old mice (Fig. 2A). Increasing the ITI did not interfere with initial learning, since all groups learned the individual arm–reward associations at similar rates (Fig. S2A), but performance in the flexibility test varied with ITI and age (Fig. 2B). Thus, young mice performed equally well in the flexibility probe when successive events had occurred up to 20 s apart during acquisition. With longer ITIs during learning, subsequent probe performance progressively declined and dropped to chance level when the ITI was 60 s. Aged mice performed well on the flexibility test when learning occurred with a 0- to 5-s ITI but performance dramatically dropped when learning occurred at longer ITIs (as early as 20 s). These findings show that separate arm experiences made during learning have to be related to one another to form a flexible memory, and such relational memory organization is limited by the temporal binding capability of each age. Thus, the findings demonstrate that (i) flexible memory expression assessed in our radial-maze task does rely on relational memory organization of individual arm experiences, just as we hypothesized, and (ii) temporal binding is a critical determinant of relational memory organization underlying flexible memory expression. Thus, flexibility per se is not impaired in aged mice, because test performance was normal when original learning occurred at the short ITI. This finding demonstrates that relational/declarative memory is normal in aging but that it is the reduction of temporal binding capability necessary for relating discontiguous events in memory which is responsible for the apparent degradation of relational/declarative memory occurring in aging.
Fig. 2.
Temporal binding sustained by CA1 is a critical determinant of declarative memory formation and its age-associated degradation: the radial-maze model in mice. (A) Protocol: In the acquisition phase, independent groups of young (3- to 4-mo-old) and aged (21- to 23-mo-old) mice learned the constant food (+)/no food (−) rewarding valence of each arm through daily sessions of 24 successive individual arm presentations, separated by an ITI of different duration among the groups. For behavioral analyses, each animal was trained until reaching the learning criterion and, 24 h after, was submitted to the flexibility probe. In this test, the reward contingencies among the arms remained unchanged but the arms were now presented by pairs to assess flexible memory expression as a model of declarative memory. For Fos analyses, groups of mice were prepared after the third training session of the acquisition phase (SI Methods). (B) Flexibility probe: Performance depends on the ITI condition under which memories were encoded, in an age-specific manner [age x ITI (0-, 20-, and 60-s ITI): F2,46 = 4.975; P = 0.0111; age effect: P = 0.0001, P = 0.3967, not significant (ns) and P = 0.9032, ns for 20-, 0-, and 60-s ITI, respectively]. Thus, flexible memory expression relies on the capability to relate individual arm visits across time intervals, capability limited to less than 60-s interevent separation in young mice [ITI (20-, 40-, and 60-s ITI) effect: F2,17 = 5.045; P = 0.019. Post hoc: 20 s vs. 40 and 60 s, P < 0.05] and to only 5-s intervals in aged mice [ITI (0, 5, 10, and 20 s) effect: F3,26 = 5.011; P = 0.0071. Post hoc: 0 s vs. 20 s, 5 s vs. 10 and 20 s: P < 0.05]. n = 7 to 10 per group. ***P < 0.001 vs. aged. (C) CA1 Fos+ cells in young and old mice: Training in the acquisition phase of the radial-maze task produces an ITI-dependent pattern of CA1 Fos activation, which resembles the one induced by trace fear conditioning in young mice, but this activation is not seen in aged mice [age x ITI (0-, 20-, and 60-s ITI): F2,59 = 5.582; P = 0.006; age effect: P = 0.0002, P = 0.2381, ns and P = 0.885, ns for 20-, 0-, and 60-s ITI, respectively]. n = 7 to 14 per group. (D) Transitory inactivation of CA1 through local lidocaine infusion during the acquisition phase spares the acquisition of individual arm valence, whichever the ITI condition (cf. Fig. S2C), but produces subsequent impairment of performance in the flexibility probe in the sole 20-s ITI condition under which temporal binding of successive learning events normally occurs [significant lidocaine x ITI interaction (F2,36 = 4.36; P = 0.0201); lidocaine effect: P < 0.0001, P = 0.4546, and P = 0.6934 in the 20-, 0-, and 60-s ITI condition, respectively], thus mimicking the aging effect. n = 6 to 8 per group. (E) Optogenetic inactivation of CA1 during the 20-s ITI between events in the acquisition phase also produces a subsequent impairment of performance in the flexibility probe. Here we used the second version of our radial-maze design (SI Methods), also used in humans (Fig. 3). The initial acquisition of separate pairs was spared by CA1 inactivation in 20-s ITI but performance was severely diminished in the “recombined” test of flexibility in the “laser on” group compared with controls [significant group x acquisition probe interaction (F1,14 = 5.502; P = 0.034)]. Thus, CA1 is needed to bridge a temporal gap, and this temporal binding function is crucial for relational organization sustaining the formation of flexible/declarative memory. n = 7 to 8 per group. *P < 0.05; **P < 0.01; ***P < 0.001. Data are presented as mean ± SEM.
Fig. S2.
Radial-maze task of relational memory in mice. See protocol and main results in Fig. 2. (A) Fos activations induced by acquisition of the radial-maze task in young and old mice in areas other than dCA1: In contrast to what is seen in dorsal dCA1 (cf. Fig. 2C), in the dorsal DG and dorsomedial (DM) striatum, Fos activation was unrelated to successful temporal binding [two-way ANOVA; for DG: ITI, age, and age x ITI [all P > 0.07, not significant (ns); for DM striatum: ITI and age x ITI (all P > 0.27, ns) but significant age effect (F1,76 = 5.155; P = 0.026), as Fos levels are overall higher in old mice than in young ones]. In dCA3, though, a pattern of activation as a function of ITI and age, similar to the one seen in dCA1 but nonsignificant, is observed [significant effects of ITI (F2,76 = 4.483; P = 0.0144) and age (F1,76 = 6.988; P = 0.01) but age x ITI (P = 0.1067, ns)]. (B) Pretraining lidocaine infusion into dorsal dCA1 during the acquisition phase of the radial-maze task in young mice produces a significant reduction of Fos activation in dCA1 only (Fos, lidocaine vs. control groups: P = 0.036 in dCA1, P = 0.2637, ns in dCA3 and P = 0.49, ns in DG) that spares the task acquisition [final performance expressed as normalized ratio of arm entry latencies: (−arm) − (+arm)/(−arm) + (+arm), two-way ANOVA with group (lidocaine, control) and ITI (0, 20, 30 s) factors, all P > 0.18, ns]. This treatment nevertheless impairs temporal binding function critical for flexible expression of spatial memories (cf. Fig. 2D). *P < 0.05; **P < 0.01. Data are presented as mean ± SEM.

Temporal binding relies on dCA1 activity during learning.

Analyses of Fos activation induced by learning the individual arm associations revealed an ITI-dependent increase in dCA1 activation, similar to that observed associated with trace fear conditioning. Thus, learning-related CA1 activation was found to depend on a combination of a demand for and success in temporal binding (Fig. 2C). In young mice, strong demand for temporal binding when probe performance was successful (20-s ITI) resulted in high levels of Fos activation in dCA1. However, there was less or no Fos activation in dCA1 either when there was no demand for temporal binding and probe performance was normal (0-s ITI) or when the interval between trials exceeded temporal binding capacity and probe performance failed (60-s ITI). In contrast to young animals, dCA1 neuronal activity was not recruited across the 20-s ITI in aged mice, corresponding to their poor subsequent probe test performance following training at this ITI. Instead, Fos activation in the dorsomedial striatum was greater in older mice, independent of the length of the ITI duration (Fig. S2B). Furthermore, neuronal activity in dCA1 is essential for temporal binding that supports subsequent flexible memory expression. First, local infusions of the anesthetic lidocaine were performed in young mice before each daily session of the initial phase but not before the flexibility-test session. We found that inactivation of dCA1 (see reduced Fos levels; Fig. S2C) during the entire acquisition phase impaired flexible memory expression when animals were trained at 20-s but not 0- or 60-s ITI (Fig. 2D), resulting in a decrement in temporal binding capacity similar to that observed in aged mice. Second, optogenetic inhibition of CA1 during the critical 20-s ITI in the acquisition phase also resulted in a subsequent impairment in the flexibility probe (Fig. 2E). This result demonstrates a dependence on dCA1 information processing to bridge a temporal gap. Finally, a procholinergic drug was found to rescue probe test performance in aged mice at the dose that concomitantly recovered activation in dCA1 at the critical 20-s ITI (Fig. S2D).
Altogether, the present findings show that a critical engagement of dCA1 is a necessary condition for bridging a temporal gap to form a relational memory organization, and indicate that the aging-related decline in declarative memory results from a compromised dCA1 function in temporal binding.

The Age-Associated Decline in Declarative Memory Depends on Temporal Binding in Humans as Well.

To test the validity of the above conclusions on the role of temporal binding in flexible/declarative memory formation for human senescence, young and aged participants selected as “cognitively normal” for their age (SI Methods) were submitted to the virtual analog of the radial-maze task previously used in mice (Fig. 3A). Experiments were approved by the following ethics committees: the CPP Aquitaine (Comité de Protection des Personnes), the CCTIRS (Comité Consultatif sur le Traitement de l'Information en Matière de Recherche dans le Domaine de la Santé), and the CNIL (Commission Nationale de l'Informatique et des Libertés). Written informed consent was obtained from all participants before any study-related procedure. Participants were divided into three intertrial interval conditions (0-, 20-, or 40-s ITI, matched for age and performance in other cognitive tests; Fig. S3A and Table S1). While all aged groups normally learned the initial pairs (Fig. 3B, Left), only those trained with a short ITI (0 or 20 s) could perform the flexibility test performance as well as younger adults (Fig. 3B, Right; replication Fig. S3B and Table S2). Thus, like in aged mice, a reduction in temporal binding causes the age-associated impairment in relational organization allowing flexible memory expression.
Fig. 3.
Age-associated decline in flexible/declarative memory depends on temporal binding in humans: the virtual radial-maze task. (A) Protocol: In the acquisition phase, young (18- to 25-y-old; n = 43) and aged (59- to 75-y-old; n = 40) participants learned the constant rewarding (+, virtual coin)/no rewarding (−) valence of each arm through successive presentations of invariant pairs (1 arm+, 1 arm−), separated by an ITI of 0, 20, or 40 s, depending on the group, until reaching the learning criterion. In the flexibility probe, the reward contingencies among the arms remained unchanged, but the arms were rearranged into novel pairings to assess flexible memory expression. Participants were matched among the ITI conditions according to their age and performance in other cognitive tests (SI Methods, Fig. S3A, and Table S1). (B) Results: In acquisition, aged participants required more training to reach the learning criterion but eventually learned the task as well as the young participants, whichever the ITI condition (Left Top: final performance age, ITI and age x ITI: all P > 0.14, not significant (ns); Left Bottom: trials to criterion age effect: F1,77 = 6.02, P = 0.0164; ITI and age x ITI: P = 0.0676, ns and P = 0.0729, ns). In contrast, in the flexibility probe, there was an age-related impairment dependent on the ITI condition under which the task was acquired [Right: flexibility age x ITI (F2,77 = 3.473, P = 0.036); significant ITI effect in aged groups (F2,37 = 3.828, P = 0.0308; post hoc 40 s vs. 0 and 20 s, P < 0.05) but not in young (P = 0.74, ns); significant age effect for 40- (P < 0.0001) but not 0- or 20-s ITI (all P > 0.057, ns)]. Thus, the age-related loss of flexibility is due to a reduction of temporal binding capability. n = 10 to 18 per group. ***P < 0.001. Data are presented as mean ± SEM.
Fig. S3.
Radial-maze task of relational memory in humans. Results taken from two studies independent of the study described in the main text (Fig. 3) confirm the conclusion that age-related loss in temporal binding function is the primary cause for the degradation of relational memory organization leading to declarative memory impairments in aging. Like in our main experiment, after having learned spatial information (learning criterion reached for initial pairs of arms), old participants indeed display normal and above chance-level performance in the flexibility probe (recombination of initial pairs) if they were trained with a 0-s intertrial interval in the acquisition stage (i.e., in experiment A, CL1-NEURO-006, baseline) but they failed in the probe when a 40-s ITI was used during acquisition (i.e., in experiment B, CL1-47445-004_OP0787) placebo group. The two experiments were performed in a clinical research center (Optimed) with the same criteria of inclusion (age, cognitive evaluation) as in our main study (cf. Table S2). Data are presented as mean ± SEM.
Table S1.
Participant demographic and neuropsychological testing information: Experiment aimed at testing ITI variation in young and normal old participants
Demographic information and neuropsychological testingYoung ITI 0 sYoung ITI 20 sYoung ITI 40 sOld ITI 0 sOld ITI 20 sOld ITI 40 s
Demographic information      
 N151018121315
 Age, y24.2 (0.88)23 (0.63)24.8 (0.88)64.5 (1.19)68.1 (1.48)64.5 (1.14)
 Gender, m/f9/63/79/95/75/86/9
 Education, y13.7 (0.43)14.2 (0.35)14.7 (0.57)14.7 (1.14)13.7 (1.05)13.2 (0.7)
Neuropsychological testing      
 MMSE29.1 (0.2)29.6 (0.14)29.2 (0.28)
 Grober & Buschke score, total free recall/4842.4 (0.47)40.2 (1.23)40.8 (0.60)35.1 (1.77)34.7 (1.35)35.1 (1.36)
Data are presented as mean (±SEM).
Table S2.
Participant demographic and neuropsychological testing information
Demographic information and neuropsychological testingOld ITI 0 sOld ITI 40 s
Demographic information  
 N4820
 Age68.9 (0.4)68.8 (0.7)
 Gender, m/f20/286/14
 Education, y11.2 (0.43)11.3 (0.8)
Neuropsychological testing  
 MMSE29.5 (0.10)28.9 (0.16)
 Grober & Buschke, total free recall/4831.4 (0.75)31.7 (1.49)
Partial data from two pharmacological studies: baseline (old ITI 0 s from experiment CL1-NEURO-006) and placebo group (old ITI 40 s from experiment CL1-47445–004_OP0787). Data are presented as mean (±SEM).

SI Methods

Experiments in Mice.

Subjects.

Subjects were young adult (3- to 4-mo-old) and aged (21- to 23-mo-old) C57BL/6 male mice obtained from Charles River and housed in a standardized animal room (23 °C; lights on 7 AM to 7 PM; four or five mice per cage). They were moved to individual cages 2 wk before the beginning of experiments and, for radial-maze experiments, were submitted to a progressive and partial food deprivation (85 to 88% of their free feeding weight) throughout the duration of training.
All experiments were conducted in accordance with European Directive 2010-63-EU and with approval from the Bordeaux University Animal Care and Use Committee CCEA50 (agreement number A33-063-098; authorization N°5012035A-N°1377). All efforts were made to minimize suffering and reduce the number of animals used.

Optogenetic manipulations of dCA1 or dCA2/dCA3 activity in behaving animals.

Mice underwent two surgeries 7 to 9 d apart. First, AAV5-expressing ArchT-GFP, ChR2-GFP, or GFP [≈0.4 μL per hemisphere; University of North Carolina (UNC) Vector Core] were bilaterally injected using glass pipettes (tip diameter, 10 to 20 mm) connected to a picospritzer (Parker Hannifin) into the dCA1 (two injection sites to minimize diffusion to the extra-dCA1 area) at the following coordinates: 1.8 and 2.5 mm posterior to the Bregma, 1.3 and 2 mm lateral of midline, and 1.4 and 1.4 mm below the skull surface, according to a classical stereotaxic procedure. Maximum and minimum areas of virus injection from the anterior dorsal dCA1 (Bregma −1.34) to the posterior dorsal dCA1 (Bregma −2.7) are represented in Fig. S4A (dashed lines). Second, mice were implanted with bilateral optic fiber implants (diameter, 200 µm; numerical aperture, 0.39; flat tip; Thorlabs) directed to the dorsal dCA1 [coordinates: anteroposterior (AP) 1.8 mm, lateral (L) ±1.3 mm, and dorsoventral (DV) 1.4 mm]. Implants were fixed to the skull with Super-Bond dental cement (Sun Medical). Correct placement of fibers was visually checked on hippocampal slices to reject all mice with fiber located outside the medial part of the anterior dorsal CA1 (gray areas in Fig. S4A). A picture representative of the virus injection and the fiber position is presented in Fig. S4C. In additional experiments targeting dCA2 or dCA3 to assess the area selectivity of the effects observed with dCA1 manipulations, the procedure was identical except for the coordinates of virus injection (0.2 µL each side; AP −2.0, L ±2.5, DV 2) and fiber implantation (AP −1.8, L ±2.1, DV 1.7 for dCA2; AP −1.85, L ±2.45, DV 1.9 for dCA3). Mice implanted outside of the represented sites were excluded from the study. All mice exhibited virus expression at least in dCA2 and dCA3, as represented in Fig. S4B.
Fig. S4.
(A and B) Histology: optogenetic manipulations of dCA1 (A) and dCA2/dCA3 (B) in behaving mice. Maximum and minimum areas of virus injections (blue and red dashed lines, respectively) from the anterior dorsal CA1 (Bregma −1.34 mm) to the posterior dorsal CA1 (Bregma −2.7) and medial (gray) zone for correct implantation of the optic fiber. (C) Photomicrographs representative of the virus injection and fiber position in the anterior-dorsal CA1 (Top Left), dCA2 (Top Right), posterior-dorsal CA1 (Bottom Left), and dCA3 (Bottom Right) at 2.5× magnification. (Scale bars, 300 µm.) (D) Photomicrograph at 20× and 40× magnification showing viral expression in dCA2.
The behavioral procedure started after a 10- to 12-d recovery. All optogenetic manipulations but one were performed during the acquisition of fear conditioning and at specific moments depending on the group, as explained in the text and Fig. 1. One optogenetic manipulation was performed during the acquisition phase of our radial-maze task, as explained in the text and Fig. 2E. The light (≈6 mW per implanted fiber) was bilaterally conducted from the laser (OptoDuet 473/593 nm; IkeCool) to the mice via two fiber-optic patch cords (diameter, 200 μm; Doric Lenses), connected to a rotary joint (1 × 2 fiber-optic rotary joint; Doric Lenses) that allowed mice to freely move in the behavioral apparatus. For inactivation (ArchT), the light was continuously delivered; for activation (ChR2), the light was delivered at 5 Hz (5 ms laser on, 195 ms laser off).

Intra-dCA1 infusions of lidocaine in freely moving mice.

The same procedure is detailed in ref. 4. Animals underwent surgery for implantation of guide cannulae above the dorsal dCA1 (coordinates: AP −2.0, L ±1.3, DV −0.8), and the behavioral experiment started after a 10- to 12-d recovery. Mice were subjected to bilateral infusions of lidocaine (0.3 µL each side) or pseudoinjections (control) into the dorsal dCA1 (by placing a cannula in each guide cannula while manually restraining the awake mouse; the cannula was 1 mm longer than the guide cannula to reach dCA1) immediately before each session of acquisition of the radial-maze task but not before the flexibility-test session. As lidocaine efficacy was estimated to last 20 to 25 min, dCA1 was inhibited each training session (lasting around 25 min as well) of the acquisition phase of the radial-maze task, during almost the entire session. However, dCA1 was not inactivated at the time of flexibility testing to selectively assess the effect of dCA1 inactivation performed during the acquisition phase. Correct placement of guide cannulae/cannulae was visually checked on hippocampal slices. Mice were eliminated from the statistical analysis if their guide cannula or cannula placements were not correct. The guide cannulae/cannulae should be above/inside the medial zone of anterior dorsal dCA1 (Fig. S4A). The efficacy of lidocaine injections for selectively inactivating dCA1 [but neither dCA3 nor dentate gyrus (DG)] was verified in independent groups by performing Fos immunohistochemistry after the last training session of acquisition (cf. Fig. S2C).

Trace fear conditioning.

The equipment was as used and described in ref. 31, except for optogenetic manipulations that required the use of a conditioning chamber equipped for optogenetics (OptoPath platform; Imetronic).
Day 1: Acquisition of conditioning.
Each animal was placed in the conditioning chamber for 8 min, during which it received three pairings of a tone CS (65 dB, 1 kHz, 30 s) and a footshock US (0.3 mA, 50 Hz, 1 s) separated by a trace interval varying from 0 to 60 s. The 0-s trace condition corresponds to the standard delay conditioning, where the CS and US are contiguous.
Day 2: Retention tests.
Mice were first reexposed to the tone alone in a dark and unknown chamber (tone test; 6 min with 2-min tone in the middle) and, 2 h later, were reexposed to the conditioning chamber alone (context test; 4 min).
Animals were continuously recorded on videotape for off-line scoring of freezing behavior by an observer blinded to the experimental groups. Freezing is defined as a lack of all movement except for respiratory-related movements, and was expressed as % time spent freezing over specific periods: tone delivery during the acquisition (3 × 30 s) and tone test (2 min), the neutral conditioning chamber during the tone test (no-tone period, 2 min), and the conditioning context in the context test (4 min). Mice expressing less than 10% freezing during the last (third) tone in acquisition were rejected from the experiment.

Memory testing in the radial maze.

The apparatus was an open eight-arm radial maze, fully automatized by video tracking (Imetronic), described previously (4, 7). The task was designed specifically to assess the formation of relational memory representation of separately acquired pieces of information (6). The procedure followed in the present experiments has already been described in detail (e.g., 7). Briefly, after two sessions of habituation to the maze, each mouse was attributed a set of six adjacent arms, among which three were baited and three were not baited. The food location among the six arms remained constant over the entire experiment but the arm presentation was changed between the two phases of the task: acquisition and the flexibility test. In the acquisition phase, the arms were opened one by one successively with a waiting-time intertrial interval (ITI; 0 to 60 s depending on the group) separating the 24 successive arm visits made within a session, and a 24-h interval between sessions. Training sessions (minimum, 5; maximum, 12) were repeated until reaching the acquisition criterion (when latency to enter nonbaited arms was at least 30% longer than for baited ones). Mice failing to reach the criterion within 12 training sessions were rejected from the experiment, except in the experiment with intra-dCA1 infusions, in which the acquisition period was fixed at six sessions for all mice (to keep constant the number of intracerebral infusions performed), whether the criterion was reached in the interval or not, and in Fos experiments, in which animals were prepared for immunohistochemistry after partial acquisition (third session of acquisition). One day after the end of the acquisition phase, the flexibility-test phase was performed. In this test, the arms were opened by pairs (one baited, one nonbaited) instead of one by one. Memory flexibility is reflected in the capability to correctly choose the food-containing arm among each pair, indicating that a relational representation was formed of previous separate experiences of individual arms. The test session was made up of 20 pair presentations, and performance was indexed by the percentage of correct choices.
In the experiment consisting of optogenetic inactivation of dCA1 in young mice (presented in Fig. 2E), we chose to use the second version of our radial-maze design (6), the version which has been translated to human subjects using a virtual analog of the radial maze (17), as we wanted to further increase similarities with the study performed in humans (presented in Fig. 3). The procedure of the task for mice is similar to the one previously described except that initial acquisition consists of learning three pairs of arms, and the flexibility probe consists of recombination of two initial pairs into a novel pairing (6). Acquisition training (20 trials of one pair presentation each day) was performed (during a minimum of 5 to a maximum of 12 d) until reaching the acquisition criterion (overall >70% correct over 2 d; >67% correct for each pair) with an ITI of 20 s. Flexibility testing was performed the day following reaching the acquisition criterion. The test session consisted of 20 trials of one pair presentation (ITI, 20 s; three pairs). Eight trials with the critical pair made from recombination of two previous pairs were intermixed with two control pairs: one unchanged from acquisition, the other made up of unlearned arms. Performance was measured by the percentage of correct choices of the rewarded arm in each pair.

Fos immunostaining and counting.

Fos immunostaining and counting were performed as described (7) (c-Fos antibody; sc-52 1/5000; Santa Cruz). Mice were killed 90 min after the beginning of the acquisition of trace fear conditioning, or after the third training session in the acquisition phase of the radial-maze task. Mice subjected to fear conditioning were compared with a naive home-caged group, whereas mice trained in the radial-maze task were compared with a “treadmill” group of the same age. The treadmill (28 cm long × 5 cm wide × 20 cm high) was placed in the same room as the radial maze and mice were subjected to the same amount and duration of training sessions as the radial-maze learning groups (2 d corresponding to habituation plus 3 d corresponding to acquisition of the radial-maze task). For each animal, the number of Fos-immunoreactive neurons was counted bilaterally in each brain area studied [dorsal hippocampus (CA1, CA3, DG); dorsomedial striatum; lateral amygdala; basolateral amygdala; prefrontal cortex (infralimbic, prelimbic); entorhinal cortex] using three or four consecutive sections. The number of positive nuclei per mm2 was quantified using an Olympus BX50 microscope equipped with a computerized live imaging analysis system (Visiolab 2000 version 4.50; Biocom).

Experiments in Humans.

Ethics and participants.

Young (18- to 25-y-old; n = 44) and old (59- to 75-y-old; n = 44) participants were recruited in an academic environment, and the elderly subjects were selected as cognitively normal for their age, based on their score in the Mini Mental Score Evaluation (MMSE > 27/30) and in the Grober & Buschke test of declarative memory (G&B delayed recall > 18/48). As shown in Tables S1 and S2, the old and young people in our study were well-matched in educational level and sex ratio.
Participants were in good health as indicated by medical questionnaires and had normal or corrected-to-normal visual acuity. A history of any neuropsychiatric/neurological disorder, dementia, or any severe disease [cardiovascular, pulmonary, renal, hepatic, gastrointestinal, metabolic, ophthalmic (i.e., glaucoma), hormonal, or hematological], chronic infection, consumption of ≥40 g of alcohol per day, and/or any medicine with effects on the central nervous system (except sleeping pills no benzodiazepine) in the preceding month and pregnancy in women were all exclusionary. Experiments were approved by the following ethics committees: the CPP Aquitaine (Comité de Protection des Personnes), the CCTIRS (Comité Consultatif sur le Traitement de l'Information en Matière de Recherche dans le Domaine de la Santé), and the CNIL (Commission Nationale de l'Informatique et des Libertés).

Memory testing in the virtual radial maze.

We used a virtual analog of the radial-maze task for mice. The principle of the task was the same as the one previously described (17), but the virtual maze was redesigned and automatized (OptoPath; ANR-10-EQX-008-1; in collaboration with Imetronic) to make the test as similar as possible as the one used with the mice. Briefly, in the acquisition phase of the task, each subject was faced with successive presentations of six pairs of adjacent arms and required to visit one of the two arms each trial. In each pair, one arm always contained a reward (virtual coin) at its end, while the other arm never contained any reward. Training continued until reaching the acquisition criterion (when the number of incorrect choices of the nonrewarded arm was less than 2 over 12 consecutive trials), and a minimum of 6 trials by pair or maximum of 20 trials by pair were performed. Once the acquisition criterion was reached, the subject was then subjected to the probe test in which nothing was modified except that the arms were now presented in novel pairings (recombination of previous pairs) to assess the flexibility of memory expression. The young and aged participants were divided into three intertrial interval conditions (0, 20, or 40 s) for the acquisition phase on the basis of their performance in the Grober and Buschke test, such that the mean score in this classical test of declarative memory was similar among the ITI conditions in each age group. Participants who failed to reach the acquisition criterion were rejected from the experiment.

Statistical Analyses.

Data were analyzed using one- or two-way ANOVAs. When appropriate, post hoc comparisons were achieved by a Fisher protected least significant difference test with a significance level at P < 0.05.

Discussion

The present findings in mice and humans establish a causal connection between two well-known and age-sensitive functions of the hippocampus: temporal binding and relational organization of memories. We show that (i) temporal binding is a necessary condition for linking experiences separated by brief time intervals into an organized representation that supports flexible memory expression, characteristic of declarative memory, and (ii) temporal binding critical to performance relies on dCA1 activity across temporal gaps between experiences. Our study thus validates important hypotheses established through a history of research on hippocampal function in associations across time, memory formation, and cognitive aging (9, 12, 15, 1823).
At the psychological level, our parallel approaches in trace conditioning and radial-maze learning demonstrate that the capacity of temporal binding is a limiting factor in the formation of a relational organization associated with declarative memory. In young and aged mice, the ability to relate discontiguous experiences of individual arms into a relational memory was limited to linking experiences across the same intervals as that supporting a CS–US association in trace conditioning. Thus, in the radial-maze task, memories for individual experiences were formed through repeated exposures to each individual arm, but a relational organization was formed only when the temporal separation allowed temporal binding, a capacity measured independently in our tests on trace conditioning.
The findings identify the age-associated reduction of temporal binding capacity as the primary cause for the memory impairment. Furthermore, these results support the conclusion that the ability to form a relational organization per se is not impaired in normal aging but rather is impaired only when temporal binding between experiences is compromised. Relational organization for spatial memory was intact in aged mice and humans as long as the demand on temporal binding was minimized by the temporal proximity of learning experiences. These findings challenge the commonly held view that aging produces impairments in declarative (1, 24) and spatial learning and memory (25, 26), and suggest the possibility that these aging-sensitive declarative tasks include a demand for binding memories across time.
The present findings not only show that temporal binding is a critical determinant of declarative memory formation and its age-related decline but also provide a potential basis for reconciling two conflicting theories of hippocampal function, spatial mapping and declarative memory. Commonly, these two functions have been respectively studied in animals and in humans, making their comparison difficult. However, a common feature of spatial mapping and declarative memory is that they both allow flexible expression of memories in modified testing situations. Hence, the present observations suggest that temporal binding plays a critical role in both spatial mapping and declarative memory by supporting the ability to form relational memory organizations.
What might be the underlying mechanisms of temporal binding in dCA1? The peculiarity of temporally distant stimuli is that they cannot be associated in memory through Hebbian mechanisms of synaptic plasticity, which are known to sustain the formation of long-lasting associations among co-occurring stimuli (27). Hebbian plasticity is triggered by the coactivation of neuronal assemblies encoding each stimulus during learning. To overcome the limitation of Hebbian plasticity for linking events separated in time, the presence of time cells in the CA1 subfield (20) provides a potential mechanism by which distinct events separated in time may be bound together in memory (20, 22, 23, 28). Namely, time cells were found to fire at successive moments during an empty temporal interval between key events, so that collectively the population of cells filled in the temporal gap and bridged the time intervals between the events. Such time-cell sequences in CA1 during temporal intervals would induce Hebbian plasticity among successively activated cells, and ultimately could thereby sustain the formation of an association between temporally separated events. This hypothesis implies that CA1 cell activity across time intervals between events must be a necessary condition for encoding a long-lasting associative memory of these events, just as shown by the present studies using optogenetic manipulations. Thus, our findings are compatible with the time-cell hypothesis even though they do not directly identify “time cells” as the underlying mechanism. Indeed, it is surprising that maintaining a randomly selected subset of cells by artificial activation was sufficient to restore the age-related memory deficit. However, it is possible that artificial stimulation of a few cells in dCA1 might lead to concomitant activation of large assemblies through recurrent collateral activation, and thereby induce Hebbian plasticity in synaptic contacts within the assemblies and with the cells activated by the event occurring just before and after the interval. Thereby, optogenetic stimulation of a few dCA1 cells could enable an association to be made across the interval in aged mice lacking (spontaneous) activation of time cells that would naturally enable the temporal binding process in young adult animals. Whatever the case might be, our findings are consistent with observations on the age-related loss in the excitability of CA1 cells (18, 29, 30) as a potential source of the reduction in temporal binding capacity.
In conclusion, our study identifies the bridging of temporal intervals by dCA1 activity as a critical determinant of relational organization that sustains characteristic flexibility of declarative memory expression, and shows that a deterioration of the temporal binding mechanism is the primary cause for age-associated declarative memory impairment.

Acknowledgments

We thank all of the personnel of the Animal Facility of the Neurocentre Magendie for mouse care, Mathieu Baudonnat and Florian Jacquot for their technical assistance in certain pilot experiments, and Aline Desmedt and Ludovic Calandreau for useful discussions. This work was supported by INSERM, a Bordeaux Neurocampus grant by le Conseil Régional d’Aquitaine, an EquipEx Grant OptoPath (ANR-10-EQX-008-1), Bordeaux Science Agro, CNRS, and National Institute of Mental Health Grant MH095297. We thank the French pension fund ProBTP and its previous director, Paul Grasset, who helped us to recruit healthy aged people to perform our experiments.

Supporting Information

Supporting Information (PDF)

References

1
SL Leal, MA Yassa, Neurocognitive aging and the hippocampus across species. Trends Neurosci 38, 800–812 (2015).
2
M Bunsey, H Eichenbaum, Conservation of hippocampal memory function in rats and humans. Nature 379, 255–257 (1996).
3
N Etchamendy, A Desmedt, C Cortes-Torrea, A Marighetto, R Jaffard, Hippocampal lesions and discrimination performance of mice in the radial maze: Sparing or impairment depending on the representational demands of the task. Hippocampus 13, 197–211 (2003).
4
F Mingaud, et al., The hippocampus plays a critical role at encoding discontiguous events for subsequent declarative memory expression in mice. Hippocampus 17, 264–270 (2007).
5
N Etchamendy, et al., Alleviation of a selective age-related relational memory deficit in mice by pharmacologically induced normalization of brain retinoid signaling. J Neurosci 21, 6423–6429 (2001).
6
A Marighetto, et al., Knowing which and knowing what: A potential mouse model for age-related human declarative memory decline. Eur J Neurosci 11, 3312–3322 (1999).
7
F Mingaud, et al., Retinoid hyposignaling contributes to aging-related decline in hippocampal function in short-term/working memory organization and long-term declarative memory encoding in mice. J Neurosci 28, 279–291 (2008).
8
PR Rapp, MT Kansky, H Eichenbaum, Learning and memory for hierarchical relationships in the monkey: Effects of aging. Behav Neurosci 110, 887–897 (1996).
9
JN Rawlins, E Tsaltas, The hippocampus, time and working memory. Behav Brain Res 10, 233–262 (1983).
10
KS LaBar, JF Disterhoft, Conditioning, awareness, and the hippocampus. Hippocampus 8, 620–626 (1998).
11
RE Clark, LR Squire, Classical conditioning and brain systems: The role of awareness. Science 280, 77–81 (1998).
12
PT Huerta, LD Sun, MA Wilson, S Tonegawa, Formation of temporal memory requires NMDA receptors within CA1 pyramidal neurons. Neuron 25, 473–480 (2000).
13
PR Solomon, ER Vander Schaaf, RF Thompson, DJ Weisz, Hippocampus and trace conditioning of the rabbit’s classically conditioned nictitating membrane response. Behav Neurosci 100, 729–744 (1986).
14
JF Disterhoft, MM Oh, Alterations in intrinsic neuronal excitability during normal aging. Aging Cell 6, 327–336 (2007).
15
GV Wallenstein, H Eichenbaum, ME Hasselmo, The hippocampus as an associator of discontiguous events. Trends Neurosci 21, 317–323 (1998).
16
BP Staresina, L Davachi, Mind the gap: Binding experiences across space and time in the human hippocampus. Neuron 63, 267–276 (2009).
17
N Etchamendy, K Konishi, GB Pike, A Marighetto, VD Bohbot, Evidence for a virtual human analog of a rodent relational memory task: A study of aging and fMRI in young adults. Hippocampus 22, 869–880 (2012).
18
JF Disterhoft, WW Wu, M Ohno, Biophysical alterations of hippocampal pyramidal neurons in learning, ageing and Alzheimer’s disease. Ageing Res Rev 3, 383–406 (2004).
19
JB Hales, JB Brewer, Activity in the hippocampus and neocortical working memory regions predicts successful associative memory for temporally discontiguous events. Neuropsychologia 48, 3351–3359 (2010).
20
CJ MacDonald, KQ Lepage, UT Eden, H Eichenbaum, Hippocampal “time cells” bridge the gap in memory for discontiguous events. Neuron 71, 737–749 (2011).
21
BJ Kraus, II RJ Robinson, JA White, H Eichenbaum, ME Hasselmo, Hippocampal “time cells”: Time versus path integration. Neuron 78, 1090–1101 (2013).
22
H Eichenbaum, Memory on time. Trends Cogn Sci 17, 81–88 (2013).
23
T Kitamura, CJ Macdonald, S Tonegawa, Entorhinal-hippocampal neuronal circuits bridge temporally discontiguous events. Learn Mem 22, 438–443 (2015).
24
CL Grady, et al., Age-related reductions in human recognition memory due to impaired encoding. Science 269, 218–221 (1995).
25
SN Burke, CA Barnes, Neural plasticity in the ageing brain. Nat Rev Neurosci 7, 30–40 (2006).
26
M Gallagher, PR Rapp, The use of animal models to study the effects of aging on cognition. Annu Rev Psychol 48, 339–370 (1997).
27
S Nabavi, et al., Engineering a memory with LTD and LTP. Nature 511, 348–352 (2014).
28
T Kitamura, et al., Island cells control temporal association memory. Science 343, 896–901 (2014).
29
CC Kaczorowski, JF Disterhoft, Memory deficits are associated with impaired ability to modulate neuronal excitability in middle-aged mice. Learn Mem 16, 362–366 (2009).
30
GG Murphy, V Shah, JW Hell, AJ Silva, Investigation of age-related cognitive decline using mice as a model system: Neurophysiological correlates. Am J Geriatr Psychiatry 14, 1012–1021 (2006).
31
M Potier, et al., Temporal memory and its enhancement by estradiol requires surface dynamics of hippocampal CA1 N-methyl-D-aspartate receptors. Biol Psychiatry 79, 735–745 (2016).

Information & Authors

Information

Published in

Go to Proceedings of the National Academy of Sciences
Proceedings of the National Academy of Sciences
Vol. 114 | No. 38
September 19, 2017
PubMed: 28874586

Classifications

Submission history

Published online: September 5, 2017
Published in issue: September 19, 2017

Keywords

  1. aging
  2. relational memory
  3. optogenetics
  4. trace conditioning
  5. radial maze

Acknowledgments

We thank all of the personnel of the Animal Facility of the Neurocentre Magendie for mouse care, Mathieu Baudonnat and Florian Jacquot for their technical assistance in certain pilot experiments, and Aline Desmedt and Ludovic Calandreau for useful discussions. This work was supported by INSERM, a Bordeaux Neurocampus grant by le Conseil Régional d’Aquitaine, an EquipEx Grant OptoPath (ANR-10-EQX-008-1), Bordeaux Science Agro, CNRS, and National Institute of Mental Health Grant MH095297. We thank the French pension fund ProBTP and its previous director, Paul Grasset, who helped us to recruit healthy aged people to perform our experiments.

Notes

This article is a PNAS Direct Submission.

Authors

Affiliations

Azza Sellami1
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Alice Shaam Al Abed1
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Laurent Brayda-Bruno1
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Nicole Etchamendy1
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Stéphane Valério1
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Marie Oulé
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Laura Pantaléon
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Valérie Lamothe
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Mylène Potier
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Katy Bernard
Institut de Recherche Internationale Servier, 92150 Suresnes, France;
Maritza Jabourian
Institut de Recherche Internationale Servier, 92150 Suresnes, France;
Cyril Herry
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Nicole Mons
Université de Bordeaux, F-33000 Bordeaux, France;
Institut de Neurosciences Cognitives et Intégratives d’Aquitaine, UMR 5287, CNRS, F-33600 Pessac, France;
Pier-Vincenzo Piazza
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;
Howard Eichenbaum
Center for Memory and Brain, Boston University, Boston, MA 02215
Deceased July 21, 2017.
Aline Marighetto3 [email protected]
Neurocentre Magendie, Physiopathologie de la Plasticité Neuronale, U1215, INSERM, F-33000 Bordeaux, France;
Université de Bordeaux, F-33000 Bordeaux, France;

Notes

3
To whom correspondence should be addressed. Email: [email protected].
Author contributions: N.E., K.B., M.J., and A.M. designed research; A.S., A.S.A., L.B.-B., S.V., M.O., L.P., V.L., and M.P. performed research; A.S., A.S.A., L.B.-B., N.E., S.V., C.H., and N.M. contributed new reagents/analytic tools; A.S., A.S.A., L.B.-B., N.E., and A.M. analyzed data; and N.M., P.-V.P., H.E., and A.M. wrote the paper.
1
A.S., A.S.A., L.B.-B., N.E., and S.V. contributed equally to this work.

Competing Interests

The authors declare no conflict of interest.

Metrics & Citations

Metrics

Note: The article usage is presented with a three- to four-day delay and will update daily once available. Due to ths delay, usage data will not appear immediately following publication. Citation information is sourced from Crossref Cited-by service.


Citation statements




Altmetrics

Citations

If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.

Cited by

    Loading...

    View Options

    View options

    PDF format

    Download this article as a PDF file

    DOWNLOAD PDF

    Get Access

    Login options

    Check if you have access through your login credentials or your institution to get full access on this article.

    Personal login Institutional Login

    Recommend to a librarian

    Recommend PNAS to a Librarian

    Purchase options

    Purchase this article to access the full text.

    Single Article Purchase

    Temporal binding function of dorsal CA1 is critical for declarative memory formation
    Proceedings of the National Academy of Sciences
    • Vol. 114
    • No. 38
    • pp. 9991-E8130

    Media

    Figures

    Tables

    Other

    Share

    Share

    Share article link

    Share on social media