Precise timing of ERK phosphorylation/dephosphorylation determines the outcome of trial repetition during long-term memory formation

Significance

Repetition is a fundamental feature of memory induction. According to the prevailing view, during memory induction, individual stimuli produce only short-lived, subthreshold effects in neurons, which, upon repetition, summate, build up, and eventually result in long-term memory. In our work, we present evidence that this view is incomplete: Rather than producing identical subthreshold effects, training trials can have distinct functional roles, such as “priming” and “confirmation,” depending on their position in a sequence. During the induction of long-term synaptic plasticity in Aplysia, strong stimuli can play both roles, whereas weak stimuli can only be priming but not confirming. We propose that this represents a mechanism to encode not only repetition but also escalation of potentially threatening stimuli.

Abstract

Two-trial learning in Aplysia reveals nonlinear interactions between training trials: A single trial has no effect, but two precisely spaced trials induce long-term memory. Extracellularly regulated kinase (ERK) activity is essential for intertrial interactions, but the mechanism remains unresolved. A combination of immunochemical and optogenetic tools reveals unexpected complexity of ERK signaling during the induction of long-term synaptic facilitation by two spaced pulses of serotonin (5-hydroxytryptamine, 5HT). Specifically, dual ERK phosphorylation at its activating TxY motif is accompanied by dephosphorylation at the pT position, leading to a buildup of inactive, singly phosphorylated pY-ERK. Phosphorylation and dephosphorylation occur concurrently but scale differently with varying 5HT concentrations, predicting that mixed two-trial protocols involving both “strong” and “weak” 5HT pulses should be sensitive to the precise order and timing of trials. Indeed, long-term synaptic facilitation is induced only when weak pulses precede strong, not vice versa. This may represent a physiological mechanism to prioritize memory of escalating threats.

Repeated-trial learning is a fundamental form of memory acquisition exhibited by virtually all animals, including humans, and even by nonneural organisms such as bacteria and plants (1). Effects of training trials are generally taken to be additive, with each successive trial taking the system further away from a homeostatic baseline, eventually resulting in long-term molecular, cellular, and behavioral changes. However, a simplified, minimalistic two-trial paradigm for inducing long-term memory (LTM) in Aplysia exposes the importance of nonadditive interactions between training events: In this system, a single trial does not result in LTM, whereas two trials separated by 45 min produce robust LTM (2–4). The molecular pathways that account for this nonlinearity remain to be identified.

One candidate factor that could mediate a nonlinear transition to LTM is extracellularly regulated kinase (ERK, p42/44 mitogen-activated protein kinase [MAPK]), a key cellular enzyme responsible for many critical state transitions in virtually all eukaryotic cells (5, 6). In Aplysia, ERK is required for LTM and its cellular substrate, long-term synaptic facilitation (LTF). ERK is also essential for other forms of long-term synaptic plasticity in a wide range of model systems (7). Sustained phosphorylation of ERK at its activating TxY motif (Fig. 1B) is one of the most common biomarkers of long-term plasticity and memory (2, 8–13), including even immunological memory (14). In the context of repeated-trial learning, it is generally believed that isolated stimuli that do not induce LTM produce only transient activation of neuronal ERK, whereas repeated trains of stimuli that do induce LTM result in ERK’s sustained activation (15–17). In Aplysia, a single training event (such as a tail shock, for the induction of sensitization) leads to ERK phosphorylation that peaks 45 min later, followed by a rapid decay. But if the second training event occurs at 45 min, ERK phosphorylation is sustained over the course of at least several hours (3, 4, 18). Thus, the induction of ERK phosphorylation correlates with the induction of LTF/LTM, and the elevation of ERK activity above baseline has been proposed as the cause of downstream long-term molecular changes (4, 19, 20). This interpretation suggests that multiple trials are required, at least in part, to build up sufficient levels of ERK activity.

Fig. 1.

Distinct P-ERK antibodies reveal differential presynaptic responses to 5HT. (A) Overview of the two-trial training paradigm in Aplysia. (A, Top) A tactile stimulus delivered to the animal’s tail causes tail withdrawal mediated by a predominantly monosynaptic reflex arc (SN–MN). Training stimuli (electric shocks) produce a global release of 5HT throughout the Aplysia CNS, causing sensitization of the reflex. These responses can be modeled using direct 5HT applications to SN–MN cocultures. Schematic diagram of an MN paired with three SNs represents the image from C. (A, Bottom) A single training trial (or application of 5HT) does not induce LTM or LTF and causes only transient ERK phosphorylation in ganglia. Two trials (or 5HT applications) spaced 45 min apart induce a sustained elevation of P-ERK, as well as LTM and LTF lasting >24 h. (B) Activating TxY motif of MAPKs (Left) and the principal P-ERK antibodies used in this study (Right). panP-ERK (CST, 9101) recognizes all forms of ERK phosphorylated at the TxY motif, whereas dualP-ERK (Sigma, M8159) is selective for dually phosphorylated, and thus active, ERK. (C) Immunofluorescence in fixed SN–MN cocultures reveals distinct localization of panP-ERK and dualP-ERK signals. Note that the z plane is below the nuclei to highlight neurites. (Scale bars, 50 µm.) (D) Changes in panP-ERK and dualP-ERK in presynaptic SNs in response to 5HT treatment. (D, Top) Overview of the treatment protocols. (D, Bottom) Immunofluorescence targeting panP-ERK and dualP-ERK and in MN-paired SNs. The overlay includes DAPI signal to indicate the location of the nucleus. (Scale bars, 10 µm.) (E) Quantification of changes in panP-ERK and dualP-ERK densities, and the dual:panP-ERK density ratio in the SN nucleus (n = 16 to 20). (F) Quantification of changes in panP-ERK and dualP-ERK densities, and the dual:panP-ERK density ratio in the SN cell-body cytosol (n = 16 to 20). Asterisks indicate significance in Dunnett’s post hoc tests following one-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. All data are shown as means ± SEM.

In the current study, we have employed tools for inducing and examining ERK phosphorylation to directly ask whether this interpretation is correct, and have discovered additional complexity of ERK signaling during LTF induction. ERK, as all MAPKs, is activated not by single but by double phosphorylation of the TxY motif, involving a mechanistically unique (21), yet extremely well conserved (22, 23) modification of both threonine and tyrosine performed by a single enzyme, MAPK/ERK kinase (MEK) (24–27). Importantly, neither of the singly phosphorylated forms of ERK (pY-ERK or pT-ERK) possesses enzymatic activity. We now show that during LTF induction, much of the previously observed surge of phospho-ERK (P-ERK) is accounted for by singly phosphorylated, and therefore inactive pY-ERK, resulting from concurrent dephosphorylation of ERK at the threonine position. By contrast, levels of dually phosphorylated, enzymatically active ERK are reduced below baseline if only a single training trial is administered, but they remain relatively unchanged after two spaced trials. This suggests the surprising conclusion that two spaced training trials are in fact required to maintain ERK activity at a steady level for a sufficient period of time (despite concurrent dephosphorylation), rather than to elevate ERK activity above the baseline. Several functional predictions about LTF induction arise from this conclusion. In this paper, we report that these predictions are borne out both by electrophysiological and by optogenetic experiments. Collectively, our results provide insights into the molecular processing mediated by ERK that is critical for encoding the number, timing, and order of stimuli in long-term memory formation.

Results

DualP-ERK Does Not Account for the Surge of P-ERK during LTF Induction.

Two-trial LTM in Aplysia can be studied on multiple levels (Fig. 1A). At the behavioral level, two training shocks delivered to the animal’s tail result in LTM for sensitization of the tail withdrawal reflex (elicited by a mild tactile stimulus to the tail). At the level of the central nervous system (CNS), each shock results in global release of the neuromodulator serotonin (5-hydroxytryptamine, 5HT). 5HT, in turn, causes facilitation in central neurons, including a monosynaptic circuit consisting of sensory neurons (SNs) and motor neurons (MNs) (28, 29). This synaptic facilitation can be further studied by directly applying 5HT to reconstituted microcircuits consisting of cultured SNs paired with MNs. At all levels of analysis, a single shock/pulse of 5HT does not result in LTM/LTF, whereas two spaced shocks/pulses induce LTM/LTF (2–4, 30). We have previously shown that all these forms of LTM/LTF induction correlate with the dynamics of P-ERK induction, but these studies have not resolved the contribution of individual forms of P-ERK (2–4).

Thus, in the present paper, we employed the following strategy for immunochemical analysis of P-ERK during LTF induction. Our method relies on dual staining in SN–MN cocultures, fixed at various time points, using a combination of 1) a monoclonal dualP-ERK–specific antibody and 2) a polyclonal antibody with broader P-ERK specificity, which we term panP-ERK (Fig. 1B). SN–MN cocultures were treated with either a single 5-min pulse of 5HT, or two pulses separated by 45 min (Fig. 1E). ERK phosphorylation was measured immunochemically 45 min after the first pulse [a time point when a transient surge in ERK phosphorylation is expected (3)], and 90 min after the second pulse [135 min; a time point when sustained ERK phosphorylation is expected after two but not one pulse (2–4, 30)]. Indeed, panP-ERK showed the expected pattern (Fig. 1 D–F): Signal density was elevated in both the nucleus and cytosol 45 min after a single pulse of 5HT but, unless the second trial was administered, it decayed by the late (135 min) time point (here and throughout Results, see Dataset S1 for descriptive statistics). Interestingly, this was not the pattern observed for dualP-ERK: Its signal did not increase significantly above baseline at any of the time points measured after either one or two pulses of 5HT. Instead, a significant decrease in the cytosolic level of dualP-ERK was observed 45 min after the first pulse, which was then restored by the second pulse. In all cases, 5HT treatment resulted in a robust decrease in the dual:panP-ERK ratio (Fig. 1 E and F), an internally controlled metric representing the proportion of active, dually phosphorylated ERK within the overall pool of P-ERK. Thus, the nonlinear induction of LTF by the two-trial paradigm correlates with overall P-ERK dynamics (represented by panP-ERK) but, surprisingly, not with the dynamics of dualP-ERK.

DualP-ERK Induction Is Detectable and Is Required for LTF.

A possible explanation of our results is that the antibody used to detect dualP-ERK failed to capture its induction. Perhaps, the dualP-ERK antibody easily saturates with its antigen, or fails to recognize Aplysia ERK, making the effects observed for panP-ERK more representative of ERK activation. While both signals are reduced in response to a treatment by the MEK inhibitor U0126 (SI Appendix, Fig. S5), it is important to determine that the dualP-ERK antibody can detect an increase in the signal. To establish this, ERK must be selectively activated independent of the other targets of 5HT. We therefore microinjected SNs with a plasmid encoding photosensitive MEK (psMEK), a construct in which constitutively active MEK is caged within a dimer of the protein Dronpa, which dissociates in response to 502-nm light, uncaging MEK activity (31) (Fig. 2A). When cells expressing psMEK were irradiated with 502-nm light for 1 or 24 h, ERK phosphorylation was elevated significantly in psMEK-expressing cells compared with cells expressing the construct but not exposed to light (Fig. 2 B and C). Notably, both panP-ERK and dualP-ERK signals were strongly elevated to a comparable extent. Both signals were correlated with the level of expression of psMEK, confirming that both antibodies are able to capture a strong increase in ERK phosphorylation (SI Appendix, Fig. S1). These results support the conclusion that the lack of induction of dualP-ERK, and the disproportionate induction of panP-ERK in response to 5HT, accurately reflects the status of ERK phosphorylation in the SNs.

Fig. 2.

Complete ERK phosphorylation is detectable and necessary for LTF. (A) Isolated SNs were microinjected with a DNA construct encoding constitutively active MEK caged within a Dronpa dimer (psMEK). Irradiation with 502-nm light uncages psMEK, causing ERK phosphorylation. (B) Changes in panP-ERK and dualP-ERK levels in psMEK-expressing cells in response to 1- or 24-h irradiation with 502-nm light. (B, Left) Dronpa autofluorescence. (Scale bars, 10 µM.) (C) Quantification of changes in panP-ERK and dualP-ERK in the nucleus and cytosol of psMEK-expressing cells following light irradiation (n = 10 to 21). (D) LTF was induced in SN–MN cocultures using two spaced applications of 5HT. The MEK inhibitor U0126 was applied during either the first or the second pulse of 5HT. SN–MN excitatory postsynaptic potentials (EPSPs) were recorded intracellularly in MNs before (pretest) and 16 to 24 h after the treatment (posttest). (E) Changes in EPSP amplitudes before and after 5HT treatment with or without U0126 (n = 7 to 22). (Scale bars, 50 ms/5 mV.) (F) Quantification of changes. Asterisks indicate significance in Dunnett’s post hoc tests following one-way ANOVA. **P < 0.01, ***P < 0.001, ****P < 0.0001. All data are shown as means ± SEM.

We next asked whether ongoing phosphorylation of ERK was required for two-trial LTF. To that end, LTF was induced in SN–MN cocultures by applying two spaced pulses of 5HT, while the MEK inhibitor U0126 was added to culture media during either the first or the second pulse (Fig. 2D). In both cases, induction of LTF was completely blocked (Fig. 2 E and F). Since U0126 is an inhibitor of MEK, rather than ERK itself, this result indicates that not only ERK activity but active phosphorylation of ERK must occur throughout the duration of training to ultimately induce LTF. Since we have previously confirmed that dualP-ERK levels remain close to baseline during LTF induction (Fig. 1 E and F), we conclude that active phosphorylation of ERK during LTF induction is required to maintain dualP-ERK at this baseline level.

High (but Not Low) Concentrations of 5HT Induce an Excess of pY-ERK over DualP-ERK.

If dualP-ERK remains close to baseline during LTF induction, what accounts for the dramatic increase in panP-ERK in response to 5HT? The two possibilities are increases in either pY-ERK or pT-ERK. To distinguish between these singly phosphorylated forms of ERK, we employed two other monoclonal antibodies that specifically recognize them (32) (Fig. 3A). We observed that pY-ERK levels were significantly increased in SNs following a 5HT pulse (50 µM), whereas pT-ERK levels were not (Fig. 3 B and C). pY-ERK staining was also spatially similar to panP-ERK staining: Both panP-ERK and pY-ERK signals displayed mostly diffuse cytoplasmic distribution, in contrast to dualP-ERK and pT-ERK, which were predominantly located near the cell membrane. Taken together, these findings indicate that the elevated panP-ERK signal observed in response to a training pulse represents an accumulation of predominantly pY-ERK.

Fig. 3.

pY-ERK, but not pT-ERK, increases after 5HT treatment. (A) singleP-ERK–specific antibodies used in this experiment selectively recognize either pY-ERK or pT-ERK. (B) Subcellular distribution and changes in pY-ERK and pT-ERK in isolated SNs treated with a single pulse of 5HT at 50 µM. (Scale bars, 10 µm.) (C) Quantification of changes in nuclear pY-ERK and pT-ERK in response to 5HT (n = 23 to 41). The asterisk indicates significance in a two-tailed unpaired Student’s t test. *P < 0.05. All data are shown as means ± SEM.

Next, we investigated how the short-term dynamics of dualP-ERK and panP-ERK scale in proportion to 5HT concentration, namely the “strength” of the training pulse. For these experiments, isolated SNs, rather than SN–MN cocultures, were used to increase throughput. The responses of isolated SNs to 5HT differed slightly from SNs paired with MNs. Specifically, we observed a modest induction of dualP-ERK in response to a standard “strong” pulse, which was not observed in SN–MN cocultures (Fig. 4 A and D). The surge in panP-ERK was, however, much greater proportionally—thus, the dual:panP-ERK ratio rapidly decreased, as previously observed in SN–MN cocultures (Fig. 4 A, D, and G). Interestingly, an inverse effect was briefly seen at very early time points (15 to 30 s): an increase in dual:panP-ERK ratio above baseline before its eventual drop to negative values (SI Appendix, Fig. S2).

Fig. 4.

Forms of P-ERK show distinct 5HT concentration dependence. (A) Isolated SNs were treated with one or two pulses of 5HT at 50 µM and fixed with ice-cold formaldehyde at various time points, and their panP-ERK and dualP-ERK content was analyzed by immunofluorescence. (Scale bars, 10 µm.) (B) Same for 5HT at 10 µM. (C) Same for 5HT at 2 µM. (D) Quantification of nuclear (Left) and cytosolic (Right) changes in panP-ERK and dualP-ERK densities in cell bodies of SNs treated with 5HT at 50 µM (n = 25 to 82). (E) Same for 5HT at 10 µM (n = 45 to 55). (F) Same for 5HT at 2 µM (n = 34 to 47). (G) Quantification of nuclear (blue) and cytosolic (black) dual:panP-ERK ratios in cell bodies of SNs treated with 5HT at 50 µM. (H) Same for 5HT at 10 µM. (I) Same for 5HT at 2 µM. Asterisks at time points indicate significant deviation from baseline in single-sample t tests. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. All data are shown as means ± SEM.

When a lower concentration of 5HT (10 µM) was used, the effect was more clearly biphasic: The initial increase in dual:panP-ERK ratio above baseline was more pronounced and slower, peaking at ∼5 min, and then reducing below the baseline by 15 to 30 min and remaining low (Fig. 4 B, E, and H). When an even lower concentration (2 µM), “weak” pulse of 5HT was used, the first phase became the predominant one: The dual:panP-ERK ratio remained above the baseline for at least 45 min (Fig. 4 C, F, and I).

Thus, although a weak pulse produces less overall ERK phosphorylation (panP-ERK), it also produces an excess of dualP-ERK over pY-ERK, whereas strong pulses produce an excess of pY-ERK over dualP-ERK. This suggests, counterintuitively, that weak pulses might in fact be more effective at generating active, dualP-ERK over time.

Bidirectional Effects of 5HT Concentrations Depend on PP2A/PP1 and PKA.

To interpret the above results, it is essential to resolve the mechanism: What accounts for the induction of pY-ERK concurrent with dualP-ERK? There are two possibilities: 1) nonprocessive phosphorylation, or 2) partial dephosphorylation. The first involves increased production of pY-ERK directly from unphosphorylated ERK as a result of the first step in a distributed, two-collision reaction mechanism between ERK and MEK. The second possibility involves increased production of dualP-ERK coupled with its increased dephosphorylation at the threonine position. To rule out the first possibility, we performed experiments in which we manipulated and monitored levels of total ERK, since an increase in the quantity of unphosphorylated ERK (for example, by subcellular redistribution) could in principle “dilute” the substrate pool and therefore increase the probability of the first phosphorylation step (unphosphorylated ERK→pY-ERK) at the expense of the second step (pY-ERK→dualP-ERK). The results indicate that overexpressing ERK can indeed increase the quantity of pY-ERK (SI Appendix, Fig. S3). However, changes in total ERK quantity cannot account for the rapid, bidirectional changes in dual:panP-ERK ratio seen with strong and weak 5HT pulses, since we observed no difference in how these treatments affect the levels of total ERK at such timescales (SI Appendix, Fig. S4).

Thus, the second explanation is more likely: 5HT induces not only phosphorylation but also partial dephosphorylation of ERK at the threonine position. Importantly, MAPK phosphatases such as PP2A (33–35) can be close to saturation under physiological conditions (36), suggesting the potential for positive regulation. To examine whether 5HT indeed induces ERK dephosphorylation, we asked if phosphatase inhibition could prevent a decrease in dual:panP-ERK ratio in response to 5HT (Fig. 5A). Indeed, okadaic acid (OA), the inhibitor of PP2A and PP1, blocked the reduction in dual:panP-ERK ratio 45 min after the strong 5HT pulse. Importantly, OA did not affect ERK phosphorylation on its own (Fig. 5B). These findings suggest that increased dephosphorylation of ERK at the threonine position, likely by induction of PP2A activity, accounts for the buildup of pY-ERK in response to 5HT. Inhibition of calcineurin/PP2B, another Ser/Thr phosphatase involved in limiting the effects of training trials (37), did not have the same effect (SI Appendix, Fig. S6).

Fig. 5.

(A) Hypothetical model of intertrial interactions between low- and high-concentration pulses of 5HT. 5HT at 2 µM is sufficient to activate ERK phosphorylation, but not sufficient to activate PKA, which induces pT dephosphorylation. A high-concentration (“strong”) pulse of 5HT induces more ERK phosphorylation, but also disproportionately activates PKA and its downstream Ser/Thr phosphatase, leading to a reduction in dual:panP-ERK ratio. (B₁) SNs were treated with the phosphatase inhibitor OA or the PKA inhibitor H89 with or without a single pulse of 5HT (50 µM). (B₂) PanP-ERK and dualP-ERK levels in SNs treated as in B₁. (Scale bars, 10 µm.) (B₃) Quantification of changes in B2 (n = 33–40). (C₁) SNs were treated with the adenylate cyclase activator forskolin with or without a single pulse of 5HT (2 µM). (C₂) PanP-ERK and dualP-ERK levels in SNs treated as in C₁. (Scale bars, 10 µm.) (C₃) quantification of changes in C₂ (n = 42–50). Asterisks above bars indicate significant deviation from baseline in a single-sample t test. Asterisks between bars indicate significance in Dunnett’s post hoc tests following one-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. All data are shown as means ± SEM.

A well-described mechanism for inducing PP2A involves the cyclic adenosine monophosphate (cAMP)–dependent protein kinase (PKA) (38, 39), itself a well-known target of 5HT (40, 41). Indeed, a treatment with the PKA inhibitor H89 prevented the reduction in dual:panP-ERK ratio similar to OA (Fig. 5B). Thus, PKA activity leads to increased dephosphorylation of ERK, and the balance of PKA and ERK activation must therefore determine the changes in dual:panP-ERK ratio. If this notion is correct, it would lead to the prediction that elevating cAMP levels in combination with a weak pulse of 5HT should mimic the effects of a strong pulse, causing a decrease, rather than increase, in the dual:panP-ERK ratio. We therefore paired a weak pulse of 5HT with forskolin, the activator of adenylyl cyclase and hence PKA (Fig. 5 A and C). While the weak 5HT treatment by itself increased the dual:panP-ERK ratio as previously observed, this effect was reversed by the addition of forskolin: Instead of an increase, the dual:panP-ERK ratio was reduced below the baseline, similar to the effect of the strong 5HT pulse. Taken together, these experiments confirm that PKA plays a key role in shaping the dynamics of ERK phosphorylation.

In summary, during LTF induction, increased phosphorylation of ERK is opposed by its increased dephosphorylation at the pT position. These two processes scale differently with 5HT concentration: A weak pulse induces less overall phosphorylation, but it also induces disproportionally less dephosphorylation than the strong pulse, as revealed by their opposing effects on the dual:panP-ERK ratio.

ERK Phosphorylation and Dephosphorylation Occur Concurrently.

We next asked whether ERK phosphorylation and dephosphorylation occur concurrently during LTF induction, or whether the two processes are separated in time. To that end, the inhibitor U0126 was added to the culture medium either 20 min before or 20 min after a strong 5HT pulse (SI Appendix, Fig. S5A). When no inhibitor was present, 20 min after the pulse the dual:panP-ERK ratio was already significantly reduced, and this reduction remained stable at the 45-min time point (Fig. 4G). The addition of U0126 at the 20-min time point, however, reduced both panP-ERK and dualP-ERK levels observed at 45 min similar to the addition of the inhibitor prior to the 5HT pulse (SI Appendix, Fig. S5 B and C). This indicates that both phosphorylation and dephosphorylation are ongoing even between 20 and 45 min, when the equilibrium between the two processes is already established, and the ratio of dual:panP-ERK is stabilized. Thus, a 5HT pulse induces a dynamic “tug of war” between phosphorylation and dephosphorylation of ERK, with the balance of forces determined by 5HT concentration.

A Weak Trial Must Precede a Strong Trial to Induce LTF.

Several surprising predictions arise from the results described so far. First, since the concentration of 5HT determines both the extent and the balance of ERK phosphorylation and dephosphorylation, mixed weak and strong protocols should have distinct effects on LTF induction depending on the order of trials. If the weak trial follows the strong trial, the phosphorylation of ERK induced by the weak trial should have little effect on dephosphorylation already induced by the preceding strong trial. If, however, the weak trial precedes the strong trial, then a modest increase in ERK phosphorylation caused by the initial weak trial should have a larger effect because of less opposition by phosphatases. So a “weak–strong” protocol should be more likely to induce LTF than a “strong–weak” protocol.

To directly test this prediction, we investigated LTF induction in SN–MN cocultures using the following treatment protocols: 1) “weak–weak” (Fig. 6A₁), 2) “weak–strong” (Fig. 6A₂), 3) “strong–weak” (Fig. 6A₃), and 4) “strong–strong,” namely the default two-trial protocol (Fig. 6A₄). The weak–weak and strong–strong protocols showed expected effects: No LTF was induced by the weak–weak treatment, whereas the standard strong–strong treatment resulted in LTF. Strikingly, the weak–strong paradigm also produced robust LTF. In fact, the weak–strong paradigm consistently induced greater LTF than the strong–strong paradigm (SI Appendix, Fig. S7), particularly when the intertrial interval equaled 45 min, corresponding to optimal trial spacing during LTM formation (4). It is difficult to explain such an effect without concluding that a strong pulse must activate some LTF-suppressive factor (such as a phosphatase) disproportionately from the weak pulse (Fig. 5B).

Fig. 6.

Mixed LTF protocols are sensitive to trial order. (A) LTF was induced in SN–MN cocultures using two spaced applications of 5HT (intertrial interval [ITI] 45 min) at 2 or 50 µM in various combinations. (A, Top Left) Overview of the paradigms used. (A, Top Right) EPSPs were recorded before (pre) and 16 to 24 h after the treatment (post). (A, Bottom) Quantification of changes in EPSP amplitudes (n = 11 to 16). Asterisks above bars indicate significant deviation from baseline in a single-sample t test. Asterisks between bars indicate significance in Dunnett’s post hoc tests following one-way ANOVA. (B) Same as A, except LTF was induced using five spaced applications of 5HT (ITI 15 min) at various concentrations (n = 14 to 18). Asterisks above bars indicate significant deviation from baseline in a single-sample t test. Asterisks between bars indicate significance in a two-tailed unpaired Student’s t test. (C, Top Left) Overview of treatment protocols. LTF was induced in SN–MN cocultures using the weak–strong paradigm. PKA was activated using a pulse containing forskolin and the phosphodiesterase inhibitor IBMX, applied 45 min before or 45 min after the first weak pulse of 5HT. (C, Right) EPSPs were recorded before (pre) and 16 to 24 h after the treatment (post). (C, Bottom) Quantification of changes (n = 15 to 17). (D, Top Left) Overview of treatment protocols. ERK phosphorylation was activated in psMEK-expressing presynaptic SNs by 1-h irradiation with 502-nm light either alone or in combination with a single 5HT pulse (50 µM) delivered immediately before or immediately after irradiation. (D, Top Right) EPSPs were recorded before (pre) and 16 to 24 h after the treatment (post). (D, Bottom) Quantification of changes (n = 15 to 36). (Scale bars, 50 ms/5 mV.) Asterisks in C and D indicate significance in two-tailed unpaired Student’s t tests. *P < 0.05, **P < 0.01, ***P < 0.001. All data are shown as means ± SEM.

Most intriguing, however, is the observation that the inverse, strong–weak paradigm induced almost no LTF (Fig. 6A₃). This strong–weak sequence of trials was no better at inducing LTF than the weak–weak paradigm (Fig. 6A₁). Such a prominent difference between the weak–strong and strong–weak protocols clearly underscores the nonadditive nature of trial interactions and confirms the prediction stemming from our analysis of P-ERK dynamics (Fig. 4).

To test the generality of this observation, we examined the effects of various combinations of weak/strong trials in a paradigm consisting of five trials separated by 15 min (Fig. 6B), often used in studies examining LTF and LTM (42–45). SN–MN cocultures received one of the following treatments: 1) four weak pulses followed by one strong (Fig. 6B₁); 2) five weak pulses (Fig. 6B₂); 3) one strong pulse followed by four weak (Fig. 6B₃); and a single strong pulse (Fig. 6B₄). Again, the weak–strong paradigm (Fig. 6B₁) elicited significant LTF, whereas the strong–weak paradigm (Fig. 6B₃) failed to do so. As expected, the “all-weak” paradigm failed to induce LTF (Fig. 6B₂), as did the single strong trial by itself (Fig. 6B₄). Thus, our prediction of the differential effect of trial order on LTF generalizes across a range of treatment protocols.

PKA and ERK Differentially Affect LTF Depending on Their Relative Time of Activation.

The second prediction stemming from our analysis of ERK phosphorylation and dephosphorylation is that activating the MEK–ERK and the cAMP–PKA cascades during the LTF induction protocol should have differential effect depending on precise timing. Both cascades are required for LTF, but our results show that they oppose each other in the P-ERK tug of war. Early activation of PKA would shorten the time of ERK activation, and so should be less likely to induce LTF than late activation of PKA, whereas early activation of ERK should be more likely to induce LTF.

The weak–strong LTF induction paradigm provides an ideal system to test these predictions. Because the success of LTF induction by this protocol must rely on early ERK phosphorylation unopposed by phosphatase activation, elevating cAMP levels prior to the weak trial should have an LTF-suppressive effect, whereas activating it during or after the strong trial should not have such an effect. To test this prediction, we treated SN–MN cocultures with a combination of forskolin and the phosphodiesterase inhibitor IBMX, either 45 min before (Fig. 6C₂) or 45 min after the weak trial, namely during the second, strong trial (Fig. 6C₃). Activation of PKA during the latter strong trial had no effect on subsequent LTF (Fig. 6 C₃ and C₄). However, as predicted, preactivation of PKA before the weak trial had an LTF-suppressive effect (Fig. 6C₂). The cAMP–PKA axis of signaling can therefore oppose LTF induction at specific, critical time intervals.

Pharmacological manipulation of ERK is currently limited, since no manipulable dedicated second messengers exist for ERK. Thus, its activity cannot be isolated from the 5HT signaling network as easily as the activity of PKA. However, optogenetic activation of ERK using the psMEK construct (Fig. 2) provides an opportunity to target it specifically, at a precise time, and independent of 5HT (Fig. 6D). To test our final prediction that ERK activation should have a stronger LTF-inducing effect early in the treatment protocol, we expressed psMEK in SNs cocultured with MNs. We then irradiated the cultures for 1 h with activating light. This treatment alone did not result in LTF, indicating that ERK phosphorylation in SNs is not by itself sufficient to induce LTF. However, when light irradiation was followed by a strong 5HT pulse, LTF was readily induced (Fig. 6D₂). These data show that ERK activation, isolated from other effects of 5HT, is a sufficient substitute for the first trial during two-trial LTF induction. Strikingly, when the strong trial was followed by the same light irradiation, LTF was not induced—in other words, ERK activation is not sufficient as a substitute for the second trial, just as the weak trial is insufficient as a second trial in the strong–weak protocol (Fig. 6A₃).

In summary, we used a combination of immunochemical and optogenetic tools to reveal unexpected complexity in the dynamics of ERK signaling during LTF induction. These cell biological findings led us to confirm several predictions regarding the two-trial training protocol: namely, its unexpected sensitivity to trial order, and the critical importance of precisely timed signaling by the cAMP–PKA and MEK–ERK cascades.

Discussion

Why does it take multiple trials to induce a long-term memory where one trial fails to do so? The prevailing view on repeated-trial learning is summative: Each trial, insufficient by itself, takes the system further and further away from homeostasis, and eventually the change becomes long-lasting (46, 47). We here show that such summation of individual subthreshold effects cannot fully account for intertrial interactions. Specifically, simple summation fails to explain the unexpected sensitivity of mixed weak and strong LTF induction protocols to the order of trials. This effect is better explained by considering training trials as having distinct functional roles, priming and confirmation. The advantage of the two-trial protocol we employ is a clear separation between the two roles; in protocols involving more trials, the roles of individual trials can be mixed.

Encoding repeated but not isolated events is advantageous because repeated events are more likely to be relevant in the future (47). Single-trial learning can in some cases result in long-term memory (48–52), but most protocols for LTM induction involve repetition. In such cases, rather than failing to cross a threshold, a single trial may be actively restricted in its ability to induce a lasting memory, unless its effects are confirmed by subsequent trials. Such confirmatory trials do not necessarily need to take the system further away from homeostasis—instead, they might act by preventing the self-limiting arrest of the processes already set in motion by the priming trial.

Results presented in the current paper are in line with the latter interpretation. ERK activity is traditionally thought to build up in neurons during repeated-trial learning. We demonstrate that in the Aplysia SN–MN circuit, ERK activity in fact stays close to baseline, whereas much of the previously observed P-ERK surge results from a dynamic tug of war between phosphorylation and dephosphorylation producing an excess of (inactive) pY-ERK. The second trial does not take ERK activity further away from baseline but rather it prevents ERK inhibition due to excess dephosphorylation. Critically, the two processes scale differently with the concentration of the agonist, 5HT, which means that weak and strong pulses of 5HT have qualitatively different effects.

A weak pulse (2 µM 5HT) increases ERK activity without a corresponding induction of dephosphorylation. The same effect is achieved by selectively activating ERK using the psMEK construct. These treatments can successfully fulfill the role of priming (whether in a two-trial protocol, five-trial protocol, or a single-trial protocol paired with optogenetic activation of psMEK). However, the priming trial must ultimately be followed by a confirmatory trial, for which only a strong pulse (50 µM 5HT) is sufficient. The reverse order (strong–weak) does not induce LTF because the weak trial, as well as optogenetic activation of psMEK, fails to act as “confirmation” (Fig. 7).

Fig. 7.

Hypothetical model for intertrial interactions. Following a single strong trial, pT dephosphorylation occurs concurrently with dual phosphorylation and ultimately outlasts it, leading to a buildup of pY/panP-ERK and the eventual reduction of dualP-ERK below baseline. A second strong trial is sufficient to restore dualP-ERK levels to close to baseline levels as the buildup of panP-ERK is sustained. If a weak trial precedes a strong trial, it creates a time window of ERK activation without a corresponding induction of dephosphorylation, sufficient to “prime” the cell for LTF. However, if the weak trial follows the strong trial, it inherits the still-elevated rates of dephosphorylation, which cancels out its weaker induction of phosphorylation and fails to induce LTF.

What could be the physiological role of such an asymmetric system? One attractive hypothesis is that it favors the formation of long-term responses to strengthening, but not weakening, stimuli. The SN–MN microcircuit recapitulates a reflex arc for defensive withdrawal, and the release of endogenous 5HT in the Aplysia CNS can be seen as a neuronal reflection of a threat response. It is reasonable to expect that defensive memory allocation should favor escalating threats over waning ones. Just as repeated events have more predictive power than isolated events, events that repeat with increasing intensity are more relevant to the future than events that decrease in effect.

Regardless of the functional interpretation, at the cell biological level, our work shows that given the complex temporal dynamics of ERK phosphorylation/dephosphorylation, it is vital to specifically identify and track both single phosphorylation and the activating dual phosphorylation of the enzyme. We and others have previously implicated ERK activity in the duration of intertrial interactions during repeated-trial training (3, 4, 53). However, the dualP-ERK antibody used in our previous studies (CST, 4370) cross-reacts with singly phosphorylated ERK, obscuring the precise nature of the signal. Indeed, the subcellular distribution of this signal in ganglia SNs resembles a mix of dualP-ERK and panP-ERK signals seen in the current study (4). Using tools that can distinguish between various phospho states of ERK, we now add an additional dimension to the interpretation of these early results. For example, we have demonstrated that physiological outcomes can result not only from the induction of ERK activity but from its reduction below the homeostatic baseline or maintenance at a steady level. This finding probably applies to various systems beyond memory and learning. In addition, we have provided evidence that singly phosphorylated ERK, while inactive, is dynamically regulated, and plays a major part in the overall dynamics of ERK phosphorylation. This conclusion also has far-reaching implications, since phosphorylation of ERK and other MAP kinases is a widely used physiological indicator across biology.

It remains an open question whether the pool of singleP-ERK induced in response to 5HT has a functional role. An intriguing possibility is that a buildup of this form of ERK may increase the rate at which dualP-ERK is subsequently generated in response to ongoing events. Phosphorylation of ERK (and other MAPKs) has been shown to be, at least in some cases, distributed: It requires two separate, independent collisions between the substrate (ERK) and the enzyme (MEK) (54–56). pY-ERK is an intermediate product in this two-step process, and its accumulation may therefore increase the probability of the second phosphorylation step (pY-ERK→dualP-ERK) (57).

In conclusion, the results presented here demonstrate a causal role for ERK in nonadditive intertrial interactions, reinterpret the role of ERK induction in repeated-trial learning, and critically inform the future use of P-ERK levels as a physiological biomarker. On a more general level, the results update a simple additive model for intertrial interactions (58–60) and demonstrate that individual training trials can have distinct functions in the integration of salient information during learning.

Materials and Methods

Culturing of Aplysia neurons, 5HT treatments, electrophysiology, DNA microinjection, and immunofluorescence staining were carried out according to previously published protocols (18, 61, 62), with minor modifications. Key primary ERK-specific antibodies employed in this study include panP-ERK (CST, 9101), dualP-ERK (Sigma, M8159), total ERK (CST, 4696), pT-ERK (Sigma, M7802), and pY-ERK (Sigma, M3682). All data are shown as means ± SEM. Full details of experimental procedures and statistical analyses are provided in SI Appendix.

Data, Materials, and Software Availability

All study data are included in the article and/or supporting information.