Pattern detection in the TGFβ cascade controls the induction of long-term synaptic plasticity

Significance

Stimulus repetition in the environment leads to effective long-term memory (LTM) formation in the brain, but the cellular and molecular mechanisms underlying this phenomenon are not well understood. We find that temporal patterning within the transforming growth factor β (TGFβ) signaling system provides a unique sensor for stimulus repetition, via proteolytic activation of the growth factor proligand, that enables the induction of long-term synaptic plasticity critical for LTM. We propose that the timing of TGFβ signaling during learning enables interaction between two separate training events, ultimately resulting in LTM.

Abstract

Transforming growth factor β (TGFβ) is required for long-term memory (LTM) for sensitization in Aplysia. When LTM is induced using a two-trial training protocol, TGFβ inhibition only blocks LTM when administrated at the second, not the first trial. Here, we show that TGFβ acts as a “repetition detector” during the induction of two-trial LTM. Secretion of the biologically inert TGFβ proligand must coincide with its proteolytic activation by the Bone morphogenetic protein-1 (BMP-1/Tolloid) metalloprotease, which occurs specifically during trial two of our two-trial training paradigm. This paradigm establishes long-term synaptic facilitation (LTF), the cellular correlate of LTM. BMP-1 application paired with a single serotonin (5HT) pulse induced LTF, whereas neither a single 5HT pulse nor BMP-1 alone effectively did so. On the other hand, inhibition of endogenous BMP-1 activity blocked the induction of two-trial LTF. These results suggest a unique role for TGFβ in the interaction of repeated trials: during learning, repeated stimuli engage separate steps of the TGFβ cascade that together are necessary for the induction of long-lasting memories.

Memories form through specific temporal interactions between stimuli, such as the repetition of a stimulus, the coincidence of multiple stimuli, and the coincidence of a stimulus and an internal state (1). In order to detect and respond to such time-based patterns, the brain must possess molecular subsystems that operate with temporal specificity.

One superfamily of molecules that could potentially act as molecular time keepers in memory formation are growth factors (GFs) (2). GFs, which are canonically considered molecular orchestrators of development, have also been found to be critically involved in adult plasticity required for memory formation (2–10). While the role of GFs in memory induction and consolidation is well recognized, few studies explore the temporal aspects of GF signaling required for memory formation. As a result, GF signaling is often viewed as an outcome of learning and thus is considered to be downstream of the temporal computation necessary for memory formation. Here, we show that secretion and activation of a GF can itself be a critical component of the temporal choreography that leads to memory formation.

Transforming growth factor β (TGFβ) is a GF whose multistep signaling and activation is well suited to act as a molecular timekeeper in memory formation. TGFβ is a secreted pleiotropic GF involved in a wide range of cell-specific regulation throughout the animal kingdom (11–13). Some GFs such as fibroblast growth factor (FGF) are immediately bioactive upon secretion and are able to quickly induce signal transduction (14). In contrast, TGFβ is secreted from the cell in association with latency-associated peptide (LAP), which holds the TGFβ ligand in an immature, inactive form (15–17). TGFβ further associates with the latent TGFβ-binding protein (LTBP) to form the large latent complex (LLC) and is anchored to the extracellular matrix (ECM) upon secretion (15, 16). TGFβ can then be cleaved by metalloproteases, such as BMP-1 and MMP-1, which mobilizes the ligand from both the ECM and LAP, allowing for its bioactivity (18). Thus, processing of TGFβ is a multistep procedure that requires precise orchestration of both secretion and activation, in order for downstream signaling to occur (Fig. 1). This multistage feature of TGFβ activation is well-poised to mediate the temporal integration that is necessary for multitrial learning, whereby different steps of the pathway could be induced at different rates or by different trials, making the availability of active TGFβ a product of trial integration.

Fig. 1.

Multistep activation of latent TGFβ via metalloprotease BMP-1/Tolloid. Schematic illustrates the structure of latent TGFβ (Inset) as well as the release and activation of latent TGFβ through BMP-1 proteolytic activity. (1) Latent TGFβ is released from the cell and then is sequestered in the extracellular matrix. (2) TGFβ must be activated through proteolytic processing via BMP-1/Tolloid in order to free TGFβ and allow it (3) bind to its receptor.

In the present paper, we quantify and manipulate TGFβ signaling induced by two-trial training and show that it is indeed a product of intertrial interaction. We first show that subthreshold stimulation (a single pulse of 5HT), when paired with the protease BMP-1, induces LTF in sensory and motor neuron cocultures (SN+MN). We further demonstrate that treatment with human recombinant BMP-1 leads to persistent TGFβ signaling when paired with a single pulse of 5HT. Collectively, these data demonstrate a key interaction between the two training trials, which requires two temporally distinct molecular steps to engage effective TGFβ signaling and LTF.

Results

Synaptic Gain of Function in the TGFβ Signaling Cascade.

We have previously shown that GFs can operate with a striking degree of temporal and spatial specificity during LTM induction (9). In these studies, we utilized a two-trial training paradigm in which two trials are separated by a specific intertrial interval of 45 min to induce LTM for sensitization (19). This paradigm is based upon behavioral LTM training for sensitization of the tail-elicited withdrawal reflex: a defensive retraction of the tail in response to a mild mechanical stimulus (19). In Aplysia, tail shock-induced sensitization is accompanied by global release of the neuromodulator 5HT in the central nervous system (20–22). This results in synaptic facilitation of the SN+MN synapse that governs this reflex (19, 23).

Previous studies found that TGFβ inhibitors only block LTM during trial two of our two-trial training paradigm, suggesting that the activity of TGFβ may be involved in trial integration. One step in TGFβ’s cascade that could be the source of such integration is the cleavage of latent TGFβ by a metalloprotease (Fig. 1). Latent TGFβ could be produced and secreted in response to the first trial but proteolytically activated only after trial two, which would then allow LTM formation.

To test this hypothesis, we examined the induction of two-trial LTF in SN+MN cocultures, a cellular correlate of LTM. Neurons were treated with either one pulse of 5HT, which alone is ineffective in inducing LTF, or two spaced pulses of 5HT, which are highly effective in producing LTF (Fig. 2 A and B). Following treatment, cells were incubated with or without recombinant human BMP-1 for 1 h. EPSP amplitudes were measured before and 20 to 24 h after the treatments, and changes were expressed as log10 of the post:pre ratio. Two pulses of 5HT (5 min each trial, ITI = 45 min) produced significant LTF, while 1 pulse alone did not produce significant facilitation (Fig. 2 B and C, 2 × 5HT: 0.29 ± 0.08, n = 19; 1 × 5HT: −0.03 ± 0.04, n = 20; 1 × 5HT + BMP-1: 0.16 ± 0.02, two data points removed as outliers, n = 32, BMP-1: 0.02 ± 0.03, n = 26, ANOVA: F = 10.6, ****P < 0.0001, Tukey’s multiple comparison results: 2 × 5HT vs. 1 × 5HT: mean difference = −0.32, 95% CI = −0.49 to −0.15, ****P < 0.0001). Treatment of SN+MN pairs with BMP-1 and a single 5HT pulse, however, resulted in significant synaptic facilitation compared to a single pulse alone (Fig. 2 B and C, 1 × 5HT + BMP-1 vs. 1 × 5HT: mean difference = −0.19, 95% CI = −0.34 to −0.04, **P = 0.006, 1 × 5HT + BMP-1 vs. 2 × 5HT: mean difference = 0.13, 95% CI = −0.03 to 0.3, P = 0.14, 1 × 5HT + BMP-1 vs. BMP-1: mean difference = 0.14, 95% CI = 0.004 to 0.3, *P = 0.04). Addition of BMP-1 alone did not result in LTF (Fig. 2 B and C, 2 × 5HT vs. BMP-1: mean difference = 0.3, 95% CI = 0.11 to 0.43, ***P = 0.0001, 1 × 5HT vs. BMP-1: mean difference = −0.05, 95% CI = −0.2 to 0.11, P = 0.85). Thus, BMP-1 treatment produces a gain of function in the induction of LTF but only when paired with a single pulse of 5HT. These findings support the hypothesis that the extracellular proteolytic activation of an inactive TGFβ proligand serves to bind the two trials together: without trial one, the latent proligand is not available, but without trial two, it is not activated, other than by an externally supplied protease.

Fig. 2.

BMP-1 paired with a single pulse of 5HT promotes the induction of LTF. (A) Schematic of the experimental paradigm. LTF was recorded in cocultures of SNs and MNs. SNs were stimulated extracellularly to elicit a single action potential, and EPSP amplitudes were measured in motor neurons by intracellular recording (Left). After prerecordings, cultures were subjected to one (1 × 5HT) or two (2 × 5HT) pulses of 5HT at 50 µM, each lasting 5 min (ITI = 45 min), followed by 1 h incubation with or without BMP-1 (5 µg/mL, purple). Intracellular recordings were repeated 20 to 24 h after training (Right). (B) Representative EPSP traces recorded for each condition (pre-) and 20 to 24 h after (post-) treatment with 5HT (Left). (Scale bar, 10 mV/50 ms.) (C) One pulse of 5HT paired with BMP-1 induces LTF comparable to 2 × 5HT. Changes in EPSP amplitudes were expressed as log10 [post:pre]. n = 19 to 32. Asterisks indicate significance in Tukey’s post hoc tests following ANOVA tests. *P < 0.05, **P < 0.01, and ****P < 0.0001. All data shown are mean ± SEM.

Sequestering TGFβ During Trial Two Blocks LTF.

Since external BMP-1 induced LTF with subthreshold training (Fig. 2 B and C), we then sought to verify whether the effect was indeed due to the activation of TGFβ. To examine this question, we first confirmed that TGFβ was required for two-trial LTF. We utilized the soluble TGFβ receptor (TGFβ-SR) which sequesters endogenous TGFβ ligand and thereby inhibits TGFβ signaling [Fig. 3A (2, 9, 24)]. Our hypothesis predicts that BMP-1 activates TGFβ during trial two. Application of TGFβ-SR was applied before (t = 30 min), during (t = 45 min), and after trial two (A t = 60 min, B t = 1 h 45 min, Fig. 3A). Indeed, application of TGFβ-SR during and after trial two blocked LTF, indicating that release of active TGFβ must occur during and after this trial in order for LTF to occur (Fig. 3 B and C, untreated = 0.041 ± 0.06, n = 7; 2 × 5HT = 0.5 ± 0.1, n = 14; 2 × 5HT A = 0.15 ± 0.07 n = 12; 2 × 5HT B = −0.09 ± 0.08, n = 11; ANOVA: F = 9.9, ****P < 0.0001, Tukey’s multiple comparison results: 2 × 5HT vs. 2 × 5HT A = mean difference:0.36, 95% CI = 0.06 to 0.7, *P = 0.015; 2 × 5HT vs. 2 × 5HT B: mean difference = 0.61, 95% CI = 0.3 to 0.92, ****P < 0.0001, untreated vs. 2 × 5HT A: mean difference = −0.11, 95% CI = −0.5 to 0.3, P = 0.9, untreated vs. 2 × 5HT B: mean difference = 0.13, 95% CI = −0.24 to 0.51, P = 0.8, untreated vs. 2 × 5HT: mean difference = −0.5, 95% CI = −0.83 to 0.11, **P = 0.006). Additionally, when comparing the time and duration of TGFβ-SR incubation, B (B t = 1 h 45 min) did not produce a greater block of LTF compared to A (2 × 5HT A vs. 2 × 5HT B: mean difference = 0.24, 95% CI = −0.08 to 0.6, P = 0.2). Together, these data suggest that blocking TGFβ signaling at any point during or after LTF training impairs LTF.

Fig. 3.

Gain of function effect of BMP-1 on LTF is mediated by the TGFβ signaling cascade. (A) Sequestering TGFβ before, during, and after trial two. Cultures were incubated with TGFβ-soluble receptor (TGFβ-SR, 5 µg/mL, orange) 15 min before, during, and after trial two for either 15 min (A) or for an hour after trial two (B). (B) Sequestering TGFβ before, during, and after trial two inhibits LTF. Representative EPSP traces recorded for each condition (pre-) and 20 to 24 h after (post-) treatment with 5HT. n = 7 to 14. (Scale bar, 10 mV/50 ms.) (C) Changes in EPSP amplitudes were expressed as log10 [post:pre]. (D) Application of TGFβ soluble receptor during BMP-1. LTF was recorded in cocultures of SNs and MNs as in Fig. 2. After prerecordings, cultures were subjected to one (1 × 5HT) pulse of 5HT at 50 µM, lasting 5 min, followed by 1 h incubation with BMP-1 (5 µg/mL, purple), with or without TGFβ-SR. (E) Application of TGFβ soluble receptor blocks BMP-1’s gain of function effect on LTF. Representative EPSP traces recorded for each condition (pre-) and 20 to 24 h after (post-) treatment with 5HT. n = 13 to 15. (Scale bar, 10 mV/50 ms.) (F) Changes in EPSP amplitudes were expressed as log10 [post:pre]. Asterisks for (B and C) indicate significance in Tukey’s post hoc tests following ANOVA tests. Asterisks for (E and F) indicate significance in two-tailed unpaired Student’s t test. Data are represented as mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. All data shown are mean ± SEM.

BMP-1 Proteolytically Activates Latent TGFβ During the Induction of LTF.

Thus far, we have established that bioactive TGFβ is needed during the second training trial to induce LTF and if trial two is replaced with metalloprotease BMP-1, LTF is induced. We next tested whether the BMP-1-induced gain of function was due to TGFβ activation. We paired a single pulse of 5HT with a 1 h incubation in media containing BMP-1 with or without TGFβ-SR (Fig. 3D, 1 × 5HT + BMP-1: 0.26 ± 0.07, one data point removed as an outlier, n = 15, 1 × 5HT + BMP-1 +TGFβ-SR: −0.1 ± 0.05, n = 13). We found that application of TGFβ-SR blocked the gain of function observed with a single 5HT pulse paired with BMP-1 treatment (Fig. 3 E and F, 1 × 5HT + BMP vs. 1 × 5HT + BMP-1 +TGFβ-SR: t = 4.11, df = 26, 95% CI = −0.54 to 0.18, ***P = 0.0003). Our findings support the conclusion that BMP-1 specifically activates latent TGFβ in the induction of LTF.

Blocking Proteolytic Cleavage of TGFβ by Inhibiting BMP-1 Blocks the Induction of LTF.

Our data support a model wherein protease BMP-1 mediates the induction of LTF when paired with subthreshold 5HT stimulation. We therefore posited that inhibiting BMP-1 (thereby blocking proteolytic cleavage of latent TGFβ) should block two-trial LTF. To examine this, we utilized two different methods of blocking BMP-1 and applied these blockers throughout training and 24 h afterward to fully assess potential changes to synaptic LTF. The first method involved using a calcium-saturated divalent cation chelator (calcium-saturated ethylenediaminetetraacetic acid, EDTA·Ca), and the second method involved using a BMP-1 specific small-molecule inhibitor, UK383-367. BMP-1 is a zinc-dependent metalloprotease (25, 26). Sequestration of Zn²⁺ ions should therefore block the proteolytic activation of TGFβ. EDTA, which has a higher affinity for Zn²⁺ ions compared to other common divalent metals such as Mg²⁺ and Ca²⁺, has been used to effectively block metalloprotease activity and its saturation with Ca²⁺ prevents the blockade of synaptic transmission (27–30). The second method of blocking BMP-1 involved using UK383-367, a small molecule inhibitor that directly targets the astacin catalytic domain of BMP-1 which binds the Zn²⁺ cofactor (31). UK383-367 has been shown in previous studies to be a highly effective and targeted blocker of BMP-1 (31, 32).

SN+MN pairs were treated with two pulses of 5HT followed by incubation with or without EDTA·Ca ±ZnCl₂ for 24 h (Fig. 4A, 2 × 5HT = 0.22 ± 0.04, n = 23, 2 × 5HT + EDTA·Ca = 0.03 ± 0.04, one data point removed as an outlier, n = 22, 2 × 5HT + ZnCl₂ + EDTA·Ca = 0.4 ± 0.04, three data points removed as outliers, n = 13, ANOVA: F = 17.05, ****P < 0.0001). EDTA·Ca completely blocked LTF induced by two 5HT pulses, indicating that a divalent cation-dependent, extracellular enzymatic activity is critical for two-trial LTF (Tukey’s multiple comparison results: 2 × 5HT vs. 2 × 5HT+EDTA·Ca: mean difference = 0.19, 95% CI = 0.07 to 0.32, **P = 0.0014). However, when an excess of ZnCl₂ was included in the incubation with EDTA·Ca, LTF was restored and comparable to that observed with two trials, confirming that the effect of EDTA·Ca is due to zinc chelation (Fig. 4 B and C). We did find that two trials with ZnCl₂ & EDTA·Ca gave a slightly stronger LTF compared to two trials alone, which suggests that the excess ZnCl₂ may have enhanced proteolytic processing of latent TGFβ (2 × 5HT vs. 2 × 5HT+ ZnCl₂ +EDTA·Ca: mean difference = −0.18, 95% CI = −0.34 to 0.02, *P = 0.023). When we compare 2 × 5HT+EDTA·Ca to 2 × 5HT+ ZnCl2 +EDTA·Ca, 2 × 5HT+ ZnCl₂ +EDTA·Ca does indeed give strong, significant LTF (2 × 5HT+ ZnCl₂ +EDTA·Ca vs. 2 × 5HT+EDTA·Ca: mean difference = −0.4, 95% CI = −0.5 to −0.21, ****P < 0.0001). Thus, restoration of the zinc cofactor resulted in the induction of robust LTF.

Fig. 4.

Inhibiting endogenous BMP-1 blocks LTF. (A) Two-trial LTF is zinc-dependent. After prerecordings, cultures were subjected to two (2 × 5HT) pulses of 5HT at 50 µM, lasting 5 min (ITI = 45 min). Experimental groups also received EDTA·Ca [(10 mM), blue, bar not visible due to tight range], with or without ZnCl₂ [(15 mM), green] during training, followed by 20 to 24 h of incubation. (B) Zinc chelation during and after 2 × 5HT treatment blocks LTF. Representative EPSP traces recorded for each condition (pre-) and 20 to 24 h after (post-) treatment with 5HT. n = 10 to 23. (Scale bar, 10 mV/50 ms.) (C) Changes in EPSP amplitudes were expressed as log10 [post:pre]. (D) Two-trial LTF induction in the presence of BMP-1 inhibitor UK383-367. Prereordered cultures underwent 2 × 5HT (50 µM, ITI = 45 min) with or without 20 to 24 h inhibitor UK (UK383-367, 10 µM, pink) application. Representative EPSP traces recorded for each condition (pre-) and 20 to 24 h after (post-) treatment with 5HT. (Scale bar, 10 mV/50 ms.) (E) BMP-1 inhibitor UK383-367 blocks LTF. Representative EPSP traces recorded for each condition (pre-) and 20 to 24 h after (post-) treatment with 5HT. n = 15 to 20. (Scale bar, 10 mV/50 ms.) (F) Changes in EPSP amplitudes were expressed as log10 [post:pre]. Asterisks for (B and C) indicate significance in Tukey’s post hoc tests following ANOVA tests. Asterisks for (E and F) indicate significance in one-tailed unpaired Student’s t test. Data are represented as mean ± SEM. *P < 0.05, **P < 0.01, and ****P < 0.0001. All data shown are mean ± SEM.

We next applied the selective BMP-1 inhibitor UK 383-367 during two-trial training and incubated with inhibitor for 24 h (Fig. 4D, 2 × 5HT: 0.2 ± 0.07, n = 20, 2 × 5HT + UK: 0.03 ± 0.05, one data point removed as outlier, n = 15). We then compared these results to two-trial training plus vehicle. Consistent with our EDTA·Ca findings (Fig. 4 B and C), blocking BMP-1 with UK 383-367 blocked the induction of LTF (Student’s t test results: t = 1.9, df = 33, 95% CI = −0.4 to 0.02, *P = 0.03, one-tailed t test, predicted by the direction of the effect of previous results with blocker EDTA·Ca, Fig. 4 A–C). Taken collectively, two distinct BMP-1 inhibitors confirm that the protease is necessary for LTF.

Cytoplasmic-to-Nuclear Smad Translocation Is a Reporter for TGFβ Activity in Aplysia SNs.

We observed a synaptic gain of function when we paired subthreshold stimulation of one trial of 5HT with BMP-1 application which resulted in LTF (Fig. 2). In the next series of experiments, we examined the molecular mechanism underlying this gain of function by quantifying TGFβ activity over time through tracking movement of Smad, a downstream TGFβ-directed mediator.

In mammalian systems, TGFβ receptor (TGFβ-R) activation causes translocation of a downstream mediator, Receptor Smad (also referred to as a R-Smad, Smad2/3, Smad) from the cytosol into the nucleus (33) (Fig. 5A). We posited that Smad translocation could serve as a reporter of TGFβ activity in Aplysia SNs, which would allow us to determine whether TGFβ signaling is differentially engaged by trials one and two. We designed two experiments to test whether Smad translocation could be used as a reporter for TGFβ activity: i) artificially inducing TGFβ signaling by applying TGFβ ligand and ii) blocking TGFβ signaling with TGFβ-SR to examine the corresponding changes in the amount of Smad translocation.

Fig. 5.

BMP-1 paired with a single 5HT pulse activates the TGFβ cascade. (A) Schematic illustrates nuclear translocation of R-Smad proteins in response to TGFβ-R activation in the canonical TGFβ signaling pathway. (B) Western blot with untreated Aplysia ganglia lysates using Smad2/3 antibody (1:1,000, Santa Cruz) detects band at the predicted molecular weight for ApSmad2/3 (~41 kDa). A duplicate sample and empty lanes to the right of the protein sample were cropped out of the gel image. (C) Experimental paradigm. Cultured SNs were incubated with recombinant human TGFβ (100 ng/mL) for 15 or 45 min and then stained for ApSmad2/3. (D and E) Smad translocates in response to TGFβ-ligand application. The ratio of nuclear:cytosolic Smad immunofluorescence revealed an increase in the amount of Smad that had translocated into the nucleus in both incubation periods. (Scale bar here and for all images, 10 µm.) n = 31 to 44. Asterisks indicate significance in Tukey’s post hoc tests following ANOVA tests. ****P < 0.0001. All data shown are mean ± SEM. (F) Experimental paradigm. SN cultures were subjected to two (2 × 5HT) pulses of 5HT at 50 µM, each lasting 5 min (ITI = 45 min) with or without TGFβ soluble receptor (5 µg/mL) incubation, before, during, and after trial two (total incubation time = 30 min). (G and H) Smad translocation is blocked by soluble TGFβ receptor application during trial two. n = 21 to 28, for ICC experiments, “n” represents the number of individual SNs cells. Asterisks indicate significance in Tukey’s post hoc tests following ANOVA tests. *P < 0.05, and ****P < 0.0001. All data shown are mean ± SEM. (I) Experimental paradigm. SN cultures were exposed to one (1 × 5HT) 5-min pulse of 5HT at 50 µM, with or without recombinant human BMP-1 (5 µg/mL) incubation for 1 h. (J and K) BMP-1 induces an increase in Smad translocation but only when paired with one pulse of 5HT. Student’s t test was used. n = 14 to 46. Asterisks indicate significance in Dunn’s post hoc tests following Kruskal–Wallis tests. ****P < 0.0001. All data shown are mean ± SEM.

First, we treated cultured Aplysia SNs for 15 or 45 min with TGFβ to induce TGFβ-R activation. This TGFβ ligand has been used previously to study TGFβ signaling in Aplysia (2, 34). Following treatment, cells were processed for immunocytochemistry probing for Smad2/3 using an antibody which we found to recognize a single band at the predicted molecular weight of Aplysia Smad2/3 (ApSmad2/3, Fig. 5B). We then measured levels of ApSmad2/3 in the nucleus and cytoplasm by immunofluorescence. Following prolonged application of TGFβ, there was a clear overall increase in both the cytoplasmic and nuclear ApSmad2/3 expression (Fig. 5 C–E, vehicle control = −0.02 ± 0.03, n = 31, TGFβ 15’ = 0.19 ± 0.02, n = 38, TGFβ 45’ = 0.21 ± 0.02, n = 44, ANOVA: F = 23.08, ****P < 0.0001). We observed a significant redistribution of ApSmad2/3 immunoreactivity from the cytosol to the nucleus compared to the vehicle-treated control, indicating that ApSmad2/3 translocation is a reporter for TGFβ-induced activation in Aplysia SNs (Fig. 5 D and E, TGFβ 15’ vs. vehicle: mean difference = −0.2, 95% CI = −0.3 to 0.1, ****P < 0.0001, TGFβ 45’ vs. vehicle: mean difference = −0.2, 95% CI = −0.3 to 0.14, ****P < 0.0001).

To block TGFβ signaling, we again utilized TGFβ-SR application. Two-trial training was administered to cultured SNs, and, during the second trial, TGFβ-SR was applied (vehicle control: −0.01 ± 0.03, n = 21, 2 × 5HT: 0.3 ± 0.03, n = 28, 2 × 5HT+TGFβ-SR: 0.1 ± 0.03, n = 28, ANOVA: F = 24.86, ****P < 0.0001). As expected, inclusion of the TGFβ-SR during trial two significantly decreased ApSmad2/3 translocation compared to the two trials alone (Fig. 5 F–H, 2 × 5HT vs. 2 × 5HT+TGFβ-SR: mean difference = −0.19, 95% CI = −0.3 to −0.09, ****P < 0.0001, 2 × 5HT vs. vehicle: difference = −0.3, 95% CI = −0.4 to −0.2, ****P < 0.0001). We also found that pairing TGFβ-SR with two-trial training gave a slight increase in Smad translocation over the vehicle control, suggesting that there was still some residual TGFβ that the TGFβ-SR was not able to sequester during training (2 × 5HT+TGFβ-SR vs. vehicle: difference = −0.1, 95% CI = −0.2 to −0.003, *P = 0.04). Taken together, these data indicate that ApSmad2/3 translocation is a reliable bioassay for TGFβ activity in Aplysia SNs.

Metalloprotease BMP-1 Paired with a Single Training Trial Induces Increased TGFβ Activity.

We next examined whether the ApSmad2/3 translocation bioassay we developed could reflect the synaptic gain of function we observed previously with BMP-1 (Fig. 2 A–C). Cultured SNs were exposed to one of four protocols: i) a single training trial, ii) a single training trial followed by incubation with human recombinant BMP-1, iii) BMP-1 application alone, or iv) vehicle control. (1 × 5HT: 0.09 ± 0.03, one data point removed as an outlier, n = 17, 1 × 5HT+BMP-1: 0.3 ± 0.02, six data points removed as outliers, n = 46, BMP-1 = −0.006 ± 0.05, two data points removed as outliers, n = 21, vehicle = −0.02 ± 0.04, n = 14, Kruskal–Wallis: KW statistic = 56.98, ****P < 0.0001). Treatment of SNs with a combination of a single training trial and recombinant human BMP-1 significantly induced ApSmad2/3 nuclear translocation (Fig. 5 I–K, Data are from Dunn’s post hoc tests following Kruskal–Wallis tests. 1 × 5HT+BMP-1 vs. vehicle: mean difference = −49.42, ****P < 0.0001, 1 × 5HT+BMP-1 vs. BMP-1: mean difference = −44.27, ****P < 0.0001, 1 × 5HT+BMP-1 vs. 1 × 5HT: mean difference = 35.07, ****P < 0.0001). BMP-1 incubation alone did not induce translocation and was not significantly different from the vehicle control (BMP-1 vs. vehicle control: mean difference = −5.14, P > 0.99). While a single trial produced a modest and statistically insignificant increase in ApSmad2/3 translocation compared to vehicle control, adding the protease to the system produced a significant increase in the amount of TGFβ signaling (Fig. 5J, 1 × 5HT vs. vehicle: difference = −14.34, P = 0.97, 1 × 5HT+BMP-1 vs. 1 × 5HT: mean difference = 35.07, ****P < 0.0001, 1 × 5HT-1 vs. BMP-1: mean difference = −9.2, P > 0.99, BMP-1 vs. vehicle: mean difference = −5.14, P > 0.99). These results reveal a significant molecular gain of function in the effects of 5HT paired with BMP-1 on ApSmad2/3 translocation and add further support to the hypothesis that proteolytic activation of TGFβ by BMP-1 underlies intertrial interactions toward the effective induction of LTM.

Discussion

Our results reveal how the multistep arrangement of the TGFβ signaling pathway critically controls the induction of long-term synaptic plasticity. We have identified a key step in the pathway that involves an interaction between the release of inert TGFβ and its cleavage by the protease BMP-1. These two steps must temporally coincide if LTF is to be induced. To identify the mechanism underlying this interaction, we combined electrophysiological recordings from SN+MN pairs with a culture-based bioassay to examine TGFβ signaling at the synaptic and molecular levels. Our model begins with the secretion of latent LAP-bound TGFβ and BMP-1 from the SN, in response to 5HT signaling (Fig. 6). Next, latent TGFβ is docked in the extracellular matrix where it is associated with the latent TGFβ binding protein to form the large latent complex. TGFβ must then be cleaved by BMP-1 to liberate the ligand from both the extracellular matrix and from the LAP. Only after this proteolytic cleavage can TGFβ become bioactive and, by binding to its receptor on the SN, induce intracellular downstream signaling. Our data suggest that proteolytic activation of latent TGFβ occurs specifically during trial two and results in an increase in the rate of TGFβ signaling that is necessary for LTM. Accordingly, we are able to induce both LTF and increases in TGFβ signaling with a subthreshold stimulus (a single training trial) coupled with the addition of BMP-1. TGFβ signaling, and GF signaling in general, has long been implicated in memory formation. However, our results provide additional insights into the precise molecular choreography that must occur within the TGFβ cascade to induce long-term synaptic changes. These experiments provide unexpected evidence of the necessity of patterned activity within the TGFβ signaling cascade for the induction of LTM.

Fig. 6.

Proposed model for the timing of TGFβ activity that ultimately leads to LTM formation. (1) SN releases BMP-1 or latent TGFβ in response to 5HT from trial one. (2) Latent TGFβ docked in the extracellular matrix is released due to BMP-1’s proteolytic activity occurring during trial two. (3) Liberated TGFβ ligand binds and activates its receptor. (4) Activated TGFβ-R phosphorylates R-Smads, which binds co-Smad and translocates to the nucleus. (5). Once in the nucleus, R-Smad-co-Smad dimer binds to DNA and acts as a cotranscription factor.

Temporal and Spatial Aspects of TGFβ Activation.

An important question raised by our data concerns the timing of the molecular events that occur during two-trial training. When are latent TGFβ and BMP-1 released into the system? One potential mechanism for this intertrial interaction could be that trial one triggers the slow continual release of latent TGFβ that must build up over a 45-min period. Trial two then deploys protease BMP-1 causing fast activation of latent TGFβ in the extracellular matrix. Alternatively, BMP-1 could be slowly released by the first trial while inactive TGFβ is quickly discharged by either of the trials. BMP-1 could also be activated by means other than secretion. At present, our data support the hypothesis that trial one releases latent TGFβ that is then sequestered in the extracellular matrix (Fig. 5). We are able to induce high levels of TGFβ signaling in the ApSmad2/3 assay with a single training trial paired with a 5-min pulse of BMP-1. Typically, this level of activity would only occur with two trials of 5HT. This finding suggests that a majority of latent TGFβ is released during the 5-min period of trial one. Currently, it is not known when BMP-1 is released during two-trial training. Based on our findings, we predict that trial two triggers BMP-1 release. Indeed, this prediction is supported by previous findings in which BMP-/Tolloid mRNA levels in Aplysia are increased following long-term memory induction (35). Further experiments charting and quantifying the release of active TGFβ as well as BMP-1 will be required to develop a more nuanced understanding of the molecular choreography underlying TGFβ signaling.

Our present findings performed in both isolated SNs and SN+MN cocultures also raise the question of the cell type that could be the source of both latent TGFβ and BMP-1. In the ApSmad2/3 bioassay experiments, it is clear that SNs are the source of latent TGFβ, as SNs were the only cell type used in this protocol. We found that TGFβ signaling increased with two training trials or with one training trial with exogenous BMP-1 application. These findings suggest that SNs are acting in a cell-autonomous fashion by releasing latent TGFβ, resulting in stronger TGFβ signaling, but only if BMP-1 is coincidentally present. This hypothesis must be integrated with the data resulting from the addition of MNs to the SN+MN LTF recordings. Previous work in mammalian systems has found that one effect of TGFβ is to induce further expression of BMP-1, resulting in a positive feedback loop promoting TGFβ activity (18). It is possible that the MNs may be the source of either BMP-1 or TGFβ and that a coincidence must occur between MN and SN in order for TGFβ signaling to be sustained and LTF to occur. Interestingly, previous work done in Aplysia demonstrates that BMP-1 mRNA is expressed basally at the highest level in SNs compared to other tissues and its gene expression is enhanced with 5HT application and behavioral training (35). Given these findings, SNs may in fact increase their mRNA expression of BMP-1/Tolloid as well be its source during our two-trial system. Future studies are naturally required to confirm this hypothesis.

BMP-1 metalloprotease is known to cleave latent, extracellular matrix–bound TGFβ before the ligand can signal via its receptor (18, 36). BMP-1 has been previously studied during memory formation in Aplysia. Liu et al. (35) characterized an Aplysia homolog of the Drosophila metalloprotease Tolloid, which is a homolog of human BMP-1. It is reasonable to expect that this is the protease mediating the cleavage of latent TGFβ in our system, which we have been referring to as BMP-1. Interestingly, Liu et al. (35) found that BMP-1 mRNA is up-regulated in Aplysia by sensitizing stimuli and identified BMP-1 protein within the neurons’ secretory pathway. Their findings thus lend additional support to our hypothesis that the induction and secretion of BMP-1 may control the onset of TGFβ signaling and account for its delayed engagement (35). It must be noted, however, that it is unknown whether BMP-1/Tolloid protein levels are affected by sensitization training in Aplysia. Future studies quantifying changes in the BMP-1/ Tolloid protein levels in response to 5HT application would verify whether the production of BMP-1/Tolloid is in fact up-regulated with training.

A question that arises from our studies is whether the TGFβ patterned activity may be the key signaling pathway that occurs during the 45-min ITI between our two trials. While our findings address the necessity of two training trials for LTF induction, they do not explain the specific requirement of the 45-min ITI. We have previously found that a 45-min interval separating two trials induces LTM for sensitization while neither a 15- or 60-min interval does so in intact Aplysia (37). Studies of SN+MN cocultures demonstrated that a 45-min ITI is the most effective interval to inducing LTF; however, other intervals are also permissive to a lesser extent in that specific system (38). Thus, the distinction between 45 and 60 min is not relevant to our cell culture studies, in which this ITI is more broadly permissive and not as strict as it is in intact Aplysia behavior.

Extracellular Matrix.

The ECM plays a critical role in the mammalian TGFβ signaling cascade. The ECM docks latent TGFβ that will eventually be cleaved by metalloproteases. Based on the ECM’s central role in TGFβ activation, it is important to consider the nature of the ECM in our culture system. We have previously established that cultured Aplysia cells alter their morphology in response to application of extracellular matrix substrata, while others have found that Aplysia neurons express an integrin-like ECM receptor (39, 40). Furthermore, detailed anatomical studies revealed an extracellular matrix that surrounds the Aplysia ganglia, part of which is the sheath (41). The sheath, a thick layer of connective tissue surrounding the ganglia, was found to contain basement membranes which is a layer of the ECM rich in laminin (41). Collectively, these studies demonstrate that the Aplysia nervous system possesses an ECM. While a variety of Aplysia neurons have been extensively studied in culture, a detailed investigation into the structure and function of this ECM in culture remains to be undertaken. Anecdotally, cells disassociated for culture are isolated with a thick layer of connective tissue that adheres to the isolating electrode and occasionally contains embedded glial cells. Detailed studies of the composition of this isolated tissue may reveal it to be a portion of the ECM. That said, in our experiments, TGFβ behaves exactly as predicted from many other studies in several experimental systems (it is activated via BMP-1 application, an effect which gives rise to Smad translocation and is blocked by inhibitors of signaling) it is reasonable to posit that an ECM-like structure is functionally active in our culture system.

Smad Signaling and TGFβ-R.

In mammalian systems, TGFβ-R activation initiates downstream signaling through receptor-associated Smads. Smads are intracellular mediators of TGFβ signaling that, upon activation, translocate into the nucleus and directly regulate gene expression. There is a high level of specificity with Smad activation during TGFβ signaling. Previous work has demonstrated that both the level of TGFβ-R activation and its duration are correlated with the level and persistence of activated Smad complexes in the nucleus (42–44). Since it is the primary downstream signal transducer of TGFβ that is exclusively activated by TGFβ, we reasoned that Smad could serve as a reliable reporter of TGFβ activity. We thus established a cell culture–based bioassay that monitors the effects of TGFβ based on the nuclear translocation of Smad proteins (Fig. 4). Using this assay, we found that TGFβ signaling is differentially engaged after two-trial training through an increase in the rate of signaling, and reasoned that the protease BMP-1 could be a potential broker of this differential engagement (Fig. 4). Smad, in addition to providing a reliable readout for TGFβ activity, may itself be a driving factor that brokers temporal information instructive for memory formation. Smads are key transcription regulators that bind directly to DNA and interact with DNA-binding cofactors including CREB binding protein (CBP). Mutations in the gene encoding CBP have been found to cause deficits in long-term plasticity, learning, and memory (45–47). Additionally, injecting CBP-siRNA into Aplysia SNs of SN+MN cocultures, resulted in impaired LTF (48). Future studies blocking endogenous Smad via siRNA in the SNs could inform the question of whether Smad is necessary for LTF and therefore for LTM. An important question emerging from the present work concerns the nature of the gene expression generated in response to Smad’s activity in the nucleus. RNA sequencing after Smad nuclear translocation, as well as following the blockade of Smad with siRNA, could provide further insights into the specific genes that are targeted by TGFβ signaling to support long-term memory formation.

Independent of whether TGFβ availability and activity are modulated by intertrial interactions, the differential engagement of downstream TGFβ-dependent mechanisms in response to two (but not one) training trials could also be controlled at the level of TGFβ receptor synthesis or localization (49). Thus, it will be critical for future studies to determine whether protein and mRNA levels of TGFβ-R are regulated by individual training trials during memory formation. These experiments will significantly inform our overall model of TGFβ signaling regulation during LTM formation.

TGFβ’s Downstream Signaling.

An additional area of interest developing from the present study concerns the downstream effector proteins that TGFβ may interact with either directly or indirectly. We are particularly interested in TGFβ’s relation to kinases like ERK and transcription factors like CREB. Extracellular signal–regulated kinase (ERK, p42/44 mitogen-activated protein kinase, MAPK) is necessary for both LTM and LTF in multiple model systems, including Aplysia (23, 50, 51). In addition, ERK activation has been connected to GF signaling in Aplysia (9, 24, 52). Sharma et al. (24) demonstrated that GF TrkB-like ligand is required for memory formation through an ERK-dependent mechanism (24). Previous studies have also found that incubation of ganglia or SN+MN cocultures with mammalian TGFβ or brain-derived neurotrophic factor (BDNF) promotes MAPK activation (2, 34, 53, 54). Collectively, these studies demonstrate that TGFβ is a potent activator of kinases like ERK and that their interaction is key for memory formation and synaptic plasticity. Another target of interest in relation to TGFβ is cAMP response element-binding protein (CREB) whose transcriptional activation is required for LTF and memory (55, 56). Evidence from the mammalian system demonstrates that Smad interacts directly with the C terminus of coactivator p300/CREB binding protein (CBP) ultimately resulting in transcriptional regulation (57, 58). Additional experiments performed in mouse embryonic fibroblast cultures found that TGFβ activation induced CREB phosphorylation in p38 MAPK-dependent manner (59). Therefore, an interesting area of further research would be to see whether repeated trial training performed in Aplysia results in an increased Smad-p300/CBP interaction. Analysis of the genes activated by this association during memory formation could provide valuable insight into the ways in which TGFβ signaling influences memory formation at the transcriptional level.

In conclusion, our current results establish TGFβ as a molecular timekeeper that decodes trial repetition during the induction of LTM. On a broader level, these data suggest a role for GF signaling as a molecular platform for pattern detection in memory formation.

Materials and Methods

Experimental Model and Subject Details.

Aplysia californica were acquired from the University of Miami National Resource for Aplysia. All animals were allowed to acclimate for 1 to 2 d in circulating tanks with artificial seawater (ASW, Instant Ocean) at 15 °C before culturing.

Reagents.

Human recombinant BMP-1/PCP, TGFβ, and TGFβ-SR are from R&D Systems. UK383,367 is from Tocris. Smad2/3 antibody is from Santa Cruz Biotechnics. All other reagents were from Sigma unless indicated otherwise. The following concentrations were used for each drug, diluted in the solvent recommended by the manufacturer: 5HT, 50 µM in 1:1 solution of L15:ASW, TGFβ-SR, 5 µg/mL, recombinant human TGFβ-1 protein, 5 µg/mL, vehicle control, 0.1% bovine serum albumin (BSA) in phosphate-buffered saline (PBS), BMP-1, 5 µg/mL, EDTA·Ca, 10 mM, ZnCl₂, 15 mM, UK383-367, 10 µM.

Cell Culture Preparation.

SNs were cultured according to previously described protocols (60, 61) For isolated SNs, mariculture-raised A. californica (60 to 100 g, Aplysia Resource Growout Facility, University of Miami) were anesthetized by injection of MgCl₂ (369 mM MgCl₂ and 10 mM Tris–HCl H 7.6), pleural-pedal ganglia removed, and incubated for 3 h at 22 °C in a solution containing 5 U/mL dispase (Gibco) in salt-adjusted L15 (Leibovitz) medium (Sigma; supplemented with 264 mM NaCl, 26 mM MgSO₄, 27 mM MgCl₂, 5 mM KCl, 2 mM NaHCO₃, 11 mM CaCl₂, 15 mM N-2-hydroxyethylpiperazine-N-2-ethane sulfonic acid (HEPES), 35 mM glucose, 100 U/mL penicillin, and 0.1 mg/mL streptomycin). Ganglia were washed in ASW (460 mM NaCl, 55 mM MgCl₂, 11 mM CaCl₂, 10 mM KCl, and 10 mM Tris–HCl pH 7.6) and desheathed in a 1:1 solution of L15:ASW. L7 MNs were identified based on morphological characteristics and location within the ganglia. SNs were identified based on size and location within the ventral sensory cluster in the pleural ganglion and extracted from the SN cluster of the pleural ganglion. All neurons were individually extracted with a glass microelectrode (62). After extraction, neurons were transferred to and sparsely distributed on a glass bottom dish precoated with poly-L-lysine (MW ≥ 300,000 Da; 0.75 mg/mL hydrobromide in 0.1 M sodium borate pH 8.2, 2 h at 37 °C) followed by a solution containing poly-D-lysine and laminin (Cell Applications), 2 h at room temperature. Approximately 10 to 15 cells were seeded per plate, to mitigate potential cell death. Typically, each plate would have 10 cells at minimum. All n numbers represent the number of cells used in each experimental group. Cells were cultured in a humidified atmosphere at 16 °C in full culture medium, [50% salt-adjusted L15, 2 mM L-glutamine, and 50% Aplysia hemolymph (collected in the spring, pooled from ~10 wild-caught animals, and stored at −80 °C)].

For LTF experiments, SNs and MNs were cultured as previously described (62). MNs were plated first and allowed to attach for 24 h at room temperature, followed by addition of SNs in physical contact to the MN neurites. Media were changed on day 3 post-SN neuron plating. Electrophysiological LTF recordings were carried out on days 4 and 5 post-SN neuron plating in 1:1 L15:ASW. Recording L15:ASW was exchanged with fresh media after training.

Ganglion Preparation.

Pleural-pedal ganglia were dissected from anesthetized A. californica (150 to 250 g). In a 1:1 solution of MgCl₂ and ASW, the pleural SN cluster and SN-MN neuropil were exposed. Ganglia were perfused with ASW for at least 1.5 h to clear MgCl₂ before processing for Western Blot.

Western Blot.

Ganglion samples were lysed and snap-frozen in RIPA buffer (Sigma Millipore) with 1× Halt! protease and phosphatase inhibitor cocktail (Invitrogen), loaded onto 4 to 12% Bis-Tris gels (Novex; Invitrogen) for SDS-PAGE, and then transferred to nitrocellulose membranes for incubation with primary antibodies and secondary antibodies: anti-Smad2/3 (1:1,000; Santa Cruz Biotechnology, sc-133098) and anti-mouse IgG IRDye 680L (1:10,000; LICOR Biosciences, #926-68 020). Protein levels and band weights were assessed using the LICOR imaging system and EmpiriaStudio suite software.

Smad Translocation Assay.

SNs were cultured as described above. After 4 DIV, 7 plates of 10 to 20 SNs were randomly separated into four different groups. Two plates were designated into one of three experimental groups depending on the experiment (e.g., 2 × 5HT, 2 × 5HT + TGFβ-SR, No Treatment). One plate was used as a “secondaries-only” control to determine background fluorescence of SNs without primary Smad antibody, and the average fluorescence from these secondaries-only plates was subtracted from the experimental plates.

The immunocytochemistry protocol used here was adapted from Bougie et al. (63) and Kukushkin et al. (38). Following treatment according to the experiment paradigm, the cells were fixed for 30 min at room temperature with ice-cold 4% formaldehyde in 30% sucrose/Tris-buffered saline (TBS) (38). Cells were then permeabilized with 0.2% Triton X-100 in 30% sucrose/TBS for 30 min and then washed three times for 5 min each with TBS (63). Free aldehydes were then quenched with TBS/50 mM NH₄Cl for 15 min, followed by five washes with TBS at 5 min each wash. Excess TBS was vacuumed out of the dish leaving the center well of the culture dish intact with ~ 200 µL of TBS covering the cells. TBS was then exchanged with 10% normal goat serum in TBS to block nonspecific binding. Cells were incubated with blocking serum for 1 h at room temperature. After blocking, cells were then incubated for 72 h at 4 °C with anti-Smad2/3 (1:200, Cell Signaling Technology, #3102) in 10% normal goat serum. After three washes with TBS, a secondary antibody conjugated to Alexa Fluor 488 (1:500; Thermo) was applied for 2 h at room temperature in the dark. Cells were then washed three times in TBS and mounted using ProLong Gold Antifade Mountant with DAPI (Life Technologies). Slides were allowed to cure overnight, and then, the edges were sealed with clear polish (OPI). Images were obtained with a Leica SP8 confocal microscope using an Olympus 63× oil-immersion lens. Images taken through the middle of the nucleus according to DAPI staining. Mean fluorescent intensity from the cytosolic and nuclear compartments of each cell was measured using the FIJI image processing package.

Electrophysiology.

Intracellular recording was performed using glass microelectrodes containing 3 M KCl with electrode resistance of 5 to 12 MΩ. Data were sampled at 1 kHz without filtering. For LTF experiments, media were exchanged to 1:1 ASW:L15 before impaling motor neurons and hyperpolarizing their membrane to −75 mV. SNs were then stimulated with a blunt silver-coated (SPI Supplies) extracellular electrode filled with ASW to elicit a single spike, and EPSPs measured as pretests. After recording cells were subjected to 5HT treatments. 5HT was added to culture dishes at a final concentration of 50 µM, and washed out 5 min later by perfusion with 1:1 ASW:L15. For LTF induction, this procedure was repeated one more time with an intertrial interval of 45 min. Following the final 5HT washout, 1:1 L15:ASW were replaced by freshly prepared full culture medium and incubated as above for 16 to 20 h. The media were then again exchanged to 1:1 ASW:L15, and the intracellular recordings repeated.

Statistical Analyses.

Statistical analysis and data plotting was conducted with GraphPad Prism 9 software. All data within the text are shown as mean ± SEM. Data were first assessed for outliers using the ROUT method resulting in the removal of <4% of all data points, which are reported in the text. ICC data were transformed by taking a log of the ratio between the experimental sample indicated and an untreated control, log(E:C), for parametric analyses. For LTF recordings, log10 of the post:pre EPSP amplitude ratio was used for parametric analyses. Normal distribution was verified using the Kolmogorov–Smirnov test. For individual comparisons with normally distributed data, unpaired two-tailed unpaired Student’s t tests were used, unless otherwise noted in the text. For multiple comparisons with normally distributed data, ordinary one-way ANOVA was followed by Tukey’s post hoc tests when evaluating comparisons between a series of samples. If distributions of data were deemed not normal after the Kolmogorov–Smirnov test, then multiple comparisons were assessed using nonparametric statistics. Kruskal–Wallis analyses were used for multiple comparisons in non-normal datasets, followed by Dunn’s multiple comparisons tests. In LTF experiments, MNs were paired with 3 to 4 SNs, the responses from which were averaged and treated as a single value; thus, in these experiments, “n” represents individual MNs. In ICC experiments using Smad translocation, “n” represents the number of individual SN cells. Sample sizes and specific statistical tests used can be found in the figure legends.

Supplementary Material

Dataset S01 (XLSX)

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix.