Signaling from synapse to nucleus: the logic behind the mechanisms Karl Deisseroth, Paul G Mermelstein y, Houhui Xia z and Richard W Tsien z#

Transcription

1 1 Signaling from synapse to nucleus: the logic behind the mechanisms Karl Deisseroth, Paul G Mermelstein y, Houhui Xia z and Richard W Tsien z# Signaling from synapse to nucleus is vital for activity-dependent control of neuronal gene expression and represents a sophisticated form of neural computation. The nature of specific signal initiators, nuclear translocators and effectors has become increasingly clear, and supports the idea that the nucleus is able to make sense of a surprising amount of fast synaptic information through intricate biochemical mechanisms. Information transfer to the nucleus can be conveyed by physical translocation of messengers at various stages within the multiple signal transduction cascades that are set in motion by a Ca 2þ rise near the surface membrane. The key role of synapse-to-nucleus signaling in circadian rhythms, long-term memory, and neuronal survival sheds light on the logical underpinning of these signaling mechanisms. Addresses Department of Psychiatry and Behavioral Sciences, Stanford University School of Medicine, Stanford CA 94305, USA deissero@stanford.edu y Department of Neuroscience, University of Minnesota, Jackson Hall, 321 Church Street South East, Minneapolis, MN 55455, USA pmerm@umn.edu z Department of Molecular and Cellular Physiology, Stanford University School of Medicine, Stanford CA 94305, USA houhui@stanford.edu # rwtsien@stanford.edu This review comes from a themed issue on Signalling mechanisms Edited by Morgan Sheng and Terrance P Snutch /03/$ see front matter ß 2003 Elsevier Science Ltd. All rights reserved. DOI /S (03)00076-X Abbreviations BDNF brain-derived neurotrophic factor CaM calmodulin CaMK calmodulin-dependent protein kinase CREB Ca 2þ /camp responsive element binding protein DREAM downstream regulatory element antagonistic modulator NFAT nuclear factor of activated T-cells NMDAR N-methyl-D-aspartate receptor NPY neuropeptide-y PACAP pituitary adenylyl cyclase activating peptide PKA protein kinase A SCN suprachiasmatic nucleus Introduction Neuroscientists have long appreciated the computations that ion channels perform at the surface membranes of neurons. More recently, it has emerged that the complexity of this electrical signal processing is rivaled by the intricacies of biochemical signal processing that occur below the membrane, within elaborate networks of cytoplasmic signaling proteins. However, only within the past few years has it become clear that even deeper within the neuron lies perhaps the most formidable engine of computation of all [1]. The nucleus of the mammalian neuron creates, shapes, links and responds to these electrical and biochemical communication streams by actively controlling the expression of tens of thousands of genes a more diverse repertoire than that presented by any other known cell type. Until recently, little was understood about the specific role of the nucleus, apart from the notion that it could be involved in the maintenance of long-term changes in neuronal function [2]; it was not at all obvious how synapses, as myriad sources of fast electrical signals, could communicate even a tiny fraction of their rich information stream to the distant and apparently slow nucleus. It has now become evident that the nucleus is able to make sense of a surprising amount of fast synaptic information through intricate biochemical mechanisms [3 5]. Furthermore, as we review here, there is now a much clearer understanding of the crucial role of synapse-to-nucleus signaling in certain key neuronal processes, namely circadian rhythms, longterm memory and neuronal survival. This functional and behavioral level of understanding now anchors our interpretation of the underlying logic of the nuclear signaling mechanisms employed. An emerging theme as we show here is that the complexity of neuronal biochemistry may be necessary to generate signal processing streams (from the rapid ion channel kinetics that initially shape electrical information flow to the slower biochemical effectors) that are specific enough to convey an information-laden signal to the nucleus. Many signaling pathways to the nucleus What signaling entity physically translocates to the nucleus to support the transfer of information from the neuronal cell membrane? In their pioneering paper on excitation transcription coupling, Morgan and Curran [6] put forward a hypothesis for how Ca 2þ entry through L-type Ca 2þ channels might activate expression of c-fos, an immediate early gene (IEG; IEGs constitute the first wave of genes to be turned on, often within minutes, after a stimulus). They suggested that a calmodulin (CaM)-sensitive kinase would phosphorylate a transcription factor, initially positioned in the cytoplasm, causing it to move into the nucleus to activate gene expression. This first speculation was prescient in broad outline and provocative CONEUR 55

2 2 Signalling mechanisms in highlighting the following fundamental questions. First, what are the crucial and specific signal initiators? Second, what are the nuclear translocators, and why are they selected for the task? Third and finally, what is the basis for stimulus specificity and reliable information transfer? By focusing on the pathways triggered by an intracellular Ca 2þ rise downstream of synaptic activity, we now know that the signal from the cytoplasm to the nucleus can be conveyed in several different ways (Figure 1). The translocation can take place at various stages within signal transduction cascades that are set in motion by the Ca 2þ rise. At one extreme, as hypothesized by Morgan and Curran, the transcription factor itself, the terminal factor in the biochemical cascades, can move to the nucleus upon changes in its phosphorylation state. The classic example is a member of the nuclear factor of activated T- cells (NFAT) family of transcription factors. Originally characterized in immune tissue, the NFATc group plays a crucial role in neuronal plasticity as well as vascular development and muscular hypertrophy [7]. NFATc4 is expressed within hippocampal neurons, and undergoes a striking translocation from the cytosol to the nucleus upon the opening of L-type Ca 2þ channels [8]. The translocation step is dependent upon calcineurin-mediated dephosphorylation of NFATc4, which causes the unmasking of multiple nuclear localization signals and the active transport of the transcription factor through the nuclear pore complex. At the other extreme, a mechanism may exist that transduces directly the free Ca 2þ transient that can arise in the nucleus after neuronal excitation [9]. Thetranscription factor downstream regulatory element antagonistic modulator (DREAM) [10], a Ca 2þ binding protein that contains three active Ca 2þ binding motifs (E-F hands) [11 ], is abundant in the nucleus. It has been proposed that nuclear DREAM remains bound to a downstream regulatory element (DRE) that acts as a gene silencer when nuclear Ca 2þ is low, but dissociates upon elevation of Ca 2þ causing DRE derepression, activation of downstream genes such as that which encodes prodynorphin [10] and attenuation of pain signaling in vivo [12 ]. It will be interesting to see if DREAM function can be formally linked to nuclear Ca 2þ levels, and to understand why DREAM is present outside the nucleus as well as within it. Are there particular advantages for the signaling mechanisms at either one or the other of these extremes? On the one hand, the nuclear Ca 2þ mechanism that is exemplified by DREAM is straightforward, fast and simple, but it might be a suboptimal choice in general as the nucleus would suffer from a severe loss of specificity with regards to the source of Ca 2þ and the nature of the initial membrane signal. On the other hand, using a highly processed nuclear signal like the transcription factor itself (as in NFAT translocation) allows for a high level of signaling specificity, but in placing the bulk of the signal amplification and processing outside of nucleus more pressure is placed on nuclear transport mechanisms. As electrophysiologists know, it is unwise to amplify a signal before passing it through a channel that can be saturated, as it could result in serious distortion of the signal. Activation of DREAM and NFATc4 define the extremes of information transfer from the plasma membrane to the nucleus, whereas Ca 2þ /camp responsive element binding protein (CREB) signaling mechanisms typically exemplify the middle ground. In this case, the translocating message is not conveyed by free Ca 2þ moving from the ion channels to the nucleus, or by the transcription factor CREB, but by intermediate players in the pathway. In fact, CREB remains constitutively bound to CRE (Ca 2þ /camp responsive element) regulatory sites controlling target genes, and becomes activated by phosphorylation on its Ser-133 residue (with modulatory influences exerted by other residues). A fast CaM-dependent kinase cascade and a slow MAP kinase cascade (likely to be camp-modulated) converge on Ser-133 following surface membrane depolarization and Ca 2þ influx, and work together to promote CREB-dependent gene expression [3,13]. How are these two signals transduced? It is well established that in response to physiological synaptic activity in hippocampal pyramidal neurons, Ca 2þ entry through both N-methyl-D-aspartate receptors (NMDARs) and L-type Ca 2þ channels initiates the nuclear signaling to CREB [14 16]. Nevertheless, Ca 2þ ions can be excluded as the (Figure 1 Legend) Signaling from the membrane to the nucleus: multiple strategies for information transfer. (a) The transcription factors NFATc4, CREB and DREAM are all activated following increases in intracellular Ca 2þ, yet each relies upon a different mode of information transfer. At rest, NFATc4 is localized to the cytosol, allowing the transcription factor to be targeted to specific regions of the cell. This would allow for heightened signaling specificity. Indeed, Ca 2þ entry through L-type calcium channels activates this transcription factor preferentially relative to other voltagegated Ca 2þ channels. Yet, this mode of synapse-to-nucleus signaling is limited in both speed and signal amplification. Conversely, DREAM can be activated by general rises in intracellular Ca 2þ, allowing for rapid information transfer. However, this pathway is limited in regards to signal specificity. CREB activation is typically initiated by nuclear translocation of the calcium sensor CaM or by an activating kinase, providing a combination of the advantages found for both NFATc4 and DREAM signaling. (b) A snapshot image of the early stages (0 5 min) of activity-dependent gene expression. Following rises in intracellular Ca 2þ, DREAM dissociates from DNA resulting in the lifting of transcriptional repression. Within seconds of Ca 2þ entry through L-type Ca 2þ channels and NMDA receptors, CaM translocates to the nucleus, supporting CREB phosphorylation through activation of CaMKIV. Nearly as rapid, NFATc4 also undergoes translocation to the nucleus following its dephosphorylation by calcineurin (CaN). Other pathways outlined in (a), including the MAPK/PKA pathways, subsequently begin to exert their influence. Abbreviations, VGCC, voltage-gated calcium channel.

3 Signaling from synapse to nucleus Deisseroth, Mermelstein, Xia and Tsien 3 Figure 1 (a) Information transfer to the nucleus SURFACE/CYTOSOLIC EVENTS Ca2+ rise CaM Ca2+ rise CaM Ca2+ rise CaM Ca2+ rise CaM CaN NUCLEAR EVENTS Interaction with NFATn: NFAT-transcription NFATc4 CaMKK/ CaMKIV CREB phosphorylation MAPK/ Rsk AC CREB phosphorylation CREB phosphorylation PKA Ca2+ rise Binding to DREAM (b) Interaction with CBP: CRE-transcription Removal from DNA: DREAM repression lifted Early stages of activity-dependent signaling to the nucleus NFATc4 Ca2+ CaN L-type VGCC NFATc4 MAPK CaMKIV PKA CREB AC CaM DREAM NMDAR DREAM Current Opinion in Neurobiology

4 4 Signalling mechanisms typical nuclear messenger in these cells as CREB activation at Ser-133 shows marked Ca 2þ channel-type specificity. Working alongside Ca 2þ stores, many Ca 2þ influx channels (including L-type, N-type, and P/Q-type Ca 2þ channels as well as NMDA- and AMPA-type glutamate receptors) can cause a marked elevation in the bulk cytoplasmic and nuclear Ca 2þ levels, and can play important roles in the neuron [17]. However, most of these are ineffective at directly activating CREB, with the exception of the L-type channel (privileged among the voltagegated channels [15,16,18 ]) and the NMDA receptor [14 16]. Therefore, although nuclear Ca 2þ elevations are seen during relatively strong neuronal activity [9] and may support the signaling in some cases [19,20], nuclear Ca 2þ is not typically sufficient to activate CREB [15,18 ]. Furthermore, it appears that nuclear Ca 2þ elevations are not necessary, as suppression of global cytoplasmic and nuclear Ca 2þ elevations with intracellular loading of the Ca 2þ chelator EGTA generally does not block Ca 2þ - activated CREB Ser-133 activation [4,14,21 ]. These findings, together with results that demonstrate micrometer-scale localization of L-type Ca 2þ channels within the plasma membrane [22 ] and a CaM-binding motif on the L-type channel itself [18 ] are both crucial for CREB phosphorylation, have revealed that Ca 2þ typically exerts its effects on nuclear CREB activation by acting within a domain <1 mm in radius from its source of entry [14]. The privileged roles of the NMDA receptor and the L- type channel [14 16] may be attributable to their selective ability to recruit CaM signaling [15]. While both channel types can directly bind CaM, both are also specifically localized at or near synapses in which additional pools of CaM are present for recruitment. Either or both mechanisms may underlie the selective ability of these channels to initiate locally Ca 2þ /CaM-activated MAP kinase signaling to the nucleus [13,18,21,23] and nuclear translocation of CaM [15,20,24,25 ]. The translocation of CaM to the nucleus is dependent upon its ability to bind to a CaM kinase [20,25,26]. This binding would stabilize Ca 2þ CaM, although locally mobilized CaM could also subsequently make use of Ca 2þ from other sources [19,20], perhaps from nuclear Ca 2þ stores or even cytoplasmic Ca 2þ waves. Use of supportive Ca 2þ in this way would still preserve the specificity of information carried by the selective synaptic activation of the NMDA receptor and the L-type channel, as only these channel types would be able to mobilize the CaM necessary to utilize efficiently the available Ca 2þ. Despite the high concentrations of detectable CaM in all cellular compartments, free or available CaM is generally limiting in intracellular biochemical reactions [27]. The resulting increase in available nuclear CaM results in strong activation of nuclear CaM-dependent protein kinase kinase (CaMKK) and its target CaMKIV. These kinases are ultimately responsible for the early wave of Ser-133 CREB phosphorylation [13,28,29,30,31]. With stronger stimulation, the slower MAPK-dependent signaling pathway comes into play [13]. In this case, Ca 2þ /CaM leads to local activation of the small GTP binding protein Ras, stimulation and nuclear translocation of mitogen-activated protein kinase (MAPK) and/or pp90 ribosomal protein S6 kinase (Rsk) [3,18,32], in some cases with cooperation of protein kinase A (PKA). In summary, both CaMK and MAPK pathways to CREB generally use a downstream Ca 2þ effector as the translocating nuclear messenger. This thereby preserves information about the specific Ca 2þ source, while also leaving ample room for signal amplification in the nucleus a combination of features that have advantages over the more extreme scenarios. Clearly then, the transfer of information from the membrane surface of a neuron into the nucleus can be achieved by physical translocation at various points in the signal transduction pathway. Understanding the choice of strategy and the associated information processing trade-offs can best be attempted in the context of the functional output of the pathway. Therefore, next we will review recent work in intact animals describing the disruption of nuclear signaling pathways and their high-level behavioral consequences: that is, generation of circadian rhythms, formation of long-term fear memories, and survival of brain cells. Circadian rhythms The generation and entrainment of circadian rhythms, of nearly universal importance in biology, provides an outstanding example of the functional importance of synapse-to-nucleus signaling. Many aspects of circadian rhythm generation are hardwired, in that they occur without specifically requiring neuronal activity, and arise in a network of transcription factors that generate rhythmicity from negative feedback signaling. In the suprachiasmatic nucleus (SCN) of the mammalian hypothalamus, for example (Figure 2), multiple proteins are involved in the clockwork mechanism, including dimers involving the transcription factors of CLOCK and BMAL, and the period gene products mper1, mper2 and mper3. This endogenous rhythm is robust but not precise, however, and numerous other circumstances, including seasonal and migratory changes in patterns of light and dark, demand that the cycle be entrainable. In fact, exposure to light before the end of a subjective night, a signal indicating that the rhythm needs adjustment, triggers rapid induction of the genes for mper1 (within min) and mper2 (within 2 h), leading to an efficient readjustment of the circadian rhythm. How are these clock genes activated? In addition to E-box elements (regulated by the CLOCK/BMAL system), both mper1 and mper2 have activity-sensitive CRE promoter elements that are regulated by the CRE-binding protein CREB [33]. Evidence is mounting that CREBmediated control of these genes referees circadian rhythm

5 Signaling from synapse to nucleus Deisseroth, Mermelstein, Xia and Tsien 5 Figure 2 Output to SCN NMDAR Input from retina SCN neuron Synaptic transmission from ganglion cells L-type VGCC CRY mper1 TIME SCN ACTIVITY RESULT EFFECT Z Z Minimal NMDAR, Minimal L-type VGCC mper1 activity expression Z Z Z Simple spikes: pacemaker activity Activation of NMDAR and L-type VGCC: CREB phosph Strong mper1 expression INTERNAL CLOCK S133 P S142 P CLK BMAL CREB CRE mper1 E-Box Complex spikes: APs and EPSPs Current Opinion in Neurobiology Activity-dependent regulation of circadian rhythms. Light-induced activation of retinal ganglion cells results in the excitation of SCN neurons. The increases in synaptic neurotransmission result in the opening of L-type calcium channels and NMDA receptors, directly leading to CREB phosphorylation and the transcription of several clock genes including mper1. These changes in gene expression result in the entrainment of the internal clock to the external environment. cycling and entrainment. First, CREB activation by phosphorylation of the Ser-133 residue follows an active circadian rhythm (peaking near subjective dawn) and can be directly triggered by entraining stimuli, such as photic stimulation during darkness [16]. Second, expression of the CREB-dependent genes c-fos, mper1 and mper2 correlates with CREB activation, and is blocked when a key modulatory site (Ser-142) on CREB is mutated [34 ]. Third and finally, this CREB mutation attenuates entrainment of the circadian oscillator with respect to either phase advances or phase delays [34 ]. Taken together, these data show that activity-dependent control of a nuclear transcription factor (CREB) mediates a very important complex behavior (entrainment of circadian rhythm). What are the key design constraints on this activitydependent signaling pathway? First, to be useful, the pathway should be fast by comparison with the circadian cycle (minutes), although signaling on a scale of seconds may be unnecessary. It need not resolve closely spaced stimuli (after all, closely spaced flashes of daylight were hardly a factor in circadian evolution!). Second, in responding to electrophysiological events, the pathway would ideally be sensitive to synaptic input from afferents that carry sensory information from the retina, but not to the continually free-running spiking of SCN neurons, which is intrinsic even in isolated neurons. Third, as non-photic inputs such as feeding or locomotor activity modulate entrainment in some cases, the pathway must allow for the convergence of multiple stimuli, not just retinal glutamatergic afferents. Every one of these design requirements seems to be fulfilled by aspects of synapse-to-nucleus signaling in SCN neurons. First, phosphorylation of CREB Ser-133 does occur within about 10 min of light flashes or glutamatergic stimulation of SCN neurons, followed by mper1

6 6 Signalling mechanisms induction within min [34 ]. This suggests the involvement of either the fast CaMKIV signaling pathway leading to phosphorylation of CREB on Ser-133 or the slower CaM kinase/map kinase-dependent pathway, both of which can be efficacious at 10 min [13]. In fact, inhibitors of either CaM kinase or MAP kinase block CREB activation in SCN neurons [35]; in the future, circadian rhythmicity and fast signaling to CREB might also be studied in the SCN of CaMKIV knockout mice. Second, with regards to selective coupling to particular electrical events at the membrane, certain pathways of nuclear signaling in hypothalamic neurons have been found to respond relatively selectively to membrane potentials generated by synaptic input in preference to action potentials alone [36], as in various ganglionic neurons [37,38] and hippocampal pyramidal neurons [14,39]. It has become increasingly clear how gene expression avoids indiscriminate activation by the continuous tonic spiking of these cells. CREB signaling in hypothalamic SCN neurons responds to retinal glutamatergic afferents through a pathway initiated by NMDAtype glutamate receptors and voltage-gated L-type Ca 2þ channels. Activation of the NMDARs requires synaptic glutamate, and therefore needs more than just spikes to occur. Among the many voltage-gated Ca 2þ channels, only the L-type channel has been found to couple to mper1 [40]. If this remains true when the contributions of the other channels are rigorously examined [15], it suggests that there is a possibility that L-type channels help to filter out brief single action potentials by virtue of their voltage- and time-dependent kinetics [39]. This selectivity would probably not be possible if the nuclear message were conveyed by Ca 2þ ions for example, as strongly spike-responsive Ca 2þ channels (N-type or P/Qtype) also generate large increases in bulk cytoplasmic and nuclear Ca 2þ. The use of CaM as a synapse-tonucleus messenger [15] would avoid this difficulty, a possibility that needs to be tested in SCN neurons. Third, signaling in the SCN appears well equipped to integrate modulatory input from non-photic CNS systems, including that of neuropeptide-y (NPY; provided by afferents from the thalamus), serotonin, and pituitary adenylyl cyclase activating peptide (PACAP). All three systems appear to mediate behavioral influences on circadian rhythmicity, and are known to use G protein signaling to the camp/pka pathway as a means of exerting control over intracellular events (PACAP positively, NPY negatively, and serotonin being capable of both). Indeed, application of forskolin, a direct activator of the camp/ PKA pathway, promotes mper1 expression in SCN neurons [41,42], and expression of dominant negative PKA inhibits CRE-dependent gene expression in SCN neurons [41]. PKA could influence nuclear signaling in several different ways, for example, by direct phosphorylation of CREB residue Ser-133, by stimulating a CREB transcriptional coactivator such as CBP or p300 [43], or by potentiating MAPK signaling [23]. Equally intriguing is the possibility that PKA-mediated signaling could act through direct modulation of ion channels, thereby regulating the ability of L-type Ca 2þ channels to respond to the tonic electrical activity of SCN neurons. This would support the convergence of diverse inputs and allow a non-photic input to recruit exactly the same signaling pathway as a light stimulus, even without retinally driven synaptic input. Indeed, it is known that the ability of L-type channels to filter out action potentials is conditional, in that direct potentiation of ion flux restores some responsiveness to brief depolarizations [39]. It is also likely that spike width broadening or highfrequency bursts of action potentials would be able to conditionally overcome this filter [44]. To test the validity of this idea in SCN circadian rhythmicity, it would be important to conduct electrophysiological studies of SCN action potentials and the resulting L-type currents at 378C [45]. Testing at this temperature would avoid the opposing effects on spike width and channel gating kinetics that are present under even slightly lower temperatures, and under a range of waveform and modulation conditions. The potential modulation of action potential filtering could also be tested in vivo, by using L-type antagonists and potentiators to see if direct manipulation of L-type channels alters the behavioral responses to nonphotic inputs through NPY, serotonin or PACAP. Learning and memory exemplified by fear conditioning Memory formation is another vital high-level behavioral function that depends on nuclear events [2,46,47]. Again, knowledge of the behavioral output of the nuclear event can greatly illuminate our understanding of the logic that underlies the nuclear signaling. Much attention has been directed towards the hippocampus, which not only supports explicit memory formation in tasks like spatial learning but also contributes crucially to contextual fear conditioning, in which the animal learns to associate a complex, generally benign environment with an aversive stimulus (Figure 3). The amygdala mediates cued fear memories, formed when an aversive stimulus is associated with a single stimulus, such as an auditory tone. In both cases, subsequent presentation of the harmless stimulus alone will trigger fear behavior such as freezing. Neurons in both the amygdala and the hippocampus display robust activation of CREB during these learning tasks [25,48], and in both structures, genetic interference with CREB function prevents storage of long-term fear memories [49,50 ]. Additionally, in either one or the other region, memory deficits associated with reduced CREB function can be partially overcome by spaced repeated training, whereas high-level CREB function can allow long-term memories to form with single training sessions [50,51,52 ]. Given these similarities, several predictions can be made. First, that the hippocampus and amygdala would employ

7 Signaling from synapse to nucleus Deisseroth, Mermelstein, Xia and Tsien 7 Figure 3 HIPPOCAMPUS AMYGDALA Contextual information Cue information + Pain/fear Stimulus One or massed exposure One exposure under intense conditions Multiple spaced exposures Proper conclusion Spurious or rare, possibly an unimportant association Possibly spurious or rare but too important to ignore Clearly related and association should be learned Response Mild CaM/CaMKIV-mediated CREB phosphorylation Strong CaM/CaMKIV CREB phosphorylation Support from MAPK/PKA genetic or pharmacological alteration in CREB signaling Strong CaM/CaMKIV CREB phosphorylation Support from MAPK/PKA Inactivation of CaN Facilitation of CaM signaling Result No CREB-mediated long-term memory CREB-mediated long-term memory Current Opinion in Neurobiology Activity-dependent fear conditioning within the hippocampus and the amygdala. The pairing of contextual and cue information with an adverse stimulus results in heightened CREB activation within the hippocampus and the amygdala. Dependent upon the stimulus train, one or more CREB signaling pathways are recruited to mediate long-term memory. With either genetic or pharmacological manipulation of the CREB signaling pathways, changes in learning and memory can be achieved. a common molecular mechanism of signaling to CREB. Second, that CREB should respond strongly to synaptic input that bears incoming processed sensory information in both cases. Third, that activity-dependent signaling events that lead to CREB activation should display considerable speed and temporal resolution, as temporal relationships among events are important in elucidating their causal relationship. Fourth and finally, that stimuli that are spaced and repeated, those most appropriate for associations by the brain should be particularly advantaged in activating CREB by comparison with single stimulus bouts. In the hippocampus, CREB signaling triggered by physiological electrical activity is indeed selectively dependent on Ca 2þ entry through L-type calcium channels and

8 8 Signalling mechanisms NMDA receptors [14 16]. This pattern of source-specificity confers selective responsiveness to synaptic activity for the reasons discussed above. The role of these pathways for Ca 2þ entry in amygdalar CREB activation has not been tested yet, but L-type Ca 2þ channels and NMDA receptors are both required for amygdalar formation of long-term fear memories [53,54], which are CREB-dependent. In both the hippocampus and the amygdala neuronal activity causes rapid nuclear translocation of CaM and CaMKIV-dependent CREB phosphorylation [15,25,28,29,30], further reinforcing the similarities in CREB signaling. Furthermore, it has been demonstrated in both structures that the CaM/CaMKIV pathway is required for establishing CREB-dependent fear memories [25,29 ]. As noted above, the employment of a downstream effector of Ca 2þ as a nuclear messenger, rather than free Ca 2þ itself, facilitates Ca 2þ source discrimination by the nucleus [2,5,15,18 ]. The use of CaM as the messenger, which is capable of significant nuclear translocation within 15 s [20], sacrifices little in the way of speed. Interestingly, the CaM nuclear translocation pathway, widely employed by non-neuronal cell types as well as neurons [15,24,26,27,55 58], is now also known to be almost ubiquitous throughout the mammalian brain, including hippocampus, amygdala, somatosensory cortex, insular cortex, anterior cingulate cortex and cerebellum [15,20,25 ]. However, an exception has been found, three groups have determined that there is little CaM translocation in a substructure of the hippocampal formation called the dentate gyrus [19,20,25 ]. Although the dentate gyrus is anatomically positioned as the gateway to the hippocampus proper, its precise role is unknown. Dentate granule neurons differ from hippocampal pyramidal cells in many ways, including their characteristics of development, regeneration, morphology and synaptic plasticity. It would be interesting to find out if the role of these neurons at the systems level were somehow linked to unusual features of their subcellular signaling. What are the advantages of this pathway? Use of the CaM/CaMKIV pathway to CREB lends itself well not just to speed but also to implementing repetitive spaced stimulus responsiveness. Brief mild stimulation of hippocampal neurons that selectively activates the calmodulin/camkiv pathway to CREB [13], alsocreatesa primed state in which subsequent similar stimuli are more potent and even faster in their effects on nuclear CREB [20]. The increased speed of CREB formation suggests that the priming effect might be attributable to the nuclear use of persistently translocated CaM. After nuclear translocation, return of CaM to the cytoplasm proceeds gradually over about 1 hour [15]. This extended time course fits reasonably well with behavioral data on spaced stimuli [51,52 ]. As noted earlier, CREB responds to multiple converging signaling pathways in hippocampal neurons, including a slower but more prolonged MAPK-mediated activation pathway that overlaps with the calmodulin/camkiv pathway [3 5,13], and a calcineurin/protein phosphatase-1-mediated CREB deactivation pathway [31,59]. Flexible association of aversive stimuli with a wide range of environmental situations of different importance and temporal structure may require such complexity of signaling. The MAPK pathway to CREB has poorer temporal resolution than the CAMK pathway (slower in onset, and more persistent), but it has several interesting properties, including further prolongation by spaced repeated stimulation [23] or PKA activation [23,32,60], and selective responsiveness to stronger stimulation [13]. Indeed, activation of the MAPK pathway can be involved in forming both amygdalar [61] and hippocampal [62] fear memories. Similarly, the calcineurin/pp1 pathway responds to repeated stimuli by potentiating CREB function in this case through activity-dependent inactivation of calcineurin [31,63]. Notably, direct calcineurin inhibition potentiates memory formation in the amygdala [64 ] and the hippocampus [65 ]. Taken together, these data collected in the amygdala and the hippocampus reveal a clear pattern stimuli of greater behavioral significance are endowed with a more potent means of activating CREB. In this manner, the nuclear computer capitalizes on the diverse characteristics of its many incoming signaling pathways, including sensitivity and temporal resolution, to calculate the statistical significance of candidate associations for storage in long-term memory. Control of neuronal survival and death Neurons receive specific instructions to survive or to die, the correct choice being crucial for development and mature functioning of the brain. The decision to die is particularly interesting, as neurons reach this all-or-none choice after weighing a wide array of external inputs, including synaptic events. This is reminiscent of faster electrophysiological choices made between quiescence and firing an action potential. The nucleus plays a key role in the survival/death computation, largely through activity-dependent regulation of transcription factors such as CREB, MEF2 and NF-kB. An emerging theme is the highly conditional influence of these transcription factors, in that they promote survival in some circumstances but not in others. It has been known for years that stimulation of electrical activity at mild to rapid rates (for example, 5Hz to 50Hz electrical stimuli) works through both NMDA receptors and L-type Ca 2þ channels to activate nuclear CREB [14,15]. Accordingly, activation of CREB would not be expected to promote cell death in the normal brain. Indeed, CREB generally favors neuronal survival. Postnatal

9 Signaling from synapse to nucleus Deisseroth, Mermelstein, Xia and Tsien 9 genetic inactivation of CREB (along with camp response element modulator [CREM]) causes widespread apoptosis of CNS neurons [66]. Other studies have shown that a low level of activation of NMDA receptors or L-type Ca 2þ channels promotes survival of cerebellar granule neurons, presumably through the activity of CaMKIV and CREB [67]. CREB also mediates the survival-promoting effects of brain-derived neurotrophic factor (BDNF) [68,69] and other trophic factors. On the other hand, excitatory stimuli that are known to be highly toxic to mature neurons, such as steady prolonged application of glutamate, promote general cellular dysfunction and death [70 72]. Furthermore, it has been suggested that under conditions of very strong and long excitation, CREB can actually promote apoptosis by acting through a nuclear p53 cascade [73]. Perhaps this action of CREB is aimed at eliminating neurons whose function is deranged by excessive stimulation, thus maintaining the integrity of CNS function. This may be reminiscent of the other effects of p53, guardian of the genome, which prevents mitosis and potential tumorigenesis in the setting of DNA damage. Transcription factors from the MEF2 family [74], especially MEF2A, -C and -D, also appear to promote survival of cerebellar granule cells and cerebral cortical neurons in response to NMDA application or mild depolarization [75 77]. MEF2 function can be elevated directly by calcineurin and indirectly by CaMKIV [74], once again underscoring the importance of CaM signaling to the nucleus [15,31]. Additional activity-dependent signals provided by p38 MAP kinases and NFAT converge to promote MEF2 function [74]. However, a pro-apoptotic role for modified MEF2 proteins has also been identified [75,77]. Cleavage of these transcription factors follows activation of caspases, generating MEF2 fragments capable of DNA binding but lacking transactivation potential. These dominant-negative factors then strongly promote apoptosis. Thus, as in the situation with CREB, the influence of the transcription factor in favoring or disfavoring survival depends on the cellular context. The NF-kB family of transcription factors is strongly expressed in the CNS, especially in the hippocampus and the cerebellum. They are activated by growth factors and cytokines as well as by depolarization or glutamate application, which leads to nuclear translocation and in general, promotion of neuronal survival [78]. However, several studies have identified a pro-apoptotic role for NF-kB. Though considerable controversy remains, the net effect of NF-kB appears to depend on the type of stimulation, the degree of NF-kB activation, and the particular NF-kB subunits that are expressed. Interestingly, beyond simple nuclear translocation [6] of NF-kB [79] (Figure 1), the p65 subunit of NF-kB can be further activated by phosphorylation at Ser-276, a residue whose amino acid sequence environment is nearly identical to that of CREB Ser-133. It has been proposed [79] that the site can support modulation of NF-kB function by CaM- KIV or Rsk kinases, leading to disparate effects on survival or death that depend on the profile of kinase activation. Here again, as with CREB and MEF2, we see an emerging role of nuclear computation in the integration of rich streams of incoming information to reach a crucial decision between cell survival and cell death. Concluding remarks and future prospects We have deliberately focused on mechanisms that convey an initial message from the neuronal surface to a nuclear transcription factor, but in which the final outcome in gene expression also depends on multiple additional signaling events, as illustrated below for the cases of CREB- and NFAT-dependent transcription. In the case of CREB, attainment of the Ser-133 activated state is further modulated by additional activity-dependent events, including calcineurin-regulated dephosphorylation of Ser-133 as well as the phosphorylation of Ser- 142 discussed above. Stabilization of the CREB transcriptional complex is likely to be further affected by activitydependent regulation of co-activators, including the CREB binding protein (CBP) and possibly the closely related p300 [43]. With respect to activation of NFATdependent transcription, required co-activators include activator protein-1 (AP1) (composed partly by the CREBregulated factor c-fos) and MEF2 [80], which is CaMregulated in its own right as noted above. Neuronal activity not only triggers nuclear translocation of NFATc4 but it additionally regulates NFATc4-dependent transcription by delaying nuclear export of NFAT through L- type channel mediated inhibition of glycogen synthase kinase 3b [8]. The result is prolonged NFATc4 activation and increased NFAT-dependent gene expression [8]; this is analogous to CRE-mediated gene expression when CREB activity is extended by activity-dependent inhibition of CREB dephosphorylation [31,63]. Judging by these examples of signaling complexity, we have only seen the tip of the iceberg with respect to the convergence or apposition of multiple mechanisms. Intricate networks of downstream genes represent the output of nuclear processing. A large number of genes show activity modulation in neurons, and many of these have plausible behavioral roles. Within hippocampal neurons, activation of NFATc4 leads to an upregulation of inositol 1,4,5-triphosphate receptor type 1 (IP3R1) [8] and BDNF [81]; both proteins are known to influence learning and memory [82 84]. CREB likewise drives BDNF expression [69] as well as c-fos and other factors implicated in neuronal survival and memory formation. In circadian rhythm generation, the CREB-regulated mper genes are well-characterized targets of retinal activity and have clear behavioral roles. But exactly how the wide array of activity-regulated genes generate meaningful

10 10 Signalling mechanisms changes in the brain s neural networks is still not understood. Though many aspects of nuclear computation remain shrouded in mystery, it is clear that a combination of in vitro analysis and behavioral investigation can begin to illuminate the rationale for the unique biochemical complexity of neurons, and the elegant logic behind the design of nuclear signaling pathways. Acknowledgements RW Tsien is supported by grants from the National Institute of General Medical Sciences (58234), National Institute of Neurological Disorders and Stroke (24067) and National Institute of Mental Health (48108). PG Mermelstein is supported by a grant from the National Institute of Neurological Disorders and Stroke (41302), and a grant from the Whitehall Foundation. K Deisseroth and PG Mermelstein contributed equally to the manuscript. References and recommended reading Papers of particular interest, published within the annual period of review, have been highlighted as: of special interest of outstanding interest 1. Bito H, Deisseroth K, Tsien RW: Ca 2þ -dependent regulation in neuronal gene expression. Curr Opin Neurobiol 1997, 7: Bailey CH, Bartsch D, Kandel ER: Toward a molecular definition of long-term memory storage. Proc Natl Acad Sci USA 1996, 93: West AE, Griffith EC, Greenberg ME: Regulation of transcription factors by neuronal activity. Nat Rev Neurosci 2002, 3: Bradley J, Finkbeiner S: An evaluation of specificity in activitydependent gene expression in neurons. Prog Neurobiol 2002, 67: Lonze BE, Ginty DD: Function and regulation of CREB family transcription factors in the nervous system. Neuron 2002, 35: Morgan JI, Curran T: Role of ion flux in the control of c-fos expression. Nature 1986, 322: Crabtree GR, Olson EN: NFAT signaling: choreographing the social lives of cells. Cell 2002, 109:S67-S Graef IA, Mermelstein PG, Stankunas K, Neilson JR, Deisseroth K, Tsien RW, Crabtree GR: L-type calcium channels and GSK-3 regulate the activity of NF-ATc4 in hippocampal neurons. Nature 1999, 401: Lipscombe D, Madison DV, Poenie M, Reuter H, Tsien RW, Tsien RY: Imaging of cytosolic Ca 2þ transients arising from Ca 2þ stores and Ca 2þ channels in sympathetic neurons. Neuron 1988, 1: Carrion AM, Link WA, Ledo F, Mellstrom B, Naranjo JR: DREAM is aca 2þ -regulated transcriptional repressor. Nature 1999, 398: Osawa M, Tong KI, Lilliehook C, Wasco W, Buxbaum JD, Cheng HY, Penninger JM, Ikura M, Ames JB: Calcium-regulated DNA binding and oligomerization of the neuronal calciumsensing protein, calsenilin/dream/kchip3. J Biol Chem 2001, 276: Osawa et al. [11 ] and Cheng et al. [12 ] characterized key aspects of DREAM [10], a nuclear and cytoplasmic protein that can bind Ca 2þ ions (K d 14 mm). Though nuclear Ca 2þ levels probably do not reach high micromolar concentrations, one of the functions of DREAM is to act as a transcription factor, and it has also been hypothesized to act as a direct target of nuclear Ca 2þ ions. Early demonstrations of the other two main mechanistic paradigms in Ca 2þ -dependent nuclear signaling in neurons were obtained for translocation of the transcription factor itself [8] and translocation of an intermediate messenger [15]. 12. Cheng HY, Pitcher GM, Laviolette SR, Whishaw IQ, Tong KI, Kockeritz LK, Wada T, Joza NA, Crackower M, Goncalves J et al.: DREAM is a critical transcriptional repressor for pain modulation. Cell 2002, 108: See annotation to Osawa et al. [11 ]. 13. Wu GY, Deisseroth K, Tsien RW: Activity-dependent CREB phosphorylation: convergence of a fast, sensitive calmodulin kinase pathway and a slow, less sensitive mitogen- activated protein kinase pathway. Proc Natl Acad Sci USA 2001, 98: Deisseroth K, Bito H, Tsien RW: Signaling from synapse to nucleus: postsynaptic CREB phosphorylation during multiple forms of hippocampal synaptic plasticity. Neuron 1996, 16: Deisseroth K, Heist EK, Tsien RW: Translocation of calmodulin to the nucleus supports CREB phosphorylation in hippocampal neurons. Nature 1998, 392: Ginty DD, Kornhauser JM, Thompson MA, Bading H, Mayo KE, Takahashi JS, Greenberg ME: Regulation of CREB phosphorylation in the suprachiasmatic nucleus by light and a circadian clock. Science 1993, 260: Snutch TP, Sutton KG, Zamponi GW: Voltage-dependent calcium channels beyond dihydropyridine antagonists. Curr Opin Pharmacol 2001, 1: Dolmetsch RE, Pajvani U, Fife K, Spotts JM, Greenberg ME: Signaling to the nucleus by an L-type calcium channelcalmodulin complex through the MAP kinase pathway. Science 2001, 294: The studies by Deisseroth et al., Dolmetsch et al., Hardingham et al. and Weick et al. [14,18,21,22 ] all demonstrate that Ca 2þ typically acts locally, near the channels in the surface membrane, to initiate nuclear signaling. This result was first demonstrated [14] using a combination of fast and slow intracellular Ca 2þ buffers to show that Ca 2þ can act within about 1 mm of the channel source to activate nuclear CREB. Similar results were later obtained focusing on the slow pathway to CREB [21 ]. It subsequently became clear that a CaM-binding site directly on the L-type channel helps to confer the privileged ability of the L-type channel to signal to nuclear CREB [18 ], and that micron-scale localization of the L-type channel, and therefore micron-scale local signaling, is likewise crucial for signaling to CREB [22 ]. 19. Hardingham GE, Arnold FJ, Bading H: Nuclear calcium signaling controls CREB-mediated gene expression triggered by synaptic activity. Nat Neurosci 2001, 4: Mermelstein PG, Deisseroth K, Dasgupta N, Isaksen AL, Tsien RW: Calmodulin priming: nuclear translocation of a calmodulin complex and the memory of prior neuronal activity. Proc Natl Acad Sci USA 2001, 98: Hardingham GE, Arnold FJ, Bading H: A calcium microdomain near NMDA receptors: on switch for ERK-dependent synapseto-nucleus communication. Nat Neurosci 2001, 4: See annotation for Dolmetsch et al. [18 ]. 22. Weick JP, Groth RD, Isaksen AL, Mermelstein PG: Interactions with PDZ proteins are required for L-type calcium channels to activate camp response element binding protein-dependent gene expression. J Neurosci 2003, 23: See annotation for Dolmetsch et al. [18 ]. 23. Wu GY, Deisseroth K, Tsien RW: Spaced stimuli stabilize MAPK pathway activation and its effects on dendritic morphology. Nat Neurosci 2001, 4: Wang D, Tolbert LM, Carlson KW, Sadee W: Nuclear Ca 2þ / calmodulin translocation activated by mu-opioid (OP3) receptor. J Neurochem 2000, 74: Wei F, Qiu CS, Liauw J, Robinson DA, Ho N, Chatila T, Zhuo M: Calcium calmodulin-dependent protein kinase IV is required for fear memory. Nat Neurosci 2002, 5: The studies by Wei et al. and Kang et al. [25,29 ] demonstrated the crucial importance of the calmodulin CaMKIV pathway to CREB in memory formation in the amygdala and hippocampus. 26. Liao B, Paschal BM, Luby-Phelps K: Mechanism of Ca 2þ - dependent nuclear accumulation of calmodulin. Proc Natl Acad Sci USA 1999, 96: Teruel MN, Chen W, Persechini A, Meyer T: Differential codes for free Ca 2þ -calmodulin signals in nucleus and cytosol. Curr Biol 2000, 10:86-94.

11 Signaling from synapse to nucleus Deisseroth, Mermelstein, Xia and Tsien Ribar TJ, Rodriguiz RM, Khiroug L, Wetsel WC, Augustine GJ, Means AR: Cerebellar defects in Ca 2þ /calmodulin kinase IVdeficient mice. J Neurosci 2000, 20:RC107: Kang H, Sun LD, Atkins CM, Soderling TR, Wilson MA, Tonegawa S: An important role of neural activity-dependent CaMKIV signaling in the consolidation of long-term memory. Cell 2001, 106: See annotation for Wei et al. [25 ]. 30. Ho N, Liauw JA, Blaeser F, Wei F, Hanissian S, Muglia LM, Wozniak DF, Nardi A, Arvin KL, Holtzman DM et al.: Impaired synaptic plasticity and camp response element-binding protein activation in Ca 2þ /calmodulin-dependent protein kinase type IV/Gr- deficient mice. J Neurosci 2000, 20: Bito H, Deisseroth K, Tsien RW: CREB phosphorylation and dephosphorylation: a Ca 2þ - and stimulus duration-dependent switch for hippocampal gene expression. Cell 1996, 87: Impey S, Obrietan K, Wong ST, Poser S, Yano S, Wayman G, Deloulme JC, Chan G, Storm DR: Cross talk between ERK and PKA is required for Ca 2þ stimulation of CREB- dependent transcription and ERK nuclear translocation. Neuron 1998, 21: King DP, Takahashi JS: Molecular genetics of circadian rhythms in mammals. Annu Rev Neurosci 2000, 23: Gau D, Lemberger T, von Gall C, Kretz O, Le Minh N, Gass P, Schmid W, Schibler U, Korf HW, Schutz G: Phosphorylation of CREB Ser142 regulates light-induced phase shifts of the circadian clock. Neuron 2002, 34: This paper provided the first direct causal evidence linking CREB modulation to the important behavioral process of circadian rhythm entrainment. A crucial aspect of this link may lie in channel biophysics. Luckman et al. [36], Mermelstein et al. [39] and Deisseroth et al. [14] demonstrated that in hypothalamic and hippocampal neurons, nuclear signaling can filter out action potentials and respond selectively to synaptic potentials. A biophysical explanation was provided [39] linked to the activation gating kinetics of L-type calcium channels, a property that may be crucial in the behavioral roles of CREB (Figure 2). 35. Schurov IL, Hepworth TJ, Hastings MH: Dopaminergic signalling in the rodent neonatal suprachiasmatic nucleus identifies a role for protein kinase A and mitogen-activated protein kinase in circadian entrainment. Eur J Neurosci 2002, 15: Luckman SM, Dyball RE, Leng G: Induction of c-fos expression in hypothalamic magnocellular neurons requires synaptic activation and not simply increased spike activity. J Neurosci 1994, 14: Chalazonitis A, Zigmond RE: Effects of synaptic and antidromic stimulation on tyrosine hydroxylase activity in the rat superior cervical ganglion. J Physiol 1980, 300: Brosenitsch TA, Salgado-Commissariat D, Kunze DL, Katz DM: A role for L-type calcium channels in developmental regulation of transmitter phenotype in primary sensory neurons. J Neurosci 1998, 18: Mermelstein PG, Bito H, Deisseroth K, Tsien RW: Critical dependence of camp response element-binding protein phosphorylation on L-type calcium channels supports a selective response to EPSPs in preference to action potentials. J Neurosci 2000, 20: Akiyama M, Minami Y, Nakajima T, Moriya T, Shibata S: Calcium and pituitary adenylate cyclase-activating polypeptide induced expression of circadian clock gene mper1 in the mouse cerebellar granule cell culture. J Neurochem 2001, 78: Obrietan K, Impey S, Smith D, Athos J, Storm DR: Circadian regulation of camp response element-mediated gene expression in the suprachiasmatic nuclei. J Biol Chem 1999, 274: Travnickova-Bendova Z, Cermakian N, Reppert SM, Sassone- Corsi P: Bimodal regulation of mperiod promoters by CREBdependent signaling and CLOCK/BMAL1 activity. Proc Natl Acad Sci USA 2002, 99: Deisseroth K, Tsien RW: Dynamic multiphosphorylation passwords for activity-dependent gene expression. Neuron 2002, 34: Dudek SM, Fields RD: Somatic action potentials are sufficient for late-phase LTP-related cell signaling. Proc Natl Acad Sci USA 2002, 99: LiuZ,OtsuY,MurphyTH:Analysis of the preference of the L-type VSCCs for different firing patterns in HEK cells and hippocampal neurons. Soc Neurosci Abstr 1999, 707: Schafe GE, Nader K, Blair HT, LeDoux JE: Memory consolidation of Pavlovian fear conditioning: a cellular and molecular perspective. Trends Neurosci 2001, 24: Milner B, Squire LR, Kandel ER: Cognitive neuroscience and the study of memory. Neuron 1998, 20: Impey S, Smith DM, Obrietan K, Donahue R, Wade C, Storm DR: Stimulation of camp response element (CRE)-mediated transcription during contextual learning. Nat Neurosci 1998, 1: Bourtchuladze R, Frenguelli B, Blendy J, Cioffi D, Schutz G, Silva AJ: Deficient long-term memory in mice with a targeted mutation of the camp-responsive element-binding protein. Cell 1994, 79: Kida S, Josselyn SA, de Ortiz SP, Kogan JH, Chevere I, Masushige S, Silva AJ: CREB required for the stability of new and reactivated fear memories. Nat Neurosci 2002, 5: This study, along with that of Kogan et al. and Josselyn et al. [51,52 ], demonstrates the importance of CREB in fear memory formation in the amygdala and the hippocampus, as well as the use of spaced training protocols to overcome insufficient CREB activation and lead to robust memory formation in both structures. 51. Kogan JH, Frankland PW, Blendy JA, Coblentz J, Marowitz Z, Schutz G, Silva AJ: Spaced training induces normal long-term memory in CREB mutant mice. Curr Biol 1997, 7: Josselyn SA, Shi C, Carlezon WA Jr, Neve RL, Nestler EJ, Davis M: Long-term memory is facilitated by camp response elementbinding protein overexpression in the amygdala. J Neurosci 2001, 21: See annotation to Kida et al. [50 ]. 53. Bauer EP, Schafe GE, LeDoux JE: NMDA receptors and L-type voltage-gated calcium channels contribute to long-term potentiation and different components of fear memory formation in the lateral amygdala. J Neurosci 2002, 22: Weisskopf MG, Bauer EP, LeDoux JE: L-type voltage-gated calcium channels mediate NMDA-independent associative long-term potentiation at thalamic input synapses to the amygdala. J Neurosci 1999, 19: Vendrell M, Pujol MJ, Tusell JM, Serratosa J: Effect of different convulsants on calmodulin levels and proto-oncogene c-fos expression in the central nervous system. Brain Res Mol Brain Res 1992, 14: Milikan JM, Bolsover SR: Use of fluorescently labelled calmodulins as tools to measure subcellular calmodulin activation in living dorsal root ganglion cells. Pflugers Arch 2000, 439: Luby-Phelps K, Hori M, Phelps JM, Won D: Ca 2þ -regulated dynamic compartmentalization of calmodulin in living smooth muscle cells. J Biol Chem 1995, 270: Craske M, Takeo T, Gerasimenko O, Vaillant C, Torok K, Petersen OH, Tepikin AV: Hormone-induced secretory and nuclear translocation of calmodulin: oscillations of calmodulin concentration with the nucleus as an integrator. Proc Natl Acad Sci USA 1999, 96: Sala C, Rudolph-Correia S, Sheng M: Developmentally regulated NMDA receptor-dependent dephosphorylation of camp response element-binding protein (CREB) in hippocampal neurons. J Neurosci 2000, 20: Martin KC, Michael D, Rose JC, Barad M, Casadio A, Zhu H, Kandel ER: MAP kinase translocates into the nucleus of the