frontiers in REVIEW ARTICLE SYSTEMS NEUROSCIENCE published:20 May 2014 dot10.3389/fnsys.2014.00090 Pharmacological enhancement of memory or cognition in normal subjects Gary Lynch12*,Conor D.Cox2 and Christine M.Gall2 Department of Psychiatry and Human Behavior,University of Califomia,Irvine,CA.USA Department of Anatomy and Neurobiology.University of California,Irvine,CA.USA Edited by: The possibility of expanding memory or cognitive capabilities above the levels in high Mikhail Lebedev,Duke University, functioning individuals is a topic of intense discussion among scientists and in society USA at large.The majority of animal studies use behavioral endpoint measures;this has Reviewed by: loan Opris,Wake Forest University, produced valuable information but limited predictability for human outcomes.Accordingly. USA several groups are pursuing a complementary strategy with treatments targeting synaptic Rafael Roesler,Federal University of events associated with memory encoding or forebrain network operations.Transcription Rio Grande do Sul,Brazil and translation figure prominently in substrate work directed at enhancement.Notably. Sam Deadwyler,Wake Forest University Health Sciences,USA the question of why new proteins would be needed for a now-forming memory given Maryam Farahmandfar,Tehran that learning-driven synthesis presumably occurred throughout the immediate past has University of Medical Sciences,Iran been largely ignored.Despite this conceptual problem,and some controversy,recent *Correspondence studies have reinvigorated the idea that selective gene manipulation is a plausible Gary Lynch,Department of route to enhancement.Efforts to improve memory by facilitating synaptic encoding Psychiatry and Human Behavior, Gillespie Neuroscience Research of information have also progressed,in part due of breakthroughs on mechanisms Facility,University of California,837 that stabilize learning-related,long-term potentiation (LTP).These advances point to a Health Science Road,Irvine,CA, reductionistic hypothesis for a diversity of experimental results on enhancement,and 92697.1275,US4 identify under-explored possibilities.Cognitive enhancement remains an elusive goal,in e-mail:glynch@uci.edu part due to the difficulty of defining the target.The popular view of cognition as a collection of definable computations seems to miss the fluid,integrative process experienced by high functioning individuals.The neurobiological approach obviates these psychological issues to directly test the consequences of improving throughput in networks underlying higher order behaviors.The few relevant studies testing drugs that selectively promote excitatory transmission indicate that it is possible to expand cortical networks engaged by complex tasks and that this is accompanied by capabilities not found in normal animals. Keywords:cognitive enhancement,learning,long term potentiation,ampakine,synaptic plasticity,BDNF,F-actin, positive AMPA receptor modulators INTRODUCTION of a system that,while capable of periodically focusing on spe- The present review concerns three topics,two of which involve cific tasks,usually works by integrating a vast amount of disparate terms-enhancement and cognition-that are not sharply material into a product accessible to consciousness.A true cogni- defined.Usage of the former seems straightforward when applied tive enhancer might therefore take the form of a treatment that to memory,although it is often unclear whether accelerated increases the speed or capacity of this assembly process. acquisition or an increase in encoding strength is intended.But Memory enhancement,as suggested,appears to be a much applied to cognition,claims for enhancement face the great prob- more tractable problem.Retention is easily measured as is the lem of how to quantify something for which there is no consensus amount of training needed to produce a given score in a test sub- measurement system.The difficulty can be reduced by focusing sequent to learning.But a curious problem emerges here:few of on cognitive activities of a type that can be described in compu- the many pharmacological agents that produce robust enhance- tational terms.This,however,raises questions about the extent ment of memory in animals are found to have positive effects in to which the sampled process is representative,or a major com- humans.This observation has become the subject of intense pub- ponent,of cognition as the term is typically used.In response,it lic discussion,perhaps with growing skepticism about the utility could reasonably be argued that cognition is a collection of semi- of animal studies on memory enhancement.Some neuroscien- independent operations (e.g.,categorization,value assignment) tists have argued that the"failure to predict"problem reflects the (Sugrue et al.,2005;Tsunada and Sawaguchi,2012)but this seems widespread use of paradigms that have little relevance to human unsatisfactory because the phenomenon is experienced as being, learning.These workers have devised ingenious protocols that can if not unitary,then at least strongly coherent.Electrophysiological be used in rodents and with minor modifications in humans(e.g., and brain imaging results showing coordinated activity across Bari et al.,2008;Demeter et al.,2008;Eichenbaum and Robitsek, broad stretches of neocortex provide some support for the idea 2009;Zeeb et al.,2009;Demeter and Sarter,2013).There is every Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8 Article 90 1
REVIEW ARTICLE published: 20 May 2014 doi: 10.3389/fnsys.2014.00090 Pharmacological enhancement of memory or cognition in normal subjects Gary Lynch1,2*, Conor D. Cox2 and Christine M. Gall 2 1 Department of Psychiatry and Human Behavior, University of California, Irvine, CA, USA 2 Department of Anatomy and Neurobiology, University of California, Irvine, CA, USA Edited by: Mikhail Lebedev, Duke University, USA Reviewed by: Ioan Opris, Wake Forest University, USA Rafael Roesler, Federal University of Rio Grande do Sul, Brazil Sam Deadwyler, Wake Forest University Health Sciences, USA Maryam Farahmandfar, Tehran University of Medical Sciences, Iran *Correspondence: Gary Lynch, Department of Psychiatry and Human Behavior, Gillespie Neuroscience Research Facility, University of California, 837 Health Science Road, Irvine, CA, 92697-1275, USA e-mail: glynch@uci.edu The possibility of expanding memory or cognitive capabilities above the levels in high functioning individuals is a topic of intense discussion among scientists and in society at large. The majority of animal studies use behavioral endpoint measures; this has produced valuable information but limited predictability for human outcomes. Accordingly, several groups are pursuing a complementary strategy with treatments targeting synaptic events associated with memory encoding or forebrain network operations. Transcription and translation figure prominently in substrate work directed at enhancement. Notably, the question of why new proteins would be needed for a now-forming memory given that learning-driven synthesis presumably occurred throughout the immediate past has been largely ignored. Despite this conceptual problem, and some controversy, recent studies have reinvigorated the idea that selective gene manipulation is a plausible route to enhancement. Efforts to improve memory by facilitating synaptic encoding of information have also progressed, in part due of breakthroughs on mechanisms that stabilize learning-related, long-term potentiation (LTP). These advances point to a reductionistic hypothesis for a diversity of experimental results on enhancement, and identify under-explored possibilities. Cognitive enhancement remains an elusive goal, in part due to the difficulty of defining the target. The popular view of cognition as a collection of definable computations seems to miss the fluid, integrative process experienced by high functioning individuals. The neurobiological approach obviates these psychological issues to directly test the consequences of improving throughput in networks underlying higher order behaviors. The few relevant studies testing drugs that selectively promote excitatory transmission indicate that it is possible to expand cortical networks engaged by complex tasks and that this is accompanied by capabilities not found in normal animals. Keywords: cognitive enhancement, learning, long term potentiation, ampakine, synaptic plasticity, BDNF, F-actin, positive AMPA receptor modulators INTRODUCTION The present review concerns three topics, two of which involve terms—enhancement and cognition—that are not sharply defined. Usage of the former seems straightforward when applied to memory, although it is often unclear whether accelerated acquisition or an increase in encoding strength is intended. But applied to cognition, claims for enhancement face the great problem of how to quantify something for which there is no consensus measurement system. The difficulty can be reduced by focusing on cognitive activities of a type that can be described in computational terms. This, however, raises questions about the extent to which the sampled process is representative, or a major component, of cognition as the term is typically used. In response, it could reasonably be argued that cognition is a collection of semiindependent operations (e.g., categorization, value assignment) (Sugrue et al., 2005; Tsunada and Sawaguchi, 2012) but this seems unsatisfactory because the phenomenon is experienced as being, if not unitary, then at least strongly coherent. Electrophysiological and brain imaging results showing coordinated activity across broad stretches of neocortex provide some support for the idea of a system that, while capable of periodically focusing on specific tasks, usually works by integrating a vast amount of disparate material into a product accessible to consciousness. A true cognitive enhancer might therefore take the form of a treatment that increases the speed or capacity of this assembly process. Memory enhancement, as suggested, appears to be a much more tractable problem. Retention is easily measured as is the amount of training needed to produce a given score in a test subsequent to learning. But a curious problem emerges here: few of the many pharmacological agents that produce robust enhancement of memory in animals are found to have positive effects in humans. This observation has become the subject of intense public discussion, perhaps with growing skepticism about the utility of animal studies on memory enhancement. Some neuroscientists have argued that the “failure to predict” problem reflects the widespread use of paradigms that have little relevance to human learning. These workers have devised ingenious protocols that can be used in rodents and with minor modifications in humans (e.g., Bari et al., 2008; Demeter et al., 2008; Eichenbaum and Robitsek, 2009; Zeeb et al., 2009; Demeter and Sarter, 2013). There is every Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 1 SYSTEMS NEUROSCIENCE
Lynch et al. Cognition enhancement in normal subjects reason to assume that these efforts will ultimately narrow the et al.,2001;Plath et al.,2006;Katche et al.,2010,2012).However gap in cross-species comparisons.But there is a more funda- the basic idea that new protein synthesis is critical to memory for- mental issue from comparative biology that could underlie the mation has been controversial since its introduction more than failure-to-predict problem:humans are enormously encephalized 50 years ago (Abraham and Williams,2008;Gold,2008).Much of animals and rodents aren't (neocortex makes up at least 77%of the dispute revolves around the necessary prediction that protein brain volume in human and just 31%in rat;Stephan et al.,1981;synthesis inhibitors will selectively block recently acquired mem- Swanson,1995).Encephalization is hypothesized to result in a ory;most papers report this result but others do not,or argue that shift of functions from lower brain to cortex;from this perspec- observed disruptions to encoding are due to factors unrelated to tive,humans may be using networks of a very different kind than synthesis(Routtenberg,2008;Gold and Wrenn,2012). those employed by rodents to solve similar problems. Beyond this,the protein synthesis argument faces certain con- An alternative to behaviorally based approaches to developing ceptual problems.Learning is a continuous process in humans, enhancers would be to focus on the neurobiological substrates of and likely other mammals,with new encoding occurring many memory and cognition.This seems feasible in the case of mem- times a minute,as is evident with episodic memory.People recog- ory because of the tremendous progress that has been made in nize or recall a remarkable number of serial events when queried identifying synaptic mechanisms that encode information.There after a 90 min movie.Unless we make the very unlikely assump- is no good reason to think that these processes differ significantly tion that each item of information is encoded on a different between mammalian species and indeed comparative studies sug- neuron,it is difficult to see why,after hours of producing pro- gest that certain essential elements are evolutionarily ancient teins needed for consolidation,a given cell would need further (Crystal and Glanzman,2013).It follows from this that treat- synthesis to stabilize a now forming memory.Along this line,it ments acting on memory substrates in rodents are likely to have has been argued that animals exposed to an enriched environ- similar actions in human brain.Cognition again represents a ment which would entail constitutively elevated basal activity,and much more challenging problem.However,the universally held thus activity-driven protein synthesis,may not require additional assumption that cognitive operations arise from the transient synthesis to support LTP(Abraham and Williams,2008)and the formation of telencephalic networks points to a relatively sim- related encoding of hippocampus-dependent memories.There is, ple idea for enhancement.Communication within and between however,a special case in which transcription and/or broadly cortical regions is mediated by glutamatergic transmission;if distributed translation could be required to securely encode a so,then agents that augment the release of glutamate,or the specific memory;namely,a circumstance in which continuous post-synaptic response to it,should facilitate the formation of learning of similar material does not precede the new instance. cognition's substrates. Under these conditions,consolidation could depend upon pro- The following sections consider attempts to develop enhancers teins generated by the isolated learning episode.Note that this via actions on (i)different aspects of the complex machinery scenario loosely describes the great majority of animal studies underlying learning-related synaptic modifications,or(ii)com- testing for the contributions of protein synthesis.Certain of these munication within and between cortical networks. arguments make relatively straightforward,readily tested predic- tions.For example,animals with a well-developed learning set MEMORY ENHANCEMENT could be given protein synthesis inhibitors after learning a single Most research on memory enhancement deals with psychological problem with or without having dealt with many such problems events that precede the actual encoding of information.There is in the preceding hours.Such a paradigm can be achieved for rats for example a very large literature describing attempts,typically using two-odor discriminations.If continual learning obviates the using chemical agents,to increase the speed of learning by mod- need for problem-specific synthesis,then the blockers should have ulating arousal and attention (Lynch et al.,2011).It has become no effect in a group given many trials prior to being introduced common to refer to resultant improvements as cognitive enhance- to the new test items ment,presumably because key elements of cognition are being There is a variant of the translation hypothesis that addresses manipulated,but there are reasons to question this assumption the problem of why prior synthesis doesn't provide a sufficient (see below).There is a smaller,but rapidly growing,body of work supply of proteins for current learning.This involves the ample directed at the machinery responsible for converting patterns of evidence for dendritic (local)translation from already in place afferent activity into the long lasting increases in synaptic strength mRNAs.One could posit a set of conditions in which new synthe- assumed to encode specific information.This section evaluates sis,even after recent experience,needs to occur post-acquisition the latter material. for transfer into long-term storage;e.g.,(1)translation occurs within very small dendritic compartments;(2)such active regions GENE EXPRESSION AND PROTEIN SYNTHESIS are only found in the immediate vicinity of recently modified Work in this area begins with the hypothesis that learning triggers synapses;and(3)newly formed proteins do not diffuse to any the transcription or local translation of proteins that serve to con- great degree.These circumstances would reduce the probabil- solidate the newly acquired memories,something that can take ity that proteins from earlier learning would be present at the anywhere from many minutes to hours.Compounds that facili-large majority of current sites.But"synaptic tagging"experi- tate production of the pertinent RNAs or proteins could accord- ments,conducted for instances where LTP in hippocampal slices ingly increase the likelihood that recent learning will lead to stable is blocked by protein synthesis inhibitors,describe results that are memory,and there are many reports of such effects (Guzowski not consistent with these postulates.Specifically,LTP induction at Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8 Article 90 2
Lynch et al. Cognition enhancement in normal subjects reason to assume that these efforts will ultimately narrow the gap in cross-species comparisons. But there is a more fundamental issue from comparative biology that could underlie the failure-to-predict problem: humans are enormously encephalized animals and rodents aren’t (neocortex makes up at least 77% of brain volume in human and just 31% in rat; Stephan et al., 1981; Swanson, 1995). Encephalization is hypothesized to result in a shift of functions from lower brain to cortex; from this perspective, humans may be using networks of a very different kind than those employed by rodents to solve similar problems. An alternative to behaviorally based approaches to developing enhancers would be to focus on the neurobiological substrates of memory and cognition. This seems feasible in the case of memory because of the tremendous progress that has been made in identifying synaptic mechanisms that encode information. There is no good reason to think that these processes differ significantly between mammalian species and indeed comparative studies suggest that certain essential elements are evolutionarily ancient (Crystal and Glanzman, 2013). It follows from this that treatments acting on memory substrates in rodents are likely to have similar actions in human brain. Cognition again represents a much more challenging problem. However, the universally held assumption that cognitive operations arise from the transient formation of telencephalic networks points to a relatively simple idea for enhancement. Communication within and between cortical regions is mediated by glutamatergic transmission; if so, then agents that augment the release of glutamate, or the post-synaptic response to it, should facilitate the formation of cognition’s substrates. The following sections consider attempts to develop enhancers via actions on (i) different aspects of the complex machinery underlying learning-related synaptic modifications, or (ii) communication within and between cortical networks. MEMORY ENHANCEMENT Most research on memory enhancement deals with psychological events that precede the actual encoding of information. There is for example a very large literature describing attempts, typically using chemical agents, to increase the speed of learning by modulating arousal and attention (Lynch et al., 2011). It has become common to refer to resultant improvements as cognitive enhancement, presumably because key elements of cognition are being manipulated, but there are reasons to question this assumption (see below). There is a smaller, but rapidly growing, body of work directed at the machinery responsible for converting patterns of afferent activity into the long lasting increases in synaptic strength assumed to encode specific information. This section evaluates the latter material. GENE EXPRESSION AND PROTEIN SYNTHESIS Work in this area begins with the hypothesis that learning triggers the transcription or local translation of proteins that serve to consolidate the newly acquired memories, something that can take anywhere from many minutes to hours. Compounds that facilitate production of the pertinent RNAs or proteins could accordingly increase the likelihood that recent learning will lead to stable memory, and there are many reports of such effects (Guzowski et al., 2001; Plath et al., 2006; Katche et al., 2010, 2012). However, the basic idea that new protein synthesis is critical to memory formation has been controversial since its introduction more than 50 years ago (Abraham and Williams, 2008; Gold, 2008). Much of the dispute revolves around the necessary prediction that protein synthesis inhibitors will selectively block recently acquired memory; most papers report this result but others do not, or argue that observed disruptions to encoding are due to factors unrelated to synthesis (Routtenberg, 2008; Gold and Wrenn, 2012). Beyond this, the protein synthesis argument faces certain conceptual problems. Learning is a continuous process in humans, and likely other mammals, with new encoding occurring many times a minute, as is evident with episodic memory. People recognize or recall a remarkable number of serial events when queried after a 90 min movie. Unless we make the very unlikely assumption that each item of information is encoded on a different neuron, it is difficult to see why, after hours of producing proteins needed for consolidation, a given cell would need further synthesis to stabilize a now forming memory. Along this line, it has been argued that animals exposed to an enriched environment which would entail constitutively elevated basal activity, and thus activity-driven protein synthesis, may not require additional synthesis to support LTP (Abraham and Williams, 2008) and the related encoding of hippocampus-dependent memories. There is, however, a special case in which transcription and/or broadly distributed translation could be required to securely encode a specific memory; namely, a circumstance in which continuous learning of similar material does not precede the new instance. Under these conditions, consolidation could depend upon proteins generated by the isolated learning episode. Note that this scenario loosely describes the great majority of animal studies testing for the contributions of protein synthesis. Certain of these arguments make relatively straightforward, readily tested predictions. For example, animals with a well-developed learning set could be given protein synthesis inhibitors after learning a single problem with or without having dealt with many such problems in the preceding hours. Such a paradigm can be achieved for rats using two-odor discriminations. If continual learning obviates the need for problem-specific synthesis, then the blockers should have no effect in a group given many trials prior to being introduced to the new test items. There is a variant of the translation hypothesis that addresses the problem of why prior synthesis doesn’t provide a sufficient supply of proteins for current learning. This involves the ample evidence for dendritic (local) translation from already in place mRNAs. One could posit a set of conditions in which new synthesis, even after recent experience, needs to occur post-acquisition for transfer into long-term storage; e.g., (1) translation occurs within very small dendritic compartments; (2) such active regions are only found in the immediate vicinity of recently modified synapses; and (3) newly formed proteins do not diffuse to any great degree. These circumstances would reduce the probability that proteins from earlier learning would be present at the large majority of current sites. But “synaptic tagging” experiments, conducted for instances where LTP in hippocampal slices is blocked by protein synthesis inhibitors, describe results that are not consistent with these postulates. Specifically, LTP induction at Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 2
Lynch et al. Cognition enhancement in normal subjects one input protects subsequently induced potentiation at a second 2003),and does not disturb already potentiated contacts as likely input to the same region from the effects of the inhibitor (Frey required for a high capacity memory system.A very large body of and Morris,1997;Shires et al.,2012).Given the small number experimental work has confirmed the tight connection between of synapses that generate EPSPs of conventional amplitudes,it LTP and diverse instances of memory (e.g.,Roman et al.,1987; is extremely likely that connections from the two inputs are,for Rioult-Pedotti et al.,2000;Whitlock et al.,2006).Moreover,LTP the most part,located on different dendritic segments.It follows is intimately related to the theta rhythm,an oscillation long asso- then that proteins from the first episode must have been synthe- ciated with learning (Buzsaki,2005;Vertes,2005;Snider et al., sized,or traveled,throughout much of the dendritic arborization, 2013);i.e.,five brief(30 ms)bursts of high frequency stimulation a point that is reinforced by evidence for tagging in the apical den- pulses(a pattern that mimics"theta bursting"during learning) drites after stimulation of basal afferents (Alarcon et al.,2006).It prove to be near optimal for inducing extremely stable LTP but will be noted that these findings align with the broad idea that only when separated by the period of the theta wave (Larson continual learning maintains relevant proteins at levels sufficient et al.,1986;Capocchi et al.,1992).The reasons for this have been for LTP-related plasticity,obviating the need for synthesis after identified (Figure 1). individual learning events. These observations suggest the possibility of enhancing learn- The above discussion concerns interpretative issues rather than ing with drugs that promote theta activity and correlated bursts the likelihood of achieving enhancement using the transcription/of high frequency discharges.Agents such as physostigmine, translation strategy.It may well be the case that increasing within- that facilitate central cholinergic transmission,promote the theta cell levels of proteins that support consolidation reduces the rhythm (Olpe et al.,1987;Hasselmo,2006)and are reported requirements for encoding persistent memories and/or increases to improve learning scores in certain experimental situations. their stability.Signaling from synapses to the nucleus or to local Notably,drugs of this type are among the few treatments protein synthesis machinery involves many steps and so is likely approved for Alzheimer's Disease (Clarke and Francis,2005; to be a variable and somewhat uncertain process.It would not Noetzli and Eap,2013).However,cholinergic systems perform be surprising,then,if the ongoing production of memory-related varied functions in brain,some of which are homeostatic in elements operates at a less than optimal rate even in high per- nature.This likely explains why drugs targeting cholinergic forming,normal subjects.In line with this,there are multiple mechanisms have not gained widespread acceptance as plausi- demonstrations that treatment with compounds that inhibit par- ble enhancers.Another approach based on theta activity involves ticular histone deacetylases,leading to increased transcription of the large hyperpolarizing potentials triggered within target neu- select gene families,can markedly enhance memory after single rons by the short train of theta bursts used to induce LTP.These training sessions (Stefanko et al.,2009;McQuown et al.,2011).after-hyperpolarizing potentials(AHPs),set in motion by cell dis- Also of interest are the numerous studies showing that selective charges,persist throughout the duration of the theta train and phosphodiesterase-4 inhibitors have potent enhancing effects on serve to counteract the depolarization needed to unblock the memory.Inhibitors of this class(e.g.,Rolipram),drive the protein voltage dependent,synaptic NMDA receptors.Influx of calcium kinase A-CREB transcription pathway implicated in learning through these receptors,followed by release of the cation from in a broad array of animals (including invertebrates),and so intracellular stores,triggers the chain of events leading to poten- is argued to be a very ancient,evolutionarily conserved mem- tiation (Figure 1).AHPs are mediated by a set of voltage-and ory substrate (Tully et al.,2003;Normann and Berger,2008).calcium-sensitive potassium channels,prominent among which Evidence that the same results obtain after extensive experience is the SK3 channel (Hosseini et al.,2001).The bee toxin apamin with similar problems in the recent past,and presumably a great blocks this channel with some selectivity and,as predicted,aug- deal of learning-driven transcription,would constitute support ments post-synaptic responses to theta burst trains;this results in for there being less than optimal production of proteins needed a striking increase in the magnitude of LTP (Kramar et al.,2004). for encoding under normal circumstances.This would certainly While a number of studies have found substantial improve- encourage the idea that enhanced protein synthesis is a viable ments in rodent learning with apamin treatment (Ikonen and route to augmented memory. Riekkinen,1999;Brennan et al.,2008;Vick et al.,2010),this is not a likely enhancer because of toxicology issues.But given SYNAPTIC PLASTICITY AND MEMORY ENHANCEMENT increasing interest in applications of channel blockers for diverse Most mechanism-based efforts directed at improving memory clinical problems,the apamin results suggest an intriguing mech- have focused on synaptic plasticity and in particular the long term anistic target for the development of enhancers.It is of note in this potentiation (LTP)effect.Researchers since the late 19th cen- regard that Brain Derived Neurotrophic Factor (BDNF),which tury have argued that the enormous capacity of memory is best appears to be released from terminals by theta bursts(Balkowiec explained by assuming that physical encoding of new information and Katz,2000;Chen et al.,2010b),also reduces AHPs at least occurs at small numbers of connections between neurons.The in rats (Kramar et al,2004).Elevating endogenous levels of discovery of LTP demonstrated that individual synapses in the this neurotrophin,which can be achieved by pharmacological cortical telencephalon do in fact possess the properties expected manipulations described later,thus provides another avenue for for a memory substrate (Bliss and Collingridge,1993;Lynch, enhancement. 1998,2004b;Morris,2003).The increase in transmission strength Identification of the initial triggers for LTP,as schematized in (magnitude of EPSCs)develops quickly,persists for a remark- Figure 1,pointed to NMDA receptor-mediated calcium influxes able period (weeks at least)(Staubli and Lynch,1987;Abraham, as a logical target for enhancement.The existence of multiple Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8 Article 90 3
Lynch et al. Cognition enhancement in normal subjects one input protects subsequently induced potentiation at a second input to the same region from the effects of the inhibitor (Frey and Morris, 1997; Shires et al., 2012). Given the small number of synapses that generate EPSPs of conventional amplitudes, it is extremely likely that connections from the two inputs are, for the most part, located on different dendritic segments. It follows then that proteins from the first episode must have been synthesized, or traveled, throughout much of the dendritic arborization, a point that is reinforced by evidence for tagging in the apical dendrites after stimulation of basal afferents (Alarcon et al., 2006). It will be noted that these findings align with the broad idea that continual learning maintains relevant proteins at levels sufficient for LTP-related plasticity, obviating the need for synthesis after individual learning events. The above discussion concerns interpretative issues rather than the likelihood of achieving enhancement using the transcription / translation strategy. It may well be the case that increasing withincell levels of proteins that support consolidation reduces the requirements for encoding persistent memories and/or increases their stability. Signaling from synapses to the nucleus or to local protein synthesis machinery involves many steps and so is likely to be a variable and somewhat uncertain process. It would not be surprising, then, if the ongoing production of memory-related elements operates at a less than optimal rate even in high performing, normal subjects. In line with this, there are multiple demonstrations that treatment with compounds that inhibit particular histone deacetylases, leading to increased transcription of select gene families, can markedly enhance memory after single training sessions (Stefanko et al., 2009; McQuown et al., 2011). Also of interest are the numerous studies showing that selective phosphodiesterase-4 inhibitors have potent enhancing effects on memory. Inhibitors of this class (e.g., Rolipram), drive the protein kinase A—CREB transcription pathway implicated in learning in a broad array of animals (including invertebrates), and so is argued to be a very ancient, evolutionarily conserved memory substrate (Tully et al., 2003; Normann and Berger, 2008). Evidence that the same results obtain after extensive experience with similar problems in the recent past, and presumably a great deal of learning-driven transcription, would constitute support for there being less than optimal production of proteins needed for encoding under normal circumstances. This would certainly encourage the idea that enhanced protein synthesis is a viable route to augmented memory. SYNAPTIC PLASTICITY AND MEMORY ENHANCEMENT Most mechanism-based efforts directed at improving memory have focused on synaptic plasticity and in particular the long term potentiation (LTP) effect. Researchers since the late 19th century have argued that the enormous capacity of memory is best explained by assuming that physical encoding of new information occurs at small numbers of connections between neurons. The discovery of LTP demonstrated that individual synapses in the cortical telencephalon do in fact possess the properties expected for a memory substrate (Bliss and Collingridge, 1993; Lynch, 1998, 2004b; Morris, 2003). The increase in transmission strength (magnitude of EPSCs) develops quickly, persists for a remarkable period (weeks at least) (Staubli and Lynch, 1987; Abraham, 2003), and does not disturb already potentiated contacts as likely required for a high capacity memory system. A very large body of experimental work has confirmed the tight connection between LTP and diverse instances of memory (e.g., Roman et al., 1987; Rioult-Pedotti et al., 2000; Whitlock et al., 2006). Moreover, LTP is intimately related to the theta rhythm, an oscillation long associated with learning (Buzsaki, 2005; Vertes, 2005; Snider et al., 2013); i.e., five brief (30 ms) bursts of high frequency stimulation pulses (a pattern that mimics “theta bursting” during learning) prove to be near optimal for inducing extremely stable LTP but only when separated by the period of the theta wave (Larson et al., 1986; Capocchi et al., 1992). The reasons for this have been identified (Figure 1). These observations suggest the possibility of enhancing learning with drugs that promote theta activity and correlated bursts of high frequency discharges. Agents such as physostigmine, that facilitate central cholinergic transmission, promote the theta rhythm (Olpe et al., 1987; Hasselmo, 2006) and are reported to improve learning scores in certain experimental situations. Notably, drugs of this type are among the few treatments approved for Alzheimer’s Disease (Clarke and Francis, 2005; Noetzli and Eap, 2013). However, cholinergic systems perform varied functions in brain, some of which are homeostatic in nature. This likely explains why drugs targeting cholinergic mechanisms have not gained widespread acceptance as plausible enhancers. Another approach based on theta activity involves the large hyperpolarizing potentials triggered within target neurons by the short train of theta bursts used to induce LTP. These after-hyperpolarizing potentials (AHPs), set in motion by cell discharges, persist throughout the duration of the theta train and serve to counteract the depolarization needed to unblock the voltage dependent, synaptic NMDA receptors. Influx of calcium through these receptors, followed by release of the cation from intracellular stores, triggers the chain of events leading to potentiation (Figure 1). AHPs are mediated by a set of voltage- and calcium-sensitive potassium channels, prominent among which is the SK3 channel (Hosseini et al., 2001). The bee toxin apamin blocks this channel with some selectivity and, as predicted, augments post-synaptic responses to theta burst trains; this results in a striking increase in the magnitude of LTP (Kramar et al., 2004). While a number of studies have found substantial improvements in rodent learning with apamin treatment (Ikonen and Riekkinen, 1999; Brennan et al., 2008; Vick et al., 2010), this is not a likely enhancer because of toxicology issues. But given increasing interest in applications of channel blockers for diverse clinical problems, the apamin results suggest an intriguing mechanistic target for the development of enhancers. It is of note in this regard that Brain Derived Neurotrophic Factor (BDNF), which appears to be released from terminals by theta bursts (Balkowiec and Katz, 2000; Chen et al., 2010b), also reduces AHPs at least in rats (Kramar et al., 2004). Elevating endogenous levels of this neurotrophin, which can be achieved by pharmacological manipulations described later, thus provides another avenue for enhancement. Identification of the initial triggers for LTP, as schematized in Figure 1, pointed to NMDA receptor-mediated calcium influxes as a logical target for enhancement. The existence of multiple Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 3
Lynch et al. Cognition enhancement in normal subjects A AMPA-Rs NMDA-Rs AHP membrane voltage first theta second theta burst burst (+NMDA) first theta burst second theta burst GABA-A GABA-A GABA GABA-B interneuron autoreceptors maximum hyperpolarization at 180 msec FIGURE 1|Why theta burst stimulation (TBS)is so effective at after-hyperpolarization (AHP).The AHP which is largely mediated by inducing LTP.TBS (Larson et al.,1986)mimics a firing pattern found in calcium and voltage dependent potassium channels,tends to counteract cortical neurons during leaming (Otto et al..1991)and elicits a robust, the depolarization produced by the burst,thus capping the magnitude of non-decremental LTP that persists for weeks (at least).(A)(left side).A NMDA receptor responses.(C-E)How the second and subsequent single stimulation pulse releases glutamate (black dots)and a partial bursts generate large depolarization and unblock NMDA receptors.(C)A membrane depolarization via current flux through AMPA receptors (dotted glutamatergic axon innervates a pyramidal cell dendrite (gray)and a line).NMDA receptors do not open because of voltage dependent block feedforward,GABAergic interneuron (orange);note that both contacts use of the ion channel (open circle).(right side)Trains of high frequency AMPA receptors (red).(D)A first theta burst triggers GABA release from stimulation cause a greater depolarization (light blue)that removes the the interneuron onto the pyramidal neuron thereby producing a channel block and thereby allows current flow through calcium permeant di-synaptic (slightly delayed)IPSC via post-synaptic GABA-A receptors NMDA receptors.Calcium is the initial trigger for LTP (B)Intracellular (orange ellipses);this shunts the EPSCs produced at neighboring recording shows that the first theta burst (four pulses at 100 Hz)in a glutamatergic synapses.The released GABA also binds to pre-synaptic, train causes a relatively modest depolarization accompanied by a single metabotrophic GABA-B auto-receptors on the releasing intemeuron spike;NMDA receptors make a very small contribution to this response. terminal (purple).(E)The auto-receptors hyperpolarize the GABAergic A second burst administered after a delay corresponding to the period of terminal and block release,an effect that reaches its maximum at the the theta wave produces a more profound depolarization with multiple period of the theta wave.A theta burst arriving at this time point spikes;this burst response contains a large NMDA receptor mediated generates an excitatory response that is only weakly counteracted by the component.Note that each theta burst in the train is followed by a large opening of post-synaptic GABA-A receptors (see B). modulatory sites (e.g.,for glycine and polyamine)on the recep- cholesterol metabolite that facilitates NMDA receptor currents tors suggested a plausible route for building positive allosteric through a novel oxysterol modulatory site and markedly increases drugs (Monaghan et al,2012).Most of this effort has been the magnitude of LTP(Paul et al,2013).The development of pos- directed toward treatments for neuropathology and psychiatric itive NMDA receptor modulators is clearly a promising area with disorders,most notably schizophrenia and depression(Labrie and regard to enhancement. Roder,2010;Dang et al.,2014),rather than memory enhance- Increasing current flux through AMPA receptors results ment.Perhaps the most widely studied agent of this type is in greater post-synaptic depolarization and thereby promotes D-cycloserine,a compound that targets the glycine binding removal of the voltage block on NMDA receptors.This sug- pocket on the receptor and facilitates channel opening(Sheinin gests that increasing AMPA receptor currents should facilitate the et al.,2001;Dravid et al.,2010).It has been known for some induction of LTP.Tests of this became possible with the inven- time that the site is important for induction of LTP(Oliver et al.,tion of AMPA receptor modulators that freely enter the brain and 1990)and,as expected from this,D-cycloserine enhances var- increase fast glutamatergic transmission(Lynch,2004a).The ini- ious forms of memory in animals (Flood et al,1992;Baxter tial positive modulators were small benzamide compounds but et al.,1994;Tsai et al.,1999;Normann and Berger,2008;Peters subsequent work from many laboratories resulted in diverse fam- and De Vries,2013).There is also evidence that the endogenous ilies of compounds that slow deactivation or desensitization (or neurosteroid pregnenolone sulfate (Wu et al,1991),and other both)of ligand bound AMPA receptors.Here we will refer to steroid-like substances(Madau et al.,2009),promote the open- all agents of this type by the term,"ampakines,"used for the ing of NMDA receptors and facilitate both LTP and memory.Also original compounds.Through a series of electrophysiological and of note,recent work led to discovery of a naturally occurring X-ray crystallography studies,the mechanism of ampakine action Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8 Article 90 4
Lynch et al. Cognition enhancement in normal subjects FIGURE 1 | Why theta burst stimulation (TBS) is so effective at inducing LTP. TBS (Larson et al., 1986) mimics a firing pattern found in cortical neurons during learning (Otto et al., 1991) and elicits a robust, non-decremental LTP that persists for weeks (at least). (A) (left side). A single stimulation pulse releases glutamate (black dots) and a partial membrane depolarization via current flux through AMPA receptors (dotted line). NMDA receptors do not open because of voltage dependent block of the ion channel (open circle). (right side) Trains of high frequency stimulation cause a greater depolarization (light blue) that removes the channel block and thereby allows current flow through calcium permeant NMDA receptors. Calcium is the initial trigger for LTP. (B) Intracellular recording shows that the first theta burst (four pulses at 100 Hz) in a train causes a relatively modest depolarization accompanied by a single spike; NMDA receptors make a very small contribution to this response. A second burst administered after a delay corresponding to the period of the theta wave produces a more profound depolarization with multiple spikes; this burst response contains a large NMDA receptor mediated component. Note that each theta burst in the train is followed by a large after-hyperpolarization (AHP). The AHP, which is largely mediated by calcium and voltage dependent potassium channels, tends to counteract the depolarization produced by the burst, thus capping the magnitude of NMDA receptor responses. (C–E) How the second and subsequent bursts generate large depolarization and unblock NMDA receptors. (C) A glutamatergic axon innervates a pyramidal cell dendrite (gray) and a feedforward, GABAergic interneuron (orange); note that both contacts use AMPA receptors (red). (D) A first theta burst triggers GABA release from the interneuron onto the pyramidal neuron thereby producing a di-synaptic (slightly delayed) IPSC via post-synaptic GABA-A receptors (orange ellipses); this shunts the EPSCs produced at neighboring glutamatergic synapses. The released GABA also binds to pre-synaptic, metabotrophic GABA-B auto-receptors on the releasing interneuron terminal (purple). (E) The auto-receptors hyperpolarize the GABAergic terminal and block release, an effect that reaches its maximum at the period of the theta wave. A theta burst arriving at this time point generates an excitatory response that is only weakly counteracted by the opening of post-synaptic GABA-A receptors (see B). modulatory sites (e.g., for glycine and polyamine) on the receptors suggested a plausible route for building positive allosteric drugs (Monaghan et al., 2012). Most of this effort has been directed toward treatments for neuropathology and psychiatric disorders, most notably schizophrenia and depression (Labrie and Roder, 2010; Dang et al., 2014), rather than memory enhancement. Perhaps the most widely studied agent of this type is D-cycloserine, a compound that targets the glycine binding pocket on the receptor and facilitates channel opening (Sheinin et al., 2001; Dravid et al., 2010). It has been known for some time that the site is important for induction of LTP (Oliver et al., 1990) and, as expected from this, D-cycloserine enhances various forms of memory in animals (Flood et al., 1992; Baxter et al., 1994; Tsai et al., 1999; Normann and Berger, 2008; Peters and De Vries, 2013). There is also evidence that the endogenous neurosteroid pregnenolone sulfate (Wu et al., 1991), and other steroid-like substances (Madau et al., 2009), promote the opening of NMDA receptors and facilitate both LTP and memory. Also of note, recent work led to discovery of a naturally occurring cholesterol metabolite that facilitates NMDA receptor currents through a novel oxysterol modulatory site and markedly increases the magnitude of LTP (Paul et al., 2013). The development of positive NMDA receptor modulators is clearly a promising area with regard to enhancement. Increasing current flux through AMPA receptors results in greater post-synaptic depolarization and thereby promotes removal of the voltage block on NMDA receptors. This suggests that increasing AMPA receptor currents should facilitate the induction of LTP. Tests of this became possible with the invention of AMPA receptor modulators that freely enter the brain and increase fast glutamatergic transmission (Lynch, 2004a). The initial positive modulators were small benzamide compounds but subsequent work from many laboratories resulted in diverse families of compounds that slow deactivation or desensitization (or both) of ligand bound AMPA receptors. Here we will refer to all agents of this type by the term, “ampakines,” used for the original compounds. Through a series of electrophysiological and X-ray crystallography studies, the mechanism of ampakine action Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 4
Lynch et al. Cognition enhancement in normal subjects is now fairly well understood.As illustrated in Figure 2,each between AMPA and NMDA receptor pharmacology:compounds subunit of the tetrameric AMPA receptor has two large extra- widely used to block the former also exhibit high affinity antag- cellular domains that form a"clamshell"that closes upon gluta- onism of the glycine modulatory site on the latter(Kessler et al., mate binding (Sun et al.,2002).Relaxation to the resting state, 1989).However,the ampakine pocket is distant to the extracel- and transmitter release,terminates current flow;this process is lular domain of AMPA receptor antagonist binding and there referred to as"deactivation."The four subunits form two dimers,is no evidence that these drugs affect NMDA receptor-gated an arrangement that can be disrupted by ligand binding;under currents. these conditions the channel closes but the transmitter is retained. Early work established that ampakines enhance both LIP and This interesting,high affinity (slow dissociation constant)state memory (Granger et al.,1993;Staubli et al.,1994),results that constitutes the desensitized condition of the receptor (Hall et al., have been multiply replicated by different groups(Lynch,2004a; 1993).It was originally thought that desensitization is the normal Lynch and Gall,2013).Versions of the drugs that simply slow route for terminating the EPSC but it now appears that deacti-deactivation lower the threshold for inducing LTP whereas those vation is responsible for the decay rate of the synaptic response. that affect both deactivation and desensitization also raise the The ampakine binding pocket is located at the dimer interface ceiling on the degree of potentiation produced by theta bursts near the hinge of the clamshell(Jin et al,2005);this strategic (Arai et al,2002).By changing rate constants for both recep- position explains how ampakines can affect both deactivation tor inactivation processes,the latter compounds lead to much and desensitization (Arai et al.,1996)(Figure 2).Apparently, longer EPSCs and thus prolonged NMDA receptor-mediated cal- the orientation of the compounds within the pocket determines cium influxes.This presumably explains their greater potency. which of the two processes is most affected.There is overlap Surprisingly,there appear to have been no studies testing for A resting bound c deactivation deactivation rate (1 ms pulse) plus membrane ampakine 的 bound,channel ampakine open binding pocket desensitization hinge'for extracellular domains also,dimer interface glutamate desensitization rate(500 ms plus ampakine pulse) ampakine bound,channel closed FIGURE 2 Mode of action for positive allosteric modulators of AMPA ms pulse of glutamate to an excised patch:delivery of the ligand causes a receptors (ampakines).(A)Schematic shows two of the four subunits that sharp influx of current that decays after rapid washout.Bound ampakines comprise the AMPA receptor tetramer in the resting state;the C-tails are not slow reopening,resulting in a significant retardation of deactivation (bottom included.Each subunit has two large extracellular domains that form a trace).(D)Prolonged stimulation of the receptor can disrupt the dimer "clamshell"containing the glutamate binding site.The hinge of the structure configuration,leading to a condition in which transmitter remains bound but is indicated by the double circle.The subunits dimerize at a zone close to the the ion channel returns to the closed state (desensitization).The upper trace hinge.The ampakine pocket (star)is strategically located at the dimer to the right describes an instance of this in which glutamate was applied for interface adjacent to the two hinges.There are thus four neurotransmitter, 500 msec.An initial influx of current was followed by decay,despite and two ampakine,sites on the full AMPA receptor.(B)Glutamate binding is continuing presence of the transmitter,to a steady state value about 1/10 of accompanied by a closing of each subunits'clamshell,resulting in opening of the peak flux.Ampakines stabilize the dimer configuration and,as predicted, the ion channel and inward flux of current.The receptor then shifts into one greatly slow desensitization-current flow continues throughout the 500 ms of two configurations;the gray arrows denote the time required for the application of glutamate.The receptor structural dynamics,including transitions in the presence (dotted)and absence (solid line)of an ampakine. interactions with an ampakine,illustrated here are based on X-ray (C)Normally,single transmission events are followed by opening of the crystallography studies (Sun et al.,2002:Jin et al.,2005);physiological data extracellular domains and release of the transmitter,a process referred to as are from patches taken from hippocampal slices (Arai et al..1996.2002:Arai 'deactivation."The upper trace to the right describes deactivation after a one and Lynch,1998). Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8 Article 905
Lynch et al. Cognition enhancement in normal subjects is now fairly well understood. As illustrated in Figure 2, each subunit of the tetrameric AMPA receptor has two large extracellular domains that form a “clamshell” that closes upon glutamate binding (Sun et al., 2002). Relaxation to the resting state, and transmitter release, terminates current flow; this process is referred to as “deactivation.” The four subunits form two dimers, an arrangement that can be disrupted by ligand binding; under these conditions the channel closes but the transmitter is retained. This interesting, high affinity (slow dissociation constant) state constitutes the desensitized condition of the receptor (Hall et al., 1993). It was originally thought that desensitization is the normal route for terminating the EPSC but it now appears that deactivation is responsible for the decay rate of the synaptic response. The ampakine binding pocket is located at the dimer interface near the hinge of the clamshell (Jin et al., 2005); this strategic position explains how ampakines can affect both deactivation and desensitization (Arai et al., 1996) (Figure 2). Apparently, the orientation of the compounds within the pocket determines which of the two processes is most affected. There is overlap between AMPA and NMDA receptor pharmacology: compounds widely used to block the former also exhibit high affinity antagonism of the glycine modulatory site on the latter (Kessler et al., 1989). However, the ampakine pocket is distant to the extracellular domain of AMPA receptor antagonist binding and there is no evidence that these drugs affect NMDA receptor-gated currents. Early work established that ampakines enhance both LTP and memory (Granger et al., 1993; Staubli et al., 1994), results that have been multiply replicated by different groups (Lynch, 2004a; Lynch and Gall, 2013). Versions of the drugs that simply slow deactivation lower the threshold for inducing LTP whereas those that affect both deactivation and desensitization also raise the ceiling on the degree of potentiation produced by theta bursts (Arai et al., 2002). By changing rate constants for both receptor inactivation processes, the latter compounds lead to much longer EPSCs and thus prolonged NMDA receptor-mediated calcium influxes. This presumably explains their greater potency. Surprisingly, there appear to have been no studies testing for FIGURE 2 | Mode of action for positive allosteric modulators of AMPA receptors (ampakines). (A) Schematic shows two of the four subunits that comprise the AMPA receptor tetramer in the resting state; the C-tails are not included. Each subunit has two large extracellular domains that form a “clamshell” containing the glutamate binding site. The hinge of the structure is indicated by the double circle. The subunits dimerize at a zone close to the hinge. The ampakine pocket (star) is strategically located at the dimer interface adjacent to the two hinges. There are thus four neurotransmitter, and two ampakine, sites on the full AMPA receptor. (B) Glutamate binding is accompanied by a closing of each subunits’ clamshell, resulting in opening of the ion channel and inward flux of current. The receptor then shifts into one of two configurations; the gray arrows denote the time required for the transitions in the presence (dotted) and absence (solid line) of an ampakine. (C) Normally, single transmission events are followed by opening of the extracellular domains and release of the transmitter, a process referred to as “deactivation.” The upper trace to the right describes deactivation after a one ms pulse of glutamate to an excised patch: delivery of the ligand causes a sharp influx of current that decays after rapid washout. Bound ampakines slow reopening, resulting in a significant retardation of deactivation (bottom trace). (D) Prolonged stimulation of the receptor can disrupt the dimer configuration, leading to a condition in which transmitter remains bound but the ion channel returns to the closed state (desensitization). The upper trace to the right describes an instance of this in which glutamate was applied for 500 msec. An initial influx of current was followed by decay, despite continuing presence of the transmitter, to a steady state value about 1/10 of the peak flux. Ampakines stabilize the dimer configuration and, as predicted, greatly slow desensitization—current flow continues throughout the 500 ms application of glutamate. The receptor structural dynamics, including interactions with an ampakine, illustrated here are based on X-ray crystallography studies (Sun et al., 2002; Jin et al., 2005); physiological data are from patches taken from hippocampal slices (Arai et al., 1996, 2002; Arai and Lynch, 1998). Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 5
Lynch et al. Cognition enhancement in normal subjects differential actions of the two functional classes of ampakine on 2007).The observed anatomical restructuring implied that induc- learning. tion events for LTP or memory result in substantial alterations Notably,the positive influence of acutely administered to the actin cytoskeleton.Tests of this,using a newly devel- ampakines on memory are reported for aged and young animals oped in situ method for labeling F-actin in hippocampal slices, (Granger et al.,1993,1996;Shors et al,1995)as wellas for a broad found that theta bursts cause a dramatic increase in the num- array of species and learning tasks (Lynch,2004a;Bernard et al., ber of spines with high concentrations of polymerized actin (Lin 2010).Very few effects in human have been published although et al.,2005;Kramar et al.,2006).The newly formed filaments one study using a short half-life,deactivation-only drug obtained proved to be unstable for a period of 5-10min,after which evidence for memory enhancement in different tasks including they were unaffected by depolymerizing agents(Rex et al.,2009, ones involving complex processing (Ingvar et al.,1997). 2010).The experimental question then became one of how the very brief AMPA and NMDA receptor events that induce LTP LEARNING-RELATED SYNAPTIC MODIFICATIONS AS A TARGET FOR lead to the formation of new actin cytoskeleton.Work using ENHANCEMENT STRATEGIES Fluorescence Deconvolution Tomography for assessing concen- The discovery of LTP (Bliss and Lomo,1973)greatly simpli-trations of activated signaling proteins at individual synapses, fied what had already been an extended search for the sub-along with the use of selective inhibitors,identified multiple, strates of memory.An early and critical clue came with electron GTPase-initiated signaling pathways involved in the assembly microscopic evidence that stable potentiation is accompanied by and stabilization of actin filament networks following theta burst changes in the morphology of dendritic spines(Lee et al,1979, stimulation (Kramar et al.,2009;Rex et al.,2009,2010;Seese 1980;Chang and Greenough,1984),an observation recently and et al.,2012).Particularly relevant to the present topic,these stud- convincingly confirmed by live imaging experiments(Matsuzaki ies also described membrane receptors that modulate the activity et al.,2004;Harvey and Svoboda,2007;Kramar et al.,2012b).of cascades leading to the cytoskeletal reorganization required for The initial studies also described results suggestive of an increase consolidation of LTP (Figure 3). in synapse size and there are now data pointing to a similar Brief treatments with BDNF partially activate at least two of effect after LTP(Chen et al,2007)and learning(Fedulov et al., the signaling pathways shown in Figure 3 and potently facilitate ..matrix ligands modulatory transmitter adhesion 000 A1 tpa ERB integrin TRKB Ca+ -◆RhoA Racl Ras calpains cdc42 Rock Lim-K PAK ERK src o p-cofilin o cortactin ◆myosin llb arp2/3 disassemble 9 min 5-10 min branching-stabilization assemble-treadmilling FIGURE 3|Signaling events responsible for reorganizing the synaptic burst stimulation.Transmitter receptors i increase calcium which stimulates cytoskeleton and consolidating LTP.The substrate map for LTP stabilization is calpain,a spine protease (Perlmutter et al,1988)that cleaves cross-linking largely based on work using hippocampal slices,although some of the steps proteins (blue lines)for the subsynaptic cytoskeleton,and ii)activates synaptic have been observed in learning studies.(A)Immunolabeled synapses adhesion receptors belonging to the integrin family (Babayan et al.,2012). surrounding the LTP site as reconstructed using Fluorescence Deconvolution Integrins then engage at least two Rho family GTPases that promote the Tomography(Seese et al.,2013):The green elements reflect immunostaining assembly of dynamic actin filaments (RhoA to cofilin and myosin)(Rex et al., for PSD95,a protein that is evenly distributed within post-synaptic densities at 2009)and,over a period of several minutes,branching and stabilization of the excitatory (glutamatergic)synapses.Phosphorylated (inactivated)cofilin was reorganized cytoskeleton.The latter processes involve Rac/Cdc42 signaling to immunolabeled with red fluorescence.Co-localization (p-Cofilin/PSD95)results cortactin and ARP2/3.The synaptic membrane also contains receptors for the in yellow labeling.The technique supports counts and size-measures for about releasable factors adenosine,estrogen,and BDNF(A1,ERB,TrkB,respectively) 40,000 synapses per image z-stack and 160,000 synapses per slice,and These receptors positively and negatively(A1)influence the signaling calculates the number of these synaptic elements that are co-localized with the pathways,probably at the level of the GTPases.Studies using neutralizing signaling protein of interest (p-Cofilin in this instance).These values are then antisera,genetic manipulations,toxins,and enzyme blockers confirm certain compared for slices that did or did not receive theta burst simulation.(B) key links in the model and show that disrupting these specific actin regulatory Schematic shows signaling pathways activated at excitatory synapses by theta pathways blocks the consolidation,but not initial expression,of LTP Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8 Article 90 6
Lynch et al. Cognition enhancement in normal subjects differential actions of the two functional classes of ampakine on learning. Notably, the positive influence of acutely administered ampakines on memory are reported for aged and young animals (Granger et al., 1993, 1996; Shors et al., 1995) as well as for a broad array of species and learning tasks (Lynch, 2004a; Bernard et al., 2010). Very few effects in human have been published although one study using a short half-life, deactivation-only drug obtained evidence for memory enhancement in different tasks including ones involving complex processing (Ingvar et al., 1997). LEARNING-RELATED SYNAPTIC MODIFICATIONS AS A TARGET FOR ENHANCEMENT STRATEGIES The discovery of LTP (Bliss and Lomo, 1973) greatly simpli- fied what had already been an extended search for the substrates of memory. An early and critical clue came with electron microscopic evidence that stable potentiation is accompanied by changes in the morphology of dendritic spines (Lee et al., 1979, 1980; Chang and Greenough, 1984), an observation recently and convincingly confirmed by live imaging experiments (Matsuzaki et al., 2004; Harvey and Svoboda, 2007; Kramar et al., 2012b). The initial studies also described results suggestive of an increase in synapse size and there are now data pointing to a similar effect after LTP (Chen et al., 2007) and learning (Fedulov et al., 2007). The observed anatomical restructuring implied that induction events for LTP or memory result in substantial alterations to the actin cytoskeleton. Tests of this, using a newly developed in situ method for labeling F-actin in hippocampal slices, found that theta bursts cause a dramatic increase in the number of spines with high concentrations of polymerized actin (Lin et al., 2005; Kramar et al., 2006). The newly formed filaments proved to be unstable for a period of 5–10 min, after which they were unaffected by depolymerizing agents (Rex et al., 2009, 2010). The experimental question then became one of how the very brief AMPA and NMDA receptor events that induce LTP lead to the formation of new actin cytoskeleton. Work using Fluorescence Deconvolution Tomography for assessing concentrations of activated signaling proteins at individual synapses, along with the use of selective inhibitors, identified multiple, GTPase-initiated signaling pathways involved in the assembly and stabilization of actin filament networks following theta burst stimulation (Kramar et al., 2009; Rex et al., 2009, 2010; Seese et al., 2012). Particularly relevant to the present topic, these studies also described membrane receptors that modulate the activity of cascades leading to the cytoskeletal reorganization required for consolidation of LTP (Figure 3). Brief treatments with BDNF partially activate at least two of the signaling pathways shown in Figure 3 and potently facilitate FIGURE 3 | Signaling events responsible for reorganizing the synaptic cytoskeleton and consolidating LTP. The substrate map for LTP stabilization is largely based on work using hippocampal slices, although some of the steps have been observed in learning studies. (A) Immunolabeled synapses surrounding the LTP site as reconstructed using Fluorescence Deconvolution Tomography (Seese et al., 2013): The green elements reflect immunostaining for PSD95, a protein that is evenly distributed within post-synaptic densities at excitatory (glutamatergic) synapses. Phosphorylated (inactivated) cofilin was immunolabeled with red fluorescence. Co-localization (p-Cofilin/PSD95) results in yellow labeling. The technique supports counts and size-measures for about 40,000 synapses per image z-stack and 160,000 synapses per slice, and calculates the number of these synaptic elements that are co-localized with the signaling protein of interest (p-Cofilin in this instance). These values are then compared for slices that did or did not receive theta burst simulation. (B) Schematic shows signaling pathways activated at excitatory synapses by theta burst stimulation. Transmitter receptors i) increase calcium which stimulates calpain, a spine protease (Perlmutter et al., 1988) that cleaves cross-linking proteins (blue lines) for the subsynaptic cytoskeleton, and ii) activates synaptic adhesion receptors belonging to the integrin family (Babayan et al., 2012). Integrins then engage at least two Rho family GTPases that promote the assembly of dynamic actin filaments (RhoA to cofilin and myosin) (Rex et al., 2009) and, over a period of several minutes, branching and stabilization of the reorganized cytoskeleton. The latter processes involve Rac/Cdc42 signaling to cortactin and ARP2/3. The synaptic membrane also contains receptors for the releasable factors adenosine, estrogen, and BDNF (A1, ERB, TrkB, respectively). These receptors positively and negatively (A1) influence the signaling pathways, probably at the level of the GTPases. Studies using neutralizing antisera, genetic manipulations, toxins, and enzyme blockers confirm certain key links in the model and show that disrupting these specific actin regulatory pathways blocks the consolidation, but not initial expression, of LTP. Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 6
Lynch et al. Cognition enhancement in normal subjects both theta burst-driven actin polymerization and LTP (Chen training(Andero et al.,2011;Bollen et al,2013);for object loca- et al.,2006;Rex et al.,2007).It seems likely that the LTP effects tion memory administration 3 h after training was also effective. reflect both direct actions on the actin regulatory pathways and These results encourage the expectation that acute systemic treat- the above noted influence on AHPs generated during the theta ment with a TrkB agonist can facilitate both initial encoding and stimulation trains(see above).Notably,scavenging extracellular mechanisms of consolidation for at least some forms of memory BDNF blocks the stabilization of LTP produced by theta burst Further work is needed to determine the range of learning and stimulation(Kovalchuk et al.,2002;Rex et al.,2007)as well as the cognitive functions that respond to this strategy and if this occurs associated activation of actin regulatory signaling and increases without deleterious side effects. in spine F-actin(Figure 4);activity-induced release of the neu- Another route for utilizing BDNF in memory studies is sug- rotrophic factor thus emerges as a key ingredient in the normal gested by the observation that transcription of the factor is production of learning-related synaptic changes.In all,increases positively regulated by neuronal activity (Isackson et al.,1991; in BDNF signaling appear to be a biologically plausible means Gall,1992).It follows from this that increases in excitatory drive for enhancing memory.Peripheral administration of the pro- to neurons,as for example produced by ampakines,should up- tein is unlikely to have robust central effects but brain permeant regulate the neurotrophin.A sizable number of studies using agonists for its synaptic TrkB receptor have been developed and individual or a series of daily injections of the positive modula- shown to improve function in varied conditions of impairment tors have confirmed this basic idea(Lauterborn et al,2000;Rex (Andero et al,2012;Schmid et al.,2012;Ding et al.,2013;Jiang et al,2006;Simmons et al.,2009;Bernard et al,2010;Haditsch et al.,2013).Reports on how these compounds affect mem- et al.,2013).The treatments rescue theta burst-induced actin ory in normals have only begun to appear but initial studies polymerization and LTP in a number of animal models of human indicate that acute systemic treatment can improve object recog- conditions in which memory loss and/or intellectual disabilities nition,object location and fear memory when given just before are prominent,including those for normal aging,low estrogen CON 0.21 CON 0.212 plus BDNF p-PAK-3 p-cofilin plus TrkB-FC PAK-3 cofilin B 2 theta bursts 10 theta bursts no BDNF BDNF labeled spines +TrkB-Fc labeled spines 40- 40 30 20 20- 10 0 0 CON TBS TBS CON TBS TBS BDNF TrkB-Fc FIGURE4 BDNF enhances theta burst induced actin signaling and alone.Pretreatment with BDNF (right image)caused a marked increase cytoskeletal assembly at hippocampal synapses.(A)Brief infusion of in TBS-induced labeling of dendritic spines relative to labeling in BDNF (60 ng/ml)into adult hippocampal slices increases phosphorylation ACSFbathed slices receiving similar stimulation.Graph:Summary of the of PAK and cofilin (immunoblots),two of the signaling proteins included number of spines with intense concentrations of Factin,as assessed in the schematic for LTP consolidation (Figure 3).Phosphorylation was using automated counting for a fixed sampling field,confirmed these blocked by addition of the extracellular BDNF scavenger TrkB-Fc (applied observations (CON,slices received low frequency stimulation only; at 0.2-2ug/ml).There were no evident effects of either treatment on *p<0.01 vs.CON).(C)Similar to (B)except that ten theta bursts levels of total PAK or cofilin.(B)Labeling of filamentous (F)actin with were used.Pretreatment with TrkB-Fc completely blocked the otherwise fluorescence-tagged phalloidin applied to slices 10min after delivery of robust increase in Factin positive spines seen with ten burst TBS. two theta bursts (TBS),a number too small to generate LTP if applied Modified from Rex et al.(2007). Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8 Article 907
Lynch et al. Cognition enhancement in normal subjects both theta burst-driven actin polymerization and LTP (Chen et al., 2006; Rex et al., 2007). It seems likely that the LTP effects reflect both direct actions on the actin regulatory pathways and the above noted influence on AHPs generated during the theta stimulation trains (see above). Notably, scavenging extracellular BDNF blocks the stabilization of LTP produced by theta burst stimulation (Kovalchuk et al., 2002; Rex et al., 2007) as well as the associated activation of actin regulatory signaling and increases in spine F-actin (Figure 4); activity-induced release of the neurotrophic factor thus emerges as a key ingredient in the normal production of learning-related synaptic changes. In all, increases in BDNF signaling appear to be a biologically plausible means for enhancing memory. Peripheral administration of the protein is unlikely to have robust central effects but brain permeant agonists for its synaptic TrkB receptor have been developed and shown to improve function in varied conditions of impairment (Andero et al., 2012; Schmid et al., 2012; Ding et al., 2013; Jiang et al., 2013). Reports on how these compounds affect memory in normals have only begun to appear but initial studies indicate that acute systemic treatment can improve object recognition, object location and fear memory when given just before training (Andero et al., 2011; Bollen et al., 2013); for object location memory administration 3 h after training was also effective. These results encourage the expectation that acute systemic treatment with a TrkB agonist can facilitate both initial encoding and mechanisms of consolidation for at least some forms of memory. Further work is needed to determine the range of learning and cognitive functions that respond to this strategy and if this occurs without deleterious side effects. Another route for utilizing BDNF in memory studies is suggested by the observation that transcription of the factor is positively regulated by neuronal activity (Isackson et al., 1991; Gall, 1992). It follows from this that increases in excitatory drive to neurons, as for example produced by ampakines, should upregulate the neurotrophin. A sizable number of studies using individual or a series of daily injections of the positive modulators have confirmed this basic idea (Lauterborn et al., 2000; Rex et al., 2006; Simmons et al., 2009; Bernard et al., 2010; Haditsch et al., 2013). The treatments rescue theta burst-induced actin polymerization and LTP in a number of animal models of human conditions in which memory loss and/or intellectual disabilities are prominent, including those for normal aging, low estrogen FIGURE 4 | BDNF enhances theta burst induced actin signaling and cytoskeletal assembly at hippocampal synapses. (A) Brief infusion of BDNF (60 ng/ml) into adult hippocampal slices increases phosphorylation of PAK and cofilin (immunoblots), two of the signaling proteins included in the schematic for LTP consolidation (Figure 3). Phosphorylation was blocked by addition of the extracellular BDNF scavenger TrkB-Fc (applied at 0.2–2µg/ml). There were no evident effects of either treatment on levels of total PAK or cofilin. (B) Labeling of filamentous (F-) actin with fluorescence-tagged phalloidin applied to slices 10 min after delivery of two theta bursts (TBS), a number too small to generate LTP if applied alone. Pretreatment with BDNF (right image) caused a marked increase in TBS-induced labeling of dendritic spines relative to labeling in ACSF-bathed slices receiving similar stimulation. Graph: Summary of the number of spines with intense concentrations of F-actin, as assessed using automated counting for a fixed sampling field, confirmed these observations (CON, slices received low frequency stimulation only; ∗p < 0.01 vs. CON). (C) Similar to (B) except that ten theta bursts were used. Pretreatment with TrkB-Fc completely blocked the otherwise robust increase in F-actin positive spines seen with ten burst TBS. Modified from Rex et al. (2007). Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 7
Lynch et al. Cognition enhancement in normal subjects levels,early stage Huntington Disease,and Angelman syndrome introduction of the word "synapse"(Cajal,1894).The idea is (Rex et al.,2006;Simmons et al.,2009;Baudry et al.,2012;Kramar intuitively attractive since such increases would clearly alter the et al.,2012a).When tested,daily injections also reduced or elimi- operation of cortical networks and thus behavior.In essence,it nated memory impairments(Simmons et al.,2009;Baudry et al., describes microscopic events that,when implemented at many 2012).Several weeks of daily ampakine treatment were shown sites,could be the physical instantiation of the macroscopic phe- to be well tolerated.They also markedly reduced pathology and nomenon of memory.From this perspective,the most direct improved motor functioning in a mouse model of early onset route to memory enhancement would involve facilitating physi- Huntington Disease (Simmons et al.,2011);subsequent work ologically produced,long lasting increases in synaptic responses. with systemic administration of a TrkB agonist obtained similar Developing what is still only an outline of the machinery that results (Simmons et al.,2013). induces,expresses,and consolidates LTP then shaped ideas about Although it is apparent that semi-chronic ampakine treat- how to produce facilitation.To some extent,it also led to a uni- ment increases BDNF protein levels,and has potent brain effects fication that is perhaps under-appreciated:an unrelated array of predicted from this,there appear to be no studies testing for enhancement candidates such as steroids,trophic factors,posi- influences of up-regulating BDNF on learning in normal,high tive modulators of glutamate receptors,and channel blockers can functioning animals.This likely reflects an assignment of greater now be seen to operate at specific levels within the same cell bio- importance to treatment than to enhancement with regard logical framework (Lynch et al.,2013).Optimistically,we may to drug development.But the exciting results obtained with be approaching a reductionistic (simplifying)conceptual event up-regulation and receptor agonists with regard to brain dis- with regard to enhancing encoding of specific pieces of infor- orders make BDNF-based strategies one of the more promis- mation.Notably,something of this kind may also be going on ing mechanism-grounded approaches to achieving memory for appreciating shared mechanistic impairments present in quite enhancement. different disorders that interfere with learning:work with a sizable The substrate map for LTP consolidation includes estrogen number of rodent models suggests that conditions with disparate receptor beta as a second membrane agent that exerts a powerful etiologies result in a common endpoint failure in cytoskeletal modulatory influence over the actin signaling leading to LTP con- reorganization(Lynch and Gall,2013). solidation.Thirty minute infusions of estrogen,at physiological But there are warning signs with regard to the possibility that concentrations,cause a modest increase in baseline transmis- the current substrate model may be overly tailored to a spe- sion in hippocampus but a striking facilitation of LTP(Cordoba cific instance of learning-related plasticity,and in particular to Montoya and Carrer,1997;Foy et al.,1999;Bi et al,2000;Kramar that found in a particular dendritic lamina(stratum radiatum) et al.,2009).Recent work showed that these effects are due to of a particular hippocampal subfield(CAl).Even within that activation of one of the actin regulatory cascades initiated by subfield,there is good evidence that the basal dendritic field theta bursts (i.e.,RhoA>ROCK>LIMK>cofilin-see Figure 3)exhibits a different form of LTP (Arai et al.,1994;Kramar and and the assembly of new filamentous actin in spine heads(Kramar Lynch,2003).And it is now well established that the peculiar et al.,2009).Unlike the case for BDNE,there are several reports mossy fiber connections between dentate gyrus and field CA3 that estrogen improves memory scores in high functioning sub- use a form of long lasting potentiation that bears little resem- jects across tasks and species (Frye et al.,2007;Liu et al.,blance to that found in apical field CAl(Staubli,1992;Schmitz 2008;Hara et al.,2014).Evidence for similar effects in humans et al.,2003).It is not unreasonable to expect that additional appears to be lacking(Grodstein,2013)although several stud- plasticity variants will be discovered as parametric studies are ies describe a decline in verbal memory with surgical menopause carried out for other telencephalic connections;e.g.,the cortico- and improvements with hormone replacement (Brinton,2009). striatal glutamatergic synapses (Jia et al.,2010)or the olfactory Beyond needing further evidence for effects in cognitively normal and associational afferents to piriform cortex (Jung et al,1990). individuals,a primary barrier to development of an estrogen- While these observations greatly complicate predictions about the based enhancement strategy lies in the fact that the steroid affects behavioral effects of putative enhancers,they also offer intrigu- many fundamental cellular processes in brain and the periphery, ing possibilities concerning specificity of action.That is,there and is known to facilitate certain types of cancer.More restricted are reasons to think that different forms of synaptic potentiation actions can be had using agonists selective for the hormone's beta may underlie different types,or aspects,of memory.An explicit receptor which is,to a degree,concentrated in brain;such agonists proposal of this type has been advanced for the basal and api- are highly effective in LTP studies (Kramar et al.,2009).Evidence cal dendrites of field CAl (Arai et al.,1994;Kramar and Lynch, that estrogen is synthesized by hippocampal neurons and that 2003):The easily induced,readily erased LTP in the basal den- hormone of local origin contributes significantly to hippocampal dritic field seems well suited for transient encoding while the synaptic plasticity(Ooishi et al.,2012;Vierk et al.,2012)should higher threshold and more rapidly stabilized form in the apical also be noted here.Thus,it may be possible to find means to field is more appropriate for long term memory.An arrangement promote normal,likely activity-dependent,estrogen actions in a of this type would be useful in addressing the problem of how regionally restricted manner. to accomplish,through repeated sampling,low noise extraction of constancies from a novel environment (apical dendrites)while INTEGRATION:MANY PATHS TO THE SAME END at the same time transiently storing a great deal of information Brain scientists had proposed increases in the strength of con- much of which can be discarded as being irrelevant(basal den- nections between neurons as the substrate of memory before the drites).In any event,testing experimental compounds on various Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8 Article 90|8
Lynch et al. Cognition enhancement in normal subjects levels, early stage Huntington Disease, and Angelman syndrome (Rex et al., 2006; Simmons et al., 2009; Baudry et al., 2012; Kramar et al., 2012a). When tested, daily injections also reduced or eliminated memory impairments (Simmons et al., 2009; Baudry et al., 2012). Several weeks of daily ampakine treatment were shown to be well tolerated. They also markedly reduced pathology and improved motor functioning in a mouse model of early onset Huntington Disease (Simmons et al., 2011); subsequent work with systemic administration of a TrkB agonist obtained similar results (Simmons et al., 2013). Although it is apparent that semi-chronic ampakine treatment increases BDNF protein levels, and has potent brain effects predicted from this, there appear to be no studies testing for influences of up-regulating BDNF on learning in normal, high functioning animals. This likely reflects an assignment of greater importance to treatment than to enhancement with regard to drug development. But the exciting results obtained with up-regulation and receptor agonists with regard to brain disorders make BDNF-based strategies one of the more promising mechanism-grounded approaches to achieving memory enhancement. The substrate map for LTP consolidation includes estrogen receptor beta as a second membrane agent that exerts a powerful modulatory influence over the actin signaling leading to LTP consolidation. Thirty minute infusions of estrogen, at physiological concentrations, cause a modest increase in baseline transmission in hippocampus but a striking facilitation of LTP (Cordoba Montoya and Carrer, 1997; Foy et al., 1999; Bi et al., 2000; Kramar et al., 2009). Recent work showed that these effects are due to activation of one of the actin regulatory cascades initiated by theta bursts (i.e., RhoA>ROCK>LIMK>cofilin—see Figure 3) and the assembly of new filamentous actin in spine heads (Kramar et al., 2009). Unlike the case for BDNF, there are several reports that estrogen improves memory scores in high functioning subjects across tasks and species (Frye et al., 2007; Liu et al., 2008; Hara et al., 2014). Evidence for similar effects in humans appears to be lacking (Grodstein, 2013) although several studies describe a decline in verbal memory with surgical menopause and improvements with hormone replacement (Brinton, 2009). Beyond needing further evidence for effects in cognitively normal individuals, a primary barrier to development of an estrogenbased enhancement strategy lies in the fact that the steroid affects many fundamental cellular processes in brain and the periphery, and is known to facilitate certain types of cancer. More restricted actions can be had using agonists selective for the hormone’s beta receptor which is, to a degree, concentrated in brain; such agonists are highly effective in LTP studies (Kramar et al., 2009). Evidence that estrogen is synthesized by hippocampal neurons and that hormone of local origin contributes significantly to hippocampal synaptic plasticity (Ooishi et al., 2012; Vierk et al., 2012) should also be noted here. Thus, it may be possible to find means to promote normal, likely activity-dependent, estrogen actions in a regionally restricted manner. INTEGRATION: MANY PATHS TO THE SAME END Brain scientists had proposed increases in the strength of connections between neurons as the substrate of memory before the introduction of the word “synapse” (Cajal, 1894). The idea is intuitively attractive since such increases would clearly alter the operation of cortical networks and thus behavior. In essence, it describes microscopic events that, when implemented at many sites, could be the physical instantiation of the macroscopic phenomenon of memory. From this perspective, the most direct route to memory enhancement would involve facilitating physiologically produced, long lasting increases in synaptic responses. Developing what is still only an outline of the machinery that induces, expresses, and consolidates LTP then shaped ideas about how to produce facilitation. To some extent, it also led to a uni- fication that is perhaps under-appreciated: an unrelated array of enhancement candidates such as steroids, trophic factors, positive modulators of glutamate receptors, and channel blockers can now be seen to operate at specific levels within the same cell biological framework (Lynch et al., 2013). Optimistically, we may be approaching a reductionistic (simplifying) conceptual event with regard to enhancing encoding of specific pieces of information. Notably, something of this kind may also be going on for appreciating shared mechanistic impairments present in quite different disorders that interfere with learning: work with a sizable number of rodent models suggests that conditions with disparate etiologies result in a common endpoint failure in cytoskeletal reorganization (Lynch and Gall, 2013). But there are warning signs with regard to the possibility that the current substrate model may be overly tailored to a specific instance of learning-related plasticity, and in particular to that found in a particular dendritic lamina (stratum radiatum) of a particular hippocampal subfield (CA1). Even within that subfield, there is good evidence that the basal dendritic field exhibits a different form of LTP (Arai et al., 1994; Kramar and Lynch, 2003). And it is now well established that the peculiar mossy fiber connections between dentate gyrus and field CA3 use a form of long lasting potentiation that bears little resemblance to that found in apical field CA1 (Staubli, 1992; Schmitz et al., 2003). It is not unreasonable to expect that additional plasticity variants will be discovered as parametric studies are carried out for other telencephalic connections; e.g., the corticostriatal glutamatergic synapses (Jia et al., 2010) or the olfactory and associational afferents to piriform cortex (Jung et al., 1990). While these observations greatly complicate predictions about the behavioral effects of putative enhancers, they also offer intriguing possibilities concerning specificity of action. That is, there are reasons to think that different forms of synaptic potentiation may underlie different types, or aspects, of memory. An explicit proposal of this type has been advanced for the basal and apical dendrites of field CA1 (Arai et al., 1994; Kramar and Lynch, 2003): The easily induced, readily erased LTP in the basal dendritic field seems well suited for transient encoding while the higher threshold and more rapidly stabilized form in the apical field is more appropriate for long term memory. An arrangement of this type would be useful in addressing the problem of how to accomplish, through repeated sampling, low noise extraction of constancies from a novel environment (apical dendrites) while at the same time transiently storing a great deal of information much of which can be discarded as being irrelevant (basal dendrites). In any event, testing experimental compounds on various Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 8
Lynch et al. Cognition enhancement in normal subjects forms of plasticity could lead to agents that target particular forms a third stimulation train produces still more potentiation but of memory. only if it is applied at least 60 min after the second.Additional work suggested that this LTP "spaced trials effect"reflects the COGNITIVE ENHANCEMENT presence of a large population of synapses with high plasticity DOES AUGMENTING MEMORY ENHANCE COGNITION? thresholds that are "primed"by the first theta episode and then Memory is such a prominent part of cognition that it seems obvi- shifted into the potentiated state by the second(Kramar et al., ous that enhancing the one will improve the other.However, 2012b).These effects fit naturally within the above described sub- there may be good computational reasons that cognitive benefit strate map for LTP:activation of synaptic integrins by a first is gained when acquisition is less than optimal in terms of speed theta burst train was followed by an approximately one hour and strength.Animals faced with new and complex circumstances period before these receptors could be re-engaged by additional need to encode regular features without storing variable,low stimulation (Babayan et al.,2012).They also set the stage for a information elements.Otherwise,as noted earlier,the resultant first test of how a drug that enhances memory affects a phys- memories will be noisy and less predictive of future encounters. iological analogue of the spaced trials effect.The results were The spaced trials effect-wherein,temporally separated train- clear:infusion of an ampakine prior to theta train #1 produced ing trials more efficiently support encoding than does a single the expected enhancement in the amplitude of LTPI but also "massed"session-can be seen as one adaptation toward better occluded further increases in the level of potentiation following capture of regularities in the learning environment(Hintzman,a second,delayed theta train administered in the absence of the 1976;Commins et al.,2003;Cepeda et al.,2006).That is,spac- drug(Kramar et al.,2012b).Thus,the ampakine enhanced ini- ing ensures that only elements that are regularly present will be tial encoding(as multiply reported)but did so at the expense of incorporated into memory while transient features will not.An effects of spaced stimulation,and presumably the computational enhancer could obviate the need for spacing by producing strong advantages associated with spacing(Lynch and Gall,2013 for a memory on an initial trial but would be expected to result in a discussion). noisy representation. The preceding example describes a situation in which phar- Tests of the above point are lacking but LTP experiments have macologically augmenting memory would likely not result in produced what may be pertinent results.The original descrip-enhanced cognition,at least in complex environments lacking tions of links between theta burst afferent stimulation and LTP explicit guidelines for effective performance.These are routine showed that,absent other manipulations,trains of ten bursts pro- circumstances in which demands on cognition are high.But a duced near maximal potentiation (Larson et al.,1986),a result great deal of cognition involves instances in which significant that led to what has become a standard paradigm.Recently,how- cues and appropriate responses are salient and predetermined, ever,it was found that a second theta train doubles the level of and irrelevant information is minimized.Under these conditions, potentiation but only if it is delayed by about 60 min(Figure 5); enhanced encoding could be of great use in building or expanding 3.2- o control path experimental path (n=7) 2.4 Ploj) 1.6 0.8 ↑TBS 0 % 80 120 160 200 240 280 320 B 3.2 个TBS1↑TBS2 (n=6) (n=5 (n=6 2.4 1.6 0.8 ↑年10min ←30min ◆40min ↑60min 0 20406080 03060901200 30 60 90 306090120 minutes FIGURE 5 A "spaced trials effect"for LTP (A)Theta burst stimulation decay over 5h of recording (means +s.e.m.s for seven slices);traces at right (TBS)produces stable LTP A single train of ten bursts was delivered to one show representative baseline (black)and potentiated (gray)responses. input to the CA1b recording field after 20 min of collecting baseline synaptic (B)Effect of a second theta train (TBS2)applied at various times after TBS1. responses to 3/min stimulation pulses in adult rat hippocampal slices.A TBS2 produced no further increases in the slopes of the responses when second input to the same dendritic region received 3/min pulses throughout delayed by 10,30,or 40 min after TBS1,but doubled the magnitude of LTP the session.Note that the potentiation of the experimental input did not when applied after a 60 min interval.Modified from Kramar et al.(2012b). Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8 Article 909
Lynch et al. Cognition enhancement in normal subjects forms of plasticity could lead to agents that target particular forms of memory. COGNITIVE ENHANCEMENT DOES AUGMENTING MEMORY ENHANCE COGNITION? Memory is such a prominent part of cognition that it seems obvious that enhancing the one will improve the other. However, there may be good computational reasons that cognitive benefit is gained when acquisition is less than optimal in terms of speed and strength. Animals faced with new and complex circumstances need to encode regular features without storing variable, low information elements. Otherwise, as noted earlier, the resultant memories will be noisy and less predictive of future encounters. The spaced trials effect—wherein, temporally separated training trials more efficiently support encoding than does a single “massed” session—can be seen as one adaptation toward better capture of regularities in the learning environment (Hintzman, 1976; Commins et al., 2003; Cepeda et al., 2006). That is, spacing ensures that only elements that are regularly present will be incorporated into memory while transient features will not. An enhancer could obviate the need for spacing by producing strong memory on an initial trial but would be expected to result in a noisy representation. Tests of the above point are lacking but LTP experiments have produced what may be pertinent results. The original descriptions of links between theta burst afferent stimulation and LTP showed that, absent other manipulations, trains of ten bursts produced near maximal potentiation (Larson et al., 1986), a result that led to what has become a standard paradigm. Recently, however, it was found that a second theta train doubles the level of potentiation but only if it is delayed by about 60 min (Figure 5); a third stimulation train produces still more potentiation but only if it is applied at least 60 min after the second. Additional work suggested that this LTP “spaced trials effect” reflects the presence of a large population of synapses with high plasticity thresholds that are “primed” by the first theta episode and then shifted into the potentiated state by the second (Kramar et al., 2012b). These effects fit naturally within the above described substrate map for LTP: activation of synaptic integrins by a first theta burst train was followed by an approximately one hour period before these receptors could be re-engaged by additional stimulation (Babayan et al., 2012). They also set the stage for a first test of how a drug that enhances memory affects a physiological analogue of the spaced trials effect. The results were clear: infusion of an ampakine prior to theta train #1 produced the expected enhancement in the amplitude of LTP1 but also occluded further increases in the level of potentiation following a second, delayed theta train administered in the absence of the drug (Kramar et al., 2012b). Thus, the ampakine enhanced initial encoding (as multiply reported) but did so at the expense of effects of spaced stimulation, and presumably the computational advantages associated with spacing (Lynch and Gall, 2013 for a discussion). The preceding example describes a situation in which pharmacologically augmenting memory would likely not result in enhanced cognition, at least in complex environments lacking explicit guidelines for effective performance. These are routine circumstances in which demands on cognition are high. But a great deal of cognition involves instances in which significant cues and appropriate responses are salient and predetermined, and irrelevant information is minimized. Under these conditions, enhanced encoding could be of great use in building or expanding FIGURE 5 | A “spaced trials effect” for LTP. (A) Theta burst stimulation (TBS) produces stable LTP. A single train of ten bursts was delivered to one input to the CA1b recording field after 20 min of collecting baseline synaptic responses to 3/min stimulation pulses in adult rat hippocampal slices. A second input to the same dendritic region received 3/min pulses throughout the session. Note that the potentiation of the experimental input did not decay over 5 h of recording (means ± s.e.m.s for seven slices); traces at right show representative baseline (black) and potentiated (gray) responses. (B) Effect of a second theta train (TBS2) applied at various times after TBS1. TBS2 produced no further increases in the slopes of the responses when delayed by 10, 30, or 40 min after TBS1, but doubled the magnitude of LTP when applied after a 60 min interval. Modified from Kramar et al. (2012b). Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 9
Lynch et al. Cognition enhancement in normal subjects cognitive structures.Thus,the effects of memory enhancement important roles played by adenosine in the periphery,including on cognition could prove to be situationally dependent with clear actions on the heart and lungs. benefits in some cases and neutral or even negative influence in Nicotinic receptors for acetylcholine are also found on glu- others. tamatergic terminals where they promote release (Wonnacott, 1997)and there is evidence that this increases network through- NETWORKS AND COGNITIVE ENHANCEMENT put (Gioanni et al.,1999).Alpha7-containing and alpha4/beta2 Discussions of neurobiological processes underlying cognition subtypes of the receptors both appear to be effective in this regard inevitably begin with the immensely complicated networks (Dickinson et al.,2008).However,the situation is complicated formed by cortical neurons,if for no other reason than a lack by the likelihood that compounds targeting nicotinic receptors of realistic alternatives.This fundamental idea suggests two paths act on cholinergic and GABAergic neurons as well (Wonnacott, to acute enhancement.First,improving throughput within estab- 1997;Alkondon and Albuquerque,2001);moreover,it is not clear lished networks should lead to faster computation and better that these receptors are present throughout glutamatergic net- utilization of cognitive time.Second,augmented synaptic com- works.In all,nicotinic receptor agonists and positive allosteric munication could allow for the transient assembly of larger modulators can be assumed to affect portions of excitatory cir- than normal networks (e.g.,incorporation of additional corti- cuitry in the telencephalon while at the same time modifying local cal regions)to deal with a particular problem,and thus the processing-via modulation of cholinergic input,interneurons, opportunity to execute more complex or even entirely novel com- and glutamatergic collaterals-at individual relays.Net effects putations.In this sense better throughput would add capabilities, will be complex but there is good evidence that the compounds perhaps the surest measure of cognitive enhancement.Increased acting on frontal networks enhance "top-down"mechanisms for plasticity might add a third route to enhanced cognition by allow- focusing attention (Sarter et al.,2009).Since pertinent drugs ing for construction of functional networks that would not likely are already in clinical trials(Holmes et al.,2011;Demeter and emerge under normal conditions;however,as noted in the pre- Sarter,2013),nicotinic compounds,and especially those target- ceding section,positive versions of such effects may be limited to ing the alpha4/beta2 receptor subtype concentrated in brain,have particular circumstances. to be seen as one of the most promising of current approaches to There are multiple manipulations that should result in cognitive enhancement. improved throughput.Communication between collections of The ampakine compounds described in the earlier section on neurons is greatly improved by synchronizing their activity,memory enhancement seem particularly appropriate for improv- something that is accomplished in the cortical telencephalon ing communication within and between cortical regions.Their by system-wide rhythms.These patterns are induced by diffuse mode of action has the virtue of relative simplicity:an exten- ascending projections from the lower brain and drugs that affect sive body of research from many laboratories has not uncovered these have predictable strong effects on rhythmic activity(Staubli any evidence for effects on targets other than AMPA receptors. and Xu,1995;Kowalczyk et al.,2013).But,as mentioned in And they produce the same facilitation of fast,excitatory trans- the discussion of memory,the diffuse systems influence a broad mission after peripheral administration as seen with infusions range of brain functions including ones that are vital to survival. into brain slices.Indeed,ampakines appear to be the only agents And so,as in the case of memory,they do not represent a promis- so far shown to cause comparable in vitro/in vivo facilitation ing avenue toward enhancement in high functioning individuals. of EPSPs.These points lead to two critical experimental ques- A more likely approach would be to increase transmitter release tions.First,does increasing monosynaptic transmission result in or post-synaptic responses to transmitter binding at the gluta- greater output from a polysynaptic network?This might seem to matergic connections used for the great bulk of intra-cortical be a foregone conclusion but each step in a series of neuronal communication. stations has local processing mechanisms (relays are not passive Adenosine,which depresses glutamate release via presynap- transferal points)dominated by an impressive collection of dif- tic Al receptors (Dunwiddie and Haas,1985),is increased in ferent types of inhibitory interneurons.These inhibitory elements the extracellular environment during repetitive firing by two respond both to inputs directly and to discharges from principle mechanisms:rapid release from post-synaptic neurons followed (glutamatergic)neurons;they also form complex local networks by slower release of ATP from glia which is then converted to among themselves.It is therefore possible that strong inputs are adenosine by ecto-5'-nucleotidase,an enzyme located on glial dampened and normalized to a degree such that the second stage membranes (Klyuch et al.,2012;Wall and Dale,2013).These of a network may not pass on a larger than normal signal in observations represent a significant part of the tripartite model the presence of an ampakine.Second,assuming augmentation (terminal bouton,spine,astrocyte)of fast,excitatory transmis- of the signal does occur,what are the functional consequences of sion (Araque et al.,1999).Selective antagonists of the Al receptor enhanced network throughput? increase glutamate release in slices and these compounds do Brain slices provide for the simplest and most compelling indeed reverse impairments in LTP in slices of middle-aged rat tests for circuit behavior because anatomically precise stimulation hippocampus (Rex et al.,2005).However,despite evidence that and recording is possible and extrinsic modulatory (cholinergic, the compounds enter the brain (Wall and Dale,2013),there etc.)inputs(cholinergic,serotonergic,etc.)that might influence has been surprisingly little work on in vivo effects after periph- downstream responses are excluded.Work of this kind has estab- eral administration.Perhaps the lack of interest with regard to lished that weak facilitation of monosynaptic transmission with network operations reflects understandable concern about the an ampakine results in a greatly amplified response from the Frontiers in Systems Neuroscience www.frontiersin.org May 2014 Volume 8|Article 90 10
Lynch et al. Cognition enhancement in normal subjects cognitive structures. Thus, the effects of memory enhancement on cognition could prove to be situationally dependent with clear benefits in some cases and neutral or even negative influence in others. NETWORKS AND COGNITIVE ENHANCEMENT Discussions of neurobiological processes underlying cognition inevitably begin with the immensely complicated networks formed by cortical neurons, if for no other reason than a lack of realistic alternatives. This fundamental idea suggests two paths to acute enhancement. First, improving throughput within established networks should lead to faster computation and better utilization of cognitive time. Second, augmented synaptic communication could allow for the transient assembly of larger than normal networks (e.g., incorporation of additional cortical regions) to deal with a particular problem, and thus the opportunity to execute more complex or even entirely novel computations. In this sense better throughput would add capabilities, perhaps the surest measure of cognitive enhancement. Increased plasticity might add a third route to enhanced cognition by allowing for construction of functional networks that would not likely emerge under normal conditions; however, as noted in the preceding section, positive versions of such effects may be limited to particular circumstances. There are multiple manipulations that should result in improved throughput. Communication between collections of neurons is greatly improved by synchronizing their activity, something that is accomplished in the cortical telencephalon by system-wide rhythms. These patterns are induced by diffuse ascending projections from the lower brain and drugs that affect these have predictable strong effects on rhythmic activity (Staubli and Xu, 1995; Kowalczyk et al., 2013). But, as mentioned in the discussion of memory, the diffuse systems influence a broad range of brain functions including ones that are vital to survival. And so, as in the case of memory, they do not represent a promising avenue toward enhancement in high functioning individuals. A more likely approach would be to increase transmitter release or post-synaptic responses to transmitter binding at the glutamatergic connections used for the great bulk of intra-cortical communication. Adenosine, which depresses glutamate release via presynaptic A1 receptors (Dunwiddie and Haas, 1985), is increased in the extracellular environment during repetitive firing by two mechanisms: rapid release from post-synaptic neurons followed by slower release of ATP from glia which is then converted to adenosine by ecto-5 -nucleotidase, an enzyme located on glial membranes (Klyuch et al., 2012; Wall and Dale, 2013). These observations represent a significant part of the tripartite model (terminal bouton, spine, astrocyte) of fast, excitatory transmission (Araque et al., 1999). Selective antagonists of the A1 receptor increase glutamate release in slices and these compounds do indeed reverse impairments in LTP in slices of middle-aged rat hippocampus (Rex et al., 2005). However, despite evidence that the compounds enter the brain (Wall and Dale, 2013), there has been surprisingly little work on in vivo effects after peripheral administration. Perhaps the lack of interest with regard to network operations reflects understandable concern about the important roles played by adenosine in the periphery, including actions on the heart and lungs. Nicotinic receptors for acetylcholine are also found on glutamatergic terminals where they promote release (Wonnacott, 1997) and there is evidence that this increases network throughput (Gioanni et al., 1999). Alpha7-containing and alpha4/beta2 subtypes of the receptors both appear to be effective in this regard (Dickinson et al., 2008). However, the situation is complicated by the likelihood that compounds targeting nicotinic receptors act on cholinergic and GABAergic neurons as well (Wonnacott, 1997; Alkondon and Albuquerque, 2001); moreover, it is not clear that these receptors are present throughout glutamatergic networks. In all, nicotinic receptor agonists and positive allosteric modulators can be assumed to affect portions of excitatory circuitry in the telencephalon while at the same time modifying local processing—via modulation of cholinergic input, interneurons, and glutamatergic collaterals—at individual relays. Net effects will be complex but there is good evidence that the compounds acting on frontal networks enhance “top-down” mechanisms for focusing attention (Sarter et al., 2009). Since pertinent drugs are already in clinical trials (Holmes et al., 2011; Demeter and Sarter, 2013), nicotinic compounds, and especially those targeting the alpha4/beta2 receptor subtype concentrated in brain, have to be seen as one of the most promising of current approaches to cognitive enhancement. The ampakine compounds described in the earlier section on memory enhancement seem particularly appropriate for improving communication within and between cortical regions. Their mode of action has the virtue of relative simplicity: an extensive body of research from many laboratories has not uncovered any evidence for effects on targets other than AMPA receptors. And they produce the same facilitation of fast, excitatory transmission after peripheral administration as seen with infusions into brain slices. Indeed, ampakines appear to be the only agents so far shown to cause comparable in vitro/in vivo facilitation of EPSPs. These points lead to two critical experimental questions. First, does increasing monosynaptic transmission result in greater output from a polysynaptic network? This might seem to be a foregone conclusion but each step in a series of neuronal stations has local processing mechanisms (relays are not passive transferal points) dominated by an impressive collection of different types of inhibitory interneurons. These inhibitory elements respond both to inputs directly and to discharges from principle (glutamatergic) neurons; they also form complex local networks among themselves. It is therefore possible that strong inputs are dampened and normalized to a degree such that the second stage of a network may not pass on a larger than normal signal in the presence of an ampakine. Second, assuming augmentation of the signal does occur, what are the functional consequences of enhanced network throughput? Brain slices provide for the simplest and most compelling tests for circuit behavior because anatomically precise stimulation and recording is possible and extrinsic modulatory (cholinergic, etc.) inputs (cholinergic, serotonergic, etc.) that might influence downstream responses are excluded. Work of this kind has established that weak facilitation of monosynaptic transmission with an ampakine results in a greatly amplified response from the Frontiers in Systems Neuroscience www.frontiersin.org May 2014 | Volume 8 | Article 90 | 10