Molecular Determinants of Synaptic Function: Molecules and Models

The synapse proteome and phosphoproteome: a new paradigm for synapse biology

S.G.N. Grant


Synapse proteomics has recently resulted in a quantum leap in knowledge of the protein composition of brain synapses and its phosphorylation. We now have the first draft picture of the synapse, comprising ∼1000 proteins. This is not matched by available methods of functional analysis either in reduced systems or in whole animals. Fewer than 20% of synapse proteome proteins have a known function in the nervous system. A concerted effort is required to establish new technical approaches before we can understand the diversity of functions conferred by the synapse proteome on the synapse, the neuron and the animal. This review will highlight this change in knowledge and discuss current technical and interpretative limitations challenged by synapse proteomics.

  • phosphoproteome
  • protein function
  • proteome
  • synapse
  • synaptic plasticity
  • synaptosome


In this era of fully sequenced genomes, the daily life of molecular biologists has been transformed by the simple expediency of looking up the sequence of the gene of interest and knowing its structure and similarity to all other genes. The era of cDNA cloning of genes with exciting functions, such as those encoding hormones and receptors, has now passed and it is the functions of the already sequenced genes that we must now discover. The presence of the genome sequence and the proteome, derived by the translation of open reading frames, presents the investigator with a set of molecules resembling an enormous list of parts for some model that requires assembling, although the instructions for this model are missing. Finding out what the instructions say and mean is now the task, rather than finding the pieces.

Proteomic profiling refers to the identification of the protein parts list of some entity, typically a multiprotein complex or subcellular organelle [1,2]. As shown in Table 1, many mammalian cell components have been profiled, and the numbers of proteins range from dozens to over 1000, with the typical range in the hundreds for many organelles. This complexity is surprising to most scientists, most of whom have worked with single-gene-based approaches. It is worth noting that these numbers, although more than expected, are likely to be underestimates because of technical limitations of sensitivity. High degrees of molecular complexity are more likely to be the rule than the exception.

View this table:
Table 1 Number of proteins in PSD, NRC and other organelles

The results of global organellar proteomic profiling studies of distinct organelles are shown (data are taken from [2]). Note that the PSD is of the highest complexity and neurotransmitter receptor complexes (NMDA receptor/MASC) are between the spliceosome and nucleolus.

Functional understanding of a molecularly complex organelle will also require a shift in expectations. In most cases, each organelle is identified with a single general function; for example, ribosomes translate mRNA into protein, or spliceosomes splice mRNA. Why are there hundreds of components if there is a seemingly simple or single function? Each part must have a very specific role and all parts need to be carefully orchestrated and co-regulated to produce a productive outcome. Thus orchestration and subtlety of function need to be found within the organization of these molecular profiles.

Profiling the synapse proteome and phosphoproteome

Proteomic profiling technology is rapidly evolving and is reviewed elsewhere [1]. In outline, the key steps are the biochemical isolation of the entity in question, followed by cleavage of proteins into peptides whose mass can be identified with a mass spectrometer, and by comparison of this mass with sequence from genome databases, the peptide can be identified as corresponding to a protein encoded by a specific gene. Thus the output of the profiling experiment is a list of proteins.

For several decades, a simple biochemical separation technique has been available for the isolation of relatively pure preparations of mammalian synapses [3]. These preparations, called synaptosomes, contain the pre-synaptic terminal and post-synaptic terminal and the machinery for release and response to neurotransmitters. Synaptosomes can be sub-fractionated into synaptic vesicles (containing neurotransmitters), the PSD (post-synaptic density, containing neurotransmitter receptors) and other fractions. In addition, complexes such as neurotransmitter receptor, adhesion protein and signal transduction complexes can be isolated from the synaptosomes using affinity procedures [46]. We have reported and compiled these sets of synapse profiling data from many laboratories [4,614] to draw up comprehensive lists of proteins found in mammalian synapses [15].

The mammalian synapse appears to have in excess of 1000 protein components [15]. A total of 1124 proteins from the PSP (post-synaptic proteome), comprising the PSD, glutamate receptor complexes and also components of the pre-synaptic membrane tethered to the post-synaptic membrane (synaptic junctional components), have been reported. We found 186 proteins within glutamate receptor – MAGUK (membrane-associated guanylate kinase) protein complexes {NRC [NMDA (N-methyl-D-aspartate) receptor complex]/MASC (MAGUK-associated signalling complex)} and nine proteins within AMPA (α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid) receptor complexes and 620 PSD proteins [15,16]. Figure 1 summarizes these different datasets and where proteins found in these different sets are common. This Venn diagram presents the model that the PSP comprises multiple multiprotein complexes that are interconnected.

Figure 1 Venn diagram illustrating the overlap of three glutamate receptor complexes with the PSD datasets

A total of 1124 proteins (see Table 2) are sorted into their respective complexes: NRC/MASC, AMPA-R (AMPA receptor complex), mGluR-5 (metabotropic receptor complex) and PSD. Further data can be found in [15].

A limitation of these synaptosome profiling experiments is that they require milligrams to grams of starting material, and brain regions or whole brains are often used. Thus anatomical diversity must be considered in interpreting the results. The ideal situation would be if the composition of individual synapses could be determined, or if there was a uniform synapse population by analogy with clonal cells. Neither of these is currently possible, and therefore one must rely on less accurate approaches such as immunostaining of cultured neurons and brain sections. Since this has not been done on a scale commensurate with the proteomic set numbers, the conclusions are not available. However, interpretation of small sets of these proteins such as NMDA receptor subunits or MAGUK proteins suggests a broadly common pattern of expression, albeit with many exceptions. For example, NMDA receptor subunits NR1, NR2A and NR2B are found in all forebrain and hindbrain regions, although the levels of NR2A and NR2B are low in the adult cerebellum, while high in hippocampus and other forebrain structures. Another limitation is that the current profiling experiments do not take into account the dynamics or rate of turnover of the proteins. Techniques to approach these issues will be discussed below.

Approximately 10% of the 1124 proteins reported had been previously reported as having post-synaptic localization, and subsequent testing of 48 proteins showed that greater than 80% could be validated at the level of Western blotting [15]. These results indicate that profiling of synaptosome fractions has dramatically increased (>5-fold) the number of proteins known at the synapse. A similar picture holds true for the synapse phosphoproteome, where large-scale identification of in vivo phosphorylation sites identified hundreds of sites, of which 92% were novel [14]. Thus profiling experiments have resulted in a huge increase in novel data on the composition and modification of proteins at the mammalian synapse.

How much do we know about synapse protein function?

Historically, functional studies of synaptic proteins were initiated by their identification and followed up with a manipulation (using pharmacological, genetic or molecular biological tools) and assay of neuronal function. The 5–10-fold increase in identification of proteins and phosphorylation sites from profiling experiments is not matched with functional neurobiological data, and therefore we must conclude that we know the function of a small percentage (<10–20%) of synapse proteins. The molecular understanding of the synapse is still in its very early phases.

Charles Sherrington not only coined the word ‘synapse’ (1897), but also described its electrophysiological functions, and, ever since, the vast majority of studies have focused on electrophysiological approaches to function. Assigning electrophysiology data to proteins has been heavily dependent on the function of the molecule. In some cases, heterologous cell-expression systems, such as Xenopus oocytes, for expressing cDNAs encoding ion channels, have been very useful. However, synapse proteomic profiling experiments show that less than 10% of proteins are channels and receptors and the majority of proteins need other approaches. Table 2 shows the functional classification of proteins found in proteomic profiling experiments of synaptic proteins. This shows that the synapse comprises many functional classes of protein and that there is probably the need for specialist techniques relevant to specific classes of protein (e.g. channels) and generic techniques that can be applied to all proteins.

View this table:
Table 2 Functional classification of synapse proteome proteins

Proteins found in PSD and MASC profiling experiments are classified into functional groups. Note the diversity of functions and that ion channels are a small percentage. Data are taken from [15].

One such generic technique is the creation of loss-of-function alleles of genes in the mouse using homologous recombination in embryonic stem cells. The first example of a gene that hitherto had no known role in synapse biology was that encoding the tyrosine kinase Fyn [17]. Null mutation in the fyn gene resulted in mice with a learning deficit and impaired synaptic plasticity. Since then, close to 200 different genes have been created by the knockout approach and their synaptic physiology examined (A. Howell, M. Cumiskey, A.J. Pocklington, J.D. Armstrong and S.G.N. Grant, unpublished work). In a subset of these data, 128 mutations out of 166 in total (77%) have effects on synaptic transmission or plasticity. This indicates that (i) many genes contribute to the function of the synapse and (ii) the total number of genes involved will be much higher. In other words, these genetic studies are in agreement with the proteomic profiling experiments in that the biology of the synapse involves many genes, certainly hundreds. Another logistically important fact is that the rate of gene knockout experiments reporting new synaptic phenotypes is approx. 25 per annum. At the current rate, it will take 40 years of knockout studies to complete a systematic study of the synapse! Clearly, this is too slow and new strategies or scaling are required.

Problems with existing functional models and interpretations

Problems in understanding the role of molecules are more than simply a problem of scale. The reductionist approaches of single-gene manipulations have presented interpretative problems that I believe are an intrinsic feature of the molecular complexity of the synapse. To illustrate this at several levels, examples will be presented from the study of synaptic plasticity.

First, a brief introduction to synaptic plasticity. Patterns of nerve cell impulses propagated as action potentials arrive at the pre-synaptic terminal of excitatory synapses and elicit release of glutamate-containing vesicles, which activate post-synaptic receptors that can be generally described as having two broad functions. The receptors that mediate the post-synaptic depolarizations that are responsible for initiating the action potential are Na+-permeable receptors known as AMPA receptors and those that activate signalling and plasticity mechanisms are NMDA receptors and mGluRs (metabotropic glutamate receptors). Proteomic profiling of these receptors isolated from brain reveals that NMDA receptors and mGluRs are assembled into large complexes of 186 proteins measuring 2–3 MDa, and AMPA receptors into much smaller complexes of approximately ten proteins [4,15]. Models of synaptic plasticity from the late 1980s proposed that neurotransmitter activation of NMDA receptors allowed Ca2+ influx, which activated CaMKII (Ca2+/calmodulin kinase II), which then phosphorylated Ser831 on the GluR1 subunit of the AMPA receptor [19,20]. The hypothesized net effect of this was to enhance AMPA receptor function, resulting in increased synaptic strength. There were many correlative experiments that were consistent with the hypothesis, such as those interfering with the kinase [21,22] and observed changes in the level of Ser831 phosphorylation. However, a key experiment was the mouse knockin, where alanine was substituted for Ser831, and the study of synaptic physiology and learning in these mice showed very little effect [23]. The simplest explanation is that the phosphorylation site was not physiologically relevant. CaMKII is known to phosphorylate dozens of proteins [24] (and presumably dozens more that are unknown) and it now seems unlikely on simple probabilities that the original model of Ser831 would have been correct and accounts for the changes in synaptic strength alone.

A similar type of problem was also seen in interpretation of experiments on trafficking of the AMPA receptor. In addition to the phosphorylation model of the AMPA receptor mentioned above, a second dominant model over the last decade has been that the GluR1 subunit of the AMPA receptor is dynamically trafficked and inserted into the post-synaptic membrane following activation of NMDA receptors [25]. One specific molecular hypothesis was that the C-terminal residue of the GluR1 subunit, which forms a binding sequence for PDZ domains [presumably of SAP97 (synapse-associated protein 97)] was necessary for the regulated trafficking of AMPA receptors and subsequent physiological modulation of synaptic transmission. However, the point mutation of this residue in knockin mice, which rendered GluR1 incapable of binding PDZ domains, had no effect on synaptic electrophysiology [25]. Again, the power of genetics for testing these models has proven its worth.

A recent paper strongly stated that the insertion/trafficking of GluR1 was proven to be the mechanism of learning in an experiment where the entire 81-amino-acid C-terminal domain of the receptor was expressed in neurons [26]. This fragment is rich in protein–protein interaction sites [including PDZ, Forkhead-associated and 14-3-3 domains and ERK (extracellular-signal-regulated kinase) and PDK (phosphoinositide-dependent kinase) docking sites] and phosphorylation sites for several kinases [PKA (protein kinase A), PDK1 and CaMKII]. The overexpression of this ‘plasticity block’ vector will interfere with these kinases and protein interactions. The conclusion from this study is that this ‘plasticity block’ vector acted by interfering with the endogenous trafficking of GluR1 to mediate its effect on behaviour. As was the case above for Ca2+-CaMKII, phosphoproteome profiling and peptide array experiments strongly support the idea that post-synaptic kinases phosphorylate many substrates and these substrates can be found in different functional classes of proteins (e.g. trafficking proteins, receptors, cytoskeleton and translational regulators) [14]. Therefore the overexpression with inhibition of many different enzymes and protein interactions cannot be simply interpreted.

In a detailed survey of the strength of phenotypes in knockout mice, we found that 75% or more knockouts have only partial effects on synaptic plasticity (A. Howell, M. Cumiskey, A.J. Pocklington, J.D. Armstrong and S.G.N. Grant, unpublished work). As mentioned above, there are >100 genes involved with synaptic plasticity and most have partial effects. This robustness to single-gene effects indicates that there must be subtlety and redundancy within the signalling mechanisms. As described elsewhere, the model most consistent with these observations is not the one with simple linear biochemical pathways, but one where there is a complex interplay in a highly connected molecular network [16,2729].

Why have so many molecules? Is it just so that the system can be more robust? If a biological mechanism is important to the animal, then robustness offers an evolutionary advantage. Learning and memory, which are behavioural forms of plasticity, are important for animal survival and it should be no surprise if these mechanisms are robust. However, in addition to the robustness of many interacting molecules, there is the secondary advantage that the molecules can have subtly different functions, which may enhance the repertoire of behaviours. To find examples of this, it is best to examine a gene family and compare the phenotypes of mice carrying mutations in the different families. We have created mutations in the adaptor proteins (MAGUK proteins) PSD-95 [3033] and SAP102 (synapse-associated protein 102), which directly bind the cytoplasmic domains of NMDA receptors. Comparison of the phenotypes of these mutants shows overlapping but different behaviours, consistent with this model. This raises the issue that to understand the function of the synapse proteome, we need to perform analyses at many levels – from genes to cognition.

Scaling functional studies of the synapse proteome

Functional studies of specific synaptic proteins can be performed at levels from gene to behaviour; however, almost all of these assays are low-throughput approaches. In addition to having multiple integrated approaches, the second important feature is to have a quantitative data and standard operating procedures [34] ( This is because the relative importance of a mutation can be explained by its connectedness in a molecular network and biochemical pathway. A comparison of experimental methods from approx. 90 different experiments on synaptic plasticity from different laboratories shows that all experiments were performed differently (A. Howell, M. Cumiskey, A.J. Pocklington, J.D. Armstrong and S.G.N. Grant, unpublished work). Although in these individual experiments the investigator concluded that the molecule in question had some role, it is not possible to compare the data to determine if any molecule is more important than any other. This illustrates the fact that finding new mutants with an already described phenotype is no longer an exciting activity – it is now important to synthesize the knowledge of these different mutants into mechanistic models that can be validated or refuted.

How can the community of neuroscientists systematically study the synapse proteome and what are the range of experiments that might be considered? This is not the place for a detailed description of possible strategies, and several examples will be addressed at biochemical, physiological and behavioural levels.

Profiling of synaptic proteins sets the stage for validation and quantification experiments. We require the systematic study of subcellular localization and anatomical localization of the mRNA and proteins in the nervous system of a mammal. Large-scale gene expression mapping is currently under way using in situ hybridization and transgenic mouse approaches [35] (; This is complemented by microarray studies of microdissected brain regions [36]. These methods are now operating at a scale that should allow data on 1000 proteins in less than 2 years. At a finer level of resolution, the ultrastructural localization of proteins within the synapse and PSD will ultimately be important. To date, this has been done using antibodies and the immunogold technique and does not appear scalable. New techniques, perhaps involving epitope tagging of transgenes, may offer an antibody-independent and systematic platform.

The protein itself can be mapped systematically at the levels of protein–protein interaction using yeast and bacterial binary-interaction approaches [37]. Manually curated protein interaction data have shown that it is possible to construct large maps of NRC/MASC [29] and PSD. This may be useful when superimposed on phosphoproteome mapping of kinase–substrate interactions using peptide arrays [14]. These and other large-scale, recombinant-based methods can rapidly match the scale of the synapse proteome. Dynamic studies of protein turnover and phosphorylation using new MS-labelling methods are also well suited.

Electrophysiology and behaviour are more problematic and require a more serious investment in resources. A generic system, such as mouse knockouts, has the advantage that both physiology and behaviour can be measured in the same organism. Moreover, the scale of knockout technology is such that it will be possible to obtain knockouts of 1000 genes in a span of 3–5 years [38,39]. Brain slice electrophysiology, which has been a mainstay of synaptic physiology for decades, can now be performed in a medium-throughput multiplexed manner using Micro Electrode arrays [40,41]. In this system, the slice sits on a glass slide with 64 printed electrodes and this slide plugs into an amplifier and stimulating head for acquisition of the data. Mouse behavioural batteries can be scaled to deal with dozens to hundreds of mutants per year.

Bioinformatics and databases of these multiple forms of data will provide a valuable community resource and provide entirely new sets of data or functional annotation to the synapse proteome. From this network, models and other statistical approaches could be applied [2729]. Perhaps it is inevitable that the synapse proteome will eventually be studied thoroughly but not systematically simply by the progressive accrual of data as performed by the current community. On the other hand, centres with large-scale expertise could be established to provide a far faster and more cost-effective option. Not only would this facilitate development of models of disease and therapeutics, but it would also allow individual laboratories to focus on bespoke and specific problems.


Supported by the Genes to Cognition Programme of the Wellcome Trust. Thanks to Mark Collins for the Figures.


  • Molecular Determinants of Synaptic Function: Molecules and Models: Focused Meeting held at Chilworth Manor and the University of Southampton, U.K., 22–23 September 2005. Organized and edited by L. Holden-Dye (Southampton, U.K.), V. O'Connor (Southampton, U.K.) and F.A. Stephenson (School of Pharmacy, London, U.K.). Sponsored by Abcam, Affiniti Research Products Ltd, BBSRC (Biotechnology and Biological Sciences Research Council), Capsant Neurotechnologies, Eli Lilly, Ferring Research Ltd, GlaxoSmithKline, Merck Sharp & Dohme Research Laboratories, Portland Press Ltd, Syngenta and UCB Pharma.

Abbreviations: AMPA, α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid; CaMKII, Ca2+/calmodulin kinase II; MAGUK, membrane-associated guanylate kinase; mGluR, metabotropic glutamate receptor; NMDA, N-methyl-D-aspartate; NRC/MASC, NMDA receptor complex/MAGUK-associated signalling complex; PDK, phosphoinositide-dependent kinase; PSD, post-synaptic density; PSP, post-synaptic proteome


View Abstract