Synaptic logistics: Competing over shared resources

High turnover rates of synaptic proteins imply that synapses constantly need to replace their constituent building blocks. This requires sophisticated supply chains and potentially exposes synapses to shortages as they compete for limited resources. Interestingly, competition in neurons has been observed at different scales. Whether it is competition of receptors for binding sites inside a single synapse or synapses fighting for resources to grow. Here we review the implications of such competition for synaptic function and plasticity. We identify multiple mechanisms that synapses use to safeguard themselves against supply shortages and identify a fundamental neurologistic trade-off governing the sizes of reserve pools of essential synaptic building blocks.


Introduction
Competition over limited resources is an essential aspect of life: from trees in a forest competing for sunlight to human societies competing for access to fossil fuels. Indeed, competition is pervasive and exists at all scales, all the way down to subcellular compartments: the chemical synapses in our brains, which must compete for resources that allow them to survive and grow stronger. Estimates suggest that excitatory synapses in the forebrain are composed of over 2000 different types of proteins (Sheng and Kim, 2011). Ensuring that all these are present in the right quantity at a neuron's thousands of synapses is a formidable logistic challenge.
Lynch and colleagues in the late 1970s made the first observations that groups of synapses compete with each other in a single neuron. They recorded in patch clamp CA1 pyramidale neurons of the hippocampus while stimulating independent afferent pathways onto the neuron, namely, inputs on the apical (Schaffer collaterals in stratum radiatum) or the basal (commissural fibers in stratium oriens) dendritic field. Interestingly, they found that potentiation of one afferent pathway leads to depression of the target cell's responses to a second test input on the other afferent pathway (Lynch et al., 1977). They called this phenomenon heterosynaptic plasticity as opposed to homosynaptic plasticity that occurs at the stimulated pathway. The strength of heterosynaptic plasticity is dependent on the frequency of stimulation. Additionally, while potentiation is restricted to the subset of activated synapses the resulting depression is neuron wide (Dunwiddie and Lynch, 1978). Similar experiments in the presence of a protein inhibitor (anisomycin or emetine) exacerbated heterosynaptic depression in the nonstimulated pathway (Fonseca et al., 2004). These results pointed to the central role of protein availability in the phenomenon. Since these early studies, heterosynaptic plasticity has been reported by numerous labs and has been found to occur beyond the CA1 region of the hippocampus (Oh et al., 2015), including in the cortex (Field et al., 2020). Thus, heterosynaptic plasticity is a general phenomenon that might reflect, at least in part, the inner challenges faced by neurons in order to maintain function from a finite amount of resources (reviewed in detail in this special issue (Wagle et al., 2023)).
A first challenge arises because most synaptic proteins have limited lifetimes of just a few days (Cohen et al., 2013;Fornasiero et al., 2018;Dörrbaum et al., 2018) and need to be constantly replaced to maintain a synapse's function. Second, a synapse needs to be provided with sufficient energy to run the processes of signal transmission and protein renewal. Third, synapses are plastic and store information by, e.g., increasing their efficacy of signal transmission through growing in size. The "raw materials" necessary for this physical growth need to be provided in sufficient quantity and in a timely fashion. This applies to both sides of a synapse, the pre-and the postsynapse, because both undergo coordinated changes (Harris and Stevens, 1989;Bartol et al., 2015). Indeed, electron microscopy on rodent brains has long revealed a strong correlation between the sizes of postsynaptic elements (e.g., postsynaptic density (PSD) area, spine head volume) and presynaptic ones (e.g. number of synaptic vesicles). Importantly, proteins or the ingredients to produce them, energy, and even physical space to grow are limited. This induces competition among synapses for these finite resources. How does the brain manage this competition and what are the fundamental limitations, trade-offs, and failure modes resulting from such competition? Here, we review what is known about how synapses compete over shared resources and its functional ramifications. Our emphasis will be on the competition for synaptic building blocksthe proteins that are required to grow and maintain functioning synapses. Competition for energy supply or physical space (e.g., dendritic real estate (Ryglewski et al., 2017)) are likely to be of similar importance, however.
The remainder of this article is organized as follows. We will begin by discussing the two synaptic supply chains that provide the presynaptic and the postsynaptic compartments with the necessary goods. Then we will introduce the different strategies that neurons use to manage inevitable fluctuations in supply and demand in order to avoid shortages and loss of function. This leads us to introduce the fundamental neurologistic tradeoff dictating how big of a reserve to maintain for critical components. This tradeoff affects how fierce the competition for resources will be. Next, we review which synaptic building blocks are likely to be bottlenecks within the synaptic supply chains. Then we will review a recent theoretical model that sheds light on the functional ramifications of this tradeoff for the specific case of synapses on the same piece of dendrite competing for neurotransmitter receptors. We conclude with a discussion of possible implications for neurodegenerative disorders.

The two synaptic supply chains
A synapse is composed of (at least) two compartments: a presynaptic one (sender) and a postsynaptic one (receiver). The presynaptic bouton is placed along the axon of the presynaptic neuron and releases neurotransmitter upon arrival of an action potential. The postsynaptic density contains neurotransmitter receptors that will bind the neurotransmitter and permit ions to flow through the postsynaptic neuron's membrane, changing its voltage. This completes the signal transmission across the synapse. While this basic communication scheme is rather simple, the pre-and postsynaptic machinery implementing it and endowing synapses with additional encoding abilities certainly is not (Reshetniak et al., 2020).
As discussed above, the proteins that form this complex pre-and postsynaptic machinery need to be replaced on a regular basis, necessitating a presynaptic and a postsynaptic supply chain. These supply chains comprise multiple routes and different protein species may make use of one or more of these, as illustrated in Fig Routes to the synapse. Schematic representation of an excitatory synaptic connection between a presynaptic neuron (light green) and a postsynaptic neuron (blue) and a local glial cell (purple). Synapses rely on two supply chains for maintaining proper functiona presynaptic and a postsynaptic one. Both comprise multiple routes. Three routes for the delivery of synaptic proteins are depicted: (1) long-distance shipping via intracellular transport of cargoes along microtubules (gray double lines) using motor proteins, (2) local protein synthesis directly at the synapse, and (3) local sharing of goods between different cell types.
route is defined for synaptic proteins originating in the soma. These need to be shipped over long distances to the synapse via either the axon (to presynaptic compartments) or via the dendrites (to postsynaptic compartments). A second route exists for proteins whose mRNA is present directly at or near the synapse so that they can be produced locally in either the presynaptic or the postsynaptic compartment (Cajigas et al., 2012;Shigeoka et al., 2016;Tushev et al., 2018;Hafner et al., 2019;Glock et al., 2021). A third route to the presynaptic compartment was recently suggested, as postsynaptic neurons and glial cells appear to transfer, respectively, proteins and ribosomes to axons (Pastuzyn et al., 2018;Müller et al., 2018). Whether such a third route also exists for the postsynaptic compartment appears to be unknown. Importantly, from the point of view of a synapse the origin of a protein or mRNA for its capture is most likely irrelevant. Nonetheless, a key aspect is to what extent a route can or cannot lead to an increase in the local concentration of proteins of interest and thus promote their synaptic incorporation. Each route has advantages and disadvantages. A main difference of somatic versus local (i.e. at or near the synapse) protein synthesis is the short-term destination of the protein: While proteins arriving from the soma first enter a shared dendritic or axonal "pool" before being captured by a particular synapse, locally synthesized proteins may be captured quickly where they are produced. Indeed, local production of proteins has the advantage that required proteins can be produced on the spotmultiple copies per single mRNAavoiding transportation costs and delays associated with moving single units (Fonkeu et al., 2019). However, local production of proteins implies that the entire translation machinery (ribosomes, mRNA, tRNA, amino acids, ATP) needs to be present locally. Importantly, just like the synaptic proteins themselves, all these components of the translation machinery also have limited lifetimes and need to be replaced continuously, which is also associated with certain costs. It is interesting to note that because the half-life of synaptic proteins, 5-7 days in vitro (Cohen et al., 2013;Dörrbaum et al., 2020), greatly exceeds the residency time of those proteins in a particular synapse, typically on the order of minutes to hours (Sturgill et al., 2009;Ziv and Fisher-Lavie, 2014), locally produced proteins ultimately join the dendritic or axonal pool and become available for capture by other synapses (Tsuriel et al., 2006;Gray et al., 2006).
It is important to note that in mammalian dendrites the copy number of mRNAs is significantly lower than the number of synapses (Kosik, 2016). For instance, hippocampal pyramidal neurons have an average spine density of 1-2 per μm. Even the most abundant mRNA in dendrites, namely the mRNA coding for CamKIIα, has a density of less than 1 mRNA per 2 μm (Tushev et al., 2018;Fonkeu et al., 2019). How do neurons deal with this apparent shortage? Electron microscopy study of dendrites from the CA1 region of the hippocampus show that under basal conditions only roughly 12 % of spines contain polyribosomestypically considered as a marker of active translation (Ostroff et al., 2002). Interestingly, 2 h after tetanic stimulation this number goes up to 39 % and density of polyribosomes in dendritic shafts decreases. More recently using live imaging, Donlin-Asp and colleagues were able to show that untranslated regions (3'UTRs) on mRNA coding for synaptic proteins like CamKIIα or PSD-95 play a central role in dendritic localization of those mRNA and allow their capture at or near active synapses (Donlin-Asp et al., 2021). Importantly, experimental and computational data show that the activation of protein synthesis in one location in the dendrite results in a new protein spread (local increase in protein concentration) that affects synapses within a dendritic domain of 18 μm Fonkeu et al., 2019). Together these data suggest that as long as a minority of spines require local protein synthesis at any given time, a small number of mRNA might be sufficient. This is because the pool of available mRNAs and ribosomes in dendrites is highly mobile and can be recruited on-demand. If the number of synapses in need for local protein synthesis increases beyond a certain threshold, however, the needs are likely not to be met by the demand. The consequences of mRNA local recruitment or shortage on homo-and heterosynaptic plasticity in dendrites are discussed in more detail in (Wagle et al., 2023).
Proteins produced in the somatic region prior to being captured by a synapse diffuse over long distances and/or are moved inside neurites by active transport on the microtubule network. Specific sets of motor proteins transport mRNAs, proteins, and entire organelles along this network, reviewed in (Burute and Kapitein, 2019), and there exist fundamental structural differences in these networks between axons and dendrites (Tas et al., 2017). Live imaging experiments studying the movement of various cargoes such as intracellullar vesicles containing AMPA-type glutamate receptors (AMPARs) or mRNA granules along microtubules in dendrites have revealed that individual cargoes are not targeted to specific postsynapses (Hangen et al., 2018;Park et al., 2010). In fact, dendritic cargoes (and a subset of axonal cargoes (Maday et al., 2014)) are moving in both antero-and retrograde directions as described by the Sushi-belt model (Doyle and Kiebler, 2011). Active synapses are known to immobilize these cargoes in their vicinity via, for instance, the phosphorylation of KIF17 by Ca(2+)/calmodulin-dependent protein kinase II (CaMKII) (Guillaud et al., 2003;Hanus et al., 2014). In order to ensure a sufficient supply of these cargoes even during periods of high demand (e.g. while the synapse undergoes LTP) a neuron may need to over-produce synaptic components. In the following section, we will look at this logistic challenge in more detail.

The fundamental neurologistic trade-off
In our economy, it is common for both the supply and demand of goods to fluctuate. Fluctuations of supply have many origins that are linked to the goods' supply routes. Fluctuations of demand can be local or global. The supply of synaptic proteins to synapses shares interesting analogies with these economic challenges. Each supply route of the two supply chains for synaptic proteins discussed above has its own characteristic fluctuations. One fundamental source of fluctuations in supply is, for instance, that transcription and translation both happen in bursts (Raj and van Oudenaarden, 2008;Wu et al., 2016). Similarly, demand will vary based on, e.g., fluctuations of overall neural activity levels or the local induction of long-term plasticity triggering growth or shrinkage of specific synapses. Importantly, such fluctuations of supply and demand invariably create a risk: if an unusually high demand occurs during a time of unusually low supply, a shortage may occur. The supply no longer meets the demand and normal function is jeopardized. For example, a shortage of readily releasable synaptic vesicles filled with neurotransmitter or a shortage of postsynaptic neurotransmitter receptors can both significantly reduce the efficacy of synaptic transmission. While presynapses are densely packed with synaptic vesicles, strikingly, only a small fraction of about 10 % of these vesicles can be used for immediate release (Kaeser and Regehr, 2017). For slow firing frequencies of up to 10 Hz at physiological temperature, neurons seems to avoid vesicle depletion by increasing their vesicle recycling rate (Fernández-Alfonso and Ryan, 2004;Rey et al., 2020). Nonetheless, single neuron recordings from macaque in multiple cortical areas (i.e. prefrontal cortex, anterior cingulate cortex, and pre-supplementary motor area) show firing rates ranging from a few Hz at rest to around 60 Hz during task performance (Gavrilov et al., 2017). Thus, it seems that presynaptic boutons may function close to the breaking point of the supply chain. It is worth mentioning that many studies showed that synaptic vesicles can be shared between adjacent presynaptic boutons (Chen et al., 2008;Darcy et al., 2006;Fernandez-Alfonso and Ryan, 2008;Krueger et al., 2003;Westphal et al., 2008). While this phenomenon seems to allow for the incorporation of new recycling vesicles in active synapses, the time scale is of the order of minutes (Staras et al., 2010).
Does a synapse need to safeguard itself against such risk of a loss of function and, if so, how does it do so? Arguably, a temporary loss or reduction of synaptic function may not matter much at the system level.
After all, many if not most cortical synapses are rather unreliable and have substantial failure rates (Borst, 2010;Rusakov et al., 2020). Release probabilities at central synapses have been reported to range from 0.09 to 0.80, with in vivo experiments suggesting that most synapses fall at the low end of this spectrum (Rosenmund et al., 1993;Oertner et al., 2002;Emptage et al., 2003;Sylantyev et al., 2013;Jensen et al., 2019). Therefore, a brief loss of function could easily go unnoticed. It has even been argued that activity-driven synaptic failures may be beneficial for overall network function (Rusakov et al., 2020). First, synaptic failures may help to save energy (Levy and Baxter, 2002). Second, they may in certain situations improve the encoding of information (Guo and Li, 2012). Third, they may promote the emergence of new neural pathways (Budak and Zochowski, 2019). However, problems are expected if (1) a reduction or loss of function persists for extended periods of time preventing growth or in an extreme case causing elimination of synapses, or (2) affects entire networks. So how can synapses (neurons, networks) safeguard themselves against loss of function due to supply chain failures or excessive demand? There are several mechanisms available.
The first line of defense against arising shortages is having multiple independent sources for the same product. For example, one can utilize both local and remote providers. This is what is observed in neurons. A large variety of synaptic proteins have their coding mRNAs localized in the somatic region and in dendrites or axons (Perez et al., 2021). Metabolic labelling allowing for the localization of the sites of protein production and translatome analysis in soma versus neurites show, to the best of our knowledge, that always both somatic and neuritic mRNAs are used for protein production (tom Dieck et al., 2015;Hafner et al., 2019;Kreis et al., 2019;Glock et al., 2021). Thus, local protein synthesis does not replace somatic synthesis but adds an extra source of proteins. Nevertheless, local protein synthesis is required in various long-lasting forms of plasticity including the late phase of long-term potentiation (Steward et al., 1998;Steward and Worley, 2001;Kang and Schuman, 1996). Importantly, local protein synthesis while providing on-the-spot a "ready-to-be captured" pool of proteins during activity, also simply participates to the global effort of providing sufficient proteins for the neuronal cytoplasm and plasma membrane surface. This was recently validated by a computational model dissecting the contribution of mRNA and protein motion in neuronal protein distribution using CamKIIα as a case study. Using realistic dynamic parameters extracted from experimental data, the authors showed that about one half of the CamKIIα proteins in neurons are synthesized in dendrites and the other half in the soma (Fonkeu et al., 2019).
The second line of defense against shortages is to modulate the "expiration date" of already made products depending on the demand and supply. Interestingly, this has also been observed in neurons. A large fraction of the synaptic proteome shows changes in protein synthesis and/or degradation during homeostatic up-and down-scaling (Dörrbaum et al., 2020). Loss of protein homeostasisthe maintenance of a functional equilibrium between protein synthesis and degradationis a hallmark of aging (López-Otín et al., 2013). Recently, Kluever and colleagues discovered that protein lifetimes in the brain systematically change during aging (Kluever et al., 2022). Indeed, the general trend is a 20 % increase in protein lifetime when comparing young mice at 5 weeks and old mice at 22 weeks. Interestingly, the effects of aging on protein lifetime is particularly strong for pathways linked to neurodegenerative disorders and of the opposite direction to the general trend.
The third line of defense is the maintenance of a "safety buffer" we call the reserve pool. The existence of such pools has been described experimentally for pre-and postsynaptic proteins (Gray et al., 2006;Tsuriel et al., 2006;Ziv and Fisher-Lavie, 2014) and even synaptic vesicles (Staras et al., 2010). Local mRNAs also form a pool of ready to be captured transcripts for the local production of proteinssee previous section. Similar to how a logistics centers store products to make sure a fluctuating demand can be met even if supply is intermittent, maintaining a reserve pool of synaptic building blocks can serve a similar goal. At the presynapse, a prominent example is the maintenance of a large pool of neurotransmitter containing vesicles. On the postsynaptic side, dendrites maintain a pool of neurotransmitter receptors that are ready to be incorporated into synapses.
How big does such a reserve pool need to be? Clearly, a reserve pool that is too small will not be effective at mitigating the risk of temporary shortages where demand exceeds supply. On the other hand, an oversized reserve pool is wasteful because (1) it binds precious resources (material, space, etc.) that could be used for other purposes and (2) it is expensive to maintain since proteins are short lived (Cohen et al., 2013;Dörrbaum et al., 2018;Fornasiero et al., 2018) note that rough estimates predict that protein synthesis contributes 30 % to the ATP budget in differentiated mammalian cells (Buttgereit and Brand, 1995). The ideal size of a reserve pool is one that balances these aspects: it must be big enough to minimize the risk of serious shortages and loss of function for extended periods of time and it should be small enough to not waste precious resources. We call this the fundamental neurologistic trade-off. In the following, we discuss some of the different solutions to this tradeoff observed at neocortical synapses and their implications in more detail.

Resource bottlenecks for synaptic function
The average copy number of different synaptic proteins can vary over several order of magnitudes, ranging from a few tens of copies for regulating synaptic membrane exocytosis proteins (RIMs) or postsynaptic receptors to thousands of copies for VAMP2 or CaMKIIa at the pre-and postsynapse, respectively (Sheng and Kim, 2011;Wilhelm et al., 2014). This reveals different strategies for the supply of synaptic proteins taking part in the same functional process. Some proteins appear to be provided in large excess while others might constitute bottlenecks for synaptic function.
On the side of the presynapse, signal fluctuations result in an essentially binary outcome: the arrival of an action potential at a bouton either triggers or not the release of neurotransmitter. As mentioned above the release probability of central synapses is rather low. However, there are also very reliable synapses in the brain: the calyx of Held, mossy fiber synapses in the CA3 region of the hippocampus, and inputs on the granule cell layer of the cerebellum. For all those synapses, high continuity of signal transduction is achieved by multiplying the number of (low release probability) release sites on the same postsynaptic target. Of note, recent work on the drosophila mushroom body indicates that transient active zone remodelling could also increase the release probability of individual boutons upon learning (Turrel et al., 2022). Interestingly, while proteins of the synaptic vesicles (Wilhelm et al., 2014) are the most abundant in presynaptic boutons, their mRNAs are not found locally (Shigeoka et al., 2016;Hafner et al., 2019). This supports what has long been proposed and observed, namely that synaptic vesicle proteins are produced at the soma (Hannah et al., 1999). Conversely, mRNAs coding for key components of the active zone are found in boutons such as Bassoon, RIMs and Munc-13 (Gumy et al., 2011;Taylor et al., 2009;Hafner et al., 2019). Strikingly, RIMs and Munc-13 have low copy numbers per bouton, only tens of copies (Wilhelm et al., 2014), but high turnover rates (Dörrbaum et al., 2018;Sun and Schuman, 2022). Is local availability of RIMs and Munc-13 a bottleneck restricting the number of synaptic vesicle release sites? If that is the case, local synthesis of a few copies of RIMs and/or Munc13 might be sufficient to trigger the formation of additional release sites. Thus, local protein synthesis of a few of these active zone components might significantly improve the reliability of neurotransmitter release at a single bouton.
On the postsynaptic side, it is the detection of neurotransmitter released in the synaptic cleft that is the main source of fluctuations. An excitatory synapse's strength is largely defined by the number of AMPARs trapped in the postsynaptic density vis-à-vis glutamate release sites when release occurs. However, those receptors are highly mobile and continuously exchange between the PSD, the perisynaptic membrane and intracellular stores (Tardin et al., 2003;Opazo et al., 2012). This is considered the main source of postsynaptic current fluctuations for a single excitatory synapse. In the following section we describe in detail the origin and functional consequences of such fluctuations. Similar to active zone proteins, glutamate receptors (i.e. AMPARs, NMDARs, mGluRs) are only present in tens of copies at a single synapse (Nair et al., 2013;Sheng and Kim, 2011). Again, those low copy number proteins have among the fastest turnover of postsynatic proteinsmuch faster than the abundant scaffolding proteins of the PSD (e.g. PSD-95, Shank1 or Homer1) (Dörrbaum et al., 2018;Sun and Schuman, 2022). Thus, on the one hand, transient disruptions of the supply chain for those receptors could have significant impact on the response to neurotransmitter release. On the other hand, supply of additional receptors is likely to constitute a bottleneck for the induction of long-term potentiation. In fact, preventing lateral diffusion of receptors using cross-linking strategies impacts synaptic function and plasticity and impedes learning (Heine et al., 2008;Penn et al., 2017).

Competition for postsynaptic building blocks shapes synaptic plasticity
The implications of the fundamental neurologistic trade-off have been studied in the context of synapses competing for resources such as AMPA receptors (Earnshaw and Bressloff, 2006;Bressloff and Earnshaw, 2007;Earnshaw and Bressloff, 2008;Czöndör et al., 2012;Triesch et al., 2018). The starting point of all those models is the experimental observation that AMPA receptors are highly dynamic, transitioning between individual synapses and a dendritic pool on time scales of seconds and minutes (Sturgill et al., 2009;Chater and Goda, 2014;Opazo et al., 2012). The first models by Earnshaw and Bressloff considered a long dendrite with a point source at one end symbolising a somatic source for receptors. Their models already reproduced interesting features observed experimentally like a fast increase in synaptic receptors after increased exocytosis in extrasynaptic membranes upon LTP induction Earnshaw and Bressloff (2006). They also highlighted the inadequacy of a unique point source of receptors and diffusion as the sole type of motion to target receptors to distal dendritic compartments (Bressloff and Earnshaw, 2007;Earnshaw and Bressloff, 2008). Czondor and colleagues modelled a local system composed of a membrane with a postsynaptic density (Czöndör et al., 2012). Kinetics of receptors were based on experimentally measured trajectories of individual AMPA receptors, including free diffusion in the extrasynaptic space, confinement in the synapse, and trapping at the postsynaptic density through reversible interactions with scaffold proteins. Their model predicted that local recycling of AMPA receptors close to the postsynaptic density, coupled to short-range surface diffusion, provides rapid control of AMPA receptor numbers at synapses (Czöndör et al., 2012).
Triesch and colleagues extended these ideas by modeling a small piece of dendrite with multiple postsynaptic densities in which the total amount of proteins in short timescales is conserved Triesch et al. (2018). The authors developed a mathematical model in which synapses are competing for AMPA receptors and/or anchoring proteins such as PSD-95. This model was built to better understand this trade-off and its implications on synaptic transmission and plasticity. Indeed, fast receptor dynamics in combination with synaptic competition for a finite amount for available receptors have a number of interesting ramifications. A first ramification is the presence of spontaneous fluctuations of synaptic efficacies. The amplitudes of these fluctuations scale with the square root of synapse size, i.e., the efficacies of larger synapses are relatively more stable, as quantified by a smaller coefficient of variation. Interestingly, careful measurements of these fluctuations allowat least in theoryto infer quantities that are not easy to access experimentally, such as the number of receptor slots in a particular synapse or the average fraction of receptor slots that are filled by receptors, the so-called filling fraction. A second ramification is a multiplicative scaling behaviour of synapses communicating with the local dendritic pool. If the size of the dendritic receptor pool is increased (or decreased), the synapses scale up (or down) their efficacies multiplicatively, while maintaining their relative efficacies. This may facilitate the homeostatic regulation of activity, while safe-guarding memories stored in the pattern of relative synaptic efficacies. A third ramification of the rapid exchange of receptors between synapses and the dendritic pool is a competition of synapses for available receptorsin particular during the induction of LTP. This competition induces a transient form of heterosynaptic plasticity such that synapses undergoing LTP grow at the expense of other synapses, which are forced to undergo a temporary depressiona mechanism reviewed by Wagle et al. in this special issue. Interestingly, the amount of this heterosynaptic depression is inversely related to the size of the receptor pool. A small receptor pool induces strong competition and strong heterosynaptic plasticity, while a large receptor pool limits the amount of heterosynaptic plasticity.
As discussed above, proper functioning of a synapse requires the maintenance of a highly complex machinery and synapses may compete over many of the required building blocksin particular those with low copy numbers. The mathematical model of Triesch et al. (2018) makes quite generic assumptions. Therefore, it should be possible to adapt it to other kinds of synaptic building blocks such as presynaptic active zone proteins or postsynaptic scaffolding molecules.

Discussion and outlook
Synapses are in constant competition for resources such as synaptic building blocks. Shortages in the supply of these building blocks may obstruct synaptic growth and thereby impair learning or even jeopardize basic synaptic function. We identified three complementary mechanisms that neurons use to manage such risks. First, they use multiple alternative routes to supply building blocks to synapses. Second, they modulate protein turnover rates. Third, they use reserve pools for essential ingredients to buffer fluctuations in supply and demand. The sizes of these reserve pools must balance two opposing objectivesreducing the risk of shortages and the associated loss of function while not wasting precious resources that could be used otherwise. We refer to this balancing act as the fundamental neurologistic trade-off.
Beyond individual neurons, what may be the effects of such competition of synapses for synaptic building blocks at the network level (Ramiro-Cortés et al., 2014)? Above we highlighted that shortages of supply may impair signal transmission (e.g., for shortage of neurotransmitter) or learning (e.g. for shortage of new AMPARs). These problems can be expected to directly translate to the network level, leading to, respectively, impaired information processing and learning. However, fierce competition of synapses for building blocks may also exert beneficial effects on network stability. In computational models, Hebbian-like synaptic plasticity mechanisms are notoriously unstable and can lead to run-away growth of individual synapses or certain connectivity patterns. Such excessive growth can make networks unstable and lead to seizure-like activity patterns. Strong competition inherently reins in such growth and therefore contributes to network stability.
While we have focused on the supply of synapses with building materials and competition for these materials, this is only one of many logistics problems that nervous systems have to solve. Triesch and colleagues introduced the term "Neurologistics" to refer to the study of this larger set of problems (Triesch et al., 2018). Here we define it as follows: Neurologistics is the study of how neurons and the entire nervous system solve various logistics problems, how they do so efficiently and robustly, trading off conflicting demands on space, time, energy and material, and how logistic failures are related to nervous system disorders.
Indeed, a potentially powerful application of this Neurologistics perspective are neurodevelopmental and neurodegenerative disorders. In particular, reduced or excessive supply of certain proteins has been linked to several mental retardation disorders (Ramiro-Cortés et al., 2014). Neurodegenerative disorders often appear to be associated with a dwindling supply of an essential resource, that will ultimately lead to the demise of neural structures (synapses, neurons) and loss of function. Importantly, an early sign of a dwindling supply of a resource will be increased competition for this resource. This suggests a general strategy through which a dwindling supply could be detected, before it leads to loss of function: by detecting increased competition. Consider the above examples of competition for synaptic building blocks. A shortage of neurotransmitter receptors in the dendritc pool would increase competition for receptors among synapses and manifest as increased levels of heterosynaptic plasticity. Similarly, dwindling supplies of neurotransmitter may affect synaptic reliability and short-term synaptic plasticity. Interestingly, both short and long-term plasticity can be induced noninvasively with brain stimulation techniques such as transcranial magnetic stimulation (TMS) (Ziemann et al., 1996;Stefan et al., 2000). This might allow the development of non-invasive techniques for detecting increased competition for different synaptic resources. Ultimately, this could open the door for improved early diagnosis of neurodegenerative disorders.
In conclusion, field of Neurologistics in general and the study of competition among synapses for vital resources in particular are still in their infancy. However, they hold great promise for deepening our understanding of healthy brain function and disruptions in neurodegenerative disorders.

Declaration of competing interest
We declare having no competing interest.

Data availability
No data was used for the research described in the article.