In neural cells, certain RNAs are targeted to dendrites by a specific RNA trafficking pathway, termed the A2 pathway, mediated by the trans-acting trafficking factor, heterogeneous nuclear ribonucleoprotein (hnRNP) A2, which binds to an 11 nucleotide cis-acting trafficking sequence, termed the hnRNP A2 response element (A2RE). RNAs containing A2RE-like sequences are recognized by hnRNP A2 in the nucleus and exported to the cytoplasm where they assemble into trafficking intermediates, termed granules, which also contain components of the translation machinery and molecular motors (cytoplasmic dynein and conventional kinesin). RNA granules move along microtubules to the cell periphery where they become localized and where the encoded protein is translated. Intracellular trafficking of RNA molecules by the A2 pathway is mediated by a complex system consisting of five different subsystems, ∼35 different molecules and ∼45 different molecular interactions. Specificity in the A2 pathway is provided by specific interactions of hnRNP A2 with different molecular partners in different subsystems. Polarity of RNA trafficking is controlled by transitions of trafficking intermediates between different subsystems. Comprehensive understanding of the A2 RNA trafficking pathway will require quantitative analysis of concentrations and diffusion constants for each of the different molecules, on rates and off rates for each of the different interactions, relevant conditional operators controlling specific interactions, and interactions of different subsystems. Once the necessary quantitative data are available, mathematical models for the different RNA trafficking subsystems can be developed using computational platforms such as the ‘Virtual Cell’. Here we describe how each of the subsystems in the A2 system functions and how the different subsystems interact to regulate RNA trafficking.
RNA trafficking is a complex process whereby specific mRNAs are localized to particular subcellular compartments, thereby maximizing translation of the encoded proteins at appropriate intracellular locations and minimizing ectopic translation elsewhere in the cell. We have characterized the A2 RNA trafficking pathway, which is mediated by an 11 nucleotide cis-acting element (GCCAAGGAGCC), termed the heterogeneous nuclear ribonucleoprotein (hnRNP) A2 response element (A2RE), found in many localized RNAs, and its cognate trans-acting trafficking factor, hnRNP A2, expressed in many different cell types (Carson et al., 1998). The A2 pathway presumably functions in all cells that express hnRNP A2 and A2RE RNAs, but has been best characterized in oligodendrocytes (Carson et al., 2001a) and neurons (Shan et al., 2003) where the extended ramified cell morphology in culture facilitates microscopic visualization of RNA trafficking in live cells.
A2RE-containing RNA molecules bind to hnRNP A2 in the nucleus and are exported to the cytoplasm where they are assembled into trafficking intermediates, termed ‘RNA granules’. In addition to A2RE RNA and hnRNP A2, granules also contain components of the translation machinery (aminoacyl-tRNA synthases, elongation factors and ribosomal subunits) and molecular motors (conventional kinesin and cytoplasmic dynein) that move the granules along microtubules (Carson et al., 2001a). The overall direction of movement of A2RE RNA granules is from the perikaryon towards the plus ends of microtubules, located in the periphery in most cells. However, analysis of granule dynamics reveals that most granules exhibit complex bi-directional motion, reflecting stochastic fluctuations of dynein and kinesin motor activities in each granule. In oligodendrocytes, pH microdomains affect trafficking of RNA granules by regulating motor activities in their vicinity (Ro and Carson, 2004). Granules are targeted to specific sites along the dendrites where they become immobilized. Translation is suppressed during transport and activated when the RNA granules reach their target sites, which correspond to sites of myelin synthesis in oligodendrocytes (Carson et al., 1998) and to sites of synapse formation in neurons (Huang et al., 2003).
The entire A2 RNA trafficking pathway, beginning with export from the nucleus and ending with localization and translation at specific sites in the dendrites, is mediated by a system of reversible molecular interactions, as shown in Figure 1. We have identified ∼35 different molecular species and ∼45 different molecular interactions involved in the A2 RNA trafficking system. The overall system can be decomposed into five subsystems, which are colour coded in Figure 1: nuclear export (cyan), granule assembly (blue/red), transport on microtubules (yellow), pH regulation (magenta), and translation regulation (green). Each subsystem consists of a group of molecules that work together to perform a specific function. The molecules in each subsystem have specific interactions with each other and are often coordinately regulated and co-localized. Individual subsystems function semi-autonomously and trafficking intermediates move sequentially from one subsystem to the next. Some molecular species appear in multiple subsystems and can participate in multiple molecular interactions linking separate subsystems, which means that the different subsystems are not completely self-autonomous. Different subsystems can interact with each other if they share common molecular components. The overall polarity of the pathway is achieved through differential subcellular distributions of key molecular components, transitions of trafficking intermediates between different subsystems and conditional operators (AND, OR, IF, ONLY) controlling specific molecular interactions. For example, protein A interacts with protein B ONLY IF protein C also interacts with protein B. Or, protein A interacts with protein B, OR protein C, but not both. In this review we will discuss how each subsystem, molecule, interaction and conditional operator affects RNA trafficking by the A2 pathway.
Molecular interaction networks of the type shown in Figure 1 provide a useful outline for describing complex cell biological systems, but contain too many variables to be solved analytically. However, if individual subsystems within the network are represented as mathematical models, they can be solved numerically provided certain quantitative parameters are known. Specifically, for each node (molecular species) in the system, the concentration, subcellular distribution and diffusion coefficient must be known, and for each connection (molecular interaction) in the system, the on rate and off rate must be known. For nodes with more than one connection (molecules that interact with more than one partner), conditional operators (AND, OR, IF, ONLY) governing those interactions must be defined. The overall pathway is defined by sequential transition of trafficking intermediates from one subsystem to the next.
Three fundamental questions about the system can be answered by mathematical modelling. The first question is whether the molecular interactions in the network actually occur in living cells. Networks of the type shown in Figure 1 are qualitative representations of molecular interactions that, in most cases, have been shown to occur either in vitro or in specialized heterologous systems (e.g. yeast two-hybrid analysis) where specific molecules are often overexpressed. Whether the same molecular interactions also occur in live cells depends on local concentrations and diffusion properties for each molecular species, and on kinetic constants for each molecular interaction in the particular cell of interest. If these parameters are known, mathematical modelling can predict the extent to which each molecular interaction actually occurs in a particular cell. The second question is whether the molecular species and interactions in the network are sufficient to account for the experimentally observed behaviour of the system. Again, if concentrations and diffusion coefficients for each molecular species, and kinetic constants for each molecular interaction are known, mathematical modelling can reveal if these are sufficient to generate the observed behaviour. The third question is how perturbations to one subsystem are propagated through other subsystems to generate complex behaviour. The complexity of most interaction networks makes it impossible to predict analytically how a perturbation in one particular subsystem will propagate through the entire system. However, if concentrations and diffusion coefficients for each molecular species, and kinetic constants for each molecular interaction are known, mathematical modelling can be used to analyse interactions between different subsystems computationally. Thus, mathematical modelling can be used to analyse complex molecular interaction systems if quantitative data on intracellular concentrations, diffusion coefficients, and kinetic constants are known.
Unfortunately, quantitative data of this type are sparse in the current cell biological literature, and mathematical modelling approaches are not easily accessible to most cell biologists. Concentrations of specific molecular species and kinetic constants for molecular interactions can be analysed in vitro by conventional biochemical techniques; however, these techniques cannot generally be applied in vivo, where conditions inside the cell (molecular crowding, tortuosity, obstructed diffusion, spatial non-uniformity) are likely to be different from the conditions in vitro. Fortunately, recent advances in fluorescence correlation spectroscopy (FCS) provide the technology to measure concentrations, diffusion and kinetic constants in living cells (Levin and Carson, 2004), and recently developed computational modelling platforms, such as the Virtual Cell (Slepchenko et al., 2002), make mathematical modelling accessible to cell biologists. The results of quantitative analysis and mathematical modelling approaches can provide insight into how RNA trafficking works in live cells.
The first step in the A2 RNA trafficking system is nuclear export of A2RE RNA. The nuclear transport subsystem (Figure 2) mediates import of macromolecules into the nucleus and export of molecules out of the nucleus. One of the key regulators of the A2 RNA trafficking pathway is hnRNP A2, which shuttles between the nucleus and the cytoplasm. In the cytoplasm, newly synthesized hnRNP A2 binds to transportin 1, and is imported through the nuclear pore into the nucleus (Siomi et al., 1997). In the nucleus, hnRNP A2 dissociates from transportin and binds to newly synthesized A2RE RNAs, which are exported to the cytoplasm by an export pathway involving several additional molecules not shown in Figure 1 (Erkmann and Kutay, 2004). Once dissociated from hnRNP A2, transportin 1 binds to Ras-related nuclear protein (Ran)GTP and is exported from the nucleus through the nuclear pore complex. In the cytoplasm, RanGTP interacts with RanBP (Ran binding protein) and RanGAP (GTPase-activating protein), which activate the GTPase activity of Ran, converting RanGTP to RanGDP and causing transportin 1 to dissociate. RanGDP, generated by hydrolysis of GTP in the cytoplasm, binds to NTF2 (nuclear transport factor 2) and is imported through the nuclear pore into the nucleus. In the nucleus, RanGDP dissociates from NTF2 and interacts with the RanGEF (guanine nucleotide exchange factor), RCC1, regenerating RanGTP. NTF2 is exported to the cytoplasm through the nuclear pore.
The nuclear transport subsystem described above explains how hnRNP A2 shuttles into and out of the nucleus. However, in the A2 RNA trafficking pathway, hnRNP A2 bound to A2RE RNA is exported to the cytoplasm and, instead of shuttling back into the nucleus, remains associated with the RNA during subsequent cytoplasmic steps in the A2 pathway including: granule assembly, transport on microtubules, localization and translation. The mechanism(s) that prevent hnRNP A2 bound to A2RE RNA from being re-imported into the nucleus by the transportin pathway are not known. One possibility is that hnRNP A2 binding is conditional, such that it can bind to EITHER transportin OR A2RE RNA, but not to both molecules simultaneously. This would prevent hnRNP A2 bound to A2RE RNA from being re-imported into the nucleus by transportin. The structural basis for such putative conditional binding is not clear since the binding sites in hnRNP A2 for transportin and for A2RE RNA have been localized to different parts of the molecule. Transportin binds to the M9 domain of hnRNP A2, which is a sequence of ∼35 amino acids located near the C terminus (Siomi et al., 1997), while A2RE RNA binds to one of two RNA binding domains (RBD I and RBD II) located in the N-terminal part of the molecule (Shan et al., 2000). Nevertheless, it is possible that binding of A2RE RNA induces a conformational change in hnRNP A2 that prevents simultaneous binding of transportin. A second possibility is that subsequent cytoplasmic steps in the A2 pathway (RNA granule assembly and microtubule transport) effectively remove hnRNP A2 from the vicinity of the nucleus, thereby preventing re-import by transportin. In this way the cytoplasmic granule assembly and/or microtubule transport subsystems may influence the function of the nuclear transport subsystem by competition for a common molecular species, hnRNP A2. A third possibility is that hnRNP A2 exported from the nucleus undergoes post-translational modification(s) that prevent its re-import into the nucleus by transportin. Other hnRNP proteins are known to undergo a variety of post-translational modifications including methylation, phosphorylation, acetylation and sumoylation. If hnRNP A2 undergoes similar post-translational modification this might prevent re-import into the nucleus. Distinguishing between these possible mechanisms for directional nuclear export of hnRNP A2 bound to A2RE RNA is important because this represents the first irreversible step in the A2 RNA trafficking pathway.
After nuclear export, the next step in the A2 pathway is granule assembly. When A2RE RNA molecules associated with hnRNP A2 are exported from the nucleus to the cytoplasm they assemble into large aggregates called RNA granules. Each granule contains multiple (up to 30) RNA molecules (Mouland et al., 2000), and the assembly process is sequence specific so that A2RE RNA molecules co-assemble into the same granules, and non-A2RE RNA molecules are segregated into a different population of granules. This implies an RNA sorting step that recognizes the A2RE sequence and an RNA aggregation step that assembles multiple A2RE RNAs into the same granule. One model for granule assembly (Figure 3) involves sequence-specific recognition of A2RE sequences by hnRNP A2 molecules, and aggregation of RNA molecules by dimerization of hnRNP A2 molecules (Carson et al., 2001a). This model is based on two observations. First, many A2RE RNAs actually contain multiple A2RE-like sequences, which means that they can potentially bind multiple hnRNP A2 molecules. Secondly, hnRNP A2 molecules can dimerize, which means that A2RE RNA molecules bound to hnRNP A2 molecules can aggregate through A2–A2 dimerization interactions. This model provides an explanation for the specificity of RNA sorting and aggregation during granule assembly. According to the model, the size of the granule (how many A2RE molecules it contains) is ultimately determined by the concentrations of A2RE RNA and hnRNP A2 in the cytoplasm and by the kinetic constants for binding of hnRNP A2 to A2RE RNA and for dimerization of hnRNP A2 molecules. These parameters also provide an explanation for granule assembly occurring in the cytoplasm but not in the nucleus. The concentration of hnRNP A2 in the nucleus (∼20 μM) is approx. 20-fold higher than the concentration in the cytoplasm (∼1 μM) (Brumwell et al., 2002). In the nucleus, the concentration of hnRNP A2 is much greater than the concentration of A2RE RNA, which means that all hnRNP A2 binding sites on A2RE RNA molecules are likely to be saturated and all dimerization sites on hnRNP A2 molecules bound to A2RE RNA are also likely to be occupied. In other words, each hnRNP A2 binding site on A2RE RNA will be occupied by a dimer of hnRNP A2. This will preclude aggregation of A2RE RNA molecules by hnRNP A2 dimerization, because no unoccupied dimerization sites are available. In the cytoplasm, the reduced hnRNP A2 concentration will result in dissociation of some hnRNP A2 molecules from A2RE RNA, so that some hnRNP A2 binding sites on A2RE RNAs will be occupied by single hnRNP A2 molecules with unoccupied dimerization sites. This will allow dimerization of hnRNP A2 molecules bound to different RNA molecules, resulting in aggregation of A2RE RNAs into granules. Thus, granule assembly may be controlled by mass action, reflecting the partition coefficient for hnRNP A2 concentrations between nucleus and cytoplasm.
The nuclear/cytoplasmic partition coefficient for hnRNP A2 is determined by the interaction of the nuclear transport subsystem with other RNA trafficking subsystems. Perturbations in one subsystem can be propagated into other subsystems, thereby altering the nuclear/cytoplasmic partition coefficient of hnRNP A2. For example, inhibition of transcription in the nucleus causes redistribution of hnRNP A2 from the nucleus to the cytoplasm (Pinol-Roma and Dreyfuss, 1993). Conversely, disruption of microtubules in the cytoplasm causes redistribution of hnRNP A2 from the cytoplasm to the nucleus (Brumwell et al., 2003). The mechanism(s) for hnRNP A2 redistribution are not clear in either case. Redistribution of hnRNP A2 after transcription inhibition suggests that hnRNP A2 is retained in the nucleus by binding to newly synthesized RNA molecules. When the level of newly synthesized RNA in the nucleus is reduced by transcription inhibition, unbound hnRNP A2 is exported from the nucleus. Redistribution of hnRNP A2 after disruption of microtubules suggests that re-import of hnRNP A2 into the nucleus is prevented by transport of RNA granules to the periphery of the cell along microtubules. When microtubules are disrupted, transport of RNA granules is inhibited, allowing re-import of hnRNP A2 into the nucleus. Understanding hnRNP A2 redistribution will require quantitative analysis and mathematical modelling, which will provide insight into how granule assembly is regulated and how different RNA trafficking subsystems interact.
Transport on microtubules
Each RNA granule contains both conventional kinesin and cytoplasmic dynein motors, which move the granule towards the microtubule plus end or minus end respectively. The overall direction of granule movement is determined by the balance of power between kinesin and dynein activities in each granule (Carson et al., 2001a). In the case of A2RE RNA granules, the overall direction of movement is anterograde, towards the plus ends of microtubules in the periphery, indicating that the balance of power in these granules favours kinesin. Conversely, non-A2RE RNA granules are retained near the minus ends of microtubules in the perikaryon, indicating that the balance of power in these granules favours dynein. The only known differences in composition between A2RE and non-A2RE RNA granules are the sequences of the cargo RNAs and the specific proteins that bind to those RNAs. A2RE RNA granules contain hnRNP A2 while non-A2RE RNA granules do not. Therefore, hnRNP A2, in addition to mediating granule assembly, may also regulate the balance of power between kinesin and dynein in favour of kinesin. The mechanism for such regulation is not known.
A yeast two-hybrid screen for proteins that interact with hnRNP A2 identified TOG (tumour overexpressed gene), a microtubule associated protein, as a molecular partner for hnRNP A2 (L. Kosturko, M. Maggipinto, C. D'Sa, J.H. Carson, E. Barbarese, unpublished data). In other systems, orthologues of TOG are believed to regulate motor activities during spindle formation in mitosis (Hiroyuki et al., 2001). If TOG also regulates motor activities in RNA granules, it is possible that TOG binds to hnRNP A2 in A2RE RNA granules and shifts the balance of power in these granules in favour of kinesin. In this way, specific sequences (such as A2RE) in the RNA cargo of each granule may determine the overall direction of transport by recruiting specific RNA binding proteins (such as hnRNP A2) that alter the balance of power between kinesin and dynein by binding to specific regulatory proteins (such as TOG). In other words, the direction of RNA granule transport is determined by interactions between the granule assembly subsystem, which controls which granules contain hnRNP A2, and the microtubule transport subsystem, which is mediated by molecular motors that may be regulated by the hnRNP A2 binding protein, TOG.
Analysis of trajectories of individual RNA granules reveals periods of sustained uni-directional movement interspersed with periods of constrained bi-directional movement (Carson et al., 2001a). These complex granule dynamics can be explained by the model outlined in Figure 4 (Carson et al., 2001b). Each motor associated with the granule can be in one of three states relative to the microtubule: unbound, bound but inactive, or active. The direction of movement at each instant is determined by which motor(s), kinesin or dynein, is active. Bi-directional movement is generated by stochastic activation and inactivation of kinesin and/or dynein motors. The cloud of inactive motors associated with each granule exerts a stalling strain on the active motor(s), increasing the probability of inactivation of the active motor(s) and constraining the movement of the granule along the microtubule. Conversely, the active motor(s) associated with each granule exert an escape strain on the cloud of inactive motors, increasing the probability of dissociation of inactive motors from the microtubule. Stalling and escape strains are propagated through elastic properties of the granule. According to this model, sustained uni-directional movement occurs when the cloud of inactive motors dissociates from the microtubule, leaving the active motor free to move in one direction until the cloud reassociates with the microtubule. Thus, the movement of each granule is determined by kinetic constants for binding of individual motors to microtubules, activation/inactivation rates for each motor and opposing strains between inactive and active motors.
The overall velocity of movement of RNA granules fluctuates along the length of the dendrite, with relatively rapid movement in certain regions interspersed with slower movement in other regions. Since the intrinsic step-size for individual motors is constant, the overall displacement of the granule over time will be determined by the length of time each motor remains active before becoming inactive, which is regulated by the stalling strain exerted by the cloud of inactive motors. Therefore, the velocity of the granule will reflect the kinetic constants for motor binding to microtubules. Velocity fluctuations as the granule moves along the dendrite suggest that motor binding to microtubules is affected by variations in the micro-environment along the dendrite.
Intracellular pH is an important regulator of many biochemical and cell biological processes. Until recently it was assumed that intracellular pH was relatively uniform in most cell types. However, studies using ratiometric pH indicator dyes in oligodendrocytes have revealed discrete alkaline and acidic pH microdomains (∼1 μm in diameter) along the dendrites (Ro and Carson, 2004). As illustrated in Figure 5, alkaline microdomains are generated in the vicinity of transport metabolons consisting of carbonic anhydrase II (CAII), which converts CO2 and H2O into H? and HCO3−, bound to the sodium/hydrogen exchanger (NHE), which exports H? ions out of the cell, leaving a local concentration of HCO3− ions. Acidic microdomains are generated in the vicinity of transport metabolons consisting of CAII bound to sodium bicarbonate cotransporter (NBC), which exports HCO3− ions out of the cell leaving a local concentration of H? ions. Since the CAII binding sites in NHE and NBC are similar, both proteins probably bind to the same region of CAII. Thus, formation of transport metabolons and generation of pH microdomains is mediated by competing molecular interactions of CAII, with either NHE or NBC, which are controlled by a conditional operator, whereby each CAII molecule can bind to EITHER NHE OR NBC, but NOT both, simultaneously.
Kinesin binding to microtubules is pH dependent with stronger binding at lower pH (Verhey et al., 1998). This means that kinesin in RNA granules will bind more strongly to microtubules in the vicinity of acidic microdomains, causing the granule to slow down, and less strongly in the vicinity of alkaline microdomains, causing the granule to speed up. In this way, the pH subsystem interacts with the microtubule transport subsystem by regulating the local activity of motor proteins.
If acidic microdomains cause granules to slow down in their vicinity they may serve as ‘attractors’ for RNA granules. If so, then the subcellular distribution of pH microdomains is an important determinant controlling the direction of RNA trafficking. As outlined above, alkaline microdomains are generated in the vicinity of CAII—NHE transport metabolons, which are more concentrated in the perikaryon, while acidic microdomains are generated in the vicinity of CAII—NBC transport metabolons, which are more concentrated in the periphery. The reason(s) for the differential subcellular distribution of NHE, NBC and CAII RNAs are not known, but may be related to differential RNA trafficking. CAII and NBC RNAs both contain A2RE-like sequences, suggesting that these RNAs are transported to the periphery by the A2 RNA trafficking pathway. There is experimental evidence that CAII RNA, at least, is localized to the peripheral dendrites in oligodendrocytes (Carson et al., 2001a). The subcellular distribution of NBC RNA has not been analysed. If both CAII and NBC RNAs bind hnRNP A2 and are co-assembled into granules that are transported to the periphery by the A2 pathway, this could lead to co-translation of CAII and NBC proteins in the periphery of the cell, facilitating local formation of CAII—NBC transport metabolons, resulting in generation of acidic metabolons, ‘attracting’ transport of more A2RE RNAs to the periphery by the A2 pathway. In this way, interactions between the granule assembly subsystem, the microtubule transport subsystem and the pH regulation subsystem could constitute a positive feedback loop in the A2 trafficking pathway, ensuring that initial formation of CAII—NBC transport metabolons in a particular dendrite would lead to generation of local acidic microdomains and further transport of A2RE RNAs to that dendrite by the A2 pathway. This could lead to differential extension and differentiation of that dendrite relative to others.
RNA trafficking provides a mechanism to target expression of specific proteins to particular subcellular compartments while minimizing ectopic expression elsewhere in the cell. To accomplish this, translation must be suppressed during transport and activated once the RNA reaches its final destination. The concept of ‘localization-dependent translation’ implies a bi-stable two-state model where granules are either transport competent and translationally inactive (state 1), or transport incompetent and translationally active (state 2). Transition from state 1 to state 2 occurs upon RNA localization. The mechanism(s) responsible for bi-stability and state transitions in the A2 trafficking pathway are not known. However, two observations suggest a possible model. The first observation is that hnRNP A2 itself functions as a specific activator of cap-dependent translation of A2RE RNA (Kwon et al., 1999). The second observation is that, in a yeast two-hybrid screen, hnRNP A2 binds to hnRNP E1 (L. Kosturko, M. Maggipinto, J.-W. Lee, J.H. Carson, E. Barbarese, unpublished observation), which, in other systems, inhibits translation of specific RNAs by blocking recruitment of the 60 S ribosomal subunit (Ostareck et al., 2001; Ostareck-Lederer and Ostareck, 2004). Combining these two observations suggests the model outlined in Figure 6, whereby when granules are in state 1, A2RE RNAs are associated with hnRNP A2, which stimulates cap-dependent recruitment of the 40 S ribosomal subunit and also binds hnRNP E1, which prevents translation by blocking subsequent recruitment of the 60 S subunit. Thus, granules in state 1 are primed for translation but are translationally inactive during transport. When granules are in state 2, hnRNP E1 dissociates from hnRNP A2, thereby activating translation by removing the block to recruitment of 60 S subunit. Transition from state 1 to state 2 requires both localization of granules at sites of synthesis and activation of protein synthesis by dissociation of hnRNP E1 from the granule. It is not known if these events are independent or somehow linked. This model illustrates how two separate subsystems in the RNA trafficking pathway, microtubule transport and translation regulation, are linked by a conditional operator such that each granule can be in EITHER one subsystem, OR the other subsystem, BUT NOT BOTH subsystems simultaneously.
Movement of a particular granule from the microtubule transport system to the translation system involves a transition from one stable state (transport) to a second stable state (translation). Furthermore, this state transition must be highly co-operative for all RNA molecules in each granule. That is, in any one granule, all the RNA molecules should be either translationally inactive or translationally active. Otherwise, translation could be activated on some RNA molecules while the granule was still in transit, which would result in ectopic expression of the encoded proteins in inappropriate subcellular compartments. If all A2RE RNA molecules in a single granule are inactive during transport and become active simultaneously upon localization, hnRNP E1 dissociation from hnRNP A2 must be highly co-operative. The molecular mechanism(s) that regulates co-operative hnRNP E1 dissociation from hnRNP A2 and transition from state 1 to state 2 are not known.
Specificity, polarity, complexity and stochasticity
We have described how each of the species, interactions and subsystems in the A2 system functions to mediate intracellular trafficking of A2RE RNA. Specificity, polarity and complexity of the system are generated through higher order interactions among subsystems. The specificity of the A2 pathway is provided by specific binding of hnRNP A2 to A2RE sequences in specific RNAs. A2RE RNAs are specifically targeted for the A2 pathway as soon as they bind hnRNP A2 in the nucleus. Since hnRNP A2 is a component of most of the different subsystems in the A2 pathway, the specificity of subsequent steps is mediated through specific sequential interactions between hnRNP A2 and a series of molecular partners in different subsystems.
The first specificity step in the A2 pathway, where A2RE RNAs are segregated from nonA2RE RNAs, is granule assembly, which is mediated by homotypic interactions between hnRNP A2 molecules bound to different A2RE RNA molecules. The second specificity step in the pathway, where A2RE RNA granules are distinguished from non-A2RE granules, is transport along microtubules. The balance of power between kinesin and dynein activities in each granule, that determines the direction of movement, may be regulated by TOG protein, which binds to hnRNP A2 in A2RE granules. The third specificity step in the pathway, where A2RE and non-A2RE RNAs are distinguished, is translation regulation. Cap-dependent recruitment of the 40 S ribosomal subunit to A2RE RNA, that presumably occurs in the perikaryon, is enhanced by hnRNP A2 binding to the A2RE. However, subsequent recruitment of the 60 S ribosomal subunit is blocked during transport by hnRNP E1 binding to hnRNP A2. When the A2RE granules reach their destination hnRNP E1 dissociates from the granule and translation is activated. Thus, specificity in the A2 pathway is provided by specific molecular interactions between hnRNP A2 and a series of molecular partners (A2RE RNA, hnRNP A2, TOG and hnRNP E1) in different subsystems.
Although most of the individual molecular interactions in the A2 pathway are reversible, the overall pathway is essentially uni-directional from nucleus to distal dendrites. Polarity in the pathway is generated when trafficking intermediates move from one subsystem to another, or from one state to another. The first polarity step in the A2 pathway occurs when A2RE RNA with hnRNP A2 bound moves from the nuclear transport subsystem to the granule assembly subsystem. After A2RE RNA/hnRNP A2 is exported from the nucleus to the cytoplasm, re-import is prevented by aggregation into RNA granules, which are too large to fit through the nuclear pore. Thus, assembly into granules effectively sequesters A2RE RNA and hnRNP A2 molecules, preventing them from re-entering the nuclear transport subsystem. The second polarity step in the A2 pathway occurs when A2RE RNA/hnRNP A2 moves from the granule assembly subsystem to the microtubule transport subsystem. When a growing RNA granule reaches a critical mass, the number of kinesin and dynein molecular motors associated with the granule becomes sufficient to move the granule along microtubules away from the perikaryon. This prevents further granule growth, which is hnRNP A2 dependent, because the concentration of hnRNP A2 in the periphery is much less than the concentration in the perikaryon. Thus, microtubule transport of RNA granules effectively sequesters A2RE RNA and hnRNP A2 molecules from the granule assembly subsystem. The third polarity step in the A2 pathway occurs when the RNA granule becomes localized and moves from the microtubule transport subsystem to the translation subsystem. The mechanism(s) responsible for localization of A2RE RNA granules and activation of translation are not known. Since a variety of different RNAs contain A2RE-like sequences and are believed to be transported by the A2 pathway in different systems, it is possible that different localization mechanisms are utilized by different RNAs. Granules are EITHER transport competent OR translation competent, which are mutually exclusive bi-stable states for the granule. Transition between these two bi-stable states involves co-operative dissociation of all hnRNP E1 molecules from all hnRNP A2 molecules in each granule, which makes the transition essentially irreversible. Thus, polarity of the A2 pathway is provided by subsystem transitions and bi-stable state transitions. Surprisingly, because of the bi-directional nature of granule movement on microtubules, the microtubule transport subsystem does not appear to contribute significant polarity to the A2 system.
The A2 pathway exhibits a variety of complex behaviours, which are not readily explained by the system in Figure 1. These include: variations in granule size, constrained and unconstrained bi-directional movement along microtubules, and localization at specific sites along the dendrites. We have emphasized the importance of quantifying concentrations and diffusion coefficients for each molecular species and kinetic constants for each molecular interaction as a prerequisite for reaction diffusion modelling. However, this type of modelling assumes large numbers of molecules and continuous distributions, which may not be appropriate for some subsystems in the A2 pathway. Since each granule contains a limited number of RNA molecules, binding proteins, and molecular motors, there may be significant stochasticity in the behaviour of individual granules. In other words, although the average behaviour of many RNA granules may be predictable, the behaviour of individual granules may be unpredictable and may exhibit large variations from the average behaviour, particularly in regard to granule growth and bi-directional movement along microtubules. Furthermore, the limited number of granules per cell and bi-stability in the state of each granule may result in stochastic quantal characteristics for granule localization and translation. Because of the small number of molecules per granule (Mouland et al., 2000) and the small number of granules per cell (Ainger et al., 1993), there is a significant component of stochasticity in the A2 pathway, which may be responsible for some of the complex behaviour observed experimentally. Accordingly, stochastic approaches may be more appropriate than reaction diffusion approaches for modelling the granule assembly, microtubule transport and translation regulation subsystems. Although different subsystems may require different modelling approaches, it will eventually be necessary to integrate the different subsystem models into a global model for the overall A2 system in order to understand how perturbations in one subsystem are propagated into other subsystems and how interactions among different subsystems can generate the complex behaviour observed experimentally in the A2 trafficking pathway.
The A2 RNA trafficking pathway has been extensively characterized in neural cells. Conventional biochemical and cell biological approaches have delineated the steps in the pathway and identified molecules and interactions that are involved at each step. However, this information by itself is not sufficient to explain the complex behaviour of the system. Here we have defined different subsystems within the A2 pathway, and described how each individual subsystem functions, how trafficking intermediates move from one subsystem to the next, and how different subsystems interact. Several recurrent themes emerge from this analysis. First, specificity at different steps in the A2 pathway is provided by specific binding of hnRNP A2 to A2RE RNA and subsequent sequential interactions of hnRNP A2 with a series of different molecular partners in different subsystems. In several cases, specificity is provided by conditional operators affecting different molecular interactions. Secondly, polarity of the A2 pathway is generated by transitions of trafficking intermediates from one subsystem to the next or from one bi-stable state to a second. Thirdly, complex behaviour emerges when different subsystems interact with each other and when perturbations in one subsystem are propagated into other subsystems. It is not clear if these themes are specific for the A2 trafficking system or whether they may represent general principles applicable to other cell biological systems.
The A2 system described here has been characterized almost exclusively in qualitative terms, and current models for how different subsystems function are essentially qualitative. Further progress will require mathematical modelling of individual subsystems and of the system as a whole, which will require quantitative measurements of concentration and diffusion coefficients for each molecule, and determination of kinetic constants and conditional operators for each molecular interaction in the system. Once these quantitative parameters are known, mathematical modelling of individual subsystems can provide rigorous testing for current models. Combining subsytem models into global models can provide insight into the complex behaviour of the A2 trafficking system.
This work was supported by National Institutes of Health (NIH) grants NS15190 and RR13186 to J.H.C. and NIH grant NS19943 and National Multiple Sclerosis Society (NMSS) grant MSRG2843 to E.B. The review is based on work by Dr Hongyi Cui, Dr Hyunah Ro, Mr Yuanzheng Gao, Mr Marius Ifrim, Mr George Korza, Ms Allison Paradise and Dr Mikhail Levin in J.H.C.'s lab, Mr Mike Maggipinto, Dr Victor Francone and Dr Linda Kosturko in E.B.'s lab, and Dr Boris Slepchenko in the Department of Cell Biology at University of Connecticut Health Center.
RNA granule: An intracellular trafficking intermediate containing specific RNA molecules, cognate RNA binding proteins, components of the translation machinery and molecular motors.
pH microdomain: A spatially restricted local minimum (or maximum) in intracellular pH generated by the activity of transport metabolons moving specific ions into or out of the cell.
Subsystem: A group of molecules that interact with each other to perform a specific function in the cell.
Conditional operator: When one molecule can interact with two or more partners, those interactions are described by a conditonal operator. For example molecule A can interact with EITHER molecule B OR molecule C BUT NOT both, or molecule A can interact with molecule B ONLY IF molecule B also interacts with molecule C.