Cytoskeletal signaling: is memory encoded in microtubule lattices by CaMKII phosphorylation?

Memory is attributed to strengthened synaptic connections among particular brain neurons, yet synaptic membrane components are transient, whereas memories can endure. This suggests synaptic information is encoded and ‘hard-wired’ elsewhere, e.g. at molecular levels within the post-synaptic neuron. In long-term potentiation (LTP), a cellular and molecular model for memory, post-synaptic calcium ion (Ca2+) flux activates the hexagonal Ca2+-calmodulin dependent kinase II (CaMKII), a dodacameric holoenzyme containing 2 hexagonal sets of 6 kinase domains. Each kinase domain can either phosphorylate substrate proteins, or not (i.e. encoding one bit). Thus each set of extended CaMKII kinases can potentially encode synaptic Ca2+ information via phosphorylation as ordered arrays of binary ‘bits’. Candidate sites for CaMKII phosphorylation-encoded molecular memory include microtubules (MTs), cylindrical organelles whose surfaces represent a regular lattice with a pattern of hexagonal polymers of the protein tubulin. Using molecular mechanics modeling and electrostatic profiling, we find that spatial dimensions and geometry of the extended CaMKII kinase domains precisely match those of MT hexagonal lattices. This suggests sets of six CaMKII kinase domains phosphorylate hexagonal MT lattice neighborhoods collectively, e.g. conveying synaptic information as ordered arrays of six “bits”, and thus “bytes”, with 64 to 5,281 possible bit states per CaMKII-MT byte. Signaling and encoding in MTs and other cytoskeletal structures offer rapid, robust solid-state information processing which may reflect a general code for MT-based memory and information processing within neurons and other eukaryotic cells.

The brain's ability to learn and store memory is understood in terms of changes in synaptic connections between neurons: ‘synaptic plasticity’ [1]. This is supported by the paradigm of ‘long-term potentiation’ (LTP) in which repetitive pre-synaptic stimulation increases post-synaptic sensitivity and strengthens synapses (e.g. the adage “neurons that fire together, wire together”). LTP is supported experimentally in vitro [2], [3], and may occur over many brain regions [4] as a common feature of excitatory synapses [5].

Below the level of the synapse, LTP involves complex gene expression, protein synthesis and recruitment of new receptors or even synapses [6]. Generally, however, synaptic plasticity is viewed in terms of changes in function, location and/or number of post-synaptic receptors and ion channels. However, synaptic receptors and channel proteins are transient, being synthesized and degraded in the protein lifecycle, and yet memories can last lifetimes. Therefore, information pertinent to memory must be stored elsewhere, yet remain able to regulate synaptic plasticity.

LTP involves the neurotransmitter glutamate binding to post-synaptic receptors, opening calcium channels to allow influx of calcium ions (Ca2+) to dendritic spines [7], shafts and neuronal cell body. Within dendritic spines, inflow of Ca2+ results in activation of multiple enzymes including protein kinase A [8], protein kinase C [9] and Ca2+-calmodulin kinase II (CaMKII) [10]. These enzymes, in turn, interact with (e.g. by phosphorylating) various intra-neuronal molecules, presumably for storage and processing of synaptic information. These CaMKII-phosphorylated protein structures must then, in some as-yet-unknown way, encode memory and regulate synaptic plasticity.

CaMKII has 12 kinase domains (6 on one side, 6 on the other), normally folded tightly in an inactive state to an association domain. Ca2+ influx to post-synaptic neurons binds calmodulin (CaM) to form Ca2+/CaM complexes, versatile regulators of multiple proteins and enzymes. Ca2+/CaM complexes activate CaMKII kinase regions allowing them to extend from the association domain and enable phosphorylation of a substrate protein, i.e. transfer of a phosphate group (through ATP hydrolysis) to a serine or threonine residue on a protein.

As Ca2+ influx can activate any, or all, of the 12 kinases, CaMKII has been suggested to record synaptic activity, retaining a ‘memory’ of past Ca2+ influx events in terms of activated phosphorylation states [11]. CaMKII may also auto-phosphorylate, in which one kinase phosphorylates an adjacent kinase on a single holoenzyme, resulting in prolonged CaMKII activation and possible information processing after initial Ca2+ influx. CaMKII is effectively a Ca2+ triggered synaptic memory device [12]. How and where information is encoded by CaMKII to be stored, and utilized to regulate synaptic plasticity remain open questions.

Following synaptic activity, CaMKII is rapidly distributed throughout post-synaptic neuronal dendrites and cell bodies [13]. The active sites of substrate phosphorylation on CaMKII kinase regions are referred to as ‘S’ sites and ‘T’ sites, apparently related to shorter and longer-term phosphorylation, respectively [14]. Single point mutation of amino acid threonine 286 in CaMKII results in impaired Ca2+ dependent alterations in synaptic plasticity, as well as in learning and memory in mouse models [10]. CaMKII phosphorylation of post-synaptic protein substrates is a likely mechanism for memory encoding/storage and regulation of synaptic plasticity.

Synaptic plasticity involves neuronal differentiation, movement, synaptogenesis and up- and down-regulation, all requiring, in one way or another, MTs, major structural components of the cytoskeleton. MTs are composed of tubulin, 110 kD protein hetero-dimers which self-assemble into hollow cylinders 25 nm in outer and 15 nm in inner diameter. MT walls have been crystallographically characterized as forming two types of hexagonal lattices (A-lattice and B-lattice) with helical winding patterns, including those following Fibonacci geometry [15]. Each tubulin dimer in MT lattices may occupy different states, and interact with neighbor tubulin dimer states, suggesting to a number of authors [16]–[20] that MTs process information in terms of tubulin states, and function as computational devices, e.g. molecular automata (‘microtubule automata’).

MTs in neuronal axons are arranged in continuous, uninterrupted parallel bundles. However, MTs in dendrites and neuronal cell bodies are uniquely arranged in mixed polarity, anti-parallel arrays of interrupted MTs, interconnected by MT-associated proteins (“MAPs”) including MAP2, whose activities are implicated in learning ([21] and references therein).

Various types of MAPs include those which cross-bridge MTs into scaffold-like networks defining neuronal architecture, and motor proteins (e.g. dynein, kinesin) which transport molecular cargo along MTs for delivery to specific synaptic locations. In most cells, MTs are labile and tend to undergo cycles of assembly and disassembly. However in neuronal dendrites and cell bodies, MTs are capped and stabilized by specialized MAPs including MAP-2 [22], [23].

MTs in brain neurons may also be stabilized by post-translational modifications. α-tubulin isotypes have either tyrosine, or glutamate at their C-terminal ends. Generally, MTs with abundant glutamate-terminating tubulins are stable, while MTs with many tyrosine-terminating tubulins are labile and short-lived [24], [25]. Terminal tyrosine may be removed (“detyrosination”) via tubulin carboxypeptidase, yielding glutamate-terminating tubulin and giving rise to disassembly-resistant MTs [26]. Stable MTs composed of glutamate-terminating α-tubulin appear to interact via MAPs with another, extremely stable cytoskeletal component, namely intermediate filaments [27]. Thus short-term memory stored in MTs may be further encoded and hard-wired in neurofilaments for long-term memory.

Evidence for MT-based information processing includes correlations between tubulin production and visual processing [28], MT signaling between membrane proteins and cell nucleus [29], [30], and MT-MAP alterations correlating with memory, novel learning and cytoskeletal remodeling [31]. Deficits in cytoskeletal function affect learning and memory [20], with the best-known correlation being disruption of MTs and resulting neurofibrillary tangles in brains of Alzheimer's disease patients.

In this study we performed molecular mechanics modeling and electrostatic profiling of binding and phosphorylation between activated CaMKII holoenzyme kinase domains and tubulin lattices in MTs, and analyzed their information processing capabilities.

CaMKII is a holoenzyme with twelve kinase regions in two hexagons sandwiched around a central association domain (see Figure 1 A). Activated by Ca2+/CaM, the kinases extend outward from the association domain, remaining tethered by linker regions (see Figure 1 B).

The phosphorylation process takes place on the inner surface of the kinase domains, in ‘hydrophobic groove’ regions characterized by valine 208 and tryptophan 237 (and, in the inactive state, occupied by threonine 286 on the autoregulatory region). The S-T hydrophobic phosphorylation site is exemplified by valine 208. Symmetric arrangement of the kinase regions provides a unique electrostatic profile in which each kinase region presents a positive potential region surrounded by negatively charged surfaces (see Figure 1 C). Substrates with reciprocal patterns of negatively charged regions on a positive background are excellent candidates for attractive electrostatic interactions.

As previously described, MTs are cylindrical polymers of hetero-dimer tubulin proteins, each composed of α- and β-tubulin monomers (see Figure 2 A). Tubulin dimers self-assemble into MTs, which are hollow cylinders of 13 linear tubulin chains known as ‘protofilaments’ (see Figure 2B). Side-to-side electrostatic interactions between tubulins in parallel protofilaments comprising the cylinder result in two types of skewed hexagonal lattices and helical winding pathways, the so-called A-lattice and B-lattice [32] (see Figure 2 C & D).

Overall, tubulin has a large net negative surface charge with almost half of the negatively charged residues located on the C-terminal tail of each monomer, which extend from the surface into the cytoplasm (see Figure 2A). This net electrostatic charge is, obviously, neutralized by counter-ions in the cytoplasm. Surface charges differ between α- and β monomers resulting in a net dipole moment pointing from the α- towards the β-monomer. Thus, MTs have net dipole moments, reflecting alignment of tubulin dipoles, and a unique skewed hexagonal electrostatic pattern of highly negatively charged regions surrounded by a positive background, dependent on the MT lattice type (see Figure 2 C & D).

Generally, CaMKII shows a preference for phosphorylating protein substrates at the amino acid sequence arginine-X-X-serine/threonine, where X can be any amino acid (although exceptions occur [33]). Wandosell et al. [34] showed that free, unpolymerized α- and β-tubulin are phosphorylated by activated CaMKII on or near the C-terminal region (beyond residue 306), resulting in tubulin conformational change, inhibition of assembly and inability to bind MT-associated protein 2 (MAP2). Multiple serine and threonine residues capable of being phosphorylated exist in this region (see Figure 2) however, the exact CaMKII phosphorylation site, or sites, on tubulin are unknown. Several sites on α- and β-tubulin follow this consensus sequence, however only one falls within the C-terminal region [35]. That site involves threonine 312 located not on the surface, but buried in the cleft between the α- and β-monomers. Another potential site is serine 444 on the βIII-tubulin C-terminal tail. Experiments suggest that phosphorylation of mammalian β-tubulin is restricted to the βIII isotype [27], which is found predominately in neurons. While alternate sites on βIII-tubulin have been argued for [36], it is clear that phosphorylation of tubulin by CaMKII holoenzymes occurs, though the precise sites remain unclear.

Size and geometry of the activated hexagonal CaMKII holoenzyme and the two types of hexagonal lattices (A and B) in MTs are identical (see Figure 3). The CaMKII hexagon is approximately 20 nm in breadth (see Figure 3 A), and the maximum distances between individual tubulin monomer in both the A lattice and B lattice MT neighborhoods are the same (see Figure 3 B). With minimal realignment in the linkers between association and kinase domains to account for MT curvature and lattice asymmetry, CaMKII precisely matches the MT A-lattice geometry, i.e. 6 extended kinases can interface collectively with 6 tubulins (see Figure 3 C – upper panel). Overlaying CaMKII with the 9-tubulin neighborhood of the MT B lattice requires slightly greater flexibility to precisely match the geometry (see Figure 3 C – middle and lower panels). This geometric matching permits activated CaMKII holoenzymes to bind to MT surfaces (see Figure 4 A & B).

The electrostatic pattern formed by a neighborhood of tubulin dimers on a MT surface (see Figure 2 C & D) shows highly negative charged regions surrounded by a less pronounced positive background, dependent on the MT lattice type (see Figure 2 C & D). These electrostatic fingerprints are complementary to those formed by the 6 CaMKII holoenzyme kinase domains making the two natural substrates for interaction. Alignment of the CaMKII holoenzyme with tubulin dimers in the A-lattice MT arrangement yields converging electric field lines indicating a mutually attractive interaction (see Figure 4 C & D). Considering the positive potential region of an individual kinase domain to be 1 kT/e and the negative surface of an individual tubulin monomer to be −10 e gives an electrostatic attraction of 10 kT, or 6 kcal/mol at 310 K, for each kinase-monomer interaction, the lower bound of binding. This attraction reaches a maximum bound of 60 kT (36 kcal/mol) when all 6 kinase regions are involved. This range of attraction is much stronger than thermal vibrational energy and indicates a significant association.

If each extended kinase can either phosphorylate at the S-T site on a tubulin substrate, or not, the process effectively conveys one bit of information (e.g. no phosphorylation = 0, phosphorylation = 1). Each set of six extended kinases on either side of a CaMKII holoenzyme can thus act collectively as 6 bits of information. Ordered arrays of bits are termed ‘bytes’.

Three hypothetical scenarios for CaMKII information encoding in MT lattice are considered (see Figure 5). In the first situation only β-tubulins in the 7-tubulin neighborhood patch of an A-lattice MT may be phosphorylated, giving two possible states – no phosphorylation (0), and phosphorylation (1) (see Figure 5 A). The scenario would be identical if only α-tubulin phosphorylation was considered. The central dimer is not considered for phosphorylation, so 6 dimers are available. Thus, there are 26 possible encoding states for a single CaMKII-MT interaction resulting in the storage of 64 bits of information. This case, however, only accounts for either α- or β-tubulin phosphorylation, not both. In the second scenario each tubulin dimer is considered to have three possible states – no phosphorylation (0), β-tubulin phosphorylation (1), or α-tubulin phosphorylation (2) (see Figure 5 B). These are ternary states, or ‘trits’ (rather than bits). Six possible sites on the A-lattice yield 36 = 729 possible states. The third scenario considers the 9-tubulin B-lattice neighborhood with ternary states. As in the previous scenarios the central dimer is not considered available for phosphorylation. In this case, 6 tubulin dimers out of 8 may be phosphorylated in three possible ways. The total number of possible states for the B lattice neighborhood is thus 36–28−8(27) = 5281 unique states.

Each phosphorylation event requires hydrolysis of a single ATP molecule, releasing ∼20 kT, or 12 kcal/mol of free energy at physiological temperature for a single kinase-monomer interaction, suggesting a robust molecular mechanism for information encoding. The daily energy usage of the human body requires the production of more than 2×1026 molecules of ATP [37]. Brain processes consume approximately one fifth this amount using 5×1020 ATP molecules per second. During a calcium influx there are ∼100 CaMKII subunits activated/phosphorylated per synaptic dendrite [38], each using a single ATP molecule. Taking calcium influx signals at a frequency of 100 Hz, activating 100 CaMKII subunits per synapse, and ∼104 synapses per neuron, one neuron would require 108 ATP/second for maximal encoding. With 1011 neurons in the brain this amounts to 1019 ATP/second, equivalent to 1019 bits, at a cost of approximately 2% of the brain's total energy consumption.

Metabolic energy for the proposed memory encoding by CaMKII phosphorylation of MTs should be distinguished from significantly higher metabolic costs of more coarse-grained forms of neuronal information processing. For example, hydrolysis of 104 ATP molecules is required to transmit one bit of information at a chemical synapse, and hydrolysis of 106 to 107 ATPs is needed for graded signals or spike coding [39].

We presume memory encoded by CaMKII phosphorylation alters, programs and provides a background for ongoing MT-based information processing e.g. microtubule automata, finer-grained processes underlying membrane potentials and synaptic activities. Information processing in post-synaptic dendritic and somatic MTs can thus: 1) integrate inputs to firing threshold at the proximal axon, and 2) regulate synaptic plasticity. In the following, we describe 6 possible mechanisms by which CaMKII-encoded tubulin phosphorylation in MT lattices could regulate cytoplasmic, membrane and neuronal structure and function leading to cognition and behavior.

Each tubulin dimer has two C-terminal ‘tails’, one from each monomer (see Figure 6). Modeling has shown that C-terminal tails can exist in multiple states, and that the tails may dynamically oscillate between extended conformations, involving interaction with water and ions (the “up-state”) (see Figure 6A), and folded conformations, involving interactions with the tubulin body (the “down-state”) (see Figure 6B) [40]. Thus, under proper conditions the negatively charged tails extend outward into the cytoplasm, able to alter local electric fields, as well as ionic and chemical states. CaMKII-induced phosphorylation of particular tubulins may cause dynamical (e.g. hexagonal) patterns of C-terminal extension, able to precipitate reaction-diffusion waves. Reaction-diffusion patterns known as Turing patterns [41], [42] are thought to function in intra-cellular information processing.

MAPs bind on the surface of individual tubulins in the region where phosphorylation occurs [27], [39], and removal of phosphate groups on tubulin results in decreased MT assembly in the presence of MAP2 [43]. These findings suggest that tubulin phosphorylation promotes MAP-MT interaction [27]. CaMKII induced phosphorylation patterns on MTs would thus both promote MAP-MT binding in general, and also serve as templates for attachment of MAPs at specific sites. MAPs attached in this way can form bridges between MTs, and thus cytoskeletal scaffolding determining MT spacing, neuronal architecture and synaptic location (Figure 7 A & B).

Figure 7 C shows pathways of phosphorylated tubulins following MT lattice geometry, intersecting at various tubulin sites. In Figure 7 D the intersection points are sites for MAP attachment. Patterns of phosphorylated tubulins can determine cytoskeletal morphology and cellular function.

In the absence of CaMKII, MAPs bind to MTs in vitro in super-helical lattice geometric patterns [44]–[46]. These may reflect resonance nodes of MT lattice vibrations with which CaMKII phosphorylation and other information modes interact.

Motor proteins dynein and kinesin move (in opposite directions) along MTs (using ATP as fuel) to transport and deliver components and precursors to specific synaptic locations. While MTs are assumed to function as passive guides, like railroad tracks, for motor proteins, the guidance mechanism by which they follow specific paths, e.g. through branching dendrites and interrupted MTs is unknown. Phosphorylation patterns on MT lattices could guide motor proteins to specific locations (see Figure 8 A–C).