STDP in a Bistable Synapse Model Based on CaMKII and Associated Signaling Pathways

Abstract

The calcium/calmodulin-dependent protein kinase II (CaMKII) plays a key role in the induction of long-term postsynaptic modifications following calcium entry. Experiments suggest that these long-term synaptic changes are all-or-none switch-like events between discrete states. The biochemical network involving CaMKII and its regulating protein signaling cascade has been hypothesized to durably maintain the evoked synaptic state in the form of a bistable switch. However, it is still unclear whether experimental LTP/LTD protocols lead to corresponding transitions between the two states in realistic models of such a network. We present a detailed biochemical model of the CaMKII autophosphorylation and the protein signaling cascade governing the CaMKII dephosphorylation. As previously shown, two stable states of the CaMKII phosphorylation level exist at resting intracellular calcium concentration, and high calcium transients can switch the system from the weakly phosphorylated (DOWN) to the highly phosphorylated (UP) state of the CaMKII (similar to a LTP event). We show here that increased CaMKII dephosphorylation activity at intermediate Ca2+ concentrations can lead to switching from the UP to the DOWN state (similar to a LTD event). This can be achieved if protein phosphatase activity promoting CaMKII dephosphorylation activates at lower Ca2+ levels than kinase activity. Finally, it is shown that the CaMKII system can qualitatively reproduce results of plasticity outcomes in response to spike-timing dependent plasticity (STDP) and presynaptic stimulation protocols. This shows that the CaMKII protein network can account for both induction, through LTP/LTD-like transitions, and storage, due to its bistability, of synaptic changes.

Calcium control of triphasic hippocampal STDP

“…For example, the magnitude of weight change induced by different stimulation protocols in this model is generally smaller than that observed empirically, despite the fact that the relative frequency and conductance of potentiated and depressed synapses are matched to published data [36, 37, 44; see Methods]. Furthermore, specific AMPAr and NMDAr conductances are not affected by changes in putative synaptic weight which, although in line with many theoretical studies of plasticity induction, is clearly at odds with the situation in vivo [18,[42][43][44][45][46][47][48]. It has been demonstrated that changes in AMPAr conductance associated with the expression of long-term synaptic plasticity are accompanied by concomitant changes in NMDAr conductance [69,76].…”

“…The slow time constant of Calcium influx through the NMDA receptor is also shorter than that measured experimentally, which has been shown to correspond well with the time course of deactivation following glutamate release [~150ms; 57]. However, that study measured the time course of Calcium influx following unitary NMDAr activation, whereas we are concerned with NMDAr kinetics when significant post-synaptic depolarisation follows glutamate binding by ~50-100ms, which may be significantly more complex [46].…”

“…It has also been suggested that one of several alternative mechanisms may serve as the second coincidence detector required for the induction of LTD by spike and triplet pairing stimulation protocols [42,45,66]. By elaborating the secondary messenger pathways in the plasticity model to more closely replicate the activity of CaM, CaMKII, PP1, PP2A, I1 and other proteins implicated in plasticity expression, as well as additional sources of internal and external Calcium influx, it may be possible to more accurately appraise the contributions made by bAPs to synaptic plasticity [7,46,67]. However, the complexity of such models and associated computational cost might also render them unsuitable for network level simulation.…”

“…Firstly, we aim to ascertain whether the Calcium control hypothesis -which has been demonstrated to successfully reproduce earlier STDP data obtained in culture (Figure 1a), as well as that induced by other activity patterns -can be revised to account for this joint dependency [18,[41][42][43][44][45][46][47][48]. The experimental data we aim to replicate can be characterised by considering the effects of two different stimulation protocols -pairing 100 single preand post-synaptic spikes (hereafter referred to as 'spike pairing') at low frequencies (0.1-5Hz), which generates a depression-only learning rule ( Figure 1b); or pairing a single pre-synaptic spike with two post-synaptic spikes (hereafter referred to as 'triplet pairing'), which generates a triphasic bidirectional learning rule after 100 pairings at a frequency of 5Hz (Figure 1c), an unsaturated potentiation-only rule after 30 pairings at 5Hz (Figure 1d), or mild depression after 100 pairings at a frequency of 0.5Hz (data not shown).…”

“…The kinetics of kinase and phosphotase activation also differ significantly, as LTP can be rapidly induced by appropriate patterns of activity while LTD requires prolonged stimulation [15,17,29]. Computational modelling of synaptic plasticity has demonstrated that the dynamics of Calcium influx through the NMDA receptor (NMDAr) is sufficient to account for empirical data obtained using multiple stimulation protocols, integrating an array of experimental results within a single theoretical framework [18,[41][42][43][44][45][46][47][48]. However, these models as yet fail to assimilate or replicate various aspects of more recently obtained synaptic plasticity data such as those described above.…”