This subproject is one of many research subprojects utilizing theresources provided by a Center grant funded by NIH/NCRR. The subproject andinvestigator (PI) may have received primary funding from another NIH source,and thus could be represented in other CRISP entries. The institution listed isfor the Center, which is not necessarily the institution for the investigator.Monte Carlo simulation of presynaptic calcium dynamics and neurotransmitterrelease. This computational project is directed by J. Stiles and is beingcarried out by John Pattillo, a post-doc in his lab. Using realistic nerveterminal ultrastructure and data such as that described in II.A.1.c, MCellsimulations of active zone calcium dynamics encompass actionpotential-activation of voltage-gated calcium channels, stochastic calcium ionentry and diffusion, calcium binding to sensor sites on arrays of synapticvesicles, and prediction of vesicle fusion and resulting transmitter release. To our knowledge, this is the only study to date that has included the 3-Dstructure of an entire presynaptic active zone, and that has used multipleexperimental constraints to enable quantitative predictions, e.g., the numberof calcium-binding sites on synaptic vesicles, and the relationship betweennumber of binding sites, number of sites that must be bound to initiateneurotransmitter release, and the importance of active zone spatialorganization.In brief, a supralinear (~4th order)1 relationship (CRR) between extracellularCa2+ ([Ca2+]o) and transmitter release indicates that multiple Ca2+ ions arerequired for fusion of a synaptic vesicle (SV), but how this empiricalobservation relates to the stoichiometry and architecture of voltage-gated Ca2+channels (VGCCs), Ca2+ binding sites, and SVs is unclear. We created aspatially realistic model of a frog neuromuscular active zone (AZ), and usedMCell to simulate action potential (AP)-induced Ca2+ influx through VGCCs, Ca2+binding to SVs, and several models of Ca2+-dependent SV fusion. We variedspatial parameters to simultaneously reproduce 3 experimental observations: 1.) average release probability (pr) per trial per AZ at physiological[Ca2+]o; 2.) the distribution of release latencies (Ldis); and 3.) the 4th order CRR. Also, a 4-state VGCC model reproduced macroscopic Ca2+ current kinetics, andthe on and off rates for Ca2+ binding were based on the synaptotagmin-1 C2Adomain. Given all these constraints, we obtained a surprisingly unique set ofmodel parameters and several counter-intuitive predictions. With a VGCC:SVstoichiometry of 1:1 (supported by the experimental and mathematical modelingdata outlined above), each SV contains ~20 Ca2+ binding sites, and 6 sites mustbe bound simultaneously to induce fusion. Alternative models were either muchtoo Ca2+-insensitive to reproduce pr or could not simultaneously reproduce Ldisand CRR. These results demonstrate the dramatic sensitivity of CRR, pr, andLdis to presynaptic architecture, and suggest that vesicle fusion may require avariety of SNARE protein and membrane lipid binding sites for Ca2+. This workhas been published in abstract form (Pattillo et al., 2004), and several fulllength manuscripts are in preparation. This project has required something on the scale of 105 simulations to date,primarily run on the PSC HP GS1280 machine(s), for which we are one of thepreferred user groups. This machine is based on latest-generation Alpha EV7processors, large shared memory, and outstanding memory bandwidth, and isoptimally suited to our Monte Carlo algorithms and run-time optimizationswithin MCell. Specifically, MCell simulations require larges amounts of memorywith random access patterns. In addition, this project admirably demonstratesthe advantages to MCell's unique Monte Carlo algorithms for bimolecularinteractions. The spatial dimensions of the active zone are tightly confined,and our simulations show that the average calcium concentration in the vicinityof vesicular binding sites corresponds to less than a single ion at any instantin time. Despite these conditions, MCell is able to accurately simulate thesecalcium dynamics with a time step on the sub-microsecond scale, rather than thesub-nanosecond scale (as would be required with less sophisticated algorithmsfor bimolecular interactions). Thus, this project has been possible onlythrough a combination of optimized algorithms coupled with outstandinglydesigned and supported hardware.Computational ChallengesThese simulation have been performed using PSC's Marvel systems. Within thisstudy, we are usually running one 'project' at any given time. Each 'project'includes 24 'sets' of simulations, and each 'set' requires 500-1000 separate(embarrassingly parallel) simulations, each of which runs in 3 GBytes of RAM. Because of the Marvel's outstanding memory bandwidth and MCell's frequentrandom memory accesses, our simulations run very efficiently even compared toother more recent processors running at higher clock speeds. Perhaps even moreimportant, we have never had any problems related to compilers or operatingsystem issues. This is especially impressive given that each 'project'generates up to 48 million output files that would consume up to 2.4 TBytes ofdisk space, except that we post-process the results on-the-fly, obtaining areduction of ~1000-fold before transfer to mass storage. Without a stablesystem combining large memory, outstanding memory bandwidth, fast I/O, andreliable transfer to mass storage, our projects probably could not have beendone.Publications:Pattillo, JM, Meriney, SD, and Stiles, JR., 2004, in press, Spatially realisticMonte Carlo simulations predict calcium dynamics underlying transmitter releaseat a neuromuscular active zone. Soc. Neurosci. Abst.Footnotes:1. The calcium source is generally more than one channel, each of which isat a different distance from the vesicle that happens to fuse. The calciumsensing (binding) sites are arrayed around the base of each vesicle. Thecalcium gradient is very steep and different (in space and time) from eachchannel to each sensor. Thus it is very different from a situation in whichmultiple binding sites are each responding to the same calcium signal. Theapparent cooperativity also depends on how we define the fusion model, e.g.,the results are different depending on whether or not we require ~6 sites to bebound simultaneously or just to have been bound at some point in time.
Showing the most recent 10 out of 292 publications
