TITLE Multisite synapse

COMMENT
-----------------------------------------------------------------------------
Multi-site synapse with independent release sites. Each site operates independently
and releases a vesicle upon presynaptic depolarization with a probability 
determined by the history of activity, using the Dittman and Regehr (1998, 2000)
model.

Revised from coh2.mod, coh3.mod, and coh4.mod.
The Dittman and Regeher (1998, 2000) release model with
facilitation closely fits  Auditory Nerve data from mouse over
a wide range of frequencies.
The model DOES NOT include the postsynaptic receptors or desensitization, since
these should be treated separately (couple XMTR to an AMPA receptor model,
such as the Trussell-Raman model)

Range variables:
nZones: is the number of active zones simulated in this calyx model. Each zone
        can be connected to a separate PSD.
F (0.4): The base release probability
k0 (1/1.75): /s, baseline recovery rate from depletion (slow rate)
kmax (1/0.025): /s, maximal recovery rate from depletion (fast rate)
td (0.05) : time constant for fast calcium-dependent recovery, sec
kd (0.7) : affinity of fast recovery process for calcium sensor
kf (0.5) : affinity of facilitation process
tf (0.01) : rate of facilitation process (slow) seconds
dD (0.02): calcium that drives recovery (ca influx per AP)
dF (0.02): calcium that drives facilitation

Added latency and variable delay (latstd, latency standard deviation in msec)
around the mean spike time. 4/5/2011 pbm.

Version 4 uses a log-normal distribution to determine release latencies. 
The calculation is built-in instead of being passed through an array. 
The lognormal distribution describes the individual vesicle release time
course at this synapse as measured by Isaacson and Walmsley, 1996. Note that
they used a gamma distribution in some plots, but the lognormal distribution 
seems to fit their published data at least as well. 
The parameters of the distribution, as well as the release latency,
are controlled by an exponential function whose parameters are initialized at
run time. 
10/19/2011 Paul B. Manis, UNC Chapel Hill

ENDCOMMENT

DEFINE MAX_ZONES 1000  : maximum number of zones in this model
DEFINE EVENT_N 10000   : number of entries in the Event Distribution (e.g., as sampled)

INDEPENDENT {t FROM 0 TO 1 WITH 1 (ms)}

NEURON {
    THREADSAFE
    POINT_PROCESS MultiSiteSynapse
    RANGE F, k0, kmax, taud, kd, tauf, kf
    RANGE nZones, multisite, rseed, latency, latstd, debug
    RANGE dD, dF, XMTR, glu, CaDi, CaFi
    RANGE Fn, Dn
    RANGE TTotal
    RANGE nRequests, nReleases
    RANGE Identifier : just a number so we can report which instance is active
    RANGE tau_g, amp_g
    : Distributions for stochastic release and testing (Sept, Oct, 2011):
    RANGE EventLatencies, EventTime : returns the first EVENT_N latencies and absolute times at which they were used
    RANGE ev_index : count in the EventLatencies (in case we are "short")
    : parameters for latency shift during repetitive stimulation (Oct 19, 2011)
    RANGE Dep_Flag : Depression flag (0 to remove depression; 1 to allow DKR control of facilitation and depression)
    RANGE Lat_Flag, Lat_t0, Lat_A0, Lat_tau : Lat_Flag  = 0 means fixed latency (set by "latency" above)
                                            : otherwise, latency = latency for t < Lat_t0
                                            :            latency = latency + Lat_A0*(1-exp(-(t-Lat_t0)/Lat_tau))
    : parameters for lognorm distribution shift during repetitive stimulation (Oct 19, 2011)
    RANGE LN_Flag, LN_t0, LN_A0, LN_tau : LN_Flag  = 0 means fixed sigma as well
                                            : otherwise, sigma = latstd for t < LN_t0
                                            :            sigma = latstd + LN_A0*(1-exp(-(t-LN_t0)/LN_tau))

    : externally assigned pointers to RNG functions
    POINTER uniform_rng  : for deciding the number of active synapses when multisite==0
}

UNITS {
    (mA) = (milliamp)
    (mV) = (millivolt)
    (mM) = (milli/liter)
    (uM) = (micro/liter)
}

PARAMETER {
    dt      (ms)
    amp_g = 1.0 (mM)          : amplitude of transmitter pulse
    tau_g = 0.5    (ms)    : duration of transmitter pulse
    dD = 0.02 (1)       : calcium influx driving recovery per AP
    dF = 0.02 (1)       : calcium influx driving facilitation per AP
    F  = 0.5 (1)        : basal facilitation
    k0 = 0.0005714(/ms)       : slow recovery from depletion (1.0/1.75)
    kmax = 0.040 (/ms)      : fast recovery from depletion (1/0.025)
    taud = 50.0 (ms)        : time constant for fast calcium dependent recovery
    kd = 0.7 (1)       : affinity of fast recovery process for calcium sensor
    tauf = 10.0 (ms)        : rate of slow facilitation process
    kf = 0.5 (1)       : affinity of slow facilitation process
    : taus = 1 (ms)    : defined by DKR but not used here 
    : ks = 0.5 (1)     
    : glu = 1 (mM)
    rseed (1)        : random number generator seed (for SCOP module)
    latency = 0.0 (ms)
    latstd = 0.0 (ms)
    
    : Time course of latency shift in release during repetitive stimulation
    Lat_Flag = 0 (1) : 0 means fixed latency, 1 means lognormal distribution
    Lat_t0 = 0.0 (ms) : minimum time since simulation start before changes in latency are calculated
    Lat_A0 = 0.0 (ms) : size of latency shift from t0 to infinity
    Lat_tau = 100.0 (ms) : rate of change of latency shift (from fit of a+b(1-exp(-t/tau)))
    : Statistical control of log-normal release shape over time during repetitive stimulation
    LN_Flag = 0 (1) : 0 means fixed values for all time
    LN_t0 = 0.0 (ms) : : minimum time since simulation start before changes in distribution are calculated
    LN_A0 = 0.0 (ms) : size of change in sigma from t0 to infinity
    LN_tau = 100.0 (ms) : rate of change of sigma over time (from fit of a+b*(1-exp(-t/tau)))
    
    : control flags     - if debug is 1, show all, if 2, just "some"
    debug = 0
    Identifier = 0
    Dep_Flag = 1 (1) : 1 means use depression calculations; 0 means always set release probability to F
}

ASSIGNED {
    : Externally set assignments
    nZones (1)    : number of zones in the model
    multisite (1)  : whether zones are modeled individually (1) or as a single, variable-amplitude zone (0)
    nRequests (1) 
    nReleases (1)
    EventLatencies[EVENT_N] (0)
    EventTime[EVENT_N] (0)
    tRelease[MAX_ZONES] (ms)    : time of last release

    : Internal calculated variables
    Fn (1)
    Dn (1)
    CaDn (1)
    CaFn (1)
    CaDi (1)
    CaFi (1)
    eta (1)
    tSpike (ms)            : time of last spike
    tstep(ms)
    TTotal(0)
    tspike (ms)
    latzone (ms)
    vesicleLatency (ms)
    sigma (ms)
    gindex (0)
    ev_index (0)
    scrand (0)    
    
    uniform_rng
}

: Function prototypes needed to assign RNG function pointers
VERBATIM
double nrn_random_pick(void* r);
void* nrn_random_arg(int argpos);
ENDVERBATIM

: Return a pick from uniform distribution.
: (distribution parameters are set externally)
FUNCTION rand_uniform() {
VERBATIM
    _lrand_uniform = nrn_random_pick(_p_uniform_rng);
ENDVERBATIM
}

: Function to allow RNG to be externally set
PROCEDURE setUniformRNG() {
VERBATIM
 {
    void** pv = (void**)(&_p_uniform_rng);
    *pv = nrn_random_arg(1);
 }
ENDVERBATIM
}


STATE {
    XMTR[MAX_ZONES]       (mM)       : per-zone neurotransmitter concentration
    N_ACTIVE[MAX_ZONES]       (1)    : number of zones actively releasing
}

INITIAL {
:    VERBATIM
:        fprintf(stdout, "MultiSiteSynapse: Calyx #%d Initialized with Random Seed: %d\n", (int)Identifier, (int)rseed);
:    ENDVERBATIM

    TTotal = 0
    nRequests = 0
    nReleases = 0
    set_seed(rseed)
    tSpike = -1e9
    latzone = 0.0
    sigma = 0.0
    vesicleLatency = 0.0
    gindex = 0
    ev_index = 0
    scrand = 0.0
    CaDi = 1.0
    CaFi = 0.0
    CaDn = 1.0
    CaFn = 0.0
    Fn = F
    Dn = 1.0
    FROM i = 0 TO (nZones-1) {
        XMTR[i] = 0
        N_ACTIVE[i] = 1
        tRelease[i] = tSpike
    }
    update_dkr(t-tSpike)
}

BREAKPOINT {
    SOLVE release
}

LOCAL tz, n_relzones, amp
PROCEDURE release() {
    : Once released, the transmitter packet has a defined smooth time course in the "cleft"
    : represented by the product of rising and falling exponentials.
    : update glutamate in cleft
    if (multisite == 1) {
        : Update glutamate waveform for each active release zone
        n_relzones = nZones
    }
    else {
        : Update aggregate glutamate waveform for only the first release zone
        n_relzones = 1
    }
    
    FROM i = 0 TO (nZones-1) { : for each zone in the synapse
        if (t >= tRelease[i] && t < tRelease[i] + 5.0 * tau_g) {
            tz = t - tRelease[i] : time since onset of release
            : calculate glutamate waveform (Xie & Manis 2013 Supplementary Eq. 1)
            XMTR[i] = amp_g * (1.0-exp(-tz/(tau_g/3.0))) * exp(-(tz-(tau_g/3.0))/tau_g) 
        }
        else {
            XMTR[i] = 0
        }
    }
}


PROCEDURE update_dkr(tstep (ms)) {
    : update the facilitation and depletion variables
    : from the Dittman-Regehr model. 
    : Updates are done with each new presynaptic AP event.
    if(tstep > 0.0) {
        CaDi = CaDi + dD
        CaFi = CaFi + dF
        CaDn = CaDi * exp (-tstep/taud)
        CaFn = CaFi * exp (-tstep/tauf)
        eta = (kd/CaDi + 1.0)/(kd/CaDi + exp(-tstep/taud))
        eta = eta^(-(kmax-k0)*taud)
        Dn = 1.0-(1.0-(1.0-Fn)*Dn)*exp(-k0*tstep)*eta
        Fn = F + (1.0-F)/(1.0+kf/CaFn)
        CaDi = CaDn
        CaFi = CaFn
    }
    if (Dep_Flag == 0) { : no depression
        Dn = 1.0  : set to 1
        Fn = F   : set to initial value to set release probability constant
    }
:    VERBATIM
:        if (debug >= 2 ){
:            fprintf(stdout, "update start t = %f ts=%f: F=%7.2f CaDi = %g CaFi = %g\n", \
:            t, tstep, F, CaDi, CaFi);
:            fprintf(stdout, "    vars:  taud=%g: tauf=%g kd = %g kmax= %g\n", taud, tauf, kd, kmax);
:            fprintf(stdout, "    CaDi = %g CaFi = %g\n", CaDi, CaFi);
:            fprintf(stdout, "    CaDn = %g CaFn = %g\n", CaDn, CaFn);
:            fprintf(stdout, "    eta: %g\n", eta);
:            fprintf(stdout, "    Fn=%7.2f Dn: %7.2f  CaDi = %g CaFi = %g,\n", \
:            Fn, Dn, CaDi, CaFi);
:        }
:    ENDVERBATIM
}


NET_RECEIVE(weight) {
    : A spike has been received; process synaptic release
    
    : First, update DKR state to determine new release probability
    update_dkr(t - tSpike)
    tSpike = t      : save the time of spike
    
    TTotal = 0 : reset total transmitter from this calyx for each release
    nRequests = nRequests + 1 : count the number of inputs that we received
    
    : Next, process vesicle release using new release probability
    if (multisite == 1) {
        release_multisite()
    }
    else {
        release_singlesite()
    }
}


PROCEDURE release_multisite() {
    : Vesicle release procedure for multi-site terminal.
    : Loops over multiple zones using release probability Fn*Dn to decide whether
    : each site will release, and selecting an appropriate release latency. 
    
    : The synapse can release one vesicle per AP per zone, with a probability 0<p<1.
    : The probability, p, is defined by the time evolution of a Dittman-Regher model
    : of release, whose parameters are set during initialization.
    : The vesicle can be released over a variable time interval defined by a lognormal
    : distribution, plus a fixed latency.
    
    FROM i = 0 TO (nZones-1) { : for each zone in the synapse
        if(tRelease[i] < t) {
            scrand = scop_random()
            : look to make release if we have not already (single vesicle per zone per spike)
            : check for release and release probability - assume infinite supply of vesicles
            if (scrand  < Fn*Dn) { 
                nReleases = nReleases + 1 : count number of releases since inception
                TTotal = TTotal + 1     : count total releases this trial.

:               Compute the median latency for this vesicle.
                if (Lat_Flag == 0 || t < Lat_t0) {
                    vesicleLatency = latency : use a fixed value
                }
                else {
                    vesicleLatency = latency + Lat_A0*(1-exp(-(t-Lat_t0)/Lat_tau)) : latency rises during train
                }
:               Now compute distribution around that latency
:               if LN_Flag is 1, we adjust the sigma values over time, otherwise we just use a fixed value.
:               The following math applies:
:               lognormal dist = exp(X), where X = gaussian(mu, sigma).
:               The median of the lognormal dist. is e^mu; Note that if mu is 0, then the median is 1.0
:               The mode of the lognormal dist is e^(u-sigma^2). 
:               Note that normrand is a SCoP function, see http://cns.iaf.cnrs-gif.fr/files/scopman.html
                if (LN_Flag == 0 || t < LN_t0) { : use fixed sigma in lognormal distribution for all time.
                    if (latstd > 0.0) {
                        latzone = normrand(0.0, latstd) : set a latency for the zone with one draw from the distribution
                        latzone = exp(latzone) - 1.0 + vesicleLatency : the latency should not be too short.... 
                    }
                    else {
                        latzone = vesicleLatency : fixed value
                    }
                }
                else {
                    sigma = latstd + LN_A0*(1-exp(-(t-LN_t0)/LN_tau)) : time-dependent std shift
                    latzone = normrand(0.0, sigma)
                    latzone = exp(latzone)-1.0 + vesicleLatency
                }
                if (latzone < 0.0) { : this is to be safe... must have causality.
                    latzone = 0.0
                }
                if (ev_index < EVENT_N) { : save event distribution list for verification
                    EventLatencies[ev_index] = latzone
                    EventTime[ev_index] = t
                    ev_index = ev_index + 1
                }
                
                : release time for this event
                tRelease[i] = t + latzone
            }
        }
    }
}


PROCEDURE release_singlesite() {
    LOCAL pr
    tRelease[0] = t
    pr = Fn * Dn
    FROM i = 0 TO (nZones-1) {
        if (rand_uniform() < pr) { 
            TTotal = TTotal + 1     : count total releases this trial.
        }
    }
    : Tell PSD to multiply its final current by the number of active zones
    N_ACTIVE[0] = TTotal
    printf("Release: %f\n", TTotal)
}