from moose_nerp.prototypes.util import NamedDict
from moose_nerp.prototypes.chan_proto import (
AlphaBetaChannelParams,
StandardMooseTauInfChannelParams,
TauInfMinChannelParams,
ZChannelParams,
BKChannelParams,
ChannelSettings,
TypicalOneD,
TwoD,
)
#contains all gating parameters and reversal potentials
# Gate equations have the form:
# AlphaBetaChannelParams (specify forward and backward transition rates):
# alpha(v) or beta(v) = (rate + B * v) / (C + exp((v + vhalf) / vslope))
# OR
# StandardMooseTauInfChannelParams (specify steady state and time constants):
# tau(v) or inf(v) = (rate + B * v) / (C + exp((v + vhalf) / vslope))
# OR
# TauInfMinChannelParams (specify steady state and time constants with non-zero minimum - useful for tau):
# inf(v) = min + max / (1 + exp((v + vhalf) / vslope))
# tau(v) = taumin + tauVdep / (1 + exp((v + tauVhalf) / tauVslope))
# or if tau_power=2: tau(v) = taumin + tauVdep / (1 + exp((v + tauVhalf) / tauVslope))* 1 / (1 + exp((v + tauVhalf) / -tauVslope))
#
# where v is membrane potential in volts, vhalf and vslope have units of volts
# C, min and max are dimensionless; and C should be either +1, 0 or -1
# Rate has units of per sec, and B has units of per sec per volt
# taumin and tauVdep have units of per sec
#
#units for membrane potential: volts
clrev=-60e-3
krev=-90e-3
narev=50e-3
carev=140e-3 #assumes CaExt=2 mM and CaIn=50e-3
ZpowCDI=2
VMIN = -120e-3
VMAX = 50e-3
VDIVS = 3401 #0.5 mV steps
#units for calcium concentration: mM
CAMIN = 0.01e-3 #10 nM
CAMAX = 40e-3 #40 uM, might want to go up to 100 uM with spines
CADIVS = 4001 #10 nM steps
#mtau: Ogata fig 5, no qfactor accounted in mtau, 1.2 will improve spike shape
#activation minf fits Ogata 1990 figure 3C (which is cubed root)
#inactivation hinf fits Ogata 1990 figure 6B
#htau fits the main -50 through -10 slope of Ogata figure 9 (log tau), but a qfact of 2 is already taken into account.
qfactNaF = 2.5
Na_m_params = TauInfMinChannelParams(T_min=7.998875749153482e-05,
T_vdep=0.0016817636262595197,
T_vhalf=-0.05308849295680853,
T_vslope=0.008,
SS_min=0.0,
SS_vdep=1.0,
SS_vhalf=-0.01608849295680853,
SS_vslope=-0.01,
T_power=2)
Na_h_params = TauInfMinChannelParams(T_min=0.00027542296648548886,
T_vdep=0.0012001000718321953,
T_vhalf=-0.034845326172449564,
T_vslope=0.003,
SS_min=0.0,
SS_vdep=1.0,
SS_vhalf=-0.05284532617244956,
SS_vslope=0.006,
T_power=1)
NaFparam = ChannelSettings(Xpow=3, Ypow=1, Zpow=0, Erev=narev, name='NaF')
#This is from Migliore.
'''
KDrparam = ChannelSettings(Xpow=1, Ypow=0, Zpow=0, Erev=krev, name='KDr')
KDr_X_params = AlphaBetaChannelParams(A_rate = 28.2,
A_B = 0,
A_C = 0.0,
A_vhalf = 0,
A_vslope = -12.5e-3,
B_rate = 6.78,
B_B = 0.0,
B_C = 0.0,
B_vhalf = 0.0,
B_vslope = 33.5e-3)
KDr_Y_params = []
'''
Krpparam = ChannelSettings(Xpow=2, Ypow=1, Zpow=0, Erev=krev, name='Krp')
#Act tuned to fit Nisenbaum 1996 fig6C (minf^2) and fig 8C (mtau)
qfactKrp=3 #Used by RE
Krp_X_params = AlphaBetaChannelParams(A_rate=93.17594708227949,
A_B=0.0,
A_C=0.0,
A_vhalf=-0.0023501473737281243,
A_vslope=-0.02,
B_rate=13.976392062341922,
B_B=0.0,
B_C=0.0,
B_vhalf=-0.0023501473737281243,
B_vslope=0.04)
# tuned to fit Nisenbaum 1996 fig 9D (hinf, 87% inactivating) and 9B (htau)
Krp_Y_params = TauInfMinChannelParams(T_min=1.6771670474810307,
T_vdep=24.22574624139267,
T_vhalf=-0.044350147373728124,
T_vslope=0.013000000000000001,
SS_min=0.13,
SS_vdep=0.87,
SS_vhalf=-0.05835014737372812,
SS_vslope=0.015,
T_power=1)
Kirparam = ChannelSettings(Xpow=1, Ypow=0, Zpow=0, Erev=krev, name='Kir')
qfactKir = 1
Kir_X_params = AlphaBetaChannelParams(A_rate=0.008,
A_B=0,
A_C=0.0,
A_vhalf=0,
A_vslope=0.011,
B_rate=1000,
B_B=0.0,
B_C=1.0,
B_vhalf=-0.04,
B_vslope=-0.04)
KaFparam = ChannelSettings(Xpow=2, Ypow=1, Zpow=0, Erev=krev, name='KaF')
# activation constants for alphas and betas (obtained by
# matching m2 to Tkatch et al., 2000 Figs 2c, and mtau to fig 2b)
qfactKaF = 2
KaF_X_params = AlphaBetaChannelParams(A_rate=7078.191955696018,
A_B=0.0,
A_C=1.0,
A_vhalf=0.019718735644534503,
A_vslope=-0.013,
B_rate=1769.5479889240046,
B_B=0.0,
B_C=1.0,
B_vhalf=-0.0002812643554654941,
B_vslope=0.011)
#inactivation consts for alphas and betas obtained by matching Tkatch et al., 2000 Fig 3b,
#and tau voltage dependence consistent with their value for V=0 in fig 3c.
#slowing down inact improves spike shape tremendously
KaF_Y_params = AlphaBetaChannelParams(A_rate=383.3791829132746,
A_B=0.0,
A_C=1.0,
A_vhalf=0.13455520966793152,
A_vslope=0.022,
B_rate=237.32997037488425,
B_B=0.0,
B_C=1.0,
B_vhalf=0.06855520966793152,
B_vslope=-0.011)
#KaS based on Shen 2004 data and Wolf 2005 model code. Note that the Wolf model
#code on ModelDB notes that, via personal correspondance with Shen 2004 author,
#parameters in Shen 2004 were misreported, and are subsequently corrected in
#the Wolf 2005 code. Alpha/Beta channel params were fit to be similar to the
#Wolf steady state and tau equations--see fitKaS.py for fitting script.
#Note: Fit looks ok with these parameters but could stand to be improved.
KaSparam = ChannelSettings(Xpow=2, Ypow=1, Zpow=0, Erev=krev, name='KaS')
qfactKaS = 3 # Shen 2004/Wolf 2005
KaS_X_params = AlphaBetaChannelParams(A_rate=131874.00929401233,
A_B=0.0,
A_C=1,
A_vhalf=-0.10285571937267687,
A_vslope=-0.0162951634,
B_rate=2080740907.394687,
B_B=0.0,
B_C=1,
B_vhalf=0.3832628144273231,
B_vslope=0.0218235302)
KaS_Y_params = AlphaBetaChannelParams(A_rate=111685872.76089874,
A_B=0.0,
A_C=1.0,
A_vhalf=1.225808650743561,
A_vslope=0.0645391447,
B_rate=5.615951122169981,
B_B=0.0,
B_C=1.0,
B_vhalf=0.004444253545561,
B_vslope=-0.0262013787)
#SS values from Churchill and MacVicar, assuming Xpow = 1
##time constants extrapolated from scarce measurements - Song & Surmeier
#SS values measured by Kasai and Neher are quite similar, except
#they use Xpow=2 to fit, thus params would be different
#they have nice time constant measurements which are ~2x slower than above
#but they measured at room temp, so qfact=1 on S&S and qfact=2 on K&N would equate them
#Vdep inact by Kasai ~10% and very slow (>50 ms). For now, have no Vdep inact.
# CDI measured by Kasai
#Note that CaL13 for D1 has mvhalf 10 mV more negative than for D2
#CaL12 does not differ between D1 and D2.
CaL12param = ChannelSettings(Xpow=1, Ypow=1, Zpow=ZpowCDI, Erev=carev, name='CaL12')
qfactCaL = 2
CaL12_X_params = AlphaBetaChannelParams(A_rate=-1760.132,
A_B=-440000.0,
A_C=-1.0,
A_vhalf=0.0040003,
A_vslope=-0.008,
B_rate=-568.0426,
B_B=142000.0,
B_C=-1.0,
B_vhalf=-0.0040003,
B_vslope=0.005)
CaL12_Y_params = TauInfMinChannelParams(T_min=0.02215,
T_vdep=0,
T_vhalf=0.0040003,
T_vslope=-0.0075,
SS_min=0.83,
SS_vdep=0.17,
SS_vhalf=-0.055,
SS_vslope=0.008,
T_power=1)
# Using Xpow=1 produced too high a basal calcium,
# so used Xpow=2 and retuned params - much better basal calcium
CaL13param = ChannelSettings(Xpow=1, Ypow=1, Zpow=ZpowCDI, Erev=carev, name='CaL13')
CaL13_X_params = AlphaBetaChannelParams(A_rate=3000,
A_B=0,
A_C=1,
A_vhalf=-0.005,
A_vslope=-0.018,
B_rate=4000,
B_B=0,
B_C=1.0,
B_vhalf=0.052,
B_vslope=0.008)
CaL13_Y_params = TauInfMinChannelParams(T_min=0.02215,
T_vdep=0,
T_vhalf=0.037,
T_vslope=0.005,
SS_min=0,
SS_vdep=1,
SS_vhalf=-0.037,
SS_vslope=0.005,
T_power=1)
#Params from McRory J Biol Chem, alpha1I subunit
CaTparam = ChannelSettings(Xpow=3, Ypow=1, Zpow=0, Erev=carev, name='CaT')
qfactCaT = 2
CaT_X_params = AlphaBetaChannelParams(A_rate=2000,
A_B=0.0,
A_C=0.0,
A_vhalf=0.0,
A_vslope=-0.019,
B_rate=2673.0099,
B_B=33000,
B_C=-1.0,
B_vhalf=0.0810003,
B_vslope=0.00712)
#Original inactivation ws too slow compared to activation, made closder the alpha1G
CaT_Y_params = AlphaBetaChannelParams(A_rate=7684.020399999999,
A_B=68000,
A_C=-1.0,
A_vhalf=0.1130003,
A_vslope=0.00512,
B_rate=640,
B_B=0,
B_C=0.0,
B_vhalf=0.0,
B_vslope=-0.017)
# CaN SS parameters tuned so m2 fits Bargas and Surmeier 1994 boltzmann curve
# CaN tau from kasai 1992.
# Kasai measures calcium dependent inactivation
#McNaughton has act and inact, tau and ss for human CaN
CaNparam = ChannelSettings(Xpow=2, Ypow=1, Zpow=ZpowCDI, Erev=carev, name='CaN')
qfactCaN = 2
CaN_X_params = AlphaBetaChannelParams(A_rate=608,
A_B=0,
A_C=0.0,
A_vhalf=0.0,
A_vslope=-0.014,
B_rate=1499.5231680000002,
B_B=105600,
B_C=-1.0,
B_vhalf=0.01420003,
B_vslope=0.01)
CaN_Y_params = TauInfMinChannelParams(T_min=0.035,
T_vdep=0,
T_vhalf=0.0,
T_vslope=-0.014,
SS_min=0.79,
SS_vdep=0.21,
SS_vhalf=-0.0748,
SS_vslope=0.0065,
T_power=1)
# CaR SS (Act and Inact) parameters from Foerhing et al., 2000
# Was Xpow=3 taken into account during fit?
# CaR tau from a few measurements from pyramidal neurons by Foerhing
# CaR inact tau from Brevi 2001
#Inact params are a bit too steep for ss, and not steep enough for tau
CaRparam = ChannelSettings(Xpow=3, Ypow=1, Zpow=ZpowCDI, Erev=carev, name='CaR')
qfactCaR = 2
CaR_X_params = AlphaBetaChannelParams(A_rate=480,
A_B=0,
A_C=0.0,
A_vhalf=0.0,
A_vslope=-0.028,
B_rate=2528000.0,
B_B=16000000.0,
B_C=-1.0,
B_vhalf=0.158,
B_vslope=0.0136)
CaR_Y_params = AlphaBetaChannelParams(A_rate=1100.0,
A_B=10000,
A_C=-1.0,
A_vhalf=0.11,
A_vslope=0.017,
B_rate=20,
B_B=0,
B_C=0.0,
B_vhalf=0.0,
B_vslope=-0.03)
#Reference: Maylie Bond Herson Lee Adelman 2004, Fig 2 steady state
#Fast component has tau~4 ms; not used: slow tau = 70 ms
#Fast component, tau=4.9ms from Hirschberg et al., 1998 figure 13.
SKparam = ChannelSettings(Xpow=0, Ypow=0, Zpow=1, Erev=krev, name='SKCa')
SK_Z_params = ZChannelParams(Kd=0.00057,
power=5.2,
tau=0.0049,
taumax=0,
tau_power=0,
cahalf=0)
#Reference: Berkefeld et al. Science 2006 314(5799):615-20. Moczydlowski and Latorre 1983, J. Gen. Physiol. 82:511-542.
BKparam = ChannelSettings(Xpow=1, Ypow=0, Zpow=0, Erev=krev, name='BKCa')
BK_X_params=[BKChannelParams(alphabeta=480, K=0.18, delta=-0.84),
BKChannelParams(alphabeta=280, K=0.011, delta=-1.0)]
#These CDI params can be used with every channel, make ZpowCDI=2
#If ZpowCDI=0 the CDI will not be used, power=-4 is to transform
#(Ca/Kd)^pow/(1+(Ca/Kd)^pow) to 1/(1+(ca/Kd)^-pow)
CDI_Z_params = ZChannelParams(Kd=0.00012,
power=-4,
tau=0.142,
taumax=0,
tau_power=0,
cahalf=0)
# Calcium-activated chloride channel params
# Reference: Pifferi, S., Dibattista, M., & Menini, A. (2009). TMEM16B induces chloride
# currents activated by calcium in mammalian cells. Pflügers Archiv-European
# Journal of Physiology, 458(6), 1023-1038.
CaCCparam = ChannelSettings(Xpow=0, Ypow=0, Zpow=1, Erev=clrev, name='CaCC')
CaCC_Z_params = ZChannelParams(Kd=0.00183,
power=2.3,
tau=0.013,
taumax=0,
tau_power=0,
cahalf=0)
#Dictionary of "standard" channels, to create channels using a loop
#NaF doesn't fit since it uses different prototype form
#will need separate dictionary for BK
Channels = NamedDict(
'Channels',
Krp = TypicalOneD(Krpparam, Krp_X_params,Krp_Y_params),
KaF = TypicalOneD(KaFparam, KaF_X_params, KaF_Y_params),
KaS = TypicalOneD(KaSparam, KaS_X_params, KaS_Y_params),
Kir = TypicalOneD(Kirparam, Kir_X_params, []),
CaL12 = TypicalOneD(CaL12param,CaL12_X_params, CaL12_Y_params, CDI_Z_params, calciumPermeable=True),
CaL13 = TypicalOneD(CaL13param, CaL13_X_params,CaL13_Y_params, CDI_Z_params, calciumPermeable=True),
CaN = TypicalOneD(CaNparam, CaN_X_params, CaN_Y_params, CDI_Z_params, calciumPermeable=True),
CaR = TypicalOneD(CaRparam, CaR_X_params, CaR_Y_params, CDI_Z_params, calciumPermeable=True),
CaT = TypicalOneD(CaTparam,CaT_X_params, CaT_Y_params, [], calciumPermeable=True),
SKCa = TypicalOneD(SKparam, [], [], SK_Z_params, calciumDependent=True),
NaF = TypicalOneD(NaFparam, Na_m_params, Na_h_params),
BKCa = TwoD(BKparam, BK_X_params, calciumDependent=True),
CaCC = TypicalOneD(CaCCparam, [], [], CaCC_Z_params, calciumDependent=True),
)