Вы находитесь на странице: 1из 12

SOLUTIONS TO 580.

222 HOMEWORK #2, 2008 1) Write the two forms of the equation:
dx = (V )(1 x) (V )x dt and dx x (V ) x = dt x (V )

Rearrange the left one into the same form as the right:

dx = ( + )x dt x + = 1 + x (V ) = +

and

x =

1 +

2) Save the following m-file as "betah.m" in the HH_files directory. Then hh should run. This version is vectorized. function bh = betah(V) % BETAH(V) - HH beta function for h variable bh = 1./(exp(-(V+30)/10)+1); return The unvectorized version follows (they are different, note the .); if you don't believe it, try calling the following with a vector for V. It is worthwhile explaining this to the students function bh = betah(V) % BETAH(V) - HH beta function for h variable. bh = 1/(exp(-(V+30)/10)+1); return

Problem 5 The full HH equations are as follows:


dV 1 = [ Iext GK n 4 (V E K ) GNa m 3 h(V E Na ) Gleak (V E leak )] dt C dn n (V ) n = dt n (V ) dm m (V ) m = dt m (V ) dh h (V ) h = dt h (V )

where Iext is an externally applied current, V is the membrane potential, and n, m, and h are the HH gating variables. With the substitutions given in the problem, the equations become:
dV 1 3 = [ Iext GK n 4 (V E K ) GNa m (V )(0.95 1.1n)(V E Na ) Gleak (V E leak )] dt C dn n (V ) n = dt n (V )

There are now only two differential equations. Replacing m and h by a function of V and a function of n means that the differential equations for m and h are no longer necessary. The form of the dV/dt equation is the same as before, only now with the substitutions for m and h. Another way of thinking about the dm/dt and dh/dt equations is that they can still be solved as part of the simulation, but because their state variables m and h no longer appear in the other equations, they are irrelevant. Problem 6 There is no reason why the assumptions given above should be true. In fact, the system simulated with the second set of differential equations above is an approximation of the full HH equations, which shares many qualitative properties (although not detailed quantitative ones, as we will see). The purpose of this exercise is to produce a simpler set of equations that approximates the properties of the HH equations, but is easier to understand. To see how accurate the approximations are, plot the exact and approximate functions on the same axes during an action potential. To do this, first compute an action potential using hh as follows.
>> hh Current (uA/cm^2): 10 Starting at ?? ms: 0 . . and ending at ?? ms: 1 Max simulation time (ms): 20 Current I.C.s: V=-60; n=0.318; m=0.053; h=0.596. Enter new ICs, in order (1-4 as '-60.1 0.4 0.05 0.5'): ICs to be used: V=-60, n=0.318, m=0.053, h=0.596. Final state: V=-59.6999, n=0.312651, m=0.0544757, h=0.604715. Again, plot Vars, G-conductances, I-currents, Quit? q

This leaves the state variables (V, n, m, h) from the simulation in the matrix yv for times in the vector tim. Each column of yv is one of the state variables, in the order V, n, m, h. Note that the solution is not found at evenly spaced points in time (check the tim vector), because of the way that the differential-equation solvers work, so you should always plot the time axis from tim. To compare m with m(V), first compute m(V) at the time points. The easiest way to do this is to use the minfhh() function that was provided with the hh m-files as follows.
>> minfv = minfhh(yv(:,1));

If you have trouble parsing the Matlab shorthands involved in this code, take it apart as follows. yv(:,1) is the first column of the yv matrix, containing the values of membrane potential resulting from the simulation. m is a function of V only, so can be computed as minfhh(yv(:,1)). minfhh is written so that if its input is a vector, it returns an identically sized vector whose elements are minf() computed from the corresponding elements of the input. minfv is then the values of m at each of the time points during the solution. The one-line code above is equivalent to the following, in which all calculations are done on scalar variables:
nn = size(yv,1); minfv = zeros(nn,1); for jn=1:nn vv = yv(jn,1); minfv(jn) = minfhh(vv); end % How many time points? % Set aside storage for

% get the jnth membr. potential % compute m

Now if you didnt happen to notice that minfhh was in the set of files, you could have done the same calculation using the alphm() and betam() functions as follows:
>> minfv = alphm(yv(:,1))./(alphm(yv(:,1)) + betam(yv(:,1)));

Or you could have written your own code for m(V) from the HH equations supplied at the end of the problem set. All of these give exactly the same result. These m values can be compared to the actual m values resulting from the simulation, which are in yv(:,3).
>> plot(tim, yv(:,3)); hold on; plot(tim, minfv, g);

Similarly, to compare h with (0.95-1.1n),


>> plot(tim, yv(:,4)); hold on; plot(tim, 0.95-1.1*yv(:,2), g);

The comparisons are shown in the two plots below.

In each case, the actual state variable, from the simulation of the full four differential equation set, is shown in blue and the approximation, computed from another state variable, is shown in green. The approximations are not exact, but are close. This method is very important in the qualitative study of non-linear equations. It depends on the fact that the four differential equations have rather different time scales. The m equation is quite fast, with a time constant around 1 ms, much shorter than the 10 ms time scale of the other variables. Similarly, n and h have almost the same time constant, as is evident from the plot of the state variables during an action potential, either in the course notes or the problem statement. Because m is fast, it is a reasonable approximation to say that mm(V). We know that the dm/dt equation causes m to track m, all we are assuming is that the tracking is so fast, that the approximation mm(V) is a good one. Because n and h have about the same time scale, it is reasonable to approximate one of them as a function of the other. The linear approximation used is an arbitrary curve fit to the data from the full model (plot n versus h in a simulation of the AP from the full HH set). More details about this method can be found in Kepler TB, Abbott LF, Marder E. Reduction of conductance-based neuron models. Biol Cybern 66:381-7 (1992). Problem 7 Action potentials can be computed from the two models using the usual command sequence given at the start of Problem 6. The results are plotted below, where both action potentials are initiated by a 1 ms long 10 A/cm2 current pulse. The blue plot is the AP produced by the full model, the green plot is the AP from the reduced model. One way in which the models differ is their current thresholds. As a result, the reduced model is farther above threshold at 10 mA than is the full model. That difference causes the shorter latency of the green AP (the threshold of the reduced system is smaller). The dotted green AP is the AP of the reduced model delayed by enough to make the peaks occur at the same time, to allow better comparison of AP shapes. There are two major differences, other than those associated with the threshold shift: 1) the AP amplitude is larger in the reduced model; and 2) the rise and fall time are faster in the reduced model. The reason for the differences can be inferred from an examination of the HH variables in the two models. These are plotted superimposed in

Comparison of action potentials produced by the full (blue) and reduced (green) models.

Comparison of HH variables during the AP for the full (solid lines) and reduced (dashed lines) HH models

the second plot. The HH variables from the full model are shown as solid lines and those from the reduced model as dashed lines (actually, these are plots of n, m(V), and (0.951.1n for the reduced model). The values in the reduced model are delayed by the same amount as for the AP waveform. As expected from the results of Problem 6, m rises and falls faster than m, which largely accounts for the faster rise and fall of V in the reduced model. In addition, m(V) reaches a larger peak, which is probably the major factor in the larger peak size of the AP in the reduced model. n and h both vary slightly less from their resting values in the reduced model, but it is not clear that there is any strong effect of those differences. Problem 8 The phase plane is plotted below (plot(yv(:,1), yv(:,2))). This plot allows a different look at the action potential, as described in the next paragraph.

The small plots at right show the trajectories of V and n during an action potential. The large plot at left is a phase plane plot of n (ordinate) versus V (abscissa). The colored dots identify corresponding points in the two sets of plots. The arrows show the direction of the trajectory in the phase plane. Note that time is not explicitly represented in the phase plane and distances in the phase plane bear no correspondence to equal increments in time (e.g. compare the distance from black to orange with the distance from orange to green). The green dot is at the resting potential. During the initial phase of the action potential (green to red), V increases very rapidly under the effect of m (V) with little change in n. During the next phase (red to black), n increases driven by the increase in n(V) produced by the depolarization. As n increases, V undergoes a corresponding decrease. Between the black and orange dots, m and V decrease rapidly. Finally, the afterhyperpolarization (orange to green) occurs as n returns slowly to its resting value. V increases slightly during this time. The different time constants of the two differential equations in the reduced HH model are clearly shown in these plots.

Problem 9 The nullclines are the points where dV/dt=0 and dn/dt=0. For the dn/dt equation, things are straightforward:

dn n (V ) n = =0 dt n (V ) n iso = n (Viso )
Thus, dn/dt=0 on a simple plot of n (V) in the phase plane. The nullcline for dV/dt=0 is more complicated:

dV 3 = 0 = Iext GK n 4 (V E K ) GNa m (V )(0.95 1.1n)(V E Na ) Gleak (V E leak ) dt


3 Iext 0.95GNa m (Viso )(Viso E Na ) Gleak (Viso E leak )

4 3 GK n iso (Viso E K ) 1.1GNa m (Viso ) n iso (Viso E Na ) =

This equation cannot be simplified further, but still describes a set of points in the (V, n) plane where dV/dt=0. A Matlab function (Visohhred()) was provided to compute this set of points for you, but it is worth looking at the behavior of the equations a little to convince youself that we did not make a horrible mistake in writing this function. The plot at right shows dV/dt plotted versus n at 12 different values of V (and Iext=0). Clearly over the range of V and n important in this problem, dV/dt is zero for only one n value at each value of V (the intercepts of the plots with the dashed axis at dV/dt=0). The function Visohhred() finds these zeros, given a set of V values and a fixed external current.

Problem 10 The requested phase-plane plot is at right. The blue trajectory is the phase portrait of the action potential, i.e. a plot of n versus V for the signals during an action potential, the same curve you plotted in Problem 8. The magenta curve is the nullcline for dn/dt=0; it is just a plot of n(V) versus V, as described above. The green plot is the nullcline for dV/dt=0. It was computed from the function provided. Assuming that you already have the basic phase-plane on screen (resulting from Problem 8 above), the nullclines are added with the following steps: >> vv = [-80:1:60]'; % Compute a range of voltages >> niso1 = ninfhh(vv); % The dn/dt=0 nullcline >> hold on; plot(vv, niso1, 'm') % Plot it on the existing phase plane >> niso2 = Visohhred(vv, 0); %Compute the dV/dt=0 nullcline >> plot(vv, niso2, 'g') % and plot it Now this plot provides further insight into the HH equations. First, notice that there is one spot where the two nullclines intersect, marked by the red dot. You will have to look a the graph carefully to see that there is not another intersection near V=-35 mV and n=0.68. At this point, all the time derivatives in the system are zero, so the system should be stable at this point, if placed there. This point is called an equilibrium point in mathematics and the resting potential in biology. The direction of trajectories in the phase plane can be inferred by looking at the action potential trajectory in the plots above or by computing a few values of dn/dt and dV/dt. Another way is to compute a few trajectories starting at various places in the phase plane. These are shown in the next figure (done by starting hhred at various initial values, shown by the Xs, with zero applied current). From this plot, we can infer that the system moves to the right in the phase plane below the green nullcline and to the left above it. Correspondingly, computation of dV/dt shows that dV/dt is positive below the green nullcline and negative above it. Similarly, the system moves

downward when it is to the left of the magenta nullcline and upward when it is to the right. Correspondingly, dn/dt is negative to the left and positive to the right of the magenta nullcline. The movements along the n direction are sometimes hard to see because dV/dt is quite a bit larger than dn/dt for this system, so the system tends to move horizontally except when it is near the dV/dt nullcline. However, the behavior of dn/dt can be seen clearly when trajectories are close to the green nullcline. Note that a qualitative idea of the behavior of the system could be gotten by plotting arrows in the phase plane on a dense grid of points. Often it is possible to figure out approximately how nonlinear systems will behave from such a grid of arrows. However, they can be misleading, as for the trajectory starting at (70, 0.3) which bypasses the resting point, even though nearby trajectories end up at rest. Notice that the action potential trajectories cross the nullclines moving horizontally or vertically. The reason for this is that at the dV/dt nullcline (green), dV/dt=0, so the system must be moving vertically when it crosses the green nullcline. Similarly, at the dn/dt nullcline, dn/dt=0, so the system must be moving horizontally. (Sorry, there was a misstatement in the problem statement about this!). Problem 11 The result of running the model with a current of 50 A/cm2 is that a steady train of action potentials is seen, as in the figure below. The spike train will continue as long as the current is applied. To understand why the model now spikes continuously, consider the phase plane in the second plot below. The blue curve shows the trajectory of the action potentials. They form a continuous loop, with one turn around the loop for each action potential in the train. The isoclines are also shown. Adding the current does not affect the dn/dt isocline (magenta), but it shifts the dV/dt isocline vertically. There is still an equilibrium point, but it is not in the usual resting place (near 60 mV). In fact the initial condition used to produce this train of action potentials (V=30.598, n=0.7236) is very close to the new equilibrium point. Apparently, the system is no long stable near its equilibrium point. Because there is only one equilibrium point in the phase plane,

there is no steady potential at which the system can sit, so it ends up in the steady spike train observed. This sort of steady oscillation is a characteristic of many non-linear systems, like membrane potentials, and is called a l i m i t cycle. It seems counterintuitive that the system has an equilibrium point where all its time derivatives are zero but the system is not stable. This is a property frequently observed in nonlinear systems. Behavior near an equilibrium point like this can be predicted, but that theory is beyond the range of these homework problems (see the reference in the problem set for enlightenment). Another way to put it is, never turn your back on a non-linear system.

Вам также может понравиться