Biological

Cybernetics

© Spnnger-Verlag 1986

A Spike Generator Mechanism Model

Simulates Utricular Afferents Response to Sinusoidal Vibrations

R. W. Budelli, E. Soto, M. T. Gonzalez-Estrada, and O. Macadar

Departamento de Ciencias Fisiologicas, lnstituto de Ciencias de la Universidad Autónoma de Puebla, Puebla,

Pue., México and Instituto de Investigaciones Biológicas, Montevideo, Uruguay.

Abstract. Using a model of spike generator mechanism (SGM) with a variable threshold we simulate the

responses of articular afferents to sinusoidal vibrations. It reproduces the phase locking characteristics

(bifurcations diagrams) and the stimulus frequency firing rate relationships of different types of utrieular

affereuts. We estimate the model parameters selecting the values which best fit the experimental results and

we compare them with those from basic mechanisms involved in articular codification.

INTRODUCTION

In a previous paper (Budelli and Macadar 1979) we studied the response of articular afferents to sinusoidal

vibrations. We classified these fibers in three types. Type I afferents fire spontaneously, respond to tilts in a

phasic-tonic way and follow stimulation frequency, increasing or decreasing their firing rate when vibrated

sinusoidally. Type II afferents differ from type I in relation to response to vibrations. They always increase

the firing rate when vibrated, whatever the stimulation frequency. Type 111 afferents do not fire

spontaneously. Our interest is to understand how the differences between type I and II afferents are

originated.

Several mechanisms produce utricular codification (Aidley 7971; Davis 1961): stimulus modification,

transducdon, afferent synapses and spike generation. All of them may be involved in the generation of

response characteristics from the different types of utricular afferents. In this paper we study the capacity of

the spike generation mechanism ISOM) to produce these characteristics and, furthermore, the differences

between type I and II fibers. Using a model of SGM with variable threshold (Perkel et al. 1981; Torras 1985)

which is in good agreement with the classic theory of spike generation (Mounteaettle 1980; Aidley 1971;

Adrian and Zottermau 1926; Erlanger and Gasser 1937), we simulate the responses of articular afferents to

sinusoidal vibrations. With different values of the model's parameters we may reproduce the behavior of the

different types of afferents. The values of these parameters (compared with those experimentally determined)

may indicate which process is or is not involved in the physiological basis of repetitive firing of utricular

afferents.

The Model

Firstly, we assume that the place where spikes are generated is a well defined region of the afferent fiber, on

which synaptic potentials exert a direct influence. We also assume that the membrane potential and the

threshold is uniform in this region. These conditions define a unieompaetmeutal model (Perkel et al. 1981). It

also includes a relative refractory period (RRP) following each spike (Fig. 1) modelled through a time

dependent increase of the threshold (T in Fig. 1):

Fig. 1. The model The membranepotcntial V Is a function of the time t, The threshold T increases after each spike (S).

The afferent will produce a spike when the membrane potential (V) reaches the threshold (T). The decay of the threshold

to its asymptotic potential (T

0

) follows an exponential function

where To is the steady state threshold and f is a decreasing function with limit 0 when t tends to infinit. In our

simulations we will use:

f (t)=b exp(-(t-t

0

)τ),

where to is the time of the last spike and 6 is reset after each spike to the actual value of fit) plus a fixed

amount (B). These last assumptions are conventional; 1201 we shall use them for two reasons. On the ono

hand, computations are asier with exponential functions, sincef(t+dt)=f(t) exp(dt) and we may calculate ,fat

fixed intervals multiplying it by the fixed mount exp(dt) without using exponentials each time. On the other

hand, under these assumptions, the model is equivalent, regardless of a linear filter, to the leaky integrator

(Seharstein 1979; Budelli 1985). We may therefore interpret the results in relation to the biophysical basis of

this model.

The membrane potential of the region where the spikes are generated is a function of the synaptic

activity which roduces the generator potential. Since the synaptic activity depends upon the stimulus, the

membrane potential of the Region where the spikes generate is a function of the stimulus. As a first

approach, in order to compare the model with the experimental results, we will suppose that the relation

ship between stimulus and generator potential is linear. If we assume that the filter transforming the stimulus

in the generator potential is a leaky in tegrator, with the same time constant of the evolution ofthe threshold,

the model reduces itselfto the classical leaky integration and fire model (Scharstein 1979; Budelli 1985). We

will work with the relationship between the generator potential and the spikes, and thus we will not deal with

the transfer function between the stimulus and the generator potential. Methods: We simulate the behavior of

the model for different parameters with standard computational methods (Perfect et al. 1981) using a Digital

computer PDP 11/34.

RESULTS

There are two types of spontaneously firing utricular afferente with different responses to sinusoidal vibra

tions (Budelli and Macadar 1979). When type I afferents are stimulated with frequencies close to their

spontaneous firing rate, the s s spikes always discharge close to a fixed phase of the stimulus (phase lacking)

(Fig. 2 left); the cycle histogram presents a sharp peak and the fiber may increase or decrease its firing rate

as a function of the stimulus frequency. Nevertheless, hen the stimulus value is increased

above a critical value, the phase locking paradoxically disappears and the cycle histogram becomes flatter

Fig. 2. Response of typosl and it utncular aRCrcnts. The left column shows results from a type I afferent and the right

one, from a type 11. The first row shows raw records of responses to emasondal annotation: type l ancient phase-locks

and type 11 does not. The second row shows the corresponding cycle histograms Thethnd row

showsoyclehnstogramscorrespondmg to a larger stimulus: the type 1 afferent loses the phase locking

(Fig. 2 left, bottom). Type IT afferents do not show phase locking (Fig. 2 right), the cycle histogram does not

present sharp peaks and the firing rate can not be reduced below its spontaneous rate.

Some results from type 1 afferents can be reproduced by the leaky integrator model (Rescigno et al.

1970; Stein et al. 1972; Barbi et al. 1975; Poggio and Tome 1977; Ascoliet al. 1977; Rescigno 1978;

Angelini and Petracchi 1978; Angelina et al. 7980; Zeevi and Bruekstein 1981). Some aspects of the behavior

of type 11 afferents can be reproduced by the perfect integrator model (Bayly 1968; Knight 7972; Stein et al.

1972). Nevertheless, this model produces meaningless results. For example: a negative stimulus will produce

a fixed delay of the next spike quite independently of the moment in which it was released. In order to present

a more convincing model we have to be able to simulate bath types of afferents with a single model, selecting

different numerical values for the parameters. Furthermore, we are attempting to estimate the model

parameters, looking for those that fit the best to the experimental results. This estimation will provide

information related to the physical baste of the spike generation mechanism.

Phase Locking

We determined the parameters of the stimulus (amplitude, A, and frequency, f) under which phase locking

occurs, for different time constants (r) of the function

f(t)=K'exp(-(t-t

o

)/τ),

where to is the time of firing of the last spike and

K=B- Σ

i

exp(-(t

o

-ti)/ τ) i=∞,…,0,

where t, are the firing times of the previous spikes and B is the increase of the threshold following each spike.

We usually understand by phase-locking the stable firing of a spike in a fixed phase of each cycle of the

stimulus. This concept was extended to the case where n spikes fire in fixed phases of m cycles, and we say

there is a (n, m) phase locking.

The bifurcation diagrams (Fig. 3) show the regions of the stimulus parameters (A, f) where some simple m,

in) phase lockings are produced for different rates (τ/P) between the model time constant (r) and the

spontaneous firing period (P). The stimulus amplitude is expressed as a percentage of the asymptotic value of

the threshold (To). The stimulus frequency is expressed as a percentage of the firing rate. In general, the

regions of phase locking do not overlap, corresponding to "Arnold's tongues", which in turn correspond to

the results predicted when using small stimulus for a wide range of firing models (Arnold 1965). This figure

also shows that for small time constants (r'P=0.1) the phase lockingregions are narrow and that they widen

ass, P increases. For intermediate values of the time constant (τ/P=1) the regions widen at small amplitudes

of the stimulus, but they narrow at larger amplitudes. With further increases of the time constant (τ /P>3) the

regions vanish.

These results show that the model reproduces the behavior of both types of afferents. Obviously, it

simulates the behavior of type I fibers, since our model is equivalent to the leaky integrator. For i,P=1. it also

simulates the lost of (1,1) phase locking for a stimulus frequency close to the firing rate when the stimulus

amplitude increases above a critical value (Fig. 2, Fi&3 from Budelli and Macadar 1979; Ange

Fig. 3. Bifurcation diagrams, The figure shows for xT equal l0 10, 3.1.03, and 01 the bdurcenon diagrams Them arc

several types of phase-locking on, am where n os the number of,p,k,, produced in fixed phases of or cycles

Fig.4. Minimum frequency obtained with u, 1)phase-locking as a function of TIP

lini and Petracchi 1978). For TIP > 3 it also reproduces the results of type II afferents. Since the regions of

phase locking in these cases are very narrow, the probability of selecting the stimulus parameters in the (1,1)

phase lacking region is very low. Furthermore, when in these units phase locking is obtained (Fig. 3 from

Macadar and Budelli 1984) it is so for frequencies above the spontaneous firing rate.

Figure 4 shows the minimum frequency with (1,1) phase locking as a function of τ/P. This figure allows

an estimation of τ/P. If, for example, the minimum frequency allowing (1,1) phase locking is 70 %, we may

estimate 0.3 < τ/P < 1. Since, as we have reported, most of type I fibers can not phase lock with frequencies

below 70% of their spontaneous firing (Budelli and Macadar 1979), we estimate TIP > 0.3.

The model simulates the spontaneous firing of utricular afierents (types I and II). With an adequate

selection of parameters we were also able to simulate the phase lacking of type I fibers and its absence in

type II afferents. Quantitative changes in the model parameters may produce qualitative changes in its

behavior. Furthermore, we showed that for many pairs of integers the model produces (m, n) phase locking.

Although this result was not reported in our experimental papers, it may probably accounts for some

multimodal cycle histograms (Budelli and Macadar 1979).

Stimulus Frequency - Firing Rate Relationship

When type I and II afierents were defined, one of the criteria we considered was the capacity of type I fibers

to increase or decrease their rate as a function of the stimulus frequency. On the other hand, type II afferents

always increase their firing rate whatever the stimulation frequency. Since for τ/P<1 the model produces a

(1,7) phase locking region reaching fre quency values below the firing rate, we can be certain that for these

values of τ/P it reproduces this behavior of type I aBerents. Nevertheless, we do not know what happens for

larger values of τ/P.

Figure 5 shows this relationship for TIP -10 and far several amplitudes of the stimulus. The broken

straight line passing through (1, 1) corresponds to points where the (1,1) phase locking may be produced.

Since there are no simulation points on this line, we confirmed that the (7,1) phase locking region is very

narrow. Nevertheless, the curves change their slope near this line, tending to decrease the angle of

intersection. In this case all the stimuli increased the firing rate of the model proportionally to their

amplitude.

Figure 6 shows the results for τ/P=3. In this case, there is no proportional relationship between the

increase in firing rate and the stimulus amplitude. The curves follow partially the phase locking lines. This

tendency is notorious for the (1,1) phase locking, but it is also clear far the (1, 2) and (2,1) ones. There are

also several points in the lines corresponding to phase lockings (3,1), (3, 2), and (2, 3).

Figure 7 shows the same kind of curves for τ/P=1. Here, the tendency of the simulation points to appear

in the phase locking lines is even greater. Besides the (1,1), (1, z), and (z,1) phase soakings, (z, 3), (3, z),

(3,1), and even (3,4) and (3, 5) became notorious. Moreover, this is the largest value of τ/P producing a clear

decrease of the firing rate for some frequencies (30, 90, and 190) and stimulus amplitudes.

The curves for τ/P=0.3 (Fig. 8) also have many points in the phase locking lines. For this time constant it is

possible to produce reductions in the firing rate of up to 30% for the (1,1) phase locking up to 20% for (2,1)

and up to 10 % for (1,2)

Fig. 5. Firing rate against stimulus frequency for

τ

/P=10. It shows this relationship for different amplitudes of the

stimulus, shown on each curve in relation to the difference between the asymptotic values of the membrane potential and

the threshold

Fig. 6. Firing rate against stimulus frequency for τ/P=3.

Fig. 7. Firing rate against stimulus frequency for τ/P=1.

Fig. 8. Firing rate against stimulus frequency for τ/P=0.1.

Fig. 9. Firing rate against stimulus frequency for τ/P=0.3.

or τ/P=0.1 (Fig. 9) the (1,1), (1, 2), and (2, 3) phase locking curves are notorious. 1, a smaller degree

ma

timulus frequency relationships are similar to those described for

typ

iscussion

imulation experiments show that the SGM may reproduce several characteristics of the oracular afferenls

he stimulus frequency firing rate curves show that a sinusoidal stimulus may increase (or decrease) the rate

decreases in the rate may be produced by stimuli which have negative and positive phases.

F

ny points appear in the (2, 1) and (8, 5) lines. This result shows that it is easier to obtain phase locking by

decreasing the Firing rate than by increasing it This result does not coincide with our experimental results,

thus indicating τ/P must be larger than 0.1.

It is worth noting that the firing rate - s

e I afierents if i~P=0.3 and to type H ones if τ/P=3. Results for τ/P=1 probably correspond to intermediate

results mentioned in our experimental papers (Budelli and Macadar 1979, 1981; Macadar and Budelli 1984.).

D

S

response to sinusoidal stimulation. Furthermore, an estimation of the model parameters is possible, selecting

the values τ/P which best fit the experimental results With τ/P=1 the model simulates type I aBerents and

with τ/P=3, type II. The fact that a model with different values of its parameters can reproduce the behavior

of both types of afierents is a strong argument on its behalf, since there is no need to introduce different

processes for different types of fibers. Besides, it is worth noting that quantitative changes in the parameters

produce qualitative changes m the response.

T

of firing depending upon its frequency and the model parameters. Consequently, a change in the frequency

of discharge produced by a periodic stimulus does not indicate the presence of a maintained polarization at

the level where the spikes are produced. Perkel et al. (1964) have shown that an increase in frequency of a

hyperpolarizing stimulus may paradoxically increase the firing rate of the neuron. Nevertheless, all these

changes of rate correspond to values below the basal rate (Segundo 1979; Diez-Martinez et al. 1983). When

excitatory stimulation is used, similar results are obtained, but the mean effect is an increase of the rate

(Segundo and Kohn 1987; Diez-Martinez and Segundo 1983). In our model, we prove that both increases or

Moat of the neuronal models show phase-locking. Au exception is the integrate and fire model (Knight

1972; Rescigno et al. 1970). Probably, this general behavior is a consequence of ge

neral topological

the

egulation of the repetitive firing of the vestibular

afie s, simulated in our model by an increase of the threshold. They may be related to the dynamics of the

Na`

hysical basis (the charge of the membrane capacitor), we prefer not to

e model parameters.

These estimated values can be compared with those experimentally determined for the different mechanisms

inv

ould like to thank Drs. O. DiceMarnncz, J A Brig, B Holmgrcm and R. Lara for

reading the manuscript. We also appreciate the artistic help of R. Bcraal and R. Cardozo and the

971) The Physiology of ...table calls. Cambridge University Press

ngelini F, Pelrucchl D (1978) Significance of phase-locking mennmemenls as a test for neural cneodlag

timulation. Biol Cybern 36:137-142

orems (Arnold 1965; Rescigno 1978). In this case, any model will share some general characteristics of

the phase-locking regions in the frequency-amplitude diagram (Glass and Mackey 1979; Glass and Perez

1982; Guevara et al. 7981; Grasman 1984). Then, in order to select between models, we must compare

mostly geometrical characteristics. Since our model shares this behavior, we have to compare the

bifurcations diagrams it produces with those experimentally determined. They are different in our model and

in those of Glass and Perez (1982) or Grasman (1984).

Several mechanisms could be involved in the r

rent

channel (Gunman et al. 1980), the Car' dependent K' channel (Meech and Standen 1975; Partridge

1982), the A-current channel Connor 1985), or the charge of the membrane capacitor (Hodgkin 1948; Stein

ct al. 1972)_ The determination of the time constants by the simulation of lhc experimental results can be

compared with the time constants of these processes in order to determine which one might be responsible

for the characteristics just examined.

Our model, in the conditions used m the snmulafions. behaves like the leaky integrator (Budelh 7985). Since

the leaky integrator model implies a p

use it in order to avoid the subsequent interpretation of the meaning of tire parameters: the time constant is

that of the membrane (resistance times capacity). Nevertheless, since both models behave in the same way,

we may check if our estimation of the time constant correspond to those experimentally determined in the

afferent fiber. Forukawa et al. (197R) recorded poet synaptic potentials from the nerve terminals from the

gold fish sacculus. The decay of these potentials establishes a superior boundary for the time constant of the

membrane of around 1 ms. On the other hand, we estimate τ/P between 0.3 and 3. Sine. the values of P are

usually larger than 20 ms, it must be r>6ms, reaching values even higher than 60 ms. These considerations

show that the values of r necessary to simulate the experimental results are always larger than the time

constant of the membrane. Consequently, we have to reject the biophysical basis of the leaky integrator (the

charge of the membrane capacitance) as the regulator mechanism of the repetitive discharge of the utricular

afferents. Then, we will have to look elsewhere to find the mechanism accountable for the regulation of

repetitive firing of utricular afferents. They can consist of changes in excitability related to the dynamical

behavior of the sodium current channel, or slow changes in the potassium permeability.

In this paper we show how the simulation results can be used for the estimation of th

olved in the process we are dealing with. This is a powerful method for testing alternative hypothesis

about the physical basis of any characteristic. Usually, the result can be the rejection of a hypothesis, since a

numerical coincidence between parameters estimated by a model and a basic process does not prove that the

latter is responsible for the general behavior of the system. In our case we rejected the hypothesis that the

repetitive firing of vestibular afferents is regulated by the charge of the membrane capacitor. Usually the

rejection of a hypothesis is disturbing. But, as Ironware (1902) argued "this feeling is noujustified. The

physicist who has just given up one of his hypothesis should, on the contrary, rejoyce; since by the same

token he has found an unexpected ocassion of discovery. Has the discarded hipothesis been fruitless? Far

from that, we might say, it has provided more benefit than a well stablished hypothesis. It has pointed to the

decisive experiment; and furthermore, if the experiment was the product of hazard without any supporting

hypothesis, nothing extraordinary would be detected; and we would simply catalogue another fact without

reaching any conclusion.”

Acknowledgements. We w

critically

editorial one of G Payas.

References.

Aidley DJ (1

A

models Biol Cybern 30:169 177

Angelini F, Barbs M, Chiller, S, Pctrawhi D (1980) Resonant response of a neural model and of Limulus,

ommatidae to double frequency s

Anted VI (1965) Small denominators 1. Mappings of the circunfcrence onto itself. Translation from Sere

A.M.S. 2, 46:213-284

Ascoh C, Barba M, Chiller, S, Petrarchan D (1977) Phase locked response in the Lamulus lateral eye

Theoretical and expcrlm

enlal investigation. Rini J 19.219-240

Bud 981) Dynamic properties from utrleutar alferenbs. In: Gundnerou T led) The

in

Con

Dav Some principles of sensory receptor action Phyaol Be, 41 391"16

citatory loop. Biol Cybem

ble valucdncss in crayfish stretch receptors not Cybern 49.3313

pses between hair cells and effluent nerve floes in goldfish. J Physro1276:211-226

Gra 84) The mathematical modeling of entrained biological oscillators. Bull Math Bel 46:40722

d irregular dynamics in

d(London)305:377 395

repetitive action in a non mychhnated cell.

Ma nisms ofsensory adaptation in the Isolated utnde. par, Neurol 86:147

onent mediated by calcium influx J Physiol (London) 249:211-239

oduced

Barbs M, Fanelli V, Frcdlam C, Pctracchn D (1975) The self inhibited leaky integrator transfer function and

steady state relations Emil Cybern 20:51-59

Buell E (1968) Spectral analysis of pulse frequency modulation in the ncrvoussystcm JEFF

TransBiomcdEngl5.257 265

Budelli R (1985) Codlfrsvon sensorial en el sistarm vestibular Ph, D. Thesis, Umveraidad Nacional

Autonoma de Mexico

Budelli R, Macadar O (1979) Sueloacouslic properties of utrr cddar aferem,J Neurophyslol 42:1479 1493

elli R, Macadar O (1

vesnbular system Function and morphology. Clrapd 23. IF 374-389. Springer, New York Berl

Heidelberg

nor lA(1985)Neural paecmokerssnh rhythmiaty. Annu Rev Physlol 47.17-28

is H (1961)

Dlcz-Martmcz O, Scgundo 1P (1983) Behavior ofa single neurun in a recurrent ex

47:331

Dlez-Mart,ncz O. Kohn AF, Segundo JP (1983) Pervasive locking, saturation. asymmetric rate sensitivity

and dou

Eflanger 1, Gasser HS (1937) Electrical signs ofnervous activity. University off ennsylvanla Press.

Philadelphia, Pa

Forukawa T, Hayashida Y, Matsum S (1978) Qmmad analysis of the size of excitatory post-synaptic

potentials at syna

Glass L, Mackey MC (1979) A simplemmel for phase locking of biological oscillators. 1 Math find

7.339-352

Glass L, Perez R (1982) Fine structure ofphase locking, Phys Rev Lett48:1772-1775

sman ,1 (19

Go,,,,, MRL. Glass G Shrier A (1981) Phase locking, period doubling bifurcations an

periodically stimulated cardiac cells. Science 214.1350.1353

Guttman R, Lewie S, Rinzh J (1980) Control ofrepebtvefinmg m

squid axon membranes as a model for a neural oscillator J Physi

Hodgkin AL (1948) The local electrical changes associated with

J Physiol (London) 107:165 181

Knight BW (1972) Dynamics of encoding in a population of neurons. J Gen Physic] 59:734-766

cudar O, Budelff RT984) Mecha

155

Meech R, Standen N (1975) Potassium activation in Helix aspersa neurons under voltage clamp: a

comp

Mounacastfle VP (1980) Medical Physiology. C V. Mosby, SL Louis Tatum. London

Partridge LD (1982) A mechanism for the control of low frequency repeenve firing. Cell Mol Neurobiol

2:33A5

Perked D, Schulman JH, Bullock TH, Moore GP, Segundo JP (1964) Pacemakers neurons: effect of

regularly spaced synaptic input. Science 145:61-63

Perkel D, MuOoncy B, Budelli R (1981) Quantitative methods for predicting neuronal behavior.

Neuroscrenee 6:823-837

Poggio T, Torte V (1977) A Volterra representation for some near.. models. Biol Cybern 27:113 124

Poincaré H (1902) La science Of fhypoth8se. Flammarion, Pans

Rescigno A (1978) On phase locking of pulse encoders. Bull Math hot 40.807-827

model for the discharge patterns pr

Rescigno A, Stein RB, Purple RL, Poppcle RE (1970) A neuronal

by cyclic inputs. Bull Math Biophys 32:337 353

Scharstetn H (1979) Input-output relationship of the leaky integrator neuron model J Math Bird 8:40320

Segundo JP (1979) Pacemaker synaptic interactions. modelling locking and paradoxical features. Blot

Cybern 35:55-62

Segundo JP, Kohn AF (1981) A model of excitatory synaptic rat eraelions between pacemakers. Its reality,

its generality, and the principles involved. Biol Cybcrn 40:113-126

Stem RB, French A, Holden A (1972) The frequency respon coherence and information capacity oftwo

neuronal models Brophys1 12:295-322

Torras C (1985) Pacemaker neuron model with plastic filing rate. tournament and learning ranges. Biol

Cybem 52.74-91

Zeevi YY, Bmckswin AM (1981) Adaptative neural encoder model with self inhibition threshold control.

Burl Cybcr, 70:79

-92

eceived: November 12, 1985

r Ruben Budelli

ept. Ciencias Fisológicas

ma de Puebla

R

D

D

Univesidad Autóno

Ap. Postal 406

Puebla, Pue

Mexico

## Comments 0

Log in to post a comment