notion that plasmodesmata of guard cells ibid. 63, 417 (1967); A. W.

417 (1967); A. W. Robards, Nature where t is time, (G,,} is a complete set

are missing or of little importance. Cyto- 218, 784 (1968); R. J. Hedler and J. Boerma,
Acta Bot. Neerl. 18, 99 (1969). of orthogonal functionals, and {ha} is
logically, the plasmodesmata appear 5. B. R. Voeller, in The Cell, J. Brachet and the set of "Wiener kernels" of S. The
A. E. Mirsky, Eds. (Academic Press, New
functional and probably play a major York, 1964), pp. 257-267. set {h.) completely characterizes a
role in the movement of substances in 6. F. L. M. Sheffield, Ann. Appl. Biol. 23, 506
particular system. The power density
and out of the guard cell or the transmis- (1936); W. V. Brown and C. Johnson, Sr.,
Amer. J. Bot. 49, 110 (1962); H. Meidner spectrum of the input white noise is
sion of stimuli. These observations con- and T. A. Mansfield, Physiology of Stomata
0..(f) = P, where P is the power spec-
(McGraw-Hill, New York, 1968); 0. V. S.
firm reports, based on light microscopy Heath and T. A. Mansfield, in The Physiol- trum level and f is the frequency. Lee
(12), that plasmodesmata exist between ogy of Plant Growth and Development, M.
B. Wilkins, Ed. (McGraw-Hill, New York, and Schetzen (5) have shown that hn
neighboring cells and guard cells of Di- 1969); W. W. Thomson and R. DeJoumett, is given by the expression
cotyledonous species and complement Amer. J. Bot. 57, 309 (1970); R. A. Fisher,
Plant Physiol. 47, 555 (1971).
recent electron microscopic evidence 7. R. M. Spanswick and J. W. F. Costerton, hnTl1 72, . . . rTn) =
for their existence in a monocotyledon- J. Cell Sci. 2, 451 (1967); N. C. Spitzer, J.
Cell Biol. 45, 565 (1970).
n! 1P"
ous species (13). Certain other aspects 8. W. H. Arisz and M. J. Schreuder, Proc. K. n-i
Ned. Akad. Wet. C 59, 454 (1956); W. H.
of Vicia faba and Nicotianna tobaccum
guard cell ultrastructure have been de-
Arisz, Acta Bot. Neerl. 7, 1 (1958).
9. A. R. Spurr, F. Ultrastruct. Res. 26, 31 (1969).
{y(t) -
m=o
E Gm [h., x(t)]}
scribed (14). 10. F. A. L. Clowes and B. E. Juniper, in Plant
Cells, J. H. Burnett, Ed. (Blackwell, Oxford, X(t - 71)X(l - T2) * * X(t - Tn)
J. E. PALLAS, JR. 1968), p. 269.
Southern Piedmont Conservation 11. B. E. Juniper and P. W. Barlow, Planta 89, which is simply a cross-correlation be-
352 (1969). tween functions of the input white
Research Center, Agricultural 12. W. Gardiner and A. W. Hill, Phil. Trans.
Research Service, P.O. Box 555, Roy. Soc. London Ser. B Biol. Sci. 194, 83 noise and the response of the system
(1901); F. Kienitz-Gerloff, Deut. Bot. Ges.
Watkinsville, Georgia 30677 Ber. 20, 93 (1902); K. Esau, Hilgardia 13, (6); r is in seconds.
H. H. MOLLENHAUER 427 (1941); A. Sievers, Flora 147, 263 (1959); In this report we apply the Wiener
W. Franke, Umschau 62, 501 (1962); F. E.
C. F. Kettering Research Laboratory, Litz and W. C. Kimmins, Can. J. Bot. 46, theory to a three-stage neuron chain
Yellow Springs, Ohio 45387 1603 (1968). formed by the horizontal cell, bipolar
13. P. B. Kaufman, L. B. Petering, C. S. Yocum,
D. Baic, Amer. J. Bot. 57, 33 (1970). cell, and ganglion cell in the catfish
References and Notes 14. J. E. Pallas and H. H. Mollenhauer, ibid., Ictalurus punctatus retina (7). The
in press.
1.
2.
E.
E.
Tangl, Jahrb. Wiss. Botan. 12, 170 (1879).
Strasburger, ibid. 36, 493 (1901). 15. Supported in part by PHS grant GM15492. input to this chain is an extrinsic cur-
3. L. G. Livingston, Amer. J. Bot. 51, 950 (1964). Contribution from the Southem Branch, Soil rent injected into a horizontal cell that
4. R. Buvat, C. R. Hebd. Seances Acad. Sci. 245, and Water Conservation Research Division, forms part of a laminar structure ex-
198 (1957); A. Frey-Wyssling, J. F. L6pez- Agricultural Research Service, U.S. Depart-
Saez, K. MUhlethaler, J. Ultrastruct. Res. ment of Agriculture, in cooperation with the
University of Georgia Agricultural Experi-
tending across the entire retina. This
10, 422 (1964); J. F. L6pez-Saez, G. Gimenez- ment Stations. extrinsic current gives rise to a poten-
Martin, M. C. Risuenlo, Protoplasma 61, 81
(1966); T. P. O'Brien and K. V. Thimann, 12 October 1971 tial change in the bipolar cells, which,
in turn, evokes spike disoharges from
the ganglion cells, the output element in
our study (we assume that potential
White-Noise Analysis of a Neuron Chain: change inside the horizontal cell is pro-
portional to the magnitude of the ex-
An Application of the Wiener Theory trinsic current).
A preliminary analysis for the sys-
Abstract. The Wiener theory of nonlinear system identification was applied to tem under study showed that (i) the
a three-stage neuron chain in the catfish retina in order to determine the functional series can be truncated after the second
relationship between the artificial polarization of the horizontal cell membrane order term with small error, (ii) the
potential and the resulting discharge of the ganglion cell. A mathematical model kernels should be computed for values
was obtained that can predict quantitatively, with reasonable accuracy, the non- of their arguments up to 300 msec, (iii)
linear, dynamic behavior of the neuron chain. The applicability of the method is the white-noise bandwidth should be
discussed. We conclude that this is a very powerful method in the analysis of flat from essentially 0 up to 25 hz,
information transfer in the central nervous system. and (iv) the temporal length of the
identifying experiment should exceed
Analysis techniques for linear sys- response to white Gaussian noise, since 30 seconds in order to expect less than
tems have been used extensively in the with such an input there is a non- 5 percent error in the statistical esti-
study of biological systems, often by zero probability that any given time mates of the cross-correlations (ker-
using "small signals" or by making cer- function over a finite interval of time nels).
tain assumptions about the behavior of will be closely represented by some A white-noise signal (35 seconds
the system. However, a biological sys- sample of this noise, and therefore the long) was stored on analog magnetic
tem seldom behaves linearly even un- system will effectively be tested with all tape and was concatenated with itself
der "small signal" conditions (1, 2). In possible inputs (4). ten times to form the stimulus record
fact, from the functional point of view, Considering a nonlinear system S, (350 seconds long). An electrical cir-
nonlinearities in biological systems ap- which is subjected to an input white cuit was designed so that the magnitude
pear to be a necessity as is, for ex- noise x(t) and whose output is y(t), of the current passed through the elec-
ample, the logarithmic transformation Wiener has shown that y(t) can be rep- trode in the horizontal cell was pro-
of the sensory input in order to accom- resented by the expansion portional to this white-noise signal.
modate large ranges. cc
The ganglion responses in ten experi-
Wiener (3) postulated that a non- y(t) = E G,. [h,, x(t)] ments with identical white noise were
linear system can be identified by its n=o superimposed and represented by histo-
1276 SCIENCE, VOL. 175
 A

sec

Fig. 1. The first (A) and second (B and C) order kernels for the neuron chain formed by the horizontal cell, bipolar cell, and gang-
lion cell in the catfish retina. The first order kernel hi(T) is plotted as a function of time T, and it is the impulse response of the linear
part of the transfer function of the chain. The second order kernel h2(Ti, T2), which describes the nonlinear behavior of the system, is
represented by a-three-dimensional solid made of the amplitude of the kernel plotted against Tr and T2. As seen from the display, the
second order kernel is symmetric about the diagonal line rl = T2. For a clear view of the kernel h2, the amplitude axis is reversed in
polarity in (C). View B shows the peaks well while view C permits a good view of the valleys (appearing as peaks) in h2(Ti, r2).
A peak represents a nonlinear interaction that enhances the response (facilitation), while a valley is an inhibitory nonlinear con-
tribution to the response. The grid lines are spaced 8 msec apart. The maximum of h2 is 53.0 and the minimum is -18.0 (spike/
sec)/(na sec)2.

grams in time to produce a poststimu- response of the model system shows a of predicting, with reasonable accuracy,
lus histogram of the ganglion discharge marked improvement with the addition the response of the system to any input.
in response to the white-noise input as of the nonlinear kernel (about 60 per- Some of the characteristics of the neu-
a function of time. The latter is a cent reduction of the mean square er- ron chain that are revealed by the
continuous function, and thus the dif- ror) and that the nonlinear model close- derived model are (i) it has a latency
ficulty of dealing with the discrete spike ly approximates the system response. of about 10 msec; (ii) it is under-
events is avoided. This indicates that nonlinearity is a damped; (iii) it has a cutoff frequency
Kernel h1 (Fig. IA) is the best linear very important factor in this neuron of about 11 hz and a high-frequency
approximation (in the sense of mean chain. attenuation of 12 db per octave; and
square error) to the transfer function of The set of kernels {hl, h2} defines, in (iv) it is strongly nonlinear, acting as
the system, while the set {h1, h2} (Fig. effect, the nonlinear, dynamic transfer a low-pass differentiator followed by a
1, B and C) is the best second order function of the neuron chain. This is half-wave rectifier. Finally, (v) it is
nonlinear approximation. The kernel a global model derived by testing the suggested that the bipolar cell processes
h1(T) is the impulse response of the system equally and uniformly over the the signal linearly while the nonlineari-
linear model of the neuron chain, and entire input-function space as a con- ties occur at the ganglion stage (9).
h2 (T1, T2) indicates the nonlinear inter- sequence of using white noise as input In conclusion, this powerful and gen-
action between two portions of the in- (in contrast to using the specialized eral method (Wiener theory) should
put signal, T1 and T2 seconds in the functions, sines, and pulses of linear find wide use in the study of the cen-
past, in affecting the response of the theory). As such, this model is capable tral nervous system whenever the
system at the present. For example,
h2(TIL, TO) shows that stimulus pulses
close together (about 10 msec) would Fig. 2. (Trace A) -
+20 na

produce a nonlinear interaction that White-noise input Ap

0
current to the hori-
would facilitate the response, while if zontal cell. (Trace 1-20 na

the two pulses are separated by more B) Resulting experi-

than 40 msec their nonlinear interaction mental response of
would inhibit the subsequent ganglion the ganglion cell
cell response. Indeed, this prediction [type B cell in (10)]
represented by his-
was verified experimentally (by two- tograms with a bin
pulse experiments). Examination of width of 8 msec
hl(T), h2(T1, T2) also suggests that the (average of ten ex-
system can be represented by a linear periments). (Trace C)
system followed by a nonlinear one, so First order Wiener
model response to
that a functional sequence is deter- the same white-noise
mined in the total process. Our physio- signal (hi only, that
logical interpretation of this implica- is, a linear model).
tion is that the bipolar cell processes (Trace D) Second
order Wiener model
the signal linearly while the nonlinearity response (hi and h2,
occurs at the ganglion cell stage (8). that is, a nonlinear 1.0 sec I
It can be seen in Fig. 2 that the model).
17 MARCH 1972 1277
 primary interest is to analyze the in- Institute of Technology, Cambridge, 1956);
formation transfer across several neu- D. A. George, Technical Report 355 (Re- compared with Bertine and Goldberg's
ronal stages by defining the dynamic
search Laboratory of Electronics, Massa-
chusetts Institute of Technology, Cambridge,
estimates of 3400 x 109 g/year for the
1959); J. F. Barrett. J. Electron. Control 15, amount of sulfur produced from the
input-output relations of the systems 567 (1963). combustion of oil. Their calculation for
rather than to explain the physio- 5. Y. W. Lee and M. Sehetzen, Quarterly
the amount of sulfur produced from the
Progress Report 60 (Research Laboratory of
chemical basis of interacting subcellular Electronics, Massachusetts Institute of Tech- combustion of coal results from the as-
systems. nology, Cambridge, 1961).
PANOS Z. MARMARELIS
6. Very few attempts have been made to apply
the Wiener theory. The difficulties are that
sumption that, during combustion, 5
KEN-ICHI NAKA (i) conditions for convergence of the series percent of the sulfur (and all other ele-
are not known, (ii) computation
Divisions of Applied Science creases almost exponentially with
time in-
the order ments) is mobilized. The estimate made
and Biology, of the computed kernel, and (iii) low in this technical comment is based on fig-
signal-to-noise ratio and high internal noise ures from the Study of Critical Environ-
are usually encountered in biological systems.
California Institute of Technology, See L. Stark [Neurological Control Systems
Pasadena 91109 (Plenum, New York, 1968)] and J. Katzenel-
mental Problems (SCEP) (2), which ap-
son and L. A. Gould [Inform. Control 5,
108 (1962); ibid. 7, 117 (1964)]. See also E.
parently are based on the assumption
References and Notes J. Thomas [Bell Syst. Tech. J. 50, 2797 that about 80 percent of the sulfur in
1. For cases of systems with "small signal" (1971)]. coal is released as SO2.
7. K. I. Naka, Science 171, 691 (1971).
nonlinearities see, for example, H. Spekreiise 8. F. S. Werblin and J. E. Dowling, J. Neu- Further refinement of the comparison
(2) and J. Levett [Vision Res. 10, 1347 (1971)]
and papers on the pupillary reflex system rop6ysiol. 32, 339 (1969); A. Kaneko, J. of fossil fuel emissions can be made on
and neutral threshold systems. For the many Physiol. (London) 207, 623 (1970). the basis of a recent estimate made by
biological "unidirectional rate sensitivity" 9. By a specialized nonlinear technique in which
systems, see M. Clynes and J. H. Milsum a single, no-memory nonlinearity is assumed
(2), a similar nonlinearity has been found
Berner (3) that, on a global basis, 42,000
[Biomedical Engineering Systems (McGraw-
Hill, New York, 1970), pp. 212-316] among in the ganglion cell discharge evoked by light x 109 g/year of sulfur is contributed to
many others. in the goldfish retina. rivers from rock weathering. Bertine and
2. H. Spekreijse, Vision Res. 9, 1461 (1969); 10. K. I. Naka and P. W. Nye, J. Neurophysiol.
Kybernetik 7, 22 (1970).
11.
34, 785 (1971).
We thank Dr. G. D. McCann for his support
Goldberg's estimates of 140,000 X 109
3. N. Wiener, Nonlinear Problems in Random
Theory (Massachusetts Institute of Technolo- and B. G. Elgin for his help in the construc- g/year of sulfur mobilized by weathering
gy Press, Cambridge, 1958).
4. A. G. Bose, Technical Report 309 (Research
tion of Fig. 1. Supported by PHS grant NB
03627.
apparently is the total rate of river runoff
Laboratory of Electronics, ML,sachusetts
estimated from average concentrations
13 April 1971; revised 22 November 1971 v of SO42- in river waters and annual river
flow rates. This value is in agreement
with my calculations (Table 2).
Berner (3) uses concentrations of
Sulfur Mobilization as a Result of Fossil Fuel Combustion SO42- in rivers given by Livingstone
(4). The total rate of river runoff from
An informative comparison of the perspective that they attempt to give. Ap- these data amounts to 120,000 X 109
amounts of trace elements mobilized by parently they overlooked the fact that g/year. The data compiled by Living-
weathering processes and those mobi- most of the sulfur in coal and oil is con- stone covered a period from 1848 to the
lized by the burning of fossil fuel has verted to SO2 upon combustion. In Table mid-1950's. Eriksson (5) estimates that
been given by Bertine and Goldberg (1). 1 I list various estimates of the amounts in 1960 about 10,000 X 109 g/year of
In the case of sulfur they have made an of sulfur emitted as SO2 from fossil fuels sulfur entered rivers from the application
error in the amount mobilized by fossil for the world and for the United States at of fertilizers to soil. It is doubtful that the
fuel combustion which distorts the different times. These numbers are to be total rate of river runoff given above ac-
curately reflects the fertilizer component
for 1960. Indeed, it is difficult to estimate
Table 1. Amounts of sulfur released as SO2 by fossil fuel combustion (X 109 g/year). an appropriate amount, since undoubt-
Year Area Coal Oil Total Reference edly some fertilizer component is rep-
resented in the river water data for
1937 Global 23,300 4,100 27,400 (8)
1943 Global 25,700 4,500 30,200 (8)
S042- concentrations. In this technical
1965 Global 25,500 5,500 31,000 (7) comment I assume that a total rate of
1966 United States 11,000 (9) river runoff of 120,000 X 109 g/year is
1968 United States 9,100 2,000 11,100 (2)
1968 Global* 45,000 5,000 50,000 This work applicable for 1960.
* Based on table 7.3 of (2) showing that the 1967 U.S. energy consumption from solid fuels is From data presented in the SCEP re-
20 percent of the global value and that from liquid fuels is 36 percent of the global value. port (2), I have calculated that between
1960 and 1970 the global use of fertilizer
nutrients increased by a factor of 2.6. By
Table 2. Estimates of present rates of sulfur mobilization (X 109 g/year). The first two items give assuming that the proportion of sulfur in
a new comparison of rates of mobilization based on fossil fuel combustion and weathering. The
third item includes the components of sulfur from river runoff as detailed in the text. the fertilizer has not changed over this
Source of sulfur
period, I estimate that the present contri-
Rate bution of fertilizer sulfur to river runoff
Fossil fuel mobilization 50,000 is 26,000 x 109 g/-year. This would
Weathering mobilization 42,000 bring the estimated present total amount
Sulfur entering rivers from: of sulfur in river runoff to 136,000 x
Soil application 26,000
Deposition from the atmosphere 50,000-70,000 109 g/year.
Volcanism 8,000 The sulfur compounds emitted into
Total river runoff (from above) 126,000-146,000 the atmosphere eventually are deposited
Mobilization from nonfuel industrial sources 14,000 on the surface of the earth. Various in-
Pollution sulfur mobilized (air) 64,000
Total pollution sulfur mobilized (air plus water) 90,000
vestigators (5-7) have constructed
global models for the sulfur cycle in
1278 SCIENCE, VOL. 175