1973 - Woolley PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

J. theor. Biol.

(1973) 39, 73-89

Asynchronous Division in Cell Colonies


I. General Considerations and a Linear Control Model
WARREN H. WOOLLEY~

Department of Physics and Astronomy,


University of Maryland, College Park, Maryland 20742, U.S.A.

ANDREW G. DE Roccof

Institute for Molecular Physics, University of Maryland,


College Park, Maryland 20142,
and
Physical SciencesLaboratory,
Division of Computer Researchand Technology,
National Institutes of Health, Bethesda, Maryland 20014, U.S.A.

(Received 23 December 1971, and in revisedform 20 June 1972)


Asynchronous cell division is examinedgenerallyfor featureswhich have
mathematical analogsin the field of statistical physics. A simple linear
model for the controlling chemicalkinetics is proposed,which yields the
essentialfeatures of the experimental division time distribution. These
resultssuggestan experimentby which the conceptof talandic temperature
(Goodwin, 1963)might be studiedquantitatively.

1. Introduction
Cell division is a general phenomenon, an attribute common to most cells
in spite of significant variations among them with respect to form and
function. It is easy to observe, to study quantitatively, to conceptualize:
a single cell grows and at some point divides into two (occasionally more)
daughter cells. The process of division can be observed visually without
perturbing the system and furthermore is well-defined to within a few percent
of the life cycle. From such observations one can construct a characteristic
distribution of division times (Fig. l), and this distribution is influenced
by the growth medium, by the particular strain chosen and, of course, by
7 Present address: MathematicsResearchCenter, Code 7840, Naval Research
Laboratory, Washington,DC. 20390,U.S.A.
# Author to whomreprint requestsshouldbe addressed.
73
74 W. H. WOOLLEY AND A. G. DE ROCCO

. .-
r

FIG. 1. Division time distribution. A gamma function distribution is fitted to the data of
Prescott by a method of least squares (Rubinow, 1968).
n(t)=[NOaV+‘lT(v+l)](~-to)” e-act-to), where a =0.243 min-‘, v:= 5.318, to =85 min,
No = 766.

the necessity of choosing a cell at random from a colony in steady-state,


exponential growth.
Another relatively simple class of experiments yielding equivalent infor-
mation is that in which a number of cells, no, is isolated sharing in common
the property that each divided more or less at a time t,. Various methods
have been devised to establish the initial synchronous state and to con-
tinuously monitor the total population (Burns, 1962). The outcome of such
an experiment is a growth curve, n(t), an excellent example of which is due
to Prescott (1959) and included here as Fig. 2.
At time to the normalized conditional probability that division occurs
during the interval (t, t + dt) is given by
P,(tlt = to) dt = B(t - to) dt, (1)
and the cells are fully synchronized. During the time course of the experi-
ment the colony approaches a steady-state, exponential growth characterized
by
0 t < t’
lim P,(tlt’ 6 t 6 t’+z) dt = const. t’ < t < t’ + z (2)
r-L-0
i0 t>t’+z
where normalization is insured by selection of a finite observational interval.
ASYNCHRONOUS DIVISION IN CELL COLONIES 75

FIG. 2. Growth curve. Data from Prescott (1959) on Tetraltymena geleii HS for n(t) is
shown.

At this point the colony is asynchronous, a situation which occurs in practice


after five or so generation times.
This loss of synchrony is a direct result of a non-zero variance of the
division time distribution, f(t). The relationship between n(t) and f(t) has
been discussed by Hirsch & Engelberg (1965, 1966a,b) and clearly has the
form
-$ n(t) = 2 j f(t-s) ts n(s) ds. (3)
--m
This straightforward and intuitively reasonable result has been obtained
also by more elaborate methods (Martinez, 1966) starting from the result
cited by Harris (1963),
n(t) = [l -G(t)]n, + 2 5 n(t-u) dG(u), (4)
0
in which G(r) represents the probability that a cell have a lifetime T < t.
In such an expression the initial cell colony is taken as the unit. Equation (3)
has also been obtained from the equation of von Foerster (1959), viz.

gt n(t, a) + $ n(t, a) = -A(?, a>n(t, a), 0)


where n(t, a) dt da represents the number of cells in the interval dt about t
with an age in the interval da about a, and where &t, a) is a probability
density that a cell at time t and with age a be lost by division during the
interval in question.
Each of these last two results emerges from a consideration of the distri-
bution of ages in a population. In a natural way equation (3) provides a
connection between the population curve n(t) and the age distribution f(t)
and the generalization of equation (5) described by Rubinow (1968) follows
naturally the form of equation (3). A solution for equation (3) can be
76 W. H. WOOLLEY AND A. G. DE ROCCO

obtained by iteration (Bronk, Dienes & Paskin, 1968) while the technique
employed by Hirsch & Engelberg (1965) provides relations between the
moments off(t) and the Laplace transform of i(t).
For the most part, however, studies of cell proliferation have been con-
fined to empirical and curve-fitting analyses of growth data. The work of
Rubinow (1968) comes closest to a consideration of internal control processes
when he investigates the consequences of a maturation velocity which may
or may not be constant. The final result of his study, however, is based on
f(t) and the suggested distribution of maturation rates is never expressed
explicitly.
To illustrate some of the inherent difficulties which attend an effort to
relate macroscope growth characteristics and underlying cellular controls
we describe next a simple mechanical analog, the intent of which is primarily
expository.
Consider a vertical tube containing a liquid, at the top of which we
introduce particles of such mass and density that they attain terminal velocity
yet continue as they fall to undergo random displacements due to molecular
collisions. A description of the system can be constructed by following the
trajectory of each particle and the vertical spread for all of them can be
formulated in a probabilistic manner.
How can this situation be made to resemble the appearance of asynchrony?
Suppose now that we employ an opaque tube having windows only at
selected, regular intervals along its vertical dimension. Furthermore, let
us forget all we know about the hydrodynamic behavior of particles in a
gravitational field, and at the end of the experiment disregard any identi-
fication of which data were obtained at which window. (We remark, paren-
thetically, that we do not know a priori whether the major contribution
to spreading comes from random collisions or from a distribution of terminal
velocity within a group of similar particles. We will not, however, allow a
particle to travel the full way across any given window backwards, and it
seems reasonable, therefore, that the terminal velocity distribution dominates
molecular collisions as a contributor to the spread of distribution in space.)
We begin by placing all the particles at the top at time I, (the experimental
realization for equation (I)-the J-function distribution) and subsequently
measure the time and frequency of “sightings” at the windows. The sum
over all windows of the frequency of sighting mimics

R(t) = $ in n(f),

while the frequency of sightings restricted to the first window simulatesf(t)


(cf. Fig. 3).
ASYNCHRONOUS DIVISION IN CELL COLONIES 77

FIG. 3. Spread of spatial distribution. A b-function distribution in Cl disperses as time


increases due to a distribution in C,.

It is from an essentially blind experiment such as this that we would


like to deduce as much as possible of the nature of the force compelling
progress of a particle, the medium through which it travels and the nature
of the particle itself. Much of this difficulty is also present in genuine syn-
chrony experiments.
Recent developments in cellular biochemistry have made synchrony an
especially interesting question. Specifically, a synchronous colony, were it
attainable, would be a powerful experimental system for the study of the
time evolution of chemical concentrations (metabolic, epigenetic and
genetic) during a cell cycle. The evidence to date, however, suggests that
spontaneous maintenance of a synchronous culture is unlikely and synchrony
experiments per se have been demoted largely to the status of preparatory
techniques. Most of the concise and useful data, therefore, comes from the
earliest experiments. In some cases appreciable synchrony can be obtained
if the species is entrained by external signals. Thus 85% of all divisions in
Gonyaulax polyedra (Sweeney & Hastings, 1958) can be made to occur
within a 5 hr interval spanning the end of a dark and the beginning of the
next 12 hr period of continuous bright illumination. Interestingly enough,
the synchrony is lost within 4 to 5 days in constant bright light, and in this
respect the appearance of asynchronous division is in no way atypical.
The need for synchronized cell colonies, though, has not diminished and
it has stimulated the development of a remarkable repertoire of techniques
for maintaining at least partial synchrony for many generations. Typically,
these procedures make use of periodic external stimuli, which couple
to the developmental control system of the cell producing entrainment.
Such techniques do not, however, produce genuine total synchrony, and
there is reason, even, to suspect that the external stimulus itself affects the
78 W. H. WOOLLEY AND A. G. DE ROCCO

normal cell cycle. An improvement may be achieved in those experiments


where the synchronized cell culture is released from, for example, a periodic
blocking stimulus (Petersen, Tobey & Anderson, 1969) and typical results
obtained are illustrated in Fig. 4. But in this questions arise concerning
whether the synchrony is lost before or after the cells have returned to
normal cyclic behavior in the variables of interest. These procedures of
entrainment or block-release represent, from the physical point of view,
the application of an unknown forcing function upon an unknown, probably

FIG. 4. Block release method of synchronizing. Cell count is measured in a thimidine-


synchronized cell population (Petersen et al., 1969). A =first block, B = first release,
C = second block, D = second release.

non-linear, oscillator in the anticipation that the forced and unforced phases
will be approximately equal.
Although the conditions expressed by equations (1) and (2) are satisfactory
for representing total synchrony or total asynchrony, it is less clear how
one should characterize intermediary degrees of synchrony. Percent phasing
was suggested by Zeuthen (1958) and defined as
T-T
% phasing 3 ~ x 100.
z
where z is the one-half generation time and T is the time required for that
50 % of the cells to divide which lie between 25 % and 75 %. Engelberg (1961,
ASYNCHRONOUS DIVISION IN CELL COLONIES 79

FIG. 5. Percent synchronization of a hypothetical cell culture: n, growth curve R,


normalized rate of cell division [R = d/df In (n)]R=,normalized rate of cell division for an
exponential culture having a generation time of 20 hr (R, = In 2/20 hr-I). Percentage
synchronization = 30% (Engelberg, 1961).

1964) on the other hand chooses to define synchrony (cf. Fig. 5) as


area of overlap (crosshatched) x loo
‘A synchrony 3
area under R in one doubling time ’
and has shown (Engleberg, 1961) that the area under R over a doubling
time is constant for an arbitrary growth characteristic and, furthermore, that
non-dividing cells in the population have negligible effect on the calculated
percent synchrony (Engelberg, 1964). More complicated procedures have
been suggested (Hirsch & Engelberg, 19663; Hahn, 1966) but as they are
less easy to deal with experimentally, they have not enjoyed great favor.
From what we have said so far it is a natural step to point out that
synchrony in cell colonies is reminiscent of entropy in mechanical systems.
As the distribution of division activity becomes spread evenly in time so
that no one interval is favored, percent synchrony becomes extremal. If our
cell colony were instead a system of gas particles expanding within a con-
tainer, we would have observed, as the system approached spatial uniformity
and equilibrium, that the entropy became maximized. The entropy is not
uniquely defined away from equilibrium. Neither is synchrony. This parallel
can prove useful for considering loss of synchrony as the inevitable approach
to a state of exponential growth, which we view as the “equilibrium state”
for a living cell colony. Fluctuations from this equilibrium have been
observed qualitatively and were once considered as spontaneous synchrony.
Such events have an obvious counterpart in thermal physics, and the view
which suggests itself is that loss of synchrony in a biological system and
80 W. H. WOOLLEY AND A. G. DE ROCCO

relaxation to equilibrium in a mechanical system share many features in


common.

2. Ensemble Methods in Biology


Although as early as 1937, Vito Volterra had commented on the conserved
function appropriate for population dynamics, and had given it the instructive
name “demographic energy”, it was not until the systematic studies of
Kerner (1959, 1961, 1962, 1964) that the formalism of the Gibbs’ ensemble
was introduced to investigate a system of differential equations other than
the dynamics of Newton. An excellent review of the approach to be employed
here has recently become available (Goel, Maitra & Montroll, 1971).
Kerner’s original work was concerned with the analysis of predator-prey
interactions between competing species. A similar path was followed by
Goodwin (1963) for a specialized set of differential equations characterizing
feedback control of protein synthesis in cells. This system of equations has
since been generalized (Woolley & De ROCCO, 1970) for more complicated
kinds of interactions between genetic loci and their metabolic products.
From these examples we see that the ensemble techniques of statistical
physics are general enough to be applied to other sets of differential equations
than those which gave rise originally to the body of techniques; in a larger
sense we are dealing with a statistical theory of differential (or perhaps
more generally, a differential difference) equations. In each case a parameter
playing the role of temperature arises and is associated with the mean
square value of some appropriate dynamical variables. What is not clear,
however, is whether the formal identifications with temperature, energy, etc.,
have any significance for the system in question. The possibility of a statistical
ensemble does not guarantee relevance. For the case of Goodwin, as an
example, one would hope to find a rational basis for cellular physiology,
but is is not clear that talandic temperature tells us anything useful about a
cell. It is one of the basic motivations of this work to investigate whether
a determination of the talandic temperature can be associated with the
relaxation of a synchronous cell culture to asynchronous equilibrium.

3. A Linear Control Model


In this section we shall pursue the idea that the concentration of certain
biochemical components of a cell control specified cellular functions and
relate this control to synchrony in cell colonies. In what follows we have
chosen to have cell division coupled to the control system, although the
analysis would be suitable for any other chemically regulated event in a
cell.
ASYNCHRONOUS DIVISION JN CELL COLONIES 81
Let a particular biochemical concentration, Cr, trigger division when it
reaches a value Cf. Further, let C1 be produced in proportion to a second
concentration, C,, which is constant over the cell cycle for a particular
cell. At division the daughter cells each receive a concentration C, as a
kind of inheritance. Fluctuations in both space and time are to be expected
for C, within a given cell but will not be of significance over the cel1 cycle.
If, on the other hand, a spatial fluctuation occurs at the time of division,
then it will result in a permanent change in C2 for each daughter cell.
Finally, we propose that the supply of C, be exhausted in the reaction which
induces division.
In a single cell, then, the controlling chemical species follow the differential
equations
G = PC, (6a)
c2 = 0, (6b)
and division occurs when C1 = CT in an unspecified but all-or-nothing
manner. We have chosen to write equations (6a), (6b) in a manner which
emphasizes the chemical foundations for our model, yet it is simple enough
to appreciate that they just as easily could describe the force-free motion
of a particle of mass cc. For the chemical case, p is a rate constant which
is taken to be identical for all cells of a given strain in a given milieu. The
value of C,, on the other hand, depends on the initial conditions and to this
quantity we assign a certain continuous probability density w(C,).
This distribution, w(C,), is maintained in successive generations by the
fact that cells are chosen at random from a colony which has experienced
exponential growth for some time; and this requirement is necessary for
experiments of the type discussed in this study. Interestingly enough Hirsch
& Engelberg (1966b) have referred to such colonies as “ergodic” although
the use of such an expression in the strictest sense requires some elaboration.
What such an expression should be taken to mean is this: a colony in
steady-state exponential growth has reached a distribution of C, among
its members which is maintained by random events at each succeeding
division. So Iong as a large enough sample is chosen, subject to the condition
that they have just divided, then a uniformly consistent distribution of C2
is maintained by such a choice. An experiment which selects the initial
condition of a synchronous colony according to a property identified with
division, e.g. mass, can lead to “non-ergodic” colonies.
By “ergodic”, on the other hand, we could just as easily mean that in the
multidimensional space of the family {C,, C,} the projection of the time
course of a single cell (in a colony) or a single colony (in an ensemble)
satisfies the condition that no part of the concentration space is singularly
T.B. 6
82 W. H. WOOLLEY AND A. G. DE ROCCO

avoided and that any particular region in the phase space will experience
at least one trajectory. The former interpretation is more nearly a statement
of how, after equilibrium, the phase space is filled by a ribbon of the original
density such that any trajectory in phase either lies in the region of equilibrium
or, if outside, has most certainly just left and is soon to return again to the
region analogous to the Maxwell-Boltzmann phase space for molecular
systems. In this way it is easier to appreciate the connection between OUI
description of cellular systems and the established language and results of
statistical physics. For example, if we select in a gas at equilibrium, the mole-
cules in some small volume and follow their motion, then we will not observe
a change in the momentum distribution of the group of molecules, but they
will become evenly distributed in space. Similarly, in our cellular model,
we select cells in an interval AC, near CT which nonetheless have an equili-
brium distribution in C,. Now in time, the distribution in C2 remains fixed
while the distribution in C, becomes evenly spread between C, = 0 and
Cl = c:.

4. Calculations
To employ our model for the calculation of quantities of interest is a
straightforward matter. Rather than “resetting” the clock at C, = C:, we
employ the analytic device
(j-l)Cf< C, <jCy, (j=l,2 ,... ), (7)
where j orders the generations and where the jth generation implies that
C, represents 2’-’ cells. A given cell has
cl(t) = PC, 4
C2 = constant. (8)
By assigning a distribution w(C,) to the colony and an initial delta-function
distribution to Cr, we can solve for n(t). Specifically, the number of cells
in the jth generation at time t is given by
Cz= jCl*/at
n[(j--1)C: < Cl < jC:; t] = 2j-‘no s dc, w(G). (9)
Cz=(j-l)Cl*/ut

Since each cell in this interval (or generation) represents 2j-’ cells, we find

40 = f n[(j-l>C: < C, < jCT; I]


j=l
jCt*/W
= nojz 2j-’ (10)
(j- JCl*,prdc2 w(c2)’
This result agrees with Rubinow (1968) with the identifications CT = p,
C1 = age and C2:= rate of maturation.
ASYNCHRONOUS DIVISION IN CELL COLONIES 83

By considering only the first generation it is possible to relate w(C,)


to the division time distribution function, f(t). We calculate the number
of cells of the first generation with C, > CT at a given time:

n,(C, > c:; t) = n, c,=bt (11)


w(c2) dC2*
This represents the probability that a cell in the original synchronous colony
has divided by time t. We note that
o2
f w(C,) dC2 = 1
= (12)
C2=C*‘/pt>
indicates that all the cells have divided at least once in the long time limit.
Therefore the normalized probability of division during a given interval is
f(t) dt = n, ’ dn, (C, > CT; t)
-1
= no if nl(C1 > C:; t) dt

(13)

a result given also by Rubinow (1968) although deduced by him from


different considerations. Our linear model specifies a constant rate of
maturation, CZ, for a given cell in contrast to the possibility inherent in
the work of Rubinow for a non-constant rate of maturation.
Figure 1 reveals the good agreement which is obtained between experiment
and theory whenf(t) is fitted with a gamma function distribution, a result
to be expected from a consideration of equation (3). For our model, however,
the distribution of C2 necessary to fit a gamma function distribution in the
time variable seems somewhat artificial. Specifically, in order to achieve
f(t) = 0 for 0 < t < to, w(C,) must have a high velocity cut-off, w(C,) = 0
for C, > CT/p0 as well as a zero probability for negative values of C,.
The resulting distribution is skewed to the left and cannot be related to
fluctuations about a mean value.
A more natural distribution function for C, is suggested by ensemble
techniques of statistical mechanics. Our model of a cell may be represented
by a system of differential equations, thus:
Cl = PC2

& = 0

C”=f,(C,,c,, . **, c,>, (14)


84 W. H. WOOLLEY AND A. G. DE ROCCO

where C2 is coupled to only C, and the set of functions f3 through fn are


unspecified. By application of the Lie-Koenig theorem, Kerner (1964) has
shown that the set of equations-C’, through &can be transformed into
canonical form with a Hamiltonian-like constant of the motion.
For the variables
51 = (Cl/P)--G4
52 = P(C, - G,,
tj = tj(c,, c,, . . .) C,), (j = 3,4, . . ., n),
where C, is the average value of C, over the colony, the integral of the
motion is

G(t,, . ., 5,) = k t;+g+n(L 94,. . ., t3, (15)

and from it the differential equations for the cell are given by the canonical
equations of motion. Because IZ is large we employ ensemble methods to
construct a phase space density in the customary fashion as
P = p. exp (--WI.
It is of importance to point out that our empirical system is especially
appropriate for the canonical ensemble since it contains many similar
systems (cells) weakly coupled to one another by the growth medium and
division events. The parameter /? is the Lagrange multiplier for the constraint
G = constant, and its inverse, 8 = /?-l, has been given the name “talandic
temperature” by Goodwin (1963) for kinetic equations describing cell
control.
For this ensemble, then, the expected distribution function for C2 is
w(G) G = p. ev [--W~PFL)~~I
G j6 .
. ..Sdr.exp[-PG,-n(53,...,5n>l
= W4W* exp C- W2MC2 - WI dG) (16)
which is exactly the form to be expected for an equilibrium system, but in
this case fi is no longer (kT)-’ since we are dealing with configurations not
equivalent to thermodynamic equilibrium. Although this distribution does
allow negative values for C,, the likelihood can be made arbitrarily small
by judicious selection of b and Ct, an expedient employed also in conven-
tional statistical physics (Landau & Lifshitz, 1958).
The division time distribution predicted by this model is
t-’ ev { - WWKCTIPO- WI dt
f(t) dt = m--- (17)
d t- ’ exp { -
UI~~>C(~~/PO
- CA”>d f ’
ASYNCHRONOUS DIVISION IN CELL COLONIES 85
which is obtained by inserting the explicit expression for IV(&) as expressed
by equation (16) into the more general expression equation (13). Note that
the normalization has been altered to accommodate the lower limit of inte-
gration and the parameters indicated in equation (13). By use of L’Hospital’s
Rule we conclude that for t = 0, the exponential term, exp (- l/t’) dominates
the term t * which appears in the denominator. Therefore, as t = O,f(O) = 0
as required.
There are some features of equation (17) which, in spite of its generally
satisfactory nature, nevertheless cause certain technical difficulties. For
example, the second and higher moments of the distribution are undefined,
and the mean,
cc
(0 = 0j tf(t) dt,
is a severely complicated function of the parameters. For these reasons it
is not practical to obtain values for the parameters by matching the mean
value, the variance and the skewness of the distribution with experimental
values.
We note that distributions of this type (viz. undefined moments for all
orders n > k where the nth moment is (t”) = j t”‘(f) dt) are respectable
0
mathematical constructs, though infrequently used. These are Cauchy rype
distributions named after the Cauchy distribution which has infinite mean
and all higher moments. Thus, while a given moment may be calculated
from data of a finite sample, it bears no relation to parameters in the
distribution function. Powell & Errington (1963) concluded after examination
of various species of bacteria that a distribution with infinite moments
may be necessary to describe the data for long division times.
These problems can be circumvented, however, by utilizing other charac-
teristics of the experimental distribution function. A quantity, 7, defined as
the time at which one half of the cells have divided, is easily deduced from
experimental data. So long as the integral of w(C,) on the interval (- co, 0)
is much less than unity, then ? is given by
f = G/(P~*), (18)
since
-[ NC,) dC, + p NC21 dC, = c[ NC,) G

due to symmetry. Therefore, if

jm WCC,) dG = 0,
86 W. H. WOOLLEY AND A. G. DE ROCCO

as assumed, the time required for one-half of the cells to reach C: is obtained
from equation (8) as C:(i) = ,uCz 7.
An additional quantity of interest is the “most probable division time”,
t,, which is the maximum value of the probability density function, f’(t).
Solving for t, we obtain the quadratic expression

and upon choosing the physically relevant (positive) root and expanding
we find that
CT 2cy

The distribution of division times can be expressed conveniently in terms


of ? as follows:
tp2 exp { -(P&/2)[(i/t)- I]‘)
f(t) = z- ____-.----~~

1 te2 exp (-(B~C:/2)[(i/t)-11’} dt’

-,

FIG. 6. The data of Prescott (1959) histogram and distribution function f(t) obtained
from estimates I = 110 min, 1, = 108 min. Then <t > = 111 min.
ASYNCHRONOUS DIVISION IN CELL COLONIES 87
Note that both I and the dimensionless array (@C:/2) can be obtained
from experimental data by application of the analysis outlined above.
Accordingly we have estimated values of f and t, from the data of Prescott
(1959). Figure 6 shows the resulting distribution function and the associated
data points. A more accurate procedure would be to utilize a least squares
fit and obtain unambiguous values for I and (fl$2/2). We observe that the
probability of a cell dividing before a time of 40 min is less than lo-l2
thereby vindicating the approximations concerning very high and negative
values of C,.

5. Further Remarks
Earlier studies of talandic temperature have been associated with feedback
oscillator systems in resting cells which exhibit circadian rhythms (Goodwin,
1963). It has been difficult to find for these systems a convenient way for
relating the talandic temperature to operational characteristics of the cell.
For growing cells, however, the variables associated with the talandic tem-
perature are simpler and more directly related to experimental observables.
From an analysis of the relaxation of a synchronous cell colony to asyn-
chronous growth there can be obtained-from the division time distribution
-quantitative information concerning the talandic temperature of a cell
and the underlying oscillator. In this regard we note that Goodwin (1963)
has estimated that for a typical bacterium size considerations alone suggest
that very few (perhaps only one) clocks of any reliability could be supported
by the available biochemical complexity.
There are several ways in which this model can be connected to experi-
mental growth curves. First, of course, is the fit of the division time distri-
bution. In this connection, however, it is important to point out that the
data itself does not discriminate between various forms for the distribution
function, principally because the intrinsic precision required is not easy to
obtain in practice. Nevertheless, Kubitschek (1962a,b, 1966) has already
noted that a wide variety of celI types satisfy a Gaussian distribution in the
variable t - I, an observation entirely in keeping with our model.
While it is possible to extract f from the data with exceptional accuracy,
the value oft, (i.e. the most likely time of division) is subject to considerable
uncertainty. This uncertainty is compounded in practice because the para-
meter relevant to the distribution function
y _ m2
--=--?. i
2 i-t,
from equation (19) is highly sensitive to changes in t, when i-t, < 2:as
is the case in Fig. 6. Thus a change of 1 min in t, produces a change in y
88 W. H. WOOLLEY AND A. G. DE ROCCO

on the order of at least 28 %. This is especially troublesome because Prescott’s


data is a histogram divided into units of 2 min. For this reason we see that
a least squares fit will be necessary to test this model for its basic feature
of talandic temperature.
The linear model described here does not have the feature that the talandic
temperature can be disentangled from the parameters characteristic of the
control dynamics. The dimensionless array y = &/2tI can be obtained,
however, and one would expect that any change in 9 which occurs by
application of the pulse technique suggested by Goodwin (1963) would
lead to changes in y but not in 7, which is independent of the talandic tem-
perature 0(= 8-l). A careful fit to experimental data will be necessary for
this purpose, however, since the distribution function will be characterized
by an altered width (maximum) without change in 7. These conclusions
depend upon the assumption that only 8 is changed by this technique.
Should the results prove inconclusive further study of the possible I-depen-
dence of CT, p or C, would be appropriate.
One last point of some interest can be made. If a cell with a division
time Tr > 1 were to have daughter cells with T2 such that TI > T2 > I,
then after a suitable number of divisions the value of T would approach
near to 7. There exist cells (Schaechter, Williamson, Hood & Koch, 1962),
notably bacteria, for which a cell having Tl > I produces progeny with
T, < 1. Although this negative correlation of division times does not follow
from the linear model described here, we have shown how it can arise when
the control is determined by a non-linear oscillator (a point we shall
discuss in detail in a subsequent publication). The question of choice need
not arise now, however, for there is no reason to suppose that all cells have
identical control systems for division.
What we have presented here is a method for relating a biochemical
control mechanism to an observable, macroscopic cellular function. Although
we have chosen to discuss cell division, the analysis presented is generally
applicable to chemically controlled events in a cell. This approach offers the
advantage that it is a non-destructive method for examining control systems
in tel.
The systems studied here, by techniques closely related to statistical
physics, exhibit a stability of cellular function without recourse to limit-cycle
feedback oscillators, and, moreover, tend towards a talandic equilibrium
in much the same way as a mechanical system relaxes to thermodynamic
equilibrium once released from external constraints. When more is under-
stood of biochemical control of macroscopic cellular functions, we might
expect non-equilibrium statistical methods to become a useful and important
tool.
ASYNCHRONOUS DIVISION IN CELL COLONIES 89
One of the authors (W. H. W.) is indebted to the Grandfather Mountain
Highland Games for financial support during a year of reasearch at the Institute of
Animal Genetics, Edinburgh, Scotland; to Dr. H. Kacser for his hospitality and
guidance and to the researchers at the Institute, particularly K. M. Andrews, for
contributing to the background needed for this study.

REFERENCES
BRONK, B. V., DIENES, G. J. & PASKIN, A. (1968). Biophys. J. 8, 1353.
BURNS, V. W. (1962). In Progress in Biophysics and Biophysical Chemistry, Vol. 12, Q. 1
New York: Pergamon Press.
ENGELBERG, J. (1961). Expl. Ceil Res. 23, 218.
ENGELBERG, J. (1964). Expl. Cell Res. 34, 111.
GOEL, N. S., MAITRA, S. C. & MONTROLL, E. W. (1971). Rev. mod. Phys. 43, 231.
GOODWIN, B. C. (1963). Temporal Organization in Cells. New York: Academic Press.
HAHN, G. M. (1966). Biophys. J. 6, 275.
HARRIS, T. E. (1963). The Theory of Branching Processes, Q. 140. Englewood Cliffs, New
Jersey: Prentice-Hall, Springer-Verlag.
HIRSCH, H. R. & ENGELBERG, J. (1965). J. theor. Biof. 9, 297.
HIRSCH, H. R. & ENGELBERG, J. (1966~). Bull. math. Biophys. 28, 391.
HIRSCH, H. R. & ENGELBERG, J. (19666). In Cell Synchrony (I. L. Cameron and G. M.
Padilla, eds). Chapter 2. New York: Academic Press.
KERNER, E. H. (1959). Bull. math. Biophys. 21, 217.
KERNER, E. H. (1961). Bull. math. Biophys. 23, 141.
KERNER. E. H. (1962). Ann. N.Y. Acad. Sci. 96. 975
KERNER; E. H. i1964j. Bull. math. Biophys. 26,‘333.
KUBITSCHEK, H. E. (1962~). Nature, Land. 195, 350.
KUBITXHEK, H. E. (19626). Expl. Cell Res. 26, 439.
KUBITXHEK, H. E. (1966). Nature, Land. 209, 1039.
MARTINEZ, H. M. (1966). Bull. math. Biophys. 28, 411.
PETERSEN, D. F., TOBEY, R. A. & ANDERSON, E. C. (1969). In The Cell Cycle (G. M. Padilla,
I. L. Cameron and G. L. Whitson, eds). Q. 341. New York: Academic Press.
POWELL, E. 0. & ERRINGTON, F. P. (1963). J. gen. Microbial. 31, 315.
Prmcorr, D. M. (1959). Expl. Cell Res. 16, 279.
RUBINOW, S. I. (1968). Biophys. J. 8, 1055.
SCHAECHTER, M., WILLIAMSON, J. P., HOOD, J. R. & KOCH, A. L. (1962). J. gen. Microbial.
29,421.
SWEENEY, B. M. & HASTINGS, J. W. (1958). J. Protozool. 5, 217.t
VOLTERRA, V. (1959). Ref. in Theory of Functionals and of Integral and Zntegro-Differential
Equations, Q. 24. New York: Dover Publications, Inc.
VON FOERSTER, H. (1959). In The Kinetics of Ceil Proliferation, Q. 382. New York: Grune
and Stratton.
WOOLLEY, W. H. & DE Rocco, A. G. (1970). Biophys. J. 10, 183.
ZEUTHEN, E. (1958). Adv. biol. med. Phys. 6, 37.

7 We are grateful to the referees of this journal for their remarks to us on this important
paper.

You might also like