Ecological Modelling 290 (2014) 110120

Contents lists available at ScienceDirect

Ecological Modelling
j our nal homepage: www. el sevi er . com/ l ocat e/ ecol model
Hyperbolic features of the circadian clock oscillations can explain
linearity in leaf starch dynamics and adaptation of plants to diverse
light and dark cycles
Franc ois G. Feugier

, Akiko Satake
Graduate School of Environmental Science, Biosphere Department, Hokkaido University, Kita-10, Nishi-5, Kita-ku, Sapporo 060-0810, Japan
a r t i c l e i n f o
Article history:
Available online 12 December 2013
Keywords:
Carbon starvation
Circadian rhythm
Resource management
Starch
Photoperiod
Sucrose homeostasis
a b s t r a c t
Perfect resource management is the most important factor for plants to guaranty optimal growth and
reproduction. To be constantly supplied with sucrose, their most important resource, plants store a part
of it as transitory starch during daily photosynthesis, and degrade starch back into sucrose at night, when
photosynthesis is not possible. Observations show that starch accumulation and degradation are linear,
forming a pyramid shape, and that this shape adapt to various photoperiods to constantly supply the leaf
with sucrose. Transitory starch buffer management is performed by a careful control of carbon partition-
ing, starch degradation andsucrose export key rates, to insure optimal growth whatever the photoperiod.
Few theoretical studies have addressed this intriguing starch buffer management. Here we characterize
the shape of the three key rates that allow sucrose homeostasis to persist in different photoperiods. We
built two models of sucrose and starch dynamics including the three key rates formalized as piecewise
continuous functions. First model assumed non-saturating starch degradation (linear model) whereas
the second model assumed saturation of starch degradation activity by the substrate (saturating model).
In order to obtain sucrose homeostasis we adjusted the shape of the three piecewise continuous rate
functions using gradient descent on a sucrose homeostasis payoff function. At convergence of the gra-
dient descent, we always obtained sucrose homeostasis together with the characteristic starch pyramid
prole, as observed in empirical data. Surprisingly, in the linear model starch degradation rate at night
followed a unique hyperbolic shape whatever the photoperiod, whereas for the saturating model degra-
dation rate was constant at night but globally lower for shorter light periods. As light period shortened
starch accumulated faster while degradation speed decreased, consistently with data. For shorter light
periods, sucrose homeostasis was achieved at a level lower than the optimal one. We also reproduced
an unexpected early dusk experiment in which plants adapted to 12h light/12 h dark cycle experience
earlier dusk at 8h after dawn. Linear model could reproduce the instantaneous decrease in starch degra-
dation speed during the longer night, as observed in data, whereas saturating model failed. Our results
imply that a plant employing a unique hyperbolic starch degradation rate would be able to manage starch
buffer in various photoperiod conditions, whereas a saturating starch degradation requires a simple daily
constant rate but instantaneous regulation during sudden changes in photoperiod.
2013 Elsevier B.V. All rights reserved.
1. Introduction
Although photosynthetic carbon assimilation occurs only in the
light, carbon is required at all times for growth and respiration.
Transitory starch reserve in leaves enables continuous supply of
carbon throughout the day-night cycle (Smith and Stitt, 2007;
Yazdanbakhsh et al., 2011). Starch is stored during daily photo-
synthesis inside chloroplasts and broken down at night to provide
sugars to sinks, allowing metabolismand growth regardless of the

Corresponding author. Tel.: +81 11 706 2267; fax: +81 11 706 4954.
E-mail address: francois.feugier@ees.hokudai.ac.jp (F.G. Feugier).
presence of light (Gibon et al., 2004; Smith and Stitt, 2007; Graf
and Smith, 2011). Previous studies have reported that starch is
accumulated and is degraded almost linearly, rather thanexponen-
tially, drawing a pyramid shape in the day-night cycle (Gibon et al.,
2004) (Fig. 1a). More interestingly, the pyramid shape adaptively
changes in response to the various lengths of light period (Lu et al.,
2005; SmithandStitt, 2007; Graf et al., 2010; Graf andSmith, 2011).
Starchdegradationspeedimmediately decreases while its accumu-
lation speed increases (Fig. 1a) when the light period is shortened
from16h light/8h dark to 8h light/16h dark (Lu et al., 2005; Graf
et al., 2010; Graf and Smith, 2011) and there are no symptoms of
carbon starvation throughout the subsequent longer nights (Graf
et al., 2010). More surprisingly, when a plant initially placed in a
0304-3800/$ see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ecolmodel.2013.11.011
F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120 111
Fig. 1. (a) Typical starch diurnal pattern in long and short day and (b) owchart of
the model. (a) Scheme of observed starch diurnal pattern in long day (12h/12 LD
cycle; red) and short day (6h/18h LD cycle; blue) in plant leaves. The rate of starch
synthesis is larger in short days while the degradation rate is smaller. This allows
leaves to avoid running out of starch, and therefore sucrose, during night. (b) Flow
chart of the model. Carbon is captured by photosynthesis with rate g, and a fraction
(t) is partitioned into starch while a fraction 1(t) is partitioned into sucrose.
Starch is degraded into sucrose with rate (t). Degradation can also be inhibited by
too high level of starch (dashed line) creating a saturated degradation. Sucrose is
exported and respired with rates (t) and h, respectively. The three rate functions
(t), (t) and (t) can vary during the course of a 24h cycle, but are assumed to be
24 h-periodic. (For interpretation of the references to color in this gure legend, the
reader is referred to the web version of the article.)
12h light/12h dark cycle experiences an unexpected early dusk
at 8h after dawn, an instantaneous decrease in starch degrada-
tion speed is observed, as if the plant knew that the night would
be longer than usual because dusk came earlier. The mechanism
behind the pyramid built by plants and adaptation to diverse pho-
toperiods is mysterious and little is known about the signicance
of this characteristic shape.
Feugier and Satake (2013) previously proposed the hypothesis
that the feedback between circadian clock and starch and sucrose
metabolisms is the key mechanism to generate linearity in starch
diurnal patterns and adaptation to various photoperiods. They
developed a model linking the circadian clock to sucrose and starch
dynamics. In the model, three key processes, carbon partition-
ing between sucrose and starch, starch degradation, and sucrose
export, were described as cosine functions that approximate
circadian oscillations. They demonstrated that cosine generated
oscillations adjusted in time to minimize sucrose starvation are
sufcient to reproduce almost all the starch proles observed in
experiments, showingthat thecircadianclockregulationmayoper-
ate these key rates. However, whenthe photoperiodwas as short as
8h, the model predicted non-linear decay and depletion of starch
before the end of the night, which was not consistent with the
observations.
They explained that this discrepancy between the model and
data would be caused by the choice of the cosine function to repre-
sent circadian oscillations: to create a perfect linear decay of starch
at night degradation rate should follow an accelerating increase.
Thus the portion of cosine function having this feature would pro-
duce a linear decay while others will make the decay diverge from
linear.
In the present study we relax the assumption of cosine func-
tions by using piecewise continuous rate functions whose shapes
are undetermined. We also introduce a sucrose homeostasis objec-
tive function whose minimization will characterize the shape of
the three piecewise continuous rate functions. This new frame-
work allows decreasing the number of assumptions, making the
model more general, and sucrose homeostasis introduces more
objectively the dilemma between starch accumulation to prevent
starvation at night, and sucrose export to promote growth. That is
to say, the life time dilemma of all organisms: should I spare or
should I grow?
Our aim is to characterize the optimal shape of the three time
dependent key rates that would reproduce the characteristic pyra-
mid shape of starch prole in various photoperiods without any
assumption about starch prole itself.
2. Methods
2.1. Model for sucrose and starch dynamics in leaves
We start fromthe work fromFeugier and Satake (2013). In their
model the three key rates of carbon partitioning between sucrose
and starch, starch degradation, and sucrose export are formalized
by cosine functions summarizing the oscillating regulation from
the clock. They estimated the maximum amplitude of the oscil-
lations from available data, whereas the time shift was searched
by minimization of sucrose starvation cost function using gradient
descent.
Here we use sucrose and starch dynamics from Feugier and
Satake (2013) (Fig. 1b), while the three rate functions will be rede-
ned in next section. The gross production rate of sucrose by
photosynthesis is the constant speed of carbon capture g weighted
by the light availability function L(t) taking value 1 for light and
0 for dark. We consider a constant carbon capture rate, rather
than considering naturally varying light intensities, to be consis-
tent with empirical studies. During light period a fraction (t) of
carbon assimilated by photosynthesis is partitioned into starch (C)
which accumulates in the leaf through the day and a fraction
1(t) is partitioned into sucrose (S) immediately available for
growth and respiration. No partitioning occurs at night as there is
no photosynthesis (L(t) =0). Starch is degraded into sucrose with
rate (t) providing the only source of sucrose to support leaf res-
piration and growth at night. Sucrose is exported with rate (t)
to non-photosynthetic tissues such as roots and growing organs
via the phloem. The three time dependent rates (t), (t) and (t)
are aggregate parameters of multiple processes (such as cascades
of regulations; transcription and translation of various enzymes
involved in metabolism) and are treated here as equivalent to the
activity of carbon partitioning, activity of starch degradation, and
activity of sucrose export, respectively.
Taken together, temporal dynamics of sucrose and starch con-
centrations in leaves are represented by the following equations
(Fig. 1b):

S = (1 (t))gL(t) +(t)C (h +(t))S, (1)

C = (t)gL(t) (t)C. (2)

The rate of change in sucrose concentration (S) in (1) is equal to the
fraction 1(t) of carbon ux that is distributed into sucrose path-
way plus the ux of sucrose that comes from starch breakdown
(with rate (t)), minus the loss due to leaf maintenance and res-
piration (with constant rate h) and export to non-photosynthetic
tissues (with rate (t)). Similarly the rate of change in starch con-
centration (C) in (2) is equal to the complementary fraction (t) of
the carbon ux distributed into starch pathway, minus the loss due
112 F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120
Table 1
Symbols description and values.
Symbol Description Units Value for parameter or initial condition
for simulation
C(t) Starch concentration molC6/gFW
C(0) Initial starch concentration 0
g Carbon capture rate molC6/gFW/h 6.35 (half the value in Feugier and
Satake, 2013)
h Respiration rate 1/gFW/h 0.79 (Feugier and Satake, 2013)
k Half saturation constant molC6/gFW 1 (Figure 8a fromGibon et al., 2004)
N Number of sample points 48 (equivalent to a sample point every
30min)
S(t) Sucrose concentration molC6/gFW
S(0) Initial sucrose concentration 0

S Sucrose homeostasis level molC6/gFW 2.9 (average sucrose level in Figure 1a

in Blasing et al., 2005)
(t) Starch degradation rate Linear: 1/gFW/h
saturating:
molC6/gFW/h

(0)
Initial sampling vector All components
linear: 0.5; saturating: 4
a
(t) Carbon partitioning Dimensionless

(0)
Initial sampling vector All components
linear: 0.5; saturating: 0.5
(t) Sucrose export rate 1/h

(0)
Initial sampling vector All components
linear: 0.5; saturating: 0.5
Magnitude of constraint on starch
buffer
0.005
a
We chose

(0)
large enough to prevent starch accumulating in excess due to saturation of starch degradation, at the beginning of the gradient descent:

(0)
>g

(0)
.
to starch breakdown into sucrose (with rate (t)) for remobiliza-
tion.
As opposed to the previous model ((1) and (2)) where starch
degradation is linear with C, called linear model, we also con-
sider a model where starch degradation is subject to saturation by
C (MichalisMenten kinetics). The saturating model reads:

S = (1 (t))gL(t) +(t)
C
k +C
(h +(t))S, (3)

C = (t)gL(t) (t)
C
k +C
, (4)
where k is the starch concentration at half saturation of degrada-
tion activity. We introduce this model as starch may be saturating
the process in charge of its degradation, and also because it is able
to create linear decay of C at night (L(t) =0), when Ck and (t)
constant. Therefore the saturating model is also a candidate to
reproduce the linear trends observed in starch prole data. The
constant k was set to a value (see Table 1) in the range around
which starch prole loses its linearity before dawn when looking
at available data (Gibon et al., 2004, Figure 8a).
Now we need to determine the three unknown rate functions
(t), (t) and (t) that allow the model to satisfy sucrose homeo-
stasis at a given level.
2.2. Characterization of (t), (t) and (t) satisfying sucrose
homeostasis
From observations (Gibon et al., 2004) we can see that sucrose
(or sugars) level is rather constant during 24 hours compared
to starch level. This could be the results of a ne management
of sucrose level: too low sucrose level would slow down cell
metabolism, whereas too high sucrose level would be a waste
for growth or storage organs and would also decrease photosyn-
thetic activity (Sawada et al., 2001). The result of this behavior
may produce sucrose homeostasis at an optimal emergent level
resulting from natural selection. Considering small variations in
sucrose level, or even using actual sucrose data in our modeling
would make results specic to the chosen data set, hence making
the model less general, which is what we want to avoid. Thus we
choose the homeostatic sucrose level to be the average value of
sucrose concentration over 24h from data (see Table 1), which is
less data specic than using sucrose data proles itself and still bio-
logical. Homeostasis has a clear biological meaning which makes
interpretation of our results more intuitive.
Here we try to characterize the shape of the three unknown rate
functions (t), (t) and (t) that satisfy the homeostasis hypothe-
sis without any prior assumption on their shape. Homeostasis is
stated as a payoff function that, once it is minimized, guaranties
the sucrose level to be constantly equal to the homeostasis level

S.
The payoff function is dened by the functional:
P(t
0
, t
1
) P[, , ] =
t
1

t
0
[

S S((t), (t), (t))]

2
dt, (5)
where P(t
0
, t
1
) 0; S((t), (t), (t)), as in Eq. (1), is the sucrose
concentration at time t for given functions (t), (t) and (t). The
three rate functions are considered to be periodic over T, which is
basically 24h. When t
0
corresponds to dawn, if P(t
0
, t
0
+T) =0, then
S((t),(t),(t)) is equal to

S during one whole LD cycle, and the
leaf is assumed to have a sucrose level allowing optimal metabolic
reactions, that is, not starving neither accumulating sucrose instead
of exporting it. On the other hand if P(t
0
, t
0
+T) >0, sucrose is not
equal to

S during all or part of the period T, and metabolismis not
optimal, therefore the leaf tries to minimize the payoff function
P(t
0
, t
0
+T) during the period T by adjusting the shapes of (t), (t)
and (t) at each moment in time.
To minimize the payoff function by adjusting the shape of the
three time dependent rate functions (t), (t) and (t) we use the
gradient descent method. In a previous study, cosine functions
were used for the three rates, but here we adopt a more ex-
ible method using piecewise continuous functions. We consider
the three sampling vectors

,

and

made of N sample points
over the period T; and the piecewise continuous functions (t),

(t)
and (t) being the linear interpolation of the sample points of each
F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120 113
vector respectively. Specically, linear interpolation is used to esti-
mate the functionvalues for times that donot correspondtosample
times in the sampling vectors, hence making piece linear continu-
ous functions. Given the functions are periodic we consider in fact
(t mod T),

(t mod T) and (t mod T).
Shapes of the three rate functions are updated by adjusting each
sample point in the vectors as follows:

i
(n+1)
=
i
(n)

i
(n)
(1
i
(n)
) (6)

(n+1)
i
=
(n)
i

P

n
sign

(n)
i

(
(n)
i
)
2
(7)

(n+1)
i
=
(n)
i

P

n
sign

(n)
i

(
(n)
i
)
2
, (8)
where i [1, N] indicates the ith component of the sampling vec-
tor containing N sample points; the super script n stands for the
gradient step number; is a small constant adjusted automatically
for stability during the gradient descent. Partial derivatives of the
payoff function were calculated numerically on stabilized proles
of the dynamics (i.e. dynamics ran long enough to be independent
from initial conditions, with well-developed periodic patterns of
the variables). More formally, the increment in the value of a con-
sidered sample point is proportional to the partial derivative of
the payoff function with respect to that considered sample point,
weighted by the constraint term. The constraint term, after the
derivative in (6)(8), is a numerical technic helping to restrict the
values of the sample points within the domain of denition of
their associated rate function. Explicitly, (t) [0, 1] for all t pos-
itive, so the constraint
(n)
i
(1
(n)
i
) in (6) tends to 0 as a sample
point approaches either 0 or 1, reducing the gradient step in that
direction. Also, (t) 0 and (t) 0 for all t positive, and so the con-
straints sign

(n)
i

(
(n)
i
)
2
in (7), and sign

(n)
i

(
(n)
i
)
2
in (8), tend
to 0 as a sample point approaches 0, preventing the point value
becoming negative.
Numerical procedure for calculation using Mathematica 8.0 to
characterize the shapes of the rate functions is as follows (similar
for both the linear and saturating model) using parameter values
and initial condition fromTable 1.
In the rst step of the gradient descent we solve numerically for
three days the dynamics (1) and (2) (or (3) and (4) for the saturat-
ing model) using the initial conditions of the piecewise continuous
functions (t),

(t) and (t) inorder to obtaina stabilized patternof
the dynamics. Thenwe calculate the payoff (5) for the thirdday. We
calculate the numerical partial derivative of each sample point of
thethreesamplevector. We independentlyupdatethevalueof each
sample point of each sampling vector in the direction decreasing
the payoff function using ((6)(8)). Finally we calculate the new
payoff (5) using the updated piecewise continuous functions. If the
newpayoff is smaller than the one before updating the sample vec-
tors then we run the second gradient step, that is, the procedure of
the rst gradient step but using the updated sample vectors and
the last values of S and C instead of their initial conditions, and
repeat this procedure for further gradient steps until convergence
of thegradient descent decidedwhenthefollowingcriterionis met:
(P
new
P
old
)/ <0.01.
The set of vectors

,

and

that satises this criterion is the
optimal set that minimizes the payoff function (5). For the saturat-
ing model, we performed the same procedure using Eqs. (3) and
(4) instead of (1) and (2). Basically, this method is similar to solving
numerically a variational problem.
2.3. Conditions for sucrose homeostasis and linear starch decay
with minimumstarch buffer size
The model as it is dened, may not give a unique set of solu-
tions for the three rate functions (t), (t) and (t) at sucrose
homeostasis. In fact, when P(t
0
, t
0
+T) =0, S(t) =

S for t [t
0
, t
0
+T],
which is equivalent to having sucrose dynamics (1) or (3) at steady
state, hence having sucrose at homeostasis. This implies the follow-
ing condition for the three rate functions in order to accomplish
homeostasis; setting (1) equal to 0 and solving for S(t) =

S we
obtain:

S =
gL(t)(1 (t)) +(t)C(t)
h +(t)
, (9)
which, after replacing with (2) can be rewritten as

S =
gL(t)

C
h +(t)
. (10)
Same condition as (10) can be obtained from (3) and (4) for the
saturating model. Condition (10) tells us that sucrose homeostasis
can be achieved with a linear starch prole (

C = const. provided
(t) =const. or (t) h), or a nonlinear starch proles (

C / = const.
provided (t) / = const. and is compensating changes in

C), and
this, independently of starch level C(t) itself. Importantly, condi-
tion (10) does not guaranty that starch prole will be linear at
sucrose homeostasis. Linear starch prole requires that the right
hand side of (2) is equal to a positive (negative) constant during
the day (night).
Together with condition (10), homeostasis at prescribed level

S during the complete period T also requires that carbon entering

the systemduring the day equals respiration and export during the
period T, that is:
gt
dusk
=

ShT +

S

T
0
(t) dt. (11)
Where t
dusk
is the time of dusk which is between 0 and T. In case
(11) does not hold, sucrose level cannot equal

S during the whole
period T, and can either equal

S partially, or be always below

S and
still be constant if (10) holds. If export at night is not mandatory
for the plant the integral term in (11) goes only from dawn (t =0)
to dusk (t =t
dusk
).
Given that sucrose homeostasis depends on

C, and not C(t) (pro-
vided rate functions have the proper shapes), average starch level
can take any value as long as starch prole satises (10) at all
times, and (11) holds. We need a way to remove the uncertainty
on the level of C(t) while not assuming anything on its shape and
on (t), (t) and (t) which are the functions we aim to charac-
terize. This would help narrowing down the number of admissible
rate functions when sucrose homeostasis is achieved.
Empirical data for starch show that leaves may be using the
smallest possible buffer necessary to spend the night, as it is always
almost depleted at the end of the night (Stitt and Zeeman, 2012).
End products of photosynthesis (sucrose, starch) are known to
inhibit photosynthesis (Sawada et al., 2001), which may be the
reason leaves manage to keep only the required amount of starch
in their buffer, rather than always keeping a high concentration.
Keeping a large amount of starch in leaves is also a waste for
growth. Using these facts, we include a constraint on C(t) in the
payoff functiontorestrainthenumber of solutions allowingsucrose
homeostasis. Nowthe payoff function (5) becomes:
P(t
0
, t
1
) =
t
1

t
0
[

S S((t), (t), (t))]

2
+C(t)dt, (12)
114 F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120
/
/
Fig. 2. Proles of starchandsucrose for linear andsaturating models obtainedfromthe gradient descent invarious photoperiods. (a) Starchand(b) sucrose of linear model; (c)
starch and (d) sucrose of saturating model. Blue to red colors represent the different photoperiods: 6h/18h, 8h/16h, 10h/14h and 12h/12h LDcycle conditions, respectively.
(For interpretation of the references to color in this gure legend, the reader is referred to the web version of the article.)
where is a small positive constant. The newtermin (12) after the
square bracket tends to minimize the daily total amount of starch
that sojourned as buffer in the leaf, while adding no particular
constraint onstarchprole, andnot affectingmuchsucrose homeo-
stasis. Consequently, after convergence of the gradient descent,
the resulting optimal rate functions achieving homeostasis will
be those also corresponding to the smallest possible starch buffer
allowing homeostasis.
3. Results
3.1. Linear feature in starch prole and photoperiod adaptation
As explained in Section 2.3, linear starch prole is not set explic-
itly in the model and it is not obvious to obtain such prole after
convergence of the gradient descent. For various initial conditions
sucrose homeostasis was always achieved together with a linear
starch decrease. Linearity of starch increase was more sensitive to
initial conditions in 12h/12h and 10h/14h light/dark (LD) cycles.
Thus there is no unique set of solutions for the three rate func-
tions, able to minimize the cost function (12) despite our effort to
decrease the number of candidate rate functions. However, among
the different sets, given by different initial conditions, some sets
gave results very close to empirical data, and thus we concentrated
our analysis using one suchset givenby initial conditionsuppliedin
Table 1. Results showed a linear starch prole for all photoperiods,
and both the linear and saturating models gave very similar results
(Fig. 2) at convergence of the gradient descent (see Appendix for
convergence of gradient descent). However, rate functions achiev-
ing sucrose homeostasis are very different between both models
(Fig. 3). In 12h/12h and 10h/14h LD cycles, gradient descent pro-
duceda perfect homeostasis of sucrose at the optimal level

S, which
corresponds to a payoff function (12) close to 0. Resulting starch
proles for both models are linearly increasing during the day and
linearly decreasing at night until dawn, where starch is just above
zero. For the linear model, in the morning, as starch buffering (t)
is lowand export (t) is large (Fig. 3a and c), a large fraction of the
photosynthetic products are directly exported. At the end of the
day, starch buffering is larger while export is low, to allow lling
starchbuffer for the night. Inthe saturatingmodel, the constant ux
of captured carbon g(t) (due to constant (t) in Fig. 3d) is larger
than the constant starch degradation rate (t) (saturation term in
(4) is close to 1 for large starch level) in Fig. 3e, allowing a con-
stantly increasing buffer (Fig. 2c). At night both models showlinear
starch decay (Fig. 2a and c). In the linear model this is achieved
by an increasing (t) compensating decreasing starch level. In the
saturating model starch degradation speed equals (t) (due to sat-
uration term close to 1) which is constant, creating linear starch
decay.
In shorter light periods (8h and 6h) sucrose homeostasis is still
achieved in both models, but at a level lower than

S. Again, pro-
les of starch are linearly increasing during the day and linearly
decreasing at night until dawn where starch is just above zero,
and linearity of starch increase became more robust to various
initial conditions. Interestingly, both models generate a lower but
constant sucrose level during 24h, rather than higher, closer to

S,
during the day, then lower or close to 0 at night.
For bothmodels, as light periodlengthens, slope of starchdegra-
dation at night becomes steeper, meaning that starch degradation
into sucrose is becoming faster. If ux of starchinto sucrose is large,
F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120 115
Fig. 3. Proles of the three piecewise continuous rate functions (t), (t) and (t) obtained fromthe gradient descent for linear and saturating models in various photoperiods.
(t), (t) and (t) in the linear and saturating model are shown in (a)(c) and (d)(f), respectively. Blue to red represent colors for the different photoperiods: 6h/18h, 8h/16h,
10 h/14h and 12h/12h light/dark cycle conditions, respectively. (For interpretation of the references to color in this gure legend, the reader is referred to the web version
of the article.)
then keeping sucrose level constant requires larger sucrose export.
On Fig. 3c and f we observe that export rate (t) increases as light
period increases. However, for very short light periods, export is
turned off to keep sucrose inside the leaf for the long night. In
very short photoperiods it is likely that sucrose is imported from
storage organs to avoid starvation in the leaf, or that the leaf dies
out due to insufcient sucrose for maintenance. We also observe
that during long nights starch degradation rate is lower in the sat-
urating model (Fig. 3e) while it is surprisingly unchanged in the
linear model (Fig. 3b) but still satisfying the pyramid shape and
photoperiod adaptation (Fig. 2a).
In 12h/12h LD cycle starch prole slightly loses its linearity in
both models during the day. This can happen when sucrose avail-
ability is in excess. However, as soon as light period decreases as to
induce sucrose paucity, pyramid shape emerges robustly together
with a starch buffer with minimal size to prevent any wasteful
starch accumulation.
3.2. There is a critical photoperiod below which plants cannot
maintain a safe sucrose level
Whatever the photoperiod, leaves needto buffer starchto spend
the night with a constant and safe level of sucrose. In our results,
for both models, accumulation of starch is linear and faster as
light period shortens, while decay is linear and slower as night
period lengthens, which is consistent with observations (Gibon
et al., 2004; Graf et al., 2010).
If day is shorter, night becomes longer, so more starch must be
buffered at a faster pace during the day. This can be observed in
10h/14h LD cycle (Fig. 2) where the peak of starch at 10h after
dawn is higher than the one in 12h/12h LD cycle, and slope is
steeper. Still, for a light period as short as 10h, the leaf is not expe-
riencing any metabolic problemas it is still able to export sucrose
in excess over 24h (Fig. 3c and f) and maintain sucrose level at

S (Fig. 2b and d). For even shorter light periods it is necessary to

116 F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120
buffer even more starch, and so the peak of starch in 8h/16h LD
cycle shouldbe higher thanin10h/14hLDcycle. However, nowthe
light period becomes too short to buffer enough starch at speed g
for the 16h night period while maintaining respiration even when
ceasing export during the day (Fig. 3c and f). For that reason the
peak height is lower than in 10h/14h LD cycle (Fig. 2a and c) and
so is the homeostatic sucrose level (Fig. 2b and d).
We can easily calculate the shorter light period the plant can
endure while staying constantly at

S during the total period T .
The total carbon ux captured during the light period t
dusk
that
can be buffered is the fraction that was not used by respiration nor
exported, that is t
dusk
(g h

S)

S

t
dusk
0
(t)dt. Duringthe remaining
dark period T t
dusk
, a quantity (T t
dusk
)h

S +

S

T
t
dusk
(t)dt of
sucrose is respired and exported. If starch that was accumulated
during the day is completely used by respiration and export during
the night, the former must be equal to the latter. After rearranging,
we nd (11).
However, for a critically short light period t
dusk
=t
c
, export is
stopped to allow sucrose staying at the level

S. Setting therefore
export to 0 in (11) and rearranging we nd:
t
c
=
Th

S
g
, (13)
Using parameter values from the numerical analysis we nd
t
c
=8.65h. That is, for light periods shorter than 8h 40min, sucrose
level will not be able to stay during 24h at the safe level

S, and
the leaf will start to experience metabolic slowdown. This critical
light period is an extreme value as sucrose export to other organs
is completely shut down. One can also consider that export ceases
only at night by splitting export into day and night export in the
integral termof (11), setting night export to zero and keeping daily
export in (13). Critical light period may also depend on parameters
such as plant species, photon ux intensity, carbon dioxide par-
tial pressure, etc. Based on (13) we chose a smaller value of g than
in Feugier and Satake (2013) to obtain an earlier onset of sucrose
paucity than in experimental conditions (see Table 1).
In Fig. 2b and d we clearly see that sucrose level is exactly at

S
for 12h/12h and 10h/14h LD cycles, whereas level decreases for
8h/16h which is shorter than t
c
. Finally, in 6h/18h LD cycle the
speed of carbon capture g is too slow to sustain respiration and
starch buffering together (Fig. 2b and d), even when stopping all
sucrose export as seen in Fig. 3b and e. Still, the leaf maintains
sucrose homeostasis but belowthe optimal level

S.
When light period is shorter than t
c
the constraint to pull down
starchlevel in(12) becomes useless as starchcannolonger accumu-
late in excess for nightly export. Furthermore, when export ceases
during night, linear starchdecay is guaranteedby (10) for any initial
conditions.
3.3. In the linear model, starch degradation rate at night is an
increasing hyperbolic function
The interesting feature of the linear model is the prole of (t) at
night (Fig. 3b). This prole is identical among any LDcycle, meaning
that one type of regulation for starch degradation may be suf-
cient to maintain sucrose homeostasis regardless of the length of
the night period. Whatever the starch concentration at dusk, linear
decay will occur thanks to the increase of (t) at night, producing
a constant starch degradation speed. Furthermore, starch level will
be almost the same at the end of the night whatever the starting
level at dusk (Fig. 2a).
We can nd the ideal function (t) allowing a perfect
linear decrease of C(t) at night. First, let us assume that
C(t) is a linearly decreasing function with time such as
C(t) =C
dusk
a
night
t =a
night
(t* t) with a
night
>0 being the nightly
/
Fig. 4. Proles of starch degradation in linear model with hyperbolic (t), and sat-
urating model with constant (t). When using (14) (with t
*
=25) as the degradation
rate of starch in (2) at night (L(t) = 0) with different values of starch at dusk, starch
prole (blue) is perfectly linear and reaches 0 at t =t
*
. When using a constant degra-
dation rate (t) in (4) at night (L(t) = 0) with different values of starch at dusk (same
as with the hyperbolic case), prole of starch (red) diverges fromlinear when starch
level approaches k. Also, in the second case, degradation rate (t) is constant but
must be adjusted depending on starch concentration at dusk. Therefore in this g-
ure we set (t) =(k +C(12))/t* to obtain the same slope of starch prole as for the
linear model at dusk time. Vertical dashed line represents dawn of the following
day (For interpretation of the references to color in this gure legend, the reader is
referred to the web version of the article.)
degradation speed (the slope of starch prole), C
dusk
the concen-
tration of starch at dusk and t* =C
dusk
/a
night
0 the time at which
starch concentration reaches 0. Plugging this expression into (2) at
night (L(t) =0) we obtain upon rearranging (t) =1/(t* t) which is
a hyperbola. Hence, a hyperbolic increase of (t) allows a perfect
linear decay of starch (Fig. 4). Now(2) becomes

C =
1
t

t
C (14)
Also, this hyperbolic shape of (t) insures that whatever the con-
centrationof starchat dusk, starchwill always decrease to 0 exactly
at the end of the night if t
*
corresponds to dusk time (Fig. 4). How-
ever, it is more likely for t
*
to be later than the end of the night
(otherwise the degradation rate would have to become innitely
large), allowing starch to come close but not equal to 0 at the end
of the night (Fig. 4), and then to increase again on the next day
(L(t) =1).
In the same way, by considering that C(t) =a
day
(t* +t) with
a
day
>0, t* =C
dusk
/a
day
0 and replacing into (2) for the day (L(t) =1),
we obtain (t) =(g(t)/a
day
1)/(t* +t) which is also a hyperbola if
(t) is constant, guarantying again linearity for the daily starch
increase. When (t) is not constant with time (Fig. 3a and d for
12h/12h LDcycle), depending on initial conditions, starch increase
deviates fromlinear, as slightly seen just after dawn on Fig. 2a and
c for 12h/12h LD cycle.
3.4. Starch degradation rate is mostly constant during day and
night in the saturating model
Both the linear and the saturating model produce very similar
results for sucrose and starch proles (Fig. 2) as they both satisfy
the payoff function. However, their respective rate functions are
very different (Fig. 3). Given that in the saturating model starch
degradation ux is only proportional to (t) for large starch levels
F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120 117
(due to saturation), starch prole is guaranteed to be linear if (t)
is constant.
The increase of (t) at the end of the night is due to the low
starch concentration which turns the saturating degradation into
a linear one. As a result, the saturating model behaves similarly to
the linear model at the end of the night.
Interestingly, due to the introduction of saturation of starch
degradation in the model, (t) became mostly constant during day
andnight, while not equal betweenday andnight. The function(t)
can be thought of as the product of the degradation constant by the
concentration of enzymes involved in starch degradation. As dark
period increases (light period decreases), value of (t) decreases,
that is, the concentration of degrading enzymes would be inverse
proportional to the length of dark period.
Albeit consistent with empirical starch proles for various pho-
toperiods, the saturating model can be distinguished from the
linear model by performing the experiment of early dusk explained
in the next section.
3.5. Instantaneous decrease of starch degradation speed in
unexpected early dusk reects background behavior of (t)
So far, both the linear and saturating models produced results
consistent with observations. Therefore, we tried to confront them
to other conditions in order to verify their robustness. In an experi-
ment (Graf et al., 2010), anunexpectedearly dusk 8hafter dawnfor
plants initially in 12h/12h LD cycle, was triggered. This resulted in
aninstantlyslower starchdegradationspeed(milder slopeof starch
prole at night). We tried the same condition for the linear and sat-
urating models. We usedthe three rate functions resulting fromthe
gradient descent in 12h12h LDcycle to run a simulation for 9 days,
in order for the dynamics to stabilize. Then we shortened the light
period of the 10th day from12h to 8h. Just after the early dusk, the
linear model (Fig. 5a) shortly exhibited nonlinear and fast starch
decay, and then decay became linear from 12h to the end of the
night joining the same starch concentration at dawn than in nor-
mal 12h/12hLDcycle condition. This canbe understoodby looking
at Fig. 3b in 12h/12h LD cycle. At early dusk, 8h after dawn, (t)
is still following approximately a decreasing hyperbola (because
(t) is not constant) toward 12h. For that reason, starch decay is
not linear until it meets the portion of (t) which is an increasing
hyperbola, from 12h to 24h. Given that any initial concentration
degraded following (14) converges to the same point at the same
time t
*
, starch decay reaches automatically a safe concentration at
the end of the night. We also observe that sucrose level stays lower
thanoptimal level but constant (Fig. 5a) fromthe time of early dusk.
On the other hand, the saturating model fails to bring starch
concentration to a safe level at the end of the night (Fig. 5b). Given
that (t) is constant during day and night (Fig. 3e), and that starch
decay speed is equal to (t) for large starch level, starch is rst
degraded during 8h to 12h with the values of (t) in 12h/12h LD
condition. Then from12h, value of (t) in 12h/12h LDcycle condi-
tion becomes larger. Indeed, slope of starch degradation is steeper
in 12h/12h LDcycle than in 8h/16h (Fig. 2c). Hence degradation of
starch in early dusk experiment from12h is similar to degradation
in 12h/12h LDcycle, as can be seen by the parallel slopes in Fig. 5b.
However, early dusk creates an unexpected longer night during
which starch buffer is degraded too fast, maintaining as a result
a safe sucrose level for a part of the night, followed by complete
depletion of sucrose (Fig. 5b).
3.6. Transient hyperbolic features are enough to produce linear
accumulation and decay of starch
We tried using a very simple biological oscillator to verify how
much of hyperbolic features are necessary to reproduce linear
/
/
Fig. 5. Starch degradation prole after an unexpected early dusk. Here we suddenly
shortened the light period from12h/12h LD cycle (red, same results as Fig. 2a and
c for 12h/12h LD cycle) to 8h/16h LD cycle (blue) while keeping the three rate
functions obtained fromthe gradient descent in 12h/12h LDcycle condition. Linear
model (a) still shows linear decay in 8h/16h LD cycle and sucrose level is lower
than

S but constant. Starch decay is not linear in the saturating model (b) in 8h/16h
LD cycle where starch goes to 0 before the end of the night, while sucrose level is
maintained as long as possible to

S and goes to zero when starch buffer is depleted.
(For interpretation of the references to color in this gure legend, the reader is
referred to the web version of the article.)
starch proles. We chose the simplest oscillator, namely the preda-
tor/preymodel, andgiventhat we are mostlyinterestedinverifying
how different it is from a cosine function we consider qualitative
results. Furthermore, the predator/prey model is not exactly like
the circadian clock.
When the level of predator is used instead of (t) in (2), we
can adjust parameters (oscillation period, initial prey/predator) to
obtain surprisingly realistic starch proles (Fig. 6). Starch prole is
mostly linearly increasing during the day and linearly decreasing
at night.
We also reproduced the early dusk experiment. Light period of
the third day was shortened or lengthened ranging from 6h hour
to 16h by 2h steps. We observe that the slope of starch degrada-
tion, even if not perfectly linear for shorter light periods, becomes
steeper as light period increases, and this without any additional
sophisticated regulation.
Setting to zero (t) during the day in (1-4) would also produce
a perfect linear increase if (t) is constant, while preventing starch
turnover.
4. Discussion
We built two models of sucrose and starch dynamics to
understand how plants manage their starch buffer in various
118 F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120
10 20 30 40 50 60 70
10
20
30
40
0
0
time
S
t
a
r
c
h
P
r
e
d
a
t
o
r
0.5
1.0
1.5
Fig. 6. Starch prole using a degradation rate froma simple biological oscillator. We
use the simple predator/prey model as the clock regulating starch degradation rate.
Equations are x = x xy for the prey, y = xy y for the predator and

C = 23L(t)
0.7yC for starch. x(0) =1.2, y(0) =0.5 and C(0) =3.43. Morning light is synchronized
with the peak of the predator and time is rescaled with the period of oscillations of
the predator in order for one oscillation to occur in 24h. Simulation is run for 2 days
in 12h/12hphotoperiod, andthenonset of dusk is changedonthe thirdday from6h
to 16hhours after dawn by 2h steps. Dashed line shows predator oscillations; red
line corresponds to starch at 12h/12h photoperiod; lines with colors from blue to
yellow correspond to photoperiods from6h/18h to 16h/8h by 2h steps as shown
by thick and thin gray lines above the plot. Units are arbitrary. (For interpretation of
the references to color in this gure legend, the reader is referred to the web version
of the article.)
photoperiods to avoid metabolismslowdown or waste due to inap-
propriate sucrose level. Dynamics contained three unknown rate
functions that were formalized by piecewise continuous functions
whose shapes were adjusted in order to obtain sucrose homeosta-
sis at a dened level. When homeostasis was achieved both models
reproduced for any photoperiod the empirically observed linear
increase and decrease in starch concentration.
Despite the many combinations of the three rate functions to
keep a constant safe level of sucrose, the resulting combination of
rates always produced sucrose homeostasis with linear decrease
in starch concentration while linearity of starch increase was more
sensitive to initial conditions in longer light periods. Starch prole
observed in data may therefore reect plants effort to keep a con-
stant and safe level of sucrose together with and a minimumstarch
buffer, while linearity is not a necessary condition for homeostasis.
One major point in our ndings is the simplicity plants may
employ to manage their resources in changing environments.
Whatever the variations in intensity of photosynthesis during the
day and the state of the starch buffer at dusk, if starch degradation
rate at night follows a hyperbolic increase (14), starch buffer will
never be depleted during the night, allowing, at worst lower but
never 0, sucrose homeostasis. In the case of the saturating model
(t) would have to be adjusted depending on the starch buffer size
at dusk which could actually be the case to allowthe same phe-
nomenon. A hyperbolic degradation rate function is the only shape
allowing to deal withuncertainty inthe buffer size andnight length
in an automatic fashion (Fig. 4).
The early dusk experiment (Graf et al., 2010) was useful to
discriminate between the linear and saturating model. It allowed
inferring that the hyperbolic shape of (t) plays a crucial role in the
linear model. However, it is not excluded that the saturating model
can satisfy the early dusk experiment, but in that case (t) would
require its own sophisticated regulation at early dusk. Dynamics of
the rate functions to sudden changes in the photoperiod still need
tobe consideredinour study. Indeed, we only consideredrate func-
tions found by gradient descent for given, static photoperiods, and
then used these resulting rate functions for the early dusk exper-
iment, not allowing them to change with the new photoperiod.
However, early dusk experiments may trigger in the plant some
metabolic changes affecting instantly our three key rate functions
that could for instance instantly decrease nightly (t) in the sat-
urating model. Hence daily (t) would have two different shapes,
one before and one after the sudden change in photoperiod. Still,
we could show that the hyperbolic degradation rate from the lin-
ear model can achieve the same behavior as seen in empirical data,
even with early dusk experiment, whereas saturating model would
therefore need dynamical changes in its rate functions.
In case of the saturating model, a possible adaptation mecha-
nism could be as follows. If concentration of degradation enzyme
is constant and proportional to day length, or quantity of carbon
captured into starch if capture rate is not constant, then the longer
the day (the shorter the night), the faster the degradation of starch.
Reversely, synthesis enzyme could accumulate proportionally to
night length or change depending on the deviation from the safe
sucrose level, then the longer the night (the shorter the day), the
faster the accumulation of starch on the following day. Degrading
enzyme should be quickly degraded (or deactivated) under morn-
ing light, while synthesis enzyme should be quickly degraded (or
deactivated) at dusk, in order to be again accumulated. These last
twopoints are the dynamical changes. Then, incase of suddendusk,
starch degradation is proportional to the concentration of degrada-
tion enzyme accumulated, producing an adapted slope. Also, at the
onset of early dusk, accumulation enzyme is quickly degraded (or
deactivated) and accumulated during the longer night in order for
starch to accumulate faster the next day. However, this mechanism
is independent fromthe clock.
The role of (t) happens to be very important in the models, as
it is responsible for starch accumulation and degradation speeds,
and therefore linearity of these two processes. Indeed, condition
(10) tells us that for constant or small (t), sucrose homeostasis can
be achieved given that (t) is hyperbolic (linear model) or constant
(saturating model), hence also producing a linear starch prole.
What could be the process for the plant to adapt to various light
periods? As we sawin Fig. 5, unexpected early dusk creates sucrose
paucity. Depending onthe previous nights deviationfromthe opti-
mal sucrose level, daily (t) (or the complex degrading starch and
regulated by (t)) could decrease (increase) proportionally to the
intensity of starvation to accelerate (slowdown) starch accumula-
tion on the following day, hence allowing adaptation to shorter
(longer) light periods (Fig. 2a and c). Starch synthesis enzymes
could also accumulate during the night as proposed previously for
the saturating model. This would be the role of (t).
We obtained faster starch increase through gradient descent,
not considering dynamical changes in rates functions: in shorter
light periods, we observed that (t) is lower only during the day
and (t) is higher for the linear model (Fig. 3b), while (t) is lower
during the whole LD cycle for the saturating model (Fig. 3e). As
explained earlier the saturating model may therefore exhibit a con-
stant (t) (enzymes, etc.) whose level corresponds to a specic light
period: low for short light periods, and high for long light periods.
If that is the case, the saturating model can efciently adapt to vari-
ous photoperiods as it can intrinsically produce linear starch decay,
whose slope is decided by one constant parameter depending on
the length of the light period.
In the case of the linear model, the fact that (t) is unique at
night whatever the light period, the starch and the sucrose levels,
may imply that starchdegradationis regulatedindependently from
starch or sucrose level at night. Also, daily (t) seems to be scaled
vertically, rather than shifted along time axis. These two facts may
conrmthat the circadian clock is regulating degradation of starch
(Lu et al., 2005) and possibly accumulation to avoid excess. How-
ever, there may be a mechanism to disengage starch degradation
from the clock during the day to avoid degradation together with
synthesis ((t)), leaving the clockcontinue its cycle until duskwith-
out affecting starch accumulation, hence producing a perfect linear
starch increase if (t) is constant. In our results (t) was not close to
F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120 119
zero during the day meaning starch was at the same time produced
and degraded. Preventing starch turnover during the day would
require further sophisticated regulation in our modeling, whose
eventual purpose after natural selection, may be to save ATP.
In the saturating model no rate function seems to behave in an
oscillating manner, but rather such as a step function between day
and night, which makes the interpretation of the role of the clock
less clear. Also, comparison for both models of starch simulation
proles todata is not ambiguous, whereas tryingtocharacterizethe
molecular entity that would correspond to the three rate functions
is still challenging. Eveninempirical studies, connectingstarchdata
to genes expression or proteins level is difcult as concentrations
mynot reect activities (Smithet al., 2004; SmithandStitt, 2007; Lu
et al., 2005). Hence we still needto further dene the real molecular
nature of our three key rates.
The peculiar hyperbolic shape of the degradation rate (t) in
the linear model may give an evolutionary advantage since it can
replacea sophisticatedpredictionor regulationmechanism. Aplant
having (t) (enzyme, gene expression, etc.) oscillating regularly
and robustly in time (thus certainly driven by the circadian clock),
would solve the problemof starch buffer management (Dodd et al.,
2005). Or more generally, would eliminate some uncertainty from
the environment. Obviously, the hyperbolic shape of (t) is an ideal
shape. In reality, enzymatic activities or other biochemical reac-
tions are not likely to followperfectly a hyperbolic shape. However,
biological oscillators (for instance predatorprey model in Fig. 6)
havetransient hyperbolic features that allowthelinear andadapted
starchdegradation. Givena simple biological oscillator already nat-
urally comprises hyperbolic features, an organism equipped with
such simple clock already possesses most of the tools to manage
efcientlyits resources andavoidstarvationwithout evolvingother
mechanisms. Transient hyperbolic features in the oscillating circa-
dianclockcouldparadoxicallybeatool usedtoachievehomeostasis
in organisms. However, even if behavior of starch degradation can
be explained by the current study, we still need to understand how
starch accumulation can adapt to various photoperiods.
We assumed that maximization of sucrose export while avoid-
ing sucrose starvation in exporting organs results in sucrose
homeostasis during the light dark cycle. Therefore we chose to
minimize payoff function (12). This assumption may be a strong
constraint as it does not allow any natural small perturbation of
sucrose level around

S at all times. Homeostasis may not be the
purpose of the leaf cell but the result of another, less obvious behav-
ior. Actually, sucrose and starch dynamics ((1)(4)) may certainly
be more complicated. High sucrose and starch levels inhibit carbon
capture from photosynthesis (Sawada et al., 2001). On the other
hand sucrose is mandatory in the cell to supply energy (ATP) and
recyclecoenzymes (NAD) requiredfor maintenanceof basic cellular
functions the impairment of whichwouldindirectly decrease car-
bon capture efciency. With such tradeoff between too low or too
highsucrose concentrations bothslowing downmetabolism, a best
sucrose level may emerge as dictated by natural selection. Includ-
ingsuchmore realistic dynamics inour modelingwouldproduce an
emergent best sucrose level that could allowusing more ecological
payoff functions such as maximization of sucrose export from the
adult leaves over 24h, eventually proportional to the plant tness.
Such modeling based on the maximization of a high level payoff
function (biomass investment into seeds) could provide informa-
tion on how metabolic pathways have been sculpted by natural
selection in order to become more efcient.
Author contributions
FGF proposed the model and the method, wrote the pro-
grams, ran the simulations to generate the results and wrote the
manuscript. AS corrected the manuscript. Both authors took part
in the preparation of the manuscript.
Conict of interest
Authors declare no conict of interest.
Acknowledgements
This work was supported by Grants-in-Aid from the Min-
istry of Education, Culture, Sports, Science, and Technology, Japan
(22119009).
Appendix A. Appendix
Convergence of gradient descent
In Fig. 2b (or d) for photoperiods of 12h/12h and 10h/14h we
obtain sucrose level equal to

S during 24h. Furthermore, concen-
tration of starch at dawn is close to 0, hence all starch accumulated
during the day is used at night. Therefore results correspond to a
global minimumfor the considered photoperiods.
Eq. (13) is also useful to check convergence of the gradient
descent to a global minimum for photoperiods of 8h/16h and
6h/18h LD cycles. Setting t
dusk
at 6h (or 8h), now

S becomes S
max
whichis the theoretical maximumachievable average (or constant)
sucrose level during 24h. If the sucrose level observed on Fig. 2b
(or d) corresponds to S
max
, this means that the gradient descent
as converge to a minimumof the cost function that cannot be fur-
ther reduced, i.e. the level of sucrose cannot be further increased to
reduce the cost function over 24h. Performing the calculation for
thecasewherethephotoperiodis 6h/18hwehave, withanaverage
sucrose export over 24h equal to 0.01 as seen in Fig. 3c: S
max
=
gt
dusk

S/(24h +24 ) = 6.35 6/(24 0.79 +24 0.01) = 1.98.

Same calculationfor photoperiod8h/16hgives, withanaverage
sucrose export over 24h equal to 0.035 fromFig. 3c: S
max
=2.57.
Those two values correspond very well to the value of the con-
stant sucrose level observed on Fig. 2b and d for photoperiods of
8h/16h and 6h/18h.
Furthermore, the concentrations of starch at dawn for the cor-
responding starch proles are close to 0, thus for the same reason
as previously, starch buffer size cannot be further reduced. Hence,
all results correspond to the global minimum of the homeostasis
cost function. There may be other minima with rate functions ful-
lling homeostasis condition (10). However, they cannot be lower
than the minimumfound as explained - at best they are equal - and
are less in agreement with empirical data as, for example, starch
increase may diverge fromlinear.
References
Blasing, O.E., Gibon, Y., Gnther, M., Hhne, M., Morcuende, R., Osuna, D., Thimm,
O., Usadel, B., Scheible, W.-R., Stitt, M., 2005. Sugars and circadian regulation
make major contributions to the global regulation of diurnal gene expression in
Arabidopsis. Plant Cell 17, 32573281.
Dodd, A.N., Salathia, N., Hall, A., Kvei, E., Tth, R., Nagy, F., Hibberd, J.M., Millar,
A.J., Webb, A.A.R., 2005. Plant circadian clocks increase photosynthesis, growth,
competitive advantage and survival. Science 309, 630633.
Feugier, F.G., Satake, A., 2013. Dynamical feedback between circadian clock and
sucrose availability explains adaptive response of starch metabolismto various
photoperiods. Front. Plant Sci. 3, 305.
Gibon, Y., Blsing, O.E., Palacios-Rojas, N., Pankovic, D., Hendriks, J.H.M., Fisahn, J.,
Hhne, M., Gnther, M., Stitt, M., 2004. Adjustment of diurnal starch turnover to
short days: depletion of sugar during the night leads to a temporary inhibition
of carbohydrate utilization, accumulation of sugars and post-translational acti-
vation of ADP-glucose pyrophosphorylase in the following light period. Plant J.
39, 847862.
Graf, A., Schlereth, A., Stitt, M., Smith, A.M., 2010. Circadian control of carbohydrate
availability for growth in Arabidopsis plants at night. Proc. Natl. Acad. Sci. U. S.
A. 107, 94589463.
120 F.G. Feugier, A. Satake / Ecological Modelling 290 (2014) 110120
Graf, A., Smith, A.M., 2011. Starch and the clock: the dark side of plant productivity.
Trends Plant Sci. 16, 169175.
Lu, Y., Gehan, J.P., Sharkey, T.D., 2005. Daylength and circadian effects on starch
degradation and maltose metabolism. Plant Physiol. 138, 22802291.
Sawada, S., Kuninaka, M., Watanabe, K., Sato, A., Kawamura, H., Komine, K.,
Sakamoto, T., Kasai, M., 2001. The mechanism to suppress photosynthesis
through end-product inhibition in single-rooted soybean leaves during accli-
mation to CO
2
enrichment. Plant Cell Physiol. 42 (10), 10931102.
Smith, S.M., Fulton, D.C., Chia, T., Thorneycroft, D., Chapple, A., Dunstan, H., Hylton, C.,
Zeeman, S.C., Smith, A.M., 2004. Diurnal changes in the transcriptome encoding
enzymes of starch metabolism provide evidence for both transcriptional and
posttranscriptional regulation of starch metabolismin Arabidopsis leaves. Plant
Physiol. 136, 26872699.
Smith, A.M., Stitt, M., 2007. Coordination of carbon supply and plant growth. Plant
Cell Environ. 30, 11261149.
Stitt, M., Zeeman, S.C., 2012. Starch turnover: pathways, regulation and role in
growth. Curr. Opin. Plant Biol. 15, 282292.
Yazdanbakhsh, N., Sulpice, R., Graf, A., Stitt, M., Fisahn, J., 2011. Circadian control of
root elongation and C partitioning in Arabidopsis thaliana. Plant Cell Environ.
34, 877894.