Characterizing moisture uptake and plasticization effects of water

on amorphous amylose starch models using molecular dynamics

methods

Jeffrey M. Sanders, Mayank Misra, Thomas J. L. Mustard, David J. Giesen,

Teng Zhang, John Shelley, Mathew D. Halls

Schrödinger Inc., New York, NY 10036 USA

Abstract
Dynamics and thermophysical properties of amorphous starch were explored using molecular
dynamics (MD) simulations. Using the OPLS3e force field, simulations of short amylose chains
in water were performed to determine force field accuracy. Using well-tempered metadynamics,
a free energy map of the two glycosidic angles of an amylose molecule was constructed and
compared with other modern force fields. Good agreement of torsional sampling for both
solvated and amorphous amylose starch models was observed. Using combined grand
canonical Monte Carlo (GCMC)/MD simulations, a moisture sorption isotherm curve is predicted
along with temperature dependence. Concentration-dependent activation energies for water
transport agree quantitatively with previous experiments. Finally, the plasticization effect of
moisture content on amorphous starch was investigated. Predicted glass transition temperature
(T​g​) depression as a function of moisture content is in line with experimental trends. Further, our
calculations provide a value for the dry T​g​ for amorphous starch, a value which no experimental
value is available.

1. Introduction
Starch is one of the most common food sources in modern human diets. This polymeric
carbohydrate is synthesized and stored in plants as granules consisting of two main
components: amylose and amylopectin ​(Ai and Jane 2018)​. These two biopolymers are
composed of D-glucose repeat units and, depending on their respective molecular weights, can
have significantly branched structures ​(Delcour et al. 2010)​. Amylose and amylopectin, along
with lipids and protein molecules, organize into a granular structure as alternating regions of
amorphous and crystalline rings or “shells” ​(Bertoft 2017)​. Depending on the starch plant
source, the percentage of amylose and amylopectin can vary dramatically. Physical and
chemical modifications, both enzymatic and non-enzymatic, are of significant interest as starch
components are increasingly used for food and pharmaceutical nano-encapsulation as well as
thermoplastic polymer applications ​(Zhu 2017)​ ​(Mohammadi Nafchi et al. 2013)​ ​(Singh, Kaur,
and McCarthy 2007)​.

1
 Numerous studies have focused on understanding the microstructure and the underlying
nanostructure of starch ​(Kuang et al. 2017; Gallant, Bouchet, and Baldwin 1997; Nessi et al.
2018; Vamadevan and Bertoft 2020; Ma and Boye 2018; Bertoft et al. 2016; Li and Gilbert 2018;
Biliaderis 2009; Pérez and Bertoft 2010; Cheetham and Tao 1998b)​. Despite significant
advances, the complex morphology of an individual blocklet precludes high resolution, atomic
level structural characterization. One avenue to explore, in the lack of experimental structural
information, is atomistic simulation ​(Billinge and Levin 2007)​. If the chemical constituents of a
material are known, morphological and physical properties have been successfully predicted
using simulation for a wide range of materials ​(Steinhauser and Hiermaier 2009)​. Molecular
simulations of single amylose molecules have revealed key physical aspects of lipid binding,
while more rigorous studies focusing on disaccharide dynamics have investigated the reliability
of modern force fields in simulating conformational dynamics ​(López, de Vries, and Marrink
2012; Perić-Hassler et al. 2010; Galvelis, Re, and Sugita 2017; Kuttel and Naidoo 2005; Y.-C.
Wang et al. 2014; Turupcu and Oostenbrink 2017; Momany, Willett, and Schnupf 2009; Silva et
al. 2018; Mishra et al. 2014; Kilburn et al. 2005; Cheetham, Dasgupta, and Ball 2003; Naidoo
and Kuttel, n.d.)​. While these studies have greatly enhanced our understanding of starch
nanostructure at the single molecule level in solution, they do not address morphological or
thermophysical properties of amorphous starch. To address this, the authors performed
molecular dynamics (MD) simulations of amylose in amorphous and solvated forms. Force field
accuracy is assessed using a disaccharide maltose and a short amylose chain model by
standard MD and enhanced sampling methods. As hydrophobicity is a driving force for the
starch granular organization, the authors explored the temperature-dependent effects of
moisture exposure via combined grand canonical monte carlo (GCMC)/MD ​(Vlachy and Haymet
1986; Smit and Maesen 2008; Fuchs and Cheetham 2001)​. This method is able to not only
predict water uptake at a given temperature and relative humidity condition, but also allows for
tracking of water transport in the starch polymer matrix.

2. Methods

All models and simulations were performed using the Schrödinger Materials Science Suite,
version 2020-1 ​(​Materials Science Suite​ 2020)​.

2.1 Model Construction

Molecular models for maltose and amylose polymers were built from maltose repeat
units using (𝝓,𝜳) values consistent with an extended helix conformation. The maltose molecule
was then solvated using the system builder with 611 water molecules. The same process was
applied to the degree of polymerization 12 (DP12) amylose polymer chain by adding 4504 water
molecules. For the amorphous DP12 and degree of polymerization 31 amylose (DP31) systems,
polymer chains in an extended helix conformation (Figure 1), (𝝓,𝜳) = (62.6°, 81.9°) were built in
a box with low density using the Disordered System Builder tool with a vdW scaling factor of 0.5,

2
snapped to grid configuration with steric packing and rotational disorder. Due to the amylose
molecule’s helical nature, a self-avoiding walk method to generate a higher density initial
configuration was not possible as sampling of the (𝝓,𝜳) angles would potentially favor random
coil conformations. Five independent amorphous systems were built for DP12 and DP31 using
different random seeds. The OPLS3e force field was used for all sugars and the SPC model for
water ​(“Roos et al. - 2019 - OPLS3e Extending Force Field Coverage for Drug-Li.Pdf,” n.d.;
“Berendsen et al. - 1981 - Interaction Models for Water in Relation to Protei.Pdf,” n.d.)​.

2.2 MD equilibration
All MD simulations were performed using Desmond ​(​Schrödinger Release 2020-1:
Desmond Molecular Dynamics System​ 2020)​. For the solvated maltose and DP12 system, the
following seven stage equilibration protocol was used: 1) Canonical (NVT) ensemble with
Brownian dynamics for 12 ps at 10 K with 1.0 fs time step and restraints on all solute
non-hydrogen atoms, 2) NVT ensemble using a Langevin thermostat for 12 ps at 10 K with a 2.0
fs time step and restraints on all non-hydrogen solute atoms, 3) Isothermal-isobaric (NPT)
ensemble using a Langevin thermostat and Langevin barostat for 12 ps at 10 K and 1.01325 bar
with a 2.0 fs time step and restraints on all non-hydrogen solute atoms, 4) NPT ensemble using
a Langevin thermostat and Langevin barostat for 12 ps at 300 K and 1.01325 bar with a 2.0 fs
time step and restraints on all non-hydrogen solute atoms, and 5) NPT ensemble using a
Langevin thermostat and Langevin barostat for 24 ps at 300 K and 1.01325 bar with a 2.0 fs
time step. Following equilibration, each system was simulated for 1.0 µs using a NPT ensemble
with a Nose-Hoover thermostat and a Martyna-Tobias-Klein barostat method with a 2.0 fs time
step interval at 300 K and 1.01325 bar.
For the amorphous DP12 and DP31 systems, the following relaxation protocol was used:
1) NVT ensemble with Brownian dynamics for 20 ps at 10 K with a 1.0 fs time step, 2) NPT
ensemble with Brownian dynamics for 20 ps at 100 K with a 2.0 fs time step, 3) NPT ensemble
with a Nose-Hoover thermostat and a Martyna-Tobias-Klein barostat method for 100 ps at 100K
and 1.01325 bar with a 2.0 fs time step, and 4) NPT ensemble with a Nose-Hoover thermostat
and a Martyna-Tobias-Klein barostat method for 20.0 ns at 300 K and 1.01325 bar with a 2.0 fs
time step. Following initial equilibration, a short set of cooling simulations were performed by
heating each system up to 700 K then cooling down to 300 K with a step size of 20 K and
simulation time of 10 ns at each step (rate of 500 ps/K). Each stage used a NPT ensemble with
a Nose-Hoover thermostat and a Martyna-Tobias-Klein barostat method at 1.01325 bar with a
2.0 fs time step. An additional 50 ns was run at 300 K and 1.01325 bar to sample configurations
for measuring the torsion angles and the radius of gyration (R​g​), as well as to produce the
starting structure for T​g​ and penetrant loading calculations.

2.3 Well-tempered Metadynamics calculations

Well-tempered metadynamics simulations were performed using Desmond ​(​Schrödinger
Release 2020-1: Desmond Molecular Dynamics System​ 2020)​ to characterize the
conformational preferences of maltose. The metadynamics adaptive biasing technique allows
for sampling of rare events and estimation of the free energy landscape of atomistic systems

3
(Bussi and Branduardi 2015)​. The two collective variables (CV) were defined by the principle
degrees of freedom for maltose: the 𝝓 and 𝜳 glycosidic torsion angles defined using the
standard convention: 𝝓 = O​5​-C​1​-O​1​-C​4​’ and 𝜳 = C​1​-O​1​-C​4​’-C​3​’​ ​(Quigley, Sarko, and
Marchessault 1970)​ (Figure 1). The biasing potential in these metadynamics simulations was:

[ ]
t/τ G N CV
(sα −sα (iτ G ))2
V (s, t) = ∑ wG × exp − ∑ 2σ 2α
i=1 α=1

where 𝝈 is the gaussian width, w​G is​ the barrier height, N​CV​ is the number of collective variables,
and 𝝉​G​ is the deposition time. For the case of well-tempered metadynamics, the gaussian
heights are resized to take into account the accumulated bias potential:

w j = w × exp ( V (s,tj )
k B ΔT )
Where k​B​ is Boltzmann’s constant, w is the initial bias deposition rate, and the sampling
temperature ΔT was 300 K ​(Bussi and Laio 2020)​. The Gaussian width was set to 0.05 Å, the
Gaussian height set to 0.03 kcal/mol, and the deposition time was 0.09 ps. The simulation was
run with a NPT ensemble using a Nose-Hoover thermostat and a Martyna-Tobias-Klein barostat
for 200 ns at 300 K and 1.01325 bar with a 2.0 fs time step. For the DP12 system, the torsion
angles between the 5 and 6 sugar repeat units were selected as the collective variables (CVs).
Trajectory snapshots were recorded every 100 ps. 2-D free energy surfaces were constructed
using the metadynamics analysis panel. Graphs for publication were generated using Originlab
(​Origin​ (version Version 2020), n.d.)​.

2.4 Grand canonical Monte Carlo/Molecular Dynamics penetrant loading

To simulate moisture uptake in the amorphous amylose starch, we performed penetrant
loading simulations. Penetrant loading simulations cycle through two distinct simulation stages a
large number of times as defined by the overall simulation time. The first stage is a grand
canonical Monte Carlo simulation (GCMC) during which water molecules are added or removed
in a manner that is rigorously consistent with the applied chemical potential, µ, to sample states
from the grand canonical (µVT) ensemble. A detailed derivation of the statistical mechanical
basis of GCMC methods can be found in Frenkel and Smit​(Frenkel and Smit 2002)​. The
second stage is a NPT molecular dynamics simulation to efficiently sample molecular
rearrangements and conformations within the entire system. For each iteration, 5.0 ps of MD
simulation are performed followed by 5000 GCMC steps. To generate a water absorption
isotherm, we performed simulations at the following relative humidities (RH, µ is directly related
to RH): 5, 10, 25, 75, and 100 %. These simulations were run for 1.0 µs with a NPT ensemble at
300 K and 1.01325 bar with the SPC water model. To simulate the temperature dependence of
water uptake, penetrant loading simulations were run over a temperature range of 300-400 K
with 10 K increments at 100% RH.

4
2.5 Water transport in amorphous starch
We calculated water diffusion at 11 and 23 % H​2​O wt. The studies at 11 and 23 % wt
studies employed the final structures from the 300 K and 400 K penetrant loading simulations at
100 % RH, respectively. Each system was equilibrated in the NPT ensemble at: 300, 325, 350,
375 and 400 K at 1.01325 bar for 5.0 ns. Following the equilibration, 50 ns micro-canonical
(NVE) simulations were performed from which the mean squared displacement (MSD):

2
M SD (t) = 〈||ri (t) − ri (0)|| 〉

was calculated for each water molecule based upon the center of mass of the water molecules.
The Einstein relation was used to determine diffusion coefficients using the following equation:

M SD(t)
2Dt = 3

by fitting the linear region of the MSD versus time graphs. For each simulation, the linear region
is estimated to be between 10 and 40 ns. To calculate the temperature dependence, the
average diffusion coefficient was used to construct an arrhenius plot. The activation energy (Ea)
was then calculated by linear regression from a form of the Arrhenius equation:

Ea
ln(D) = ln(D0 ) − RT

Where R is the universal gas constant and D​0​ is formally the diffusion at infinite temperature
which is treated as a fitting parameter.

2.6 Glass transition temperature prediction

To predict the glass transition temperature, simulated annealing simulations are
performed at a constant cooling rate and the density of the system is recorded during each MD
stage. For the dry DP12 and DP31 systems, the annealing process started at 800 K with 5.0 ns
simulations in the NPT ensemble with the pressure set to 1.01325 bar and simulations were
performed every 15 K increments until 100 K was reached. As high temperature simulations
increased energy fluctuations, the time step was reduced to 1.0 fs for simulations above 700 K
before switching back to the default 2.0 fs below 700 K. Following job completion, density
versus temperature data was fit to a hyperbola to estimate the T​g​. The hyperbola fitting
procedure is automated, as opposed to the traditional bi-linear fit which requires manually fitting
the linear regions of the specific volume versus temperature for the high and lower temperature
regimes ​(Patrone et al. 2016)​. For the water containing samples, the cooling rate used for the
dry systems resulted in large variances in the density values so the MD simulation time at each
temperature interval was increased to 10 ns to obtain more accurate density values. The
hyperbola fitting process was applied to all density data from 600 K to 100 K, due to decreased

5
T​g values
​ and the possibility of water forming bubbles of vapor in the system above 373 K. For
the dry samples, uncertainty quantification was performed due to inert noise at high
temperatures. Data pooling analysis of five independent trials was performed and the T​g
reported includes the between-simulation uncertainty. For details on this method, the reader is
referred to Patrone et al.​(Patrone et al. 2016)​. For the wet samples, the arithmetic mean is used
to determine T​g​ for each system. We also repeated this process for the dry DP31 system using
the 10 ns simulation time at each step to compare equal cooling rates.

A) B)

Figure 1.​ A) Glycosidic torsion angles of maltose using the standard numbering convention. B)
Solvated DP12 amylose structure.

3. Results and Discussion

3.1 Assessment of amylose dynamics in solvent with the OPLS3e force field
Carbohydrate force fields have been an area of active research due to their inherently
complex underlying potential energy surfaces and diverse applications. As the OPLS3e force
field has been shown to reproduce structural and thermodynamic properties ​(Roos et al. 2019;
Keränen et al. 2017; Pérez-Benito et al. 2018; Sindhikara and Borrelli 2018)​, the authors sought
to determine the reliability of OPLS3e for amylose, a carbohydrate bio-polymer. Previous
studies have shown that common force fields including CHARMM36, GYLCAM06,
GROMOS53A6​CARBO_R​ and OPLS-AA are capable of predicting dynamical behavior of short
amylose chains in water or other solvents ​(López, de Vries, and Marrink 2012; Perić-Hassler et
al. 2010; Cheng et al. 2018; Pereira et al. 2006; Hatcher et al. 2011; Ott and Meyer 1996;
Winger, Christen, and van Gunsteren 2009)​. A recent study​52​ reported a thorough analysis of
the aforementioned force fields and their influence on conformational complexity of DP12
amylose in water during microsecond simulations. Using quantitative geometric descriptors and
Markov state models,the authors highlighted the conformational heterogeneity among the force

6
fields studied. In particular, the a(1→4) glycosidic torsion rotation of 𝝓 through 180° was used to
describe helix to coil transitions from syn to anti conformations.
The backbone glycoside a(1→4) linkage is common to other polysaccharides, including
amylopectin and glycogen in addition to amylose ​(Varki and Chrispeels 1999)​. NMR studies
have indicated that this bond is conformationally restricted in glucans and is generally
characterized by Ramachandran maps of the (𝝓, 𝝍) torsion angles ​(Cheetham, Dasgupta, and
Ball 2003)​. Adiabatic Ramachandran maps have been used to predict simple disaccharide
models, but fail to include solvent and entropic contributions. Accounting for solvent effects is
important as it is known that water has a dramatic effect on amylose, including lowering the
glass transition temperature of amorphous maltose and amylose ​(Biliaderis 2009)​. Free energy
calculations can be used to characterize the free energy landscape of the carbohydrates in
solutions, and several studies have shown to accurately characterize both thermodynamic and
dynamic properties of di-saccharides when compared to experimental data 22,27​ ​ . In these
studies, simulations were able to confirm glycoside (1→4) linkages are conformationally
restricted as previously reported by NMR studies.
With the importance of glycosidic torsion angles on judging force field accuracy, the
authors sought to validate the OPLS3e carbohydrate force field parameters using free energy
simulations to predict the free energy landscape of the (𝝓, 𝝍) Ramachandran map.
Metadynamics, an enhanced sampling molecular dynamics method, is able to accelerate
conformational transitions of metastable states by filling in the free energy minima using a well
controlled biasing potential ​(“Bussi and Laio - 2020 - Using Metadynamics to Explore Complex
Free-Energy .Pdf,” n.d.)​. The biasing is determined by properly defining low-dimensional
descriptors, also called collective variables. If the collective variables are appropriately chosen,
they will vary between both the relevant metastable states along with the transition states
between them. Provided the metastable states in question are well separated and defined in
phase space, the free energy as a function of the CV can be constructed. If an accurate
estimator of the free energy between metastable states is required, well-tempered
metadynamics can be employed to ensure that the free energy barriers are crossed several
times during the simulation.
Starting with maltose, the disaccharide building block of amylose, both standard MD and
well-tempered metadynamics simulations were performed in water for 1.0 µs at 300 K and
1.01325 bar. The adiabatic (𝝓, 𝝍) map (Figure 2A) shows the syn conformation is heavily
favored and centered around (70°,90°), while the syn to anti transition occurs rarely for maltose
with the anti conformation sampled ~ 2.8% of total simulation time. Koneru et al. ​(Koneru, Zhu,
and Mondal 2019)​ determined that the local minima for the syn conformation for all four force
fields is centered around (70 - 90°, 80-100°). Only three force fields (CHARMM36, OPLS-AA,
and GLYCAM06) showed minor populations of the anti conformation centered around (70-80°,
-70-​-​50°), while the GROMOS53A6​CARBO_R never
​ sampled this region during the course of a
multi-microsecond simulation. We repeated the analysis for the DP12 molecule and found the
minima for the syn conformation did not change, though increased sampling around the minima
occurs, due to the increased number of torsion angles being measured (Figure 2B).

7
 A) B)

C) D)

Figure 2. Adiabatic torsion maps of amylose systems from MD simulation.​ A) maltose in

water, B) DP12 amylose in water, C) DP12 amorphous anhydrous amylose, and D) DP31
amorphous anhydrous amylose. All simulations were performed in a NPT ensemble at 300
K and 1.01325 bar.

While the adiabatic torsion maps are useful to judge conformational stability of the force
field, a complete map requires an enhanced sampling approach using modern software and
hardware. We used well-tempered metadynamics calculations with (𝝓, 𝝍) as CVs for this
purpose. In the case of maltose, only one set of torsion angles exists making the CV selection
trivial. For the case of the DP12 amylose, we chose the central torsions between repeat units 5
and 6 as the CVs. The resulting free energy maps had similar features, with the DP12 map
showing a broadening of the well along the 𝝓 coordinate (Figure 3). The wide distribution from
the DP12 system could be a result of increased flexibility by adjacent sugar units that are able to
stabilize hydrogen bonds with water. The free energy difference (ΔG) between the syn and anti
conformation for maltose and DP12 were 1.6 and 1.4 kcal/mol , respectively. These values are
slightly lower than those estimated with local elevation umbrella sampling, where the range of
​
ΔG was 1.91 to 5.2 for kcal/mol​ for different n-linked disaccharides ​(Perić-Hassler et al. 2010)​.
Another study, focused on maltose, estimated the syn to anti transition ΔG as ~ 3 kcal/mol using

8
adaptive umbrella sampling ​(Kuttel and Naidoo 2005)​. One distinction in metadynamics is
sampling of the lower energy states occurs more frequently, whereas most other adaptive
biasing methods will equally sample all of the CV space. Metadynamics also has the advantage
of more efficiently sampling CV space with lower statistical uncertainty ​(Bochicchio et al. 2015)​.
To check for convergence of the metadynamics simulations, we checked both the energy hill
heights as a function of time and changes in the free energy surface which we will refer to as
the distance of free energies, as measured by RMS difference between aligned free energy
surfaces for consecutive intervals in the simulation as a function of time. For the hill height
energies, spikes at the end of the simulation imply the system is exploring an orthogonal degree
of freedom. In both systems, we note the hill height energies are constant throughout each
simulation (data not shown). For the distance in free energies, we noticed a plateau towards the
end of the simulation, suggesting each system is jumping between the syn and anti
conformations and convergence has been reached.

A) B)

Figure 3. Free energy contour maps of glycosidic torsion angles from well-tempered
metadynamics simulations. ​A) maltose in water and B) DP12 amylose in water. For the DP12
system, (𝝓,𝜳) CVs were sampled between the 5​th​ and 6​th​ repeat units. The simulations were run
for 200 ns with a NPT ensemble at 300 K and 1.01325 bar.

Another metric used to judge the quality of carbohydrate force fields is the radius of
gyration (R​g​) of the polymer in solution. For the four aforementioned force fields, the R​g ranges
​
between 11 and 15 Å ​(Koneru, Zhu, and Mondal 2019)​. CHARMM36 and GLYCAM06 do show
coiled conformations with R​g = ​ 7 - 10 Å, suggesting collapsed conformations are favored for
some force fields. In our simulations, we found the R​g​ to have a Gaussian distribution centered
around 13 Å and no substantial secondary peak corresponding to a collapse during the course
of the 1.0 µs simulation. Both X-ray and NMR studies have reported the syn conformation as the
dominant one, suggesting coiled conformations are short lived ​(Rundle and Edwards 1943;
Cheetham and Tao 1998a)​. This led us to conclude that the OPLS3e force field is accurate at
reproducing structural properties observed in both experiments and across different force fields
in simulation.

9
3.2 Amorphous amylose properties
Native starch is composed of semi-crystalline structures of both amylose and
amylopectin. Large amorphous channels found in granules constitute the major portion of the
granular material and contain the majority of amylose in starch ​(Brown and French 1977)​.
Amorphous amylose regions have a large impact on the physicochemical properties of starch as
they are less dense than the lamellar amylopectin regions, are susceptible to chemical and
enzymatic modification and small molecules (MW < 1000) that are water-soluble can diffuse
through the amylose matrix ​(Biliaderis 2009)​. When starch granules are immersed in water, the
current hypothesis is swelling occurs due to hydration and swelling of the amorphous regions
(Nara, Mori, and Komiya 1978)​. Amylose molecules present in the amorphous channel can be
either linear or branched and the average length of the polymer chains depends on the starch
source ​(Ai and Jane 2018)​. Debranching of amylose molecules can be performed enzymatically,
resulting in linear chains that are capable of forming inclusion complexes with lipids or soluble
polymer chains depending on the degree of polymerization (DP) ​(Guzmán-Maldonado,
Paredes-López, and Biliaderis 1995)​. Arijaje and Wang​(Arijaje and Wang 2016)​ reported the
optimum degree of polymerization of debranched potato starches to be in the range of 20 to 43
for complex formation with stearic acid, while DP6-17 were too short to form an inclusion
complex. This suggests that the standard DP12 amylose molecule model is too short for
modeling amorphous amylose for lipid complex formulations. Short chain amylose and
debranched amylopectin have also been shown to reduce starch viscosity in Q11 rice ​(Zhang et
al. 2013)​. Debranched amylopectin from Q11, contains a majority of chains with DP22-54 and
few DP11-21, have a low paste viscosity which has significant impact on starch in both food and
non-food applications.
With this information in mind, we built five amorphous starch systems with amylose
DP31 molecules with extended helical conformations. Each system was then subject to an initial
equilibration then simulated annealing to 300 K. To determine how well parameterized the force
field is for amorphous amylose, density values were recorded during annealing. The
experimental density of native starch is 1.5 g/cm​3 ​and accounts for both the semi-crystalline
amylopectin and amorphous amylose ​(Nara, Mori, and Komiya 1978)​, while the experimental
density of amorphous amylose has been measured recently at 293 K to be 1.45 +/- 0.003 g/cm​3
(Özeren et al. 2020)​. Calculated density for the five replicates was 1.43 +/- 0.001 g/cm​3​ at 295
K, < 1.5% below the experimental density. et al. ​(Özeren et al. 2020)​ also reported predicted
densities for amorphous amylose from MD simulation as 1.42 g/cm​3​, in close agreement with
our observations.
We also compared the adiabatic (𝝓, 𝝍) map of the DP31 amorphous system (Figure 2D)
with the free energy map of the solvated DP12. Sampling increased around the wells for the
amorphous state, but no new metastable states were found. The syn conformation is still heavily
favored, though the distribution has broadened compared to both maltose and the solvated
systems. This data suggests amylose retains an extended helix-like structure, even in the
glassy state. ​(Kuttel and Naidoo 2005)​ made a similar observation, based on free energy
surfaces generated with umbrella sampling simulations. They concluded that solid state forms of

10
amylose would contain both syn conformations and anti regions to relieve steric strain. Our free
energy maps from solution, along with adiabatic maps of DP31 amorphous amylose are in
agreement with this hypothesis. Additionally, helical conformations in the amorphous would
promote binding to lipids, without the need for extensive conformational adjustments and would
be advantageous for formulations where inclusion complexes of lipids and amylose in the
anhydrous form are necessary.

A) B)

Figure 4. T​g​ prediction of anhydrous amorphous DP31​. A) Hyperbola fit of density as a

function of temperature. Simulated annealing simulations with a cooling rate of 0.34 ns/K. Data
points represent the average of five independent simulations. B) T​g estimation
​ from data pooling
analysis to estimate between simulation uncertainty. Errors bars represent 3𝛔.

3.3 Moisture uptake in amorphous amylose

Key to starch functionality is the interaction of water and granules. When heated in
excess water, starch granules swell and undergo the process of gelatinization ​(Biliaderis 2009)​.
During gelatinization, the molecular order of semi-crystalline blocklets and amorphous amylose
channels collapse and irresistible water uptake occurs along with solubilization and increased
viscosity. Depending on the botanical starch source, the swelling behavior can vary in terms of
temperature response and gel volumes ​(Srichuwong et al. 2005)​. A common feature that does
occur during the heating process is leaching of amylose from the amorphous phase and
subsequent aggregation which is thought to be responsible for the viscoelastic behavior of
pastes or gels.
(Gidley and Bulpin 1989)​ reported that chain length can impact the phase behavior with
shorter chain lengths (DP < 110) forming aggregates while longer chains (110 < DP < 1000)
drive gelation. Numerous experimental studies have sought to understand the complex
state/phase transitions and the effect of moisture ​(Ai and Jane 2018)​. As amorphous starch is
highly attractive for applications in thermoplastics and food encapsulation, understanding the
interactions of moisture and it’s plasticization effects are paramount.
As amorphous amylose channels are thought to be the entry point for water in starch
granules, prediction of adsorption phase equilibria could provide insight into the early stages of

11
swelling and plasticization effects. A common simulation technique for predicting absorption in
condensed systems is the grand canonical monte carlo (GCMC) method. GCMC methods can
be used to predict absorption thermodynamics, of which the details can be found in simulation
textbooks ​(Frenkel and Smit 2002)​. Using the DP31 amorphous amylose system, GCMC/MD
simulations were employed to investigate water absorption as a function of RH (Figure 5), 1.0
µs simulations were run under different RH conditions and the resulting water sorption isotherm
plot was generated. Simulations this long were necessary for water absorption to reach
equilibrium values, which is still far shorter than experimental time which occurs over several
days ​(S. Wang et al. 2015)​. The general features of a type II adsorption isotherm were present:
a sigmoid shaped curve with three regions corresponding to significant uptake at low partial
vapor pressure, small absorption at intermediate vapor pressure and high uptake at elevated
partial pressures. The maximum moisture content of ~ 11 % was observed for 100 RH at 300 K.
Historically, moisture uptake studies have used starches containing semi-crystalline amylopectin
while few studies have focused on completely amorphous amylose starch. Mali et al.​(Mali et al.
2005)​ measured moisture uptake on untreated cassava starch that matches our predictions,
despite cassava having a low amylose content (< 30 %). Song et al.​(Song et al. 2017)
measured moisture content for fully amorphous potato starch with a maximum uptake of 35 %.
Our values are off by a factor of two, but this may be due to the preparation method as a large
amount of ethanol was present during processing. Johnson and Mauer​(Johnson and Mauer
2019)​ measured moisture uptake of high amylose corn starch lyophiles, observing a maximum
of 29 % wt at 27 hours at 95 % RH and only 13.7 % at % 90 RH. The discrepancy can be
attributed not only to the length of the simulations, but also the lack of crystal amylopectin and
other material present in native starch samples. The temperature dependent moisture uptake
trend is in line with experimental observations made for different starch samples ​(Shotton and
Harb 1965)​.

A) B)

Figure 5. Moisture uptake in DP31 amorphous amylose using combined GCMC/MD

simulations. ​A) Moisture adsorption isotherm and B) temperature dependent water absorption.
All data points represent the average of five independent simulations. Each simulation was run
for 1.0 µs.

12
 Next, we sought to investigate the behavior of water diffusion in amorphous starch
models. Ingress of water into thermoplastic starch materials is of high interest due to it’s
plasticization effects ​(Mohammadi Nafchi et al. 2013)​. Transport properties of water, particularly
diffusion coefficients, are often difficult to measure experimentally as they require continuous
monitoring of sorption changes over a period of time. Magnetic resonance imaging (MRI)
techniques can be used to visualize water ingress and diffusion through polymer materials, and
the instantaneous diffusion coefficients can be determined ​(Griffiths and Stilbs 2002; Levine and
Slade 1991; Yang et al. 2020)​. Russo et al.​(Russo et al. 2007)​ reported diffusion coefficients
determined by MRI both as a function of temperature and moisture content for high amylose
starch blends. Using a Fickian diffusion model, they were able to generate Arrhenius plots for
the activation energy of water diffusion under different moisture concentrations. Activation
energy ( E a ) can be related to the limited chain flexibility and plasticiation of the polymer matrix.
MD simulations are capable of tracking the displacement of all atoms, making this an attractive
alternative for measuring diffusion of solvent and small molecules through polymer matrices.
Taking the 300 K, 100 % RH amylose water system (11 % H​2​O wt), we measured the
temperature dependence of the water diffusion coefficient. We then constructed Arrhenius plots
to determine E a (Figure 6). We repeated this for the highest moisture content system (23 %)
and found E​a​ values of 27.8 and 41.1 kJ/mol for the 11 % and 23 % moisture content systems,
respectively. These values agree with both Russo et al.’s values (18 - 55 kJ/mol) and those
reported for amorphous starch tablets (26 - 42 kJ/mol) ​(Thérien-Aubin et al. 2005)​. The strong
correlation between experiments suggests that water diffusion is primarily through amorphous
amylose regions of granules, and that time scales are fast enough to be captured with atomistic
simulation methods. This is in line with Flory’s theory which suggests that anything that is not
part of the “crystal” in semi-crystalline molecules will be driven out of it, or the crystalline region
will become amorphous. Which one of the above happens is dependent on the rate of crystal
growth ​(Flory 1989)​.

13
 A) B)

C)

Figure 6. Water transport properties in DP31 amorphous amylose. ​A) Arrehenius plot of
water diffusion for 11% wt moisture content and B) 23% wt moisture content. C) Mean Square
Displacement (MSD) for a 11% wt moisture content over five different temperatures. Diffusion
simulations were run for 50 ns using a NVE ensemble pre-equilibrated at the appropriate
temperatures.

3.4 Predicting plasticization as a function of moisture content

One of the key properties of any polymeric material is the glass transition temperature
(T​g​). A thorough understanding of the thermophysical properties of native and amorphous
granules is challenging to characterize experimentally, due to the different microstructural
features of the amorphous and crystalline regions. It is known that T​g​ has a strong dependence
on moisture content, as water acts as a plasticizer ​(Biliaderis 2009)​. Several studies have
shown increasing moisture content depressed the T​g​ of a starch sample significantly within 0-24
% water content ​(Biliaderis 2009; Georget, Parker, and Smith 1995; Zobel 1988)​. Above 30 %,
the T​g​ remains constant as the aqueous phase separates from the granules. For the case of dry
starch, experimental methods have been unable to determine T​g​ as the polymers thermally
degrade before reaching the transition point ​(Trommsdorff and Tomka 1995)​. The dry T​g​ has

14
been estimated by measuring over a range of moisture contents and extrapolating to zero, with
a range of 498-508 K 14​ ​ .
T​g​ prediction by MD simulation is an attractive method, and large scale simulations of
different polymers have shown good agreement with experiment ​(Afzal et al. 2020)​. One key
difference with MD simulation is the cooling rate is often much faster than experiment. If a T​g
prediction is being compared with an individual experiment using a controlled cooling/heating
rate, then the Williams-Landel-Ferry (WLF) empirical equation can be used. Details of the WLF
equation and its applicability have been reviewed extensively ​(Williams, Landel, and Ferry
1955)​. In the absence of consistent experimental heating/cooling data, empirical methods based
on large scale prediction can be used. Afzal et al.​(Afzal et al. 2020)​ studied 315 polymers and
found that a cooling rate of 0.25 ns/K led to T​g​ estimates that were systematically high by 21 K.
Using a cooling rate of 0.67 ns/K, simulated annealing simulations were performed for all
moisture contents generated at different temperature and RH conditions along with the dry
amorphous DP31 system. To quantify the plasticization effects of moisture content at 100 % RH
at 300 K, we compared the T​g​ and found over 100 K depression in the 11 % moisture containing
system. Our dry T​g​ prediction was 529 K, 20-30 degrees above experiment which agrees well
with the Afzal et al. (Figure 4). For wet systems, we observe the same trend of plasticization as
shown experimentally, with a higher T​g​ at each value when compared to experiment ​(Biliaderis
2009)​ (Figure 7).

A) B)

Figure 7. Plasticization effects of water on amorphous DP31 amylose.​ A) Density versus

temperature plots of dry and 11 % moisture content DP31 using a 0.67 ns/K cooling rate. B)
T​g​ as a function of moisture content. Blue data points represent the average of five
independent systems. Orange data points are experimental values reported in ​(Ai and Jane
2018)​.

4. Conclusion
Atomistic based simulations have greatly improved our understanding of condensed
matter, and have led to rapid prediction of key physical properties. MD simulations of starch

15
have provided detail regarding the dynamics of individual amylose molecules in solution using
modern force fields. The OPLS3e force field, a force field covering a large portion of chemical
space, can be used to predict dynamical properties of amorphous starch polymers in addition to
solution behavior. To extend the utility further, we show that GCMC/MD methods can be used to
generate sorption binding isotherms of water in amorphous amylose. The temperature and
relative humidity conditions will affect the equilibrium amount of water in the amylose matrix.
The plasticization effect of water on amorphous amylose starch can be calculated yielding
values that are in good agreement with experiment, allowing prediction of the previously
inaccessible dry T​g​ of amorphous starch. Analysis of water diffusion in amylose matrices follows
Fickian behavior and the activation energy is dependent on moisture content, as previously
observed experimentally. Further studies can be performed to study in-depth water-amylose
interactions and also the effects of other commonly used ingredients in starch formulations.
Emulsifiers, used for processing starch under various conditions, are known to bind to amylose
molecules but this behavior has only been observed for single molecule host-guest binding in
dilute solutions via MD simulation. Future work can be used to understand the morphology of
complex amorphous and/or semi-crystalline starch formulations and how that morphology
affects transport and thermomechanical properties.

References

Afzal, Mohammad Atif Faiz, Andrea Browning, Alexander Goldberg, MathewD. Halls, Jacob L.
Gavartin, Tsuguo Morisato, ThomasF. Hughes, David J. Giesen, and Joseph E. Goose.
2020. “High-Throughput Molecular Dynamics Simulations and Validation of
Thermophysical Properties of Polymers.” Preprint.
https://doi.org/10.26434/chemrxiv.12250229.v1.
Ai, Yongfeng, and Jay-lin Jane. 2018. “Understanding Starch Structure and Functionality.” In
Starch in Food,​ 151–78. Elsevier. https://doi.org/10.1016/B978-0-08-100868-3.00003-2.
Arijaje, Emily Oluwaseun, and Ya-Jane Wang. 2016. “Effects of Enzymatic Modifications and
Botanical Source on Starch-Stearic Acid Complex Formation: Starch and Stearic Acid
Complexation.” ​Starch - Stärke​ 68 (7–8): 700–708.
https://doi.org/10.1002/star.201500249.
“Berendsen et al. - 1981 - Interaction Models for Water in Relation to Protei.Pdf.” n.d.
Bertoft, Eric. 2017. “Understanding Starch Structure: Recent Progress.” ​Agronomy​ 7 (3): 56.
https://doi.org/10.3390/agronomy7030056.
Bertoft, Eric, George A. Annor, Xinyu Shen, Pinthip Rumpagaporn, Koushik Seetharaman, and
Bruce R. Hamaker. 2016. “Small Differences in Amylopectin Fine Structure May Explain
Large Functional Differences of Starch.” ​Carbohydrate Polymers​ 140 (April): 113–21.
https://doi.org/10.1016/j.carbpol.2015.12.025.
Biliaderis, Costas G. 2009. “Structural Transitions and Related Physical Properties of Starch.” In
Starch​, 293–372. Elsevier. https://doi.org/10.1016/B978-0-12-746275-2.00008-2.

16
Billinge, S. J. L., and I. Levin. 2007. “The Problem with Determining Atomic Structure at the
Nanoscale.” ​Science​ 316 (5824): 561–65. https://doi.org/10.1126/science.1135080.
Bochicchio, Davide, Emanuele Panizon, Riccardo Ferrando, Luca Monticelli, and Giulia Rossi.
2015. “Calculating the Free Energy of Transfer of Small Solutes into a Model Lipid
Membrane: Comparison between Metadynamics and Umbrella Sampling.” ​The Journal
of Chemical Physics​ 143 (14): 144108. https://doi.org/10.1063/1.4932159.
Brown, Stephen A., and Dexter French. 1977. “Specific Adsorption of Starch Oligosaccharides
in the Gel Phase of Starch Granules.” ​Carbohydrate Research​ 59 (1): 203–12.
https://doi.org/10.1016/S0008-6215(00)83306-6.
“Bussi and Laio - 2020 - Using Metadynamics to Explore Complex Free-Energy .Pdf.” n.d.
Bussi, Giovanni, and Davide Branduardi. 2015. “Free-Energy Calculations with Metadynamics:
Theory and Practice.” In ​Reviews in Computational Chemistry,​ edited by Abby L. Parrill
and Kenny B. Lipkowitz, 1–49. Hoboken, NJ, USA: John Wiley & Sons, Inc.
https://doi.org/10.1002/9781118889886.ch1.
Bussi, Giovanni, and Alessandro Laio. 2020. “Using Metadynamics to Explore Complex
Free-Energy Landscapes.” ​Nature Reviews Physics​ 2 (4): 200–212.
https://doi.org/10.1038/s42254-020-0153-0.
Cheetham, Norman W.H., Paramita Dasgupta, and Graham E. Ball. 2003. “NMR and Modelling
Studies of Disaccharide Conformation.” ​Carbohydrate Research​ 338 (9): 955–62.
https://doi.org/10.1016/S0008-6215(03)00069-7.
Cheetham, Norman W.H., and Leping Tao. 1998a. “Amylose Conformational Transitions in
Binary DMSO/Water Mixtures.” ​Carbohydrate Polymers​ 35 (3–4): 287–95.
https://doi.org/10.1016/S0144-8617(97)00151-3.
Cheetham, Norman W.H, and Leping Tao. 1998b. “Variation in Crystalline Type with Amylose
Content in Maize Starch Granules: An X-Ray Powder Diffraction Study.” ​Carbohydrate
Polymers​ 36 (4): 277–84. https://doi.org/10.1016/S0144-8617(98)00007-1.
Cheng, Lilin, Tao Feng, Boyu Zhang, Xiao Zhu, Bruce Hamaker, Hui Zhang, and Osvaldo
Campanella. 2018. “A Molecular Dynamics Simulation Study on the Conformational
Stability of Amylose-Linoleic Acid Complex in Water.” ​Carbohydrate Polymers​ 196
(September): 56–65. https://doi.org/10.1016/j.carbpol.2018.04.102.
Delcour, Jan A., Charlotte Bruneel, Liesbeth J. Derde, Sara V. Gomand, Bram Pareyt, Joke A.
Putseys, Edith Wilderjans, and Lieve Lamberts. 2010. “Fate of Starch in Food
Processing: From Raw Materials to Final Food Products.” ​Annual Review of Food
Science and Technology​ 1 (1): 87–111.
https://doi.org/10.1146/annurev.food.102308.124211.
Flory, Paul J. 1989. ​Statistical Mechanics of Chain Molecules​. Reprint. ed. Munich: Hanser
[u.a.].
Frenkel, Daan, and Berend Smit. 2002. ​Understanding Molecular Simulation: From Algorithms
to Applications​. 2nd ed. Computational Science Series 1. San Diego: Academic Press.
Fuchs, Alain H., and Anthony K. Cheetham. 2001. “Adsorption of Guest Molecules in Zeolitic
Materials: Computational Aspects.” ​The Journal of Physical Chemistry B​ 105 (31):
7375–83. https://doi.org/10.1021/jp010702q.
Gallant, Daniel J., Brigitte Bouchet, and Paul M. Baldwin. 1997. “Microscopy of Starch:
Evidence of a New Level of Granule Organization.” ​Carbohydrate Polymers​ 32 (3–4):
177–91. https://doi.org/10.1016/S0144-8617(97)00008-8.
Galvelis, Raimondas, Suyong Re, and Yuji Sugita. 2017. “Enhanced Conformational Sampling
of N-Glycans in Solution with Replica State Exchange Metadynamics.” ​Journal of
Chemical Theory and Computation​ 13 (5): 1934–42.

17
 https://doi.org/10.1021/acs.jctc.7b00079.
Georget, D. M. R., R. Parker, and A. C. Smith. 1995. “ASSESSMENT OF A PIN
DEFORMATION TEST FOR MEASUREMENT OF MECHANICAL PROPERTIES OF
BREAKFAST CEREAL FLAKES.” ​Journal of Texture Studies​ 26 (2): 161–74.
https://doi.org/10.1111/j.1745-4603.1995.tb00791.x.
Gidley, Michael J., and Paul V. Bulpin. 1989. “Aggregation of Amylose in Aqueous Systems:
The Effect of Chain Length on Phase Behavior and Aggregation Kinetics.”
Macromolecules​ 22 (1): 341–46. https://doi.org/10.1021/ma00191a062.
Griffiths, Peter, and Peter Stilbs. 2002. “NMR Self-Diffusion Studies of Polymeric Surfactants.”
Current Opinion in Colloid & Interface Science​ 7 (3–4): 249–52.
https://doi.org/10.1016/S1359-0294(02)00042-0.
Guzmán-Maldonado, Horacio, Octavio Paredes-López, and Costas G. Biliaderis. 1995.
“Amylolytic Enzymes and Products Derived from Starch: A Review.” ​Critical Reviews in
Food Science and Nutrition​ 35 (5): 373–403.
https://doi.org/10.1080/10408399509527706.
Hatcher, Elizabeth, Elin Säwén, Göran Widmalm, and Alexander D. MacKerell. 2011.
“Conformational Properties of Methyl β-Maltoside and Methyl α- and β-Cellobioside
Disaccharides.” ​The Journal of Physical Chemistry B​ 115 (3): 597–608.
https://doi.org/10.1021/jp109475p.
Johnson, Kathryn A., and Lisa J. Mauer. 2019. “Effects of Controlled Relative Humidity Storage
on Moisture Sorption and Amylopectin Retrogradation in Gelatinized Starch Lyophiles:
Amylopectin Retrogradation in Lyophiles….” ​Journal of Food Science​ 84 (3): 507–23.
https://doi.org/10.1111/1750-3841.14472.
Keränen, Henrik, Laura Pérez-Benito, Myriam Ciordia, Francisca Delgado, Thomas B.
Steinbrecher, Daniel Oehlrich, Herman W. T. van Vlijmen, Andrés A. Trabanco, and
Gary Tresadern. 2017. “Acylguanidine Beta Secretase 1 Inhibitors: A Combined
Experimental and Free Energy Perturbation Study.” ​Journal of Chemical Theory and
Computation​ 13 (3): 1439–53. https://doi.org/10.1021/acs.jctc.6b01141.
Kilburn, Duncan, Johanna Claude, Thomas Schweizer, Ashraf Alam, and Job Ubbink. 2005.
“Carbohydrate Polymers in Amorphous States: An Integrated Thermodynamic and
Nanostructural Investigation.” ​Biomacromolecules​ 6 (2): 864–79.
https://doi.org/10.1021/bm049355r.
Koneru, Jaya Krishna, Xiao Zhu, and Jagannath Mondal. 2019. “Quantitative Assessment of the
Conformational Heterogeneity in Amylose across Force Fields.” ​Journal of Chemical
Theory and Computation​ 15 (11): 6203–12. https://doi.org/10.1021/acs.jctc.9b00630.
Kuang, Qirong, Jinchuan Xu, Yongri Liang, Fengwei Xie, Feng Tian, Sumei Zhou, and Xingxun
Liu. 2017. “Lamellar Structure Change of Waxy Corn Starch during Gelatinization by
Time-Resolved Synchrotron SAXS.” ​Food Hydrocolloids​ 62 (January): 43–48.
https://doi.org/10.1016/j.foodhyd.2016.07.024.
Kuttel, Michelle M., and Kevin J. Naidoo. 2005. “Free Energy Surfaces for the α(1 →
4)-Glycosidic Linkage: Implications for Polysaccharide Solution Structure and
Dynamics.” ​The Journal of Physical Chemistry B​ 109 (15): 7468–74.
https://doi.org/10.1021/jp044756m.
Levine, Harry, and Louise Slade, eds. 1991. ​Water Relationships in Foods: Advances in the
1980s and Trends for the 1990s.​ Vol. 302. Advances in Experimental Medicine and
Biology. Boston, MA: Springer US. https://doi.org/10.1007/978-1-4899-0664-9.
Li, Hongyan, and Robert G. Gilbert. 2018. “Starch Molecular Structure: The Basis for an
Improved Understanding of Cooked Rice Texture.” ​Carbohydrate Polymers​ 195

18
 (September): 9–17. https://doi.org/10.1016/j.carbpol.2018.04.065.
López, Cesar A., Alex H. de Vries, and Siewert J. Marrink. 2012. “Amylose Folding under the
Influence of Lipids.” ​Carbohydrate Research​ 364 (December): 1–7.
https://doi.org/10.1016/j.carres.2012.10.007.
Ma, Zhen, and Joyce I. Boye. 2018. “Research Advances on Structural Characterization of
Resistant Starch and Its Structure-Physiological Function Relationship: A Review.”
Critical Reviews in Food Science and Nutrition​ 58 (7): 1059–83.
https://doi.org/10.1080/10408398.2016.1230537.
Mali, S., L.S. Sakanaka, F. Yamashita, and M.V.E. Grossmann. 2005. “Water Sorption and
Mechanical Properties of Cassava Starch Films and Their Relation to Plasticizing Effect.”
Carbohydrate Polymers​ 60 (3): 283–89. https://doi.org/10.1016/j.carbpol.2005.01.003.
Materials Science Suite.​ 2020. New York, NY: Schrödinger, LLC.
Mishra, Sushil Kumar, Mahmut Kara, Martin Zacharias, and Jaroslav Koča. 2014. “Enhanced
Conformational Sampling of Carbohydrates by Hamiltonian Replica-Exchange
Simulation.” ​Glycobiology​ 24 (1): 70–84. https://doi.org/10.1093/glycob/cwt093.
Mohammadi Nafchi, Abdorreza, Mahdiyeh Moradpour, Maliheh Saeidi, and Abd Karim Alias.
2013. “Thermoplastic Starches: Properties, Challenges, and Prospects.” ​Starch - Stärke
65 (1–2): 61–72. https://doi.org/10.1002/star.201200201.
Momany, Frank A., J.L. Willett, and Udo Schnupf. 2009. “Molecular Dynamics Simulations of a
Cyclic-DP-240 Amylose Fragment in a Periodic Cell: Glass Transition Temperature and
Water Diffusion.” ​Carbohydrate Polymers​ 78 (4): 978–86.
https://doi.org/10.1016/j.carbpol.2009.07.034.
Naidoo, Kevin J, and Michelle Kuttel. n.d. “Water Structure about the Dimer and Hexamer
Repeat Units of Amylose from Molecular Dynamics Computer Simulations.” ​JOURNAL
OF COMPUTATIONAL CHEMISTRY​ 22 (4): 12.
Nara, Sh., A. Mori, and T. Komiya. 1978. “Study on Relative Crystallinity of Moist Potato Starch.”
Starch - Stärke​ 30 (4): 111–14. https://doi.org/10.1002/star.19780300403.
Nessi, Veronica, Agnès Rolland-Sabaté, Denis Lourdin, Frédéric Jamme, Chloé Chevigny, and
Kamal Kansou. 2018. “Multi-Scale Characterization of Thermoplastic Starch Structure
Using Second Harmonic Generation Imaging and NMR.” ​Carbohydrate Polymers​ 194
(August): 80–88. https://doi.org/10.1016/j.carbpol.2018.04.030.
Origin​ (version Version 2020). n.d. Northampton, MA, USA: OriginLab Corporation.
Ott, Karl-Heinz, and Bernd Meyer. 1996. “Molecular Dynamics Simulations of Maltose in Water.”
Carbohydrate Research​ 281 (1): 11–34. https://doi.org/10.1016/0008-6215(95)00320-7.
Özeren, Hüsamettin D., Richard T. Olsson, Fritjof Nilsson, and Mikael S. Hedenqvist. 2020.
“Prediction of Plasticization in a Real Biopolymer System (Starch) Using Molecular
Dynamics Simulations.” ​Materials & Design​ 187 (February): 108387.
https://doi.org/10.1016/j.matdes.2019.108387.
Patrone, Paul N., Andrew Dienstfrey, Andrea R. Browning, Samuel Tucker, and Stephen
Christensen. 2016. “Uncertainty Quantification in Molecular Dynamics Studies of the
Glass Transition Temperature.” ​Polymer​ 87 (March): 246–59.
https://doi.org/10.1016/j.polymer.2016.01.074.
Pereira, Cristina S., David Kony, Riccardo Baron, Martin Müller, Wilfred F. van Gunsteren, and
Philippe H. Hünenberger. 2006. “Conformational and Dynamical Properties of
Disaccharides in Water: A Molecular Dynamics Study.” ​Biophysical Journal​ 90 (12):
4337–44. https://doi.org/10.1529/biophysj.106.081539.
Pérez, Serge, and Eric Bertoft. 2010. “The Molecular Structures of Starch Components and
Their Contribution to the Architecture of Starch Granules: A Comprehensive Review.”

19
 Starch - Stärke​ 62 (8): 389–420. https://doi.org/10.1002/star.201000013.
Pérez-Benito, Laura, Henrik Keränen, Herman van Vlijmen, and Gary Tresadern. 2018.
“Predicting Binding Free Energies of PDE2 Inhibitors. The Difficulties of Protein
Conformation.” ​Scientific Reports​ 8 (1): 4883.
https://doi.org/10.1038/s41598-018-23039-5.
Perić-Hassler, Lovorka, Halvor S. Hansen, Riccardo Baron, and Philippe H. Hünenberger. 2010.
“Conformational Properties of Glucose-Based Disaccharides Investigated Using
Molecular Dynamics Simulations with Local Elevation Umbrella Sampling.” ​Carbohydrate
Research​ 345 (12): 1781–1801. https://doi.org/10.1016/j.carres.2010.05.026.
Quigley, Gary J., A. Sarko, and R. H. Marchessault. 1970. “Crystal and Molecular Structure of
Maltose Monohydrate.” ​Journal of the American Chemical Society​ 92 (20): 5834–39.
https://doi.org/10.1021/ja00723a003.
“Roos et al. - 2019 - OPLS3e Extending Force Field Coverage for Drug-Li.Pdf.” n.d.
Roos, Katarina, Chuanjie Wu, Wolfgang Damm, Mark Reboul, James M. Stevenson, Chao Lu,
Markus K. Dahlgren, et al. 2019. “OPLS3e: Extending Force Field Coverage for
Drug-Like Small Molecules.” ​Journal of Chemical Theory and Computation​ 15 (3):
1863–74. https://doi.org/10.1021/acs.jctc.8b01026.
Rundle, R. E., and Frank C. Edwards. 1943. “The Configuration of Starch in the Starch-Iodine
Complex. IV. An X-Ray Diffraction Investigation of Butanol-Precipitated Amylose 1​​ .”
Journal of the American Chemical Society​ 65 (11): 2200–2203.
https://doi.org/10.1021/ja01251a055.
Russo, Melissa A. L., Ekaterina Strounina, Muriel Waret, Timothy Nicholson, Rowan Truss, and
Peter J. Halley. 2007. “A Study of Water Diffusion into a High-Amylose Starch Blend:
The Effect of Moisture Content and Temperature.” ​Biomacromolecules​ 8 (1): 296–301.
https://doi.org/10.1021/bm060791i.
Schrödinger Release 2020-1: Desmond Molecular Dynamics System.​ 2020. New York, NY: D.
E. Shaw Research.
Shotton, E., and N. Harb. 1965. “The Effect of Humidity and Temperature on the Equilibrium
Moisture Content of Powders.” ​Journal of Pharmacy and Pharmacology​ 17 (8): 504–8.
https://doi.org/10.1111/j.2042-7158.1965.tb07712.x.
Silva, Felipe Azevedo Rios, Maria José Araújo Sales, Leonardo Giordano Paterno, Mohamed
Ghoul, Latifa Chebil, and Elaine Rose Maia. 2018. “Molecular Dynamics Studies of
Amylose Plasticized with Brazilian Cerrado Oils: Part I.” ​Polímeros​ 28 (3): 266–74.
https://doi.org/10.1590/0104-1428.09917.
Sindhikara, Dan, and Ken Borrelli. 2018. “High Throughput Evaluation of Macrocyclization
Strategies for Conformer Stabilization.” ​Scientific Reports​ 8 (1): 6585.
https://doi.org/10.1038/s41598-018-24766-5.
Singh, Jaspreet, Lovedeep Kaur, and O.J. McCarthy. 2007. “Factors Influencing the
Physico-Chemical, Morphological, Thermal and Rheological Properties of Some
Chemically Modified Starches for Food Applications—A Review.” ​Food Hydrocolloids​ 21
(1): 1–22. https://doi.org/10.1016/j.foodhyd.2006.02.006.
Smit, Berend, and Theo L. M. Maesen. 2008. “Molecular Simulations of Zeolites: Adsorption,
Diffusion, and Shape Selectivity.” ​Chemical Reviews​ 108 (10): 4125–84.
https://doi.org/10.1021/cr8002642.
Song, Mi-Ra, Seung-Hyun Choi, Seon-Min Oh, Hui-yun Kim, Ji-Eun Bae, Cheon-Seok Park,
Byung-Yong Kim, and Moo-Yeol Baik. 2017. “Characterization of Amorphous Granular
Starches Prepared by High Hydrostatic Pressure (HHP).” ​Food Science and
Biotechnology​ 26 (3): 671–78. https://doi.org/10.1007/s10068-017-0106-2.

20
Srichuwong, Sathaporn, Titi Candra Sunarti, Takashi Mishima, Naoto Isono, and Makoto
Hisamatsu. 2005. “Starches from Different Botanical Sources II: Contribution of Starch
Structure to Swelling and Pasting Properties.” ​Carbohydrate Polymers​ 62 (1): 25–34.
https://doi.org/10.1016/j.carbpol.2005.07.003.
Steinhauser, Martin, and Stefan Hiermaier. 2009. “A Review of Computational Methods in
Materials Science: Examples from Shock-Wave and Polymer Physics.” ​International
Journal of Molecular Sciences​ 10 (12): 5135–5216.
https://doi.org/10.3390/ijms10125135.
Thérien-Aubin, Héloïse, Wilms E. Baille, Xiao Xia Zhu, and Robert H. Marchessault. 2005.
“Imaging of High-Amylose Starch Tablets. 3. Initial Diffusion and Temperature Effects.”
Biomacromolecules​ 6 (6): 3367–72. https://doi.org/10.1021/bm0503930.
Trommsdorff, U., and I. Tomka. 1995. “Structure of Amorphous Starch. 2. Molecular Interactions
with Water.” ​Macromolecules​ 28 (18): 6138–50. https://doi.org/10.1021/ma00122a022.
Turupcu, Aysegül, and Chris Oostenbrink. 2017. “Modeling of Oligosaccharides within
Glycoproteins from Free-Energy Landscapes.” ​Journal of Chemical Information and
Modeling​ 57 (9): 2222–36. https://doi.org/10.1021/acs.jcim.7b00351.
Vamadevan, Varatharajan, and Eric Bertoft. 2020. “Observations on the Impact of Amylopectin
and Amylose Structure on the Swelling of Starch Granules.” ​Food Hydrocolloids​ 103
(June): 105663. https://doi.org/10.1016/j.foodhyd.2020.105663.
Varki, A., and M. J. Chrispeels. 1999. ​Essentials of Glycobiology.​ Cold Spring Harbor
Laboratory Press. https://books.google.com/books?id=lH72FFWIIpgC.
Vlachy, V., and A. D. J. Haymet. 1986. “A Grand Canonical Monte Carlo Simulation Study of
Polyelectrolyte Solutions.” ​The Journal of Chemical Physics​ 84 (10): 5874–80.
https://doi.org/10.1063/1.449898.
Wang, Shujun, Caili Li, Les Copeland, Qing Niu, and Shuo Wang. 2015. “Starch Retrogradation:
A Comprehensive Review: Starch Retrogradation….” ​Comprehensive Reviews in Food
Science and Food Safety​ 14 (5): 568–85. https://doi.org/10.1111/1541-4337.12143.
Wang, Yao-Chun, Shin-Pon Ju, Chien-Chia Chen, Hsin-Tsung Chen, and Jin-Yuan Hsieh.
2014. “Mechanical Property Prediction of Starch/Polymer Composites by Molecular
Dynamics Simulation.” ​RSC Adv.​ 4 (22): 11475–80.
https://doi.org/10.1039/C3RA46213G.
Williams, Malcolm L., Robert F. Landel, and John D. Ferry. 1955. “The Temperature
Dependence of Relaxation Mechanisms in Amorphous Polymers and Other
Glass-Forming Liquids.” ​Journal of the American Chemical Society​ 77 (14): 3701–7.
https://doi.org/10.1021/ja01619a008.
Winger, Moritz, Markus Christen, and Wilfred F. van Gunsteren. 2009. “On the Conformational
Properties of Amylose and Cellulose Oligomers in Solution.” ​International Journal of
Carbohydrate Chemistry​ 2009: 1–8. https://doi.org/10.1155/2009/307695.
Yang, Chao, Xiao Xing, Zili Li, and Shouxin Zhang. 2020. “A Comprehensive Review on Water
Diffusion in Polymers Focusing on the Polymer–Metal Interface Combination.” ​Polymers
12 (1): 138. https://doi.org/10.3390/polym12010138.
Zhang, Changquan, Lijia Zhu, Ke Shao, Minghong Gu, and Qiaoquan Liu. 2013. “Toward
Underlying Reasons for Rice Starches Having Low Viscosity and High Amylose:
Physiochemical and Structural Characteristics: Reasons for Rice Starches Having Low
Viscosity and High Amylose.” ​Journal of the Science of Food and Agriculture​ 93 (7):
1543–51. https://doi.org/10.1002/jsfa.5987.
Zhu, Fan. 2017. “Encapsulation and Delivery of Food Ingredients Using Starch Based Systems.”
Food Chemistry​ 229 (August): 542–52. https://doi.org/10.1016/j.foodchem.2017.02.101.

21
Zobel, H. F. 1988. “Molecules to Granules: A Comprehensive Starch Review.” ​Starch - Stärke
40 (2): 44–50. https://doi.org/10.1002/star.19880400203.

Supplementary data

A) B)

C)

Supplementary Figure 1. Radius of Gyration (R​g​) for DP12 amylose. ​A) Time evolution of R​g
for solvated DP12 over the course of 1.0 µs for a NPT ensemble at 300 K and 1.01325 bar. B)
Distribution of R​g​ values for solvated DP12. C) Distribution of amorphous R​g​ for anhydrous
DP12 from 50 ns for a NPT ensemble at 300 K and 1.01325 bar.

22
 A) B)

Supplementary Figure 2. Moisture uptake as a function of time from GCMC/MD water

uptake simulations.​ Plots show water uptake as a function of time for A) various RH values
and B) various temperatures. Each curve represents the average of five independent
simulations.

23
Supplementary Figure 3. Diffusion under different water and temperature conditions​.
Output structures from the temperature dependent GCMC/MD simulations were used to predict
diffusion at a given uptake temperature. Each data point is the average of five independent
simulations.

24