See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/51847736

More than one dynamic crossover in protein hydration water

Article in Proceedings of the National Academy of Sciences · November 2011

DOI: 10.1073/pnas.1104299108 · Source: PubMed

CITATIONS READS
92 61

6 authors, including:

Marco G. Mazza Kevin Stokely

Loughborough University Boston University
136 PUBLICATIONS 2,333 CITATIONS 19 PUBLICATIONS 716 CITATIONS

SEE PROFILE SEE PROFILE

Fabio Bruni H. Eugene Stanley

Università Degli Studi Roma Tre Boston University
105 PUBLICATIONS 5,336 CITATIONS 1,773 PUBLICATIONS 183,059 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Kevin Stokely on 20 May 2014.

The user has requested enhancement of the downloaded file.

 Two dynamic crossovers in protein hydration water and their
thermodynamic interpretation

Marco G. Mazza,1 Kevin Stokely,1 Sara E. Pagnotta,2

Fabio Bruni,3 H. Eugene Stanley,1 and Giancarlo Franzese4
arXiv:0907.1810v1 [cond-mat.soft] 10 Jul 2009

1
Center for Polymer Studies and Department of Physics,
Boston University, Boston, Massachusetts 02215, USA
2
Centro de Fisica de Materiales (CSIC-UPV/EHU) -
Materials Physics Center MPC, Donostia-San Sebastian, Spain
3
Dipartimento di Fisica “E. Amaldi”, Universita’ di Roma Tre,
Istituto Nazionale per la Fisica della Materia, Unita’ di Roma Tre,
Via della Vasca Navale 84, 00146 Roma, Italy
4
Departament de Fisica Fonamental, Universitat de Barcelona,
Diagonal 647, 08028 Barcelona, Spain

1
 Abstract
Careful studies of liquid water in its supercooled region have led to many insights into the struc-
ture and behavior of water in all temperature regions. While bulk water freezes at its homogeneous
nucleation temperature of approximately 235 K, for protein hydration water, the binding of water
molecules to the protein avoids crystallization. Here we study the slow dynamics of water at low
temperatures by dielectric relaxation experiments on lysozyme hydration water over an extremely
broad range of frequencies and temperatures. We probe the temperature dependence of the re-
laxation time due to water protons in order to gain insight into the structure and connectivity of
the hydrogen bond network of water molecules adsorbed on the protein surface. We observe two
dynamic crossovers: one at about 252 K, assigned to a change of the diffusive regime of water
protons along the hydrogen bond network; the other, at about 181 K assigned to a structural
re–arrangement of the water network. We support this interpretation by comparing the experi-
mental results with Monte Carlo simulations and mean–field calculations on a cell model of water.
The model allows an interpretation in terms of thermodynamic features of water, predicting the
presence of two specific heat maxima at ambient pressure. The first is due to fluctuations in the
hydrogen bond formation, and the second, at lower temperature, due to cooperative reordering of
the hydrogen bond network.

PACS numbers:

2
 Among liquids, water is undoubtedly the most studied, as it is the medium of life and
chemistry on our planet, and its puzzling properties are a challenge for science. Many
aspects of the dynamical behavior of water, both as a bulk solvent or adsorbed on the
surface of biological macromolecules, can be tackled by dielectric relaxation spectroscopy.
This powerful technique measures the interaction of a sample with a time-dependent small-
amplitude electric field [1] and is one of the few techniques that can span a very large
frequency range, allowing the observation of dynamical processes over many time scales.
The field-induced polarization can be investigated either in the time domain or as a function
of field frequency, ω. In the latter case the response function includes all the contributions
to polarization that depend on frequency, and thus reflects the dynamics of dipole and the
displacement of charges in the sample. Recent work has focused on dielectric relaxation
experiments on hydrated proteins, aimed at the study of the dynamics of water molecules
together with the protein amino acids [2, 3]. At variance with these experiments, we will
report on the dielectric relaxation of water protons. In particular, we discuss the temperature
dependence of proton dynamics along chains of hydrogen bonded water molecules at the
interface with a globular protein. The rationale behind this approach is that measurements
of the proton mobility are closely linked to the dynamics and connectivity of the water
molecules in the protein hydration shell [4]. The dielectric relaxation of water protons is a
sensible probe for the structure of the underlying hydrogen bond network, and therefore of
particular interest in the present context. Moreover, protein tumbling can be neglected if
sample hydration level is kept below 0.4 g H2 O/g dry protein, and results can be interpreted
in the framework of well established theoretical models [5, 6, 7].
A dielectric spectrum of the hydrated lysozyme sample at 215 K is shown in Fig. 1.
The high frequency dispersion can be compared to that observed in a recent dielectric
study of the same hydrated protein [2], and we find an identical temperature dependence
(data not shown). The assignment of this relaxation process is still controversial [2, 8]. In
addition, results obtained with other experimental techniques applied to hydrated proteins
have raised an intense debate on the nature and temperature dependence of this dynamical
process [9, 10].
Here we focus on a secondary relaxation process, previously tentatively ascribed to motion
of protein side chains or to a cross–term due to the relaxation of water and protein dipoles [2].
We have shown [6] that this secondary relaxation can be resolved into two contributions

3
(Fig. 1): one effectively due to the relaxation of protein side chains, and the other to the
motion of water protons along the hydrogen bond network of water molecules in the first
hydration shell of the protein [6, 11].
Figure 2(a) shows the temperature dependence of the water proton relaxation time τ for
0.3 g H2 O/g dry protein over the temperature range 170 K to 300 K. This relaxation time
has a very strong hydration dependence [12], and in lysozyme samples with hydration lower
than 0.28 g H2 O/g dry protein it can be measured only above 230 K.
At high T , τ shows a non–Arrhenius temperature dependence, well described by the
Vogel–Fulcher–Tamman (VFT) equation τ (T ) = τ0 exp [BT /R(T − T0 )], where τ0 , BT and
T0 are fitting parameters and R is the gas constant. At about 252 K, a crossover is observed
(Fig. 2(a)), from a non–Arrhenius behavior to a second non–Arrhenius behavior.
It should be noted that at about the same temperature, namely 254 K, a dynamical
transition was found for the same hydrated protein, and described in terms of random walk
of protons along the water network changing from a sub–diffusive regime at low temperature
to a diffusive regime above T = 254 K [7].
At about 181 K we observe a second crossover (Fig. 2(a)), as the temperature dependence
of the water proton relaxation time changes from a VFT above 181 K, to Arrhenius, τ (T ) =
τ0 exp (A/RT ) below 181 K, where τ0 is a characteristic relaxation time and A is a constant
activation energy. This might be due to a slowing down of the proton mobility with a less
marked temperature dependence, possibly due to a structural re–arrangement of the water
network.
Neutron scattering experiments on the same hydrated protein revealed two dynamical
transitions, one at 220 K and the other at 150 K [13]. These transitions have been attributed
respectively to the rotational motion of interfacial water, and to proton dynamics on a local
(few Å) scale. In particular, the low temperature transition (at 150 K) was claimed to be
due to a sudden increase of the configurational entropy of the system, linked to a significant
excursion of the hydrogen bond length [13]. Deep inelastic neutron scattering experiments
on the same hydrated protein question this interpretation, as they do no reveal any change of
the hydrogen bond length and of the proton potential up to 290 K [11, 14]. 2 H-NMR studies
on hydrated proteins at a comparable hydration level as that investigated in the present
report have nevertheless indicated two dynamical transitions at about 250 K and at 180 K
due to protons in the protein hydration shell [10]. While the low temperature transition

4
has been interpreted in terms of local motion of protons in a rather rigid hydrogen bond
network, consistent with the present work, the high temperature transition is not discussed
there in the same framework [10]. Our goal here is to give a consistent thermodynamic
interpretation of both crossovers.
To gain insight into the behavior of the relaxation dynamics we perform Monte Carlo
(MC) simulations and mean–field calculations of a cell model for water. Since the protein
powder is, on average, hydrated only by a single layer of molecules, and lysozyme is a
compact globular protein which does not undergo any configurational transformation in our
experimental conditions, we approximate that the only effect of the water–protein interaction
is to reduce the hydration water to a d = 2–dimensional system with the water coordination
number fixed to four [15]. As a consequence each water molecule can form at most four
hydrogen bonds. Since we focus on the hydrogen bond dynamics of water molecules, for
our scope is not relevant if a hydrogen bond is formed with the protein or between water
molecules, but it is relevant that water molecules can form a network of hydrogen bonds.
We consider N = 104 molecules in a d = 2–dimensional volume V , with periodic boundary
conditions, divided into N equivalent cells each with one molecule i ∈ [1, N] and with volume
V /N larger than a hard–core volume v0 due to van der Waals repulsion. The interaction
Hamiltonian in the liquid phase is [16, 17]
X X
H = U0 (r) − J δσij ,σji − Jσ δσik ,σil . (1)
hi,ji (k,l)i

The first term U0 (r) denotes the sum of all the isotropic interactions (such as the van der
Waals interaction) between molecules at distance r ≡ (V /N)1/d , represented by a Lennard–
Jones potential with attractive energy ǫ plus a hard–core at distance r0 ≡ (v0 )1/d . Compar-
ison of our simulations results with the experiments presented here allow us to set ǫ = 5.8
kJ/mol, consistent with the value 5.5 kJ/mol of the estimate of the van der Waals attraction
based on isoelectronic molecules at optimal separation [18], and r0 = 2.9 Å consistent with
the expected value of the van der Waals radius [19, 20].
The second term (with J = 2.9 kJ/mol, δa,b = 1 if a = b and δa,b = 0 otherwise, and
hi, ji denoting that i and j are nearest–neighbors) accounts for the directional contribution
to the hydrogen bond energy, where σij = 1, . . . , q is a (Potts) variable representing the
orientational state of the hydrogen (or the lone e− ) of molecule i facing the lone e− (or the
hydrogen, respectively) of the molecules j. For the sake of simplicity we do not distinguish

5
between hydrogen and lone e− , associating to each molecule four equivalent bond indices
σij . Assuming that a hydrogen bond is formed when the angular deviation with respect to
[ angle is less than 30◦ , we get q = 180◦/30◦ = 6. The third
the linear bond of the OOH
term (with Jσ = 0.29 kJ/mol, and (k, l)i indicating each of the six different pairs of the
four bond indices of molecule i) represents an interaction accounting for the hydrogen bond
cooperativity and giving rise to the O–O–O correlation [21], locally driving the molecules
toward an ordered (tetrahedral in the bulk) configuration with lower energy. By defining
the energy per hydrogen bond (between σij and σji ) as the sum of the interactions in which
two bonded molecules (i and j) are participating, we get EHB = ǫ + J + mJσ /2, where
m = 0, . . . , 6 is the number of cooperative interactions in which that bond variables (σij and
σji ) are participating. With our choice of parameters the values of EHB ranges between 8.70
and 9.6 kJ/mol depending on m. However, a definition of EHB based on a cluster of 5 or 8
bonded molecules in 3D increases this range up to 17 or 18 kJ/mol, respectively. Therefore,
EHB depends on the environment (the value of m and the number of molecules bonded in
a cluster), as observed in computer simulation of the crystalline phases of ice [22], and has
values within the range 6.3 [23]— 23.3 kJ/mol [24], proposed on the base of experiments.
The total volume of the system V ≡ VMC + NHB vHB can change to accommodate the
volume expansion vHB due to hydrogen bond formation [25], where VMC > Nv0 is a dynamical
P
variable allowed to fluctuate in the simulations, and NHB ≡ <i,j> δσij ,σji is the total number
of hydrogen bonds. We adopt vHB = 0.5v0 , corresponding to a maximum hydrogen bond
distance of about 3.3 Å, consistent with other authors [20, 26].
We perform MC simulations at constant N, T and pressure P (see Methods). At low T
and high P we find a liquid–liquid first order phase transition, with negative slope in the
P –T plane, ending in a critical point at Pc = 0.18 ± 0.01 GPa and Tc = 171 ± 1 K [17, 27].
We compute the autocorrelation function CM (t) (see Methods) of the quantity that gives
the orientational order of the four bond indices of molecule i. In Fig. 2(b) we show the
calculation of the MC relaxation time, defined as the time at which CM (t) = 1/e, as a
function of 1/T for pressure P = 0.1 MPa ≃ 1 atm. We find two crossovers: a non–
Arrhenius to non–Arrhenius crossover at T ≈ 252 K and a non–Arrhenius to Arrhenius
crossover at T ≈ 181 K.
From general considerations we know that the structure of a liquid must influence its
dynamics. The Adam–Gibbs theory [28] relates a maximum in isobaric specific heat CP as

6
a function of T with a crossover in relaxation times. This can be intuitively understood if
we consider that CP ≡ T (dS/dT )P corresponds to the variation of entropy S at constant
P , (dS/dT )P . When CP is large, S has a large decrease with decreasing T , i.e. the amount
of order increases and the liquid becomes more structured. Since the dependence of the
relaxation time τ on T is stronger when the energy change needed to modify the structure
is larger, and the change of energy is proportional to CP , a maximum variation in the
structure (CPMax ) implies a change (crossover) in the way τ depends on T . Recent computer
simulations have exploited this relation [29, 30].
We calculate the isobaric specific heat CP ≡ (∂H/∂T )P , where H = hEi + P hV i is the
enthalpy, and h·i denotes the thermodynamic average. Figure 3(a) shows CP as a function
of T for P = 0.1 MPa. Two maxima in CP are clearly visible, at variance with previous
results [31]. As the liquid–liquid critical point is approached in P , the narrow maximum
at lower T diverges, while the relatively broader maximum does not, and the two maxima
merge.
To understand the origin of the two CP maxima, we write the enthalpy as the sum of two
terms
H = H HB + H Coop (2)

where H HB ≡ h−JNHB + P (VMC + NHB vHB )i and H Coop ≡ H − H HB . Hence, we consider

CP = CPHB + CPCoop where we define the component CPHB ≡ ∂H HB /∂T P and the coop-

erative component CPCoop ≡ ∂H Coop /∂T P [Fig. 3(a)]. CPHB is responsible for the broad

maximum at higher T . CPHB captures the enthalpy fluctuations due to the hydrogen bond
formation given by the terms proportional to the hydrogen bond number NHB . To show this
relation, we calculate the locus of maximum fluctuation of hNHB i, related to the maximum
of |dhNHB i/dT |P [Fig. 3(b)], and find that the temperatures of these maxima correlate very
well with the locus of maxima of CPHB .
We find in Fig. 3(a) that the maximum of CP at lower T is given by the maximum of
CPCoop . To confirm that CPCoop corresponds to the enthalpy fluctuations due to the cooperative
term in Eq. 1 proportional to Jσ , we calculate |dhNCoop i/dT |P, where hNCoop i is the average
number of molecules with complete local order of their bond indices. (In the range of T of
interest here the contribution to H of the U0 term is negligible.) We find that the locus of
maxima of |dNCoop /dT |P [Fig. 3(b)] overlaps with the locus of maxima of CPCoop . The same
qualitative behavior is predicted on the base of mean–field (MF) calculations for the model

7
[Fig. 3(c)].
The Arrhenius behavior of τ at very low T is, therefore, the consequence of the saturation
of the hydrogen bond network (|dhNHB i/dT |P ≃ 0). In both experiments and simulations,
the constant Arrhenius activation energy A for structural rearrangements (Fig. 2) is consis-
tent with the average EHB necessary to break a hydrogen bond at low T . In the high T limit
instead, the non–Arrhenius behavior of τ is a consequence of the dilution and disorder of
the hydrogen bond network and the activation energy increases for decreasing T as does the
value EHB averaged over all the molecules. At T just below the maximum of |dhNHB i/dT |P,
the rate of change of hEHB i decreases as does |dhNHB i/dT |P , but at lower T it increases again
as a consequence of the increase of |dhNCoop i/dT |P, giving rise to another non–Arrhenius
behavior of τ at intermediate T , up to reaching the maximal fluctuation, below which the
hydrogen bond network saturates and the Arrhenius behavior takes place.
The qualitative agreement with mean field results (where the dimensionality d and the
nearest-neighbor distance play no role and where the microscopic changes of the water
structure induced by the protein matrix are reflected only in the maximum number of
nearest-neighbors, here fixed to four) shows that we can expect the same behavior also for
the model in d = 3. Therefore, the model predicts that what is observed in the experiments
could be a property of water (including bulk water) and that the presence of the protein
matrix is relevant because it induces a modification in the structure of water inhibiting
crystallization, and makes low T observation possible.
Our results indicate that the thermodynamic features of water induce two dynamical
crossovers. The fluctuations of the hydrogen bond formation give rise to a broad maximum
in CP , which corresponds to the non–Arrhenius to non–Arrhenius crossover, at higher T .
The cooperative ordering (restructuring and saturation) of the hydrogen bond network,
corresponding to the sharp maximum of CP at lower T , gives rise to the non–Arrhenius
to Arrhenius crossover. Our preliminary results show that the effect of the hydration level
and the pressure is to increase the cooperative behavior of the hydrogen bonds, reducing
the range of T where the intermediate non-Arrhenius behavior is observed and reducing the
separation in T between the two CP maxima. At high hydration level in the experiments,
or at high P in the model, the intermediate dynamic regime disappears and the two CP
maxima merge. In the model the two maxima of CP approach each other by increasing P
below the liquid–liquid critical P , while progressively separate at P higher then the liquid–

8
liquid critical point.

METHODS

–Experiments: Measurements were performed using an Alpha Analyzer apparatus (Novo-

control) in the temperature range from 170 K to 300 K at constant temperature steps and
in the frequency range from 10−2 Hz to 107 Hz. Powder samples of crystallized and highly
purified lysozyme powder (from chicken egg white) were purchased from Sigma-Aldrich.
Sample pH and hydration h were adjusted to a final value of 7 and 0.3 g H2 O/g dry pro-
tein, respectively, as previously described [6]. Data analysis, described elsewhere [6], was
performed fitting the measured complex spectra, taking into account sample permittivity
(expressed as a combination of Havriliak-Negami functions and a conductivity term) and
electrode polarization effects
N
σ0 X ∆ǫj
ǫ∗m (ω) = − + ǫ∞ + α β
+ A (ıω)d−1 (3)
ıω j=1 [1 + (ıωτj ) ]
j j

where σ0 is the conductivity, ǫ∞ is the high frequency limit of the permittivity, N is the
number of relaxation processes, ∆ǫj and τj are the dielectric strength and the relaxation
time for the jth contribution, respectively, and A and d characterize the polarizability term.
–Simulations: The MC dynamics consists in updating the variables σij by means of
the Wolff algorithm [32] and updating the volume V in accordance with the acceptance
probability min (1, exp [−β (∆E + P ∆V − NRT ln(Vf /Vi ))]), where β ≡ (RT )−1 , Vi and Vf
are respectively the initial and final values of the volume, ∆V ≡ Vf − Vi , ∆E ≡ Ef − Ei
and Ei and Ef are respectively the initial and final values of the energy calculated as the
the right hand side of Eq. 1 [27].
1
P
The autocorrelation function for the quantity Mi ≡ 4 j σij , that gives the orientational
order of the four bond indices of molecule i, is defined as
N
1 X hMi (t0 + t)Mi (t0 )i − hMi i2
CM (t) ≡ . (4)
N i=1 hMi (t0 )2 i − hMi i2

To rescale the MC results on the experimental results we apply a linear rescaling on T and
P : for a temperature expressed in MC internal units T [ǫ/R], the corresponding temperature
in K is T [K] = (134.5 + 700 T [ǫ/R]) K, corresponding to ǫ = 5.8 kJ/mol, and for the

9
pressure P [GPa] = (7.56 × 10−4 + 0.257P [ǫ/v0 ]) GPa, corresponding to r0 = 2.9 Å. For the
correlation time τ we apply a logarithmic rescaling ln(τ [s]) = −31.3 + 1.74 ln τ [MC Steps],
that is consistent with the observation that the conversion from MC Steps to real time is
T -dependent, with a MC Step corresponding to a larger time at lower T .
–Mean–field (MF) calculations: Each cell is assumed to have constant volume v0 , and
is described by occupancy variable ni = 0, 1. The system is assumed homogenous, with
P
average occupancy n = i ni /N. With a cluster expansion we approximate the probability,
pσ (hσ , T, P ), of forming a hydrogen bond between neighboring molecules, expressed in terms
of P , T and the hydrogen bond density field hσ . MF expressions for the molecular entropy s0
of the variables ni , and sσ of the variables σij , are derived. In equilibrium at constant T , P ,
and chemical potential µ, the system will tend to minimize the Gibbs free energy. Hence the
molecular Gibbs free energy, g(hσ , T, P ) = −2ǫn−2(J −P vHB )npσ −6Jσ pσ +P v0 −T (s0 +sσ ),
is minimized with respect to n and hσ at each (T, P ), yielding equilibrium values neq (T, P )
and heq
σ (T, P ). Substitution of n
eq
and heq
σ into the MF expression for the volume V =

Nv0 + 2Nn2 pσ gives the full equation of state of the system.

In MF we rescale T with the experimental values as T [K] = (102.5 + 850T [ǫMF/R]) K,
corresponding to ǫMF = 706.7 kJ/mol.

[1] Kremer, F., Schönhals, A. Broadband Dielectric Spectroscopy. Springer, New York (2002).
[2] Khodadadi, S., Pawlus, S. & Sokolov, A. P. Influence of hydration on protein dynamics:
Combining dielectric and neutron scattering spectroscopy data. J. Phys. Chem. B 112, 14273–
14280 (2008).
[3] Frauenfelder, H. G. Chen, J. Berendzen, P. W. Fenimore, H. Janssonb, B. H. McMahon, I.
R. Stroe, J. Swenson & Robert D. Young, A unified model of protein dynamics. Proc. Nat.
Acad. Sci. USA 106, 5129–5134 (2009).
[4] Peyrard, M. Glass transition in protein hydration water. Phys. Rev. E 64, 011109 (2001).
[5] Rupley, J. A. & Careri, G. Protein hydration and function. Adv. Protein Chem. 41, 37–172
(1991).
[6] Bruni, F. & Pagnotta, S. E. Dielectric investigation of the temperature dependence of the
dynamics of a hydrated protein. Phys. Chem. Chem. Phys. 6, 1912–1919 (2004).

10
 [7] Pagnotta, S. E., Gargana, R., Bruni, F. & Bocedi, A. Glassy behavior of a percolative water-
protein system. Phys. Rev. E 71, 031506 (2005).
[8] Jansson, H., Bergman, R. & Swenson, J. Relation between solvent and protein dynamics as
studied by dielectric spectroscopy. J. Phys. Chem. B 109, 24134–24141 (2005).
[9] Chen, S.-H., Liu, L., Fratini, E., Baglioni, P., Faraone, A. & Mamontov, E. Observation of
fragile-to-strong dynamic crossover in protein hydration water. Proc. Natl. Acad. Sci. USA
103, 9012–9016 (2006).
[10] Vogel, M. Origins of apparent fragile-to-strong transitions of protein hydration waters. Phys.
Rev. Lett. 101, 225701 (2008).
[11] Pagnotta, S. E., Bruni, F., Senesi, R. & Pietropaolo, A. Quantum behavior of water protons
in protein hydration shell. Biophys. J. 96, 1939–1943 (2009)
[12] Pizzitutti, F. & Bruni, F. Glassy dynamics and enzymatic activity of lysozyme. Phys. Rev. E
64, 052905 (2001).
[13] Zanotti, J.-M., Gibrat, G. & Bellissent-Funel, M.-C. Hydration water rotational motion as
a source of configurational entropy driving protein dynamics. Crossovers at 150 and 220 K.
Phys. Chem. Chem. Phys. 10, 4865–4870 (2008).
[14] Senesi, R., Pietropaolo, A., Bocedi, A., Pagnotta, S. E., & Bruni, F., Proton momentum
distribution in a protein hydration shell. Phys. Rev. Lett. 98, 138102 (2007).
[15] Merzel, F., & Smith, J. C., Is the first hydration shell of lysozyme of higher density than bulk
water? Proc. Nat. Acad. Sci. USA 99, 5378 (2002).
[16] Franzese, G. & Stanley, H. E. Liquid-liquid critical point in a Hamiltonian model for water:
analytic solution. J. Phys.: Condens. Matter 14, 2201–2209 (2002).
[17] Franzese, G., Marqués, M. I. & Stanley, H. E. Intramolecular coupling as a mechanism for a
liquid-liquid phase transition. Phys. Rev. E 67, 011103 (2003).
[18] Henry, M. Nonempirical quantification of molecular interactions in supramolecular assemblies.
Chem. Phys. Chem. 3, 561-569 (2002).
[19] Narten, A. H., Danford, M. D., & Levy, H. A. X-Ray diffraction study of liquid water in the
temperature range 4-200o C. Faraday Discuss. 43, 97-107 (1967).
[20] Soper, A.K. & Ricci, M.A. Structures of high-density and low-density water. Phys. Rev. Lett.
84, 2881-2884 (2000).
[21] Ricci, M. A., Bruni, F. & Giuliani, A. “Similarities” between confined and supercooled water.

11
 Faraday Discussions 141, 347–358 (2009).
[22] Baranyai, A., Bartók, A. & Chialvo, A. A. Computer simulation of the 13 crystalline phases
of ice. J. Chem. Phys. 123, 054502 (2005).
[23] Smith, J. D., Cappa, C. D., Wilson, K. R., Messer, B. M., Cohen R. C. & Saykally, R. J.
Energetics of hydrogen bond network rearrangements in liquid water. Science 306, 851853
(2004).
[24] Suresh, S. J. & Naik, V. M. Hydrogen bond thermodynamic properties of water from dielectric
constant data. J. Chem. Phys. 113, 97279732 (2000).
[25] Debenedetti, P. G. Supercooled and glassy water. J. Phys.: Condens. Matter 15, R1669–
R1726 (2003).
[26] Martins Costa, M. T. C. Qm/MM simulations of polyols in aqueous solution. J. Mol. Struct.
(Theochem) 729, 47-52 (2005).
[27] Mazza, M. G., Stokely, K., Strekalova, E. G., Stanley, H. E. & Franzese, G. Cluster Monte
Carlo and numerical mean field analysis for the water liquid-liquid phase transition. Comp.
Phys. Comm. 180, 497–502 (2009).
[28] Adam, G. & Gibbs, J. H. On the temperature dependence of cooperative relaxation properties
in glass-forming liquids. J. Chem. Phys. 43, 139–146 (1965).
[29] Kumar, P., Yan, Z., Xu, L., Mazza, M. G., Buldyrev, S. V., Chen, S.-H., Sastry, S. & Stanley,
H. E. Glass transition in biomolecules and the liquid-liquid critical point of water. Phys. Rev.
Lett. 97, 177802 (2006).
[30] Kumar, P., Franzese, G. & Stanley, H. E. Predictions of dynamic behavior under pressure for
two scenarios to explain water anomalies. Phys. Rev. Lett. 100, 105701 (2008).
[31] Marqués. M. I. Proposed high-pressure calorimetric experiment to probe theoretical predic-
tions on the liquid-liquid critical point hypothesis. Phys. Rev. E 76, 021503 (2007).
[32] Wolff, U. Collective Monte-Carlo updating for spin systems. Phys. Rev. Lett. 62, 361–364
(1989).
We thank F. Mallamace, S. Sastry and E. G. Strekalova for discussions, NSF grant
CHE0616489 and Spanish MEC grant FIS2007-61433 for support.
Correspondence and requests for materials should be addressed to M. G. Mazza (email:
marcokyb@buphy.bu.edu).

12
 8 1.0

T=215 K
side chain relaxation

proton relaxation

main relaxation
0.8

0.6

)w( 'e
4

)w( ''e
0.4

0.2

0 0.0
-2 0 2 4 6 8
10 10 10 10 10 10

w [s
-1
]

FIG. 1: Dielectric spectrum of hydrated lysozyme sample at 215 K. The sample response function
is expressed in terms of the complex permittivity ǫ∗m (ω) = ǫ′ (ω) − ıǫ′′ (ω). This figure shows the real
(ǫ′ , • left axis) and imaginary (ǫ′′ , N right axis) components of ǫ∗m as a function of angular frequency
ω. The solid lines trough the symbols are the result of the fitting procedure in the complex plane
[6]. The three peaks related to the main relaxation, to the water proton relaxation, and to protein
side chains relaxation, contributing to the measured imaginary component, are shown. From this
figure the contribution due to sample conductivity and the term due to electrode polarization (see
Eq. 3) are omitted for clarity. These terms contribute to the low frequency tail of the measured
complex permittivity.

13
 2 2
10 10
(a) (b)
1 1
10 10
0 0
10 10
relaxation time τ [s]

relaxation time τ [s]

-1 -1
10 10
-2 -2
10 10
-3 -3
10 10
-4 -4
10 10
-5 -5
10 10
proton relaxation MC calculation
-6 -6
10 10
-7 -7
10 3.5 4.0 4.5 5.0 5.5 6.0 6.5 10 3.5 4.0 4.5 5.0 5.5 6.0 6.5
3 -1 3 -1
inverse temperature 10 /T [K ] inverse temperature 10 /T [K ]

FIG. 2: Two crossovers in the water proton relaxation time τ : a non–Arrhenius to non–Arrhenius
crossover at T ≈ 252 K and a non–Arrhenius to Arrhenius crossover at T ≈ 181 K. (a) Experimental
temperature dependence of τ (•), as a function of inverse temperature (1/T ) at ambient pressure.
Solid line is a fit to a VFT equation τ = τ0 exp[BT /R(T − T0 )], with fitting parameters τ0 =
7.8 × 10−12 s, BT = 9.4 kJ/mol, T0 = 180 K, and describes τ in the high T region. Dotted line
indicates the VFT temperature dependence in the intermediate T region, with fitting parameters
τ0 = 6.5 × 10−8 s, BT = 6.2 kJ/mol, T0 = 140 K. Dashed line is a fit with Arrhenius equation
τ = τ0 exp[A/RT ] with fitting parameters τ0 = 1.1 × 10−7 s, A = 25.2 kJ/mol, and describes the
T dependence of τ in the low T region. Notice that the two limiting behaviors, namely that at
high T and that at low T , intersects at about 232 K. (b) Relaxation time (•) from MC calculations
as a function of 1/T , for P = 0.1 MPa. The solid line is a fit to a VFT, with fitting parameters
τ0 = 1.61 × 10−8 s, BT = 5.2 kJ/mol, T0 = 181.2 K; the dotted line is also a fit to a VFT with
parameters τ0 = 7.5 × 10−10 s, BT = 15.9 kJ/mol, T0 = 95.2 K. The fitting parameters for the
Arrhenius law (dashed line) at the lowest T are τ0 = 3.3 × 10−4 , A = 13.7 kJ/mol. In both panels
the Arrhenius activation energy is consistent with the energy to break one hydrogen bond in the
experiments or the model, respectively.

14
 70

isobaric specific heat CP [J K mol ]

(b) MC

-1
(a) MC

temperature derivative [10 K ]

5

-1
60 CP
-1

-2
50 4
HB (|d<NHB>/dT|)P
CP
40 3
(|d<NCoop>/dT|)P
Coop
30 CP
2
20
1
10

0 0
150 200 250 300 350 400 150 200 250 300 350 400
temperature T [K] temperature T [K]

(c) MF
isobaric specific heat CP [J K mol ]
-1

50
Coop HB
-1

CP + CP
40 HB
CP
30

20

10

0
150 200 250 300 350 400
temperature T [K]

FIG. 3: The isobaric specific heat CP from MC simulations for P = 0.1 MPa. (a) CP (•) can
be decomposed into the component CPCoop (⋄) due to the hydrogen bond cooperativity, and the
component CPHB () due to fluctuations of hydrogen bond formation. (b) The variation with
temperature of the average number of cooperative bonds hNCoop i and of the average number of
hydrogen bonds hNHB i for P = 0.1 MPa. (|dhNCoop i/dT |)P and (|dhNHB i/dT |)P show maxima
where CP has maxima. (c) Comparison of CP from MF calculations for the case without cooperative
interaction (Jσ = 0, labeled as CPHB ) and for the case with cooperative interaction (Jσ /ǫ = 0.05,
labeled as CPCoop + CPHB ). The low–T maximum is present only in the second case, hence is due to
the cooperativity. Calculations are at P = 0.

15
View publication stats