Journal of Membrane Science 490 (2015) 152159

Contents lists available at ScienceDirect

Journal of Membrane Science

journal homepage: www.elsevier.com/locate/memsci

A simple methodology for predicting the performances of hyaluronic

acid purication by dialtration
Nadia Oueslati a,b, Pierrick Leblanc a,b, Alice Bodin a,b, Christelle Harscoat-Schiavo a,b,
Emmanuel Rondags a,b, Stphane Meunier c, Ivan Marc a,b, Romain Kapel a,b,n
a
Laboratoire Ractions et Gnie des Procds, UMR-7274, plateforme SVS, 13 rue du bois de la Champelle, F-54500 Vanduvre-ls-Nancy, France
b
Universit de Lorraine, 2 avenue de la fort de Haye, F-54505 Vanduvre-ls-Nancy, France
c
Teoxane Geneva, 105 rue de Lyon Les Charmilles, CH1203 Geneva, Switzerland

art ic l e i nf o a b s t r a c t

Article history: The performances of hyaluronic acid (HA) purication (yield, purity, permeate ow rate and productiv-
Received 12 December 2014 ity) in a semi-synthetic medium composed of salts, proteins/peptides by dialtration through a 100 kDa
Received in revised form PES membrane were studied over 10 diavolumes (DV). The HA purity went from 3% to almost 100% with
8 April 2015
a yield in the retentate close to 100% at the end of the process. The permeate ow rate showed a peculiar
Accepted 9 April 2015
Available online 7 May 2015
trend due to the inuence of charged microsolutes. DF mass balance applied to three groups of solutes
(HA, peptides/proteins and charged microsolutes) allowed a good prediction of HA yield and purity in
Keywords: the course of DF. An equation of the permeate ux based on limit ux equation, correlations between
Hyaluronic acid mass transfer coefcient mineral concentration and mineral mass balance was proposed to predict the
Dialtration
process productivity. Eventually, a methodology based on these equations was proposed to predict HA
Tangential ltration
yield, purity and productivity. The methodology was validated with an actual Streptococcus zooepide-
Modeling
Purication micus broth. Differences between calculations and experimental performances in term of yield, purity
and productivity never exceeded 10%.
& 2015 Elsevier B.V. All rights reserved.

1. Introduction
Hyaluronic acid (HA) is a linear and unbranched mucopolysac-
charide widespread in human soft tissues. Its backbone is composed
of (13) linked N-acetylglucosamine and glucuronic acid and dis-
accharide units are linked by (14) glycoside bonds. HA has var-
ious physical functions including lubrication, water homeostasis,
ltering effects and regulation of plasma protein distribution [1]. HA
is biocompatible, non-immunogenic and thus suitable for biomedi-
cal or cosmetic applications [2,3]. The worldwide demand of HA in
these industrial areas goes increasing for around a decade.
Historically, HA was obtained by extraction from rooster combs.
This process was time and cost consuming and induced high risk of
cross contaminations. Nowadays, HA is from microbial strains of group
A or D Streptococci like Streptococcus equi cultivated in controlled
bioreactors [46].
n
Corresponding author at: Laboratoire Ractions et Gnie des Procds UMR-
7274, plateforme SVS, 13 rue du bois de la Champelle, F-54500 Vanduvre-ls-
Nancy, France.
E-mail address: romain.kapel@univ-lorraine.fr (R. Kapel).
After the production step, HA has to be separated from cells and
puried from the various molecules composing the complex culture
medium. To do so, precipitation assisted by organic solvent (such as
ethanol [7,8], isopropanol [9]) or by complexation adjuvants like
cetylpiridiniumchloride is classically achieved [10]. However, this
requires either large amounts of organic solvents or further down-
stream steps to remove the impurities produced by the micr-
oorganism.
Tangential ultraltration is a well-known solvent-free separation
process widely used for the separation of microsolutes from
microsolutes in complex biological mixtures. In this case, better
macrosolute purities are reached in dialtration mode that consists
in feeding the starting tank with fresh water or buffer [1113].
Recently, Zhou et al. studied the purication of HA from a culture of
Streptococcus zooepidemicus by a combination of microltration
(MF) and ultraltration (UF). They demonstrated that MWCO
100 kDa and 300 kDa polyvinylidene uoride membranes were
appropriate for HA purication (a purication factor of 1000
associated to an HA yield of 77% was observed after the two sep-
aration steps). Interestingly, these authors used classical mass
balance equations of membrane separation to predict the HA yield
in the course of the two steps with success.

http://dx.doi.org/10.1016/j.memsci.2015.04.024
0376-7388/& 2015 Elsevier B.V. All rights reserved.
 N. Oueslati et al. / Journal of Membrane Science 490 (2015) 152159
The two other performance criteria that are HA purity and
process productivity for a targeted purity are more complicated to
predict. Indeed, it would theoretically imply to know R and initial
concentration values of every solutes and the evolution of the
permeate ux in the course of DF.
The aim of this study was to propose a simple methodology for
predicting HA yield, purity and the process productivity in the course
of the DF. To do so, DF of HA in a semi-synthetic mixture gathering
solutes commonly present in a complex HA production broth was
studied. Solutes were divided into three groups: HA, peptides/proteins
and mineral/charged microsolutes. The evolution of the group con-
centrations, permeate ux and HA yield, purity and productivity was
monitored in the course of the DF. From the experimental data, a set of
equations that couple mass balance and limit ux equations to yield,
purity and productivity was proposed. Eventually, the methodology
based on these equation was validated with an actual HA mixture
from S. zooepidemicus culture in bioreactor.
2. Material and methods
2.1. Material
Freeze-dried HA (2 MDa) was obtained from HLT biotechnology
(Javene, France). Cetyltrimetylamonium bromide (CTAB), brain heart
infusion (BHI), yeast extract, casein peptone, glucose, sodium dodecyl
sulfate (SDS), NaHCO3, MnSO4 and NaC2H3O2 were purchased from
Sigma Corporation (St-Quentin Fallavier, France). Na2HPO4, KH2PO4,
NaH2PO4, K2HPO4 and NaOH were procured from CarloErba (Val-de-
Reuil, France). MgSO4 was obtained from VWR Prolabo (Haascrode,
Belgium) and CaCl2 from Thermo Fischer Scientic (Leicestershire,
United Kingdom).
2.2. Dialtration experiments
The ltration experiments were achieved with a tangential ow
ltration system Cogent mscale (Millipore, Molsheim, France) (Fig. 1).
The membrane was a planar polyethersulfone (PES) membrane
(50 cm) with a NMWCO of 100 kDa (Molsheim, Millipore, France).
Both the transmembrane pressure and retentate ow rate were kept
constant at 2.5 bar and 36 ml min
1 respectively. The separation was
performed at room temperature. 200 ml of the HA solution were
implemented and dialtered at constant volume with deionized
water. Every diavolume, permeate ow rate was measured (dV/dt)
and 1 ml sample of retentate was taken for HA concentration,
Fig. 1. Schematic diagram of the membrane ltration system (1, 2: peristaltic
pump; 3, 4: pressure gauge; 5: magnetic stirrer; 6: valve; 7: membrane module; 8:
permeate tank; and 9: water tank).
153
conductivity, peptides/proteins and dry matter quantication. Two
different HA solutions were implemented. The rst one was a 2 MDa
HA solution at 1 g l
1 in a semi-synthetic media composed of
1 g l
1MgSO4, 10 mg l
1 MnSO4, 2.5 g l
1 NaHCO3, 4.5 g l
1
NaC2H3O2, 4.5 g l
1 KH2PO4, 7.75 g l
1 Na2HPO4, 2.5 g l
1 casein
peptone, 7.5 g l
1 yeast extract, and 10 mg l
1 CaCl2 [15]. The second
one was an actual fermentation broth containing HA produced from a
Streptococcus equi subsp. zooepidemicus Farrow and Collins (ATCCs
6580) culture.
2.3. HA-containing fermentation broth production
The HA was produced by Streptococcus equi subsp. zooepidemicus
Farrow and Collins (ATCCs 6580). The culture medium was com-
posed of: 0.5 g l
1 MgSO4, 2 g l
1 K2HPO4, 2 g l
1 KH2PO4, 0.5 g l
1
(NH4)2PO4, 50 g l
1 glucose, 10 g l
1 tryptone and 10 g l
1 yeast ext-
ract [16]. The microorganism was cultivated for about 24 h in BHI to
develop the inoculum. Then, the inoculum was added to obtain
5.107 cells ml
1. Fermented broths were produced in Applikon ADI
1032 stirrer controller (AppliKon biotechnology, Schiedam, the Neder-
lands) in 5 l bioreactor using a working volume of 4 l. The cultures
were carried out for 48 h in the conditions cited by Lai et al. [16]. The
cells were eliminated by frontal ltration.
2.4. Analytical methods
2.4.1. Peptide/peptides quantication
The protein concentration was determined by the BCA protein
assay reagent (Thermo Fischer Scientic, Leicestershire, United
Kingdom). 25 ml of sample were introduced into a microplate well.
200 ml of working reagent (50 parts of BCA reagent A with 1 part of
BCA reagent B) were added and the plate was thoroughly mixed on
a plate shaker for 30 s. Then, the plate was incubated at 37 1C for
30 min and the absorbance was read at 562 nm. The protein con-
centrations were determined thanks to a standard curve established
between 0 and 1 g l
1 with Bovine Serum Albumin (BSA).
2.4.2. HA characterization
2.4.2.1. HA quantication by cetyltrimetylammonium bromide method
(CTM). The cetyltrimetylammonium bormide (CTAB) reagent (2.5 g)
was dissolved in 100 ml of 2% (w/v) NaOH [17]. 50 ml of HA standard
solutions were introduced into 96 well plates, lled with 50 ml of 0.1 M
phosphate buffer pH 7 and which were incubated at 37 1C for 15 min
in the UVvis spectrophotometer (Multiskan Go, Thermo Scientic,
Gometz-le-Chtel, France). Then, 100 ml of CTM reagent at 37 1C were
added to each well plate, incubated for 10 min at 37 1C. The plates
were shaken for 10 s at the beginning and at the end of this
incubation. Absorbance was read at 600 nm against the blank (HA
solution replaced by 0.1 M phosphate buffer pH 7) and plotted against
HA concentrations. The slope of the standard curve was obtained by a
linear regression for HA concentration range between 0 and 0.6 g l
1.
2.4.2.2. HA size determination. The HA integrity was determined by
gel permeation chromatography-light scattering. 50 ml of HA solution
at 0.5 g l
1 were injected into a GPCLS apparatus (Malvern, Orsay,
France). The GPCLS apparatus was constituted of a pre-column, two
columns (A6000M and A7000, Malvern) connected in series, a light
scattering (LS) detector and a Refractive Index (RI) detector. The
mobile phase was phosphate buffered saline (PBS: 137 mM NaCl,
2.7 mM KCl, 10 mM Na2HPO4 and 2 mM KH2PO4) pH 7.4 at a ow rate
of 0.35 ml min
1. The universal calibration was performed with PEO-
19kD (Malvern). The Dn/Dc used to determine the HA concentration
was 0.147 ml g
1 (experimentally determined value).
154 N. Oueslati et al. / Journal of Membrane Science 490 (2015) 152159

2.4.3. Charged microsolute quantication
Semi-synthetic medium and production medium containing HA
can be roughly divided into 3 fractions: a fraction of charged mic-
rosolutes (minerals, metabolites, etc.), a fraction composed of HA and
a third fraction containing peptides/proteins. As a consequence, for
every sample, the charged solute fraction was quantied from the dry
matter, HA and peptides/proteins as follows (Eq. (1)):


Charged microsolutes Dry matter
HA aa=peptides 1
The dry matter was obtained by 48 h desiccation in drying oven at
105 1C of 5 ml solutions. The peptides/proteins and HA were det-
ermined as presented above.
considering the large difference between membrane cut-off
(100 kDa) and HA molar weight (2 MDa). This is in good accor-
dance with retention value of 1.5 MDa (4 0.9) observed by other
authors with a 100 kDa membrane at the same TMP [14]. Peptides/
proteins and charged microsolute retentions were higher than
expected. Minerals and aminoacids are both charged solutes and
at pH 7, HA carboxylic groups are also charged. So, it can be
assumed that the relatively high retention value of these solutes is
due to electrostatic interactions with HA either in the bulk or at
the polarization layer. Such interactions between minerals and
salts have already been clearly shown [19]. Nevertheless, even
with a retention value around 0.3, peptides/proteins and the cha-
rged microsolutes were satisfactorily eliminated by dialtration.

3. Results and discussion

3.1.2. Evolution of HA yield and purity in the course of DF
3.1. Purication of HA from a semi-synthetic medium
Yield and purity are two important performance criteria for the
evaluation of a separation process. The evolution of HA yield and
3.1.1. Evolution of solute concentration in the course of DF
purity in the feed tank during DF are presented in Fig. 3. A slight
The purication of 2 MDa HA at 1 g l
1 in a semi-synthetic
loss of HA, that reached 20% after 10 diavolumes was observed,
medium with a composition close to a medium obtained at the end
which is consistent with the high HA retention. The purity went
of a production step in a bioreactor was studied. The media contained
increasing quickly up to 6 diavolumes near 102% ( 73%) and got
minerals, peptone, yeast extract mainly and the HA was from
stable beyond. At this point of DF, HA yield was higher than 92%
commercial source. The overall solute composition was divided into
(76%). Then, as expected, DF in these conditions was particularly
three categories: HA (macrosolute), charged microsolutes (mainly
efcient for HA purication.
minerals/salts) and peptides/proteins (protein microsolutes). Fig. 2
The yield () of a solute is simply given by the ratio C/C0.
shows the evolution of their relative concentrations (C/C0) in the feed
Considering Eq. (2), HA yield only depends on its retention and it
tank in the course of DF (using a 100 kDa membrane at 2.5 bar
can be written as follows (Eq. (3)):
of TMP).
C/C0 of peptides/proteins and charged microsolutes decrease Vw
HA e
1
RHA V 0 3
drastically during the four rst diavolumes while HA remains
nearly constant in the retentate, mainly. This clearly indicates a
high retention (R) of HA and a low R of proteins and minerals. The
mass balance equation in the feed compartment (Eq. (2), [18])
given below was used to calculate HA, charged and amino-acids
microsolute R values (least-squares method):
C
1
RVVw
e 0 2
C0
C is the solute concentration in feed tank (g l
1), C0 is the initial
solute concentration, Vw is the volume of water added to the feed
tank (m3), V0 is the initial feed tank volume (m3) and R is the
retention. Vw/V0 corresponds to the number of diavolumes that
passed through the membrane.
RHA, Rprotein microsolutes and Rminerals microsolutes were 0.99, 0.34
and 0.33 respectively. RHA high value was not surprising

Fig. 2. Relative concentrations of 2 MDa HA, peptides/proteins and minerals in the

course of DF (100 kDa PES membrane at 2.5 bar of TMP). Vw/V0 corresponds to the
number of diavolumes that passed through the membrane. HA, peptides/proteins
and mineral relative concentrations were calculated from CTM, BCA and conduc-
tivity analysis respectively. The theoretical relative concentrations are presented in
dotted line.
Fig. 3. Yield and purity in dialtration process versus Vw/V0 (number of diavo-
lumes). Theoretical and experimental yields and purities are presented in A and B
respectively.
 N. Oueslati et al. / Journal of Membrane Science 490 (2015) 152159
The purity (P) of a solute is the ratio C/Ci (I being the i component
in the feed tank mixture). Thus, P can be expressed as follows (Eq.
(4)):

1
R
HA V V w
C 0HA e 0
P HA P

1
Ri VVw
C 0i e 0
where C0i is the initial i solute concentration, Ri is the i solute
retention and C0HA is the initial HA concentration.
Beside HA, the other components are charged microsolutes and
peptides/proteins. HA purity can be calculated at any diavolume from
their initial concentration and retention. Experimental HA yield and
purity as a function of diavolume were compared to calculated values
using Eqs. (3) and (4) with retention and initial concentrations shown
in Table 1 (Fig. 3).
Interestingly, the calculated and experimental values are similar for
both performance criteria. Relative errors for HA yield and purity did
not exceed 7% and 18% respectively. Calculated purities were slightly
lower than experimental ones but at the end of the process, experi-
mental values were above 100% indicating a probable overestimation
of HA concentration by CTM. This result indicates that the prediction
of DF performances might be calculated for any culture condition and
starting culture medium by a simple estimation of retention and assay
of initial concentrations of HA, peptides/proteins and dry matter.
3.1.3. Permeate ow rate and productivity
The third separation process performance criterion is the
productivity that corresponds to the mass of HA at a given purity
over the membrane area (A) and the corresponding DF duration
(tDF) (Eq. (5)):
C HA  V 0
Prod:
t DF  A
The permeate ux changes throughout the DF time course. As a
consequence, tDF is expressed as (Eq. (6)):
Z Vw
1
t DF  dV w 6 Eqs. (6) and (5) for productivity calculations.
V 0 Jp  A
where Jp is the permeate ux (ow rate Q p over membrane area
ratio). In our case, this equation needs to be solved numerically since Jp
varies in the course of the process.
The permeate ux in the course of DF increases up to 8 diavolumes
and gets stabilized beyond (Fig. 4). This evolution is rather unusual for
macromolecules/microsolutes separation. Indeed, according to Yee
et al. (2009), the permeate ux rate should remain constant and then
start decrease when fouling takes place [20]. In our case, microsolutes
(minerals and peptides/proteins) have relatively high retention values.
As a consequence, they participate to the polarization layer and their
decreasing concentration during the process might result in an
increase in the permeate ux. However, the gain observed in the
course of the DF (from 14.9 m s
1 to 37.5 m s
1, i.e  2.5) seems
too high to be only explained this way.
Salts are known to inuence the HA structure by shielding
electrostatic repulsion of carboxyl groups in HA molecules and
reducing the dimension of the molecule coils [19]. Thus, the
Table 1
Initial concentration (C0) and retention (R) of each component group of the
mixture. R was calculated from experimental data (Section 3.1.1), HA and aminoacid
initial concentrations were determined from CTM, and BCA assays respectively.
Charged microsolutes were deduced by difference with dry matter.
C0 (g l
1) R (dimensionless)
Peptides/proteins 1.57 0.34
Charged microsolutes 17.07 0.33
Hyaluronic acid 1.07 0.99
155
4
Fig. 4. Permeate ow rate of HA in the time course of DF. 1 g l
1 HA was
introduced in semi-synthetic, mineral media or water. The DF conditions were
constant during the DF process (2.5 bar TMP, 36 ml min
1 feed rate and room
temperature).
modication of the mixture composition in minerals during DF
might cause a structural change of HA also impacting the perme-
ate ux. DF of HA in deionized water and in a solution containing
only the mineral part of the semi-synthetic medium were ach-
ieved to check this hypothesis. Fig. 4 clearly shows that the
permeate ux of HA in deionized water was stable during the
operation, which is the classical behavior when macrosolutes are
implemented. On the contrary, when HA was dialtered with the
mineral fraction of semi-synthetic medium, the ux went increas-
ing. This clearly conrmed the impact of minerals or salts on the
unusual ux evolution, probably through a modication of HA
conformation that changed the polarization layer properties.
5 The permeate ux of 2 MDa HA at concentration ranging from
0.5 to 1.5 g l
1 does not increase with transmembrane pressures
higher than 2.5 bar (data not shown), the HA retention is close to
1 and we neglect the inuence of microsolutes on the polarization
layer. As a consequence, the limit ux (Eq. (7)) can be integrated to
 
Cw
Cp
J p k ln 7
C0
Cp
Cw stands for the macrosolute concentration at the membrane
surface (g l
1), k is a mass transfer coefcient (mm s
1) and Cp is
the macrosolute concentration (g l
1) in the permeate (which can
be neglected in the case of high retention).
The inuence of minerals on Cw and k values was investiga-
ted by measuring permeate uxes at different HA concentrations
(0.5, 1 and 1.5 g l
1) and charged microsolute concentrations
(017.07 g l
1, corresponding to conductivities ranging from 0.09 to
16.78 mS cm
1, Fig. 5A). Cw and k values were calculated by regres-
sion (using Eq. (7)) and plotted versus the conductivity (Fig. 5B and C).
k decreased largely and quickly in a conductivity range of
07.5 mS cm
1 and then tended to stabilize up. At the same time,
a slight increase of the Cw with the conductivity was observed.
Polynomial models were deduced from the experimental
evolutions of k and Cw in order to calculate the permeate ux in
the course of HA DF (Eqs. (8) and (9))
k ak  b k 8
C w aC w  b C w 9
Ak, bk, aCw, and bCw are the polynomial regression coefcients
for k and Cw equations respectively. Their values are presented in
Table 2.
These two equations were combined with Eq. (7) to give Eq.
(10) that gives an instant Jp for a known HA concentration and
156 N. Oueslati et al. / Journal of Membrane Science 490 (2015) 152159

Fig. 5. (A) Effect of the conductivity (due to charged microsolutes) on the permeate Fig.6. Semi-synthetic medium: comparison between calculated and experimental
ux. (B and C) The transfer coefcient k and the HA wall concentration Cw as a permeate uxes, JP, versus the number of diavolumes (A) and productivities versus
function of the conductivity. Coefcient values were calculated by regression from purity (B) in the course of the dialtration process.
the permeate ux experiments measured at different HA concentrations and
conductivities. Flux was measured after 30 min of total recirculation at 2.5 bars
Using Vw values necessary to get the desired purity calculated
of PTM, 36 ml min
1 feed rate and room temperature.
from Eq. (4).
Fig. 6B compares the theoretical and experimental productivity.
Table 2 The theoretical values are close to the experimental data. The
Coefcients of the polynomial models for k and Cw. regression coefcient (R2 0.999) conrms the reliability of the
predictive model.
a b
The modication of microsolute concentration has classically no
k (lm s
1) 5.4
0.30
impact on the limit permeate ux that only depends on the mac-
Cw (g l
1) 82.7 0.17 rosolute initial concentration and the mass transport coefcients. For
the purication of HA, this ux also depends on the charged
microsolute concentrations that change during the operation. Such a
behavior has never been observed to our knowledge. To further
conductivity. investigate the impact of this phenomenon on DF performances,
! productivities for given purities were calculated at different initial
 

aC w  b C w
C p charged solute concentrations (ranging from 0 to 60 g l
1, Fig. 7A) and
J p ak  bk k ln 10
C0
Cp compared with what is obtained for microsolutes that would not
affect the ux, i.e. aminoacids/peptides, in the same concentration
The evolution of conductivity during DF can be calculated using range (Fig. 7B). Fig. 7 shows that productivities are very poor for low
Eq. (2) replacing concentration terms by conductivity since it can purities for charged microsolutes compared to peptides/proteins. The
be reasonably assumed that the composition of solutes (mainly discrepancies are more obvious at low microsolute concen-
minerals/salts responsible for the overall conductivity) does not trations (below 20 g l
1). Then, the differences strongly decrease with
change during the operation and using the retention of minerals. increasing purity. The differences in productivity remain signicant for
The evolution of permeate ux using Eqs. (10) and (2) rev- 95% purity though. For high purity values, the differences are still
ealed to be close to the experimental data (R2 0.997, Fig. 6A) more marked for low microsolutes concentrations. The extent of the
conrming the good reliability of Jp calculation. Eventually, the effect of microsolute concentration is more important at concentra-
process productivity for a given HA purity, can be calculated as tions up to 20 g l
1 probably because beyond this value, charged
follows [18]: molecules have no further inuence on the permeate ux. Indeed,
considering the trends observed in Fig. 5, mass transport coefcients k
V 0  C 0HA  e
1
RHA  V0
Vw and concentrations Cw probably remain constant beyond
Productivity R Vw 1 11 16.8 mS cm
1 (over 15 g l
1 of charged microsolutes). The effect is
V 0 J  dV w
p very important for low purity values since the differences in permeate
 N. Oueslati et al. / Journal of Membrane Science 490 (2015) 152159 157

Fig. 7. Impact of the initial mineral concentration (A) and aminoacids/peptide concentration (B) on the productivity calculated for purities ranging 0100%.

uxes are particularly high for low diavolume numbers (i.e. low
purities).
3.2. Methodology for predicting DF performances and validation
with an actual HA production broth
The results described above indicated that all DF performance
criterion (yield, purity and productivity for of given purity) can be
easily calculated from a simple preliminary UF experiment and a set of
basic analysis of the HA containing broth produced in the upstream
step. The methodology is described in Fig. 8. At rst, a rapid UF of
claried broth in the conditions of the DF (permeate and retentate
recirculation). Sample of permeate and retentate is taken for BCA, dry
matter and CTM analysis (peptides, charged microsolutes and HA
quantication). These quantications are also done with the starting
claried broth to give initial concentration and retention values of HA,
charged microsolutes and peptide/proteins. The conductivity is also
taken in the starting broth in order to get mass transfer terms to apply
to the modied limit ux equation. Once every term are known,
equations can be used to calculate HA yield, purity and productivity in
the course of DF.
HA was produced by S. zooepidemicus in a bioreactor from the
culture medium with composition proposed by Lai [16]. At the end
of the upstream process, HA and microsolute initial concentrations
were determined and their retention (R) values were measured
after 30 min of permeate and retentate recirculation in the UF
conditions (2.5 bar TMP, room temperature and 36 ml min
1 of
recirculating ow rate). The initial concentrations and the reten-
tion values are presented in Table 3.
HA retention (0.95) is in good agreement with the results of the
literature since a similar R value was observed with a similar
membrane in the same operating conditions [14]. However, this
158 N. Oueslati et al. / Journal of Membrane Science 490 (2015) 152159

Fig. 8. Methodology for predicting the evolution performance criterion (yield, purity and productivity) in the course of DF.

Table 3
Initial concentrations and retention values for the microsolutes and HA contained
in the complex HA culture medium.

Solutes C0 (g l
1) R (dimensionless)

Peptides/proteins 0.99 0.01

Minerals 18.46 0.05
Hyaluronic acid 0.83 0.95

value revealed slightly lower than expected with such a difference

between solute molar size (HA 1.5 MDa) and membrane cut-off
(100 kDa). The HA size and polydispersity before and after DF
assessed by GPCLALS were similar (1.5 MDa and polydispersity
value of 1.5). Then, a small part of the whole HA network passed
through the membrane probably under the effect of molecules
produced by S. zooepidemicus that reduced the HA/HA cohesion.
The retention of microsolutes was lower than observed with the
semi-synthetic medium. Interestingly, some authors previously sho-
wed that molecules in culture medium such as peptidoglycan unit
(charged positively) interact with HA [21] reducing the interactions
(repulsion or attraction) between HA and proteins/peptides or miner-
als. The transmission of microsolutes through the membrane would
then be increased, what is observed in this case.
In any case, the yield and the purity were calculated respectively as
explained above using retentions and initial concentrations in Table 3.
The very good accordance between calculated and experimental data
is shown in Fig. 9 This clearly indicates that the methodology pro-
posed can be applied to different mixture compositions.
J and the productivity for purity ranging from 10% to 100% were
calculated. As shown in Fig. 10, calculated and experimental values Fig. 9. Natural HA mixture (microbial culture): comparison of calculated and
are very close (R2 values over 0.999). experimental yield (A) and purity (B) versus the number of diavolumes (Vw/V0).
A slight underestimation of calculated values is observed though. A Permeate ow rate and productivity.
preliminary study (data not shown) indicated that HA size has a slight
impact on the permeate ux (around 5% higher uxes for higher HA
MW in a range of 1.14.4 MDa). In our case, ux calculations were could explain the slight underestimation of the methodology. As a
from parameter regression with experimental uxes obtained using consequence, the composition of the mixture and particularly
2 MDa HA and the HA produced has a larger size. That most probably the composition of minerals did not modify much the empiric
 N. Oueslati et al. / Journal of Membrane Science 490 (2015) 152159
Fig. 10. Comparison of the calculated and theoretical permeate ow rates versus
the number of diavolumes (A) and productivities versus purity (B).
relationships proposed to calculate J and Cw and the methodology
gives a precise estimation of DF performances. If necessary, the meth-
odology could be improved by taking into account HA size.
4. Conclusion
The evolution of the three performance criteria related to the
purication of HA from a complex semi-synthetic medium was
studied during dialtration step. The results showed that in the
chosen operating conditions (chosen on the basis of literature results),
the dialtration is suitable for the HA purication from a comp-
lex mixture. Indeed, a purity value near 90% can be reached after
7 diavolumes, giving a yield higher than 90%. The mass balance
equation applied to HA, minerals and peptides/proteins fractions
provided a satisfying estimation for HA yield and purity as a function
of the water volume introduced into the feed tank.
The evolution of the permeate ux during DF was surprising since
it increased during the rst seven diavolumes before to get steady.
The mineral fraction of the mixture revealed to be responsible for this
unexpected behavior, most probably due to a modication of HA
complex network structure. In this study, an empiric correlation that
links the mineral fraction concentration to the mass transport
coefcient k and wall concentration Cw was proposed. Thus, the
prediction of the evolution of the permeate ux during DF and the
calculation of the process productivity for a desired HA purity, can be
simply obtained from these correlations, the evolution mass balance
of minerals and the classical equation of permeate ux. To our kno-
wledge, such a result was never reported for HA or any other
biomolecule. This methodology for predicting the performance of
159
HA purication during DF was successfully validated with a HA
solution obtained by microbial culture of S. zooepidemicus using a
standard culture medium.
Acknowledgments
HA materials for this work were provided by Teoxane labora-
tories, Geneva.
Appendix A. Supporting information
Supplementary data associated with this article can be found in
the online version at http://dx.doi.org/10.1016/j.memsci.2015.04.024.
References
[1] J.R.E. Fraser, T.C. Laurent, U.B.G. Laurent, Hyaluronan: its nature, distribution,
functions and turnover, J. Intern. Med. 242 (1997) 2733.
[2] G. Kogan, L. olts, R. Stern, P. Gemeiner, Hyaluronic acid: a natural biopoly-
mer with a broad range of biomedical and industrial applications, Biotechnol.
Lett. 29 (2007) 1725.
[3] J. Chen, J. Chen, Z. Xu, Rheological and biological characteristics of hyaluronic
acid derivative modied by polyethylene glycol, J. Wuhan Univ. Technol.
Mater Sci. Ed. 23 (2008) 617621.
[4] W.-C. Huang, S.-J. Chen, T.-L. Chen, Production of hyaluronic acid by repeated
batch fermentation, Biochem. Eng. J. 40 (2008) 460464.
[5] A. Shiedlin, R. Bigelow, W. Christopher, S. Arbabi, L. Yang, R.V. Maier, et al.,
Evaluation of hyaluronan from different sources: Streptococcus zooepidemicus,
Rooster Comb, Bovine Vitreous, and Human Umbilical Cord, Biomacromole-
cules 5 (2004) 21222127.
[6] T. Yamada, T. Kawasaki, Microbial synthesis of hyaluronan and chitin: new
approaches, J. Biosci. Bioeng. 99 (2005) 521528.
[7] A.M.B. Pires, A.C. Macedo, S.Y. Eguchi, M.H.A. Santana, Microbial production of
hyaluronic acid from agricultural resource derivatives, Bioresour. Technol. 101
(2010) 65066509.
[8] O. Smidsrd, A. Haug, Precipitation of acidic polysaccharides by salts in
ethanolwater mixtures, J. Polym. Sci. Part C Polym. Symp 16 (1967)
15871598.
[9] K.J. Reddy, K.T. Karunakaran, K.R.S.S. Rao, Enhanced hyaluronic acid produc-
tion by a mutant strain, 3523-7 of Streptococcus zooepidemicus, Curr. Trends
Biotechnol. Pharm. 5 (2011) 14731479.
[10] J.B. Woolcock, The capsule of Streptococcus equi, J. Gen. Microbiol. (1974)
372375.
[11] A. Veide, T. Lindbck, S.-O. Enfors, Recovery of -galactosidase from a poly
(ethyleneglycol) solution by dialtration, Enzyme Microb. Technol. 11 (11)
(1989) 744751.
[12] E. Nakkeeran, K. Venkatesh, R. Subramanian, S.U. Kumar, Purication of
Aspergillus carbonarius polygalacturonase using polymeric membranes, J.
Chem. Technol. Biotechnol. 83 (7) (2008) 957964.
[13] Z. Li, W. Youravong, A. H-Kittikun, Separation of proteases from yellown tuna
spleen by ultraltration, Bioresour. Technol. 97 (18) (2006) 23642370.
[14] H. Zhou, J. Ni, W. Huang, J. Zhang, Separation of hyaluronic acid from
fermentation broth by tangential ow microltration and ultraltration, Sep.
Purif. Technol. 52 (2006) 2938.
[15] L.K. Nielsen, W. Chen, S.M. Saldana Esteban, Production of hyaluronic acid, WO
2009/026635 A1, 2009.
[16] Z.-W. Lai, Medium formulation and impeller design on the biosynthesis of
high molecular weight hyaluronic acid by Streptococcus zooepidemicus ATCC
39920, Afr. J. Microbiol. Res. 5 (2011) 21142123.
[17] N. Oueslati, P. Leblanc, C. Harscoat-Schiavo, E. Rondags, S. Meunier, R. Kapel,
et al., CTAB turbidimetric method for assaying hyaluronic acid in complex
environments and under cross-linked form, Carbohydr. Polym. 112 (2014)
102108.
[18] G. Foley, Minimisation of process time in ultraltration and continuous
dialtration: the effect of incomplete macrosolute rejection, J. Membr. Sci.
163 (1999) 349355.
[19] Y. Mo, T. Takaya, K. Nishinari, K. Kubota, A. Okamoto, Effects of sodium
chloride, guanidine hydrochloride, and sucrose on the viscoelastic properties
of sodium hyaluronate solutions, Biopolymers 50 (1999) 2334.
[20] K.W.K. Yee, D.E. Wiley, J. Bao, A unied model of the time dependence of ux
decline for the long-term ultraltration of whey, J. Membr. Sci. 332 (2009)
6980.
[21] L.-T. Goh, Fermentation studies of hyaluronic acid production by Streptococcus
zooepidemicus. http://espace.library.uq.edu.au/view/UQ:296914, 1998
(accessed 16.06.14).