Bioseparation 10: 221–227, 2002.

© 2002 Kluwer Academic Publishers. Printed in the Netherlands.

221

Kinetic studies of clavulanic acid recovery by ion exchange

chromatography

Marlei Barboza2 , Renata M.R.G. Almeida1 & Carlos O. Hokka1,∗

1 Department of Chemical Engineering, Federal University of São Carlos, Via Washington Luiz, km 235 – Cx.
Postal: 676 – CEP: 13565-905 São Carlos, SP Brazil; 2 Department of Chemical Engineering, UNAERP, University
of Ribeirão Preto, Avenida Costábile Romano, 2201 – Cx. Postal 98 – CEP 14096-380-Ribeirão Preto, SP Brazil

Received 3 January 2001; accepted in revised form 10 April 2002

Key words: adsorption kinetics, clavulanic acid, ion-exchange chromatography, recovery

Abstract
Clavulanic acid (CA) is a beta-lactamase inhibitor produced by strains of Streptomyces clavuligerus. Nowadays, the
combination of CA with amoxycillin is the most successful example of the use of a beta-lactam antibiotic sensitive
to beta-lactamases together with an inhibitor of these enzymes. Clavulanic acid is purified from fermentation
broth by a series of steps consisting mainly of two-phase separation processes such as liquid–liquid extraction,
adsorption or ion-exchange chromatography, among others. Amberlite IRA 400, a strong anion-exchange resin,
has a very high adsorption capacity for CA (Mayer et al. 1997). This resin can be pre-treated with NaCl (chloride
cycle), to remove selectively only those anions, which are able to displace chloride from the resin or with NaOH
(hydroxyl cycle), to remove all species of anions. In order to decide the best operating conditions for CA recovery
by ion-exchange resins and then to construct a model of this separation process, batch experiments were conducted
using Amberlite IRA 400 in the chloride cycle. These runs were carried out in a 200 ml stirred tank, at two different
initial solution pH, 6.2 and 4.0; the temperature was maintained at 10 ◦ C and 20 ◦ C during adsorption and 30 ◦ C
during the desorption step. It was possible, on the basis of these batch results, to model the separation process,
including the adsorption kinetics, equilibrium data and mass transfer limitations.
Abbreviations: CA – clavulanic acid

Introduction strains of some pathogenic microorganisms resistant

The beta-lactam family is the most important group of
antibiotics and constitutes the largest part of the mult-
ibillion dollar world antibiotic market. Approximately
60% of the total worldwide production of antibiotics
is of the beta-lactam type (Ghosh et al. 1996).
Bacterial resistance is one of the greatest
hindrances to the use of chemotherapeutic agents
and antibiotics to the control of infectious diseases.
Clavulanic acid (CA) is a beta-lactam antibiotic pro-
duced by a filamentous bacteria (actinomycete) named
Streptomyces clavuligerus. The British drugs company
Beecham first reported its isolation in their laborator-
ies in 1976. CA shows a broad antibacterial spectrum
but the level of activity is relatively low. It is, however,
a potent inhibitor of the beta-lactamases produced by
to beta-lactam antibiotics. These enzymes are able to
catalyze the hydrolysis of molecules containing a beta-
lactam ring. Clavulanic acid can block this resistance,
as a result, in the presence of low concentrations of
CA, many of these beta-lactamase-producing organ-
isms are rendered almost as sensitive to penicillins and
cephalosporins now commercially available as non-
beta-lactamase-producing strains (Brown et al. 1976).
Nowadays, the combination of CA with amoxycil-
lin is the most successful example of the use of a
beta-lactam antibiotic sensitive to beta-lactamase to-
gether with an inhibitor of these enzymes (Mayer and
Deckwer 1996).
The isolation of clavulanic acid from fermentat-
ive broth is performed in a series of steps. At the
end of fermentation, the broth is first clarified by
222
filtration or centrifugation. The most important step,
primary extraction from clarified broth, is based on
several two-phase separation methods. One of these
methods is the direct extraction of the filtrate with
organic solvent, producing an organic phase contain-
ing CA, which is subsequently isolated. An alternative
process, to avoid the use of solvent, consists of chro-
matographic adsorption techniques with non-ionic or
ionic adsorbers. This technique has developed rapidly
in recent decades, mainly since the introduction of
synthetic adsorbers (Butterworth 1982).
Adsorption techniques can be used in several steps
of the separation and purification of antibiotics. One
type is ion-exchange adsorption, which involves elec-
trostatic attraction of ionic components to sites on the
adsorbent surface with opposite charge. However, CA
is a compound that does not have any strongly hy-
drophobic group and exhibits low chemical stability.
These features are responsible for the low recuperation
levels of CA during purification.
An investigation of the use of Amberlite XAD
resins in the purification of CA from fermentative
broth was carried out by Mayer et al. (1996). Pre-
liminary studies had shown a very weak interaction
between CA and the apolar surface of these resins, as
a consequence of the chemical structure of this an-
tibiotic. A new system based on ion-pair formation
was developed. In this system the XAD matrices were
tested in combination with water-soluble quaternary
ammonium salts to form ion-pairs with the acid group
of the CA molecule. Comparative tests were run, using
the ion-pairing system alongside the traditional anion
exchanger Amberlite IRA 400 in its original chloride
form. The authors concluded that the ion-pair system
was a viable alternative for the purification of CA.
The resin Amberlite XAD-4 with the quaternary am-
monium salts performed better than the ion-exchange
Amberlite IRA 400. However, the system proposed
by the authors increases the cost of the purification
process, owing to the use of quaternary ammonium
salts.
In the work reported here, in order to find the best
operating conditions and with the aim of modeling
the separation process by ion exchange resins, batch
experiments for CA recovery were conducted with
the ion-exchange resin Amberlite IRA 400. Adsorp-
tion isotherms were determined and kinetic studies,
taking into account mass transfer limitations, were
performed. A model of the CA adsorption process,
including these phenomena, is proposed.
Materials and methods
Adsorbate
Clavulanic acid was obtained from fermentation broth
produced by Streptomyces clavuligerus ATCC 27064.
Fermentation medium contained: glycerol (15.0 g);
malt extract (10.0 g); Samprosoy 90 NB
(10.0 g);
yeast extract (1.0 g) MgSO4 .7H2 O (0.75 g); K2 HPO4
(0.80 g); soybean oil (1.0 g); trace elements solution
(1 ml); distilled water (1 l), pH 6.8 with NaOH/HCl.
Before the adsorption process the fermentation broth
was pretreated as follows: first the broth was cent-
rifuged to obtain a clear solution free of cells; then
the solution pH was brought down (3.5–4.0) in order
to precipitate the soluble solids and proteins; finally
the solution containing the product was centrifuged
again and filtered through analytical paper filters, to
eliminate suspended impurities.
Preparation of stationary phase
The stationary phase was the anion exchanger Am-
berlite IRA 400, kindly supplied by Rohm and Haas.
This resin can be regenerated with NaCl (chloride
cycle) or with NaOH (hydroxyl cycle). During re-
generating cycle with NaCl it is possible to remove
only those anions, such as SO−2 − −3
4 , NO3 , PO4 that are
capable of removing the chloride from these matrices.
However, when less strongly bonding anions need to
be removed it is necessary to use the resin in its hy-
droxyl cycle, in which every anion in solution can
be removed (Rohm and Haas). In this work the resin
was pretreated with NaCl 10% (w/v), to work with
the chloride cycle; then the resin was washed several
times with deionized water to eliminate the excess
ions. NaCl 2% and 10% (w/v) was utilized as the
eluent throughout.
Analytical methods
The clavulanic acid concentrations were determined
by high-performance liquid chromatography (HPLC),
as described by Foulstone and Reading (1982). An
HPLC device with a Photodiode Array detector (Wa-
ters 996 PDA) was utilized with a reversed-phase
column (C-18 µ-Bondapak 3.9×300 mm). The HPLC
equipment was operated at 28 ◦ C, with a flow rate
of 2.5 ml/min, and was calibrated against solutions
of the pharmaceutical product ‘Clavulin’ (tablets con-
taining 125 mg of potassium clavulanate and 500 mg
of amoxycillin), produced by SmithKline Beecham

Laboratórios Ltda., Rio de Janeiro, Brazil. The mobile
phase was composed of a 0.1 M KH2 PO4 buffer solu-
tion containing 6% of methanol and phosphoric acid,
to adjust to pH 3.2.
Determination of adsorption isotherm
The adsorption isotherm run was carried out in a
shaker at 10 ◦ C until adsorption equilibrium was at-
tained. Different concentrations were obtained by di-
luting the sample in deionized water and two glasses
containing 5 ml of CA solution and 0.25 g of the
stationary phase (wet basis) were used for each con-
centration. The initial and final concentrations were
analyzed by HPLC, as described above. The Lang-
muir model was utilized to describe the equilibrium
behavior.
Batch runs
The experiments to study the adsorption and desorp-
tion kinetics of clavulanic acid in IRA 400 resin
were carried out in a 200 ml isothermic stirred glass
vessel. The vessel contained 100 ml of clavulanic
acid solution and 5.0 g of ion-exchanger resin (wet
basis). Samples were withdrawn periodically and the
antibiotic concentrations were determined.
Mathematical model
The kinetic behaviors of the adsorption and desorption
of clavulanic acid on the resin are described as follows.
Kinetic model
Adsorption of AC on the resin:
(Resin)+ Cl− + CA− k1 Complex + Cl−
−→
← −
k2
Complex = Resin+ Cl− CA−
The rate expressions describing the intrinsic kinetic
model are presented as follows:
rq = k1 Ci (qm − qi ) − k2 qi (1)
where k1 and k2 are the intrinsic kinetic constants and
Ci is the antibiotic concentration in the solution at any
position inside the resin, qi is the antibiotic concen-
tration adsorbed on specific site of the resin and qm
223
is the maximum adsorption capacity of the resin. At
adsorption equilibrium:
qm C ∗
q∗ = (2)
KD + C ∗
where q ∗ is the equilibrium concentration in the solid
phase (gAC /gres ), C ∗ is the corresponding equilibrium
concentration in the liquid phase, qm is the maximum
capacity of the resin and KD is the dissociation con-
stant. The proposed model is based on ideal surfaces
and predicts that the adsorption surface is energetically
ideal. So, Equation (2) is the Langmuir equilibrium
model.
Desorption occurs when the equilibrium is dis-
placed by changing the process conditions such as
salt concentration. In order to describe the desorption,
assumption similar to that for adsorption can be made:
Complex + Na+ Cl− k3 Resin+ Cl− + CA− + Na+
−→
The rate expression corresponding to the kinetic of
desorption is represented by Equation (3), where k3 is
the intrinsic kinetic constant related to the desorption
process.
rq = −k3 qi (3)
Mathematical modelling for the adsorption step in a
stirred tank reactor
Let Vs be the volume of adsorber immersed in a
volume V1 of liquid with CA dissolved at an initial
concentration C0 , contained in a perfectly stirred re-
actor. The solute (CA) diffuses into the resin particles
and is adsorbed until the equilibrium is reached. For
the formulation of the model it is assumed that: the
resin particles are spherical; the diffusion of CA in
the solid particles follows the Fick’s law; the diffu-
sion process is in the r direction only; the adsorption
takes place in isothermal conditions. The adsorbed CA
is assumed to be in equilibrium with that in the pore
fluid at each radial position within the particle. The
following conservation equations and boundary equa-
tions are used to describe the CA uptake kinetics for
spherical particles of radius R in a closed batch sys-
tem. With these assumptions the adsorption process
in a constant volume stirred tank can be described by
Equation (4).
dCb 3 Vs
=− ks (Cb − Cs ) (4)
dt R V1
224
Table 1. Parameters of the Langmuir isotherm
The initial condition for Equation (4) is:
T pH qm KD r2
t = 0 → Cb = C0 (5) (◦ C) −1
(gAC .gres ) (g l−1 )

The differential material balance inside the solid 10 4.0 0.0077 0.1364 0.996
particles, where adsorption takes place on the porous 10 6.2 0.0114 0.0791 0.976
surface is: 20 6.2 0.0092 0.0906 0.987

2 
∂Ci ∂ Ci 2 ∂Ci ∂qi
εp = Def εp 2
+ − (1 − p )
∂t ∂r r ∂r ∂t
(6)

The initial and boundary conditions associated

with the diffusing process inside the solid particles are,
respectively:

t = 0 → Ci = qi = 0 (7)

∂Ci ks
r=R→ = (8)
∂r εp Def (Cb| − Cs )

∂Ci
r=0→ =0 (9)
∂r
The diffusion equation inside the particles was dis-
cretized and solved using the method of orthogonal Figure 1. Non-linear regression of the experimental data of the ad-
sorption isotherm of the clavulanic acid in the pH’s of 6.2 and 4.0 in
collocation where the boundary condition referring to the temperature of 10 ◦ C and 20 ◦ C.
the film resistance was used as a collocation point
(Villadsen and Michelsen 1978). To integrate the two When the values of i in the simplex vortexes and
differential equations, the code DASSL was used (Pet- their mean values i satisfied the inequality (11) the
zold 1989). In order to determine the effective diffu- optimization was considered completed.
sion coefficient (Def ), the film coefficient (ks ) and the  1/2
m
intrinsic kinetic constant of adsorption (k1 ), the Nedler ¯ i )2
(i − 
and Mead (1965) method for parameter optimization  
 1  <e (11)
was utilized, with simultaneous numerical solution of  m+1 
the set of differential equations. The optimization was
performed by the least square method algorithm to
minimize the objective function defined by Equation where m is the number of parameters to be optimized
(10): and e is the convergence parameter, settled at 10−5 .


N
= (Cb − Cexp )2 (10)
1
Value for k3 was obtained following the same
algorithm. The bulk concentration in the tank, the
concentration inside the resin and the amount of CA
adsorbed were used as initial and boundary conditions.
The kinetic model used was that one described by
Equation (6). In this case Def and ks are considered
constants and the variable to be estimated was k3 .
Results and discussion
Equilibrium
The adsorption isotherm of CA on the Amberlite IRA
400 operated in the cloride cycle is shown in Figure
1 at two pH values. The symbol is the experimental
data and the solid line is the fit to the data, obtained by
non-linear regression, and given in Table 1. The Lang-
muir model fitted the experimental data very well and
 225
Table 2. Transport parameters and intrinsic kinetics parameters for adsorption of CA in stirred tank

Run T pH Co k1 k2 Def 105 ks 102 Bi

(◦ C) (mg l−1 ) (Lg−1 min−1 ) (min−1 ) (cm2 s−1 ) (cm s−1 ) (−)

1 10 6.2 105.1 1.70 0.13 9.0 5.3 29.4

2 10 6.2 937.0 1.70 0.13 8.7 6.7 38.5
3 10 6.2 1420.0 1.70 0.13 0.95 5.9 310.5
4 10 4.0 121.2 1.50 0.20 8.7 4.9 28.2
5 20 6.2 116.3 1.70 0.15 9.5 6.2 32.6

can describe the behavior of the ion exchange between

clavulanic acid and Cl− of the resin.
The effect of pH on the equilibrium of adsorption
was examined. The decrease of pH produces an in-
crease in the capacity of adsorption (qm values) and
in the affinity of adsorption (KD values). The ion
exchange of CA on Amberlite IRA-400Cl is favored
by decreasing the temperatures. These results indic-
ate that the adsorption of CA on the resin Amberlite
IRA-400 is exothermic.

Determination of the effective pore diffusion and

intrinsic kinetic parameters

Another key parameters are the effective pore diffu-

sion coefficient (Def ) and intrinsic kinetic parameters
(k1 and k2 ). The mass balance equations for clavulanic
acid in the fluid and in the particles (Equations (4)–(6))
and the shape of the isotherm were utilized to perform
the non-linear regression of the adsorption data. For
this regression the isotherm parameters were used. The
values of pore diffusion coefficient (Def ), the liquid
film mass transfer coefficient (ks ) and intrinsic kinetic
parameter, k1 , (k2 = k1 .KD ) were obtained by optim-
ization of the model by the least-squares method. The
parameters values obtained are presented in Table 2.
Figure 2 shows the comparison between experi-
mental and calculated values of the clavulanic acid
adsorption data.
It was observed in these experiments that when
initial pH is 6.2 or 4.0 the time to reach the adsorp-
tion equilibrium was almost the same. From the results
obtained it was observed that the effective diffusivity
is dependent on pH, temperature and initial concen-
tration of CA. This behavior was also observed by
several authors (Garke et al. 1999; Komiyama and
Smith 1974; Mayer et al. 1997). Garke et al. (1999),
studying the Lisosyme adsorption on ion exchange
resin, verified that Def decreases with the increase
Figure 2. Simulated and experimental data of the adsorption kin-
etic of clavulanic acid on the resin Amberlite IRA 400-CL at T=10
◦ C. Continuous line simulated values for pH=6.2 and dot line for
pH=4.0.
run 1;  run 2;  run 3;  run 4 (pH=4.0);  run 5.
of enzyme concentration. The authors explained that
a high protein concentration leads to an increase of
the steric interactions between the protein molecules
and the protein molecules with the resin pores. The
values obtained by Mayer et al. (1997) in a system
CA-resin also show that calculated Def decreased with
an increasing clavulanic acid initial concentration.
Komiyama and Smith (1974) concluded that the
relative importance of surface diffusion increases as
the adsorbate concentration decreases. In the present
work, two cases can be considered: first, prob-
ably other components of the medium are present in
amounts proportional to the CA concentration. These
compounds can hinder the resin pores impairing the
diffusion process; second, the effect of surface diffu-
sion can be neglected.
226
Table 3. Intrinsic kinetic parameter and effective diffusivity by desorption of CA at 30 ◦ C

Run NaCl pH k3 102 Def 102 ks 102 Bi

(%) (−) (min−1 ) (cm2 s−1 ) (cm s−1 ) (−)

1 2.0 6.2 9.5 9.0 7.7 42.5

2 2.0 6.2 9.5 9.0 5.0 27.8
3 10.0 6.2 1.0 1.0 1.0 52.6
4 2.0 4.0 9.7 8.3 6.4 38.5
5 2.0 6.2 9.0 9.0 7.2 34.4

The external film mass transfer coefficient was

constant for all runs. This probably happens due to the
fact that the range of conditions in which the runs were
carried out did not limit the process and this fact makes
no sense to the optimization simplex method. The res-
istance to mass transfer through the boundary layer
film on the spherical resin particles proved to be neg-
ligible and this phenomenon is reflected by the large
values of the Biot number obtained in the process,
listed in Table 2. The Biot number (Bi = ks R/Def )
is a relationship between mass transfer through the
external layer and mass transfer by diffusion inside
the particle. It means that if Biot is large, the slower
step, i.e. the limiting step, is the internal diffusion. The
intrinsic parameter k1 proposed in the process model
was not influenced by the temperature but with the
decrease of pH was noted the decrease value of these
Figure 3. Simulated and experimental data of the elution kinetic of
parameter. With the parameter k2 the opposite happens clavulanic acid on the resin Amberlite IRA 400-Cl with NaCl at
with relationship the k1 . Scopes (1988) shows that the T=30 ◦ C. Continuous line simulated values for pH=6.2 and dot line
pH of the microenvironment of an ion exchange resin for pH=4.0.
run 1;  run 2;  run 3;  run 4 (pH=4.0);  run 5.
is not the same as that measured in solution. This be-
havior is known as Donnan effect that can attract or
repulse the protons in the matrices of the adsorbent. due to the degradation of CA occurring during the
In general, the pH in the matrices of an anion ex- contact time of the adsorption step. However it can
change resin is one unit higher than in the solution be observed in Figure 2 that the time to reach the
around. This fact predicts that the clavulanic acid de- equilibrium in the adsorption step can be shortened.
gradation is stronger in the resin surface than in the In this way the process described in the present work
bulk. Therefore the effect of the degradation should can be utilized in systems with small contact time with
also be analyzed in these separation processes by ionic clavulanic acid minimizing the degradation effects of
change. pH.
After the adsorption step in runs 1 and 4, the The effective diffusivities, (Def ), evaluated were
step of elution with NaCl was carried out. The elu- the same for runs 1, 2 and 5. It was expected that
tion results obtained were utilized to adjust the kinetic the parameters Def were approximately the same that
parameter, k3 , the effective diffusion coefficient, Def obtained in the adsorption. The low k3 values obtained
and the film coefficient, ks . The adjusted parameters are probably due to the high electrostatic force in the
are shown in Table 3. In Figure 3 the model fitted to resin sites attracting the clavulanic acid strongly.
the experimental results is shown. Thus it can be concluded that the initial pH=6.2
As shown in Figure 3, it is not possible to recover and T=10 ◦ C during adsorption step is more advant-
the total clavulanic acid that was adsorbed by the resin ageous because clavulanic acid is more stable in these

pH and adsorption is more favorable in this temperat-
ure. During the desorption stage it was concluded that
the diffusion step limited the process.
Conclusions
Analyzing the presented results it can be concluded
that the purification process of clavulanic acid with
the ion-exchange resin Amberlite IRA 400 pretreated
with NaCl is a promising process. The clavulanic acid
degradation really occurred, but the period of time
to reach the adsorption equilibrium is small enough
to apply this process as a step of the clavulanic acid
purification process.
It was evident from the results that the model
proposed to simulate the adsorption kinetic and the
desorption step considering adsorption kinetics, equi-
librium data and mass transfer limitations fitted the
experimental data very well, providing valuable in-
formation for further process optimization.
Acknowledgements
The authors gratefully acknowledge the financial sup-
port of FAPESP (Proc. 99/07693-2, 99/03279-7 and
98/11596-0).
References
Brown AG, Butterworth D, Cole M, Hanscomb G, Hood JD, Read-
ing C and Rolinson GN (1976) Naturally ocurring β-lactam
inhibitors with antibacterial activity. J. Antibiotics 29: 668–669.
Butterworth D (1984) Clavulanic acid: properties, biosynthesis and
fermentation. In: Vandamme EJ (ed.) Biotechnology of Industrial
227
Antibiotcs., Vol. 6, (pp. 225–235). Marcel Dekker, Inc., New
York.
Foulstone M and Reading C (1982) Assay of amoxicilin and clavu-
lanic acid, the components of augmentin, in biological fluids with
high-performance liquid chromatography. Antimicrobial Agents
Chemother. 22: 753–762.
Garke G, Hartmann R, Papamichael N, Deckwer WD and Anspach
FB (1999) The influence of protein size on adsorption kinet-
ics and equilibria in ion-exchange chromatography. Separ. Sci.
Technol. 34: 2521–2538.
Ghosh AC, Bora MM and Dutta NN (1996) Development in liquid
membrane separation of beta-lactam antibiotics. Bioseparation 6:
91–105.
Gouveia ER, Baptista-Neto A, Azevedo AG, Badino-Jr AC and
Hokka CO (1999) Improvement of clavulanic acid produc-
tion by strptomyces clavuligerus in medium containing soybean
derivatives. World J. Microbiol. Biotechnol. 15: 623–627.
Komyama H and Smith JM (1974) Surface diffusion in liquid-filled
pores. AIChE J. 20: 1110–1117.
Mayer AF, Anspach FB and Deckwer WD (1996) Purification of
clavulanic acid by ion-pairing systems. Bioseparation 6: 25–39.
Mayer AF and Deckwer WD (1996) Simultaneous production and
decomposition of clavulanic acid during streptomyces clavuli-
gerus cultivation. Appl. Microbiol. Biotechnol. 45: 41–46.
Mayer AF, Hartmann R and Deckwer WD (1997) Diffusivities
of clavulanic acid in porous sorption systems with ion-pairing.
Chem. Eng. Sci. 52: 4561–4568.
Nedler JA and Mead RA (1965) Simplex method for function
minimization. Comput. J. 7: 308–315.
Petzold L (1989) Differential Algebraic System Solver (DASSL)
Subroutine. Computing and Mathematics Research Division,
Lawrence Livermore National Laboratory, Livermore, Califor-
nia.
Scopes RK (1988) Protein Purification: Principles and Practice. 2nd
edition, Springer-Verlag, New York NY, 329 pp.
Villadsen J and Michelsen ML (1978) Solution of Differential
Equation Models by Polynomial Approximation. Prentice Hall,
Englewood Cliffs, NJ, 447 pp.
Address for correspondence: C.O. Hokka, Department of Chem-
ical Engineering, Federal University of São Carlos, Via Washington
Luiz, km 235-Cx. Postal: 676-CEP: 13565-905 São Carlos, SP
Brazil. Tel.: +55-16-260-8264; E-mail: hokka@power.ufscar.br