Anda di halaman 1dari 5



DOI: 10.5504/bbeq.2012.0061


Venko Beschkov, Tsvetan Sapundzhiev, Ivan Angelov
Bulgarian Academy of Sciences, Institute of Chemical Engineering, Sofia, Bulgaria
Correspondence to: Venko Beschkov


The anaerobic conversion of glycerol by the bacterium Klebsiella sp. is a well-known process. According to its metabolism
glycerol is converted into different products, e.g. 2,3-butanediol, 1,3-propanediol, acetic acid, formic acid, lactic acid and
biogas. The latter is produced by two competitive ways: decarboxylation of acetic acid or self-reduction of carbon dioxide by
hydrogen, acetic acid, carbon dioxide and hydrogen being intermediate products of one of the competitive metabolic routes.
Which of these compounds is a desired final product depends on the purpose of the fermentation.
In the present work a non-structured mathematical model is developed to describe these competitive processes. The model
consists of nine ordinary non-linear differential equations for the kinetics of some selected more important reactions from the
metabolic pathway. The model takes into account the microbial growth, the pH drop caused by the organic acid formation and
the resulting inhibition of the methanogenic microbial activity. The pH-optima of the enzymes are approximated by Gaussian
distribution. The enzyme sensitivity toward pH-variations is presented by the half-width of the Gaussian curve.
A large set of experimental data for glycerol anaerobic digestion by Klebsiella sp. in a baffled multistage reactor was used for
verification of the model. The comparison of the experimental data with the model results revealed the number of reactor stages,
sufficient for complete methanization and to determine the optimum conditions for 2,3-butanediol or methane production.
Biotechnol. & Biotechnol. Eq. 2012, 26(5), 3244-3248
Keywords: biogas production, glycerol, enzyme activity,
mathematical modeling
Notations: Ac: concentration of acetic acid; BD: concentration
of 2,3-butanediol; bi: enzyme activity dependence on
pH; CH4: concentration of methane; CO2: concentration
of carbon dioxide; F: concentration of formic acid; Km:
saturation constant; Kx: biomass decay constant; ki: rate
constants of intermediate and final processes; pH: pH value;
Pv: concentration of piruvate; S: concentration of glycerol; t:
time; X: concentration of biomass; YX/S: yield coefficient of
micobial production
Greek symbols: : rate constant of substrate conversion; :
specific growth rate; max: maximum specific growth rate; s2:
width of the Gaussian curve at its semi-height
Subscripts: ac: acetic acid; Ac1: acetic acid decarboxylation;
BD: 2,3-butanediol; bd1: degradation rate constant for
2,3-butandiol; CH4: methane; CO2: carbon dioxide; F: formic
acid; Pv: piruvate; S: glycerol; 0: initial values

Crude glycerol is the main waste product from biodiesel
manufacturing. Its amount is equivalent to the methanol used
for the trans-esterification and exceeds the traditional market
demands. On the other hand, this waste product contains about
20 % water and it is contaminated by the catalyst and some
methanol. Since there is market demand only for pure glycerol

and its price is not high, the purification of the waste product is
not economically feasible.
There are many chemical compounds produced from
glycerol by microbial methods. Different bacteria (from the
genera Klebsiella, Clostridium, Enterobacter, etc.) are capable
of metabolizing glycerol, leading to main basic products
with some differences in the metabolic by-products (4). The
metabolic scheme for glycerol conversion by the bacteria
Klebsiella sp. is shown in Fig.1 (11). Another way for waste
glycerol utilization is to produce energy in the form of biogas
(4, 9).
This biogas could be used for partial energy supply of
the main biodiesel plant. Methanogenic bacteria can produce
methane from the metabolic products of the other bacteria
listed above. Methane can be produced either after acetic acid
decarboxylation or carbon dioxide reduction by hydrogen:

(Eq. 1)

CO2 + 4H2 CH4 + 2H2

(Eq. 2)

A major drawback of this application is the rapid

accumulation of carboxylic acids leading to strong inhibition
of the methanogenesis and shift to production of gas with
very low methane content (8, 9). Compared to the traditional
complex substrates for biomethanation (manure, municipal
solid waste, activated sludge, etc.) glycerol has a very simple
molecule and therefore it quickly yields intermediates and final
products like organic acids and alcohols (cf. Fig.1). The acids
rapidly lower the pH to inhibit the methanogenic bacteria.
Biotechnol. & Biotechnol. Eq. 26/2012/5

One possibility to overcome, at least partially, the effect of

this strong acidification on the biogas production is to separate
the zones of inhibition from the ones of methanation. For this
purpose a baffled bioreactor separated into compartments can
be used (Fig.2). It was reported earlier (5), that this type of
digesters are stable toward various disturbances in the feed,
pH, temperature variations, etc.
There are different data on mathematical modeling of
anaerobic digestion for methane production, considering more
complicated substrates, e.g. agro-waste (6, 7, 12, 13, 14). They
are all concerned with the traditional case of waste utilization
as a substrate with the well-known scheme of consecutive
hydrolysis, acidogenesis, acetogenesis and methanation. To the
best of our knowledge, there have been no efforts to describe
the methanation process with glycerol as a substrate.

Fig. 1. The metabolic pathway of glycerol anaerobic digestion by Klebsiella

sp. Reprinted from Saxena et al. (11) with permission from Elsevier.
To gas-holder
Outlet flow

Fig. 2. Sketch of the experimental set-up.

Biotechnol. & Biotechnol. Eq. 26/2012/5

The purpose of the present paper is to develop a

mathematical model for glycerol conversion into biogas and
other products and to estimate the conversion rates of these
competitive processes based on our experimental data in a
baffled multistage bioreactor.

Materials and Methods

Crude glycerol remaining from biodiesel production containing
a 20 % mass fraction of water was used. The acidity of this
substrate corresponded to pH 5.5. No pH adjustment was
carried out.
Experimental set-up
The sketch of the experimental set-up is shown in Fig. 2. It
consists of eight rectangular sections with equal volumes of
33L separated by stationary baffles with static mixers. The
reactor was initially inoculated by activated sludge using
stillage from ethanol production as a carbon source. The
inoculum and the initial feed were equal for each section. The
gas space above the liquid was common for the whole reactor.
After full development of the biomass, indicated by the release
of combustible biogas, the feeding of the reactor with crude
glycerol started in the first section. The excess liquid from
each section enters the next one through overflows or below
the separating baffles and leaves the apparatus from the end
section. The process was carried out in a fed-batch mode, with
feeding four times daily. The temperature was maintained at
32 C by thermostat with a sensor dipped in the fermentor.
Different amounts of crude glycerol (from 0.1L/day to 1L/
day) were added. The produced gas was periodically measured
after collection in a gas-holder under water. Samples from each
compartment were taken regularly. They were analyzed for the
substrate and intermediates and also for pH.
Analytical procedures
An HPLC chromatograph, Perkin Elmer Series 10, with a BioRad column for organic acid analysis (Aminex HPX-87H)
was used. The organic acids were determined by a Knauer
UV-detector at 210 nm, whereas for analyses of alcohols an
RI detector was used. Solution of 0.005 mol/L sulfuric acid
was used as a mobile phase at an elution flow rate of 0.6l/min
at 65C. The analyzed intermediates were identified by their
retention times compared with added standards. The pH was
measured off-line by pH-meter. No lactic acid was detected
in the broth. The methane content was evaluated by an Orsat
Mathematical model
As a basis for the mathematical model the metabolic pathway
(cf. Fig. 1) was used. It was partially simplified to the following
reaction routes, assuming that the other intermediate processes
are sufficiently fast and therefore not rate-determining (cf.
Fig. 3). Hence, the formation of biogas, 2,3-butanediol and
lactic acid are three competitive metabolic routes. The acid
accumulation (including that of acetic and formic acid) leads

to a pH drop which may inhibit the methanogenic activity.

For the abovementioned purpose a non-structured model was
developed, consisting of a set of non-linear ordinary differential
equations with the corresponding initial conditions:
= X K x X, = max
, t = 0, X = X 0
S + Km
= Y X/S S bS X.S, t = 0, S = S 0
= bS X.S (k BD bBD + k Ac b Ac + k F bF )XPv, t = 0, Pv = Pvo
= k F bF X.Pv k CH4 bCH4 X.F 2 k CO2 X.FbCO2 , t = 0, I = 0
= k bd bbd XPv k bd1bbd1 BD, t = 0, BD = 0
= k ac bac XPv k Ac1b Ac1 Ac, t = 0, Ac = 0
= k Ac1b Ac1 Ac + k CH4 bCH4 X.F 2 , t = 0, CH4 = 0
= k Ac1 Ac + k CO2 X.FbCO2 , t = 0, CO2 = 0
bi = exp( (pH pH i ) 2 /(2 2 )

(Eq. 3)

The net pH value for the fermentation broth was calculated

in the model based on the organic acid concentrations according
to their dissociation constants taken from the literature (10).
Since no lactic acid was detected, the overall kinetics was
considered without lactic acid formation and degradation.
Selected pH optima for the different intermediate reactions in
Enzyme (reaction)
pH-optimum Reference
Glycerol to piruvate,
Piruvate decarboxylase, for bAc
Piruvate to hydrogen and CO2, for bF
Formate to hydrogen and CO2, for bCO2
Acetate to methane, for bAc1
Hydrogen and CO2 to methane, for bCH4
2,3-Butanediol from piruvate, bBD
2,3-Butanediol biodegradation, bBD1

The system was solved using a 20-sim.3.4 dynamic

simulator (a Dutch software) coupled with an optimization
program for seeking the kinetic parameters according to the
BroydonFletcherGoldfarbShanno gradient search method
(2). As an objective function, the sum of the squares of the
differences of the experimental and the calculated values for
glycerol, pH, 2,3-butanediol and acetic acid was selected:

Fig. 3. Simplified metabolic pathway described by the mathematical model

(Eq. 3).

The used symbols are described in the Notations. Note

however, that the rate constants ki are multiplied by factors bi
taking into account the activities of each enzyme in dependence
of the pH. These dependences were approximated by Gauss
distribution curves with a pH-sensitivity presented by the
half-width s2. The pH-optima for the involved enzymes are
taken from the literature and given in Table 1. We are fully
aware that these values might not be correct because of the
different strains they are assigned to, but for the purpose of
demonstrative modeling they are sufficient.

(Eq. 4)

The number of the kinetic parameters is 12. It appeared

during the preliminary numerical experiments that the
solutions are not sensitive to six of them (i.e., a, Kx, Km, YX/S,
kCO2 and kCH4) and could be considered constant. That is why
six kinetic parameters had to be determined. Having in mind
the complexity of the problem and the lack of kinetic data
for the particular case, we assumed the obtained values as a
qualitative estimation with indication for the model adequacy
because of the constant and non-variable values of the
estimated parameters. The concentrations of the components
were presented as normalized peak area form the HPLC
chromatograms, whereas the pH values were directly taken
from the pH-meter readings.
The effect of enzyme sensitivity toward pH variations was
also studied by varying the half-width of the Gaussian curves
for enzyme activity. Two different values for s2 were selected.
Additionally, the initial piruvate concentration Pvo was varied
in the numerical experiments.

Results and Discussion

Enzyme sensitivity
The effect of enzyme sensitivity toward pH drop and its
variation in time together with the other components of the
broth are shown in Fig. 4. Lines 14 denote the comparison
Biotechnol. & Biotechnol. Eq. 26/2012/5

between the experimental data presented as peak area in

HPLC measurements and the model curves. The lines 58 are
qualitative illustration of variation for some other quantities.
The same experimental results were fitted for enzyme
activities with different half-width s. Very good coincidence
with the experimental data was observed in the time profiles for
2,3-butanediol (BD), glycerol (S), acetic acid (Ac) and pH. It
was established that there is practically no difference between
the two fits. It was demonstrated that the enzyme activity for
the rate constant kCH4 increases during the process because of
the slight increase of pH due to the conversion of formic acid
and acetic acid.

Comparison of evaluated parameters for the first two
compartments of the bioreactors (experimental data from
different run)


expected because of the advanced conversion of glycerol into

intermediates (e.g. pyruvate) in the following consecutive
compartments. It appeared that the modeling was meaningless
for the further compartments because many of the substances
vanished and the objective function became more and more
undefined. Next, the biogas release for these compartments
became less and less intensive. Hence, our modeling can
predict the number of compartments required for complete
biogas release. Another advantage of this modeling is to
estimate the time span for optimum concentrations and yields
of some products, like 2,3-butanediol, and to avoid further
conversion and degradation.

2 8

Fig. 4. Time profiles and comparison with experimental data as HPLC peak
area for compartment 1 at s2 = 0.25. Line 1 (): glycerol; line 2 (o): acetic
acid; line3 (o): 2,3-butanediol; line4 (): pH; line5: methane volume; line6:
piruvate; line 7: enzyme activity, (bCH4) of methane formation from formic
acid; line8: biomass concentration.

mmax, h-1
kF, h-1
kBD, h-1
kBD1, h-1
kAc, h-1
kAc1, h-1
Pvo, peak area

Compartment 1

Compartment 2

Variation of enzyme activity

The dependence of enzyme activities on the pH changes during
the fermentation process and their variation in time is shown in
Fig.5. Some of the activities rise, some fall depending on the
pH optima of the enzymes and the pH variation in time.

The comparison of the estimated kinetic parameters for

these two cases is shown in Table2. There is a good coincidence
with some discrepancies for the piruvateformiate conversion
rate constant kF and to the acetic acid decarboxylation kAc1.
Comparison of estimated parameters for compartment 1 at
different enzyme sensitivities s2
mmax, h-1
kF, h-1
kBD, h-1
kBD1, h-1
kAc, h-1
kAc1, h-1
Pvo, peak area

s2 = 0.25

s2 = 0.45

Estimated parameters for two different compartments

Comparison of the kinetic parameters evaluated for
compartments 1 and 2 is shown in Table3. One can see that,
except kBD, there was a reasonable coincidence of the values.
The difference in the initial piruvate concentrations was
Biotechnol. & Biotechnol. Eq. 26/2012/5


Fig. 5. Simulated variation of different enzyme activities (bi) during the

fermentation. Line1: bCH4; line2: bPv; line3: bF; line4: bBD.


The developed mathematical model, although very simplified,

can describe the glycerol competitive conversion into biogas
and 2,3-butanediol in the presence of bacteria from the genus
Klebsiella. This approach can be successfully applied for

other microbes provided the metabolic pathway is known.

Because of the very complicated metabolism and the lack of
knowledge about the very kinetics of each reaction with the
related rate constants, this modeling has mostly demonstrative
and qualitative value.
However, the modeling can serve as an indication to
estimate the number of the compartments in a baffled reactor
necessary and sufficient for full glycerol conversion into
biogas and other metabolites of practical importance.


This work was financially supported by Grant DTK-02/36/2009

from the National Science Fund of the Republic of Bulgaria.


1. Angelidaki I., W. Sanders (2004) Rev. Environ. Sci.

Biotechnol., 3, 117-129.
2. Bazaraa M.S., Sherali H.D., Shetty C.M. (1990)
Nonlinear Programming, Theory and Algorithms, John
Wiley & Sons Inc., New York.
3. BRENDA The Comprehensive Enzyme Information
System, <
4. Da Silva G.P.,Mack M. , Contiero J. (2009) Biotechnol.
Adv., 27, 30-39.
5. Grobicki A., Stuckey D.C. (1992) Water Res., 26, 371378.


6. Kalyuzhnii S.V., Fedorovich V.V. (1997) Wat. Sci.

Technol., 36, 201-208.
7. Kleinstreuer C., Poweigha T. (1982) Biotechnol. Bioeng.,
24, 1941-1951.
8. Kolesrov N., Hutan M., palkov V., Lazor M. (2010)
In: Proceedings from the 37th International Conference of
SSCHE, Tatranske Matliare, Slovakia, 1126-1139.
9. Lpez J..S., De los ngeles Martn Santos M., Prez
A.F.C., Martn A.M. (2009) Bioresource Technol., 100,
10. Nser
Grundstoffindustrie, Leipzig, pp. 354-355.
11. Saxena R., Anand P., Saran S., Isar J. (2009) Biotechnol.
Adv., 27, 895-913.
12. Schn M. (2009) Numerical modeling of anaerobic
digestion processes in agricultural biogas plant, DSc
Thesis, Leopold-Franzens-Universitt Innsbruck, pp. 2225, 32-34.
13. Simeonov I., Stoyanov S. (2003) Chem. Biochem.Eng.Q.,
17, 285-292.
14. Simeonov, I., Queinnec I. (2006) Control Engineering
Practice, 14, 799-810.
15. Stephenson M.P., Dawes E.A. (1971) J. Gen. Microbiol.,
69, 331-343.
16. Yankova S., Begova P., Beschkov V. (2010) In: Proceedings
from Linnaeus Eco-Tech10, Kalmar, Sweden, 960-964.

Biotechnol. & Biotechnol. Eq. 26/2012/5