ww.sciencedirect.com

b i om a s s an d b i o e n e r g y 4 7 ( 2 0 1 2 ) 3 1 6e3 2 3

Available online at w

ht tp: / /www.elsevier .com/locate/biombioe

Effect of ammonia on the anaerobic hydrolysis of celluloseand tributyrin

Tania V. Fernandes a,*, Karel J. Keesman b, Grietje Zeeman a, Jules B. van Lier c

aSub-department of Environmental Technology, Wageningen University, PO Box 8129, 6700 EV Wageningen, The Netherlandsb Systems & Control Group, Wageningen University, PO Box 16, 6700 AA Wageningen, The NetherlandscSection Sanitary Engineering, Delft University of Technology, PO Box 5048, 2600 GA Delft, The Netherlands

a r t i c l e i n f o

Article history:

Received 10 November 2010

Received in revised form

4 September 2012

Accepted 12 September 2012

Available online 23 October 2012

Keywords:

Anaerobic hydrolysis

Ammonia

Manure

Cellulose

Lipids

Biomass

* Corresponding author. Tel.: þ31 317 485762E-mail addresses: Tania.Fernandes@wu

(G. Zeeman), j.b.vanlier@tudelft.nl (J.B. van L0961-9534/$ e see front matter ª 2012 Elsevhttp://dx.doi.org/10.1016/j.biombioe.2012.09.0

a b s t r a c t

Ammonia nitrogen is one of the most common inhibitors in the anaerobic digestion of

complex wastes containing high concentrations of ammonia like animal manures, black-

water and waste oil from gastronomy. The inhibiting effect of ammonia on methano-

genesis has been well established. In contrast, the knowledge on the effect of ammonia on

organic matter hydrolysis is rather limited. This study focuses on evaluating the effect of

ammonia on the hydrolysis of carbohydrates and lipids, which are commonly found in

biomass. Batch digestion of tributyrin and cellulose at varying ammonia concentrations

were performed, using biomass adapted to 4.9 g NH4þeN.l�1. From this experimental

study it was concluded that total ammonia nitrogen in the range of 2.4e7.8 g NH4þeN.l�1

(283e957 mg NH3eN.l�1) does not inhibit the hydrolysis of tributyrin or cellulose. This

result is further confirmed by mathematical analysis of the estimated variation of the first-

order hydrolysis constant as a function of the total ammonia concentration.

ª 2012 Elsevier Ltd. All rights reserved.

1. Introduction This un-dissociated ammonia form, unlike the ammonium

Ammonia nitrogen is one of the most common inhibitors

during anaerobic digestion of complexwastes containing high

concentrations of ammonia and/or proteins, like animal

manure, slaughterhouse wastewater and fish industry

wastewater [1e3]. Therefore, ammonia is frequently being

referred to as a primary cause of digester failure [4]. Total

ammonia is the ammonium ion (NH4þ) plus the free ammonia

(NH3), which form depends on the pH and temperature of the

solution. For anaerobic digesters operated at pH 7 and 35 �Cless than 1% from the total ammonia, is in the free ammonia

form, however, at the same temperature, but pH 8 the free

ammonia increases to 10%, as illustrated in Fig. 1.

; fax: þ31 317 482108.r.nl (T.V. Fernandes),ier).ier Ltd. All rights reserve29

ion, is able to diffuse through cell membrane, which has been

reported to inhibit cell functioning by disrupting the proton

and potassium balance inside the cell [5], therefore making

the free ammonia as the main inhibitory form for methano-

genesis [6]. The effect of high ammonia concentrations on

methanogenesis has beenwell established. It is known that its

inhibition will cause accumulation of intermediate digestion

products like volatile fatty acids (VFA) [6,7] at the same organic

loading rate, unless the hydraulic retention time (HRT) is

increased. In the anaerobic digestion literature, it is common

to find total ammonia inhibitory concentrations between 1200

and 4000 mg NH4þeN.l�1 [1,5,8e10]. Variation in the inhibitory

concentration was shown to be strongly dependent on the

Karel.Keesman@wur.nl (K.J. Keesman), Grietje.Zeeman@wur.nl

d.

Fig. 1 e Free ammonia and ammonium percentages

present in solution at 20, 35 and 55 �C and varying pH

(calculated from equation (1)).

b i om a s s a n d b i o e n e r g y 4 7 ( 2 0 1 2 ) 3 1 6e3 2 3 317

adaptation level of the biomass [11]. In contrast to methano-

genesis, knowledge on the effect of ammonia on the hydro-

lysis is rather limited [4]. This is of major importance since

hydrolysis is considered as the rate limiting step in the

anaerobic digestion of complexwaste, such as animalmanure

and agricultural residues [12]. van Velsen [6] and Zeeman [3]

showed that the hydrolysis of complex substrates is inhibi-

ted at high ammonia concentrations, however other

compounds present in animal manure, that were diluted at

the same rate, could have also caused the inhibition. Lu et al.

[13] investigated the influence of 0e16 g ammoniaeN.l�1, at

neutral pH, on mesophilic acidogenesis of protein-rich fish

waste and concluded that up to 8 g ammoniaeN.l�1 the

acidogenic efficiency was slightly reduced, whereas 16 g

ammoniaeN.l�1 resulted in accumulation of intermediates,

such as lactic and formic acid. El-Mashad [4] found a linear

relationship between the hydrolysis rate and the free

ammonia concentration during the batch digestion of liquid

cow manure at 50 �C and 60 �C and 1e3.7 g NH4þeN.l�1. The

experiments of El-Mashad [4], however, were conducted with

sludge that was adapted to 1 g NH4þeN.l�1 only. He verified

that enzyme production and not enzyme activity was affected

by the high ammonia concentrations. It was however unclear

if, similar to methanogenic biomass [6], the acidogenic

biomass that produces the enzymes for hydrolysis could

adapt to high ammonia concentrations. According to van

Velsen [11] and Angelidaki [14] the degree of ammonia inhi-

bition depends greatly on the degree of adaptation of the

inoculum to the ammonia. Hansen et al. [15] and Zeeman [3]

showed that the interaction between free ammonia, VFA

and pH can lead to an “inhibited steady state”, which is

a condition where the process is running stable over an

extensive period of time, but with a lower methane yield and

high VFA concentrations in the digester.

As most researches on ammonia inhibition of hydrolysis

were done with animal manure diluted to different concen-

trations, the effect of ammonia alone on hydrolysis could not

be determined.

Once, on the basis of experimental data and a first-order

hydrolysis model as in Angelidaki and Sanders [16],

estimates of the hydrolysis rate have become available. These

estimates can be used to quantify the effect of ammonia on

hydrolysis via linear regression techniques. Typically, for this

last type of analyses, requiring several but preferably not too

many batch tests, the data set is rather small. Hence, for these

limited data sets, a so-called set-membership estimation

procedure is more appropriate than a statistical procedure.

During the last two decades a growing amount of literature on

set-membership estimation or bounded error approaches has

become available (see [17e21] for overviews). The key problem

in set-membership parameter estimation is not to find a single

vector with optimal parameter estimates, but a set of feasible

parameter vectors that are consistent with a given model

structure and bounded error data. This feasible parameter set

(FPS) directly reflects the uncertainty in the estimates. A

bounded error characterization, as opposed to a statistical

characterization in terms of mean, variances or probability

distributions, is favoured when the central limit theorem is

inapplicable, such as in situations with small data sets or with

heavily structured (modelling) errors.

Our present paper focuses on determining the possible

inhibiting effect of ammonia on the anaerobic hydrolysis of

carbohydrates and lipids, which are commonly found in life

stock wastes and agriculture co-substrates. To elucidate the

effect of ammonia, sludge adapted to high total ammonia

concentration of 4.9 g NH4þeN.l�1, and particulate substrates

with identical particle size were used in the inhibition batch

trials. Crystalline cellulose was the carbohydrate substrate

and tributyrin was the lipid substrate. Carbohydrates are the

main substrate during animal manure and agricultural resi-

dues digestion. Lipids are often used for co-digestion

substrates, due to their high methane yields [22]. Hydrolysis

efficiencies and rateswere used for assessment of the process.

2. Material and methods

2.1. Experimental set-up

Fourteen trials were performed in batch bottles at 35 � 2 �Cand anaerobic conditions. Seven batch trials had 1 g COD.l�1

tributyrin (Fluka, �98%; 1.96 g COD.g�1 tributyrin) as substrate

and the other seven batch trials had 1 g COD.l�1 cellulose

(Sigmacell� type 50; 1.19 g COD.g�1 cellulose), with an average

particle size of 50 mm, as substrate. Each experiment was

performed in duplicate including two controls with no

substrate addition, but having the corresponding ammonia

concentration in order to correct for the volatile fatty acids

and methane produced by the control. All test bottles were

inoculated with 100 ml of supernatant (creamy consistency

with no fibrous material) of digested pig slurry from a pig

manure co-digester operated at mesophilic conditions

(Sterksel, The Netherlands). The inoculum had 75.0 � 0.2 g TS

kg�1, 45.8 � 1.9 g VS kg�1 and 4.89 g NH4eN.l�1. In all test

bottles 100ml of 1.14M of ammoniumbicarbonate or 1.14M of

sodium bicarbonate solution was added according to the final

ammonia concentrations desired, thereby maintaining the

same ionic strength and pH. The tested ammonia concentra-

tionswere 2.4, 3.4, 4.1, 4.8, 5.8, 6.8 and 7.8 g NH4þeN.l�1. The pH

of all batch experiments was set to approximately 8 at the

b i om a s s an d b i o e n e r g y 4 7 ( 2 0 1 2 ) 3 1 6e3 2 3318

beginning of the test. Trace elements and macronutrients

were not added since they were already present in the pig

slurry. All bottles were flushed for 1 min with a gas mixture of

80% CH4 and 20% CO2 and placed in an incubator (Model 420,

Thermo Electron Corporation) and shaken at 150 rpm. All

trials were performed in 0.5 l Scott bottles (Germany) with two

sampling ports, closed with butyl rubber stoppers and

aluminium clamps, one for liquid sample to determine the

intermediate digestion products and another for gas samples

to determine the gas composition. All bottles had a total

working volume of 0.2 l and were fitted with an Oxitop� head

for gas pressure measurement, together with the gas

composition, to assess the cumulative methane production.

Gas pressure measurements with the Oxitop� system was set

for saving data every 50min for the cellulose experiments and

every 24 min for the tributyrin experiments. For the cellulose

experiments liquid and gas composition samples were taken

twice a day in the first 4 days and then at larger time intervals

according to the development of the hydrolysis step.While for

the tributyrin experiments, due to the faster hydrolysis, liquid

and gas composition samples were taken 5e3 times a day in

the first 2 days.

2.2. Chemical analysis

Total solids (TS) and volatile solids (VS) were determined

according to standard methods [23]. pH was measured with

a Multi 340i meter and SenTix P14 electrode (WTW, Germany).

NH4þeNwas determined at the beginning and end of the trials,

according to the ammonium coloration method [23]. Samples

for VFA were centrifuged at 13,000 rpm for 10 min in a Hermle

Z300 centrifuge. After centrifuging, the supernatant was

membrane filtered (0.45 mm, Schleicher & Schuell ME 25,

Germany) and used for the analysis. VFA was analysed in

a Hewlett Packard 5890A gas chromatogram equipped with

a 2m� 6 mm� 2mm glass column packed with Supelcoport,

100e120 mesh, coated with 10% Fluorad FC-431. The flow

rate of the carrier gas, i.e. nitrogen saturated with formic

acid, was 40 ml min�1, and the column pressure was 3 bar.

The temperatures of the column, the injector port, and

the flame ionization detector were 130, 200, and 280 �C,respectively. Biogas composition was analysed in a Fisions

Instrument GC 8340 gas chromatogram equipped with

a 30 m � 0.53 mm � 25 mm Molsieve column (Alltech 13,940),

and a 2� 25m� 0.53 mm� 10 mmPoraBondQ column (Varian

7354). The columns were connected in parallel. Helium was

the carrier gas and its flow rate was 42.5 ml min�1. The

temperatures of the oven, the injection port, the thermal

conductivity detector and the filament were 40, 110, 100 and

140 �C, respectively.

2.3. Calculations

Free ammonia was calculated according to Perry and Chilton

[24] and Poggi-Varaldo [25]:

NH3 �N ¼ NHþ4 �N$

266641þ 10�pH

10�

�0:1075þ

2725K

�37775

�1

: (1)

To determine the hydrolysis efficiency (%), as shown in

equation (2), the digestion intermediates were measured

during each trial:

Hydrolysisð%Þ ¼P

CODhydrolysis products; t¼x

CODtotal; t¼0� 100 (2)

where the CODhydrolysis products,t¼x is the concentration of VFA

plus CH4 produced during the cellulose and tributyrin hydro-

lysis, at the digestion time, t¼ x, expressed asmg COD.l�1. The

CODtotal t¼0 is the concentration of cellulose or tributyrin

added at the beginning of the trial, t ¼ 0, also expressed as mg

COD.l�1.

The first-order hydrolysis constant (kh in d�1) of the cellu-

lose and tributyrinwas calculated according to Angelidaki and

Sanders [16] and is here presented as equation (3).

CODtotal; t¼0$fh �PCODhydrolysis products; t¼x

CODtotal; t¼0$fh¼ e�kht (3)

where fh is the maximum biodegradable fraction of total

substrate added (cellulose or tributyrin), for completely

biodegradable substrates fh ¼ 1.

2.4. Data analysis

In a first step of the data analysis, the kh values were esti-

mated from batch data for different ammonia concentrations.

Next, the possible linear regression between the kh values and

the ammonia concentration, including the uncertainty in this

relationship, was investigated. Since the resulting small data

sets do not allow a proper statistical analysis, in what follows

a so-called set-membership approach to the second estima-

tion problem is applied.

In some experiments a lag phase in the cellulose hydrolysis

was observed. This behaviour was analysed with a time-

varying first-order hydrolysis model, where the constant khin equation (3) was replaced by kh(t).

2.4.1. Set-membership parameter estimationGiven the limited data set of kh values and corresponding

ammonia concentrations, the set-membership approach was

used (see for details [21])

Hence, instead of trying to find the optimal parameter

vector as in an ordinary least-squares approach, the goal of

set-membership parameter estimation is to find the set with

feasible parameter vectors that are consistent with the model

and the data with related error bounds. Therefore, we will not

consider the measurements as such, but define intervals for

each measurement. This approach avoids the distortion of

the original probability density function of the error after

some nonlinear transformation, because only bounds are

considered.

2.4.2. Mathematical modelIf we define the left hand-side of equation (4) as P(t) with

P(0) ¼ 1, which directly follows from the definition of

P(t), then PðtÞ ¼ e�kht. This expression is the solution of the

linear differential equation, with constant coefficient kh:

dP/dt ¼ �khP, P(0) ¼ 1, in what follows called a first-order

hydrolysis model. If, subsequently, we replace the hydrolysis

Table 1 e pH, NH3eN and kh for cellulose hydrolysis trialsat different NH4

DeN concentrations. Data expressed asmean (standard deviation). kh values include the last 5sampling points in Fig. 2A and exclude the samplingpoints in the lag phase.

NH4þeN

(g l�1)pH NH3eN

(mg l�1)kh (d�1) R2

2.4 8.16 (0.04) 333 (27) 0.095 0.98

3.4 8.17 (0.04) 471 (34) 0.144 0.99

4.1 8.16 (0.05) 546 (60) 0.103 0.99

4.8 8.16 (0.04) 621 (56) 0.104 0.91

5.8 8.16 (0.04) 729 (62) 0.104 0.90

6.8 8.16 (0.04) 842 (65) 0.061 0.80

7.8 8.16 (0.04) 957 (54) 0.088 0.94

b i om a s s a n d b i o e n e r g y 4 7 ( 2 0 1 2 ) 3 1 6e3 2 3 319

rate kh by a time-varying hydrolysis rate by k(t), which obeys

a sigmoid curve, we obtain the following time-varying first-

order hydrolysis model. This model was used for further

analysis of the trials showing a lag phase and is given by:

dPdt

¼ �kP; Pð0Þ ¼ 1

dkdt

¼ rkðK� kÞ; kð0Þ ¼ k0

(4)

where P is the dimensionless depletion of cellulose, k the time-

varying first-order hydrolysis rate in d�1, r is the rate param-

eter in d�1 related to the changes in k and which has its

maximumat k¼ 1/2 Kwith K in d�1 the upper limit on k, which

can be interpreted as kh,max. Typically, k(t) is an S-curve that

could, for instance, represent the generation and excretion of

extracellular enzymes and/or colonization of enzymes on

substrate’s surface, therefore, stimulating the hydrolysis

products formation. For the reversed curve, which becomes

relevant in case a possible inhibitor is present and which is

also time dependent in effect, we define ~k ¼ ðK� kÞ. This

transformation gives exactly the same response for P for given

values of r, K and k0. Consequently, d~k=dt ¼ �dk=dt and

~kð0Þ ¼ ðK� kð0ÞÞ, and thus equation (4) can also be written as

dPdt

¼ ðK� ~kÞP; Pð0Þ ¼ P0

�d~kdt

¼ rðK� ~kÞ~k; ~kð0Þ ¼ K� k0

(5)

Hence, instead of estimating only kh, two parameters are

now estimated, i.e. r and K. Given the time series of P, both

parameters were estimated using a nonlinear least-squares

estimation procedure (see e.g. [26]), as Matlab’s lsqnonlin.

Table 2 e pH, NH3eN and kh for tributyrin hydrolysistrials at different NH4

DeN concentrations. Data expressedas mean (standard deviation). kh values include the first 5sampling points in Fig. 4A.

NH4þeN

(g l�1)pH NH3eN

(mg l�1)kh (d�1) R2

2.4 8.06 (0.10) 299 (96) 4.86 0.96

3.4 8.10 (0.12) 444 (146) 4.73 0.96

4.1 8.06 (0.10) 506 (154) 4.64 0.96

4.8 8.05 (0.10) 577 (158) 4.53 0.93

5.8 8.06 (0.10) 713 (213) 4.54 0.85

6.8 8.05 (0.09) 809 (237) 3.58 0.90

7.8 8.04 (0.09) 876 (205) 4.70 0.98

3. Results and discussion

The ammonia and pH ranges applied in this study and pre-

sented in Tables 1 and 2 are commonly found in animal

manures such as cattle and pig [4], which are, together with

chicken manure, the main sources of manure for anaerobic

digestion in Europe [27].

3.1. Cellulose hydrolysis

Most of cellulose hydrolysis occurred in the first 11 days, as

shown by Fig. 2A. The incomplete hydrolysis is commonly

found for this type of crystalline cellulose [28,29].

What can also be seen in Fig. 2A is that for all trials the

hydrolysis of cellulose had a lag phase that, for most trials

took up to 4 days. The reason for such lag phase is not

completely clear. Song et al. [28] suggested this lag phase is the

time needed for the bacteria, that release the hydrolysing

enzymes, to occupy the crystalline surface of the cellulose.

Other authors point at the degree of accessibility and/or

crystallinity of the cellulose to the hydrolysing enzymes [30].

Yet, besides just observing that Sigmacell type 50 cellulose is

not completely hydrolysed [29,31], no relation is known

between the lag phase and the suggested factors.

Fernandes et al. (manuscript in preparation), showed no

lag phase when adding enzymes for hydrolysing crystalline

cellulose at mesophilic conditions and neutral pH. In their

experiments no leachate or animal manure was added, as it

was in the case of Song et al’s. [28] and O’Sullivan et al’s. [29]

studies. Indicating, as suggested by van Velsen [6] and Zee-

man [3], that there are compounds, other than ammonia,

present in animal manure that could cause hydrolysis inhi-

bition. This was indeed verified by Fernandes et al. (manu-

script in preparation) where crystalline cellulose hydrolysis

was inhibited by addition of humic and fulvic acids, extracted

from cow manure. The reason for the lag phase observed in

the present research could then be that produced hydrolytic

enzymes, secreted by the acidogenic bacteria, would first be

adsorbed to the inhibiting compound and only when all the

adsorbing sites were taken, the excess of enzymes, would

hydrolyse the cellulose. As the lag phases observed in this

study are identical for all trials, independently of the varying

ammonia concentrations, it can be concluded that this

inhibiting compound is not ammonia. The model to describe

this lag phase was developed in this study (equations (4) and

(5)) and its results are presented below.

The suggested mathematical model (equation (4)) shows

a good fit with the experimental data from all trials, as

depicted by Fig. 3 describing the example of cellulose hydro-

lysis trial at 4.8 g NH4þeN.l�1.

In this example, K ¼ kh,max ¼ 0.17 � 0.03 d�1 and

r ¼ 3.25 � 1.01. As mentioned before and described by the

mathematical model, the lag phase was similar for all cellu-

lose trials, independent of the ammonia concentrations,

0

20

40

60

80

100

0 5 10 15 20

Time (day)

Hy

dro

lys

is (

%)

2.4 g/l3.4 g/l4.1 g/l4.8 g/l5.8 g/l6.8 g/l7.8 g/l

y = -6E-05x + 0.1398R2 = 0.3018

00.020.040.060.08

0.10.120.140.16

250 400 550 700 850 1000NH3-N (mg.l

-1

)

kh (

d-1

)

A

B

Fig. 2 e Hydrolysis as a function of time at batch digestion

of cellulose at varying total ammonia concentrations, with

inoculum sludge adapted to 4.8 g NH4DeN.lL1 (A) and

hydrolysis rate constant (kh) of cellulose as a function of

free ammonia (B).

b i om a s s an d b i o e n e r g y 4 7 ( 2 0 1 2 ) 3 1 6e3 2 3320

indicating that ammonia does not influence the availability

and hydrolysis of cellulose to the hydrolytic enzymes.

Since during the lag phase there is negligible cellulose

hydrolysis, the data for estimating the hydrolysis rate

constant, kh, of cellulose should not include the lag phase. The

hydrolysis rate constants for cellulose hydrolysis, where the

lag phase is excluded, are presented in Table 1.

0 5 10 15 200

0.5

1

1.5

Time (d)

P

Fig. 3 e Depletion of cellulose as a function of time at batch

digestion of cellulose at total ammonia concentrations of

4.8 g NH4DeN.lL1. Symbol (D) refers to experimental data

and solid line to model predictions.

Hydrolysis of cellulose was well described by first-order

kinetics, as can be seen by the high correlation factors found

for all constants, presented in Table 1. First-order kinetics are

commonly used to describe the hydrolysis of degradable

particulate polymeric substrates [2,12].

According to the poor linear relation between the first-

order hydrolysis constants of the trials and their free

ammonia concentrations, shown in Fig. 2B, hydrolysis of

cellulose apparently is not inhibited by free ammonia. In order

to confirm such poor correlation, kh data analysis is performed

and results are presented in Section 3.3.1.

Hydrolysis rate constants of carbohydrates have been re-

ported to vary between 0.025 and 2.0 d�1 [2,32e34]. O’Sullivan

et al. [29] reported first-order hydrolysis rates between 0.015

and 0.5 d�1 by anaerobically hydrolysing a similar crystalline

cellulose, using leachate or rumen content as inoculum and

having free ammonia concentrations exceeding 300 mg l�1.

Most of their kh values were between 0.06 and 0.14 d�1,

therewith, similar to the ones found in this study.

The pH variation between samples and controls,

throughout the trials was negligible (Table 1), guaranteeing

that the pH did not impact the kh values of the different trials.

Moreover, considering the high pH sensitivity of the NH4þeN/

NH3eN speciation in the pH range 7.5e9.0, as illustrated in

Fig. 1, pH fixation was pursued during the test. Stable free

ammonia concentrations throughout the trial are crucial,

since most authors relate ammonia toxicity is to the non-

ionised free NH3 [4].

VFA in these trials remained very low, less than 5% from

the initial added cellulose, indicating that hydrolysis was

indeed the rate limiting step, as suggested by Angelidaki and

Sanders [16] discussing their batch test set-up. When hydro-

lysis rather than methanogenesis is rate limiting, hydrolysis

efficiency and hydrolysis rate can be determined by methane

formation [2]. The low amount of VFA in the batch bottles also

indicates that the methanogenic biomass was apparently

adapted to the high ammonia concentrations.

The maximum methane percentage present in the biogas

reached for these cellulose experiments was between 37 and

45%, and they were independent from the ammonia concen-

trations tested.

3.2. Tributyrin hydrolysis

In all experiments, almost all of the tributyrin was converted

within the first day of the run, as is shown in Fig. 4A. This fast

hydrolysis is in accordance with literature [2,35].

Unlike for the cellulose trials, no lag phase was observed

for the tributyrin trials.

Independently of the ammonia concentrations, tributyrin

digestion described similar curves, and in all trials, tributyrin

was fully hydrolysed within 2 days. The similar behaviour

resulted in similar hydrolysis rates, as shown in Table 2.

Tributyrin hydrolysis was well described by first-order

kinetics, as can be seen by the high correlation value (Table 2).

For determining the kh values, only the first five points were

used, where 88e100 % of the tributyrin was hydrolysed. The

remaining points were not taken into account because they

entail possible errors due to the low amount of substrate left.

According to Sanders [2], at very low substrate concentrations,

0

20

40

60

80

100

0 0.2 0.4 0.6 0.8 1 1.2

Time (day)

Hy

dro

lys

is (

%) 2.4 g/l

3.4 g/l4.1 g/l4.8 g/l5.8 g/l6.8 g/l7.8 g/l

y = -0.0011x + 5.1936

R2 = 0.301

0

1

2

3

4

5

6

250 400 550 700 850 1000NH

3-N (mg.l

-1)

kh

(d

-1)

A

B

Fig. 4 e Hydrolysis as a function of time at batch digestion

of tributyrin at varying total ammonia concentrations, with

inoculum sludge adapted to 4.8 g NH4DeN.lL1 (A) and

hydrolysis rate constant (kh) of tributyrin as a function of

free ammonia (B).

Table 3 e Error bound ( 3), potential outlier, intercept andslope of the linear trend for cellulose hydrolysis.

Iterationnr.

Nr of datapoints

Errorbound

Potentialoutlier

Intercept Slope

1 7 0.0291 471 0.1464 �0.0001

2 6 0.0177 842 0.1350 �0.0001

3 5 0.0067 333 0.1055 �0.0000

4 4 0.0038 e 0.1268 �0.0000

b i om a s s a n d b i o e n e r g y 4 7 ( 2 0 1 2 ) 3 1 6e3 2 3 321

a small error in hydrolysis products concentration has a big

impact on the kh.

The ammonia and pH ranges presented in Table 2 are

slightly lower than the ones from the cellulose hydrolysis

trials (Table 1).

Hydrolysis rate constants of lipids have been reported to

vary from 0.005 to 7.8 d�1 [2,32e34,36]. The hydrolysis rate

constants found in this study were within the above reported

range. However, this range includes different lipid sources,

which are, in a lot of cases, slower to hydrolyse than the

tributyrin here studied. Sanders [2] compared lipases activity

during digestion of palm oil and tributyrin and concluded that

the activity of the latter was much faster than the former due

to the steric hindrance of the LCFA in the triglycerides. Also

Wu and Tsai [35] reported that tributyrin is easily hydrolysed

into butyric acid.

According to the poor linear relation between the first-

order hydrolysis constants of the tributyrin trials and their

free ammonia concentrations, shown in Fig. 4B, the hydrolysis

of tributyrin was apparently not inhibited by free ammonia.

The pH chosen for these experimentswas approximately 8,

because the sludge was already adapted to this pH, which is

commonly found in manure anaerobic digesters. Moreover,

Wu and Tsai [35] studied the effect of pH on hydrolysis of

tributyrin with lipases produced by Pseudomonas fluo-

rescenes CCRC-17015, using pH ranges from 6 to 8 and

concluded that the optimum pH values for lipase-specific

activity were between 7.5 and 8.

The maximum methane percentages present in the biogas

reached for these tributyrin experiments were between 40 and

65%which is slightly higher than for the cellulose experiments.

Similar to the cellulose experiments CH4 concentrations were

independent from the ammonia concentrations tested.

3.3. Data analysis

3.3.1. Set-membership parameter estimationTo determine the accuracy of the relationship between

ammonia and kh, a set-membership estimation procedure

was implemented.

For the cellulose trials (see Table 1, Fig. 2B), in order to

obtain a non-empty FPS, a relatively large error bound of

0.0291 (d�1) had to be specified. Thisminimumerror bound for

the whole data set, together with the intercept and slope of

a linear relationship between ammonia and kh, was found

after solving a minemax optimization problem, as described

in Keesman and van Straten [37]. After considering trial 3.4 g/l

as unreliable, and thus removing the corresponding data point

from the data set, a smaller minimum error bound of 0.0177

(d�1) was needed. Removing the estimated kh value for trail

6.8 g/l even further decreased the minimum error bound to

0.0067 (d�1) (Table 3). Notice, furthermore, that even after

removing data points from the data set, the resulting slope in

the linear ammonia e kh relationship remains close to zero.

Only the intercept with the Y axes, as can be seen by the

values presented in Table 3, significantly changes.

Using an error bound of 0.06 (d�1), which is approximately

twice the minimum error bound for all trials, resulted in an

FPS as indicated by the shaded area in the top graph of Fig. 5.

Notice that the slope, including its uncertainty, is close to zero

and that the FPS contains zero slope values. This indicates that

the variation of each trial’s kh on the slope is small, therefore,

confirming that ammonia does not inhibit the hydrolysis of

cellulose. The bottom graph of Fig. 5 with the bounded linear

trend (bold lines), as a result of the presumed bounded error

properties of the data, confirms this finding. Notice that each

of the lines in the bottom graph of Fig. 5 is related to a vertex of

the FPS, as presented in the top graph of Fig. 5.

From the set-membership literature, it is well-known that

for the linear regression case, the FPS is a polytope found from

Fig. 5 e Feasible parameter set (FPS) for all cellulose

hydrolysis trials (top), as a result of the 7 (number of data

points) intersecting strips. Bounded linear trend (bold

lines) showing the uncertain and independent relationship

between ammonia and kh (bottom).Fig. 6 e Feasible parameter set (FPS) for all tributyrin

hydrolysis trials (top), as a result of the 6 (number of data

points) intersecting strips. Bounded linear trend (bold

lines) showing the uncertain and independent relationship

between ammonia and kh (bottom).

b i om a s s an d b i o e n e r g y 4 7 ( 2 0 1 2 ) 3 1 6e3 2 3322

the intersection of N (number of data points) strips in the

parameter space, as in the top graph of Fig. 5. It can even be

shown that the well-known weighted least-squares tech-

niques can be used to solve the bounded linear regression

problem [38,39]. But, in general, the FPS can be a complex, and

even unconnected, set (see [21], for details and possible

solutions).

For the tributyrin trials (see Table 2, Fig. 4B), the large error

bound of 0.6527 (d�1) was due to the large variation in the

estimated values of kh, that is from 3.58 to 4.86 d�1. After

considering trial 6.8 g/l as unreliable, and thus removing the

corresponding data point from the data set, a smaller

minimum error bound of 0.1265 (d�1) was needed. Removing

the estimated kh value for trail 7.8 g/l even further decreased

the minimum error bound to 0.0576 (d�1) (Table 4). From this

analysis, we concluded that the khe value related to the 6.8 g/l

trial is an outlier. Consequently, for further analysis we

removed this data point from the data set. Hence, after

Table 4 e Error bound ( 3), potential outlier, intercept andslope of the linear trend for tributyrin hydrolysis.

Iterationnr.

Nr of datapoints

Errorbound

Potentialoutlier

Intercept Slope

1 7 0.6527 809 4.9763 �0.0010

2 6 0.1265 876 4.8165 �0.0003

3 5 0.0576 e 5.0336 �0.0008

removing this outlier aminimumerror bound of 0.1265 (d�1) is

required (Table 4).

Using an error bound of 0.25 (d�1), which is approximately

twice the minimum error bound for the remaining six data

points, resulted in an FPS as indicated by the shaded area in

the top graph of Fig. 6. Notice that again the slope, including its

uncertainty, is close to zero and that the FPS encloses a zero

value of the slope. This indicates that the variation of each

trial’s kh on the slope is small, therefore, confirming that

ammonia does not inhibit the hydrolysis of tributyrin. The

bottom graph of Fig. 6 with the bounded linear trend (bold

lines), as a result of the presumed bounded error properties of

the data, again confirms this finding, as it contains negative

and positive slopes.

4. Conclusions

Ammonia nitrogen in the range of 2.4e7.8 g NH4þeN.l�1, cor-

responding to 283e908 mg NH3eN.l�1, does not affect hydro-

lysis of tributyrin. Also the hydrolysis of cellulose was not

affected by ammonia nitrogen concentrations in the range

of 2.4e7.8 g NH4þeN.l�1, corresponding to 333e957 mg

b i om a s s a n d b i o e n e r g y 4 7 ( 2 0 1 2 ) 3 1 6e3 2 3 323

NH3eN.l�1. Both conclusions were verified by mathematical

analysis. The mentioned ammonium nitrogen concentration

range has no impact on the lag phase observed for cellulose

hydrolysis.

Acknowledgements

The authors would like to acknowledge Katja Grolle for her

technical support and the Dutch Ministry of Economic Affairs

Agency Senter/Novem for the funding.

r e f e r e n c e s

[1] Angelidaki I, Ahring BK. Anaerobic thermophilic digestion ofmanure at different ammonia loads: effect of temperature.Water Res 1994;28(3):727.

[2] Sanders WTM. Anaerobic hydrolysis during digestion ofcomplex substrates 2001.

[3] Zeeman G. Mesophilic and psychrophilic digestion of liquidmanure. Ph.D. thesis. Wageningen: Wageningen University;1991.

[4] El-Mashad HE-MH. Solar thermophilic anaerobic reactor(STAR) for renewable energy production. Wageningen:Wageningen University; 2003.

[5] Kayhanian M. Ammonia inhibition in high-solidsbiogasification: an overview and practical solutions. EnvironTechnol 1999;20:355e65.

[6] van Velsen AFM. Anaerobic digestion of piggery waste.Wageningen: Wageningen University; 1981.

[7] Zeeman G, Wiegant WM, Koster-Treffers ME, Lettinga G. Theinfluence of the total-ammonia concentration on thethermophilic digestion of cow manure. Agric Wastes 1985;14(1):19e35.

[8] Koster IW, Lettinga G. The influence of ammonium-nitrogenon the specific activity of pelletized methanogenic sludge.Agric Wastes 1984;9(3):205.

[9] Braun R, Huber P, Meyrath J. Ammonia toxicity in liquidpiggery manure digestion. Biotechnol Lett 1981;3:159e64.

[10] Krylova NI, Khabiboulline RE, Naumova RP, Nagel MA. Theinfluence of ammonium andmethods for removal during theanaerobic treatment of poultry manure. J Chem TechnolBiotechnol 1997;70(1):99e105.

[11] van Velsen AFM. Adaptation of methanogenic sludge tohigh ammonia-nitrogen concentrations. Water Res 1979;13(10):995.

[12] Pavlostathis SG, Giraldo-Gomez E. Kinetics of anaerobictreatment. Water Sci Technol 1991;24(8):35e59.

[13] Lu F, Chen M, He PJ, Shao LM. Effects of ammonia onacidogenesis of protein-rich organic wastes. Environ Eng Sci2008 JaneFeb;25(1):114e22.

[14] Angelidaki I, Ellegaard L, Ahring BK. Applications of theanaerobic digestion process. In: Ahring BK, editor.Biomethanation II. Springer; 2003.

[15] Hansen KH, Angelidaki I, Ahring BK. Anaerobic digestion ofswine manure: inhibition by ammonia. Water Res 1998;32(1):5e12.

[16] Angelidaki I, Sanders W. Assessment of the anaerobicbiodegradability of macropollutants. Rev Environ SciBiotechnol 2004;3:117e29.

[17] Walter E. Parameter identifications with error bounds. MathComput Simul 1990;32(5e6).

[18] Norton JP. Bounded-error estimation: issue 1. Int J AdaptContr Sign Process 1994;8(1).

[19] Norton JP. Bounded-error estimation: issue 2. Int J AdaptContr Sign Process 1995;9(1).

[20] Milanese M, Norton JP, Piet-lahanier H, Walter E. Boundingapproaches to system identification; 1996. NY and London.

[21] Keesman KJ. Nonlinear-model case in bound-basedidentification, in control systems, robotics and automation.In: Unbehauen H, editor. Encyclopedia of life supportsystems (EOLSS). Oxford: UNESCO Publishing-EolssPublishers; 2003.

[22] Neves L, Oliveira R, Alves MM. Co-digestion of cow manure,food waste and intermittent input of fat. Bioresour Technol2009;100(6):1957e62.

[23] APHA. Standard methods for the examination of water andwastewater. 20th ed. Washington D.C.: American PublicHealth Association; 1998.

[24] Perry RH, Chilton CH. Chemical engineers handbook. 5th ed.New York: McGraw-Hill; 1973.

[25] Poggy-Varaldo HM, R.R.-V., Fernandez-Villagomez G,Esparza-Garcia F. Inhibition of mesophilic solid-substrateanaerobic digestion by ammonia nitrogen. Appl MicrobiolBiotechnol 1997;47:284e91.

[26] Keesman KJ. System identification: an introduction. London:Springer Verlag; 2011.

[27] Weiland P. Biomass digestion in agriculture: a successfulpathway for the energy production and waste treatment inGermany. Eng Life Sci 2006;6(3):302e9.

[28] Song H, Clarke WP, Blackall LL. Concurrent microscopicobservations and activity measurements of cellulosehydrolyzing and methanogenic populations during the batchanaerobic digestion of crystalline cellulose. BiotechnolBioeng 2005;91(3):369e78.

[29] O’Sullivan C, Burrell PC, Clarke WP, Blackall LL. The effect ofbiomass density on cellulose solubilisation rates. BioresourTechnol 2008;99(11):4723e31.

[30] Zhang Y-HP, Lynd LR. Toward an aggregated understandingof enzymatic hydrolysis of cellulose: noncomplexedcellulase systems. Biotechnol Bioeng 2004;88(7):797e824.

[31] Martins LF, Kolling D, Camassola M, Dillon AJP, Ramos LP.Comparison of Penicillium echinulatum and Trichodermareesei cellulases in relation to their activity against variouscellulosic substrates. Bioresour Technol 2008;99(5):1417e24.

[32] Veeken AAH, Hamelers BBV. Effect of substrate-seed mixingand leachate recirculation on solid state digestion ofbiowaste. Water Sci Technol 2000;41(3):255e62.

[33] Pavlostathis SG, Miller TL, Wolin MJ. Kinetics of insolublecellulose fermentation by continuous cultures ofRuminococcus albus. Appl Environ Microbiol November 1,1988;54(11):2660e3.

[34] Vavilin VA, Fernandez B, Palatsi J, Flotats X. Hydrolysiskinetics in anaerobic degradation of particulate organicmaterial: an overview. Waste Manag 2008;28(6):939e51.

[35] Wu H-S, Tsai M-J. Kinetics of tributyrin hydrolysis by lipase.Enzyme Microb Technol 2004;35(6e7):488.

[36] Batstone DJ, Keller J, Angelidaki I, Kalyuzhnyi SV,Pavlostathis SG, Rozzi A, et al. Anaerobic digestion model,vol. 1. London: IWA Publishing; 2002.

[37] Keesman KJ, Straten Gv. Identification and predictionpropagation of uncertainty in models with bounded noise.Int J Control 1989;49(1):2259e69.

[38] Keesman KJ. Weighted least squares set estimation from lN-norm bounded noise data. IEEE Trans Autom Control 1997;42(10):1456e9.

[39] Milanese M. Properties of least-squares estimates in setmembership identification. Automatica 1995;31:327e32.