Bioresource Technology 167 (2014) 495–502

Contents lists available at ScienceDirect

Bioresource Technology

journal homepage: www.elsevier .com/locate /bior tech

Application of Anaerobic Digestion Model No. 1 to describethe syntrophic acetate oxidation of poultry litter inthermophilic anaerobic digestion

http://dx.doi.org/10.1016/j.biortech.2014.06.0080960-8524/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author at: Departamento de Ingeniería Agroindustrial, Univers-idad Autónoma Chapingo, km 38.5 Carretera México-Texcoco, Chapingo, Estado deMéxico 56230, Mexico. Tel.: +52 (595) 952 1730.

E-mail address: t.espinosa.s@taurus.chapingo.mx (T. Espinosa-Solares).

Víctor Rivera-Salvador a, Irineo L. López-Cruz b, Teodoro Espinosa-Solares a,c,⇑, Juan S. Aranda-Barradas d,David H. Huber c,e, Deepak Sharma c,e, J. Ulises Toledo c

a Departamento de Ingeniería Agroindustrial, Universidad Autónoma Chapingo, Chapingo, Estado de México 56230, Mexicob Posgrado en Ingeniería Agrícola y Uso Integral del Agua, Universidad Autónoma Chapingo, Chapingo, Estado de México 56230, Mexicoc Gus R. Douglass Institute, West Virginia State University, Institute, WV 25112-1000, USAd Unidad Profesional Interdisciplinaria de Biotecnología, Instituto Politécnico Nacional, Mexico City, DF 07340, Mexicoe Department of Biology, West Virginia State University, Institute, WV 25112-1000, USA

h i g h l i g h t s

� Hydrogenotrophic methanogensdominate poultry litter anaerobicdigestion.� Syntrophic acetate oxidation was

modeled for poultry litter TAD.� Differential evolution algorithms

applied to evaluated ADM1parameters.� Evaluation of ADM1 syntrophic

acetate oxidation kinetic parameters.

g r a p h i c a l a b s t r a c t

Time [ d ]0 20 40 60 80

Vola

tile

fatty

aci

ds [m

g·L-1

]

0

200

400

600

800

1000

1200AcetatePropionate Butyrate Valerate DEA Simulated Acetate DEA Simulated Propionate DEA Simulated Butyrate DEA Simulated Valerate

a r t i c l e i n f o

Article history:Received 9 April 2014Received in revised form 2 June 2014Accepted 4 June 2014Available online 12 June 2014

Keywords:Differential evolution algorithmsNon-linear least squaresKinetic parametersParameter estimationAnaerobic digestion model

a b s t r a c t

A molecular analysis found that poultry litter anaerobic digestion was dominated by hydrogenotrophicmethanogens which suggests that bacterial acetate oxidation is the primary pathway in the thermophilicdigestion of poultry litter. IWA Anaerobic Digestion Model No. 1 (ADM1) was modified to include the bac-terial acetate oxidation process in the thermophilic anaerobic digestion (TAD). Two methods for ADM1parameter estimation were applied: manual calibration with non-linear least squares (MC-NLLS) andan automatic calibration using differential evolution algorithms (DEA). In terms of kinetic parametersfor acetate oxidizing bacteria, estimation by MC-NLLS and DEA were, respectively, km 1.12 and3.25 ± 0.56 kgCOD kgCOD

�1 d�1, KS 0.20 and 0.29 ± 0.018 kgCOD m�3 and Yac-st 0.14 and 0.10 ± 0.016 kgCOD kgCOD�1 .

Experimental and predicted volatile fatty acids and biogas composition were in good agreement.Values of BIAS, MSE or INDEX demonstrate that both methods (MC-NLLS and DEA) increased ADM1accuracy.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Anaerobic digestion has the capacity for treatment of organicwastes and energy recovery via biogas as well as reduced CO2

emissions and use of residues as fertilizer which make it a

496 V. Rivera-Salvador et al. / Bioresource Technology 167 (2014) 495–502

sustainable technology (Shen et al., 2013; Zamanzadeh et al.,2013). Methane can be produced either by aceticlastic methano-gens, which use acetate, or by hydrogenotrophic methanogenswhich utilize H2 and CO2. It has been reported that two thirds ofmethane produced by activated sludge derives from acetatewhereas just one third is obtained from hydrogenotrophic metha-nogenesis (Boone et al., 1989).

Methanogenic acetate degradation is carried out by an aceticlas-tic reaction or an anaerobic syntrophic acetate-oxidizing reaction(Hattori, 2008; Westerholm et al., 2011). In conventional anaerobicdigestion, acetate may be degraded by aceticlastic methanogenswhich limit the accumulation of acetate. In aceticlastic methano-genesis, acetate is cleaved to methyl and carboxyl groups; themethyl group is directly converted to methane by several biochem-ical reactions while the carboxyl group is oxidized to CO2 (Hattori,2008). However, an alternative pathway for acetate degradationinvolves a syntrophic relationship between acetate-oxidizingbacteria and hydrogenotrophic methanogens (Westerholm et al.,2011). This is known as syntrophic acetate oxidation.

Syntrophy is an essential intermediate step in the anaerobicconversion of organic matter to methane. In this mutualistic rela-tionship, metabolically distinct microorganisms are tightly linkedby the need to maintain the exchanged metabolites at very lowconcentrations (McInerney et al., 2009). Syntrophic metabolismrequires reverse electron transfer, close physical contact andmetabolic synchronization of the syntrophic partners (McInerneyet al., 2009). Anaerobic digestion is also an expression of thesyntrophic relationships among different microbes (Sasaki et al.,2011; Shen et al., 2013).

Syntrophic acetate oxidation is a two step methanogenesis fromacetate by a coculture where acetate is oxidized to CO2 and H2 byone organism and H2 is subsequently used by a second organism toreduce CO2 to CH4 (Westerholm et al., 2011). During the methano-genic mineralization process, syntrophic acetate oxidation isthermodynamically unfavorable (DG�0 = +104.6 kJ mol�1) (De Vriezeet al., 2012), and proceeds only if hydrogen partial pressures arekept low. By coupling hydrogen-consuming methanogens to theprocess, hydrogen partial pressures can be maintained between10 and 40 Pa (72 ± 5 Pa in some cases) (Hao et al., 2011; Schink,1997), and Gibbs free energies fluctuate near �20 kJ mol�1 (Haoet al., 2011). Thus, interspecies hydrogen transfer between second-ary fermenting bacteria and hydrogenotrophic methanogens isimportant for the oxidation of substrates such as fatty acids, etha-nol, propionate, butyrate and acetate (Hattori, 2008; Zamanzadehet al., 2013).

Therefore, hydrogenotrophic methanogens play a crucial role inconstantly removing H2 and producing methane which makes theoxidation of substrate by proton reduction energetically feasible.Hence, the syntrophic association between substrate oxidizersand hydrogenotrophic methanogens is necessary for sustainingthe overall process of anaerobic degradation (Luo et al., 2002).The acetate oxidation process is more common at thermophilictemperatures, although this process can also be developed atmesophilic conditions (Westerholm et al., 2010). However, if stressfactors arise, ammonia accumulation for instance, aceticlasticmethanogenesis can be inhibited (Schink, 1997; Wilson et al.,2012). There is a considerable difference in the acetate degradationrate for aceticlastic methanogenesis and syntrophic acetateoxidation which suggests that syntrophic acetate oxidation mightbe difficult to observe if both mechanisms are active at the sametime (Schink, 1997).

There is also evidence that syntrophic acetate-oxidizing bacte-ria are important at high ammonia levels in thermophilic anaerobicdigesters and high acid concentrations (Hao et al., 2011; Wilsonet al., 2012). Acetate-oxidizing bacteria have been found to beimportant for the start-up of methanogenesis from high organic

loadings in thermophilic anaerobic sequenced batch reactors(ASBR) at 55 �C (Hao et al., 2011). However, information about syn-trophic acetate oxidation, the organisms involved, and their role inthe methanogenic environment is currently limited, and moreresearch is required to elucidate the kinetic and ecological charac-teristics of these bacteria (Hao et al., 2011; Westerholm et al.,2011). Recently, the microbial community structure of a pilot-scalethermophilic CSTR digester stabilized on poultry litter wascharacterized (Smith et al., 2014). Based on the predominance ofhydrogenotrophic methanogens, as well as digester chemistry,bacterial acetate oxidation was proposed as the primary pathwayfor control of acetate levels in this digester.

Anaerobic Digestion Model No. 1 (ADM1) is a mechanisticmodel that explains complex substrates through their principalcomponents (Batstone et al., 2002). It includes several steps thatdescribe the biochemical and physicochemical processes involvedin the anaerobic biodegradation of organic compounds. However,syntrophic acetate oxidation is not currently a part of the modelbut some suggestions have been made by Batstone et al. (2002)to take into consideration. Mathematical modeling has become apopular support tool for design, operation and control of activatedsludge systems (Lübken et al., 2007). It can be used to predictprocess behavior in different situations and to assist operationalmanagement in order to develop strategies that will improvestability (Silva et al., 2009). These predictions can not only improveoperational decision making in agricultural biogas plants but alsoassist the planning of research experiments (Zhou et al., 2011).

Because little progress has been made in the mathematicalmodeling of bacterial acetate oxidation (Shimada et al., 2011),the objective of this work was to incorporate bacterial acetateoxidation into a well-known mathematical model. The presentpaper simulates the TAD of poultry litter by incorporating thebacterial acetate oxidation pathway in ADM1 as the main acetatedegradation process, rather than aceticlastic methanogenesis.Two methods for parameter estimation were used: differentialevolution algorithms (DEA) and a coupled method of manualcalibration and non-linear least squares (MC-NLLS).

2. Methods

2.1. Lab-scale digester

The digester that was used in this experiment was described asthe control reactor in Sharma et al. (2013). This reactor was a 10 Lglass vessel with a round-cylindrical bottom and separate ports forsampling, feeding and recirculation. The reactor was fed with a2.2% TS chicken litter manure suspension during 90 days. Thehydraulic retention time (HRT) was 15 days and the chemical oxy-gen demand (COD) was 24.77 ± 0.8 kgCOD m�3. The feedstock wassupplied at a semi-continuous rate of 0.66 L d�1. The poultry litter(manure, feathers, and wood chips) was collected from Moorefield,WV and New Market, VA.

The mixing consisted of a digestate recirculation systemwithout mechanical stirring. The digester was maintained atthermophilic conditions (56 �C) by running hot water through anexternal jacket. Temperature was monitored by a thermocouple.pH was automatically measured by a pH probe (Cole Parmer)connected in the re-circulation line. Biogas pressure wasmeasured with a pressure transducer (Omegadyne Inc., model no.PX209-015G5V) when excess biogas was released into a cylindercontaining water.

2.2. Analytical methods

For the substrate, total solids (TS) were determined using thestandard methods of APHA (1998) and chemical oxygen demand

V. Rivera-Salvador et al. / Bioresource Technology 167 (2014) 495–502 497

(COD) was measured according to the HACH water analysis hand-book (HACH, 2004). In case of the digestate, samples for volatilefatty acids (VFAs) profiles were collected twice per week. However,four consecutive days of sampling were done at the end of each30 days (two HRTs). VFA profiles were obtained for acetic, propi-onic, butyric, iso-butyric, valeric and iso-valeric acids using gaschromatography (Varian 3300 Gas Chromatograph, FID detector;glass column packed with 80/120 Carbopack B-DA/4% Carbowax20 M). Biogas composition was measured every day with a gaschromatograph (Agilent 7890 GC with HP-PLOT Q column).

2.3. Microbial diversity methods

The detection and analysis of archaeal diversity was done usingstandard methods. Archaeal diversity was sampled with PCR usingprimers AR109f and AR915r which target 16S rRNA genes (Raskinet al., 1994). To minimize bias caused by PCR artifacts, recondition-ing PCR was used (Thompson et al., 2002). PCR reactions were donein 25 lL reactions with Accuprime Taq polymerase (Invitrogen)and 1.25 ng of sample DNA. PCR was carried out with two sets ofcycles: 20 cycle reaction followed by a 5 cycle reaction. PCR condi-tions for the first round of cycles were: denaturation at 94 �C for2 min, then 20 cycles of 94 �C for 1 min, 52 �C for 1 min, and72 �C for 1.5 min, followed by a final extension at 72 �C for3 min. The second ‘‘reconditioning’’ PCR reaction was done withthe same reaction conditions except 5 cycles were used. Ampliconswere cloned using TA cloning (Life Technologies), and plasmidextractions were done with the Qiagen Miniprep kit. Clones weresequenced with an Applied Biosystems 3130xl Genetic Analyzer.Sequence analysis was done with the Ribosomal Database Project(RDP) Classifier tool.

3. Model development

3.1. Model description and implementation

ADM1 is a mechanistic model with 35 dynamic state variables,4 algebraic equations, 19 biochemical processes, 6 acid-base pro-cesses and 3 liquid-gas transfer processes (Rosen and Jeppsson,2004). With the addition of the bacterial acetate oxidation process,the model was changed to 36 dynamic state variables and 21biochemical processes. ADM1 assumes Monod-type kinetics.Therefore, acetate oxidation was also assumed to have Monod-typekinetics (Shimada et al., 2011) as shown in Eq. (1).

q11a ¼ km;ac�st �Sac

Ks;ac�st þ Sac� Xst � I13 ð1Þ

km,ac-st is the Monod maximum specific uptake rate (kgCODac

kgCODx�1 d�1) and KS,ac-st is the half saturation value for acetate-

oxidizing bacteria (Xst). The inhibition term (I13) considers low pHinhibition (IpH,aa), inhibition for limiting nitrogen, (IIN,lim) andhydrogen inhibition (Ih2,ac) (Eq. (2)).

I13 ¼ IpH;aa � IIN;lim � Ih2;ac ð2Þ

Table 1Biochemical rate coefficients and kinetic rate equations for the syntrophic acetate oxidati

Process: j Component: i

7 8 10Sac

(kgCOD m�3)Sh2

(kgCOD m�3)SIC

(kmol m�3)

20 Decay ofXac-st

�P

i¼1—9;11—25Cj � mi;20

11a Acetate oxidation �1 (1 � Yac-st) �P

i¼1—9;11—25Cj � mi;11a

Hydrogen inhibition was considered as a non-competitive inhibi-tion similar to propionate, butyrate and valerate as shown inEq. (3). According to previous work for hydrogen inhibition,hydrogen partial pressure above 40 Pa has a KI,h2,ac value of1 � 10�8 kgCOD m�3, which is the concentration of dissolvedhydrogen in liquid medium (Hattori, 2008; Schink, 1997).

Ih2;ac ¼1

1þ Sh2=KI;h2;acð3Þ

The decay of acetate-oxidizing bacteria (Eq. (4)) was implementedas a first order kinetic which was also used for the other microor-ganisms in the model (Shimada et al., 2011). kdec,Xst has the valueof 0.04 d�1 as in ADM1 for kdec of the model.

q20 ¼ kdec;Xst � Xst ð4Þ

The modification for ADM1 is summarized in Table 1 as a Petersenmatrix. A new differential equation was proposed for syntrophicacetate-oxidizing bacteria. Equations for complex matter (Xc), totalacetate (Sac) and dissolved hydrogen (Sh2) were modified due toaddition of acetate oxidation.

To avoid the modeling of aceticlastic methanogenesis, the vari-able Xac was fixed to zero (Xac = 0 kgCOD m�3) to represent theabsence of acetate-consuming methanogens in the system. Thisavoids the growth of Xac, but keeps the process of aceticlasticmethanogenesis in the model. In previous studies, the inhibitionof aceticlastic methanogens was necessary to promote acetate oxi-dation as a dominant methanogenic pathway (Wilson et al., 2012).The digester was considered to be a continuous system with aninput flow of 6.66 � 10�4 m3 d�1. The simulations were carriedout using Matlab 6.5.0 simulation platform.

3.2. Parameter estimation

Parameter values initially recommended by Batstone et al.(2002) and Rosen and Jeppsson (2004) were used. The parametersfor acetate-oxidizing bacteria (km,ac-st, KS,ac-st and Yac-st) wereadapted from Beaty and McInerney (1989) and Dwyer et al.(1988). The estimated kinetic parameters were as follows: Yfa,Yc4, Ypro, Yh2, kdis, khyd,ch, khyd,pr, khyd,li, km,su, KS,su, Ysu, km,aa, KS,aa, Yaa,km,fa, KS,fa, km,c4, KS,c4, km,pro, KS,pro, km,h2, KS,h2, km,ac-st, KS,ac-st andYac-st.

Two methods for the estimation of parameters were applied asfollows: manual calibration coupled to non-linear least squares(MC-NLLS) and differential evolution algorithms (DEA).

Manual calibration coupled to non-linear least squares: Initially, amanual calibration of the parameters was done to get a similartrend between experimental results and model outputs. After this,non-linear least squares were applied to refine the value of theparameters. The nonlinear least squares method with the algo-rithm of Levenberg–Marquardt was used. The ‘‘lsqnonlin’’ Matlabfunction was used to minimize the objective function (Eq. (5)).

f ðpÞ ¼XN

i¼1

ðyi � yiÞ2 ð5Þ

on process.

Process rate(kgCOD m�3 d�1)

11 13 25SIN

(kmol m�3)Xc

(kgCOD m�3)Xac-st

(kgCOD m�3)

(Nbac � Nxc) 1 �1 q20 ¼ kdec;Xst � Xst

�(Yac-st)�Nbac Yac-st q11a ¼ km;ac�st � SacKs;ac�stþSac

� Xst � I13

498 V. Rivera-Salvador et al. / Bioresource Technology 167 (2014) 495–502

where f (p) is a function that depends on the vector of parameters;yi is a vector of model output i as result of the vector of parametersp and yi is a vector of observed data i. The reason for a coupledmethod (MC-NLLS) is the problem of the over parametrization. Thismeans that the estimation of a large quantity of parameters (>10)by least squares leads to inaccurate estimated values and inaccuratepredictions (Makowski et al., 2006).

Differential evolution algorithms: DEA are evolutionary algo-rithms based on a population of potential solutions representedas floating-point vectors. Such algorithms are simple and efficientcompared with other global optimization methods. DEA reducesthe number of function evaluations required to converge to theglobal optimum (López-Cruz et al., 2008). The DEA parameterswere: 0.2 for crossing over probability, differential variation factorof 0.9, population size of 60, 1 � 10�6 for accuracy, 25 variables foroptimization, 1000 generations, and the use of the standard differ-ential evolution algorithm named DE/rand/1/bin.

4. Results and discussion

4.1. Substrate characterization

Poultry litter was characterized in terms of its principal compo-nents: carbohydrates, proteins, fats, volatile fatty acids and ammo-nia. The results of this analysis are shown in Table 2. It is importantto highlight the large concentration of hemicellulose (18.15%) andprotein (15.43%). Alkalinity was taken from previous results foranaerobic digestion of poultry litter in a pilot plant digester(Espinosa-Solares et al., 2006) as 6.21 kgCaCO3

m�3 in a 5.46% TSsuspension.

The substrate also contained a small concentration of lacticacid. The original ADM1 contains no modeling of the lactate degra-dation process, since concentrations as intermediates in mostdigesters are low. Batstone et al. (2002) has pointed out that lactateis degraded quickly, and it has the same stoichiometry as glucose.Additionally the biological reaction stoichiometry is not affected byits omission from ADM1.

4.2. Microbial diversity analysis

Acetate-oxidizing bacteria and hydrogenotrophic methanogenswere previously found to be present in the 40 m3 pilot plant diges-ter that was the source of inoculum used for this experiment(Smith et al., 2014). To confirm that the microbial communityhad retained these methanogen populations, archaeal diversitywas sampled in the lab-scale thermophilic digester.

Table 2Chemical composition of chicken litter manure.

Component % Dry basis

Hemicellulose 18.15Lignin 3.61Crude protein 15.43Crude fats 1.56Starch 1.51Water soluble carbohydrates 2.34Ash 7.95Ammonia-N 0.2386Neutral detergent fiber 41.99Acid detergent fiber 23.84Volatile fatty acids 1.34Acetic acid 0.6656Propionic acid 0.0315Iso-butyric acid 0.0237Butyric acid 0.1007

Archaeal community structure was analyzed with 16S rRNAclone libraries from a digester sample collected on day 30.Thirty-four 16S rRNA sequences were randomly sequenced. Phylo-genetic analysis showed that 84.8% of the sequences were highlysimilar (98% identity) to Methanothermobacter wolfeii (AB104858)(Watanabe et al., 2004), 6.1% of the sequences were related toMethanoculleus thermophilus (AB065297) (Rivard and Smith,1982) and 9% of the sequences were related to an unculturedarchaeon (FJ205778) (Kröber et al., 2009) (Table 3). Both of thesegenera are hydrogenotrophic methanogens. No known aceticlasticmethanogens, such as Methanosarcina, were detected. The absenceof aceticlastic genera in this sample indicates that bacterial acetateoxidation is probably the dominant pathway (Shimada et al., 2011;Westerholm et al., 2011) although some researchers have sug-gested that the coupled growth of syntrophic acetate-oxidizingbacteria and Methanosarcina sp. may enhance the stability of reac-tors at high ammonia concentrations (De Vrieze et al., 2012).Therefore, this analysis confirms that acetate oxidation in thisdigester occurs primarily through the hydrogenotrophic pathway.

4.3. Parameter estimation results

Parameters estimated by the two optimization methods areshown in Table 4. Most of the parameters showed similar valuesin both methods and model outputs showed similar behaviors.However, important differences were found for km,su, km,aa, km,pro

and km,h2. These differences may indicate that those parameterscould not affect model outputs but a sensitivity analysis is neces-sary to corroborate this hypothesis.

Carbohydrates (Xch) are hydrolyzed faster than proteins and lip-ids under these anaerobic conditions, which was associated to thekhyd values. Similar results have been previously considered in ananaerobic digester fed with cattle manure and renewable energycrops (Lübken et al., 2007).

When kinetic parameters for acetate-oxidizing bacteria areneeded, two main difficulties arise: isolation and the study ofkinetics of single acetate-oxidizing bacterial populations (Haoet al., 2011; Hattori, 2008). The application of a parameterestimation method, such as DEA or MC-NLLS, can be an adequatealternative procedure to calculate those kinetic parameters.However, DEA allows the estimation of more parameters than con-ventional procedures, such as least squares (Shimada et al., 2011),and reduces the problem of overparametrization. This method mayincrease the accuracy of the parameters values and the quality ofthe model prediction. Nevertheless, kinetic parameters for ace-tate-oxidizing bacteria (Xst) were calculated by DEA and coupledMC-NLLS as estimation procedures to compare both methods.

Using the DEA method, the value for km,ac-st was3.25 ± 0.57 kgCOD kgCOD

�1 d�1 and KS,ac-st was 0.29 ± 0.02 kgCOD m�3,which are similar to values derived by Dwyer et al. (1988) fromsyntrophic anaerobic bacteria in coculture with hydrogen-oxidizing methanogens. On the other hand, the values obtainedhere were similar to the ones reported by Shimada et al. (2011)for mesophilic conditions in two-phase anaerobic digestion ofwaste activated sludge. For MC-NLLS; the value of km,ac-st

(2.8 kgCOD kgCOD�1 d�1 in this paper) was slightly less than the one

(3.9 kgCOD kgCOD�1 d�1) reported by Shimada et al. (2011).

The yield parameter for syntrophic bacteria (Yac-st) wasestimated as 0.104 ± 0.016 kgCOD kgCOD

�1 by DEA and 0.14 kgCOD

kgCOD�1 by MC-NLLS, which are similar to the value reported by

Shimada et al. (2011).For some researchers, the transition of the dominant pathway

from syntrophic acetate oxidation to aceticlastic methanogenesiswas considered a minor difference. This may explain why syn-trophic oxidation has frequently been neglected or seldom evalu-ated (Hao et al., 2011). Another research group (Sasaki et al.,

Table 3Phylogenetic affiliations and relative abundance of archaeal populations in the anaerobic digester on day 30 based on 16S rRNA gene clone libraries.

Phylogenetic affiliation 16S rRNA% Similarity Methanogenesis Relative abundance at day 30

Methanothermobacter wolfeii 99.3% Hydrogenotrophic 84.8%Methanoculleus thermophilus 99.8% Hydrogenotrophic 6%Uncultured archaeon (FJ205778) 99.1% Unknown 9%

Table 4Estimated parameters of the modified ADM1 models.

Parameter Uncertainty intervala MC-NLLS estimated value DEA estimated value (meand ± SD) Units

kdis 0.24–1.0 0.6032 0.55 ± 0.37 d�1

khyd,ch 0.19–1.94 1.0675 1.05 ± 0.56 d�1

khyd,pr 0.0096–1.0 0.5051 0.88 ± 0.23 d�1

khyd,li 0.0096–0.4 0.2136 0.17 ± 0.17 d�1

km,su 27–107 37.943 85.9 ± 27.0 kgCOD kgCOD�1 d�1

KS,su 0.05–0.2 0.1228 0.15 ± 0.04 kgCOD m�3

Ysu 0.01–0.15 0.1066 0.13 ± 0.02 KgCOD kgCOD�1

km,aa 27–107 27.465 70.4 ± 29.9 kgCOD kgCOD�1 d�1

KS,aa 0.05–0.2 0.1262 0.1 ± 0.05 kgCOD m�3

Yaa 0.086–0.15 0.0917 0.098 ± 0.02 kgCOD kgCOD�1

km,fa 11–37 24.057 26.99 ± 9.7 kgCOD kgCOD�1 d�1

KS,fa 0.058–2.0 0.4022 0.076 ± 0.04 kgCOD m�3

Yfa 0.004–0.05 0.0262 0.023 ± 0.01 kgCOD kgCOD�1

km,c4 5.3–13.7 6.3884 5.39 ± 0.17 kgCOD kgCOD�1 d�1

KS,c4 0.012–0.36 0.3230 0.36 ± 0.0001 kgCOD m�3

Yc4 0.05–0.079 0.0789 0.06 ± 0.005 kgCOD kgCOD�1

km,pro 5.5–33.0 10.012 23.4 ± 6.7 kgCOD kgCOD�1 d�1

KS,pro 0.02–0.392 0.3687 0.29 ± 0.12 kgCOD m�3

Ypro 0.05–0.089 0.0592 0.058 ± 0.009 kgCOD kgCOD�1

km,h2 1.68–178.0 12.974 76.36 ± 44.34 kgCOD kgCOD�1 d�1

KS,h2 1 � 10�6–6 � 10�4 5.0 � 10�4 4 � 10�4 ± 1.95 � 10�4 kgCOD m�3

Yh2 0.014–0.183 0.1715 0.095 ± 0.05 kgCOD kgCOD�1

km,ac-stb 0.037–25.0 1.1204 3.25 ± 0.57 kgCOD kgCOD

�1 d�1

KS,ac-stb 0.0005–0.3 0.2035 0.29 ± 0.02 kgCOD m�3

Yac-stc 0.0001–1.0 0.1350 0.104 ± 0.016 kgCOD kgCOD

�1

a From Batstone et al. (2002).b From Dwyer et al. (1988).c From Beaty and McInerney (1989).d Mean of ten replicates.

V. Rivera-Salvador et al. / Bioresource Technology 167 (2014) 495–502 499

2011), working with TAD and using labeled carbon, reported thatnon-aceticlastic oxidation accounted for 74–88% of the total degra-dation of acetate; consequently the aceticlastic cleavage of acetateaccounted for 12–26%.

In the present study, the kinetic values for syntrophic bacteria(mainly km,ac-st and Yac-st) are different from those used in ADM1for aceticlastic methanogens (km,ac and Yac). km,ac is muchlarger (16 kgCOD kgCOD

�1 d�1) than km,ac-st (1.12 kgCOD kgCOD�1 d�1 and

3.25 ± 0.57 kgCOD kgCOD�1 d�1), and Yac-st (0.14 kgCOD kgCOD

�1 and0.1 kgCOD kgCOD

�1 ) is more than double Yac (0.05 kgCOD kgCOD�1 ) which

was reported in ADM1. Therefore, the omission of a syntrophicacetate oxidation pathway in this model may have different resultscompared to an aceticlastic methanogenesis pathway.

4.4. Model outputs

Similar trends were found in the model in both cases, using DEAand MC-NLLS (Fig. 1). However, manual calibration (MC), calledtrial and error, has a subjective element which means that MCdepends on the judgment of the modeler, and replication of theresults may be difficult.

For the two methods, several measures of agreement betweencalculated and measured data were determined for the variables‘‘volatile fatty acids’’, ‘‘methane partial pressure’’ and ‘‘pH’’. Foreach one, the next measurements were performed (Makowskiet al., 2006): bias, mean square error (MSE), relative root mean

square error (RRMSE), model efficiency (EF), correlation coefficient(r) and the agreement index (Index). The results of these values arein Table 5.

Both methods of parameter estimation showed a betteradjustment than the original ADM1 parameters, mainly in theVFA outputs, as can be appreciated by an index close to 1.00(0.95 for MC-NLLS and 0.85 for DEA). The values of bias indicatethat in all cases, the model under-predicts (Bias > 0) for VFA andpH, but over-predicts (Bias > 0) for methane percentage.

Actually, it is accepted that a good model would have small biasand a correlation coefficient (r) near to 1.00 (Makowski et al.,2006). Both adjustment by DEA and MC-NLLS showed a small biasand r near to one to describe VFA behavior. These values are betterin comparison with the ones obtained using the originalparameters of the ADM1. Therefore, the application of a parameterestimation method (DEA or MC.NLLS) increased the quality of theADM1.

An agreement index of 1.00 indicates a perfect model, other-wise if the model predictions are identical in all cases and equalto the average of the observed values then the index is zero(Makowski et al., 2006). In this case, both MC-NLLS and DEAincreased the quality of the original ADM1 for the VFA measureddata. Interestingly, the calculated index indicates good modelestimation for pH and methane partial pressure. However, thesevalues must be taken carefully, because the average of theobserved data and the model predictions had similar values that

Time [ d ]0 20 40 60 80

Vola

tile

fatty

aci

ds [m

g·L-1

]

0

200

400

600

800

1000

1200AcetatePropionate Butyrate Valerate DEA Simulated Acetate DEA Simulated Propionate DEA Simulated Butyrate DEA Simulated Valerate

Time [ d ]0 20 40 60 80

Vola

tile

fatty

aci

ds [m

g·L-1

]

0

200

400

600

800

1000

1200AcetatePropionate Butyrate Valerate MC-NLLS Simulated Acetate MC-NLLS Simulated Propionate MC-NLLS Simulated Butyrate MC-NLLS Simulated Valerate

Time [d]0 20 40 60 80

Bio

gas

com

posi

tion

[%]

0

20

40

60

80

100

Experimental Methane DataMC-NLLS-Simulated Methane

Time [ d ]0 20 40 60 80

Bio

gas

com

posi

tion

[%]

0

20

40

60

80

100

Experimental Methane DataDEA-Simulated Methane

(a) (b)

(d)(c)

Fig. 1. Comparison of modeled and experimental results for (a) VFA using MC-NLLS, (b) VFA using DEA, (c) biogas composition using MC-NLLS, (d) biogas compositionusing DEA.

Table 5Measures of agreement between measured and calculated values.

Variable of analysis Measure of agreement ADM1 (parameters from Batstone et al.(2002) and Rosen and Jeppsson (2004))

ADM1 (estimation by MC-NLLS) ADM1 (estimation by DEA)

Volatile fatty acids Bias (kgCOD m�3) 0.4623 0.0366 0.1368MSE ((kgCOD m�3)2) 0.3049 0.0216 0.0471RRMSE 1.0837 0.2883 0.4260EF �2.0592 0.7834 0.5272r 0.3262 0.9136 0.8551Index 0.4797 0.9494 0.8502

Methane partial pressure Bias (bar) �0.0394 �0.0085 �0.0111MSE (bar2) 0.0019 0.0082 0.0009RRMSE 0.0861 0.1769 0.0588EF �1.4699 �9.4280 �0.1526r 0.5366 �0.0328 0.0925Index 1.0000 1.0000 1.0000

pH Bias 0.1449 0.4168 0.2266MSE 0.0247 0.2054 0.0535RRMSE 0.0202 0.0582 0.0297EF �15.7067 �138.1973 -35.2749r 0.3841 0.0194 0.0097Index 0.9999 0.9995 0.9999

MSE: mean square error; RRMSE: relative root mean square error, EF: model efficiency; r: correlation coefficient; Index: index agreement.

500 V. Rivera-Salvador et al. / Bioresource Technology 167 (2014) 495–502

caused an index near to 1.00, which could affirm that the model isquasi-perfect. Nevertheless, the values of efficiency (EF) indicatedthat the model could be a worse predictor than the average ofthe observed values (EF < 0) for pH and methane partial pressureoutputs (Table 5).

In the case of methane partial pressure and pH, the model wasnot as good as for VFA. Although biogas composition was well

represented in the model, a peak-like perturbation occurred duringthe first days (Fig. 1c and d). The reasons for this behavior areunknown but are probably due to ‘‘auto-fitting’’ of the model.According to Girault et al. (2012), during a non rate-limiting acido-genic stage, the peak at the beginning of an anaerobic digestionprocess was the result of a rapid conversion into methane ofhydrogen (Girault et al., 2012).

Time [d]0 20 40 60 80

Met

hano

geni

c or

ace

tate

oxi

dizi

ng

mic

roor

gani

ms

[kg C

OD

·m-3

]

0.0

0.2

0.4

0.6

0.8

Aceticlastic methanogens (Xac)Syntrophic acetate oxidizers (Xst)Hydrogenotrophic methanogens (Xh2)

Time [d]0 20 40 60 80

Deg

radi

ng m

icro

orga

nism

s [k

g CO

D·m

-3]

0.0

0.1

0.2

0.3

0.4

0.5

0.6

Sugar degraders (Xsu)Aminoacid degraders (Xaa)Fatty acid degraders (Xfa)Butyrate and valerate degraders (Xc4)Propionate degraders (Xpro)

Fig. 2. Microbial biomass simulation during the chicken litter manure treatment.

V. Rivera-Salvador et al. / Bioresource Technology 167 (2014) 495–502 501

In terms of the outputs, simulation of volatile fatty acids(propionate, butyrate, valerate and acetate) showed a closer repre-sentation to experimental data, mainly in the first 4 HRTs (60 days)for propionate, butyrate and valerate. On the other hand, propio-nate was overestimated at the end of the process, mainly afterday 60 (Fig. 1a and b). In addition, acetate simulation did not rep-resent the dynamic behavior of experimental data at the beginningof the process.

Pseudo-stationary states were reached for all simulated volatilefatty acids after 2.66 HRT (40 days). Furthermore, for simulatedmicrobial biomass, all species reached similar results. Sugardegraders (Xsu) were the last group to reach a pseudo-stationarystate, following a significant biomass decrease (Fig. 2). The popula-tions of microorganisms that were used in the model for methaneproduction were the hydrogenotrophic methanogens (Xh2) and theacetate-oxidizing bacteria (Xst). This assumption was taken basedon the experimental results of this work which are in agreementwith the TAD reported by Sasaki et al. (2011). Aceticlastic metha-nogens (Xac) remained zero during the entire simulation time, somethane production from aceticlastic methanogenesis was notcarried out in the simulation. However, the accumulation of otherVFAs (propionate, butyrate) may have been due to the lack ofsufficient syntrophic bacteria, mainly at the beginning of thesimulation (Shi et al., 2013).

With respect to acidogenic and acetogenic bacteria, the aminoacid degraders (Xaa) and valerate and butyrate degraders (Xc4)had the highest abundances. On the other hand, long chain fattyacid degraders (Xfa) obtained the lowest abundance compared toall the active microbial biomass. This was apparently due to thelow concentration of lipids in the substrate. The greater impor-tance of the acetogenic step compared to the acidogenic step ishighlighted by the high abundance of acetogens. Lastly, the accu-mulation of organic acids demonstrated their inhibitory effect onmethane production which is undesirable in anaerobic digestion.

5. Conclusion

TAD of poultry litter was dominated by hydrogenotrophic meth-anogens, suggesting bacterial acetate oxidation as the primarypathway. For syntrophic bacteria, the values estimated by MC-NLLSwere km,ac-st 1.12 kgCOD kgCOD

�1 d�1, KS 0.20 kgCOD m�3 and Yac-st

0.14 kgDQO kgDQO�1 ; while using DEA they were km 3.25 ± 0.57 kgCOD

kgCOD�1 d�1, KS 0.29 ± 0.02 kgCOD m�3 and Yac-st 0.104 ± 0.016 kgDQO

kgDQO�1 . Values of BIAS, MSE or INDEX demonstrate that both

methods (MC-NLLS and DEA) increased ADM1 accuracy. Thus, asuccessful model was developed incorporating acetate-oxidizing

syntrophic bacteria linked to hydrogenotrophic methanogens forTAD.

Acknowledgements

The authors wish to acknowledge the financial support fromWest Virginia State University (USA), USDA Grant 2010-02276(USA), Gus R. Douglass Institute’s Agricultural and EnvironmentalResearch Station (USA) and Consejo Nacional de Ciencia yTecnología (México). Furthermore express their gratitude to Dr.Christian Rosen and to Dr. Ulf Jeppsson for providing the initialMatlab implementation of ADM1.

References

APHA, AWWA, WPCF, 1998. Standard Methods for the Examination of Water andWastewater, 20th ed. American Public Health Association, Washington, DC,USA.

Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V., Pavlostathis, S.G., Rozzi, A.,Sanders, W.T., Siegrist, H., Vavilin, V.A., 2002. In: Anaerobic Digestion Model No.1. IWA Publishing.

Beaty, P.S., McInerney, M.J., 1989. Effects of organic acid anions on the growth andmetabolism of Syntrophomonas wolfei in pure culture and in defined consortia.Appl. Environ. Microbiol. 55, 977–983.

Boone, D.R., Johnson, R.L., Liu, Y., 1989. Diffusion of the interspecies electron carriersH2 and formate in methanogenic ecosystems and its implications in themeasurement of Km for H2 or formate uptake. Appl. Environ. Microbiol. 55,1735–1741.

De Vrieze, J., Hennebel, T., Boon, N., Verstraete, W., 2012. Methanosarcina: therediscovered methanogen for heavy duty biomethanation. Bioresour. Technol.112, 1–9.

Dwyer, D.F., Weeg-Aerssens, E., Shelton, D.R., Tiedje, J.M., 1988. Bioenergeticconditions of butyrate metabolism by a syntrophic, anaerobic bacterium incoculture with hydrogen-oxidizing methanogenic and sulfidogenic bacteria.Appl. Environ. Microbiol. 54, 1354–1359.

Espinosa-Solares, T., Bombardiere, J., Chatfield, M., Domaschko, M., Easter, M.,Stafford, D.A., Castillo-Angeles, S., Castellanos-Hernandez, N., 2006.Macroscopic mass and energy balance of a pilot plant anaerobic bioreactoroperated under thermophilic conditions. Appl. Biochem. Biotechnol. 129–132,959–968.

Girault, R., Bridoux, G., Nauleau, F., Poullain, C., Buffet, J., Steyer, J.P., Sadowski, A.G.,Béline, F., 2012. A waste characterisation procedure for ADM1 implementationbased on degradation kinetics. Water Res. 46, 4099–4110.

HACH, 2004. Hach Water Analysis Handbook, 4th ed. HACH Company, Colorado,USA.

Hao, L.-P., Lü, F., He, P.-J., Li, L., Shao, L.-M., 2011. Predominant contribution ofsyntrophic acetate oxidation to thermophilic methane formation at high acetateconcentrations. Environ. Sci. Technol. 45, 508–513.

Hattori, S., 2008. Syntrophic acetate-oxidizing microbes in methanogenicenvironments. Microbes Environ. 23, 118–127.

Kröber, M., Bekel, T., Diaz, N.N., Goesmann, A., Jaenicke, S., Krause, L., Miller, D.,Runte, K.J., Viehöver, P., Pühler, A., Schlüter, A., 2009. Phylogeneticcharacterization of a biogas plant microbial community integrating clonelibrary 16S-rDNA sequences and metagenome sequence data obtained by 454-pyrosequencing. J. Biotechnol. 142, 38–49.

502 V. Rivera-Salvador et al. / Bioresource Technology 167 (2014) 495–502

López-Cruz, I.L., Rojano-Aguilar, A., Ramírez-Arias, A., 2008. Differential EvolutionAlgorithms: A Matlab Implementation. Universidad Autónoma Chapingo,Departamento de Irrigación.

Lübken, M., Wichern, M., Schlattmann, M., Gronauer, A., Horn, H., 2007. Modellingthe energy balance of an anaerobic digester fed with cattle manure andrenewable energy crops. Water Res. 41, 4085–4096.

Luo, H.-W., Zhang, H., Suzuki, T., Hattori, S., Kamagata, Y., 2002. Differentialexpression of methanogenesis genes of Methanothermobacterthermoautotrophicus (formerly Methanobacterium thermoautotrophicum) inpure culture and in cocultures with fatty acid-oxidizing syntrophs. Appl.Environ. Microbiol. 68, 1173–1179.

Makowski, D., Hillier, J., Wallach, D., Andrieu, B., Jeuffroy, M.H., 2006. Parameterestimation for crop models. In: Makowski, D., Wallach, D., Jones, J.W. (Eds.),Working with Dynamic Crop Models: Evaluation, Analysis, Parameterization,and Applications. Elsevier, The Netherlands.

McInerney, M.J., Sieber, J.R., Gunsalus, R.P., 2009. Syntrophy in anaerobic globalcarbon cycles. Curr. Opin. Biotechnol. 20, 623–632.

Raskin, L., Stromley, J.M., Rittmann, B.E., Stahl, D.A., 1994. Group-specific 16S rRNAhybridization probes to describe natural communities of methanogens. Appl.Environ. Microbiol. 60, 1232–1240.

Rivard, C.J., Smith, P.H., 1982. Isolation and characterization of a thermophilicmarine methanogenic bacterium, Methanogenium thermophilicum sp. nov. Int. J.Syst. Bacteriol. 32, 430–436.

Rosen, C., Jeppsson, U., 2004. Anaerobic Cost Benchmark Model Description.Department of Industrial Electrical Engineering and Automation, LundUniversity.

Sasaki, D., Hori, T., Haruta, S., Ueno, Y., Ishii, M., Igarashi, Y., 2011. Methanogenicpathway and community structure in a thermophilic anaerobic digestionprocess of organic solid waste. J. Biosci. Bioeng. 111, 41–46.

Schink, B., 1997. Energetics of syntrophic cooperation in methanogenic degradation.Microbiol. Mol. Biol. Rev. 61, 262–280.

Sharma, D., Espinosa-Solares, T., Huber, D.H., 2013. Thermophilic anaerobicco-digestion of poultry litter and thin stillage. Bioresour. Technol. 136,251–256.

Shen, P., Zhang, J., Zhang, J., Jiang, C., Tang, X., Li, J., Zhang, M., Wu, B., 2013. Changesin microbial community structure in two anaerobic systems to treat bagasse

spraying wastewater with and without addition of molasses alcoholwastewater. Bioresour. Technol. 131, 333–340.

Shi, J., Wang, Z., Stiverson, J.A., Yu, Z., Li, Y., 2013. Reactor performance andmicrobial community dynamics during solid-state anaerobic digestion of cornstover at mesophilic and thermophilic conditions. Bioresour. Technol. 136, 574–581.

Shimada, T., Morgenroth, E., Tandukar, M., Pavlostathis, S.G., Smith, A., Raskin, L.,Kilian, R.E., 2011. Syntrophic acetate oxidation in two-phase (acid-methane)anaerobic digesters. Water Sci. Technol. 64, 1812–1820.

Silva, F., Nadais, H., Prates, A., Arroja, L., Capela, I., 2009. Modelling of anaerobictreatment of evaporator condensate (EC) from a sulphite pulp mill using theIWA anaerobic digestion model no. 1 (ADM1). Chem. Eng. J. 148, 319–326.

Smith, A., Sharma, D., Lappin-Scott, H., Burton, S., Huber, D.H., 2014. Microbialcommunity structure of a pilot-scale thermophilic anaerobic digester treatingpoultry litter. Appl. Microbiol. Biotechnol. 98, 2321–2334.

Thompson, J.R., Marcelino, L.A., Polz, M.F., 2002. Heteroduplexes in mixed-templateamplifications: formation, consequence and elimination by ‘reconditioningPCR’. Nucleic Acids Res. 30, 2083–2088.

Watanabe, T., Asakawa, S., Nakamura, A., Nagaoka, K., Kimura, M., 2004. DGGEmethod for analyzing 16S rDNA of methanogenic archaeal community in paddyfield soil. FEMS Microbiol. Lett. 232, 153–163.

Westerholm, M., Dolfing, J., Sherry, A., Gray, N.D., Head, I.M., Schnürer, A., 2011.Quantification of syntrophic acetate-oxidizing microbial communities in biogasprocesses. Environ. Microbiol. Rep. 3, 500–505.

Westerholm, M., Roos, S., Schnürer, A., 2010. Syntrophaceticus schinkii gen. nov., sp.nov., an anaerobic, syntrophic acetate-oxidizing bacterium isolated from amesophilic anaerobic filter. FEMS Microbiol. Lett. 309, 100–104.

Wilson, C.A., Novak, J., Takacs, I., Wett, B., Murthy, S., 2012. The kinetics of processdependent ammonia inhibition of methanogenesis from acetic acid. Water Res.46, 6247–6256.

Zamanzadeh, M., Parker, W.J., Verastegui, Y., Neufeld, J.D., 2013. Biokinetics andbacterial communities of propionate oxidizing bacteria in phased anaerobicsludge digestion systems. Water Res. 47, 1558–1569.

Zhou, H., Löffler, D., Kranert, M., 2011. Model-based predictions of anaerobicdigestion of agricultural substrates for biogas production. Bioresour. Technol.102, 10819–10828.