Modelling the population dynamics and metabolic diversity of organisms relevant in anaerobic/anoxic/aerobic enhanced biological phosphorus removal processes

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 6

Avai lab le a t www.sc iencedi rec t .com

journa l homepage : www.e lsev ie r . com/ loca te /wat res

Modelling the population dynamics and metabolic diversity oforganisms relevant in anaerobic/anoxic/aerobic enhancedbiological phosphorus removal processes

A. Oehmen a,*, C.M. Lopez-Vazquez b, G. Carvalho a,c, M.A.M. Reis a,M.C.M. van Loosdrecht d

aREQUIMTE/CQFB, Chemistry Department, FCT, Universidade Nova de Lisboa, 2829-516 Caparica, PortugalbDepartment of Urban Water and Sanitation, UNESCO-IHE Institute for Water Education, Wesvest 7, 2611 AX Delft, The Netherlandsc Instituto de Biologia Experimental e Tecnologica (IBET), Av. da Republica EAN, 2780-157 Oeiras, PortugaldDepartment of Biotechnology, Delft University of Technology, Julianalaan 67, 2628 BC Delft, The Netherlands

a r t i c l e i n f o

Article history:

Received 9 March 2010

Received in revised form

31 May 2010

Accepted 7 June 2010

Available online 12 June 2010

Keywords:

Polyphosphate accumulating

organisms (PAO)

Glycogen accumulating organisms

(GAO)

Kinetics

Model calibration

Candidatus Accumulibacter

Phosphatis clades

Fluorescence in situ hybridisation

(FISH)

* Corresponding author. Tel.: þ351 212 948 3E-mail address: [email protected] (A.

0043-1354/$ e see front matter ª 2010 Elsevdoi:10.1016/j.watres.2010.06.017

a b s t r a c t

In this study, enhancedbiological phosphorus removal (EBPR)metabolicmodels are expanded

in order to incorporate the competition between polyphosphate accumulating organisms

(PAOs) and glycogen accumulating organisms (GAOs) under sequential anaerobic/anoxic/

aerobic conditions, which are representative of most full-scale EBPR plants. Since PAOs and

GAOs display different denitrification tendencies, which is dependent on the phylogenetic

identity of the organism, the model was separated into six distinct biomass groups, consti-

tuting Accumulibacter Types I and II, as well as denitrifying and non-denitrifying Competibacter

and Defluviicoccus GAOs. Denitrification was modelled as a multi-step process, with nitrate

(NO3),nitrite (NO2),nitrousoxide (N2O)anddi-nitrogengas (N2) being theprimarycomponents.

The model was calibrated and validated using literature data from enriched cultures of PAOs

andGAOs, obtainingagooddescriptionof theobservedbiochemical transformations.A strong

correlationwasobservedbetweenAccumulibacterTypes I and II, andnitrate-reducingandnon-

nitrate-reducing PAOs, respectively, where the abundance of each PAO subgroup was well

predicted by the model during an acclimatisation period from anaerobiceaerobic to anaero-

biceanoxic conditions. Interestingly, a strong interdependency was observed between the

anaerobic, anoxic and aerobic kinetic parameters of PAOs and GAOs. This could be exploited

whenmetabolicmodels are calibrated, since all of these parameters should be changed by an

identical factor from their default value. Factors that influence these kinetic parameters

include the fraction of active biomass, relative aerobic/anoxic fraction and the ratio of acetyl-

CoA to propionyl-CoA. Employing a metabolic approach was found to be advantageous in

describing the performance and population dynamics in such complexmicrobial ecosystems.

ª 2010 Elsevier Ltd. All rights reserved.

1. Introduction phosphorus (P) removal are known as the polyphosphate

In the enhanced biological phosphorus removal (EBPR)

process, the group of organisms primarily responsible for

85; fax: þ351 212 948 550.Oehmen).ier Ltd. All rights reserved

accumulating organisms (PAOs). In order to promote the

development of PAO and, consequently, P removal, anaerobic

followed by anoxic and/or aerobic conditions are generally

.

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 64474

employed. PAOs are able to take up carbon sources such as

volatile fatty acids (VFAs) anaerobically and store them as

polyhydroxyalkanoates (PHAs), providing them a selective

advantage over most ordinary heterotrophs. However,

glycogen accumulating organisms (GAOs) are also capable of

anaerobic VFA uptake and therefore can also be enriched

under similar conditions as PAOs, consuming the generally

limited VFA supply, without contributing to P removal. In

recent years, the competition between PAO and GAO has been

studied intensively due to (a) its impact on phosphorus

removal performance and efficiency, and (b) PAO-dominated

systems have the potential to decrease operational costs

throughminimising the addition of supplemental additives (e.g.

chemical precipitants, organic carbon sources) necessary

to achieve sufficient P removal (Oehmen et al., 2007a).

The addition of these chemicals is also undesirable since

they result in higher sludge generation, increasing sludge

disposal costs.

Among numerous operational and environmental factors,

the carbon source (Pijuan et al., 2004; Oehmen et al., 2005b; Lu

et al., 2006), pH (Filipe et al., 2001b; Schuler and Jenkins, 2002;

Oehmen et al., 2005a) and temperature (Panswad et al., 2003;

Lopez-Vazquez et al., 2007b, 2009a) have been observed to

have a profound impact on the PAOeGAO competition.

Recently, Lopez-Vazquez et al. (2009b) formulated a metabolic

model that incorporates the combined effects of carbon

source, pH and temperature on the metabolism of key EBPR

microorganisms under anaerobiceaerobic conditions: specif-

ically, Accumulibacter (PAO), Competibacter (GAO) and Defluvii-

coccus (GAO). In full-scale plants the EBPR process is invariably

combined with nitrogen (N) removal. Different groups of PAOs

and GAOs have shown varying denitrification capacities (Zeng

et al., 2003c; Carvalho et al., 2007; Wang et al., 2008) that may

have an important impact on their competition. This served as

the motivation for the present study.

It has been postulated that denitrifying PAOs (or DPAOs)

able to reduce nitrate correlatewell with Type IAccumulibacter,

while non-DPAOs (or simply, PAOs) that are unable to reduce

nitrate but able to reduce nitrite have been correlated with

Type II Accumulibacter (Carvalho et al., 2007; Flowers et al.,

2009; Oehmen et al., 2010). Due to the fact that Type I and

Type II Accumulibacter correlate strongly with the so-called

DPAO and PAO, respectively, we have adopted the terms PAOI

and PAOII to differentiate between their different denitrifica-

tion tendencies in this manuscript. This was done to avoid

confusion, since both organisms appear capable of nitrite

reduction, while the only difference is that PAOI are capable of

nitrate reduction as well. The term “PAO” seems better suited

as a more general term to describe all organisms that

contribute to enhanced biological phosphorus removal in

activated sludge systems.

Kong et al. (2006) hypothesised that the different

subgroups of Competibacter also display varying denitrifying

capacities: (i) capable of nitrate and nitrite reduction

(subgroup 6), (ii) able to reduce nitrate only (subgroups 1, 4 and

5) and (iii) unable to denitrify (subgroups 3 and 7). Wang et al.

(2008) showed that an enrichment of Defluviicoccus Cluster I

was able to reduce nitrate, but not nitrite, while Burow et al.

(2007) suggested that Defluviicoccus Cluster II was unable to

denitrify. It is clear from these studies that the denitrification

activity of PAOs and GAOs depends on the abundance of the

different subgroups enriched.

By expanding the metabolic model developed by Lopez-

Vazquez et al. (2009b), the present study focuses on the calibra-

tion and validation of a metabolic model developed to describe

thebiochemical activityof 6microbial groupsof PAOsandGAOs,

namely the nitrate-reducing and non-nitrate-reducing Accumu-

libacter (i.e. PAOI and PAOII, respectively), denitrifying and non-

denitrifying Competibacter (DGB and GB, respectively) and deni-

trifying and non-denitrifying Defluviicoccus (DDEF and DEF,

respectively). Strategies aimed at facilitating the calibration of

metabolic models based on a small number of parameters are

also addressed in this study. Since model calibration is a very

important and challenging issue in activated sludge modelling,

ensuring the ease of metabolic model calibration is crucial in

order to increase its potential utility in practice. Further, the

ability of thismodel to assess the population dynamics of DPAO

and PAO inmicrobial enrichments will be illustrated.

2. Materials and methods

2.1. Model development

From a physiological perspective, the metabolic model

developed in this study incorporates the different capabilities

of PAOs and GAOs to denitrify by separating them into

multiple distinct groups. For this purpose, denitrification by

PAOs and GAOs was modelled as a multi-step process from

nitrate to nitrite, followed by nitrite to N2O, and finally N2O to

N2. The stoichiometric matrix for PAOI and PAOII is shown in

Appendix A, and that for GAOs andDGAOs (including GB, DGB,

DEF and DDEF) is shown in Appendix B. In summary (see

Fig. 1), PAOI are assumed to be capable of NO3, NO2 and N2O

reduction, while PAOII are capable of NO2 and N2O reduction

only, as suggested from their metagenome (Garcia Martin

et al., 2006). DGB is also considered to be capable of NO3,

NO2 and N2O reduction, while DDEF is capable of NO3 reduc-

tion only and GB and DEF are not capable of denitrifying.

These properties are consistentwith the results obtained from

the aforementioned literature studies.

It should be noted that PAOs and GAOs have been found to

require a brief acclimation period (e4e5 h) to induce denitrifi-

cation enzymes after being exclusively exposed to anaerobic/

aerobic conditions (Kuba et al., 1996b; Zeng et al., 2003a;Wang

et al., 2008). In most systems, the organisms are exposed to

anaerobic/anoxic/aerobic conditions, continuously exposing

the bacteria to denitrifying conditions. Since the goal of this

study was to model PAO and GAO steady-state metabolism,

enzyme inductionwasnot incorporated into themodel.Nitrite

accumulation (in the formof free nitrous acid: SHNO2) is known

to inhibit P uptake by PAOs (Saito et al., 2004; Zhou et al., 2007),

and can lead to the undesirable production of N2O (a powerful

greenhouse gas) (Zhou et al., 2008). These aspects have been

incorporated into the kinetic equations of the model.

2.2. Anaerobic stoichiometry

The anaerobic stoichiometry of PAOI and PAOII is modelled

identically. The anaerobic processes consist of VFA uptake (as

Fig. 1 e Extent of denitrification for each organism modelled in this study.

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 6 4475

well asmaintenance), where both acetate (SAc) and propionate

(SProp) uptake are considered, since these form the major

fraction of VFA in wastewater (Oehmen et al., 2007a). VFA

uptake by PAOI and PAOII is coupled with polyphosphate

hydrolysis (XPP) and phosphate release (SPO4), glycogen

degradation (XGly) and PHA production (XPHA). The anaerobic

processes of GAOs and DGAOs are similar to PAOs, except that

polyphosphate (poly-P) hydrolysis and release as phosphate

(P) does not occur. The anaerobic stoichiometric parameters

are shown in Appendix C and are identical to the model of

Lopez-Vazquez et al. (2009b).

2.3. Aerobic and anoxic stoichiometry

Aerobically and anoxically, PAOs degrade PHA for P uptake

and poly-P production, glycogen production and biomass

growth (modelled here as the resultant of the total PHA

degraded minus the PHA utilised for poly-P and glycogen

production, see Murnleitner et al., 1997 and Lopez-Vazquez

et al., 2009b), with the sole difference being the electron

acceptor utilised for ATP production. Themetabolism of GAOs

is similar, without the reaction describing poly-P storage. The

aerobic and anoxic yield coefficients are a function of the P/O

ratio (d and dN, also known as YNADH_ATP), which represents

the ATP produced per NADH oxidised during oxidative phos-

phorylation. The ATP production is dependent on the electron

acceptor, and is lower under anoxic conditions as compared to

aerobic conditions (Smolders et al., 1994; Kuba et al., 1996a).

The aerobic and anoxic parameters used in this study are

detailed in Appendix D. The yield coefficients are also a func-

tion of the phosphate transport energy (e and eN), the ATP

necessary for biomass synthesis (K1 and K2) and the

percentage of acetyl-CoA* and propionyl-CoA* (l and b) in the

PHA polymer (Smolders et al., 1994; Kuba et al., 1996a; Zeng

et al., 2003b). The l and b coefficients are based on the

relative composition of the PHA polymer produced under

anaerobic conditions, including polyhydroxybutyrate (PHB),

polyhydroxyvalerate (PHV) and polyhydroxy-2-methylvalerate

(PH2MV). Since these PHA fractions are determined by the

fraction of acetate and propionate in the influent wastewater,

the carbon source affects the anaerobic, anoxic and aerobic

stoichiometric yields. Similarly to Zeng et al. (2003b), it was

assumed that there was no preferential utilisation of

a particular PHA fraction for any particular aerobic or anoxic

metabolic reaction.

The NOx-based yield coefficients were calculated from the

PHA-based yield coefficients through redox balancing, as per-

formed in Lopez-Vazquez et al. (2009b) for the oxygen-based

yields (see Appendices A and B). This calculation incorporates

the degree of reduction of the PHA polymer and biomass

formula, where the biomass composition found by Zeng et al.

(2003b) was assumed for both PAOs and GAOs. In analogy to

theapproach takenbyKubaet al. (1996a), theNOx-basedanoxic

yieldsaremultipliedbya factorcorresponding to thedifference

in the number of available electrons per mole of electron

acceptor (see Appendices A and B): 4 e�/mol of O2 reduced,

2 e�/mol of N reduced from NO3 to NO2, 2 e�/mol of N reduced

from NO2 to N2O, and 1 e�/mol of N reduced from N2O to N2.

2.4. Kinetic model

The kinetic expressions for the anaerobic, anoxic and aerobic

processes of each organism are detailed in Appendix E. The

structure of these expressions is similar to previous studies

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 64476

(Murnleitner et al., 1997; Meijer et al., 2002; Lopez-Vazquez

et al., 2009b), except for the inclusion of:

(1) A switching function in the acetate and propionate uptake

processes (below) to limit the specific fraction of PHA

storage by PAOs and GAOs to a pre-specified value, fPHA,

max, in order to avoid situations where PHA may accumu-

late beyond realistic levels. This is similar to the approach

of de Kreuk et al. (2007).

fPHA;max � fX;PHA

fPHA;max � fX;PHA þ Ks;fPHA

where:

fX,PHA¼ the specific fraction of PHA stored by each organism

(PAO or GAO)

fPHA,max¼ the maximum specific fraction of PHA able to be

stored by each organism

Ks,fPHA¼ the half-saturation coefficient for the specific PHA

fraction

(2) An exponential function, e�170$SHNO2 , was added to the

anoxic poly-P formation expressions, in order to account

for inhibition by free nitrous acid (Zhou et al., 2007). It

should be noted that while nitrite/SHNO2 has also been

observed to inhibit aerobic P uptake (Yoshida et al., 2009),

these expressions were not amended since this was

beyond the scope of the present study.

(3) An exponential function, e�760$SHNO2 , was added to the N2O

reduction process of PHA degradation by PAOI and PAOII,

in order to account for the inhibitory effect of SHNO2 on N2O

reduction (Zhou et al., 2008). It should be pointed out that

the mechanisms affecting N2O reduction in PAOs and

GAOs require further investigation.

(4) Switching functions, SAc=SAc þ SProp and SProp=SAc þ SPropwere added to the anaerobic acetate and propionate

uptake expressions, respectively, in order to account for

the competition by PAOs and GAOs for carbon source

when both are present simultaneously, and preventing the

VFA uptake rate from doubling when both substrates are

present concurrently. The validity of this approach was

considered beyond the scope of this study, since the

Table 1 e Parameter estimates for DPAOs and DGAOs obtainedvalidation.

Carbon source Organism Cycle VFA uptake Glycogen

qVFA_PHA

C-mol/C-mol hqGLY

C-mol/C-

Model calibration

Propionate PAOI AN/AX 0.100 0.0025

Acetate DGBa AN/AX 0.050 0.0190

Acetate DDEF AN/AX 0.028 0.0025

Model validation

Acetate PAOI & DGBa AN/AX 0.050 & 0.050 0.0100 & 0

Acetate DGBa AN/AX 0.050 0.0190

a Denotes granular sludge bioreactor.

bioreactors that were described by the model were fed

with single substrates only.

(5) A switching function, SNO2=SNO2 þ SNO3, was added to the

nitrite reduction kinetics for PHA degradation, glycogen

production and maintenance of DGBs. The purpose of this

switching function is to account for the observation that

DGBs have been observed to display a preference towards

nitrate reduction vs. nitrite reduction (Zeng et al., 2003c).

This may be due to the fact that more Competibacter

subgroups have been found to be capable of nitrate

reduction as compared to nitrite reduction (Kong et al.,

2006). This type of switching function is a common

feature of other modelling strategies incorporating multi-

step denitrification (Sin et al., 2008).

The kinetic parameters for the expressions of Appendix E

are detailed in Table 1 and Appendix F.

2.5. Model calibration/validation

This integratedmetabolicmodelwas implemented inAquasim

(Reichert, 1994), where it was defined as a sequencing batch

reactor (SBR) with alternating anaerobiceanoxiceaerobic-

settling-idle stages, similar to the approach of Lopez-Vazquez

et al. (2009b).

The model calibration was performed by adjusting only 4

kinetic parameters: the VFA uptake rate (qVFA_PHA), glycogen

production rate (qGLY), PHA degradation rate (qPHA) and poly-P

formation rate (qPO4_PP). Experimental data (i.e. acetate or

propionate, glycogen, PHA and phosphate) from PAO or GAO

systems operated under anaerobic/anoxic conditions (Zeng

et al., 2003c; Carvalho et al., 2007; Wang et al., 2008) were

used for fitting these parameters. The simulation length was

set to a period of 3 sludge retention times (SRT) in order to

describe the steady-state behaviour of PAO and DGB (Zeng

et al., 2003c; Carvalho et al., 2007). The initial poly-P fraction

was set to half themaximum fraction in order to ensure that it

was neither limiting, nor approaching the saturation level,

while the initial glycogen and PHA levels were defined

according to their initial measured values. Due to the very low

anoxic activity of the DDEF biomass group (Wang et al., 2008),

the operation of this bioreactor necessitated the washout of

during model calibration and those applied during model

prod. PHA deg. Poly-P Form. Experimental Data

mol hqPHA

C-mol/C-mol hqPO4_PP

P-mol/C-mol h

0.050 0.0008 Carvalho et al. (2007)

0.210 Zeng et al. (2003c)

0.100 Wang et al. (2008)

.0190 0.150 & 0.210 0.0025 & 0 Zeng et al. (2003a)

0.210 Zeng et al. (2003c)

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 6 4477

the VFA and NOx at the end of the anaerobic and anoxic

phases, respectively, while no sludge was wasted for this

system. This suggested that a steady-state under anaerobic/

anoxic conditions was not achieved for this system, which is

consistent with the model simulations (see Section 3.1). Thus,

the parameter estimates correspond to the first cycle. In this

cycle, the total amount of acetate taken up by the culture of

Wang et al. (2008) was defined as the initial acetate concen-

tration. Similarly to the FISH quantification results from this

study, the sludge was simulated as 85% DDEF and the

remainder as DGB.

The parameter estimates obtained from model calibration

were validated through simulating the results of amixed PAO/

DGB enrichment under anaerobic/anoxic conditions with

nitrate feeding (Zeng et al., 2003a) and a DGB enrichment

under anaerobic/anoxic conditions with nitrite feeding (Zeng

et al., 2003c). Since Zeng et al. (2003c) showed that N2O was

the main product of denitrification in their system, the N2O

reduction processes were not included during these model

calibration or validation studies. The N2O reduction processes

were included in the other simulation studies despite the fact

that N2O was not measured (Zeng et al., 2003a; Carvalho et al.,

2007; Wang et al., 2008). There is a clear need to better eluci-

date the activation/inactivation of N2O reductase.

2.6. Simulation studies

The estimation of the kinetic parameters qVFA_PHA, qGLY, qPHA

and qPO4_PP from denitrifying PAO/GAO studies was compared

with estimates from other PAO/GAO studies (Filipe et al.,

2001a; Zeng et al., 2003b; Oehmen et al., 2005a, 2006; Dai

et al., 2007; Lopez-Vazquez et al., 2008; Bengtsson, 2009), in

order to assess any inter-relationships between these

parameters. In this set of simulations, the goal was to assess

the maximum specific initial rate of PHA degradation,

glycogen production and poly-P formation, rather than assess

steady-state performance, thus the simulations were run over

one cycle only. ThemaximumVFA uptake rate was calculated

through linear regression directly from the experimental data

of each study. Further, a sensitivity analysiswas performed on

each of these calibrated parameters in order to assess the

effect of varying each parameter by �10% and �50% (see

Appendix G). This was coupled with an error analysis in order

to quantify the agreement between themodel predictions and

the experimental data (see Appendix G). Percent error was

assessed through the normalised root mean squared devia-

tion (NRMSD), as shown in the following equation:

NRMSD ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPn

i¼1ðxmeas;i�xpred;iÞ2n

rxmeas;max � xmeas;min

(6)

where xmeas and xpred represent the measured and predicted

concentrations of a given variable (e.g. VFA, PHA, P, glycogen)

with n data points and xmeas,min and xmeas,max represent the

minimum and maximum measured concentrations in that

dataset. In the sensitivity analysis presented in Appendix G,

the effect of a �10% change in parameter on the propagation

of error over an extended simulation period (3 SRT) is also

presented and discussed.

Additionally, the acclimatisation of the propionate-fed

EBPR system of Carvalho et al. (2007) from anaerobic/aerobic

conditions to anaerobic/anoxic conditions was simulated in

accordance with the operational conditions of this study. The

initial PAOI and PAOII concentrations were estimated through

fluorescence in situ hybridisation (FISH) analysis using probes

Acc-I-444 and Acc-II-444 (Flowers et al., 2009), which target

Accumulibacter clade IA and others, and clades IIA, IIC and IID,

respectively. Details of the FISH quantification procedure with

these probes are specified in Oehmen et al. (2010). The corre-

lation between the measured Type I Accumulibacter by FISH

and the model-simulated PAOI, as well as Type II Accumu-

libacter with PAOII, was explored.

The minimum anoxic and aerobic SRT were calculated

according to the following equation (Brdjanovic et al., 1998) for

both acetate-fed (Murnleitner et al., 1997) and propionate-fed

(Carvalho et al., 2007) anaerobic/anoxic systems, as well as

acetate-fed (Brdjanovic et al., 1998) and propionate-fed

(Oehmen et al., 2005a) anaerobic/aerobic systems.

SRTmin ¼YVFA PHA

YVFA X$t24

fPHA;max ��fPHA;max

13� qPHA$t

3

�3

where:

YVFA_PHA¼ the anaerobic PHA yield from VFA uptake.

YVFA_X¼ the anoxic (or aerobic) biomass growth from

anaerobic VFA uptake.

qPHA¼ the maximum specific PHA degradation rate.

t¼ the total aerobic or anoxic phase time (in h) per day.

3. Results

3.1. Model calibration

The results of the model calibration for enriched cultures of

PAOs and GAOs operated under anaerobic/anoxic conditions

are displayed in Fig. 2. There was a good description of the

conversion processes in each reactor. The propionate-fed PAO

reactor of Carvalho et al. (2007) and acetate-fed DGB reactor of

Zenget al. (2003c)were simulated for a periodof 3 SRTs inorder

to describe the steady-state behaviour of these systems (Fig. 2a

and b, respectively). In addition to the good description of the

VFA, PHA, glycogen, phosphate and nitrate/nitrite data, it was

found that the final biomass concentration (56.9 and 89.8 C-

mmol/L, for PAO and DGB, respectively) agreed within 10% of

the measured value for each system (50.9 and 89.9 C-mmol/L,

for PAO and DGB, respectively). This is particularly significant

considering that the biomass growth rate is not explicitly

defined in the model, but is calculated from the remainder of

the PHA not used for glycogen or polyphosphate production.

This supports the validity of themodel structure. Furthermore,

it was noteworthy that the competition between nitrate and

nitrite reduction observed for DGBs was well described after

including the switching function SNO2=SNO2 þ SNO3, with no

other changes necessary in the model structure or anoxic

kinetic parameters. The multiple expressions for each anoxic

metabolic process that was necessitated by the inclusion of

multi-step denitrification in the model of PAOs and DGBs

0

4

8

12

16

20

72 0 7 22 72 4 7 26 72 8

Ti me ( hr )

C, P

, N

m

m o

l /L

X_ PH A X_ PH A_me as X_ Gl y X_ Gl y_ me as S_ HP r S_ HP r_ me as S_ NO 3 S_ NO 3_ me as S_ PO 4 S_ PO 4_ me as

0

5

10

15

20

25

30

143 4 1 435 14 36 14 37 14 38 14 39 14 40

Ti me ( hr )

C, N

m

mo

l/L

X_ PH A X_ PH A_ me as X_ Gl y X_ Gl y_ me as S_ HA c S_ HA c_ me as S_ NO 3 S_ NO 3_ me as S_ NO 2 S_ NO 2_ me as

0

2

4

6

8

10

0 1 2 3 4 5 6

Ti me ( hr )

C, N

m

mo

l/ L

X_PH A X_PH A_ me as X_Gl y X_Gl y_ me as S_ HA c S_ HA c_ me as S_ NO 3 S_ NO 3_ me as S_ NO 2 S_ NO 2_ me as

a

b

c

Fig. 2 e Model calibration description of the biomass behaviour in a) an enriched PAO (Accumulibacter) culture fed with

propionate b) an enriched DGB (Competibacter) culture fed with acetate c) an enriched DDEF (Defluviicoccus) culture fed with

acetate. Lines indicate model description and symbols indicate experimental measurements.

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 64478

contained identical anoxic parameters for PHA degradation

(qPHA), glycogen production (qGLY) and P uptake (qPO4_PP). Each

of these parameters can thus be viewed as the rate that

describes the overall anoxic metabolism of each organism,

independentof theactualdenitrificationelectronacceptor (e.g.

NO3� or NO2

�).Fig. 2c shows the calibrated model description of the

system ofWang et al. (2008), consistingmainly of DDEF.While

a good description of the data was achieved, the parameter

estimates associated with the activity of DDEF were assessed

from a single cycle, and the simulation was not able to be

sustained over an extended period. This is consistent with the

operational conditions employed to sustain the culture of

Wang et al. (2008), and likely due to the very low glycogen

concentration contained in the biomass. Notably, the initial

glycogen concentration of this culture was approximately one

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 6 4479

order of magnitude smaller than a similarly enriched Deflu-

viicoccus culture operated under anaerobiceaerobic conditions

(Dai et al., 2007). It appears that the low anoxic glycogen

production is a limiting factor towards sustaining a DDEF

culture, which is consistent with previous findings for Type II

Accumulibacter enrichments (Carvalho et al., 2007). The

commonality between DDEF and Type II Accumulibacter is an

incomplete denitrification pathway; however, it is unclear

how this impacts their regulation of glycogen production. In

simulations with DDEF containing a higher initial glycogen

concentration, their glycogen content was observed to

decrease over time (data not shown), which appears to agree

with the experimental results.

The slower kinetic activity of DDEF as compared to both PAO

and DGB indicates that these organisms are unlikely to be

competitive under anaerobic/anoxic conditions as compared to

PAOsandDGBs (seeTable 1). Thus, theerror associatedwith the

kineticparameterestimationsofDDEFduringanoxicconditions

(see Appendix G) is likely to be of minor practical relevance.

3.2. Model validation

Formodel validation purposes, two additional reactor systems

were simulated using the calibrated parameters. Firstly,

the data from the anaerobic/anoxic acetate-fed reactor of

0

4

8

12

16

20

0 1 2 3

Ti me ( hr )

C, P

, N

m

m o

l /L

0

5

10

15

20

25

30

0 1 2 3

Ti me ( hr )

C, N

m

mo

l/L

a

b

Fig. 3 e Model validation description of the biomass behaviour i

enriched DGB culture metabolising nitrite anoxically. Lines indi

measurements.

Zeng et al. (2003a) is described in Fig. 3a. The anaerobic stoi-

chiometry suggested that both PAOs and DGAOs were present

in this sludge. The sludge was estimated to contain 70% PAO

and 30%DGB, according to themethod of Lopez-Vazquez et al.

(2007a), which is based on the P release/acetate uptake ratio

(YPO4_Ac). Furthermore, Zeng et al. (2003a) observed that the

specific acetate uptake rate (qAc_PHA) was much lower as

compared to previous studies, which was attributed to the

biomass aggregation in the form of granules. Nevertheless, it

has been found that a granular sludge system can be well

modelled using the sameparameters as for flocculent biomass

(de Kreuk et al., 2007; Xavier et al., 2007). The lower maximal

activity of biomass observed in granular systems is likely due

to accumulation of inert biomass in the core of the granule.

We decided not to further increase the current complexity of

ourmodel by including a (also unknown) biomass inactivation

parameter. We solved it by adjusting the kinetic parameters

for granular biomass in the anaerobic/anoxic acetate-fed PAO

system previously calibrated by Murnleitner et al. (1997)

(Fig. 3a). It should be noted that the kinetic parameters of

acetate-fed DGBs (Fig. 2b) were also obtained from a system

containing granules with a similar size (1e3 mm) (Zeng et al.,

2003c).

The kinetic parameters employed for PAO and DGB in

Fig. 3a are shown in Table 1. Since qAc_PHA has been observed

4 5 6

X_PH A X_PH A_ me as X_Gl y X_Gl y_ me as S_ HA c S_ HA c_ me as S_ NO 3 S_ NO 3_ me as S_PO 4 S_PO4_ me as S_ NO 2 S_ NO 2_ me as

4 5 6

X_ PH A X_ PH A_me as X_ Gl y X_ Gl y_ me as S_ HA c S_ HA c_me as S_ NO 2 S_ NO 2_ me as

n a) a 70% PAOe30% DGB culture fed with acetate and b) an

cate model description and symbols indicate experimental

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 64480

to be similar between PAO and GB in anaerobic/aerobic

systems (Lopez-Vazquez et al., 2009b), it was assumed that

this trendwould also hold under anaerobic/anoxic conditions.

The anoxic kinetic parameters of Murnleitner et al. (1997)

were then reduced by the same factor (half) as qAc_PHA. The

justification for this decision is further explored in Section 3.3.

From Fig. 3a, an acceptable fit between the model

predictions and experimental data was found (see also

Appendix G), supporting the validity of the kinetic parame-

ters in a sludge containing both PAO and DGB. Interestingly,

nitrite was observed to accumulate in this system; however,

the amount was lower than the level of nitrite accumulation

observed in the DGB culture of Zeng et al. (2003c). This effect

was well described by the model when incorporating both

the PAO and DGB fractions and suggests that the fraction of

DGB contained in this sludge was likely responsible for the

nitrite accumulation. Saito et al. (2004) previously observed

that nitrite accumulation in an EBPR system was linked with

the proliferation of GAOs, which is consistent with this

result. It is possible that enriched PAO cultures not con-

taining GAOs are less likely to lead to nitrite accumulation

under anoxic conditions, assuming that Type II Accumu-

libacter are also present in the sludge (see Section 3.5) and

able to reduce nitrite but not nitrate. This is also significant

because anoxic nitrite (SHNO2) accumulation has been linked

to N2O production in both PAOs (Zhou et al., 2008) and GAOs

(Zeng et al., 2003c), thus minimising nitrite accumulation is

desirable towards minimising greenhouse gas emissions.

While N2O production was not found from the model

predictions of Fig. 3a, it should be noted that a quantifiable

relationship between SHNO2 accumulation and N2O produc-

tion remains to be elucidated for GAOs, whereas it was

developed for PAOs (Zhou et al., 2008). Predicting N2O

production by PAOs and GAOs was not one of the objectives

of this study, as more experimentally-based results would

first be needed in order to develop a proper model

description.

Furthermore, the acetate-fed DGB reactor of Zeng et al.

(2003c) was simulated with nitrite instead of nitrate as the

electron acceptor (Fig. 3b). The model predictions present

a close fit with the experimental data, supporting the validity

of the model to describe the metabolism associated with both

nitrate and nitrite reduction by DGB.

y = 4.17x - 0.03R2 = 0.98

0.00.20.40.60.81.01.21.41.61.8

0.00 0.10 0.20 0.30 0.40

qGAO,Ac_PHA (C-mol/C-mol hr)

qG

AO

,P

HA (C

-m

ol/C

-m

ol h

r)

a b

Fig. 4 e a) Relationship between anaerobic acetate uptake and

production rate; by enriched GAO cultures (Competibacter and D

parameter value.

3.3. The relationship between anaerobic VFA uptakerate and aerobic/anoxic kinetics of PAOs and GAOs

The results described above suggest that multiplying each

kinetic parameter (qVFA_PHA, qPHA, qGly, qPO4_PP) by an identical

factor can be a useful means of calibrating the kinetic rates of

metabolic models. The validity of this strategy was evaluated

by comparing the kinetic rates of PAO and GAO enriched

cultures from literature studies.

Fig. 4a illustrates the relationship between the specific

anaerobic acetate uptake rate (qAc_PHA) and specific aerobic or

anoxic PHA consumption rate (qPHA) for enriched GAO

cultures. It should be pointed out that each of these enriched

cultures, fed with acetate as the sole carbon source, was

dominated with either Competibacter or Defluviicoccus, and was

operated under either anaerobic/aerobic (Filipe et al., 2001a;

Zeng et al., 2003b; Oehmen et al., 2006; Dai et al., 2007;

Lopez-Vazquez et al., 2008; Bengtsson, 2009) or anaerobic/

anoxic conditions (Zeng et al., 2003c; Wang et al., 2008) in

systems with flocular or granular sludge. A linear relationship

was observed between these two parameters, as was also

observed between qAc_PHA and qGly (Fig. 4b). This result

suggests interdependency between the kinetic parameters.

The reason for these differences in kinetic parameters could

be related to disparities in the overall activity of the biomass in

each system. It is clear that the kinetic activity will be

proportional to the quantity of GAOs enriched in the sludge.

The overall specific activity will also be dependent on the

quantity of inactive cells and inert particulates. Indeed,

Moussa et al. (2005) showed that the active biomass fraction

had a clear impact on the kinetics of nitrifying SBR systems. In

addition to the flocular structure of the biomass, other factors

impacting the active biomass fraction include the sludge

retention time (longer SRT leads to a higher accumulation of

inert material resulting in a lower fraction of active cells), the

aerobic/anoxic fraction (cells are less active under anoxic vs.

aerobic conditions due to their lower efficiency in respiration)

and substrate availability (i.e. organic loading rate can impact

the availability of the substrate to the cells), even in very

highly enriched cultures (e.g. 80e90%).

Similarly, the relationship between the anaerobic and

aerobic/anoxic parameters of qVFA_PHA, qPHA, qGly and qPO4_PP

was analysed for enriched PAO cultures (Fig. 5). In this case,

y = 0.98x - 0.02R2 = 0.95

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0 0.1 0.2 0.3 0.4

qGAO,Ac_PHA (C-mol/C-mol hr)

qG

AO

,G

ly (C

-m

ol/C

-m

ol h

r)

aerobic or anoxic a) PHA degradation rate and b) glycogen

efluviicoccus). Error bars correspond to a 10% change in

y = 2.85x - 0.23R2 = 0.99

0.000.050.100.150.200.250.300.350.400.450.50

0. 00 0.05 0.10 0.15 0.20 0.25

qPAO,Prop_PHA (C-mol/C-mol hr)

qP

AO

,P

HA (C

-m

ol/C

-m

ol h

r)

y = 0.16x - 0.01R2 = 0.95

0.000

0.005

0.010

0.015

0.020

0.025

0 0.05 0.1 0.15 0.2 0.25

qPAO,Prop_PHA (C-mol/C-mol hr)

qP

AO

,G

ly (C

-m

ol/C

-m

ol h

r)

y = 0.034x - 0.002R2 = 0.959

0.000

0.001

0.002

0.003

0.004

0.005

0.006

0 0.05 0.1 0.15 0.2 0.25

qP

AO

,P

O4

_P

P (P

-m

ol/C

-m

ol h

r)

y = 0.08x - 0.00R2 = 0.74

0.0000.0020.0040.0060.0080.0100.0120.0140.016

0 0.05 0.1 0.15 0.2 0.25

qPAO,Prop_PHA (C-mol/C-mol hr)

µPA

O (C

-m

ol/C

-m

ol h

r)

a b

d c

qPAO,Prop_PHA (C-mol/C-mol hr)

Fig. 5 e a) Relationship between anaerobic propionate uptake and aerobic or anoxic a) PHA degradation rate, b)

polyphosphate formation rate, c) glycogen production rate, and d) biomass growth rate; by enriched PAO cultures

(Accumulibacter). Error bars correspond to a 10% change in parameter value.

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 6 4481

the sludges were enriched with Accumulibacter using propio-

nate as the sole carbon source, under either anaerobic/anoxic

conditions (Carvalho et al., 2007) or anaerobic/aerobic condi-

tions at a series of different pHs, which was found to alter the

kinetic activity of the biomass (Oehmen et al., 2005a). A linear

relationship was observed between qVFA_PHA and qPHA (Fig. 5a),

qPO4_PP (Fig. 5b), and qGly (Fig. 5c). It could be noted that one

point seemed to fall outside the trend in Fig. 5b and c, which

corresponded to the study performed at a pH of 8.5. This

deviation was more prominent when plotting the maximum

biomass growth rate (mPAO) vs. qPHA (Fig. 5d) for these studies

(mPAO could be calculated from the aerobic ammonia removal

rate by Oehmen et al., 2005a since a nitrification inhibitor was

present in the media). This result suggests that a high pH

constituted a “stressful condition” that altered PAO metabo-

lism. Furthermore, the pH of 8.5 coincidedwith a higher-than-

expected glycogen production at the expense of biomass

growth, which is consistent with the hypothesis put forth by

Murnleitner et al. (1997) that PAOs (or GAOs) favour the

replenishment of storage polymers over a high biomass

growth rate for their survival.

It should also be noted that the carbon source fed anaer-

obically (i.e. acetate or propionate) not only has an impact on

the anaerobic kinetics, but also the anoxic kinetics, whereby

the propionate-fed systems consistently display lower anoxic

kinetic rates. Indeed, the aerobic kinetic parameters

describing anaerobic/aerobic propionate-fed reactors were

also lower than those associated with acetate-fed anaerobic/

aerobic reactors (Lopez-Vazquez et al., 2009b). The reason is

likely due to a lower conversion rate for different PHA frac-

tions produced with each carbon source. After PHB, PHV and

PH2MV are converted to acetyl-CoA and propionyl-CoA,

acetyl-CoA can be directly catabolised through the TCA cycle,

while propionyl-CoA must first be converted into acetyl-CoA

(via e.g. pyruvate) before proceeding through the catabolic

pathway (Zeng et al., 2003b; Oehmen et al., 2007b). This extra

metabolic step likely leads to slower kinetic rates, necessi-

tating the expression of the aerobic (or anoxic) rates as

a function of the acetate/propionate ratio (Lopez-Vazquez

et al., 2009b). This property of acetyl-CoA vs. propionyl-CoA

metabolism has also been shown for aerobic PHA producing

cultures, in addition to PAOeGAO systems (Dias et al., 2008;

Lopez-Vazquez et al., 2009b).

The results of the sensitivity and error analysis (Appendix

G) showed that an increase or decrease of 10% in the

parameters displayed in Figs. 4 and 5 led to a small increase

(�3%) in the deviation between the experimental measure-

ments and model predictions (as calculated by NRMSD, see

Section 2.6). A change of 50% to the parameter estimates led

to a more substantial increase in the NRMSD of approxi-

mately 20%. Further, a 10% change in parameter value did

not lead to significant steady-state error propagation.

Therefore, a change in the kinetic parameters of approxi-

mately 10% does not have a significant impact on the

capability of the model to describe PAO or GAO metabolism.

Despite the fact that extraordinary conditions (e.g. high pH)

could alter PAO or GAO metabolism, the linear relationship

found in Figs. 4 and 5 suggest that these three or four main

kinetic parameters can in most situations be described

through assessment of only one parameter (qVFA_PHA), and

the interdependency between the anaerobic and aerobic/

anoxic parameters can be used to calculate the remaining

parameters. We hypothesise that the kinetics of these

organisms are constant on a per cell basis, while the

apparent rates observed in each system are a function of the

relative level of activity of the culture.

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 64482

3.4. The relationship between VFA uptake rate andaerobic/anoxic fraction

As stated above, the anaerobic VFA uptake rate is influenced

by the subsequent phase (aerobic or anoxic) that was

employed in previous cycles. The explanation for this result

could stem from either (1) a change in the microbial pop-

ulation (e.g. PAO or DGAO subgroup) that became adapted to

these different operational conditions or (2) a direct influence

of the aerobic/anoxic time fraction on the metabolism of the

same population. To clarify this point, the biomass composi-

tion in the propionate-fed bioreactor operated by Carvalho

et al. (2007) was studied along the acclimatisation period

from anaerobic/aerobic to anaerobic/anoxic conditions. From

Table 2, it can be observed that the anaerobic VFA uptake rate

steadily decreased throughout acclimatisation, as the length

of the anoxic phase was increased. However, an insignificant

change in the population structure of Accumulibacter was

found while the aerobic phase was applied; PAOII persisted

while the PAOI population did not increase. The anoxic P

uptake rate during this period also did not increase, which is

in agreement with the stable Accumulibacter population

observed. Onlywhen the aerobic phasewas eliminated did the

PAOI population and anoxic P uptake rate increase signifi-

cantly (at the expense of the PAOII population). It is note-

worthy that this correlation between anoxic P uptake rate and

Type I Accumulibacter further supports the hypothesis that

Type I Accumulibacter are nitrate-reducing PAOs and Type II

Accumulibacter are non-nitrate-reducing PAOs.

Since the dynamics of the microbial populations con-

trasted with the steady decline in propionate uptake rate

throughout acclimatisation, it was concluded that the aerobic/

anoxic fraction is proportional to qVFA_PHA. Based on this

result, qVFA_PHA was related to the aerobic/anoxic fraction

through the following relationship:

qVFA PHA ¼ qVFA PHA;AN AX þ�qVFA PHA;AN OX � qVFA PHA;AN AX

�$fOX

(7)

where fox represents the aerobic phase length divided by the

sum of the aerobic and anoxic phase lengths.

Table 2 shows that the qVFA_PHA,AN_AX is approximately half

that of the qVFA_PHA,AN_OX. This is consistentwith the results of

Kuba et al. (1996a), who also observed that the qVFA_PHA

decreased by half after adapting an acetate-fed anaerobic/

aerobic system to anaerobic/anoxic conditions.

Table 2 e Change in VFA uptake rate along acclimatisation frobioreactor operated by Carvalho et al. (2007).

Stage ofacclimatisation

Propionateuptake rate

FISHquantification (%)

C-mol/(C-mol h)

Acc-I-444 Acc-II

e 0.26 5 82

I 0.21 8 81

II 0.16 5 74

III 0.12 44 31

3.5. Describing the population dynamics of Type I andType II Accumulibacter

The applicability of this model was further tested through

running a set of simulations under the same operational

conditions as employed in the propionate-fed EBPR reactor of

Carvalho et al. (2007) during the acclimatisation from anaer-

obic/aerobic conditions to anaerobic/anoxic conditions. Since

GAOs were found to represent a very small portion of the

biomass (<5%) throughout the study, their activity was

assumed negligible, and only the PAOI and PAOII processes

were considered. The initial PAOI and PAOII concentrations

corresponded to the abundance of Type I and Type II Accu-

mulibacter, respectively, measured by FISH. The results are

displayed in Fig. 6, and show that the changes in PAOI and

PAOII abundance predicted by the model correlate very

strongly with the dynamics of Type I and Type II Accumu-

libacter. This result supports the validity of the model to

describe the population dynamics of both PAOs under

anaerobic/anoxic/aerobic conditions and further supports the

hypothesis that Type I Accumulibacter are nitrate-reducing

DPAOs, and Type II Accumulibacter are non-nitrate-reducing

PAOs. Moreover, the strong correlation between model

predictions and FISH quantification highlight the utility of the

FISH probes developed by Flowers et al. (2009) to distinguish

between DPAO and PAO. The strong agreement between the

total Type I and Type II Accumulibacter as quantified by each

individual probe with the total Accumulibacter population as

assessed by the PAOmix probes of Crocetti et al. (2000) shows

that the Type I and Type II probes covered the entire Accu-

mulibacter population present in this sludge.

The results from Fig. 6 also illustrate that the acclimatisa-

tion phase between anaerobic/aerobic and fully anaerobic/

anoxic conditions had little impact on the relative fraction of

PAOI and PAOII. The small number of PAOI present in the

sludge may have been responsible for the reduction of the

5e10 mgN/L in the 1e2 h anoxic phase, where the addition of

nitrate to the reactor was regulated based on the nitrate

demand of the culture (i.e. the quantity that could be removed

without leaking into the aerobic and subsequent anaerobic

periods). Only when the aerobic phase was completely elimi-

nated did the nitrate demand increase substantially (50 mgN/

L), and correspondingly, the PAOI population also increased.

These results are in good agreement with the model predic-

tions of Filipe and Daigger (1999), who proposed that PAO

outcompete DPAO because of their higher efficiency under

m anaerobic/aerobic to anaerobic/anoxic conditions in the

Anoxic Puptake rate

Anoxicfraction (%)

Aerobicfraction (%)

-444 Pmmol/(gVSS h)

0.10 0 100

0.10 25 75

0.13 50 50

0.63 100 0

0

10

20

30

40

50

60

70

80

90

100

0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 90

TI me ( d)

Ab

u n

da

n ce/ T

o ta l B

acter ia

( %

)

A ccu mu li bact er I PA OI ( M ode l) A ccu mu li bact er I I PAO II (M odel ) A ccu mu li bact er I +I I PA Om ix

Step 1 Step 2 Step 3

Fig. 6 e FISH quantification of Accumulibacter Type I and Type II and total Accumulibacter (Types ID II, PAOmix) in

comparison with model predictions along the acclimatisation from anaerobiceaerobic to anaerobiceanoxic conditions. Day

0: aerobic fraction[ 100%; Step 1: anoxic fraction[ 25%, aerobic fraction[ 75%; Step 2: anoxic fraction[ 50%, aerobic

fraction[ 50%; Step 3: anoxic fraction[ 100%.

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 6 4483

aerobic conditions as compared to anoxic conditions. Indeed,

the efficiency of oxidative phosphorylation (YNADH_ATP) under

anoxic conditions is approximately half the aerobic value

(Murnleitner et al., 1997; see Appendix D), and the anoxic

kinetics are also far lower than the aerobic kinetics (Appendix

G). In an effort to estimate the critical length of the anoxic

phase necessary to enrich for PAOI, the minimum SRT was

calculated according to the method of Brdjanovic et al. (1998)

for anaerobic/anoxic and anaerobic/aerobic systems fed with

either acetate or propionate. A close agreement was observed

between the calculated minimum anoxic SRT at 20 �C for

acetate (4.3 d) and propionate (4.7 d), as well as the minimum

aerobic SRT in each case (1.6 d and 1.7 d, respectively). Since

PAOI will also grow under aerobic conditions the minimum

anoxic SRT for PAOI selection could be lower in an integrated

anaerobic/anoxic/aerobic system. The critical anoxic phase

length for PAOI enrichment over PAOII still requires further

research, especially since the actual competitive difference

between the two organisms is likely small and e.g. PAOII could

compete with PAOI for nitrite in denitrification.

Indeed, it is noteworthy that the PAOII population was not

completely washed out in Fig. 6, even after 4 SRTs of biore-

actor operation andmodel simulation. In order to assess if this

effect was due to an insufficient bioreactor operation/simu-

lation time, the model was then simulated under the same

conditions for an additional 4 SRTs. It was found that the PAOI

and PAOII populations remained similar, and the relative

fraction achieved at steady-state was 55% PAOI and 45%

PAOII. The fact that PAOs did not completely washout under

anaerobic/anoxic conditions supports the hypothesis that

PAOII are able to use nitrite (GarciaMartin et al., 2006), and can

thus survive by scavenging the nitrite produced from nitrate

reducers (e.g. PAOI). This finding could also potentially explain

why nitrite was observed to accumulate in an enrichment of

DGBs (Zeng et al., 2003c), but not in the case of the PAO reactor

of Carvalho et al. (2007), as both PAOI and PAOII may have

been responsible for the nitrite reduction, while only the PAOI

were able to perform nitrate reduction.

4. Discussion

4.1. Modelling the metabolic activity of PAOI and PAOII

It has been often debated in literature whether or not PAO and

DPAO are different organisms, and more recently, if PAOs are

separated into 2 groups, whereby one group is able to denitrify

from nitrate to di-nitrogen gas, and the other from nitrite to

di-nitrogen gas. The stoichiometry and kinetics of two such

AccumulibacterTypeswere described in themetabolicmodel of

this study, and provided an ideal platform to test this

hypothesis. The activity predicted by the model for each

Accumulibacter Type agreed very well with the abundance of

each Type assessed by oligonucleotide probes (see Fig. 6). We

propose that this model can be useful in describing the

differences in Accumulibacter metabolic activity, providing

a valuable tool in assessing the impact of different Accumu-

libacter Types on EBPR performance. This will lead to advances

in our knowledge about how PAOs function, and our ability to

control the process.

Ekama and Wentzel (1999) have previously recognized the

challenge in modelling the kinetics of DPAOs, as previous

attempts to model this process always required calibration to

specific case studies and were of low predictive value. This

work provides evidence showing that the model developed in

this study can indeed be used to predict the growth and

activity of PAOI and PAOII based on the operational conditions

employed, thus providing an advantage over other modelling

strategies. Further, this work opens up the possibility for new

design and control strategies of full-scale EBPR systems based

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 64484

on achieving a desired microbial community. For example, in

COD-limited situations, it may be more desirable to enrich

PAOI as they can provide both denitrification and P removal

simultaneously. Under situations with a high influent P

concentration, it may be more desirable to maximise the PHA

driven for aerobic P uptake, which has a higher metabolic

efficiency. Future work should examine the practical appli-

cability of enriching each PAO Type in full-scale systems.

4.2. Factors affecting the kinetic parameters of metabolicmodels

The correlation between anaerobic and aerobic (or anoxic)

kinetic parameters of PAOs and GAOs was investigated in this

study. The linear relationships shown in Figs. 4 and 5 suggest

that these correlations may be useful when calibrating the

apparent PAO and GAO activity, since each aerobic/anoxic

parameter is essentially a linear function of the anaerobic VFA

uptake rate. Measuring the VFA uptake rate is experimentally

much easier then e.g. the glycogen formation or PHA

consumption rate. In calibrating a full-scale system the

default parameters for the different kinetic constants should

be changed with the same percentage as the VFA uptake rate.

This study represents the first time that such a correlation

between kinetic parameters of PAOs and GAOs has been

reported, to the best of our knowledge.

In order to confidently make use of this information, it is

important to understand the factors affecting the kinetic

parameters, and the reason why they are subject to change in

certain situations. The relative quantity of active/inactive

biomass seems to be one of the major regulatory causes,

which is consistent with the findings of Moussa et al. (2005).

Hao et al. (2010) demonstrated that decay processes have

a stronger effect on the specific activity of PAOs and GAOs (i.e.

kinetic rates lowered under famine conditions) rather than on

cell death (i.e. only small decreases in the biomass concen-

tration was observed), which further highlights the impor-

tance of the active biomass fraction when describing the

activity of these systems. The fact that the anoxic/aerobic

fraction affects the kinetic parameters, including the anaer-

obic VFA uptake rate, may also be a reflection of relative

biomass activity, since cells are less efficientmetabolising NOx

as compared to oxygen, leading to a lower overall biomass

activity under anoxic conditions. The influent carbon source

composition is another factor affecting aerobic/anoxic

kinetics, likely due to its influence on the PHA composition,

and more directly, the lower rate of propionyl-CoA metabo-

lism as compared to acetyl-CoA.

5. Conclusions

A metabolic model describing the steady-state growth,

activity and competition of PAOs and GAOs under anaero-

biceanoxiceaerobic conditions was developed, calibrated and

experimentally validated. The main conclusions from this

work are:

1) The model was able to describe well the metabolic trans-

formations occurring in enriched PAO or GAO cultures

under anaerobic/anoxic conditions, as well as mixed PAO/

GAO cultures.

2) The population dynamics of PAOI and PAOII were

successfully predicted along the acclimatisation from

anaerobic/aerobic to anaerobic/anoxic conditions. Indeed,

a strong correlation was observed between the abundance

of Accumulibacter Types I and II in the sludge and the pre-

dicted PAOI and PAOII population, suggesting clear differ-

ences in the ability of each group to denitrify.

3) Interdependency was observed between the calibrated

kinetic parameters for the apparent anaerobic, anoxic and

aerobic activities of PAOs and GAOs. This suggests that

although the kinetics of these cells on a per cell basis are

likely constant, the apparent rates are subject to change

depending on the activity level of each group of organisms

in the sludge. This inter-relationship could be exploited as

a practical means of calibrating metabolic models to

describe the apparent activity in activated sludge systems

subjected to different operational conditions.

Nomenclature

The notation used in this article is in accordance with the new

standardised framework for wastewater treatment modelling

notation (Corominas et al., 2010). Descriptions of the model

parameters are provided in the Appendices.

Acknowledgements

The authors would like to thank an anonymous reviewer for

their helpful comments. The Fundacao para a Ciencia e

a Tecnologia is acknowledged for grant SFRH/BPD/30800/2006.

Appendix. Supplementary data

Supplementary data associated with this article can be found,

in the online version, at doi:10.1016/j.watres.2010.06.017.

r e f e r e n c e s

Bengtsson, S., 2009. The utilization of glycogen accumulatingorganisms for mixed culture production ofpolyhydroxyalkanoates. Biotechnology and Bioengineering104 (4), 698e708.

Brdjanovic, D., van Loosdrecht, M.C.M., Hooijmans, C.M.,Alaerts, G.J., Heijnen, J.J., 1998. Minimal aerobic sludgeretention time in biological phosphorus removal systems.Biotechnology and Bioengineering 60 (3), 326e332.

Burow, L.C., Kong, Y.H., Nielsen, J.L., Blackall, L.L., Nielsen, P.H.,2007. Abundance and ecophysiology of Defluviicoccus spp.,glycogen accumulating organisms in full-scale wastewatertreatment processes. Microbiology 153 (1), 178e185.

Carvalho, G., Lemos, P.C., Oehmen, A., Reis, M.A.M., 2007.Denitrifying phosphorus removal: linking the processperformance with the microbial community structure. WaterResearch 41 (19), 4383e4396.

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 6 4485

Corominas, L., Rieger, L., Takacs, I., Ekama, G., Hauduc, H.,Vanrolleghem, P.A., Oehmen, A., Gernaey, K.V., VanLoosdrecht, M.C.M., Comeau, Y., 2010. New framework forstandardized notation in wastewater treatment modelling.Water Science and Technology 61 (4), 841e857.

Crocetti, G.R., Hugenholtz, P., Bond, P.L., Schuler, A., Keller, J.,Jenkins, D., Blackall, L.L., 2000. Identification ofpolyphosphate-accumulating organisms and design of 16SrRNA-directed probes for their detection and quantitation.Applied and Environmental Microbiology 66 (3), 1175e1182.

Dai, Y., Yuan, Z.G., Wang, X., Oehmen, A., Keller, J., 2007.Anaerobic metabolism of Defluviicoccus vanus related glycogenaccumulating organisms (GAOs) with acetate and propionateas carbon sources. Water Research 41 (9), 1885e1896.

Dias, J., Oehmen, A., Serafim, L.S., Lemos, P.C., Reis, M.A.M.,Oliveira, R., 2008. Metabolic modelling ofpolyhydroxyalkanoate copolymers production by mixedmicrobial cultures. BMC Systems Biology 2, 59.

Ekama, G.A., Wentzel, M.C., 1999. Difficulties and developmentsin biological nutrient removal technology and modelling.Water Science and Technology 39 (6), 1e11.

Filipe, C.D.M., Daigger, G.T., 1999. Evaluation of the capacity ofphosphorus-accumulating organisms to use nitrate andoxygen as final electron acceptors: a theoretical study onpopulation dynamics. Water Environment Research 71 (6),1140e1150.

Filipe, C.D.M.,Daigger, G.T., Grady,C.P.L., 2001a.Ametabolicmodelfor acetate uptake under anaerobic conditions by glycogenaccumulating organisms: stoichiometry, kinetics, and theeffect of pH. Biotechnology and Bioengineering 76 (1), 17e31.

Filipe, C.D.M., Daigger, G.T., Grady, C.P.L., 2001b. pH as a keyfactor in the competition between glycogen-accumulatingorganisms and phosphorus-accumulating organisms. WaterEnvironment Research 73 (2), 223e232.

Flowers, J.J., He, S., Yilmaz, S., Noguera, D.R., McMahon, K.D.,2009. Denitrification capabilities of two biological phosphorusremoval sludges dominated by different ‘CandidatusAccumulibacter’ clades. Environmental Microbiology Reports1 (6), 583e588.

Garcia Martin, H., Ivanova, N., Kunin, V., Warnecke, F., Barry, K.,McHardy, A.C., Yeates, C., He, S., Salamov, A., Szeto, E.,Dalin, E., Putnam, N., Shapiro, H.J., Pangilinan, J.L.,Rigoutsos, I., Kyrpides, N.C., Blackall, L.L., McMahon, K.D.,Hugenholtz, P., 2006. Metagenomic analysis of two enhancedbiological phosphorus removal (EBPR) sludge communities.Nature Biotechnology 24 (10), 1263e1269.

Hao, X.D., Wang, Q.L., Cao, Y.L., van Loosdrecht, M.C.M., 2010.Experimental evaluation of decrease in the activities ofpolyphosphate/glycogen-accumulating organisms due to celldeath and activity decay in activated sludge. Biotechnologyand Bioengineering. doi:10.1002/bit.22703.

Kong, Y.H., Xia, Y., Nielsen, J.L., Nielsen, P.H., 2006. Ecophysiologyof a group of uncultured gammaproteobacterial glycogen-accumulating organisms in full-scale enhanced biologicalphosphorus removal wastewater treatment plants.Environmental Microbiology 8 (3), 479e489.

de Kreuk, M.K., Picioreanu, C., Hosseini, M., Xavier, J.B., vanLoosdrecht, M.C.M., 2007. Kinetic model of a granular sludgeSBR: influences on nutrient removal. Biotechnology andBioengineering 97 (4), 801e815.

Kuba, T., Murnleitner, E., vanLoosdrecht, M.C.M., Heijnen, J.J.,1996a. A metabolic model for biological phosphorus removalby denitrifying organisms. Biotechnology and Bioengineering52 (6), 685e695.

Kuba, T., vanLoosdrecht, M.C.M., Heijnen, J.J., 1996b. Effect ofcyclic oxygen exposure on the activity of denitrifyingphosphorus removing bacteria. Water Science andTechnology 34 (1e2), 33e40.

Lopez-Vazquez, C.M., Hooijmans, C.M., Brdjanovic, D., Gijzen, H.J.,van Loosdrecht, M.C.M., 2007a. A practical method forquantification of phosphorus- and glycogen-accumulatingorganism populations in activated sludge systems. WaterEnvironment Research 79 (13), 2487e2498.

Lopez-Vazquez, C.M., Song, Y.I., Hooijmans, C.M., Brdjanovic, D.,Moussa, M.S., Gijzen, H.J., van Loosdrecht, M.C.M., 2007b.Short-term temperature effects on the anaerobic metabolismof glycogen accumulating organisms. Biotechnology andBioengineering 97 (3), 483e495.

Lopez-Vazquez, C.M., Song, Y.I., Hooijmans, C.M., Brdjanovic, D.,Moussa, M.S., Gijzen, H.J., van Loosdrecht, M.C.M., 2008.Temperature effects on the aerobic metabolism of glycogen-accumulating organisms. Biotechnology and Bioengineering101 (2), 295e306.

Lopez-Vazquez, C.M., Hooijmans, C.M., Brdjanovic, D., Gijzen, H.J.,van Loosdrecht, M.C.M., 2009a. Temperature effects onglycogen accumulating organisms. Water Research 43 (11),2852e2864.

Lopez-Vazquez, C.M., Oehmen, A., Hooijmans, C.M.,Brdjanovic, D., Gijzen, H.J., Yuan, Z.G., Van Loosdrecht, M.C.M.,2009b. Modeling the PAOeGAO competition: effects ofcarbon source, pH and temperature. Water Research 43 (2),450e462.

Lu, H., Oehmen, A., Virdis, B., Keller, J., Yuan, Z.G., 2006. Obtaininghighly enriched cultures of Candidatus AccumulibacterPhosphatis through alternating carbon sources. WaterResearch 40 (20), 3838e3848.

Meijer, S.C.F., van Loosdrecht, M.C.M., Heijnen, J.J., 2002.Modelling the start-up of a full-scale biological phosphorousand nitrogen removing WWTP. Water Research 36 (19),4667e4682.

Moussa, M.S., Hooijmans, C.M., Lubberding, H.J., Gijzen, H.J., vanLoosdrecht, M.C.M., 2005. Modelling nitrification,heterotrophic growth and predation in activated sludge.Water Research 39 (20), 5080e5098.

Murnleitner, E., Kuba, T., vanLoosdrecht, M.C.M., Heijnen, J.J.,1997. An integrated metabolic model for the aerobic anddenitrifying biological phosphorus removal. Biotechnologyand Bioengineering 54 (5), 434e450.

Oehmen, A., Vives, M.T., Lu, H., Yuan, Z.G., Keller, J., 2005a. Theeffect of pH on the competition between polyphosphate-accumulating organisms and glycogen-accumulatingorganisms. Water Research 39 (15), 3727e3737.

Oehmen, A., Yuan, Z.G., Blackall, L.L., Keller, J., 2005b.Comparison of acetate and propionate uptake bypolyphosphate accumulating organisms and glycogenaccumulating organisms. Biotechnology and Bioengineering91 (2), 162e168.

Oehmen, A., Saunders, A.M., Vives, M.T., Yuan, Z.G.,Keller, J., 2006. Competition between polyphosphate andglycogen accumulating organisms in enhanced biologicalphosphorus removal systems with acetate andpropionate as carbon sources. Journal of Biotechnology123 (1), 22e32.

Oehmen, A., Lemos, P.C., Carvalho, G., Yuan, Z.G., Keller, J.,Blackall, L.L., Reis, M.A.M., 2007a. Advances in enhancedbiological phosphorus removal: from micro to macro scale.Water Research 41 (11), 2271e2300.

Oehmen, A., Zeng, R.J., Keller, J., Yuan, Z.G., 2007b. Modeling theaerobic metabolism of polyphosphate accumulatingorganisms enriched with propionate as a carbon source.Water Environment Research 79 (13), 2477e2486.

Oehmen, A., Carvalho, G., Freitas, F., Reis, M.A.M., 2010.Assessing the abundance and activity of denitrifyingpolyphosphate accumulating organisms through molecularand chemical techniques. Water Science and Technology 61(8), 2061e2068.

wat e r r e s e a r c h 4 4 ( 2 0 1 0 ) 4 4 7 3e4 4 8 64486

Panswad, T., Doungchai, A., Anotai, J., 2003. Temperature effecton microbial community of enhanced biological phosphorusremoval system. Water Research 37 (2), 409e415.

Pijuan, M., Saunders, A.M., Guisasola, A., Baeza, J.A., Casas, C.,Blackall, L.L., 2004. Enhanced biological phosphorus removal ina sequencing batch reactor using propionate as the sole carbonsource. Biotechnology and Bioengineering 85 (1), 56e67.

Reichert, P., 1994. Aquasim e a tool for simulation and data-analysis of aquatic systems. Water Science and Technology 30(2), 21e30.

Saito, T., Brdjanovic, D., van Loosdrecht, M.C.M., 2004. Effect ofnitrite on phosphate uptake by phosphate accumulatingorganisms. Water Research 38 (17), 3760e3768.

Schuler, A.J., Jenkins, D., 2002. Effects of pH on enhancedbiological phosphorus removal metabolisms. Water Scienceand Technology 46 (4e5), 171e178.

Sin, G., Kaelin, D., Kampschreur, M.J., Takacs, I., Wett, B.,Gernaey, K.V., Rieger, L., Siegrist, H., Van Loosdrecht, M.C.M.,2008. Modelling nitrite in wastewater treatment systems:a discussion of different modelling concepts. Water Scienceand Technology 58 (6), 1155e1171.

Smolders, G.J.F., Vandermeij, J., Vanloosdrecht, M.C.M., Heijnen, J.J., 1994. Stoichiometric model of the aerobic metabolism of thebiological phosphorus removal process. Biotechnology andBioengineering 44 (7), 837e848.

Wang, X.L., Zeng, R.J., Dai, Y., Peng, Y.Z., Yuan, Z.G., 2008. Thedenitrification capability of cluster 1 Defluviioccus vanus-related glycogen-accumulating organisms. Biotechnology andBioengineering 99 (6), 1329e1336.

Xavier, J.B., De Kreuk, M.K., Picioreanu, C., Van Loosdrecht, M.C.M., 2007. Multi-scale individual-based model of microbial andbioconversion dynamics in aerobic granular sludge.Environmental Science & Technology 41 (18), 6410e6417.

Yoshida, Y., Kim, Y., Saito, T., Tanaka, K., 2009. Development ofthe modified activated sludge model describing nitriteinhibition of aerobic phosphate uptake. Water Science andTechnology 59 (4), 621e630.

Zeng, R.J., Saunders, A.M., Yuan, Z., Blackall, L.L., Keller, J., 2003a.Identification and comparison of aerobic and denitrifyingpolyphosphate-accumulating organisms. Biotechnology andBioengineering 83 (2), 140e148.

Zeng, R.J., van Loosdrecht, M.C.M., Yuan, Z.G., Keller, J., 2003b.Metabolic model for glycogen-accumulating organisms inanaerobic/aerobic activated sludge systems. Biotechnologyand Bioengineering 81 (1), 92e105.

Zeng, R.J., Yuan, Z.G., Keller, J., 2003c. Enrichment of denitrifyingglycogen-accumulating organisms in anaerobic/anoxicactivated sludge system. Biotechnology and Bioengineering 81(4), 397e404.

Zhou, Y., Pijuan, M., Yuan, Z.G., 2007. Free nitrous acid inhibitionon anoxic phosphorus uptake and denitrification by poly-phosphate accumulating organisms. Biotechnology andBioengineering 98 (4), 903e912.

Zhou, Y., Pijuan, M., Zeng, R.J., Yuan, Z., 2008. Free nitrousacid inhibition on nitrous oxide reduction bya denitrifying-enhanced biological phosphorus removalsludge. Environmental Science & Technology 42 (22),8260e8265.