Accepted Manuscript Metabolic modelling of full-scale enhanced biological phosphorus removal sludge Ana B. Lanham, Adrian Oehmen, Aaron M. Saunders, Gilda Carvalho, Per H. Nielsen, Maria A.M. Reis PII:
S0043-1354(14)00603-4
DOI:
10.1016/j.watres.2014.08.036
Reference:
WR 10843
To appear in:
Water Research
Received Date: 20 May 2014 Revised Date:
1 August 2014
Accepted Date: 23 August 2014
Please cite this article as: Lanham, A.B., Oehmen, A., Saunders, A.M., Carvalho, G., Nielsen, P.H., Reis, M.A.M., Metabolic modelling of full-scale enhanced biological phosphorus removal sludge, Water Research (2014), doi: 10.1016/j.watres.2014.08.036. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT 1
Metabolic modelling of full-scale enhanced biological
2
phosphorus removal sludge
3
Ana B. Lanhama, Adrian Oehmena,*, Aaron M. Saundersc, Gilda Carvalhoa,b, Per H.
4
Nielsenc, Maria A. M. Reisa
5
a
6
[email protected]; [email protected]; [email protected];
7
[email protected]
8
b Instituto de Biologia Experimental e Tecnológica, Apt.12, 2781-901 Oeiras, Portugal
9
c Center for Microbial Communities, Department of Biotechnology, Chemistry and
SC
RI PT
REQUIMTE/CQFB, Chemistry Department FCT-UNL, 2829-516 Caparica, Portugal,
Environmental Engineering, Aalborg University, Denmark, [email protected];
11
[email protected]
12
* Corresponding author – Tel: +351212948571 /Fax: +351212948550
M AN U
10
13 14 15
Abstract
16
This study investigates, for the first time, the application of metabolic models
17
incorporating
18
accumulating organisms (GAOs) towards describing the biochemical transformations of
19
full-scale enhanced biological phosphorus removal (EBPR) activated sludge from
20
wastewater treatment plants (WWTPs). For this purpose, it was required to modify
21
previous metabolic models applied to lab-scale systems by incorporating the anaerobic
22
utilisation of the TCA cycle and the aerobic maintenance processes based on sequential
23
utilisation of polyhydroxyalkanoates, followed by glycogen and polyphosphate. The
24
abundance of the PAO and GAO populations quantified by fluorescence in situ
25
hybridisation served as the initial conditions of each biomass fraction, whereby the
26
models were able to describe accurately the experimental data. The kinetic rates were
27
found to change among the four different WWTPs studied or even in the same plant
28
during different seasons, either suggesting the presence of additional PAO or GAO
29
organisms, or varying microbial activities for the same organisms. Nevertheless, these
30
variations in kinetic rates were largely found to be proportional to the difference in
31
acetate uptake rate, suggesting a viable means of calibrating the metabolic model. The
32
application of the metabolic model to full-scale sludge also revealed that different
accumulating
organisms
(PAOs)
and
glycogen
AC C
EP
TE D
polyphosphate
ACCEPTED MANUSCRIPT Accumulibacter clades likely possess different acetate uptake mechanisms, as a
34
correlation was observed between the energetic requirement for acetate transport across
35
the cell membrane with the diversity of Accumulibacter present. Using the model as a
36
predictive tool, it was shown that lower acetate concentrations in the feed as well as
37
longer aerobic retention times favour the dominance of the TCA metabolism over
38
glycolysis, which could explain why the anaerobic TCA pathway seems to be more
39
relevant in full-scale WWTPs than in lab-scale systems.
40
RI PT
33
Keywords : metabolic modelling; polyphosphate accumulating organisms (PAOs);
42
glycogen accumulating organisms (GAOs); anaerobic TCA cycle; glycolysis;
43
maintenance processes
SC
41
44
M AN U
45
1
Introduction
47
The biological removal of phosphate (also known as enhanced biological phosphorus
48
removal or EBPR) has been incorporated into several wastewater treatment plant
49
(WWTP) configurations and provides a more economical and sustainable alternative to
50
chemical precipitation methods of P removal (Mino et al., 1998; Oehmen et al., 2007).
TE D
46
51
For more than 20 years, an effort has been made to develop and apply activated sludge
53
models (ASM) to describe and predict the activated sludge processes, which are a useful
54
tool for plant design and optimisation (Henze et al., 2000). While ASM models use a
55
grey-box approach and focus on macroscopic processes, a different modelling approach,
56
relying on metabolic and biochemical pathways, describes the energy, redox and mass
57
balances of the cell processes (Smolders et al., 1994a). When comparing the two
58
strategies, ASM models require a plant-tailored calibration procedure that can affect a
59
higher number of variables, whereas metabolic models have been reported to require a
60
simpler calibration procedure, since all of the equations for the microbial processes are
61
inter-dependant (Seviour et al., 2010). Both approaches have been combined in the
62
Technical University of Delft model (TUDP) and successfully applied for full-scale
63
WWTPs, describing anaerobic, anoxic and aerobic processes of polyphosphate
64
accumulating organisms (PAOs) (Van Veldhuizen et al., 1999; Brdjanovic et al., 2000;
65
Meijer et al., 2001).
AC C
EP
52
ACCEPTED MANUSCRIPT 66 EBPR is carried out by PAOs using three different internal storage compounds that
68
generate energy and/or reducing power, i.e., polyphosphate (poly-P), glycogen and
69
polyhydroxyalkanoate (PHA). Anaerobically, volatile fatty acids (VFAs) like acetate
70
and propionate are converted into PHA by consuming poly-P and glycogen, while PHA
71
is then consumed for P uptake and glycogen production under anoxic and aerobic
72
conditions. Additionally, PAOs have to withstand competition from glycogen
73
accumulating organisms (GAOs), for which external parameters, such as temperature,
74
pH, COD:P ratio and carbon source, play a significant role (Oehmen et al., 2007).
75
Hence, the initial metabolic models developed for PAOs (Smolders et al., 1994a;
76
Smolders et al., 1995; Kuba et al., 1996; Murnleitner et al., 1997), were expanded to
77
include the metabolic pathways of two main GAOs, i.e., Competibacter and
78
Defluviicoccus vanus-related organisms, as well as the effects of temperature, carbon
79
source and pH on their metabolism (Lopez-Vazquez et al., 2009). Moreover, the
80
denitrification capacities of Accumulibacter, the main PAO known, and Competibacter-
81
and Defluviicoccus-GAOs were further incorporated in the model by Oehmen et al.
82
(2010b).
M AN U
SC
RI PT
67
TE D
83
However, these new additions have only been validated in lab-scale systems containing
85
enrichments of PAOs and GAOs and have not previously been tested on full-scale
86
sludge. While simplified metabolic model calibration strategies have been proposed
87
based on lab-scale results (Oehmen et al., 2010b), it is necessary to test these theories
88
using full-scale sludge in order to evaluate their applicability to more complex
89
situations. It is noteworthy to mention that when modelling the performance of full-
90
scale systems, there is an added complexity, since not only could there be unknown
91
PAOs and/or GAOs whose contribution to the phosphorus removal process is still
92
unknown, but known PAOs such as Tetrasphaera could be active, whose metabolism
93
related to EBPR is still largely unclear. Adding to this complexity is the fact that
94
wastewater influents contain a wider diversity of organic carbon sources that are subject
95
to much variability, and PAOs and GAOs make up a much smaller fraction of the total
96
microbial community in full-scale sludge as compared to lab-scale systems.
AC C
EP
84
97 98
Although PAOs have been typically modelled as using the glycolysis pathway as their
99
sole source of anaerobic reducing power generation, it has been suggested that the role
ACCEPTED MANUSCRIPT of the anaerobic TCA cycle in real WWTPs might be greater than expected as compared
101
to lab-scale results (Pijuan et al., 2008; Lanham et al., 2013). In fact, Zhou et al. (2009)
102
have shown that the TCA might have a particularly prominent role when PAOs face
103
conditions of glycogen shortage. Since WWTPs deal with variable influent
104
compositions and often with limited carbon substrate availability, this might be the
105
reason for a greater reliance on the TCA cycle in WWTPs as opposed to lab-scale
106
reactors (Lanham et al., 2013). Therefore, in order to improve the applicability of
107
metabolic models, particularly with respect to full-scale situations, the relevance of
108
incorporating the TCA cycle activity into the model should be assessed.
RI PT
100
SC
109
Furthermore, in previous metabolic models the aerobic maintenance processes predict
111
cell decay at low PHA levels, which is not consistent with experimental findings.
112
Experiments on the endogenous metabolism of PAOs (Lopez et al., 2006; Lu et al.,
113
2007; Vargas et al., 2013) observed that the aerobic maintenance processes were
114
dependant on glycogen and polyphosphate degradation following PHA depletion, with
115
minimal cell decay. This is a particularly relevant factor to include when applying the
116
model to full-scale systems, where the level of polymers stored by the sludge is much
117
lower as compared to lab-scale systems.
TE D
118
M AN U
110
In this study, a simplified version of the metabolic models previously developed by
120
Lopez-Vazquez et al. (2009) and Oehmen et al. (2010b) was adapted in order to
121
incorporate the anaerobic TCA utilisation of PAOs, in addition to the previously
122
implemented glycolytic pathway. The resulting model was tested by describing the
123
anaerobic/aerobic chemical transformations observed in activated sludge batch tests fed
124
with acetate as carbon source. The tests were carried out with sludge from four different
125
EBPR WWTPs with differing microbial compositions (including different fractions of
126
Accumulibacter, Competibacter and Defluviicoccus) and metabolisms, as shown in
127
Lanham et al. (2013). Special attention was paid to the required calibration procedure
128
necessary in order to describe the activity of each biomass, and where possible,
129
simplified calibration procedures that could be applicable to the modelling industry
130
were evaluated. In addition, theoretical simulation studies were conducted between
131
PAOs using solely the TCA cycle anaerobically (PAO_TCA) and PAOs using
132
glycolysis (PAO_Glyc) in order to better understand the conditions which may lead to
133
the use of one metabolic pathway over the other. Thus, this study is also relevant to
AC C
EP
119
ACCEPTED MANUSCRIPT 134
improve our knowledge about factors that influence the microbial metabolism in EBPR
135
systems, which is necessary in order to better understand and optimise the performance
136
of the process.
137 138
2
Materials and Methods
139
2.1
140
Four WWTPs were modelled in order to describe the anaerobic and aerobic chemical
141
transformations observed in activated sludge tested in lab-scale batch tests, fed with
142
acetate, at neutral pH (7) and at 20°C. The WWTPs studied had either an A2/O
143
configuration (Portuguese WWTPs: PT_1 and PT_2) or a Biodenitro configuration
144
coupled to a return sludge side-stream hydrolysis process (RSS) (Danish WWTPs:
145
DK_1 and DK_2). Experiments were carried out in winter and summer for the
146
Portuguese WWTPs and only in winter for the Danish WWTPs (Lanham et al., 2013).
147
Two of the WWTPs (PT_1 and DK_2) only had Accumulibacter-PAOs (approximately
148
4% as determined by quantitative fluorescence in situ hybridisation (qFISH)), whereas
149
PT_2 had significant amounts of Defluviicoccus- and Competibacter-GAOs (4-8%).
150
DK_1 presented one time point with almost 1% of Competibacter. Also, Type I and
151
Type II Accumulibacter were quantified using specific oligonucleotide probes, Acc-I-
152
444 and Acc-II-444 targeting type-I and type-II Accumulibacter-PAOs (Flowers et al.,
153
2009). The biovolume of these Accumulibacter types was determined as described in
154
Lanham et al., 2013. A complete account of the WWTPs characterisation, the batch test
155
results and the microbial population quantification can be seen in Lanham et al. (2013).
158
RI PT
SC
M AN U
TE D
EP
AC C
156 157
Experimental results
2.2
Model description
159
The model developed in this study was based on a simplified version of previous
160
metabolic models defined by Lopez-Vazquez et al. (2009) and Oehmen et al. (2010b)
161
and was compiled using AQUASIM software (v. 2.1, Reichert (1994)).
162
The present model focused only on acetate as the sole external carbon source, converted
163
into polyhydroxyalkanoate (PHA). It does not address pH nor temperature dependencies
164
(since these were controlled at 7 and 20oC in all batch tests) and aims at describing the
ACCEPTED MANUSCRIPT 165
anaerobic and aerobic transformations of Accumulibacter-PAOs (abbreviated to ACC)
166
and Competibacter and Defluviicoccus-GAOs (abbreviated to GB and DEF).
167 The model describes the acetate ( SHAc ) and the phosphate ( SPO4 ) concentrations
169
observed in the medium, the concentration of PHA ( X PHA ), glycogen ( X GLY ) and
170
polyphosphate ( X PP ) inside the bacterial cells, as well as the concentration of PAOs and
171
GAOs ( X ACC , X GB or X DEF ). The initial values for these parameters in the model were
172
based on the experimental results. The initial concentrations of the PAO and GAO
173
biomass fractions were based on the active biomass concentrations (as given by the
174
volatile suspended solids (VSS) minus the organic storage polymers, PHA and
175
glycogen, cf., Smolders et al. (1994a), converted to CmM via the biomass formula of
176
CH1.84O0.5N0.19 (Zeng et al., 2003a)), and multiplied by the fraction of Accumulibacter
177
(XACC), Competibacter (XGB)and Defluviicoccus (XDEF) detected by qFISH (Oehmen et
178
al., 2010b). In systems containing PAOs and GAOs simultaneously, their initial
179
fraction of glycogen and PHA was estimated based on the specific anaerobic yields for
180
each compound (cf., Table 1) in each type of organism as exemplified in Eq. 1 for the
181
GB initial PHA fraction of Competibacter ( X PHA ,ini ).
182 183
X
GB PHA ,ini
= X PHA,ini ×
GB fGB × YPHA , HAc
GB ACC DEF f GB × YPHA , HAc + f ACC × YPHA, HAc + f DEF × YPHA, HAc
(1)
EP
184
TE D
M AN U
SC
RI PT
168
where f GB , f ACC and f DEF are the fraction of each of these organisms as exemplified in
186
Eq. 2 for Competibacter.
187 188
AC C
185
f GB =
X GB
X GB + X ACC + X DEF
(2)
189 190
The anaerobic stoichiometry of PAOs was based on acetate uptake coupled with
191
polyphosphate hydrolysis and phosphate release, glycogen degradation and PHA
192
production. Anaerobic maintenance processes were modelled as polyphosphate
193
hydrolysis and subsequent phosphate release, which was followed by glycogen
ACCEPTED MANUSCRIPT 194
degradation, if low polyphosphate levels were attained. A complete account of the
195
anaerobic reactions and kinetics in the model is given in Appendix A-I.
196 The anaerobic yields were defined based on the utilisation of the anaerobic TCA cycle
198
or the glycolysis pathway, as determined experimentally by the anaerobic glycogen per
199
acetate yield obtained in the activated sludge batch tests (Lanham et al., 2013). The
200
overall reactions for the TCA cycle or the glycolysis stoichiometry were based on the
201
Comeau-Wentzel (Comeau et al., 1986; Wentzel et al., 1986) and Mino (Mino et al.,
202
1987) models, respectively, as presented by Smolders et al. (1994) and shown in Eqs. 3
203
and 4, respectively. Greater explanation of the incorporation of the TCA cycle
204
stoichiometry is detailed in Section 3.1.1.
205
SC
RI PT
197
The overall acetate uptake reaction where the TCA cycle is incorporated is shown
207
below (C-mol basis):
M AN U
206
208 209
1 Acetate + (0.5 + α ACC ) HPO3 + ( − 0.5 + α ACC ) H 2O → 3 0.89 PHA + 0.11CO2 + (0.5 + α ACC ) H 3 PO4
TE D
210
(3)
211
The overall acetate uptake reaction where glycolysis is incorporated is shown below
212
(conversion of acetate, glycogen and polyphosphate into PHB, CO2 and phosphate):
213
215 216 217
5 1.33PHA + 0.17CO2 + (0.25 + α ACC ) H 3 PO4 + − (0.25 + α ACC ) H 2O 12
(4)
AC C
214
EP
Acetate + 0.5Glycogen + (0.25 + α ACC ) HPO3 →
where αACC is the energy of transport of one C-mol of acetate across the cell membrane.
218
GAOs were modelled in the same way as PAOs, except with a different stoichiometry
219
(see Table 1) and excluding the processes dependant on polyphosphate or phosphate.
220
The overall anaerobic acetate uptake of GAOs is shown in Eq. 5 (Filipe et al., 2001a;
221
Zeng et al., 2002):
222
5 1 Acetate+ (1 + 2α GAO )Glycogen→ 1.75 + α GAO PHA + 0.25 + α GAO CO2 (5) 3 3
ACCEPTED MANUSCRIPT 223 The aerobic stoichiometry and kinetics of the PAO model are based on the method
225
proposed by Murnleitner et al. (1997), where PHA is degraded to contribute to
226
phosphate uptake, polyphosphate formation and glycogen replenishment. Growth is
227
determined in the model from the difference between the degraded PHA and the
228
glycogen and polyphosphate produced. The aerobic stoichiometry depends on the ATP
229
production yield per NADH oxidised with oxygen through oxidative phosphorylation,
230
which is known as the P/O ratio or δ. The same processes are applied to the GAO
231
models with the exception of the phosphate uptake and the polyphosphate formation
232
processes. As explained in more detail in Section 3.1.3, the aerobic maintenance
233
processes rely sequentially on PHA, glycogen and polyphosphate as energy sources.
234
The kinetic processes and parameters utilised in this model, summarised in Appendix
235
A-H, were mostly consistent with those presented in Lopez-Vazquez et al. (2009) and
236
Oehmen et al. (2010b), except for the differences specified in Section 3.1.
M AN U
SC
RI PT
224
237 2.3
Model calibration and validation strategy
239
The PAO model was first calibrated using the averaged results from two replicate batch
240
tests of plant PT_1, in winter time, at standard conditions (acetate as the carbon source,
241
pH=7 and 20ºC). The calibration procedure was performed on the key kinetic
242
max parameters, namely the maximum anaerobic acetate uptake rate ( qHAc ), the aerobic
243
polyphosphate formation rate ( q PP ), the aerobic PHA consumption rate ( q PHA ) and the
244
aerobic glycogen production rate ( qGLY ).The calibrated values were determined based
245
on a deviation between the experimental values and the modelled results of less than
246
10%, calculated by the normalised root mean squared deviation (NRMSD) (cf. 2.4).
247
The resulting calibrated model was then used to describe the experimental results
248
obtained for the other experiments at different time-points or in different WWTPs. In
249
order to achieve this, the strategy detailed in Oehmen et al. (2010b) was used, where
250
max qHAc was first recalibrated, if needed, followed by adjusting all other aerobic kinetic
251
rates proportionally. During this validation procedure, the initial concentrations of
252
acetate, phosphate, PHA, glycogen and polyphosphate were set to the initial
253
experimental values.
AC C
EP
TE D
238
ACCEPTED MANUSCRIPT 254
The Competibacter model and the Defluviicoccus model were first calibrated in one
255
batch test of plant DK_1 and two averaged replicate batch tests of plant PT_2 (summer),
256
respectively. The same procedure as for PAOs was applied to adjust the models to the
257
experimental results of plant PT_2 (winter).
258 2.4
Error and sensitivity analysis
260
The deviation for n data points of the values predicted by the model ( ximod el at time
261
point i ) and the experimental values ( xiexp at time point i ) was determined by
262
calculating the normalised root mean squared deviation (NRMSD) as stated in Eq. 6: n
∑ (x
exp i
i =1
NRMSD =
− ximod el ) 2
n exp − xmin
M AN U
263
SC
RI PT
259
exp max
x
264
exp exp where xmax and xmin are the maximal and minimal experimental values measured for
265
that parameter.
266
(6)
Sensitivity analyses were performed to assess the impact of the aerobic PHA
268
degradation rate and of the initial polyphosphate concentration by varying the value of
269
these parameters by ± 50% and then determining the effect on growth, levels of storage
270
compounds (PHA, glycogen and polyphosphate) and on phosphorus removal in long-
271
term simulations of 40 d.
272
274 275
2.5
Simulation studies
AC C
273
EP
TE D
267
276
Long-term simulations were performed by defining a sequencing batch reactor (SBR) in
277
AQUASIM as a mixed reactor compartment (maximum volume = 8L), with a sludge
278
retention time (SRT) of 10 d, a hydraulic retention time of approximately 0.6 d, and a 7-
279
h cycle. The simulations were executed for 40 d to ensure that repeatable profiles were
280
achieved (4 x SRT), similar to what is described in Oehmen et al. (2010b).
281
ACCEPTED MANUSCRIPT The simulations were performed either for sensitivity analysis purposes (see 2.4) or for
283
microbial competition estimates. In the latter, simulations were performed to determine
284
the theoretical competition between the two anaerobic metabolisms that are at the focus
285
of this work. Two theoretical distinct PAO organisms were defined, one using only the
286
anaerobic TCA metabolism and the other only using the glycolysis metabolism. While it
287
is recognised that PAOs can potentially perform both processes simultaneously, this
288
approach allowed examination of the factors that influence the competition between
289
different metabolic pathways. Due to hypotheses suggested by Lanham et al. (2013), the
290
competition between the two metabolisms was assessed at different acetate
291
concentrations in the feed, from 0.25 to 5 C-mM, and at different aerobic phase
292
durations, from 3 to 12 h. Simulations that varied the acetate concentration were
293
performed at a constant aerobic phase duration of 4 h and simulations that varied the
294
duration of the aerobic phase were performed at a constant acetate concentration in the
295
feed of 2 C-mM.
296
3
298
3.1
299
Results and Discussion Model development
TE D
297
M AN U
SC
RI PT
282
The goal of this study was primarily to adapt metabolic models developed for lab-scale
301
systems to describe the processes occurring in full-scale sludge. Although lab-scale
302
systems are essential to achieve a deeper understanding about key metabolic
303
transformations, the higher complexity of full-scale WWTPs might require adjustments
304
to the models developed upon lab-scale data, regarding the diversity of influent
305
characteristics, configurations and operating conditions existing in WWTPs. In
306
particular for EBPR systems, the diversity of PAOs and GAOs and of their metabolic
307
activities in real plants might be much higher than that obtained in lab-scale reactors.
308
Moreover, the enrichments obtained in the latter can contain an abundance of these
309
specific populations that are amplified as compared to full-scale systems, thus diluting
310
possible competition and synergetic interactions of flanking populations. In this study,
311
some parameters and concepts had to be adjusted from the models developed by Lopez-
312
Vazquez et al. (2009) and Oehmen et al. (2010b), as detailed below.
313
AC C
EP
300
ACCEPTED MANUSCRIPT 314
3.1.1
Incorporating the TCA cycle stoichiometry
315 Both Pijuan et al. (2008) and Lanham et al. (2013), two EBPR studies concerning full-
317
scale activated sludge, have observed the partial utilisation of the anaerobic TCA cycle
318
to different extents, as shown by the stoichiometric yields, which, as reviewed by
319
Oehmen et al. (2010a), can indicate that different biochemical pathways are being
320
employed. Smolders et al. (1994b) proposed two metabolic models for the anaerobic
321
metabolism of PAOs, one using the TCA cycle and the other employing glycolysis. The
322
anaerobic TCA cycle utilisation is associated with a higher phosphorus release to
323
ACC acetate uptake yield ( YPO 4, HAc ) and a lower glycogen consumption and PHA production
324
ACC ACC per acetate uptake ( YGLY , HAc and YPHA, HAc , respectively) than what is observed when using
325
the glycolytic pathway (Smolders et al., 1994b, cf., Table 1). Lanham et al. (2013)
326
observed different degrees of TCA cycle utilisation in different plants, and even in the
327
same plant at different periods in time, which were proportional to the glycogen
328
consumption to acetate uptake yield. Furthermore, this study verified correlations of the
329
glycogen consumption yield coefficients with the PHA production and P release yields
330
that supported the glycolysis and TCA models. Therefore, the incorporation of the TCA
331
ACC ACC cycle metabolism was performed by adjusting the YPO 4, HAc and the YPHA, HAc yields such
332
ACC that they are dependent on the YGLY , HAc yields observed in the batch tests.
SC
M AN U
TE D
333
RI PT
316
Another important aspect is that, within the cases where the anaerobic TCA cycle was
335
ACC relevant, a discrepancy was observed between some of the YPO values obtained 4 , HAc
336
experimentally and those predicted by the TCA model. The reason for this could likely
337
ACC stem from the fact that YPO is dependent on the energetic requirement for acetate 4 , HAc
338
uptake across the cell membrane (αACC), which may vary depending on whether PAOs
339
employ the TCA cycle or glycolysis. Thus, the αACC value that Smolders et al. (1994)
340
found using the glycolysis pathway (via a lab-scale culture) is not necessarily
341
transferable to the TCA cycle model. This energetic requirement (αACC), is known to be
342
dependent on pH (Smolders et al., 1994a; Filipe et al., 2001b) but also on the VFA
343
uptake mechanism of the cell (Oehmen et al., 2010a). Since pH was controlled at 7 in
344
all experiments, we hypothesised that the reason for this occasional discrepancy was
345
due to the αACC parameter changing as a function of the PAO population, with different
AC C
EP
334
ACCEPTED MANUSCRIPT Accumulibacter sub-groups possessing different VFA uptake mechanisms. Supporting
347
this hypothesis is the fact that a correlation was observed in this study between the αACC
348
parameter and the abundance of Accumulibacter Type I and Type II vs. total
349
Accumulibacter (Table 2). In cases where total Accumulibacter correlates closely
350
(between 75 and 100%) with the sum of Type I plus Type II, a higher αACC, and
351
ACC consequently a higher YPO was observed. When Type I plus Type II accounted for 4 , HAc
352
between 25 and 75% of the total Accumulibacter population, the αACC agrees well with
353
that predicted by the glycolysis model. Interestingly, Zhou et al. (2009) also found a
354
ACC value than the one predicted by Smolders et al. (1994) when the TCA higher YPO 4 , HAc
355
cycle was active, corroborating the results of this study. Therefore, two new terms were
356
ACC introduced into the model to describe the value of YPO 4, HAc in the case of the glycolysis
357
and TCA cycle pathways (Eqs. 7 and 8). The first term, defined in equation 7, accounts
358
ACC for a higher YPO 4, HAc value when the TCA cycle is active instead of glycolysis, due to the
359
necessity to increase the hydrolysis of polyphosphate to accommodate for the ATP
360
production that is no longer being generated by glycolysis. This is dependent on the
361
ACC observed glycogen per acetate yield ( YGLY , HAc ) and the model predicted yields via
362
glycolysis and the TCA cycle shown in Table 1. The second term impacts the αACC
363
parameter, as shown in equation 8. This term is expressed as a function of f ACCI , ACCII ,
364
which describes the composition of the Accumulibacter population (i.e. whether the
365
total Accumulibacter population assessed through the PAOMIX probes was mainly
366
comprised of Type I plus Type II as assessed through the FISH probes, or of other
367
clades). Furthermore, it is important to note that this increase in ATP requirement for
368
acetate transport was not observed when the glycolysis pathway was active, but only
369
when the TCA cycle was used and the total Accumulibacter population was well
370
described by the sum of Type I and Type II, as reflected in equation 8 through
371
ACC dependence on the YGLY , HAc term. This suggests that different Accumulibacter sub-
372
populations possess different acetate uptake mechanisms. Additionally, when
373
comparing the VFA uptake energetic requirements of PAOs and GAOs (
374
Table 3), it can be observed that α appears to be dependent on the level of glycolysis
375
activity of the cell, in addition to the VFA uptake mechanism. For instance, PAOs and
376
GAOs are capable of using additional VFA uptake mechanisms besides the direct use of
AC C
EP
TE D
M AN U
SC
RI PT
346
ACCEPTED MANUSCRIPT ATP (see Oehmen et al. (2010a) for a review). These mechanisms become more
378
prevalent when higher quantities of glycogen are used by the cell, and thus, they are
379
more significant in GAOs than PAOs. Furthermore, Defluviicoccus GAOs have been
380
found to have two simultaneous extra proton motive force (pmf) generation capacities
381
(i.e. through fumarate reductase and methylmalonyl-CoA decarboxylase), while
382
Competibacter GAOs only have one (through fumarate reductase) (Saunders et al.
383
2007; Burow et al. 2008). This correlates very well with the ATP requirement for
384
transport found in these organisms through metabolic models (Filipe et al. 2001;
385
Oehmen et al. 2006).
RI PT
377
387
ACC YPO 4, HAc =
SC
386 ACC YGLY 1 , HAc ACC _ Gly ACC _ TCA ACC Y + (YGLY − ) × (1 − ) + α HAc , HAc GLY , HAc ACC _ Gly 4 YGLY , HAc
389
α
ACC HAc
M AN U
388
(7)
ACC YGLY 1 = −1.1 + 0.19 × pH + (1 − ACC,_HAc ) × f ACCI , ACCII 4 YGLY , HAcGly
390
(8)
where f ACCI , ACCII is 1 for a higher coverage of total Accumulibacter by Types I and II
392
(75-100% of total Accumulibacter) and 0 for a lower coverage (25-75% of total
393
Accumulibacter). As previously reported by Smolders et al. (1994a), the range of α is
394
constrained between 0 and 0.5 mol ATP / C-mol acetate uptake. The values of the yields
395
for the two Accumulibacter metabolisms (ACC_TCA and ACC_Gly) are stated in Table
396
1.
EP
TE D
391
397
ACC ACC YPHA , HAc was defined using a similar strategy as for YPO4 , HAc (Eq. 9), where more PHA is
399
produced via the glycolysis pathway per mole of acetate uptake as compared to the TCA
400
cycle.
401
AC C
398
ACC PHA, HAc
Y
=Y
ACC _ TCA PHA , HAc
+ (Y
ACC _ Gly PHA, HAc
−Y
ACC _ TCA PHA , HAc
)×
ACC YGLY , HAc ACC _ Gly YGLY , HAc
(9)
402 403
3.1.2
Anaerobic maintenance processes
404
ANA The anaerobic maintenance coefficient ( mACC ) was determined in a blank test without
405
the addition of acetate, and its value was determined by calculating the rate of ATP
ACCEPTED MANUSCRIPT production from the rate of P release observed during this test. The batch tests
407
conducted in WWTPs that carried out a chemical phosphate precipitation polishing step,
408
namely DK_1 and DK_2, presented a relatively high anaerobic P release, which could
409
be partially due to anaerobic dissolution of iron-phosphate precipitates caused by iron-
410
reducing bacteria (Nielsen, 1996). Since it was not possible to distinguish between the P
411
release attributed to anaerobic maintenance and chemical dissolution in the tests
412
performed, it was opted to model this additional P-release as an increased anaerobic
413
maintenance coefficient, instead of adding a new chemical P release process (Table 4).
414
ANA Therefore, the anaerobic maintenance coefficient ( mACC ) should be considered as a
415
lumped parameter for the Danish WWTPs.
SC
RI PT
406
416 3.1.3
418
In previous models, the aerobic maintenance was modelled as a function of cell decay,
419
which is indirectly related to PHA degradation (Murnleitner et al., 1997). For this
420
reason, at low PHA concentrations, the model predicted cell decay, which contrasts with
421
studies where the endogenous mechanisms of PAOs and GAOs were characterised
422
(Lopez et al., 2006; Lu et al., 2007; Vargas et al. 2013). Based on the results of
423
Brdjanovic et al. (1998), Lopez et al. (2006), Lu et al. (2007) and Vargas et al. (2013),
424
the aerobic maintenance process was expanded to include a sequential dependence on
425
PHA, glycogen and then polyphosphate. These studies have shown that PAOs tend to
426
rely on energy generation from their storage polymers prior to cell decay, an effect that
427
is of high importance particularly in the substrate-limited conditions that normally exist
428
in full-scale WWTPs, unlike lab-scale systems. The kinetic equations of the three
429
sequential aerobic maintenance processes are defined in Eqs. 10 to 12:
430
AER ACC mACC , PHA = mPHA × X ACC ×
TE D
EP
AC C
431 432
Aerobic maintenance processes
M AN U
417
m
AER ACC ,GLY
=m
ACC GLY
ACC X PHA ACC ACC X PHA + K PHA
(10)
ACC ACC X GLY X PHA × X ACC × 1 − ACC × ACC ACC ACC X PHA + K PHA X GLY + KGLY
(11)
433 434
ACC X PHA AER ACC mACC = m × X × 1 − , PP PP ACC ACC ACC X PHA + K PHA
ACC X GLY × 1 − ACC ACC X GLY + KGLY
ACC X PP × ACC ACC X PP + K PP
(12)
ACCEPTED MANUSCRIPT 435 436
AER AER AER where, mPHA , mGLY and mPP are the aerobic maintenance coefficients based on PHA,
437
AER glycogen and polyphosphate respectively. mPHA was defined according to Lopez-
438
AER AER Vazquez et al. (2009) (Eq.15), while mGLY and mPP are defined in this study. From
439
Eqs. 13 and 14, the degradation of one C-mol of glycogen or of one P-mol of
440
polyphosphate produces
441
proportional to how much glycogen or polyphosphate would be needed to produce the
442
AER ATP requirements for aerobic maintenance, defined as the constant variable mATP
443
(Smolders et al., 1994a), hence obtaining the expressions presented in Eqs. 16 and 17.
444 Glycogen degradation to produce ATP (C-mol basis):
M AN U
445 446 447
1 2 1 1 1+ δ Glycogen + O2 → PHA + CO2 + H 2O + ATP 4 3 3 3 2
448
450 451
HPO3 + H2 0 → ATP + H2 PO4
452 AER mPHA =
AER 12 × mATP 6 + 27δ
454
456 457
AER 2 × mATP 1+ δ
AC C
455
EP
453
(13)
Polyphosphate degradation to produce ATP (Smolders et al. 1994b):
TE D
449
SC
RI PT
1+ δ or 1 ATP-mol, respectively. These coefficients are 2
AER mGLY =
AER AER mPP = mATP
(14)
(15)
(16)
(17)
458 459
The stoichiometric matrix for these maintenance coefficients is given in Table 5.
460 461
It was assumed that the aerobic maintenance processes would be similar in GAOs, as in
462
PAOs, i.e. with the sequential utilisation of PHA and glycogen for aerobic maintenance,
ACCEPTED MANUSCRIPT 463
excluding of course the process dependent on polyphosphate. This is consistent with the
464
results found by Vargas et al. (2013), where GAOs aerobic maintenance processes
465
relied on glycogen degradation following PHA depletion.
466 467
3.2
Model calibration and validation in the different WWTPs
RI PT
468
The PAO model was first calibrated using results from PT_1 during the winter (Figure
470
1). The strategy used in this study was to input the abundance of Accumulibacter, as
471
determined by quantitative FISH, multiplied by the active biomass concentration
472
(Oehmen et al., 2010b). This allowed to account for the microbial community covered
473
by the model, which in the case of a WWTP sludge is only a minor fraction of the total
474
community (in this study, usually approximately 5% of the total population, and never
475
surpassing 12%). This step alone yielded a fairly correct description of the chemical
476
transformations, with some adjustment required with respect to the kinetic parameters.
M AN U
SC
469
477
Similarly to lab-scale studies, the calibration procedure involved the adjustment of the
479
max , the q PHA , the q PP and the qGLY parameters (Table 6). In general, the rates were qHAc
480
slightly lower than the values obtained by Lopez-Vazquez et al. (2009), except for
481
glycogen, which had a slightly faster rate than found in lab-scale systems (Table 6).
482
After the calibration procedure, the model fitted very well with the experimental results,
483
with error values below 5% for PHA, glycogen and P, as determined by the NRMSD
484
(Table 8). The strong fitting of the model to the data further supports the anaerobic TCA
485
cycle activity in some sludges (e.g. Figure 1a). It should be noted that both the
486
glycolysis and TCA models are based on a steady-state assumption for intermediates
487
including NADH, ATP and acetyl-CoA. Some studies have found that the levels of
488
some intermediates in cells (e.g. NADH) can fluctuate over time and space in WWTPs
489
(Wos and Pollard, 2009). Nevertheless, if NADH were to be consumed anaerobically in
490
PAOs, without generation through the TCA cycle (or glycolysis), then acetate would be
491
completely converted to PHA at a ratio of 1 C-mol PHA/C-mol HAc (since some
492
acetyl-CoA must be oxidised to CO2 to generate reducing equivalents via the TCA
493
cycle), leading to an underestimation of anaerobic PHA production by the model. Such
494
a discrepancy was not observed in this study, which validates the hypothesis that the
495
TCA cycle and/or glycolysis were responsible for anaerobic NADH generation.
AC C
EP
TE D
478
ACCEPTED MANUSCRIPT 496 Other minor modifications from the lab-scale models are shown in Table 4, notably of
498
max which was the need to increase the maximal glycogen fraction ( fGLY ), which was found
499
to exceed the previously estimated threshold (0.27 C-mol/C-mol as used in Lopez-
500
Vazquez et al. (2009)) when measured in the activated sludge. This parameter was
501
introduced in the model to prevent the prediction of an unrealistically high level of
502
glycogen accumulation (Meijer et al., 2002). However, considering the low quantity of
503
PAOs and GAOs in full-scale sludge, the measured glycogen value might be more
504
highly influenced by the presence of other hydrolysable sugar-polymers resulting, for
505
example, from exopolymeric substances, from the presence of other EBPR bacteria (e.g.
506
Tetrasphaera or unidentified populations) or other populations present in the sludge. It
507
is noticeable that despite the fact that glycogen levels are approximately 1 C-mmol/L in
508
sludge from PT_1, the microorganisms hardly consume glycogen anaerobically (Figure
509
1), which could suggest a depletion of their internal storage reserves and that the
510
remaining glycogen quantified resulted from other glucose sources.
M AN U
SC
RI PT
497
511
The calibration of the GAO models necessitated the inclusion of PAOs as well, since in
513
the set of WWTPs modelled, there was no situation where GAOs, either Competibacter
514
or Defluviicoccus, were present without PAOs. The DK_1 winter_3 experiment, with a
515
population of Accumulibacter and some Competibacter, was used for calibrating the
516
Competibacter model, whereas PT_2 summer, with a population containing
517
Accumulibacter, Competibacter and Defluviicoccus was used to calibrate the
518
Defluviicoccus model. Despite this challenge, the phosphate profile was effectively
519
described by introducing, as initial concentrations, the GAO and Accumulibacter
520
abundance in the proportions determined by qFISH, confirming once more the success
521
of this methodology (Figure 1 and Table 8). The stoichiometry of glycogen and PHA
522
was also generally well described. It should be noted that the calibrated anaerobic and
523
aerobic kinetic parameters of Competibacter were consistently higher than those of
524
Defluviicoccus (Table 7), which is consistent with the results of Lopez-Vazquez et al.
525
(2009) and Oehmen et al. (2010b). Indeed, these studies showed that it was necessary to
526
model each group of GAOs independently, primarily due to their different carbon
527
source preferences (Competibacter were able to take up acetate at a much faster rate
528
than propionate, while Defluviicoccus preferred propionate uptake).
AC C
EP
TE D
512
ACCEPTED MANUSCRIPT 529 The next step was to apply the calibrated models to other experiments. Once more, the
531
strategy of inputting the PAO and GAO abundance as determined by FISH proved very
532
effective in determining the overall cycling of the target parameters (Figure 2 and Table
533
8). However, in some situations the model required some adjustments in order to
534
correctly describe the experimental data. When applying the model to the Danish plants,
535
the anaerobic maintenance coefficient had to be adjusted in order to also account for a
536
higher phosphorus release due to the presence of chemical precipitants (see 3.1.2 and
537
Table 4). Additionally, the acetate consumption rate had to be adjusted for most
538
experiments. The aerobic kinetic parameters were also adjusted but in the same
539
proportion as for the acetate uptake rate as shown in Table 6 and Table 7. This
540
corroborates the findings from lab-scale studies presented in (Oehmen et al., 2010b).
541
Interestingly, different kinetics were observed in Portuguese plants when comparing
542
summertime (faster) and wintertime (slower) experiments. In Denmark, despite the fact
543
that the experiments were all performed in winter-like conditions, DK_1 revealed the
544
same kinetics as the summertime Portuguese experiments, whereas DK_2 agreed with
545
the Portuguese wintertime kinetics (Table 6 and Table 7), suggesting this effect might
546
not entirely be related to seasonal effects, but could reflect either different levels of
547
activity of PAO and GAO cells, or the presence of other PAOs and GAOs besides
548
Accumulibacter, Competibacter and Defluviicoccus.
SC
M AN U
TE D
549
RI PT
530
One exception to the proportionality between parameters was the aerobic PHA
551
degradation rate, which remained constant in most experiments (except in DK_2, Table
552
6) and therefore was independent of the overall activity factor discussed above. Since
553
PHA is not an exclusive polymer of PAOs and GAOs, it could be produced or
554
consumed by other organisms present in the sludge, thus influencing the rates observed.
555
Nevertheless, Meijer et al. (2002) already discussed that the model was more sensitive
556
to stoichiometry than to kinetics, in steady state conditions. In fact, when varying the
557
aerobic PHA degradation rate by ± 50%, a difference of less than 10% was observed for
558
the PAO biomass concentration and
S0043-1354(14)00603-4
DOI:
10.1016/j.watres.2014.08.036
Reference:
WR 10843
To appear in:
Water Research
Received Date: 20 May 2014 Revised Date:
1 August 2014
Accepted Date: 23 August 2014
Please cite this article as: Lanham, A.B., Oehmen, A., Saunders, A.M., Carvalho, G., Nielsen, P.H., Reis, M.A.M., Metabolic modelling of full-scale enhanced biological phosphorus removal sludge, Water Research (2014), doi: 10.1016/j.watres.2014.08.036. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT 1
Metabolic modelling of full-scale enhanced biological
2
phosphorus removal sludge
3
Ana B. Lanhama, Adrian Oehmena,*, Aaron M. Saundersc, Gilda Carvalhoa,b, Per H.
4
Nielsenc, Maria A. M. Reisa
5
a
6
[email protected]; [email protected]; [email protected];
7
[email protected]
8
b Instituto de Biologia Experimental e Tecnológica, Apt.12, 2781-901 Oeiras, Portugal
9
c Center for Microbial Communities, Department of Biotechnology, Chemistry and
SC
RI PT
REQUIMTE/CQFB, Chemistry Department FCT-UNL, 2829-516 Caparica, Portugal,
Environmental Engineering, Aalborg University, Denmark, [email protected];
11
[email protected]
12
* Corresponding author – Tel: +351212948571 /Fax: +351212948550
M AN U
10
13 14 15
Abstract
16
This study investigates, for the first time, the application of metabolic models
17
incorporating
18
accumulating organisms (GAOs) towards describing the biochemical transformations of
19
full-scale enhanced biological phosphorus removal (EBPR) activated sludge from
20
wastewater treatment plants (WWTPs). For this purpose, it was required to modify
21
previous metabolic models applied to lab-scale systems by incorporating the anaerobic
22
utilisation of the TCA cycle and the aerobic maintenance processes based on sequential
23
utilisation of polyhydroxyalkanoates, followed by glycogen and polyphosphate. The
24
abundance of the PAO and GAO populations quantified by fluorescence in situ
25
hybridisation served as the initial conditions of each biomass fraction, whereby the
26
models were able to describe accurately the experimental data. The kinetic rates were
27
found to change among the four different WWTPs studied or even in the same plant
28
during different seasons, either suggesting the presence of additional PAO or GAO
29
organisms, or varying microbial activities for the same organisms. Nevertheless, these
30
variations in kinetic rates were largely found to be proportional to the difference in
31
acetate uptake rate, suggesting a viable means of calibrating the metabolic model. The
32
application of the metabolic model to full-scale sludge also revealed that different
accumulating
organisms
(PAOs)
and
glycogen
AC C
EP
TE D
polyphosphate
ACCEPTED MANUSCRIPT Accumulibacter clades likely possess different acetate uptake mechanisms, as a
34
correlation was observed between the energetic requirement for acetate transport across
35
the cell membrane with the diversity of Accumulibacter present. Using the model as a
36
predictive tool, it was shown that lower acetate concentrations in the feed as well as
37
longer aerobic retention times favour the dominance of the TCA metabolism over
38
glycolysis, which could explain why the anaerobic TCA pathway seems to be more
39
relevant in full-scale WWTPs than in lab-scale systems.
40
RI PT
33
Keywords : metabolic modelling; polyphosphate accumulating organisms (PAOs);
42
glycogen accumulating organisms (GAOs); anaerobic TCA cycle; glycolysis;
43
maintenance processes
SC
41
44
M AN U
45
1
Introduction
47
The biological removal of phosphate (also known as enhanced biological phosphorus
48
removal or EBPR) has been incorporated into several wastewater treatment plant
49
(WWTP) configurations and provides a more economical and sustainable alternative to
50
chemical precipitation methods of P removal (Mino et al., 1998; Oehmen et al., 2007).
TE D
46
51
For more than 20 years, an effort has been made to develop and apply activated sludge
53
models (ASM) to describe and predict the activated sludge processes, which are a useful
54
tool for plant design and optimisation (Henze et al., 2000). While ASM models use a
55
grey-box approach and focus on macroscopic processes, a different modelling approach,
56
relying on metabolic and biochemical pathways, describes the energy, redox and mass
57
balances of the cell processes (Smolders et al., 1994a). When comparing the two
58
strategies, ASM models require a plant-tailored calibration procedure that can affect a
59
higher number of variables, whereas metabolic models have been reported to require a
60
simpler calibration procedure, since all of the equations for the microbial processes are
61
inter-dependant (Seviour et al., 2010). Both approaches have been combined in the
62
Technical University of Delft model (TUDP) and successfully applied for full-scale
63
WWTPs, describing anaerobic, anoxic and aerobic processes of polyphosphate
64
accumulating organisms (PAOs) (Van Veldhuizen et al., 1999; Brdjanovic et al., 2000;
65
Meijer et al., 2001).
AC C
EP
52
ACCEPTED MANUSCRIPT 66 EBPR is carried out by PAOs using three different internal storage compounds that
68
generate energy and/or reducing power, i.e., polyphosphate (poly-P), glycogen and
69
polyhydroxyalkanoate (PHA). Anaerobically, volatile fatty acids (VFAs) like acetate
70
and propionate are converted into PHA by consuming poly-P and glycogen, while PHA
71
is then consumed for P uptake and glycogen production under anoxic and aerobic
72
conditions. Additionally, PAOs have to withstand competition from glycogen
73
accumulating organisms (GAOs), for which external parameters, such as temperature,
74
pH, COD:P ratio and carbon source, play a significant role (Oehmen et al., 2007).
75
Hence, the initial metabolic models developed for PAOs (Smolders et al., 1994a;
76
Smolders et al., 1995; Kuba et al., 1996; Murnleitner et al., 1997), were expanded to
77
include the metabolic pathways of two main GAOs, i.e., Competibacter and
78
Defluviicoccus vanus-related organisms, as well as the effects of temperature, carbon
79
source and pH on their metabolism (Lopez-Vazquez et al., 2009). Moreover, the
80
denitrification capacities of Accumulibacter, the main PAO known, and Competibacter-
81
and Defluviicoccus-GAOs were further incorporated in the model by Oehmen et al.
82
(2010b).
M AN U
SC
RI PT
67
TE D
83
However, these new additions have only been validated in lab-scale systems containing
85
enrichments of PAOs and GAOs and have not previously been tested on full-scale
86
sludge. While simplified metabolic model calibration strategies have been proposed
87
based on lab-scale results (Oehmen et al., 2010b), it is necessary to test these theories
88
using full-scale sludge in order to evaluate their applicability to more complex
89
situations. It is noteworthy to mention that when modelling the performance of full-
90
scale systems, there is an added complexity, since not only could there be unknown
91
PAOs and/or GAOs whose contribution to the phosphorus removal process is still
92
unknown, but known PAOs such as Tetrasphaera could be active, whose metabolism
93
related to EBPR is still largely unclear. Adding to this complexity is the fact that
94
wastewater influents contain a wider diversity of organic carbon sources that are subject
95
to much variability, and PAOs and GAOs make up a much smaller fraction of the total
96
microbial community in full-scale sludge as compared to lab-scale systems.
AC C
EP
84
97 98
Although PAOs have been typically modelled as using the glycolysis pathway as their
99
sole source of anaerobic reducing power generation, it has been suggested that the role
ACCEPTED MANUSCRIPT of the anaerobic TCA cycle in real WWTPs might be greater than expected as compared
101
to lab-scale results (Pijuan et al., 2008; Lanham et al., 2013). In fact, Zhou et al. (2009)
102
have shown that the TCA might have a particularly prominent role when PAOs face
103
conditions of glycogen shortage. Since WWTPs deal with variable influent
104
compositions and often with limited carbon substrate availability, this might be the
105
reason for a greater reliance on the TCA cycle in WWTPs as opposed to lab-scale
106
reactors (Lanham et al., 2013). Therefore, in order to improve the applicability of
107
metabolic models, particularly with respect to full-scale situations, the relevance of
108
incorporating the TCA cycle activity into the model should be assessed.
RI PT
100
SC
109
Furthermore, in previous metabolic models the aerobic maintenance processes predict
111
cell decay at low PHA levels, which is not consistent with experimental findings.
112
Experiments on the endogenous metabolism of PAOs (Lopez et al., 2006; Lu et al.,
113
2007; Vargas et al., 2013) observed that the aerobic maintenance processes were
114
dependant on glycogen and polyphosphate degradation following PHA depletion, with
115
minimal cell decay. This is a particularly relevant factor to include when applying the
116
model to full-scale systems, where the level of polymers stored by the sludge is much
117
lower as compared to lab-scale systems.
TE D
118
M AN U
110
In this study, a simplified version of the metabolic models previously developed by
120
Lopez-Vazquez et al. (2009) and Oehmen et al. (2010b) was adapted in order to
121
incorporate the anaerobic TCA utilisation of PAOs, in addition to the previously
122
implemented glycolytic pathway. The resulting model was tested by describing the
123
anaerobic/aerobic chemical transformations observed in activated sludge batch tests fed
124
with acetate as carbon source. The tests were carried out with sludge from four different
125
EBPR WWTPs with differing microbial compositions (including different fractions of
126
Accumulibacter, Competibacter and Defluviicoccus) and metabolisms, as shown in
127
Lanham et al. (2013). Special attention was paid to the required calibration procedure
128
necessary in order to describe the activity of each biomass, and where possible,
129
simplified calibration procedures that could be applicable to the modelling industry
130
were evaluated. In addition, theoretical simulation studies were conducted between
131
PAOs using solely the TCA cycle anaerobically (PAO_TCA) and PAOs using
132
glycolysis (PAO_Glyc) in order to better understand the conditions which may lead to
133
the use of one metabolic pathway over the other. Thus, this study is also relevant to
AC C
EP
119
ACCEPTED MANUSCRIPT 134
improve our knowledge about factors that influence the microbial metabolism in EBPR
135
systems, which is necessary in order to better understand and optimise the performance
136
of the process.
137 138
2
Materials and Methods
139
2.1
140
Four WWTPs were modelled in order to describe the anaerobic and aerobic chemical
141
transformations observed in activated sludge tested in lab-scale batch tests, fed with
142
acetate, at neutral pH (7) and at 20°C. The WWTPs studied had either an A2/O
143
configuration (Portuguese WWTPs: PT_1 and PT_2) or a Biodenitro configuration
144
coupled to a return sludge side-stream hydrolysis process (RSS) (Danish WWTPs:
145
DK_1 and DK_2). Experiments were carried out in winter and summer for the
146
Portuguese WWTPs and only in winter for the Danish WWTPs (Lanham et al., 2013).
147
Two of the WWTPs (PT_1 and DK_2) only had Accumulibacter-PAOs (approximately
148
4% as determined by quantitative fluorescence in situ hybridisation (qFISH)), whereas
149
PT_2 had significant amounts of Defluviicoccus- and Competibacter-GAOs (4-8%).
150
DK_1 presented one time point with almost 1% of Competibacter. Also, Type I and
151
Type II Accumulibacter were quantified using specific oligonucleotide probes, Acc-I-
152
444 and Acc-II-444 targeting type-I and type-II Accumulibacter-PAOs (Flowers et al.,
153
2009). The biovolume of these Accumulibacter types was determined as described in
154
Lanham et al., 2013. A complete account of the WWTPs characterisation, the batch test
155
results and the microbial population quantification can be seen in Lanham et al. (2013).
158
RI PT
SC
M AN U
TE D
EP
AC C
156 157
Experimental results
2.2
Model description
159
The model developed in this study was based on a simplified version of previous
160
metabolic models defined by Lopez-Vazquez et al. (2009) and Oehmen et al. (2010b)
161
and was compiled using AQUASIM software (v. 2.1, Reichert (1994)).
162
The present model focused only on acetate as the sole external carbon source, converted
163
into polyhydroxyalkanoate (PHA). It does not address pH nor temperature dependencies
164
(since these were controlled at 7 and 20oC in all batch tests) and aims at describing the
ACCEPTED MANUSCRIPT 165
anaerobic and aerobic transformations of Accumulibacter-PAOs (abbreviated to ACC)
166
and Competibacter and Defluviicoccus-GAOs (abbreviated to GB and DEF).
167 The model describes the acetate ( SHAc ) and the phosphate ( SPO4 ) concentrations
169
observed in the medium, the concentration of PHA ( X PHA ), glycogen ( X GLY ) and
170
polyphosphate ( X PP ) inside the bacterial cells, as well as the concentration of PAOs and
171
GAOs ( X ACC , X GB or X DEF ). The initial values for these parameters in the model were
172
based on the experimental results. The initial concentrations of the PAO and GAO
173
biomass fractions were based on the active biomass concentrations (as given by the
174
volatile suspended solids (VSS) minus the organic storage polymers, PHA and
175
glycogen, cf., Smolders et al. (1994a), converted to CmM via the biomass formula of
176
CH1.84O0.5N0.19 (Zeng et al., 2003a)), and multiplied by the fraction of Accumulibacter
177
(XACC), Competibacter (XGB)and Defluviicoccus (XDEF) detected by qFISH (Oehmen et
178
al., 2010b). In systems containing PAOs and GAOs simultaneously, their initial
179
fraction of glycogen and PHA was estimated based on the specific anaerobic yields for
180
each compound (cf., Table 1) in each type of organism as exemplified in Eq. 1 for the
181
GB initial PHA fraction of Competibacter ( X PHA ,ini ).
182 183
X
GB PHA ,ini
= X PHA,ini ×
GB fGB × YPHA , HAc
GB ACC DEF f GB × YPHA , HAc + f ACC × YPHA, HAc + f DEF × YPHA, HAc
(1)
EP
184
TE D
M AN U
SC
RI PT
168
where f GB , f ACC and f DEF are the fraction of each of these organisms as exemplified in
186
Eq. 2 for Competibacter.
187 188
AC C
185
f GB =
X GB
X GB + X ACC + X DEF
(2)
189 190
The anaerobic stoichiometry of PAOs was based on acetate uptake coupled with
191
polyphosphate hydrolysis and phosphate release, glycogen degradation and PHA
192
production. Anaerobic maintenance processes were modelled as polyphosphate
193
hydrolysis and subsequent phosphate release, which was followed by glycogen
ACCEPTED MANUSCRIPT 194
degradation, if low polyphosphate levels were attained. A complete account of the
195
anaerobic reactions and kinetics in the model is given in Appendix A-I.
196 The anaerobic yields were defined based on the utilisation of the anaerobic TCA cycle
198
or the glycolysis pathway, as determined experimentally by the anaerobic glycogen per
199
acetate yield obtained in the activated sludge batch tests (Lanham et al., 2013). The
200
overall reactions for the TCA cycle or the glycolysis stoichiometry were based on the
201
Comeau-Wentzel (Comeau et al., 1986; Wentzel et al., 1986) and Mino (Mino et al.,
202
1987) models, respectively, as presented by Smolders et al. (1994) and shown in Eqs. 3
203
and 4, respectively. Greater explanation of the incorporation of the TCA cycle
204
stoichiometry is detailed in Section 3.1.1.
205
SC
RI PT
197
The overall acetate uptake reaction where the TCA cycle is incorporated is shown
207
below (C-mol basis):
M AN U
206
208 209
1 Acetate + (0.5 + α ACC ) HPO3 + ( − 0.5 + α ACC ) H 2O → 3 0.89 PHA + 0.11CO2 + (0.5 + α ACC ) H 3 PO4
TE D
210
(3)
211
The overall acetate uptake reaction where glycolysis is incorporated is shown below
212
(conversion of acetate, glycogen and polyphosphate into PHB, CO2 and phosphate):
213
215 216 217
5 1.33PHA + 0.17CO2 + (0.25 + α ACC ) H 3 PO4 + − (0.25 + α ACC ) H 2O 12
(4)
AC C
214
EP
Acetate + 0.5Glycogen + (0.25 + α ACC ) HPO3 →
where αACC is the energy of transport of one C-mol of acetate across the cell membrane.
218
GAOs were modelled in the same way as PAOs, except with a different stoichiometry
219
(see Table 1) and excluding the processes dependant on polyphosphate or phosphate.
220
The overall anaerobic acetate uptake of GAOs is shown in Eq. 5 (Filipe et al., 2001a;
221
Zeng et al., 2002):
222
5 1 Acetate+ (1 + 2α GAO )Glycogen→ 1.75 + α GAO PHA + 0.25 + α GAO CO2 (5) 3 3
ACCEPTED MANUSCRIPT 223 The aerobic stoichiometry and kinetics of the PAO model are based on the method
225
proposed by Murnleitner et al. (1997), where PHA is degraded to contribute to
226
phosphate uptake, polyphosphate formation and glycogen replenishment. Growth is
227
determined in the model from the difference between the degraded PHA and the
228
glycogen and polyphosphate produced. The aerobic stoichiometry depends on the ATP
229
production yield per NADH oxidised with oxygen through oxidative phosphorylation,
230
which is known as the P/O ratio or δ. The same processes are applied to the GAO
231
models with the exception of the phosphate uptake and the polyphosphate formation
232
processes. As explained in more detail in Section 3.1.3, the aerobic maintenance
233
processes rely sequentially on PHA, glycogen and polyphosphate as energy sources.
234
The kinetic processes and parameters utilised in this model, summarised in Appendix
235
A-H, were mostly consistent with those presented in Lopez-Vazquez et al. (2009) and
236
Oehmen et al. (2010b), except for the differences specified in Section 3.1.
M AN U
SC
RI PT
224
237 2.3
Model calibration and validation strategy
239
The PAO model was first calibrated using the averaged results from two replicate batch
240
tests of plant PT_1, in winter time, at standard conditions (acetate as the carbon source,
241
pH=7 and 20ºC). The calibration procedure was performed on the key kinetic
242
max parameters, namely the maximum anaerobic acetate uptake rate ( qHAc ), the aerobic
243
polyphosphate formation rate ( q PP ), the aerobic PHA consumption rate ( q PHA ) and the
244
aerobic glycogen production rate ( qGLY ).The calibrated values were determined based
245
on a deviation between the experimental values and the modelled results of less than
246
10%, calculated by the normalised root mean squared deviation (NRMSD) (cf. 2.4).
247
The resulting calibrated model was then used to describe the experimental results
248
obtained for the other experiments at different time-points or in different WWTPs. In
249
order to achieve this, the strategy detailed in Oehmen et al. (2010b) was used, where
250
max qHAc was first recalibrated, if needed, followed by adjusting all other aerobic kinetic
251
rates proportionally. During this validation procedure, the initial concentrations of
252
acetate, phosphate, PHA, glycogen and polyphosphate were set to the initial
253
experimental values.
AC C
EP
TE D
238
ACCEPTED MANUSCRIPT 254
The Competibacter model and the Defluviicoccus model were first calibrated in one
255
batch test of plant DK_1 and two averaged replicate batch tests of plant PT_2 (summer),
256
respectively. The same procedure as for PAOs was applied to adjust the models to the
257
experimental results of plant PT_2 (winter).
258 2.4
Error and sensitivity analysis
260
The deviation for n data points of the values predicted by the model ( ximod el at time
261
point i ) and the experimental values ( xiexp at time point i ) was determined by
262
calculating the normalised root mean squared deviation (NRMSD) as stated in Eq. 6: n
∑ (x
exp i
i =1
NRMSD =
− ximod el ) 2
n exp − xmin
M AN U
263
SC
RI PT
259
exp max
x
264
exp exp where xmax and xmin are the maximal and minimal experimental values measured for
265
that parameter.
266
(6)
Sensitivity analyses were performed to assess the impact of the aerobic PHA
268
degradation rate and of the initial polyphosphate concentration by varying the value of
269
these parameters by ± 50% and then determining the effect on growth, levels of storage
270
compounds (PHA, glycogen and polyphosphate) and on phosphorus removal in long-
271
term simulations of 40 d.
272
274 275
2.5
Simulation studies
AC C
273
EP
TE D
267
276
Long-term simulations were performed by defining a sequencing batch reactor (SBR) in
277
AQUASIM as a mixed reactor compartment (maximum volume = 8L), with a sludge
278
retention time (SRT) of 10 d, a hydraulic retention time of approximately 0.6 d, and a 7-
279
h cycle. The simulations were executed for 40 d to ensure that repeatable profiles were
280
achieved (4 x SRT), similar to what is described in Oehmen et al. (2010b).
281
ACCEPTED MANUSCRIPT The simulations were performed either for sensitivity analysis purposes (see 2.4) or for
283
microbial competition estimates. In the latter, simulations were performed to determine
284
the theoretical competition between the two anaerobic metabolisms that are at the focus
285
of this work. Two theoretical distinct PAO organisms were defined, one using only the
286
anaerobic TCA metabolism and the other only using the glycolysis metabolism. While it
287
is recognised that PAOs can potentially perform both processes simultaneously, this
288
approach allowed examination of the factors that influence the competition between
289
different metabolic pathways. Due to hypotheses suggested by Lanham et al. (2013), the
290
competition between the two metabolisms was assessed at different acetate
291
concentrations in the feed, from 0.25 to 5 C-mM, and at different aerobic phase
292
durations, from 3 to 12 h. Simulations that varied the acetate concentration were
293
performed at a constant aerobic phase duration of 4 h and simulations that varied the
294
duration of the aerobic phase were performed at a constant acetate concentration in the
295
feed of 2 C-mM.
296
3
298
3.1
299
Results and Discussion Model development
TE D
297
M AN U
SC
RI PT
282
The goal of this study was primarily to adapt metabolic models developed for lab-scale
301
systems to describe the processes occurring in full-scale sludge. Although lab-scale
302
systems are essential to achieve a deeper understanding about key metabolic
303
transformations, the higher complexity of full-scale WWTPs might require adjustments
304
to the models developed upon lab-scale data, regarding the diversity of influent
305
characteristics, configurations and operating conditions existing in WWTPs. In
306
particular for EBPR systems, the diversity of PAOs and GAOs and of their metabolic
307
activities in real plants might be much higher than that obtained in lab-scale reactors.
308
Moreover, the enrichments obtained in the latter can contain an abundance of these
309
specific populations that are amplified as compared to full-scale systems, thus diluting
310
possible competition and synergetic interactions of flanking populations. In this study,
311
some parameters and concepts had to be adjusted from the models developed by Lopez-
312
Vazquez et al. (2009) and Oehmen et al. (2010b), as detailed below.
313
AC C
EP
300
ACCEPTED MANUSCRIPT 314
3.1.1
Incorporating the TCA cycle stoichiometry
315 Both Pijuan et al. (2008) and Lanham et al. (2013), two EBPR studies concerning full-
317
scale activated sludge, have observed the partial utilisation of the anaerobic TCA cycle
318
to different extents, as shown by the stoichiometric yields, which, as reviewed by
319
Oehmen et al. (2010a), can indicate that different biochemical pathways are being
320
employed. Smolders et al. (1994b) proposed two metabolic models for the anaerobic
321
metabolism of PAOs, one using the TCA cycle and the other employing glycolysis. The
322
anaerobic TCA cycle utilisation is associated with a higher phosphorus release to
323
ACC acetate uptake yield ( YPO 4, HAc ) and a lower glycogen consumption and PHA production
324
ACC ACC per acetate uptake ( YGLY , HAc and YPHA, HAc , respectively) than what is observed when using
325
the glycolytic pathway (Smolders et al., 1994b, cf., Table 1). Lanham et al. (2013)
326
observed different degrees of TCA cycle utilisation in different plants, and even in the
327
same plant at different periods in time, which were proportional to the glycogen
328
consumption to acetate uptake yield. Furthermore, this study verified correlations of the
329
glycogen consumption yield coefficients with the PHA production and P release yields
330
that supported the glycolysis and TCA models. Therefore, the incorporation of the TCA
331
ACC ACC cycle metabolism was performed by adjusting the YPO 4, HAc and the YPHA, HAc yields such
332
ACC that they are dependent on the YGLY , HAc yields observed in the batch tests.
SC
M AN U
TE D
333
RI PT
316
Another important aspect is that, within the cases where the anaerobic TCA cycle was
335
ACC relevant, a discrepancy was observed between some of the YPO values obtained 4 , HAc
336
experimentally and those predicted by the TCA model. The reason for this could likely
337
ACC stem from the fact that YPO is dependent on the energetic requirement for acetate 4 , HAc
338
uptake across the cell membrane (αACC), which may vary depending on whether PAOs
339
employ the TCA cycle or glycolysis. Thus, the αACC value that Smolders et al. (1994)
340
found using the glycolysis pathway (via a lab-scale culture) is not necessarily
341
transferable to the TCA cycle model. This energetic requirement (αACC), is known to be
342
dependent on pH (Smolders et al., 1994a; Filipe et al., 2001b) but also on the VFA
343
uptake mechanism of the cell (Oehmen et al., 2010a). Since pH was controlled at 7 in
344
all experiments, we hypothesised that the reason for this occasional discrepancy was
345
due to the αACC parameter changing as a function of the PAO population, with different
AC C
EP
334
ACCEPTED MANUSCRIPT Accumulibacter sub-groups possessing different VFA uptake mechanisms. Supporting
347
this hypothesis is the fact that a correlation was observed in this study between the αACC
348
parameter and the abundance of Accumulibacter Type I and Type II vs. total
349
Accumulibacter (Table 2). In cases where total Accumulibacter correlates closely
350
(between 75 and 100%) with the sum of Type I plus Type II, a higher αACC, and
351
ACC consequently a higher YPO was observed. When Type I plus Type II accounted for 4 , HAc
352
between 25 and 75% of the total Accumulibacter population, the αACC agrees well with
353
that predicted by the glycolysis model. Interestingly, Zhou et al. (2009) also found a
354
ACC value than the one predicted by Smolders et al. (1994) when the TCA higher YPO 4 , HAc
355
cycle was active, corroborating the results of this study. Therefore, two new terms were
356
ACC introduced into the model to describe the value of YPO 4, HAc in the case of the glycolysis
357
and TCA cycle pathways (Eqs. 7 and 8). The first term, defined in equation 7, accounts
358
ACC for a higher YPO 4, HAc value when the TCA cycle is active instead of glycolysis, due to the
359
necessity to increase the hydrolysis of polyphosphate to accommodate for the ATP
360
production that is no longer being generated by glycolysis. This is dependent on the
361
ACC observed glycogen per acetate yield ( YGLY , HAc ) and the model predicted yields via
362
glycolysis and the TCA cycle shown in Table 1. The second term impacts the αACC
363
parameter, as shown in equation 8. This term is expressed as a function of f ACCI , ACCII ,
364
which describes the composition of the Accumulibacter population (i.e. whether the
365
total Accumulibacter population assessed through the PAOMIX probes was mainly
366
comprised of Type I plus Type II as assessed through the FISH probes, or of other
367
clades). Furthermore, it is important to note that this increase in ATP requirement for
368
acetate transport was not observed when the glycolysis pathway was active, but only
369
when the TCA cycle was used and the total Accumulibacter population was well
370
described by the sum of Type I and Type II, as reflected in equation 8 through
371
ACC dependence on the YGLY , HAc term. This suggests that different Accumulibacter sub-
372
populations possess different acetate uptake mechanisms. Additionally, when
373
comparing the VFA uptake energetic requirements of PAOs and GAOs (
374
Table 3), it can be observed that α appears to be dependent on the level of glycolysis
375
activity of the cell, in addition to the VFA uptake mechanism. For instance, PAOs and
376
GAOs are capable of using additional VFA uptake mechanisms besides the direct use of
AC C
EP
TE D
M AN U
SC
RI PT
346
ACCEPTED MANUSCRIPT ATP (see Oehmen et al. (2010a) for a review). These mechanisms become more
378
prevalent when higher quantities of glycogen are used by the cell, and thus, they are
379
more significant in GAOs than PAOs. Furthermore, Defluviicoccus GAOs have been
380
found to have two simultaneous extra proton motive force (pmf) generation capacities
381
(i.e. through fumarate reductase and methylmalonyl-CoA decarboxylase), while
382
Competibacter GAOs only have one (through fumarate reductase) (Saunders et al.
383
2007; Burow et al. 2008). This correlates very well with the ATP requirement for
384
transport found in these organisms through metabolic models (Filipe et al. 2001;
385
Oehmen et al. 2006).
RI PT
377
387
ACC YPO 4, HAc =
SC
386 ACC YGLY 1 , HAc ACC _ Gly ACC _ TCA ACC Y + (YGLY − ) × (1 − ) + α HAc , HAc GLY , HAc ACC _ Gly 4 YGLY , HAc
389
α
ACC HAc
M AN U
388
(7)
ACC YGLY 1 = −1.1 + 0.19 × pH + (1 − ACC,_HAc ) × f ACCI , ACCII 4 YGLY , HAcGly
390
(8)
where f ACCI , ACCII is 1 for a higher coverage of total Accumulibacter by Types I and II
392
(75-100% of total Accumulibacter) and 0 for a lower coverage (25-75% of total
393
Accumulibacter). As previously reported by Smolders et al. (1994a), the range of α is
394
constrained between 0 and 0.5 mol ATP / C-mol acetate uptake. The values of the yields
395
for the two Accumulibacter metabolisms (ACC_TCA and ACC_Gly) are stated in Table
396
1.
EP
TE D
391
397
ACC ACC YPHA , HAc was defined using a similar strategy as for YPO4 , HAc (Eq. 9), where more PHA is
399
produced via the glycolysis pathway per mole of acetate uptake as compared to the TCA
400
cycle.
401
AC C
398
ACC PHA, HAc
Y
=Y
ACC _ TCA PHA , HAc
+ (Y
ACC _ Gly PHA, HAc
−Y
ACC _ TCA PHA , HAc
)×
ACC YGLY , HAc ACC _ Gly YGLY , HAc
(9)
402 403
3.1.2
Anaerobic maintenance processes
404
ANA The anaerobic maintenance coefficient ( mACC ) was determined in a blank test without
405
the addition of acetate, and its value was determined by calculating the rate of ATP
ACCEPTED MANUSCRIPT production from the rate of P release observed during this test. The batch tests
407
conducted in WWTPs that carried out a chemical phosphate precipitation polishing step,
408
namely DK_1 and DK_2, presented a relatively high anaerobic P release, which could
409
be partially due to anaerobic dissolution of iron-phosphate precipitates caused by iron-
410
reducing bacteria (Nielsen, 1996). Since it was not possible to distinguish between the P
411
release attributed to anaerobic maintenance and chemical dissolution in the tests
412
performed, it was opted to model this additional P-release as an increased anaerobic
413
maintenance coefficient, instead of adding a new chemical P release process (Table 4).
414
ANA Therefore, the anaerobic maintenance coefficient ( mACC ) should be considered as a
415
lumped parameter for the Danish WWTPs.
SC
RI PT
406
416 3.1.3
418
In previous models, the aerobic maintenance was modelled as a function of cell decay,
419
which is indirectly related to PHA degradation (Murnleitner et al., 1997). For this
420
reason, at low PHA concentrations, the model predicted cell decay, which contrasts with
421
studies where the endogenous mechanisms of PAOs and GAOs were characterised
422
(Lopez et al., 2006; Lu et al., 2007; Vargas et al. 2013). Based on the results of
423
Brdjanovic et al. (1998), Lopez et al. (2006), Lu et al. (2007) and Vargas et al. (2013),
424
the aerobic maintenance process was expanded to include a sequential dependence on
425
PHA, glycogen and then polyphosphate. These studies have shown that PAOs tend to
426
rely on energy generation from their storage polymers prior to cell decay, an effect that
427
is of high importance particularly in the substrate-limited conditions that normally exist
428
in full-scale WWTPs, unlike lab-scale systems. The kinetic equations of the three
429
sequential aerobic maintenance processes are defined in Eqs. 10 to 12:
430
AER ACC mACC , PHA = mPHA × X ACC ×
TE D
EP
AC C
431 432
Aerobic maintenance processes
M AN U
417
m
AER ACC ,GLY
=m
ACC GLY
ACC X PHA ACC ACC X PHA + K PHA
(10)
ACC ACC X GLY X PHA × X ACC × 1 − ACC × ACC ACC ACC X PHA + K PHA X GLY + KGLY
(11)
433 434
ACC X PHA AER ACC mACC = m × X × 1 − , PP PP ACC ACC ACC X PHA + K PHA
ACC X GLY × 1 − ACC ACC X GLY + KGLY
ACC X PP × ACC ACC X PP + K PP
(12)
ACCEPTED MANUSCRIPT 435 436
AER AER AER where, mPHA , mGLY and mPP are the aerobic maintenance coefficients based on PHA,
437
AER glycogen and polyphosphate respectively. mPHA was defined according to Lopez-
438
AER AER Vazquez et al. (2009) (Eq.15), while mGLY and mPP are defined in this study. From
439
Eqs. 13 and 14, the degradation of one C-mol of glycogen or of one P-mol of
440
polyphosphate produces
441
proportional to how much glycogen or polyphosphate would be needed to produce the
442
AER ATP requirements for aerobic maintenance, defined as the constant variable mATP
443
(Smolders et al., 1994a), hence obtaining the expressions presented in Eqs. 16 and 17.
444 Glycogen degradation to produce ATP (C-mol basis):
M AN U
445 446 447
1 2 1 1 1+ δ Glycogen + O2 → PHA + CO2 + H 2O + ATP 4 3 3 3 2
448
450 451
HPO3 + H2 0 → ATP + H2 PO4
452 AER mPHA =
AER 12 × mATP 6 + 27δ
454
456 457
AER 2 × mATP 1+ δ
AC C
455
EP
453
(13)
Polyphosphate degradation to produce ATP (Smolders et al. 1994b):
TE D
449
SC
RI PT
1+ δ or 1 ATP-mol, respectively. These coefficients are 2
AER mGLY =
AER AER mPP = mATP
(14)
(15)
(16)
(17)
458 459
The stoichiometric matrix for these maintenance coefficients is given in Table 5.
460 461
It was assumed that the aerobic maintenance processes would be similar in GAOs, as in
462
PAOs, i.e. with the sequential utilisation of PHA and glycogen for aerobic maintenance,
ACCEPTED MANUSCRIPT 463
excluding of course the process dependent on polyphosphate. This is consistent with the
464
results found by Vargas et al. (2013), where GAOs aerobic maintenance processes
465
relied on glycogen degradation following PHA depletion.
466 467
3.2
Model calibration and validation in the different WWTPs
RI PT
468
The PAO model was first calibrated using results from PT_1 during the winter (Figure
470
1). The strategy used in this study was to input the abundance of Accumulibacter, as
471
determined by quantitative FISH, multiplied by the active biomass concentration
472
(Oehmen et al., 2010b). This allowed to account for the microbial community covered
473
by the model, which in the case of a WWTP sludge is only a minor fraction of the total
474
community (in this study, usually approximately 5% of the total population, and never
475
surpassing 12%). This step alone yielded a fairly correct description of the chemical
476
transformations, with some adjustment required with respect to the kinetic parameters.
M AN U
SC
469
477
Similarly to lab-scale studies, the calibration procedure involved the adjustment of the
479
max , the q PHA , the q PP and the qGLY parameters (Table 6). In general, the rates were qHAc
480
slightly lower than the values obtained by Lopez-Vazquez et al. (2009), except for
481
glycogen, which had a slightly faster rate than found in lab-scale systems (Table 6).
482
After the calibration procedure, the model fitted very well with the experimental results,
483
with error values below 5% for PHA, glycogen and P, as determined by the NRMSD
484
(Table 8). The strong fitting of the model to the data further supports the anaerobic TCA
485
cycle activity in some sludges (e.g. Figure 1a). It should be noted that both the
486
glycolysis and TCA models are based on a steady-state assumption for intermediates
487
including NADH, ATP and acetyl-CoA. Some studies have found that the levels of
488
some intermediates in cells (e.g. NADH) can fluctuate over time and space in WWTPs
489
(Wos and Pollard, 2009). Nevertheless, if NADH were to be consumed anaerobically in
490
PAOs, without generation through the TCA cycle (or glycolysis), then acetate would be
491
completely converted to PHA at a ratio of 1 C-mol PHA/C-mol HAc (since some
492
acetyl-CoA must be oxidised to CO2 to generate reducing equivalents via the TCA
493
cycle), leading to an underestimation of anaerobic PHA production by the model. Such
494
a discrepancy was not observed in this study, which validates the hypothesis that the
495
TCA cycle and/or glycolysis were responsible for anaerobic NADH generation.
AC C
EP
TE D
478
ACCEPTED MANUSCRIPT 496 Other minor modifications from the lab-scale models are shown in Table 4, notably of
498
max which was the need to increase the maximal glycogen fraction ( fGLY ), which was found
499
to exceed the previously estimated threshold (0.27 C-mol/C-mol as used in Lopez-
500
Vazquez et al. (2009)) when measured in the activated sludge. This parameter was
501
introduced in the model to prevent the prediction of an unrealistically high level of
502
glycogen accumulation (Meijer et al., 2002). However, considering the low quantity of
503
PAOs and GAOs in full-scale sludge, the measured glycogen value might be more
504
highly influenced by the presence of other hydrolysable sugar-polymers resulting, for
505
example, from exopolymeric substances, from the presence of other EBPR bacteria (e.g.
506
Tetrasphaera or unidentified populations) or other populations present in the sludge. It
507
is noticeable that despite the fact that glycogen levels are approximately 1 C-mmol/L in
508
sludge from PT_1, the microorganisms hardly consume glycogen anaerobically (Figure
509
1), which could suggest a depletion of their internal storage reserves and that the
510
remaining glycogen quantified resulted from other glucose sources.
M AN U
SC
RI PT
497
511
The calibration of the GAO models necessitated the inclusion of PAOs as well, since in
513
the set of WWTPs modelled, there was no situation where GAOs, either Competibacter
514
or Defluviicoccus, were present without PAOs. The DK_1 winter_3 experiment, with a
515
population of Accumulibacter and some Competibacter, was used for calibrating the
516
Competibacter model, whereas PT_2 summer, with a population containing
517
Accumulibacter, Competibacter and Defluviicoccus was used to calibrate the
518
Defluviicoccus model. Despite this challenge, the phosphate profile was effectively
519
described by introducing, as initial concentrations, the GAO and Accumulibacter
520
abundance in the proportions determined by qFISH, confirming once more the success
521
of this methodology (Figure 1 and Table 8). The stoichiometry of glycogen and PHA
522
was also generally well described. It should be noted that the calibrated anaerobic and
523
aerobic kinetic parameters of Competibacter were consistently higher than those of
524
Defluviicoccus (Table 7), which is consistent with the results of Lopez-Vazquez et al.
525
(2009) and Oehmen et al. (2010b). Indeed, these studies showed that it was necessary to
526
model each group of GAOs independently, primarily due to their different carbon
527
source preferences (Competibacter were able to take up acetate at a much faster rate
528
than propionate, while Defluviicoccus preferred propionate uptake).
AC C
EP
TE D
512
ACCEPTED MANUSCRIPT 529 The next step was to apply the calibrated models to other experiments. Once more, the
531
strategy of inputting the PAO and GAO abundance as determined by FISH proved very
532
effective in determining the overall cycling of the target parameters (Figure 2 and Table
533
8). However, in some situations the model required some adjustments in order to
534
correctly describe the experimental data. When applying the model to the Danish plants,
535
the anaerobic maintenance coefficient had to be adjusted in order to also account for a
536
higher phosphorus release due to the presence of chemical precipitants (see 3.1.2 and
537
Table 4). Additionally, the acetate consumption rate had to be adjusted for most
538
experiments. The aerobic kinetic parameters were also adjusted but in the same
539
proportion as for the acetate uptake rate as shown in Table 6 and Table 7. This
540
corroborates the findings from lab-scale studies presented in (Oehmen et al., 2010b).
541
Interestingly, different kinetics were observed in Portuguese plants when comparing
542
summertime (faster) and wintertime (slower) experiments. In Denmark, despite the fact
543
that the experiments were all performed in winter-like conditions, DK_1 revealed the
544
same kinetics as the summertime Portuguese experiments, whereas DK_2 agreed with
545
the Portuguese wintertime kinetics (Table 6 and Table 7), suggesting this effect might
546
not entirely be related to seasonal effects, but could reflect either different levels of
547
activity of PAO and GAO cells, or the presence of other PAOs and GAOs besides
548
Accumulibacter, Competibacter and Defluviicoccus.
SC
M AN U
TE D
549
RI PT
530
One exception to the proportionality between parameters was the aerobic PHA
551
degradation rate, which remained constant in most experiments (except in DK_2, Table
552
6) and therefore was independent of the overall activity factor discussed above. Since
553
PHA is not an exclusive polymer of PAOs and GAOs, it could be produced or
554
consumed by other organisms present in the sludge, thus influencing the rates observed.
555
Nevertheless, Meijer et al. (2002) already discussed that the model was more sensitive
556
to stoichiometry than to kinetics, in steady state conditions. In fact, when varying the
557
aerobic PHA degradation rate by ± 50%, a difference of less than 10% was observed for
558
the PAO biomass concentration and
