Towards functional molecular fingerprintsemi_1898 1717..1727
Maxime Dumont,1* Jérôme Harmand,1,2
Alain Rapaport2 and Jean-Jacques Godon1
1Laboratoire de Biotechnologie de l’Environnement,INRA UR050, Avenue des Etangs, Narbonne, France.2INRA-INRIA MERE Research Team, UMR-ASB, 2place Viala, Montpellier, France.
Summary
One of the most important challenges in microbialecology is to determine the ecological function ofdominant microbial populations in their environment.In this paper we propose a generic method couplingfingerprinting and mathematical tools to achieve thefunctional assigning of bacteria detected in microbialconsortia. This approach was tested on a nitrificationbioprocess where two functions carried out by twodifferent communities could be clearly distinguished.The mathematical theory of observers of dynamicalsystems has been used to design a dynamic estima-tor of the active biomass concentration of each func-tional community from the available measurementson nitrifying performance. Then, the combinationof phylotypes obtained by fingerprinting that bestapproximated the estimated trajectories of eachfunctional biomass was selected through a randomoptimization method. By this way, a nitritation ornitratation function was assigned to each phylotypedetected in the ecosystem by means of functionalmolecular fingerprints. The results obtained by thisapproach were successfully compared with the infor-mation obtained from 16S rDNA identification. Thisoriginal approach can be used on any biosysteminvolving n successive cascading bioreactions per-formed by n communities.
Introduction
In recent years, the use of molecular tools for a betterunderstanding of microbial communities has beensteadily on the increase. Fingerprint pattern analysis,mostly based on 16S rDNA sequences, appears to be oneof the best established molecular tools in microbialecology. The fingerprints, obtained by several techniques
such as denaturing gradient gel electrophoresis (DGGE)or single strand conformation polymorphism (SSCP), canbe considered as giving an overall picture of the wholemicrobial ecosystem. In such pictures, bacterial commu-nities appear as discrete bands through DGGE or asdiscrete peaks through SSCP, both of which emerge fromthe background signal. Progress in the analysis ofmolecular fingerprints has enabled researchers to extractincreasing amounts of information about microbial eco-systems (Marzorati et al., 2008). First, peaks or bandsdetected on these fingerprints were used to estimate andanalyse the species richness, structure and dynamics ofmicrobial communities (Muyzer and Smalla, 1998; Loiselet al., 2006). Second, statistical analyses were carried outto determine the influence of environmental parameterson microbial community structure (Fromin et al., 2002).Subsequently, the range-weighted richness reflecting thecarrying capacity of an ecosystem was determined (Mar-zorati et al., 2008). Recently, a newly defined ecologicalindex based on information given by molecular fingerprint-ing has also been proposed for characterizing microbialecosystems (Marzorati et al., 2008). Nevertheless, theinformation coded in the fingerprints have not yet beenexploited for possibly determining the ecological func-tion(s) of dominant microbial populations in their environ-ment: in other words, ‘who does what’, which nowrepresents one of the most important challenges inmicrobial ecology.
In this paper, we present a new generic method, cou-pling molecular fingerprints and mathematical tools, forallocating such functional roles to bacteria in their envi-ronment. To test this original approach, we have con-ducted experiments on a nitrifying ecosystem in which twofunctions (i.e. nitritation and nitratation) carried out by twospecific microbial communities [the ammonium-oxidizingbacteria (AOB) community for nitritation and the nitrite-oxidizing bacteria (NOB) community for nitratation] couldbe clearly distinguished. Moreover, nitrifying bacteriapresent a strong link between their phylogeny and biologi-cal functions, which enabled us to confront resultsobtained by our mathematical method with the resultsgiven by 16S rDNA identification.
Results
Applied strategy
The procedure applied in the present study is presented inFig. 1.
First, in a given microbial ecosystem, bioreactions weredescribed on the basis of measurements of physicalinputs (input substrate concentrations and flow rates) andoutputs (the substrate and the total biomass concentra-tions) (Fig. 1, I).
Second, thanks to an understanding of the bioreactionunder consideration, a mass-balance model of the‘system’ was built up (here, two reactions were studied,nitritation and nitratation, each one carried out by a differ-ent microbial community: the AOB community for nitrita-tion and the NOB community for nitratation) that describesthe time evolution of the concentrations of each popula-tion (Fig. 1, II).
Third, the establishment of this dynamical model hasallowed us to design a dynamical estimator based on thetheory of observers. More precisely, the goal of anobserver is to reconstruct or estimate the unmeasuredvariables of the model from the knowledge of inputs andonline measurements of the system, which are here thebiomass concentrations of each functional community inthe system (i.e. in the present case, the AOB and NOBconcentrations) (Fig. 1, III).
Finally, from the estimates of the functional biomassesdelivered by the observer, along with the dynamics ofphylotype abundance revealed by molecular finger-printing, functional assigning for each phylotype wasperformed by way of an optimization method. Functionalmolecular fingerprints were the final result of thisapproach (Fig. 1, IV).
This strategy was applied to data obtained from twinnitrifying chemostats (denoted A and B) which were main-tained for 525 days under disturbed environmentalparameters.
Dynamics of nitritation and nitratation functions
Nitrification is a cascading bioreaction where two func-tions are involved: nitritation, consisting in the oxidation ofammonium nitrogen into nitrites, and nitratation, the oxi-dation of nitrites into nitrates. Due to regular disturbancessuch as changes in operating temperature, variations ininput substrate concentration, changes in the flow rate orbiotic perturbations resulting from the addition of biomass,these two functions were fulfilled differently throughoutthe experimental period in the chemostats (Fig. 2).
From their inoculation up to day 183, only nitrite pro-duction was performed with high efficiency in both chemo-stats. A start of the nitratation function was obtained aftermodification of the operating temperature from 30°C to25°C on day 183. At this time, both chemostats displayeddifferences in nitrifying performance. In chemostat A,nitrate production at first followed the increase in inputsubstrate concentration applied on day 198. After thisinitial increase, however, nitrate production quicklydecreased, its total collapse leading to nitrite accumula-tion. In chemostat B, nitrate production seemed to beinhibited by the increase of input substrate concentrationbut was maintained, in contrast to chemostat A. In addi-
Fig. 1. Theoretical scheme of themethodology used for functional communityassigning of bacteria in their environmentusing observers and molecular fingerprinting.(I) Bioreactions in a given microbialecosystem. (II) Mathematical modelling of thebioreactions. (III) Estimate of functionalbiomasses using a mathematical observer.(IV) Functional assigning of molecular speciespresent in the microbial ecosystem.
Macroscopicmass-balance
model
Mathematicalobserver
I
II
III
IVStatistical
distribution
Abioticsparameters
Reaction scheme:NH4
+ → NO2- + NO3
-
Dynamic of populational measurments
Validation of the model
Estimates dynamics of functional biomasses
Functional molecular fingerprint
Dynamic of functional measurments
Dynamic of functional measurments
STUPTUOSTUPNI ENTITY
1718 M. Dumont, J. Harmand, A. Rapaport and J.-J. Godon
tion, the nitratation function in B followed the increase ininput substrate concentration made on day 198 and waslost only after a major increase in the flow rate introducedon day 337. After restoration of this environmental param-eter on day 372, the twin chemostats stopped showingfunctional divergence. Effectively, the nitratation functionreappeared in both chemostats, maintaining high perfor-mance right to the end of the experimental procedure (day525).
Dynamics of microbial communities estimated by themathematical observer
In a nitrifying process, each effective function (i.e. nitrita-tion and nitratation) is carried out by a specific microbialcommunity. The nitritation reaction is done by AOBwhereas the nitratation reaction is done by NOB.
The AOB and NOB concentrations, which best explainthe nitrifying performances, were estimated using dedi-cated observers as described in Dumont and colleagues(2008) (refer to Experimental procedures for details)taking into account the total biomass measurements(Fig. 3).
Estimated active AOB and NOB biomass showed evo-lutions similar to those of nitrite and nitrate concentrationsrespectively. During the first period of the process, whereonly the nitritation function was observed in both chemo-stats, the estimated AOB concentration displayed equiva-lent values to total biomass measurements whereas theconcentration of NOB appeared to be nil. During the
period when nitrates were produced, NOB biomass reg-istered relatively weak values in comparison with AOBbiomass.
Dynamics of phylotypes detected bymolecular fingerprints
During the experiment, 42 phylotypes were detected inchemostat A whereas 41 were detected in chemostat B,from 132 and 136 SSCP profiles respectively (Fig. 4).
After inoculation, the same phylotype, corresponding topeak 38 on the SSCP profiles (Table 1), was dominant inchemostats A and B with high relative abundance (70% oftotal biomass on average). For a short period of time,coexistence with another dominant phylotype (the samein both chemostats), corresponding to peak 35, wasobserved in response to a temporary break in the flow ratecorresponding to the first disturbance (Fig. 4). The bioticdisturbance made on day 121, consisting in the addition ofnitrifying biomass (AOB and NOB) and which led to a lightdecrease in nitritation performance, had a major effect onthe relative abundance of the dominant phylotype (peak38) in both chemostats. Whereas in chemostat A thisphylotype showed very responsive structural resilience, inchemostat B the phylotype continued to decrease slowlybut remained dominant despite its weak relative abun-dance (above 20% of total biomass). Modification of theoperating temperature from 30°C to 25°C on day 181 ledto the emergence of another phylotype from the back-ground, corresponding to peak 41, which decreased
0.0
0.5
1.0
1.5
2.0
2.5
0 50 100 150 200 250 300 350 400 450 500
A
0.0
0.5
1.0
1.5
2.0
2.5
0 50 100 150 200 250 300 350 400 450 500
B
In g N l
Days
-1
Fig. 2. Nitrifying performances in the two regularly disturbed chemostats A and B. Black continuous line with diamond-shaped signsrepresents nitrite, dotted black line with round signs represents nitrates and dotted grey line with triangular signs represents residualammonium concentrations expressed in g N l-1 throughout the duration of experiments (in days). Disturbances introduced during the kineticsin both reactors are indicated as follows: circle with arrow represents increasing or decreasing flow rate, black arrow represents bioticdisturbance, a black triangular sign represents increasing or decreasing substrate concentration and black star represents a decrease inoperating temperature from 30°C to 25°C.
quickly, probably due to the increase of the input substrateconcentration on day 198. After this modification, themicrobial structures of chemostats A and B diverged. Inchemostat A, the initial dominant phylotype (peak 38)remained so, with high relative abundance; whereas inchemostat B, a marked successive alternation of majorphylotypes was observed. These differences in thedynamics of microbial diversity might explain the differ-ences in nitrification performance observed in bothchemostats at this time. After this period of divergence,both chemostats presented the same structural pattern asafter inoculation, with an important domination of phylo-type 38. This domination was nevertheless upset after thelast modification to input substrate concentration, carriedout on day 432, when coexistence between several majorphylotypes was observed. Finally, the second biotic dis-turbance, made on day 503 and consisting in the additionof NOB biomass, engendered a switch in the major phy-lotype in both chemostats. Phylotype 38 decreased andthe phylotype 36 became dominant until the end of thekinetic in both chemostats.
Statistical assigning of phylotypes toa functional community
The combination of phylotype concentrations detected bySSCP which best approximated to the estimated concen-trations of functional communities were looked for bymeans of a random optimization method. Thus, theprobability for each phylotype of belonging to the
nitritation or the nitratation community was obtained(Table 1).
Then the phylotypes were regrouped into three classes:AOB community, NOB community or not determined(Table 2).
For chemostat A, functional community allocation wasperformed for 35 phylotypes (Table 1). Nineteen molecu-lar species, representing around 65% of total biomass,were regrouped in the AOB community (Tables 1 and 2).These phylotypes were the major phylotypes detected inthis chemostat, in terms of maximum relative abundanceand length of presence during the lifetime of the experi-ments (Table 2). Sixteen, representing around 26%, wereregrouped in the NOB community and seven phylotypes,representing around 9% of total biomass, were notassigned to either the AOB or the NOB community(Tables 1 and 2).
The results obtained for chemostat B were similar tothose obtained for chemostat A. Functional allocation wasperformed for 33 phylotypes (Table 1). Seventeen phylo-types representing around 60% of the total biomass wereregrouped in the AOB community; 16 representing around24% were regrouped in the NOB community; and eightphylotypes representing around 15% of total biomasswere not assigned to either the AOB or the NOB commu-nity (Tables 1 and 2).
Molecular assigning of phylotypes to bacterial species
Assigning phylotypes detected by molecular fingerprintingto bacterial species was carried out by the sequencing of
0.0
0.2
0.4
0.6
0 50 100 150 200 250 300 350 400 450 500
0.0
0.2
0.4
0.6
0 50 100 150 200 250 300 350 400 450 500
A
B
In g N l-1
Days
Fig. 3. Evolution of functional communities and total biomass concentrations in both chemostats A and B. Grey continuous line with roundsigns represents dynamics of total biomass obtained by measurements, dotted black line with diamond-shaped signs represents activebiomass of AOB community generated by observers, and dotted black line with triangular signs represents active biomass of NOBcommunities generated by observers expressed in g N l-1. Disturbances made during the kinetics in both reactors are indicated as describedin Fig. 2.
1720 M. Dumont, J. Harmand, A. Rapaport and J.-J. Godon
16S rDNA of clones coming from two samples: one fromchemostat A on day 153, the second from chemostat B onday 273 (Fig. 4). Thirteen phylotypes, observed in bothchemostats through their SSCP peak migration, wereidentified (Table 3).
These phylotypes represented 64% and 61.5% of thetotal fingerprint area in samples A and B respectively.
Among these 13 phylotypes, only four were clearly asso-ciated with a nitrification function. One of the phylotypeswas identified as a Nitrosomonas sp. (Table 3: peaknumber 13) corresponding to an autotrophic nitritationspecies (AOB community) and three were identified asautotrophic nitratation species (NOB community): a Nitro-spirae sp. and two Nitrobacter sp.
Table 1. Statistical assigning of phylotypes to the AOB or the NOB community.
Mean relative abundance was calculated for each phylotype during its period (in days) of presence. Phylotypes highlighted in black represent theAOB community, phylotypes highlighted in grey represent the NOB community and phylotypes highlighted in white represent no AOB and no NOBfunctional community. Bold peak numbers with an asterisk refer to assigned peaks (Fig. 4 and Table 3).
Comparison between statistical and molecular assigning
The results given by 16S rDNA sequencing for thesephylotypes are in accordance with the results obtainedby our mathematical technique (Table 3). Nitrosomonaswas assigned to the AOB community with a probabilityof 100% in both chemostats whereas Nitrospirae and
the two Nitrobacter were assigned to the NOB commu-nity with a probability of 86%, 100% and 66% in chemo-stat A and 57%, 100% and 74% in chemostat Brespectively.
Nine other phylotypes identified by 16S rDNA appearedto be heterotrophic bacteria but their function in the nitri-fying ecosystem was not ascertained. So mathematical
0.0
0.2
0.4
0.6
0.8
0 50 100 150 200 250 300 350 400 450 500
0.0
0.2
0.4
0.6
0.8
0 50 100 150 200 250 300 350 400 450 500
A
B
Relative abundance
Days
38
36
3638
35
41
41
35
29 14 23
25
19
29
35
36
38
11
38
352
30
Fig. 4. Relative abundance dynamics of the phylotypes detected by fingerprints. Disturbances created during the kinetics in both reactors areindicated as described in Fig. 2. Molecular fingerprints in boxed insert corresponding to samples from which 16S rDNA sequencing wascarried out. The numbers under some peaks refer to identifications obtained by the method presented in Tables 1 and 3.
Table 2. Results of the K mean analysis made on functional assignation percentages obtained for each phylotypes detected in the chemostatsA and B.
Chemostat Functional communityNumber ofphylotypes
Centroid of each cluster(percentage of functionalassignment) Mean distance
functional assigning obtained by our approach cannot becompared with 16S rDNA for these heterotrophicbacteria.
Discussion
The functional assigning of molecular fingerprints wasobtained with the help of dedicated mathematical observ-ers. This method required only the results of functionalmeasurements along with the dynamics of the microbialcommunity obtained by molecular fingerprinting. Thismethod was tested on a nitrifying microbial communitywhere two functions performed by two different commu-nities could be clearly distinguished. Taking into accountthe link for nitrifying bacteria between phylogeny andbiological function, the results obtained by our mathemati-cal approach were confronted with those obtained by 16SrDNA identification. These results can be assessed atdifferent levels.
At the community level, the results showed on averagea total relative abundance of 65% and 60% for AOB andof 26 and 24% for NOB, in chemostats A and B respec-tively (Table 2). These results appear to be in the samerange for each community, generally estimated at 2/3–1/3
for AOB and NOB, respectively, in the nitrogen removalprocess (Li et al., 2007).
At the phylotype level, a nitrifying ecosystem waschosen due to the strong link between phylogeny andbiological functions which characterizes nitrifying bacteria(Purkhold et al., 2000) and which has permitted anexperimental validation of our mathematical approach.Despite this assertion, only four of the phylotypes couldbe assigned unambiguously to a functional group by 16SrRNA gene sequence analysis and the same results wereobtained for them using our mathematical approaches.This small number of phylotypes clearly associated with anitrification function highlights the limitations of this stan-dard molecular method in addressing the question of ‘whodoes what’, even in such a favourable case as nitrification.The four assigned bacteria: a Nitrosomonas (AOB), aNitrospirae and two Nitrobacter (NOB) were those mostoften cited in the bibliography concerning nitrificationbioprocess (Schroeder, 1985). The dynamics of thesephylotypes throughout the 523 days corresponded tothe dynamics of functional measurements. Effectively,the phylotype identified as a Nitrospirae (Table 3: peaknumber 2) showed maximal abundance in both chemo-stats just after the modification of operating temperature,
Table 3. Results of 16S rDNA sequencing identification carried out for 13 phylotypes present in both chemostats and their functional assigningobtained by the mathematical approach.
No. of peaks Chemostat
Blast information Nitrification function
Phylogenetic identification % of similarity Accession number Clonesa From 16S rDNA From observer
a. The number of clones identified under the same name in the database, compared with the total number of clones obtained for each peak.Phylotypes highlighted in black represent the AOB community, phylotypes highlighted in grey represent the NOB community and phylotypeshighlighted in white represent no AOB and no NOB functional community.
corresponding to the start of nitrate production. In thesame way, the first identified Nitrobacter (Table 3: peaknumber 5) displayed high maximum relative abundance,with 11.2% in chemostat A and 26.7% in chemostat B,maxima which were reached during the final period whenenvironmental parameters corresponded to a stabilizationof nitrate production in both chemostats. The other phy-lotype can be phylogenetically identified by sequencingbut cannot be related to functional groups, as previousstudies have already shown (Egli et al., 2003; Bougardet al., 2006). These undetermined phylotypes can be con-sidered as AOB, NOB or associated with either AOB orNOB. The example of peak number 35 (Table 3), whichhad been identified as a Bacteroidetes sp., is interesting(Fig. 4). This phylotype grew after the first environmentalmodification (when the flow rate was temporarily stopped)and replaced as dominant the phylotype identified asNitrosomonas sp. (AOB) without any consequences onnitrite production. Bacteroides were never found as AOBwhereas our mathematical approach predicted that thisphylotype belongs to an AOB community with a 100%probability. Nevertheless, different strains of bacteria suchas Pseudomonas, Bacillus, Diaphorobacter, Alcaligenes,Tiosphaera, Comamonas . . . were already identified asheterotrophic AOB (Su et al., 2006; Lin et al., 2007;Ahmad et al., 2008; Hayatsu et al., 2008). These bacteriahave the same capability for ammonium oxidation asautotrophic AOB (Kim et al., 2005). In the same way,heterotrophic NOB were also described (Khardenaviset al., 2007).
Another assumption to explain the results obtained forthese heterotrophic bacteria is to consider that these dif-ferent phylotypes have no nitrification function but were inclose interaction with the AOB or NOB phylotypes. That iswhy, in the present study, it seems to be more appropriateto talk about functional community assignment ratherthan functional assignment. Effectively, through the math-ematical method presented here, we cannot determineexactly ‘who does what’ (i.e. functional assigning), corre-sponding to which species performed nitrite or nitrateproduction. Nevertheless, we can determine ‘who isinvolved in what’ (i.e. functional community assigning), or,in other words, which species interact to perform a givenfunction in the ecosystem. Moreover, molecular tech-niques can be coupled advantageously with our math-ematical approach. Effectively, these techniques, suchas cloning-sequencing used here to valid the results orstable isotope probing make it possible to determine ‘whodoes what’ in an ecosystem whereas our method,enabled us to determine ‘who is involved in what’. So,coupling both approaches can be used to determinewhich species interact to perform a given function in anecosystem and, among these species, which are directlyactive in a given function.
Functional community assigning by mathematicalapproach could be applied to many other microbial eco-systems but nevertheless this method, coupling math-ematical observers and various molecular fingerprints,shows some limitations. The first is the requirement ofclearly defined functions which can be measured bychemical, enzymatic, physical, mechanical or othermeans. These measurements should enable to design asatisfactory reaction scheme with mass-balance consid-erations. Indeed, functional community assignmentsgiven by our mathematical approach depend on theequation describing the studied system. The secondlimitation is due to molecular fingerprint saturation. Inhigh diversity ecosystems, sequence co-migrationsarise. These co-migration events make molecular finger-printing useless for ascertaining the dynamics of thephylotypes detected (Jossi et al., 2006; Loisel et al.,2006).
The main advantage of functional community assigningis that mathematical observers are independent of thekinetics of the process (e.g. growth rate, temperature{).These types of mathematical observers have proved to besuitable for various biological processes (Alcaraz-Gonzalez et al., 2005). Another interesting advantage ofthis technique, compared with standard fitting methodssuch as the least-square one, is to obtain a filtering of thedata. The estimation is provided recursively with time,guaranteeing that the estimation error converges towardszero, despite measurement noise or some uncertaintieson the model. This leads to good robustness properties.
Moreover, only direct functional measurements andyield coefficients, which can be quite easily evaluatedfrom experiments or from the literature, are needed togenerate the active biomass of each functional commu-nity using these observers.
The other advantage of this approach is the possibilityit gives of assigning an observed function to a known orunknown phylotype directly within a complex microbialcommunity. Moreover, such assigning of functions in situcan be carried out for given environmental parameters orfor biotic interactions. In microbial ecology, such assign-ing opens the door to interactive models and, thus, to abetter understanding of the links between structural diver-sity and ecosystem functioning. In ecosystem engineer-ing, it could be used to control the optimization ofbioprocesses by testing different environmental condi-tions or assembled communities.
Thus, this approach can easily be used for all biopro-cesses which function on a general mass-balance modelof n successive cascading bioreactions carried out by ncommunities [i.e. in which the product of the ith reaction isthe substrate of the (i + 1)th reaction]. Such processesinclude anaerobic digestion, cheese ecosystems, waste-water treatment, digestive tract activity . . . etc.
1724 M. Dumont, J. Harmand, A. Rapaport and J.-J. Godon
Bioreactor conditions and macroscopic measurements
The experimental set-up consisted of two continuously mixed6.5 l (working volume) all-glass chemostats inoculatedbeforehand with activated sludge from the municipal sewagetreatment plant at Coursan (Aude, France). Both chemostats(A and B) were operated in strictly identical conditions over2 years. Air flow rate was maximum to ensure good fluidiza-tion and provide enough oxygen for the nitrification processwhereas pH was measured and maintained around 7 by theautomatic addition of an alkaline solution. Chemostats A andB were fuelled by the same synthetic mineral medium com-posed of ammonium sulfate (with concentration varying from0.5 to 2 g l-1) as the nitrogen source and a mineral solution.
The total biomass was measured by calculating the weightof 50 ml of sample after drying 24 h at 105°C, minus theweight of the synthetic medium dried under the same condi-tions and the weights of nitrite and nitrate.
Chemical analysis consisted of off-line quantification ofresidual ammonium, nitrites and nitrates by an ion chroma-tography system (Dionex 100) using conductivity detection.
Sampling and extraction of total genomic DNA
Twenty millilitres of samples was collected from the middle ofeach chemostat three times per hydraulic retention time. Thesamples were centrifuged at 13 000 r.p.m. (10 min, 20°C).Supernatants were collected for chemical analysis whereaspellets were resuspended in 1 ml of 4 M guanidine thiocyan-ate Tris-HCl 0.1 M at pH 7.5 and 300 ml of 10% N-lauroylsarcosine. Aliquots of 500 ml were placed in 2 ml screw-capmicrocentrifuge tubes and stored at -20°C before DNAextraction. The extraction and purification of total genomicDNA were performed with the Qiagen DNA stool mini kit,following the manufacturer’s instructions.
Single strand conformation polymorphism analysisand 16S rDNA identification
For SSCP analysis, a short fragment (200 bp) of the V3region of the 16S rDNA gene was PCR amplified using theuniversal bacterial primers W49 (ACGGTCCAGACTCCTACGGG) and 6-FAM labelled W104 (TTACCGCGGCTGCTGGCAC) (Eurogentec, Belgium), and Pfu Turbo DNApolymerase (Stratagene, Holland), as described by Wery andcolleagues (2008). Single strand conformation polymorphismcapillary electrophoresis with an ABI 3100 genetic analyser(Applied Biosystems) was done using the protocol describedby Wery and colleagues (2008). Single strand conformationpolymorphism raw data were exported into the easy-to-handle csv format using the Chromagna shareware (devel-oped by Dr Mark J. Miller at the US National Institute ofHealth) and statistics were performed using SAFUM (Zembet al., 2007) and the Matlab 6.5 software (MathWorks). Thedynamics of each phylotype were obtained from the kineticsof molecular fingerprints using SAFUM (Zemb et al., 2007).Using the internal SSCP standard (Rox), this open-sourceprogram first aligns all the fingerprints of the kinetics. Then it
calculates the area under each peak for each fingerprint. Thetotal area generated by the signal is normalized to 1 so thatthe relative abundance of a peak can be compared from onefingerprint to another and in this way the dynamics of theabundance of each peak (i.e. phylotype) can be obtained.Finally, the proportion of the background (i.e. differencebetween total area and area of all peaks) was subtractedfrom the total biomass measurement; then the abundance ofeach peak was multiplied by the total biomass remaining inorder to obtain the dynamics of the concentration of eachphylotype. Identification of bacterial peaks revealed on theSSCP profiles was obtained as described in Dabert and col-leagues (2001). Each sequence was identified by correlationto the closest species in the sequence database (GenBank),using the BLAST algorithm.
Mathematics tools
Macroscopic model used. Bioreactions were considered asdynamic systems with defined inputs (flow rate, substrateconcentrations . . .) and outputs (concentrations of reactioncomponents). From such a systemic point of view, a macro-scopic mass-balance model can be developed as follows:
dXdt
S D X
dXdt
S D X
dSdt
SY
X S
AA A
BB B
A
AA in
= ( ) −( )
= ( ) −( )
= −( )
+ −
μ
μ
μ
1
2
1 1 SS D
dSdt
SY
XS
YX S D
dSdt
SY
X S D
1
2 1 22
2 23
( )
= −( )
−( )
−
=( )
−
μ μ
μ
A
AA
B
BB
B
BB
(1)
where: XA and XB represent the concentrations of AOB andNOB, Sin, S1, S2 and S3 are, respectively, the ammoniuminput, residual ammonium, nitrites and nitrates measuredconcentrations, YA and YB are the yield coefficients of AOBand NOB, mA and mB are the growth rates of AOB and NOB, Dis the dilution rate (ratio of the input flow rate and the volume).Design of the observers. The macroscopic model used cangenerate AOB and NOB concentrations from functional mea-surements (Sin, S1, S2 and S3) but requires unknown bioreac-tion kinetics, mA and mB. To counter this drawback, a newsystem called an ‘observer’ was designed based on the mac-roscopic model in order to estimate XA and XB without priorknowledge of the growth rates mA and mB.
An observer (Bastin and Dochain, 1990) is a mathematicalentity originating from the theory of dynamical systems. In thepresent case, it was built up using the macroscopic mass-balance model [1]. We used the mass invariance property ofthe model [1] in considering the following changes ofvariables:
ZXY
S ZXY
S S1 1 2 1 2= + = + +A
A
B
B
and
The derivatives of these new variables permit to obtain thetwo independent following systems:
From these two systems, two independent observers can bederive which guarantees the convergence of the estimation:lim A At X X→∞ − =ˆ 0 that and lim B Bt X X→∞ − =ˆ 0. These twoobservers use the available inputs S1 and S2 to obtain andestimates XA and XB independently of mA and mB as follows:
dZdt
D Z SdZdt
D Z S
X Y Z S
ˆˆ
ˆˆ
ˆ ˆ ; ˆ
11
22
1 1
= − −( ) = − −( )= −( )
in in
A A
and
and XX Y Z S SB B= − −( )ˆ2 1 2
Indeed, it can be shown that X̂A and X̂B convergetowards XA and XB. Because of this property, such systemswere called ‘observer’ or sometime ‘software sensor’. In addi-tion, taking advantage that we measure the total biomass:XT = XA + XB, the robustness and the performance of theseobservers can be improved in coupling them in the followingway:
dZdt
D Z S G X X Xˆ
ˆ ˆ ˆ11 1= − −( ) + + −( )in A B T
dZdt
D Z S G X X Xˆ
ˆ ˆ ˆ22 2= − −( ) + + −( )in A B T
where: G1 and G2 are tuning parameters (see Dumont et al.,2008 for more details).
Optimization procedure
The final step in the procedure consisted in an optimizationprocedure to find the combination of phylotypes detected bySSCP which best explained the biomass trajectories gener-ated by the observer. The total number of possible combina-tions of assignments being too high to test all of them, insteada random selection method has been used. A random lot of10 phylotypes was taken from the total number of detectedphylotypes and this sampling process was repeated 10 000times. This proportion was chosen after a preliminary studyhad shown such values were sufficient to approximate thefunctional community without any loss of information and,also, because random optimization with these values ap-peared to be feasible in terms of computer calculation time.For each random lot, the combination of these 10 molecularspecies, which best approximated the active biomass of thefunctional communities, enabled us to assign them to eitherthe NOB or the AOB community. After 10 000 repetitions ofrandom selection of lots of 10 phylotypes, the probability ofeach phylotype belonging to the AOB or the NOB communitywas ascertained. A statistical analysis of these probabilities inthe form of a K mean analysis was carried out to obtain thethree functional assignation groups: AOB community, NOBcommunity, not determined.
Acknowledgements
The authors gratefully acknowledge the help of NicolasBernet, Valérie Bru, Bart Haegeman, Jérôme Hamelin,
Claude Lobry and Frank Poly whose support made this studypossible.
References
Ahmad, N., Xu, H., Chen, L., Liu, Z., and Liu, S. (2008)Enhanced biological nutrient removal by the alliance of aheterotrophic nitrifying strain with a nitrogen removing eco-system. J Environ Sci 20: 216–223.
Alcaraz-Gonzalez, V., Salazar-Pena, R., Gonzalez-Alvarez,V., Gouzé, J.L., and Steyer, J.P. (2005) A tunable multi-variable nonlinear robust observer for biological systems.C R Biol 328: 317–325.
Bastin, G., and Dochain, D. (1990) On-Line Estimation andAdaptative Control of Bioreactors. Amsterdam, the Nether-lands: Elsevier.
Bougard, D., Bernet, N., Dabert, P., Delgenes, J.P., andSteyer, J.P. (2006) Influence of closed loop control onmicrobial diversity in a nitrification process. Water SciTechnol 53: 85–93.
Dabert, P., Sialve, B., Delgenes, J.P., Moletta, R., andGodon, J.J. (2001) Characterisation of the microbial 16SrDNA diversity of an aerobic phosphorus-removal ecosys-tem and monitoring of its transition to nitrate respiration.Appl Microbiol Biotechnol 55: 500–509.
Dumont, M., Rapaport, A., Harmand, J., Benyahia, B., andGodon, J.J. (2008) Observers for microbial ecology – Howincluding molecular data into bioprocess modeling? In Pro-ceeding of the 16th Mediterranean Conference on Controland Automation, June 25–27, Congress Centre, Ajaccio,Corsica, France, pp. 1381–1386.
Egli, K., Bosshard, F., Werlen, C., Lais, P., Siegrist, H.,Zehnder, A.J.B., and van der Meer, J.R. (2003) Microbialcomposition and structure of a rotating biological contactorbiofilm treating ammonium-rich wastewater without organiccarbon. Microb Ecol 45: 419–432.
Fromin, N., Hamelin, J., Tarnawski, S., Roesti, D., Jourdain-Miserey, K., Forestier, N., et al. (2002) Statistical analysisof denaturing gel electrophoresis (DGE) fingerprinting pat-terns. Environ Microbiol 4: 634–643.
Hayatsu, M., Tago, K., and Saito, M. (2008) Various players inthe nitrogen cycle: diversity and functions of the microor-ganisms involved in nitrification and denitrification. Soil SciPlant Nutr 54: 33–45.
Jossi, M., Fromin, N., Tarnawski, S., Kohler, F., Gillet, F.,Aragno, M., and Hamelin, J. (2006) How elevated pCO2modifies total and metabolically active bacterial commu-nities in the rhizosphere of two perennial grasses grownunder field conditions. FEMS Microb Ecol 55: 339–350.
Khardenavis, A.A., Kapley, A., and Purohit, H.J. (2007)Simultaneous nitrification and denitrification by diverseDiaphorobacter sp. Appl Microbiol Biotechnol 77: 403–409.
Kim, J.K., Park, K.J., Cho, K.S., Nam, S.W., Park, T.J., andBajpai, R. (2005) Aerobic nitrification-denitrification by het-erotrophic Bacillus strains. Bioresour Technol 96: 1897–1906.
Li, B., Irvin, S., and Baker, K. (2007) The variation of nitrifyingbacterial population sizes in a sequencing batch reactor(SBR) treating low, mid, high concentrated synthetic waste-water. J Environ Eng Sci 6: 651–666.
1726 M. Dumont, J. Harmand, A. Rapaport and J.-J. Godon
Lin, Y., Kong, H., He, Y., Liu, B., Inamori, Y., and Yan, L.(2007) Isolation and characterization of a new het-erotrophic nitrifying Bacillus sp strain. Biomed Environ Sci20: 450–455.
Loisel, P., Harmand, J., Zemb, O., Latrille, E., Lobry, C.,Delgenes, J.P., and Godon, J.J. (2006) Denaturing gradi-ent electrophoresis (DGE) and single-strand conformationpolymorphism (SSCP) molecular fingerprintings revisitedby simulation and used as a tool to measure microbialdiversity. Environ Microbiol 8: 720–731.
Marzorati, M., Wittebolle, L., Boon, N., Daffonchio, D., andVerstraete, W. (2008) How to get more out of molecularfingerprints: practical tools for microbial ecology. EnvironMicrobiol 10: 1571–1581.
Muyzer, G., and Smalla, K. (1998) Application of denaturinggradient gel electrophoresis (DGGE) and temperature gra-dient gel electrophoresis (TGGE) in microbial ecology.Antonie Van Leeuwenhoek Int J Gen Mol Microbiol 73:127–141.
Purkhold, U., Pommerening-Roser, A., Juretschko, S.,
Schmid, M.C., Koops, H.P., and Wagner, M. (2000) Phy-logeny of all recognized species of ammonia oxidizersbased on comparative 16S rRNA and amoA sequenceanalysis: implications for molecular diversity surveys. ApplEnviron Microbiol 66: 5368–5382.
Su, J.J., Yech, K.S., and Tseng, P.W. (2006) A strain ofpseudomonas sp. Isolated from piggery wastewater treat-ment systems with heterotrophic nitrification capability inTaiwan. Curr Microbiol 53: 77–81.
Wery, N., Bru-Adam, V., Minervini, C., Delgenes, J.P., Gar-relly, L., and Godon, J.J. (2008) Dynamics of Legionellaspp. & bacterial populations during the proliferation ofL. pneumophila in a cooling tower facility. Appl EnvironMicrobiol 4: 30–37.
Zemb, O., Haegeman, B., Lebaron, P., and Godon, J.J.(2007) SAFUM: Statistical analysis of SSCP fingerprintsusing PCA projections, dendograms and diversity estima-tors. Mol Ecol Notes 7: 767–770.
