Rapid quantification of intracellular PHA using infrared spectroscopy: An application in mixed cultures

Journal of Biotechnology 150 (2010) 372–379

Contents lists available at ScienceDirect

Journal of Biotechnology journal homepage: www.elsevier.com/locate/jbiotec

Rapid quantification of intracellular PHA using infrared spectroscopy: An application in mixed cultures Mónica V. Arcos-Hernandez a,b , Nicholas Gurieff a , Steven Pratt a,b,∗ , Per Magnusson c , Alan Werker c , Alejandro Vargas d , Paul Lant b a

University of Queensland, Advanced Water Management Centre, UQ, St. Lucia, Brisbane, QLD 4072, Australia University of Queensland, School of Chemical Engineering, St. Lucia, QLD 4072, Australia AnoxKaldnes AB, Klosterängsvägen 11A, SE-226 47 Lund, Sweden d Laboratorio de Investigación en Procesos Avanzados de Tratamiento de Aguas; Instituto de Ingeniería, Universidad Nacional Autónoma de México 76230 Querétaro, Mexico b c

a r t i c l e

i n f o

Article history: Received 24 May 2010 Received in revised form 6 September 2010 Accepted 10 September 2010

Keywords: Infrared spectroscopy Polyhydroxyalkanoates Mixed cultures

a b s t r a c t Fourier transform infrared (FT-IR) spectroscopy is proposed for a method for rapid quantification of polyhydroxyalkanoates (PHA) in mixed culture bacterial systems. Spectra from 122 samples from a wide range of PHA production systems were studied. The spectra were collected in a library that was used to calibrate a partial least squares (PLS) model linking FT-IR spectra with PHA content in the biomass. The library of spectra contained samples with a range of total PHA content (0.03–0.58 w/w) as well as varying compositions (poly-(3-hydroxyvalerate) (3HV) content of 0–63% in Cmol basis). A robust PLS model was developed using calibration data from a diverse range of systems and PHA content. Coupling this model with FT-IR spectra has been shown to be applicable for predicting PHA content in mixed culture production systems. The method was used to reliably determine PHA content in biomass from a new, independent PHA production system with a standard error of prediction (RMSEP) value of 0.023 w/w, despite the complexity of the matrices. This method reduces the analytical time for PHA quantification down to under 30 min (5 min handling time was achieved when FT-IR equipment was immediately available), and eliminates hazardous waste by-products. The work has demonstrated a level of accuracy and reproducibility in quantifying PHA in mixed culture systems similar to that obtained from the GC analytical technique. Further work is required to enable the use of the method to analyze crystallinity related factors that may be useful towards quantifying poly-(3-hydroxybutyrate) and poly-(3-hydroxyvalerate) (3HB/3HV) composition. The method has been shown to be suitable for rapid quantification in large scale applications and in its present form is reliable for routine process monitoring. © 2010 Elsevier B.V. All rights reserved.

1. Introduction One of the most promising biomaterials to emerge in recent times is the polyhydroxyalkanoate (PHA) family of polymers as they exhibit comparable physical properties to conventional polymers and are completely biodegradable. These polymers are produced in most bacteria as intracellular carbon and energy storage compounds and are especially prevalent in bacterial communities that have been exposed to transitory growth conditions (Anderson and Dawes, 1990; Bengtsson, 2009; Lee, 1996). The traditional method of PHA production has involved the use of pure culture biotechnology methods including the use of genetically modified organisms

∗ Corresponding author at: University of Queensland, Advanced Water Management Centre, UQ, St. Lucia, Brisbane, QLD 4072, Australia. Tel.: +61 07 3346 7843; fax: +61 07 3346 7843. E-mail address: [email protected] (S. Pratt). 0168-1656/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jbiotec.2010.09.939

with highly refined carbon substrates. In more recent times, it has been realized that mixed cultures, derived from wastewater treatment plants, have the ability to accumulate significant amounts of PHA while using waste products as the carbon source. This has led to a number of studies aimed at understanding and optimizing mixed culture PHA production and the complex microbial community interactions involved (Bengtsson et al., 2010; Lemos et al., 2006; Serafim et al., 2008; Takabatake et al., 2000). The PHA content in the biomass is an important indicator of the economic and environmental sustainability of mixed culture PHA production (Gurieff and Lant, 2007). Real time knowledge of the level of PHA accumulation has the potential to optimize not only the accumulation of PHA, but also the downstream processing of the accumulated biomass. Currently, the most common quantification method used is gas chromatography (GC) combined with the acidic alcoholysis sample preparation method (Braunegg et al., 1978). The original method relied on chloroform extraction, but there has been recent emphasis on avoiding

M.V. Arcos-Hernandez et al. / Journal of Biotechnology 150 (2010) 372–379

the requirement for chlorinated solvents (e.g. Werker et al., 2008) without need of chlorinated solvents. However, this analytical approach requires a number of sample preparation steps and can take up to 24 h. It also produces hazardous waste products including solvents and acids. Although chromatographic methods can simultaneous yield important information about PHA monomer composition and, for example, changes in biomass polysaccharide content (Werker et al., 2008), when PHA content is the sole interest, these and other digestion and chromatographic methods are onerous and do not lend themselves to more proactive feedback in the process control. The lack of a rapid, reliable method for quantifying PHA in mixed culture systems is a major obstacle to achieving on-line control and optimization. Infrared spectroscopy (IR) has been used for the detection and quantification of PHA in intact bacterial cells taken from pure culture systems (Hong et al., 1999; Jarute et al., 2004; Kansiz et al., 2000). The advantages of FT-IR quantification include small sample size (∼0.4 mg of biomass), rapid analysis, no solvents, and minimal sample preparation (Kansiz et al., 2000). While the prediction models used by Kansiz et al. (2000) and Jarute et al. (2004) for 3PHB accumulation in genetically modified Escherichia coli were found to be very accurate, even when using an on-line flow cell (Jarute et al., 2004), systematic evaluation of these techniques for use on more complex mixed microbial cultures accumulating PHA has not been reported in the literature. In the present investigation we acquired a large database of attenuated total reflection (ATR) spectra from activated sludge biomass enriched in Sweden and Australia for purposes of PHA production from wastewater. The data was obtained via experiments of varied duration, ranging from short, one-off batch experiments, to yearlong studies on enriched activated sludge. In these systems it was of interest to routinely and rapidly quantify intracellular PHA that was dominated by P (3-HB-co-3HV) with varying 3HB/3HV compositions in intact cells of the mixed cultures. Since the many components in the complex, dried biomass matrix exhibit overlapping absorbance features in these spectra, successful application of this method required appropriate data conditioning for modeling (Kornmann et al., 2003). Partial least squares (PLS) was used to extract quantitative information of PHA content from the spectra (Schenk et al., 2007) with reference to independent quantification by gas chromatography. The purpose of the present investigation was to determine the utility of Fourier transform infrared (FT-IR) spectroscopy, coupled with an appropriate calibration model, as a robust and rapid tool for quantifying intracellular PHA in mixed cultures. 2. Materials and methods 2.1. FT-IR method The method proposed in this paper is described by means of a flow chart shown in Fig. 1 2.1.1. Biomass samples A variety of mixed culture PHA production systems, with differing operational strategies, and a variety of wastewater feeds, was included in this study. Mixed culture PHA production is conventionally a two-stage process. The first stage is enrichment for populations of bacteria expressing the phenotype of PHA accumulation. In the second stage, biomass harvested from enrichment reactors is fed with readily degradable substrates in order to achieve intracellular PHA content at levels sometimes in excess of 50% w/w (Dias et al., 2006; Gurieff, 2007; Lemos et al., 2006; Beccari et al., 2009). In this work, samples were obtained from the second-stage, accumulation reactors.

373

Fig. 1. Flow chart for using FT-IR to quantify mixed culture PHA. Solid line is flow for data used for calibration models. Dashed line indicates data from reference method for calibration. GC—gas chromatography and PLS—partial least squares.

The database of biomass for ATR spectra used for this study is arranged in 6 different datasets (A–F). The origin of the biomass and the distinctions between operational conditions of the systems defining the 6 datasets are reported in Table 1. In Table 1, respective datasets (A–F) refer to PHA accumulation experiments performed under the same conditions with the same wastewater. For example, for dataset A, biomass samples were obtained from 5 batch accumulation experiments from an Aerobic Dynamic Feeding (ADF) system, producing 17 samples in total with a range of PHA content. The seed sludge was enriched using wastewater from a cannery (60% volatile fatty acids (VFA) content) in a sequencing batch reactor operating with a hydraulic retention time (HRT) of 1 day and sludge retention time (SRT) of 4 days. The organic loading rate was 2 g COD/L/d. Some of the batch accumulations were operated for 8 h with a synthetic feed (70% acetic acid, 30% propionic acid) to produce P (3HB-co-3HV) having high 3HVcontent from 26% to 62% Cmol basis, while others were operated for 6 h with a fermented wastewater to produce P (3HB-co-3HV) having significantly lower 3HV content from 0% to 9% Cmol basis. By including biomass samples from disparate sources and containing a range of 3HV-content it was anticipated that a more universal FT-IR calibration model for PHA quantification could be established.

374

Table 1 Datasets for PLS Regression. Data set

a

17 6 5 42 47 85

Enrichment reactor

Feed wastewater

Duration [h] Sample times [h]

Strategy control

Feed concentration [g COD/L]

Seed sludge

WW

Seed sludge Cycle HRT [d]

SRT [d]

OLR [g COD/L/d]

Sync Syn Synk FDWWl Syn Syn

8 8 8 6 8–32 8–32

DOd DO DO DO DO DO

25 16 16 50–60 25 25

Lab scale ADFe Lab scale ADF Lab scale ADF Lab scale ADF Full–scale WWTPn Full-scale WWTP

Cannery (S1)f Cannery (S2)j Cannery(S1) FDWW – –

BNRg BNR BNR BNR (Lund) – –

4 2 4 4 – –

2 2 2 2 – –

0–8 0–6 0.5–6 0–6 0–32 0–32

ADFh ADF ADF ADF – –

1 1 1 1 – –

Content CDWa

Composition Instrument PHVb

0.07–0.49 0.02–0.44 0.09–0.42 0.03–0.55 0.04–0.30 0.04–0.30

0.26 –0.43 0.30–0.55 0.29–0.62 0.00–0.09 0.34–0.62 0.34–0.62

Nicoleti Nicolet Nicolet Bruker Alfa Nicolet Nicolet

CDW—cell dry weight [mg PHA/mg SSV]. Estimated composition of the obtained PHA, column values represent range of values of content of PHV in the sample, % Cmol. c Synthetic feed: 70% acetic acid, 30% propionic acid. d DO—A 1 L aerated fed batch reactor design for accumulation experiments. The control strategy uses a DO control point to initiate the addition of a consistent dose of carbon source (DO set point: 4.1 [mg/L]). Each batch accumulation ran for a maximum of 8 h. The control system allowed for the consistent addition of the carbon source for as long as the biomass was actively metabolizing it. e Accumulation bioreactors were seeded with excess biomass produced from a lab-scale enrichment reactor operated with aerobic dynamic feeding strategy such as in (d). f S1—enrichment SBR fed with prefermented cannery wastewater (60% VFA content). g BNR—the reactors were seeded with return waste activated sludge from a local advanced biological nutrient removal sewage treatment plant (Luggage Point, Brisbane). h ADF—aerobic dynamic feeding. Cycle: 6 [h] (5 [min] aerobic feed, 335 [min] aerobic react, 15 [min] settle and 5 [min] decant). Operating procedures for S1 and S2 were very similar, with only the carbon source used changing between reactors. The characteristics of the wastewaters and the aims of the experiments were different. The cycles were the same for both reactors, but for S1 mixed liquor was removed once every 4 cycles at the end of the aerated reaction phase while in S2 this was done every second cycle. i Spectrometer with a single bounce diamond attenuated total reflection (ATR) cell Thermo Electron Nicolet 6700 or Bruker Alfa. j S2—enrichment SBR used in carbon source composition variation experiments, fed with highly diluted cannery wastewater with 60% of the organic loading provided by synthetic VFAs. The cannery wastewaters were sampled from a local cannery prefermented on-site, with the samples being taken from the pre-acidification tank effluent prior to entering an UASB. The biodegradable non-VFA SCOD component of the wastewater was composed of sucrose and fructose. k The feed composition for this batch test was slightly different with 75% of acetic acid and 25% of propionic acid. l FDWW—Fermented Dairy WW a wastewater from a local cheese producer, was used as the substrate. The organic content of this dairy industry wastewater is dominated by milk sugar (lactose 85% of dried solids) followed by raw milk protein (3% of dried solids), and amino acids (1.1% of dry solids). Concentrated permeate (70–80 g COD/L) was fermented within a 4L well-mixed chemostat (8 day HRT and 30 ◦ C). pH was controlled at 6 dosing 5 M NaOH. VFA production was such that effluent COD was 90–100% volatile fatty acids dominated by butyric and acetic acids, with some minor amount of propionic acids. m Direct accumulation (DA) of PHA by pulse feed dosing in waste activated sludge from a Sewage Treatment Plant (Brisbane, Australia) without previous enrichment. This data set was used either for calibrations or predictions. The feed used was a synthetic wastewater with 25 [g COD/L] and a composition of 70% of acetic acid and 30% propionic acid in COD basis. n Accumulation bioreactors were seeded with waste activated sludge directed sample from a full-scale carbon removal Sewage Treatment Plant (Brisbane, Australia). o Direct accumulation (DA) of PHA by pulse feed dosing in waste activated sludge from a Sewage Treatment Plant (Brisbane, Australia) without previous enrichment. This data set was used for predictions. The feed used was a synthetic wastewater with 25 [g COD/L] and a composition of 70% of acetic acid and 30% propionic acid in COD basis. b

M.V. Arcos-Hernandez et al. / Journal of Biotechnology 150 (2010) 372–379

A (5 exp) B (1 exp) C (1 exp) D (8 exp) E (8 exp)m F (10 exp)o

# Samples Accumulation reactor

M.V. Arcos-Hernandez et al. / Journal of Biotechnology 150 (2010) 372–379

P (3HB-co-3HV) is the most common type of co-polymer being produced in these mixed-culture processes. Other co-polymers of PHA are possible for mixed-cultures (Pisco et al., 2009) but these were not included for the present investigation. Datasets A–D are biomass harvested from diverse laboratory scale systems and datasets E and F were derived from biomass obtained from a full-scale wastewater treatment process that contained PHA accumulating biomass. Biomass rich in PHA was sourced from accumulation reactors from two different laboratories: The University of Queensland (UQ), Australia, and AnoxKaldnes AB (AKAB), Lund, Sweden. Datasets from A, B, C, E and F are from UQ and data set D is from AKAB. 2.1.2. Sample preparation Sample preparation was performed as outlined in (Kansiz et al., 2000) with slight modifications. During different batch accumulation experiments, samples were taken over a range of times between 0 and 32 h for GC and IR analyses. For the IR analysis, 1 mL of sample was collected and centrifuged at 15000 × g for 3–5 min. Then the supernatant was removed and the cell pellet washed with 90 ␮L of isotonic saline (9 g/L NaCl) and re-centrifuged. The supernatant was once more removed and the final pellet re-suspended in 80 ␮L of isotonic saline. AKAB samples were not washed. After centrifugation a small amount of wet solid biomass was collected using a spatula and smeared out in a thin layer on a microscopic glass. The samples were dried at 105 ◦ C for 5 min to obtain a solid sample for spectra acquisition. 2.1.3 Reference analysis Gas chromatography (GC). For GC analysis, 10 mL samples were prepared according to Braunegg et al. (1978) with modifications proposed by Oehmen et al. (2005). 10 mL samples were taken with 10 drops of 30–40% formaldehyde added to stop all activity. The samples were centrifuged at 5000 rpm for 5 min. The supernatant was removed and the pellet dried in an oven at 110 ◦ C until dry. Next, 2 mL of acidified methanol with benzoic acid as internal standard (3% w/w of H2 SO4 ) and 2 mL of chloroform were added followed by digestion of the sample for 20 h at 100 ◦ C. After cooling to room temperature, 1 mL of MilliQ water was added to allow phase separation. After 1 h settling, the organic phase was transferred to a vial for GC analysis. GC analysis at AKAB was performed according to Werker et al. (2008). Acidic alcoholysis of dried microbial biomass using 3:1 butanol to concentrated (37%) hydrochloric acid at 100 ◦ C for 8 h hydrolyze and derivatize microbial storage products and membrane lipids. Esters of the hydroxyalkanoates, carbohydrates converted to levulinic acid and long chain microbial fatty acids were extracted into hexane for gas chromatographic analysis and quantification. 2.1.4 Spectra acquisition (FT-IR) For each sample, two to five different deposits were tested. The IR spectra were recorded with one of the two FT-IR spectrometers described in Table 1. ATR technique is the technique of choice for spectrum acquisition because of the strong midinfrared absorption of water and ensures a short optical path for measurement without putting restraints on the geometry of the unit hosting the ATR element (Jarute et al., 2004). The scanning conditions were spectra range of 4000 cm−1 and 400 cm−1 , 24–32 scans, a resolution of 4 cm−1 . Three to ten replicate spectra were recorded for each sample to assess precision and ensure representative spectra. The mean of the replicas of each sample was then used for model calibration due to the good reproducibility of the IR spectra (analysis not shown). Atmospheric compensation was also performed.

375

2.1.5 Data preprocessing An automatic baseline correction algorithm was used in all spectra to avoid errors due to baseline shifts. The thickness of the samples was not controlled, and sample thickness varied. Normalization to the sum of the absolute value of all variables included between the bands Amide Band I and Amide Band II was performed using 1-Norm algorithm from the software PLS Toolbox® V.5.2.2 (Eigenvector Research, Inc.). These two bands were chosen because it was assumed that they do not contribute to the PHA concentration. This step produced spectra with consistent absorbance values in this range of wave number for all samples. Also, this has been shown to account for some of the variation due to changes in sample thickness (Kansiz et al., 2000). Fig. 2 shows the data normalized to the selected bands. 2.1.6 Data analysis The data of reference analysis and its correspondent spectra were analyzed by partial least squares (PLS) regression using PLS Toolbox® V.5.2.2 (Eigenvector Research, Inc.) for Matlab® and the Matlab® software. Statistical analysis of spectra included detection of possible outliers. To discard outliers, the Hotelling (T2 ) and Q residuals were calculated to “quantify” the distance in the model space and the residual space respectively (measurement of the variation inside the model and variation outside the model for each sample) (Beebe et al., 1998). Also, graphs of leverage and the studentized residuals helped to indicate the lack of fit of reference values with their respective spectra. 2.1.7 Calibration model After outlier removal, the remaining data was used to generate calibration models. The calibration models related the spectra to the independently assessed content of PHA by sample digestion, extraction and gas chromatography. The number of latent variables or factors for each model to be included in the PLS method was evaluated by means of cross-validation (one-leave-out and random subsets selection). Cross-validation can be used to check the statistical significance of the latent variables or factors. Some objects are kept out and the y-values (i.e. PHA content) for the excluded objects are predicted from the corresponding x-vectors (absorbance matrix). This is repeated a number of times until each object has been kept out once and once only (objects can refer to one sample or subsets of samples) (Tang and Li, 2003). Models are characterized by three evaluation parameters: root-mean-squared error of calibration (RMSEC), the root-mean-squared error of crossvalidation (RMSECV) as well as R2 (correlation coefficient calculated from a regression over the measured data and predicted data plot) to test the goodness of fit. These data were also coupled with an analysis of error for the GC method. The calculation of GC-method error was performed during a batch test, where 5 replicates of each sample were analyzed. For these samples, the GC reported mean PHA dry-cell content of 15.5%, 34.6%, 46%, 48% and 47% PHA respectively with a standard deviation of 2.0% and a standard error of 0.5% PHA. 2.1.8 Validation tests Validation tests were performed by choosing random subsets of samples. A random subset of 10 samples was selected to be left out of the calibration model and then was predicted as described previously for cross-validation. Five tests of this nature were performed for each calibration. 2.1.9 Prediction tests Independent prediction for a set of samples of known PHA content was carried out. In this case the root-mean-square error of prediction (RMSEP) was calculated.

376

M.V. Arcos-Hernandez et al. / Journal of Biotechnology 150 (2010) 372–379

Fig. 2. Representative spectra from a batch experiment (4) from data set A with varying content of PHA, A.U.—arbitrary units. C O band of ester functional groups is a narrow high strong peak centered around ∼1735 (cm−1 ).

3. Results and discussion 3.1. PHA production The biomass samples, taken from batch accumulation experiments, were all characterized by an increase of PHA content over time. Copolymer formation P (3HB-co-3HV) was expected due to the feed composition (see Table 1). Due to varying feed composition, PHA representing a wide range of 3HV content in the polymer was accumulated. Because of the time varying PHA content in each accumulation experiment, the datasets provide for a range of PHA content for each distinct biomass source. The PHA composition obtained in each subset of data is reported in Table 1. 3.2. FT-IR spectra of mixed cultures containing PHA The spectra in Fig. 2 illustrate the trends in a given batch accumulation experiment (Experiment 4 of dataset A) for biomass producing PHA (3HB-co-3HV). The bands associated with PHA and other bio-molecule markers (i.e. cellular protein) can be clearly distinguished. It is observed that each individual spectrum directly relates to the relative concentrations of the specific components of the sample (Jarute et al., 2004; Kansiz et al., 2000), which increase with time during a batch accumulation experiment.

The spectra all reveal peaks that have been previously associated with the presence of PHB in pure cultures (Jarute et al., 2004; Kansiz et al., 2000, 2007) at about 1730 cm−1 , as well as in the range from 1200 cm−1 to 900 cm−1 . For example, Misra et al. (2000) reported key PHA bands at 1724 cm−1 and strong bands at 1280–1300 cm−1 and Hong et al. (1999) reported that a band at between 1728 cm−1 and 1744 cm−1 was characteristic of PHA. The exact peak location of the bands is known to vary with the crystallinity of the PHA and with the polymer chain length (Hong et al., 1999). As well as the PHA bands, distinctive peaks are present at 1637 cm−1 and at 1536 cm−1 . These have been reported as the Amide Band I and the Amide Band II, respectively, and are associated with cellular proteins (Kansiz et al., 2000). The absence of the C O band in samples of biomass known not to contain PHA, confirmed that the IR spectra of mixed cultures could be used to screen the PHA production. 3.3. Quantification of PHA in mixed cultures Datasets A–E in Table 1 are used to produce five calibration models (M1–M5), which relate FT-IR spectra with intracellular PHA content. M1 is a general model that was developed using samples from datasets A–E. M2–M5 were developed using samples from distinct subsets of the data (Table 1 defines the subsets).

Table 2 Summary of results from different calibrations. Model

Calibration datasets

M1

A, B, C, D, E

M2 M3 M4 M5

A, B, C, E A, B, C, D D E

Factors

R2

RMSEC [w/w]

RMSECV [w/w]

Test datasets

RMSEP [w/w]

113

4

0.980

0.023

0.026

0.023 0.023

70 71 38 47

4 4 3 3

0.985 0.911 0.911 0.980

0.019 0.030 0.023 0.011

0.023 0.037 0.027 0.012

Random subsets (10 samples, 3 iterations) F (4) D (43) E (47) A, B, C, E(47) A, B, C, D (66) A, B, C (28)

Samples

0.059 0.042 0.094 0.13 0.07

M1—general calibration using all datasets available (A, B, C, D and E) corresponding to samples of mixed cultures able to store PHA with different contents and composition (%HV). M2—datasets A, B, C and E correspond to samples obtained from two production techniques (ADF and DA), variable PHA contents and %HV content of 26–62% in Cmol basis. M3—datasets A, B, C and D correspond to samples obtained from same production technique (ADF). Data sets D showed PHV compositions of 0–9%. Cmol basis in contrast to a range of 30–62% Cmol Basis of datasets A, B and C. M4—dataset D corresponds to samples obtained from same production technique (ADF) with PHV compositions of 0–9% Cmol basis. M5—dataset E corresponds to samples obtained from same production technique (DA).

M.V. Arcos-Hernandez et al. / Journal of Biotechnology 150 (2010) 372–379

377

Fig. 3. Plot of measured versus predicted PHA content from the unique calibration model (M1). All datasets included—dataset A, dataset B, dataset C, dataset D and dataset E. R2 = 0.980, RMSEC = 0.026 and RMSECV = 0.029. RMSEP calculated for random dataset validation test of 0.023.

Table 2 shows a summary of the calibration models. The correlation (R2 ) between the predicted and measured PHA content, and the standard error of calibration (RMSEC), validation (RMSECV) and prediction (RMSEP) for each model are reported. Because the FTIR method relies directly on measurements obtained from GC, an absolute error cannot be assessed. However, considering the error from GC, which was calculated in this study to be in the order of 2.0% w/w, it may be that a significant fraction of errors associated with the models is due to errors in reference measurements. This can be supported by the fact that when spectra from several deposits of the same sample were obtained, the predicted values for all spectra were practically the same with negligible differences (data not shown). All of the models showed strong correlation (R2 > 0.90) between the predicted and measured PHA content. The standard errors of calibration and validation were lowest for the model developed from samples from dataset E (M5) (0.011 w/w and 0.012 w/w respectively) and highest for the model developed from samples from dataset D (M4) (0.030 w/w and 0.037 w/w respectively). The general model, M1, developed using samples from different sources of biomass (datasets A–E), showed a RMSEC and RMSECV of 0.023 w/w and 0.026 w/w respectively with a very high R2 of 0.98. The model was used to predict PHA content in samples from datasets A-E that were unseen in the calibration. The RMSEP for these samples was as low as 0.023 w/w. The model M1 was also used to predict PHA content in four samples from dataset F, an entirely independent set of biomass samples. The RMSEP for these independent samples was again 0.023 w/w. The models developed using samples from the distinct subsets, or limited combinations of subsets, were less effective in relating spectra to PHA content in independent samples. The RMSEP for these models all exceeded 0.04 w/w, suggesting that development of a general calibration model benefits from the inclusion of multiple, distinct datasets. In terms of PHA content prediction, the RMSEP associated with the general model, M1, of 0.023 w/w is higher than that observed with pure cultures where a RMSEP as little as 0.0149 w/w has been reported (Kansiz et al., 2000). However, considering the complex matrices of the mixed culture samples, as well as the variations between datasets, the RMSEP and RMSECV of 0.026 w/w are low enough to demonstrate the utility of the general calibration model for interpreting FT-IR spectra for the prediction of PHA content in samples from mixed culture PHA production processes.

Fig. 3 is a plot of PHA content predicted using the general model, M1, versus measured PHA content. The predictions of PHA content in samples from the various subsets are shown. The figure highlights that FT-IR spectra are reliable for quantifying PHA, regardless of its composition, in mixed cultures with varying characteristics. 3.4. Characteristics of mixed culture PHA It is suspected that FT-IR spectra can be used to characterize PHA in mixed cultures. The carbonyl stretch (C O) band, located at 1744–1722 cm−1 , provides information on the degree of crystallinity of the polyesters (Jarute et al., 2004). In this work, the peak of this band was observed to vary in position depending on the origin of the sample. For data sets A, B, C, E, and F this band was observed to be relatively broad and centered at 1735–1740 cm−1 . In contrast, dataset D presented a broad band from 1710 to 1760 cm−1 with the maximum crest located at 1725 cm−1 . With reference to other findings (Kansiz et al., 2007; Xu et al., 2002), the band type obtained for samples from dataset D could be related to crystalline phases of PHA. Similar analysis has been also reported for PHB produced in E. coli (Jarute et al., 2004) where the analysis of location of the main peak (∼1740 cm−1 ) in the IR spectra led to the conclusion that PHB in that biomass was amorphous. Additionally, bands at 980 cm−1 and 1230 cm−1 were observed in dataset D, which are also thought to be related to the crystalline phase (Xu et al., 2002). The PLS analysis reveals information on the significance of all of the aforementioned bands for each of the datasets. Fig. 4 shows the loadings for the first latent variable or factor selected for models M2 (dataset D not included) and M4 (only dataset D included). The main band differences at 980, 1230 and 1725 cm−1 were in agreement with that found by Kansiz et al. (2007) and Xu et al. (2002), corresponding to the interpreted differences between the crystalline and amorphous phases of PHA in the biomass. With respect to PHA composition, varying 3HV content is thought not to cause significant variations in the spectra, although it has been observed that the crystallinity has a tendency to decrease with increasing 3HV content (Galego et al., 2000). In this manner, 3HV content of the PHA may be inferred by subtle trends of band shifts due to differences in crystallinity. The relative content of 3HV in dataset D, which apparently showed bands related with crystalline morphological state, was considerably lower than that observed for the other datasets. There may be opportunity

378

M.V. Arcos-Hernandez et al. / Journal of Biotechnology 150 (2010) 372–379

Fig. 4. Loadings plots for factor 1 of model M2 (dashed line) and of model M4 (continuous black). Bands assigned to crystalline phase found in model M4 corresponding to dataset D. Amorphous phase band characterized the loadings for model M4 where dataset D was not included.

to further develop FT-IR methodologies to actually quantify PHA composition. 4. Conclusions This work has demonstrated the utility of FT-IR for prediction of PHA content in mixed culture systems. The method significantly reduces the analytical time for PHA quantification compared to current common practice. This efficiency lends FT-IR towards more direct application in process control and monitoring of mixedculture PHA production processes. It was suggested that 3HV induced sample crystallinity differences influenced the spectra band positions. Quantitative assessment of this variation is part of the ongoing research activity. Notwithstanding, this work shows that a robust FT-IR calibration model for predicting PHA content can be developed. This is evidenced by low residual errors obtained for predictions of PHA content in biomass from different sources, with disparate feeds causing significantly different copolymer content. Acknowledgments M.A. thanks UNAM-DGEP for the financial support during the research visit to The University of Queensland, CONACYT for the graduate studies scholarship in Mexico and Australia, and the Advanced Water Management Centre, The University of Queensland for financial support during this study. The project was also partially funded by CONACYT under grant number J-46097-Y. M.V.A., S.P., N.G. and P.L. thank the Australian Research Council for funding through grant number DP0452860. The authors thank Dr. Stephen Coombs and Dr. Gordon Xu, Australian Institute for Bioengineering & Nanotechnology (AIBN), The University of Queensland for assistance and support with the FT-IR analysis. References Anderson, A.J., Dawes, E.A., 1990. Occurrence, metabolism, metabolic role, and industrial uses of bacterial polyhydroxyalkanoates. Microbiol. Mol. Biol. Rev. 54, 450–472. Beccari, M., Bertin, L., Dionisi, D., Fava, F., Lampis, S., Majone, M., Valentino, F., Vallini, G., Villano, M., 2009. Exploiting olive oil mill effluents as a renewable resource for production of biodegradable polymers through a combined anaerobic-aerobic process. J. Chem. Technol. Biotechnol. 84, 901–908. Beebe, K., Pell, R.J., Seasholtz, M.B., 1998. Chemometrics: a Practical Guide. John Wiley & Sons, New York.

Bengtsson, S., 2009. The utilization of glycogen accumulating organisms for mixed culture production of polyhydroxyalkanoates. Biotechnol. Bioeng. 104, 698–708. Bengtsson, S., Pisco, A.R., Johansson, P., Lemos, P.C., Reis, M.A.M., 2010. Molecular weight and thermal properties of polyhydroxyalkanoates produced from fermented sugar molasses by open mixed cultures. Journal of Biotechnology 147, 172–179. Braunegg, G., Sonnleitner, B., Lafferty, R., 1978. Rapid gas-chromatographic method for determination of poly-beta-hydroxybutyric acid in microbial biomass. Eur. J. Appl. Microbiol. Biotechnol. 6, 29–37. Dias, J.M.L., Lemos, P.C., Serafim, L.S., Oliveira, C., Eiroa, M., Albuquerque, M.G.E., Ramos, A.M., Oliveira, R., Reis, M.A.M., 2006. Recent advances in polyhydroxyalkanoate production by mixed aerobic cultures: from the substrate to the final product. Macromol. Biosci. 6, 885–906. Galego, N., Rozsa, C., Sánchez, R., Fung, J., Analía, V., Santo Tomás, J., 2000. Characterization and application of poly([beta]-hydroxyalkanoates) family as composite biomaterials. Polym. Test. 19, 485–492. Gurieff, N., 2007. Production of Biodegradable Polyhydroxyalkanoate Polymers Using Advanced Biological Wastewater Treatment Process Technology. Advanced Water Management Centre, University of Queensland, Brisbane, p. 145. Gurieff, N., Lant, P., 2007. Comparative life cycle assessment and financial analysis of mixed culture polyhydroxyalkanoate production. Bioresour. Technol. 98, 3393–3403. Hong, K., Sun, S., Tian, W., Chen, G.Q., Huang, W., 1999. A rapid method for detecting bacterial polyhydroxyalkanoates in intact cells by Fourier transform infrared spectroscopy. Appl. Microbiol. Biotechnol. 51, 523–526. Jarute, G., Kainz, A., Schroll, G., Baena, J.R., Lendl, B., 2004. On-line determination of the intracellular poly(␤-hydroxybutyric acid) content in transformed Escherichia coli and glucose during PHB production using stopped-flow attenuated total reflection FT-IR spectrometry. Anal. Chem. 76, 6353–6358. Kansiz, M., Billman-Jacobe, H., McNaughton, D., 2000. Quantitative determination of the biodegradable polymer poly(␤-hydroxybutyrate) in a recombinant Escherichia coli strain by use of mid-infrared spectroscopy and multivariative statistics. Appl. Environ. Microbiol. 66, 3415–3420. Kansiz, M., Domínguez-Vidal, A., McNaughton, D., Lendl, B., 2007. Fourier-transform infrared (FTIR) spectroscopy for monitoring and determining the degree of crystallisation of polyhydroxyalkanoates (PHAs). Anal. Bioanal. Chem. 388, 1207–1213. Kornmann, H., Rhiel, M., Cannizzaro, C., Marison, I., von Stockar, U., 2003. Methodology for real-time, multianalyte monitoring of fermentations using an in-situ mid-infrared sensor. Biotechnol. Bioeng. 82, 702–709. Lee, S.Y., 1996. Bacterial polyhydroxyalkanoates. Biotechnol. Bioeng. 49, 1–14. Lemos, P.C., Serafim, L.S., Reis, M.A.M., 2006. Synthesis of polyhydroxyalkanoates from different short-chain fatty acids by mixed cultures submitted to aerobic dynamic feeding. J. Biotechnol. 122, 226–238. Misra, A.K., Thakur, M.S., Srinivas, P., Karanth, N.G., 2000. Screening of poly-␤hydroxybutyrate-producing microorganisms using Fourier transform infrared spectroscopy. Biotechnology Letters 22, 1217–1219. Oehmen, A., Keller-Lehmann, B., Zeng, R.J., Yuan, Z., Keller, J., 2005. Optimisation of poly-B-hydroxyalkanoate analysis using gas chromatography for enhanced biological phosphorus removal systems. Journal of Chromatography A 1070, 131–136. Pisco, A.R., Bengtsson, S., Werker, A., Reis, M.A.M., Lemos, P.C., 2009. Enrichment of glycogen-accumulating organisms producing polyhydroxyalkanoates (PHA)

M.V. Arcos-Hernandez et al. / Journal of Biotechnology 150 (2010) 372–379 from fermented molasses: evolution in community structure and PHA production. Appl. Environ. Microbiol., 02408-02486. Schenk, J., Marison, I.W., von Stockar, U., 2007. Simplified Fourier-transform midinfrared spectroscopy calibration based on a spectra library for the on-line monitoring of bioprocesses. Anal. Chim. Acta 591, 132–140. Serafim, L.S., Lemos, P.C., Torres, C., Reis, M.A.M., Ramos, A.M., 2008. The influence of process parameters on the characteristics of polyhydroxyalkanoates produced by mixed cultures. Macromol. Biosci. 8, 355–366. Takabatake, H., Satoh, H., Mino, T., Matsuo, T., 2000. Recovery of biodegradable plastics from activated sludge process. Water Sci. Technol., 351–356.

379

Tang, K., Li, T., 2003. Comparison of different partial least-squares methods in quantitative structure–activity relationships. Anal. Chim. Acta 476, 85–92. Werker, A., Lind, P., Bengtsson, S., Nordström, F., 2008. Chlorinated-solvent-free gas chromatographic analysis of biomass containing polyhydroxyalkanoates. Water Res. 42, 2517–2526. Xu, J., Guo, B.-H., Yang, R., Wu, Q., Chen, G.-Q., Zhang, Z.-M., 2002. In situ FTIR study on melting and crystallization of polyhydroxyalkanoates. Polymer 43, 6893–6899.