SEARCH

SEARCH BY CITATION

Keywords:

  • recirculating aquaculture systems;
  • Ion Torrent sequencing;
  • biofilter;
  • off-flavours;
  • nitrification

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

A variety of factors affecting water quality in recirculating aquaculture systems (RAS) are associated with the occurrence of off-flavours. In this study, we report the impact of water quality on the bacterial diversity and the occurrence of the geosmin-synthesis gene (geoA) in two RAS units operated for 252 days. Unit 2 displayed a higher level of turbidity and phosphate, which affected the fresh water quality compared with unit 1. In the biofilter, nitrification is one of the major processes by which high water quality is maintained. The bacterial population observed in the unit 1 biofilter was more stable throughout the experiment, with a higher level of nitrifying bacteria compared with the unit 2 biofilter. Geosmin appeared in fish flesh after 84 days in unit 2, whereas it appeared in unit 1 after 168 days, but at a much lower level. The geoA gene was detected in both units, 28 days prior to the detection of geosmin in fish flesh. In addition, we detected sequences associated with Sorangium and Nannocystis (Myxococcales): members of these genera are known to produce geosmin. These sequences were observed at an earlier time in unit 2 and at a higher level than in unit 1. This study confirms the advantages of new molecular methods to understand the occurrence of geosmin production in RAS.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Recirculating aquaculture systems (RAS) are complex, multicompartmental systems used in fish farming. RAS are typically indoor systems that allow farmers to control environmental conditions year-round; they represent an ideal alternative to open fish culture systems (van Rijn, 1996). This system may assist in minimizing environmental problems by reducing water demand and managing effluent discharge (Gutierrez-Wing & Malone, 2006). One problem associated with RAS is the accumulation of off-flavour compounds; the presence of geosmin in trout fillets produces a musty/earthy flavour and odour in the final product and can negatively impact the quality of aquaculture products (Hanson, 2003; Klausen et al., 2005; Smith et al., 2008). Although geosmin is not toxic, this compound is characterized by an exceptionally low detection threshold by human senses and thereby causes the fish to be perceived as unsafe by consumers (Cook et al., 2001; Schrader & Rimando, 2003). Some microorganisms such as Streptomyces, Myxobacteria and Cyanobacteria are responsible for the synthesis of this compound in streams and aquaculture systems (Dickschat et al., 2005; Robin et al., 2006; Guttman & van Rijn, 2008; Schrader & Summerfelt, 2010). Streptomyces and Myxobacteria form spores that both enable the organisms to survive adverse conditions (Weyland, 1969; Vos & Velicer, 2008) and contribute to their widespread distribution by wind and water (Lloyd, 1969; Goodfellow & Williams, 1983). Different factors such as the concentrations of phosphate, micronutrients or organic matter all affecting the water quality are associated with the occurrence of off-flavours and the growth of off-flavour-producing microorganisms within the RAS operation. (Schrader & Blevins, 2001; Robertson et al., 2006; Guttman & van Rijn, 2008).

RAS often include compartments such as (1) a microbial biofilter for nitrifying activity in the conversion of ammonium (inline image) to nitrite (inline image) and then to nitrate (inline image), a less toxic compound, for mineralization of organic matter by heterotrophic bacteria, and sometimes for denitrification (Barak et al., 2003); (2) a physical filtration component (sand filter) to concentrate particulate organic matter derived from uneaten food and excreta of fish, thereby maintaining high-quality water; (3) and multiple fish tanks containing microbial communities originating from the fish being cultured. In different water treatment systems, nitrification failure is a frequent occurrence, because nitrifiers are inhibited by several environmental factors, including low temperature, extreme pH, low dissolved oxygen concentration and a wide variety of chemical inhibitors (Prosser, 1989). Accumulation of different nitrogen sources in RAS can be toxic for fish (Colt & Armstrong, 1979; Russo & Thurston, 1991) and affect nitrifying bacterial populations potentially reducing the nitrification efficiency (Prosser, 1989). Controlling the nitrification process by a better understanding of nitrifying populations could increase production volume, reduce discharged water and enhance profitability for system managers.

Changes in the bacterial community observed in the biofilter media or in digestion/sedimentation tanks in different RAS have been analysed previously using molecular approaches, such as polymerase chain reaction-denaturing gradient gel electrophoresis (PCR-DGGE) and clone library analyses that target the 16S rRNA gene (Cytryn et al., 2003; Tal et al., 2003; Sugita et al., 2005; Matos et al., 2011). These molecular techniques allow for the analysis of major bacterial populations, sometimes only corresponding to only one percentage of the total bacterial community (Fromin et al., 2002). New high-throughput sequencing methods now allow for the examination of a larger proportion of the bacterial community. Massive parallel sequencing techniques of short regions of the 16S rRNA gene now allows for the examination of rare bacterial populations in aquatic samples (Campbell et al., 2011), such as nitrifying bacteria or geosmin-producing microorganisms.

To improve the development of environmentally sustainable farming using RAS, it is imperative: (1) to determine the structure of nitrifying populations in microbial biofilters, and factors affecting these populations and/or nitrification efficiency, (2) to identify, localize and quantify off-flavour producers in all compartments of two RAS and (3) to determine which environmental criteria can explain the development of off-flavour events. We hypothesized that we can identify the off-flavour producers using next-generation sequencing technology. As reported in many cases, Streptomyces species could explain the occurrence of off-flavours that represent one of the major problems in RAS production. Development of these species can be mediated by the degradation of the water quality with the increase in organic particles and different nutrient concentrations. We also hypothesized that the community composition of the nitrifying bacterial populations in biofilter compartments is influenced by similar criteria without any impact on the nitrification efficiency, due to functional redundancy.

This study was aimed to describe the spatial and temporal variability of bacterial diversity among the compartments of two RAS units used to farm rainbow trout that were fed with two different diets. We used Ion Torrent sequencing of the V5 region of the 16S rRNA genes to focus our analysis on nitrifying bacterial populations in the biofilter and on geosmin-producing bacterial populations observed in the fish tanks, biofilters and sand filters from samples taken every 28 days for 252 consecutive days. We also determined the level of the geosmin-synthesis gene (geoA) by real-time PCR (qPCR) in these compartments during the operation of two RAS units along with the level on geosmin in fish fillets.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Experimental system

The indoor RAS units located in the Laboratoire Régional des Sciences Aquatiques (LARSA, Université Laval, QC, Canada) have been previously described by Auffret et al. (2011). Briefly, two identical systems placed in two different rooms without any contact or possible cross contamination were composed of four fish tanks per unit (water volume: 1.5 m3 per system). They were operated in a recirculation mode for rainbow trout culture (Oncorhynchus mykiss). Prior to placing the fish into the RAS units, fresh water was filtered (45 μm) and UV treatment was applied to eliminate pathogenic strains. The oxygen level (90–100% saturation), pH (7.4 ± 0.4) and water temperature (15.2 ± 0.4 °C) were monitored every day for 252 days to ensure that these variables remained constant. The biofilters (trickling filter) were composed of expanded polystyrene, and the biofilm media were obtained directly from the manufacturer (Aquabiotech, Coaticook, QC, Canada) and were inoculated with a nitrification/denitrification commercial bacterial consortium (Bacta-Pur® N3000; IET-Aquarecherche Ltée, North Hatley QC, Canada) 3 months before the introduction of fish. After this period, 100 rainbow trout (O. mykiss) per tank were introduced into the RAS, and fish biomass in RAS units was maintained at 25 kg m−3 by harvesting excess fish every 28 days. The final number of fish decreased during the experiment from 70 to 11 fish per tank in unit 1 and to 16 fish per tank in unit 2. The absence of off-flavour compounds and off-flavour producers prior to the introduction of fish was confirmed in fish tissue by solid-phase micro-extraction gas chromatography–mass spectrometry (SPME–GC–MS) and in RAS compartments by monitoring the geosmin synthesis gene (geoA) by real-time PCR. Fish were fed with the ‘Skretting Orient’ diet (Skretting, St Andrews, NB, Canada) in unit 1 and the ‘Martin Mills Classic’ diet (Martin Mills Inc., Elmira, ON, Canada) in unit 2 having a similar composition (data not shown). Fish were fed 5 days per week (every day of the week except Friday and Saturday) with a similar amount of fish food in both units. System sand filters were backwashed five times per week to remove solids accumulated in the sand filter. The backwashes were scheduled on the days that followed fish feeding.

Physicochemical parameter monitoring

Turbidity (triplicate) and total organic carbon (TOC) were measured directly from the effluent of the fish culture tanks every 7 and 14 days, respectively. A volume of 30 mL per sample was filtered (0.45 μm) and then analysed for TOC content using a total organic C analyser (Shimatzu, model TOC-V) at 680 °C. Organic carbon was calculated as the difference between total carbon and total inorganic carbon. The turbidity was measured using a HACH 2100N laboratory turbidimeter, and the turbidity was reported in nephelometric turbidity units (NTU).

Determination of the concentration of inorganic anions and cations was performed on duplicate samples that were derived from the effluent of the fish culture tanks. Water samples (30 mL) were filtered through a 0.45-μm filter, and the filtrates were analysed (50 μL volume injection) using ion chromatography (ICS-3000; Dionex Corporation, Sunnyvale, CA). Cations (ammonium) were analysed with an IonPac AS17 column (4 × 250 mm) with a KOH eluent gradient (9–30 mM, for 16 min) with 1 mL min−1 flow rate at 30 °C. Anions (nitrate, nitrite, phosphate) were analysed with an IonPac CS16 column (5 × 250 mm) with methanesulphonic acid (42 mM) as eluent with 1 mL min−1 flow rate for 16 min at 30 °C.

Fillet preparation and quantification of off-flavour compounds in fish flesh

During the first period (day 0 to day 84), three fish from different tanks per unit were collected. From day 112 to 252, twelve fish from each unit were collected every 28 days. These fish were euthanized and immediately filleted, and the skin was removed. Slices of rainbow trout flesh (no skin) were cut from dorsal to ventral direction across the fillet to yield a 20 ± 0.1 g portion of flesh. Samples were cut starting from the anterior end such that any unused fillet always remained at the tail end of the fish. Two fillets per fish were obtained and placed in separate plastic bags (Ziploc™; SC Johnson Canada, Brantford, ON, Canada). Fillets were then frozen until SPME–GC–MS analysis. The method of Lloyd & Grimm (1999) was used to analyse geosmin and 2-methylisoborneol (MIB) from the fillets and was modified as follows. Each fillet was placed into a glass distillation flask. The flask was then heated in a microwave oven (Daewoo, Seoul, Korea; model TMW-1100EC) for 4 min 45 s at power level ‘4’ while purging with 80 mL L−1 of N2 gas. The collected distillate was cooled in a polystyrene box filled with ice, and the volume was adjusted to 25 mL using nanopure water. Each 25 mL sample was placed into a 40-mL glass vial containing 6 g of NaCl. Each vial was sealed with a crimp cap. The vials were stored at 4 °C until the SPME–GC–MS analysis. The protocol for detecting the geosmin and MIB using GC–MS was previously described by Auffret et al. (2011). The limit of quantification was 1 ng kg−1 in fish flesh, whereas the limit of detection by human senses is approximately 600 ng kg−1 (Cook et al., 2001; Schrader & Rimando, 2003).

Sample collection for microbiological analysis

Water samples from the fish tanks (250 mL) and the sand filters (duplicate of 50 mL) as well as the biomass from the biofilters (biofilm from the media) were collected every 28 days for 252 days and stored in 50-mL Falcon tubes at −20 °C before DNA extraction. Prior to the storage at −20 °C, water samples from the fish tanks were filtered (ca. 250 mL per tank, with 0.22-μm filters). At days 196, 224 and 252, the attached biomass (biofilm) collected from the sand filter was added to the water sample tubes (duplicate of 50 mL) for DNA extraction.

DNA extraction and PCR amplification

The filters from the fish tank water samples stored at −20 °C were cut into small pieces and distributed into two 2-mL tubes containing 250-mg glass beads (0.25–0.50 mm) and 500 μL TEN [50 mM Tris–HCl pH 8.0, 100 mM ethylenediaminetetraacetic acid (EDTA) pH 8.0 and 150 mM NaCl]. The biofilter media were washed with 5 mL of phosphate buffer (20 mM, pH 7.2) and vortexed twice at maximum speed for 10 s to detach the biomass. The suspension was then centrifuged at 16 060 g for 15 min. The sand filter samples were split into different Falcon tubes and centrifuged at 16 060 g for 15 min. A large majority of the supernatant was removed and the residual water samples served to harvest a maximum of biomass inside each tube but also to disperse the cell pellets from the biofilter and the sand filter. These water samples were then transferred in 2-mL tubes containing 500 μL TEN and 250-mg glass beads (0.25–0.50 mm). The cells were disrupted twice for 20 s at speed 5.0 using a FastPrep homogenizer (MP Biomedicals Qbiogene, Solon, OH) and placed on ice. The homogenate was centrifuged for 15 min at 16 060 g. The supernatant was then extracted once using phenol–chloroform–isoamyl alcohol (25 : 24 : 1) and once using chloroform–isoamyl alcohol (49 : 1). DNA was precipitated overnight at −20 °C by adding ammonium acetate (0.1 M final concentration), 5 μL of a diluted solution (1 : 50) of glycogen at 5 mg mL−1 (Ambion, Austin, TX) and two volumes of ethanol. After centrifugation at 16 060 g for 15 min, the DNA pellet was washed once with 70% ethanol and then with 95% ethanol and was dissolved in 50 μL TE (10 mM Tris–HCl, 0.1 mM EDTA, pH 8).

The PCR conditions to detect geoA- and MIB-synthesis genes were the same as previously described by Auffret et al. (2011).

The quantification of total DNA was performed in triplicate with the Quant-iT™ PicoGreen® kit (Invitrogen, Burlington, ON, Canada) on 1 μL DNA samples according to the manufacturer's specifications. qPCR was performed according to Auffret et al. (2011) in triplicate using a Rotor Gene 6000 instrument (Corbett Research, Mortlake, NSW, Australia). Each qPCR was performed in 20 μL with the Perfecta® SYBR® Green Fast Mix (QuantaBiosciences, Gaithersburg, MD), 250 nM of each primer (AMgeoF/R) (Auffret et al., 2011) and 10 ng of total DNA previously quantified by Quant-iT™ PicoGreen® kit. The amplification conditions were as follows: preheating at 95 °C for 10 min followed by 40 cycles of 95 °C for 15 s, 66 °C for 60 s and 72 °C for 15 s. Melting curves were performed to confirm the purity of the amplified product.

To perform Ion Torrent sequencing (Rothberg et al., 2011), DNA samples were PCR amplified using the adapter–multiplex identifier primer combinations described in the Supporting Information (Table S1; Baker et al., 2003). Each primer combination reaction was performed on two different DNA extracts. The sequence-specific 16S rRNA gene primers used were E786F and U926R, which amplify the position 786–926 (V5) of the 16S rRNA gene of Escherichia coli. The PCR conditions consisted of one denaturation cycle (5 min at 95 °C); 30 cycles of denaturation (30 s at 95 °C), annealing (30 s at 55 °C) and elongation (45 s at 72 °C); and one elongation cycle (10 min at 72 °C). PCRs were carried out in 25 μL volumes containing 1 μL of template DNA, 0.625 μL of each 16S rRNA gene primer (20 pmol μL−1), 0.5 μL 10 mM deoxynucleoside triphosphates, 2.5 mM MgCl2 and 0.625 U of Taq DNA polymerase in 2.5 μL of the 10 ×  Taq DNA polymerase buffer provided (GE Healthcare, Baie d'Urfé, Canada). The PCR amplicons were gel purified using a QIAquick Gel Extraction kit (Qiagen, Mississauga, ON, Canada), quantified by PicoGreen, diluted and pooled at equimolar concentrations. Pooled amplicons were sequenced on two 314 chips using the Ion Torrent PGM™ sequencer according to the manufacturer's instructions (Life Technologies, Carlsbad, CA). We analysed the DNA sequences using a local version of the RDP pyrosequencing pipeline (http://pyro.cme.msu.edu/). The set-up of this custom-made Perl script pipeline kept all sequences with a minimal sequence length of 75 bp and a maximal length of homopolymers at 5. First, sff files were transformed into fasta and qual files, and we removed all sequences that contained undetermined bases (N) had an average expected quality score lower than 17 or were shorter than 75 bp. The sequences were then aligned and binned according to their 5-bp multiplex identifier (only accepting perfect matches), and the multiplex identifiers were removed. We identified 55 distinct data sets, each data set corresponding to one sample. The sequences of each data set were individually classified using the local RDP Multi-Classifier tool with a 50% bootstrap cut-off recommended for short sequences (Claesson et al., 2009). We obtained reads of 100–200 bases that reduce the likelihood of generating chimeras and increases the likelihood of detecting low-abundance taxa (Huber et al., 2009). The weight normalized Unifrac distances between each sample pair were calculated using the FastUnifrac website based on the GreenGenes core data set (Hamady et al., 2010).

Statistical analysis

All statistical analyses were performed using the R package (R Foundation for Statistical Computing, Vienna, Austria). The principal coordinate analyses were carried out using the ‘cmdscale’ function of the ‘vegan’ package based on Unifrac (16S rRNA gene) distances. For Spearman correlation, the ‘cor.test’ function was employed.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Monitoring of water quality

We initiated this experiment with two identical sterilized RAS units supplemented with new biofilm attachment media inside the biofilter system and inoculated with a commercial bacterial consortium for nitrification, denitrification and mineralization processes. Because the occurrence of off-flavours in RAS is not a predictable event, we used two different commercial diets with similar compositions in the two units that were operated in parallel. The diet used in unit 2 generated more friable faeces. In both units, the turbidity fluctuated between 0.3 and 1.5 (NTU) for the first 200 days and increased afterwards (Fig. 1a). It was overall higher in unit 2 than unit 1. The TOC concentrations were similar in both units (Fig. 1a). A fourfold increase was observed after 14 days and then the concentrations fluctuated between 8–18 mg L−1. The phosphate concentration in unit 1 fluctuated between 0.31 and 0.55 mg L−1 for the first 140 days (Fig. 1b). In unit 2, phosphate concentrations were overall higher than unit 1, reaching 0.85 mg L−1 at day 42 and over 1 mg L−1 after 152 days. In both units, phosphate increased steadily afterwards to reach over 3 mg L−1 by the end of the experiment (Fig. 1b). The concentrations of other compounds and several micronutrients were similar in both units throughout the experiment (data not shown).

image

Figure 1. Evolution of the turbidity, the TOC and the phosphate concentrations in the two units. Water samples were taken from the fish tanks. (a) □ (Unit 1) and ♦ (Unit 2), turbidity (NTU). Δ (Unit 1) and × (Unit 2), TOC. (b) Phosphate concentration. Δ, Unit 1 and ×, Unit 2.

Download figure to PowerPoint

In the biofilter, nitrification is one of the major processes by which high water quality is maintained. Prior to the introduction of fish, the nitrification processes needed to be evaluated. High concentrations of both nitrate and nitrite were detected in both units (Fig. 2a and b), thereby reflecting the transformation of ammonium by nitrifying bacteria. After the fish were introduced, the ammonium concentrations increased for 17 days (2.1 mg L−1 for unit 1 vs. 1.1 mg L−1 for unit 2) and then decreased to undetectable levels after 56 days (Fig. 2a). Nitrate concentrations, which were similar in both units, decreased for 42 days and fluctuated between 50 and 150 mg L−1 thereafter (Fig. 2b). As observed with the nitrate measurements, the nitrite concentrations decreased during the first 40 days and remained below the toxic level (1 mg L−1) throughout the experiments (Fig. 2a).

image

Figure 2. Evolution of ammonium, nitrite and nitrate concentrations in both units. Water samples were taken from the fish tanks. (a) Ammonium: □ (Unit 1) and ♦ (Unit 2). Nitrite: Δ (Unit 1) and × (Unit 2). (b) Nitrate: □ (Unit 1) and ♦ (Unit 2).

Download figure to PowerPoint

Detection of geosmin and MIB compounds by GC–MS

The fish showed the presence of geosmin at 549 ng kg−1 in the flesh before their introduction into the RAS. In unit 1, the concentration of geosmin decreased to 147 ng kg−1, reflecting the quality of water that helped to flush geosmin from the fish (Fig. 3). However, after day 140, the geosmin concentration increased to reach 803 ng kg−1 by 224 days. The situation was contrasted in unit 2, in which the geosmin concentration increased significantly between days 56 and 84 (3162 ng kg−1) and reached a maximum at 196 days (3845 ng kg−1), which corresponded to nearly fivefold the maximum geosmin concentration found in unit 1. After this period, the presence of geosmin in the fish flesh decreased (2181 ng kg−1) after 252 days. We also noticed a decrease in geosmin at day 252 (540 ng kg−1) in unit 1. The MIB was never detected in the fish flesh in either unit.

image

Figure 3. Quantification of geosmin in the fish flesh and the occurrence of geoA. Geosmin concentrations were measured using SPME–GC–MS. The mean values obtained from the quantification analysis of three fish flesh from day 0 to day 84 and twelve fish from day 112 to the end of the study are reported. ■, Unit 1; ♦, Unit 2. * and ¤, Positive PCR signal for geoA in at least one of the water samples (see Table 1).

Download figure to PowerPoint

Detection of MIB- and geosmin-synthesis genes

Water samples from the fish tanks and from the sand filter, and the biomass from the biofilter were collected at the same time that the fish were taken for the determination of geosmin concentration in the fish flesh. Total DNA was extracted from these samples and the levels of geoA were determined using qPCR (Table 1 and Fig. 3). In unit 2, a low level of geoA was detected in the biofilter and the sand filter at day 56, just prior to the increase in geosmin concentration in the fish flesh. At day 140, the level of geoA was approximately 165 times higher than at days 56 and 84 in the biofilter. In the sand filter, a 60-fold increase was observed at day 196 compared with day 56. At days 140 and 168, geoA was detected for the first time in the fish tank at a level 5 times higher than in the biofilter. Towards the end of the operation, the level of geoA decreased in all three locations at levels ranging from 114 to 1910 copies per 10 ng of total DNA. This decrease paralleled the decrease in geosmin that was observed in the fish flesh.

Table 1. Quantification of geoA by real-time PCR of the RAS samples
geoA (SD) copy number per 10 ng of genomic DNA
Sampling time (days)Fish tankBiofilteraSand filterb
Unit 1Unit 2Unit 1Unit 2Unit 1Unit 2
  1. N.D., not detected; N.M, not measured.

  2. a

    The biomass was detached from the biofilm attachment carriers that were present in the biofilter.

  3. b

    The biomass was collected after centrifugation of the water sample.

  4. c

    A backwash process was performed to remove the excess biomass (biofilm) that was present in sand filter. This biomass was used for the qPCR assays.

0N.D.N.D.N.D.N.D.N.D.N.D.
28N.D.N.D.N.D.N.D.N.D.N.D.
56N.D.N.D.N.D.721 ± 91N.D.1757 ± 205
84N.D.N.D.N.D.3137 ± 242N.D.N.D.
112N.D.N.D.N.D.N.D.N.D.N.D.
140N.D.53 200 ± 1572791 ± 3511 947 ± 2252N.D.N.D.
1685530 ± 28963 500 ± 21 721N.D.N.D.N.D.N.D.
196N.M.N.M.N.M.N.M.N.D.10 633 ± 208c
2242100 ± 191N.D.1210 ± 30933 ± 411910 ± 1361500 ± 240
252N.D.1850 ± 171N.D.1217 ± 586291 ± 31114 ± 38c

In unit 1, geoA was detected at a low level in the fish tank only at day 224 (Table 2). In the sand filter, geoA was only detected at the end of the operation. However, in the biofilter, geoA was detected at day 140, but its level was 15 times lower than in unit 2 at the same time. This detection of geoA was just prior the slight increase in geosmin in the fish flesh of unit 1. At day 168, the level of geoA increased eightfold, which paralleled the increase in geosmin in unit 1. As observed with unit 2, geoA decreased in unit 1 towards the end of the operation such that it was not detected in the fish tank and the biofilter and remained at 291 copies per 10 ng of total DNA in the sand filter.

Table 2. Spearman correlation between the major bacterial populations detected in the biofilters and physicochemical factors
 Spearman factorPhysicochemical factor
r a P b
  1. Only significant (P < 0.05) correlations are shown in bold.

  2. a

    r is the correlation factor with a value between −1 and +1.

  3. b

    P is a second correlation factor that was either equivalent or below 0.05 for significant analysis.

  4. c

    Bacterial genus with a proportion higher than 1% in at least one sample.

Unit 1 (genus)c
Acidovorax 0.8670.004Nitrate
Aquabacterium −0.7160.036Nitrate
Flavobacterium 0.9790.000004Nitrate
Prolixibacter −0.8580.003Nitrate
Pseudomonas 0.7300.025Nitrate
Nitrosospira −0.730 0.039 Nitrite
Polynucleobacter −0.7610.028Nitrite
0.9270.0009Phosphate
Mycetocola −0.7900.019Nitrite
0.9700.00006Phosphate
Nitrospira −0.874 0.004 Nitrite
  0.738 0.045 Phosphate
Nitrobacter 0.837 0.009 Phosphate
Sorangium 0.730 0.039 Phosphate
Dyadobacter 0.7820.012Turbidity
Niabella 0.7740.014TOC
Devosia 0.8090.008TOC
Rhodovulum 0.7160.036TOC
Legionella 0.8830.003TOC
Unit 2
Polynucleobacter −0.7280.026Nitrate
Hyphomicrobium −0.8030.009Nitrate
Flavobacterium 0.8530.003Nitrate
Nitrospira 0.730 0.025 Phosphate
Mycetecola 0.7020.034Turbidity
Nitrosospira 0.803 0.009 TOC
Novosphingobium 0.7310.025TOC
Adhaeribacter 0.6780.044TOC
Rhodovulum 0.7310.025TOC
Niabella 0.7190.028TOC

The MIB-synthesis gene tpc was not detected in the units at any time during the study.

Community structure of nitrifying bacterial populations

Ion Torrent sequencing targeting the V5 region of the 16S rRNA genes yielded 46 974 classifiable 16S rRNA gene sequences (run 1) from unit 1 and 16 528 sequences (run 2) from unit 2. We explained the variability in the number of 16S rRNA gene sequences by different efficiencies between the two runs. This result corresponds to an average of 1154 16S rRNA gene sequences per sample with an average read length of 137 bases. Unifrac analysis was performed to determine the similarity of the bacterial community structure between the three compartments and verify the stability of the bacterial populations over 252 days (Fig. S1). However, we focused our analysis on the biofilter compartment that represents the core of RAS and performs the nitrification process. The bacterial community observed in the biofilter of unit 1 was similar and stable throughout the experiment as illustrated by a tight clustering in one quadrant. More variations in the bacterial community were observed in unit 2 (Fig. 4). Complete analysis of the bacterial composition of all samples in the three compartments and the complete Unifrac analysis are detailed in the Supporting Information (Fig. S1; Table S2–S4).

image

Figure 4. Unifrac analysis of the total bacterial community found in biofilter (BF) of both units. □, Biofilter Unit 1; ♦, Biofilter Unit 2. Circle: T = 0 or BF1: Biofilter Unit 1.

Download figure to PowerPoint

We performed an extended study of the biofilter to analyse the structure of nitrifying bacterial populations throughout the experiment and determine which environmental factors could affect the presence of these populations. The occurrence of nitrifying bacteria showed that Nitrosomonas (Nitrosomonadaceae) was the primary ammonium oxidizer at days 28 and 56 (3.6% and 4.0%, respectively) in the biofilter of unit 1 (Fig. 5a). Their proportions decreased below 1% after these time points and reached 0.03% by the end of the experiment. At day 112, the abundance of Nitrosopira (Nitrosomonadaceae) was equivalent to that of Nitrosomonas, which reached a maximum at day 140 (1.4%) and subsequently decreased to 0.3%. Nitrospira increased in proportion and represented the dominant nitrite-oxidizing bacteria by the end of the operating time (up to 6.6%).

image

Figure 5. Evolution of nitrifying bacterial populations (percentage of relative abundance) identified in both units. (a) Unit1; (b) Unit2. White column: Nitrosomonas; bright grey column: Nitrosospira; grey column: Nitrospira; black column: Nitrobacter.

Download figure to PowerPoint

In unit 2, the overall nitrifying population (including Nitrobacter) proportion was significantly lower than that observed in unit 1 (0.5% vs. 3.7% of the overall sequences in the respective biofilter). Nitrosospira was the most abundant ammonium oxidizer early during the experiments (1.7% at day 56) (Fig. 5b). No known nitrifiers were detected at days 112, 140 and 168. At the end of the experiment, Nitrospira was the most abundant nitrite oxidizer, which reached a proportion of 1% at day 252. The higher proportion of Nitrospira was observed in the unit 2 fish tank at the end of the operating time. Nitrobacter was poorly detected at day 28 (0.3%) and day 84 (0.1%) in units 1 and 2, respectively.

A Spearman correlation analysis was performed to determine which RAS conditions had an impact on the nitrifying bacterial populations (Table 2). In both units, Nitrosomonas proportions did not display a correlation with any of the environmental factors tested. A negative correlation was observed between nitrite levels and the ammonia oxidizer, Nitrosospira, in unit 1, while the presence of this genus displayed a positive correlation with TOC content in unit 2. Nitrospira proportions displayed a positive correlation with phosphate levels in both units and a negative correlation with nitrite levels in unit 1.

Detection of putative geosmin producers

Expected sequences previously reported in other aquatic systems and associated with other geosmin producers, such as actinomycetes or cyanobacteria, were not detected by high-throughput sequencing. However, sequences associated with Sorangium and Nannocystis (Myxococcales) were identified in both units (Table 3). Some Sorangium and Nannocystis species were reported to be geosmin producers (see Discussion). These Myxococcales were encountered in more samples in unit 2, and in general, in higher proportions than in unit 1. In unit 1, Sorangium was first detected on day 140 in the biofilter and later in the fish tank and sand filter samples. Nannocystis was also detected at day 224 in the sand filter sample, although the relative bacteria abundance was 16-fold lower than the value obtained for Sorangium. In unit 2, Sorangium was first detected on day 56 in the biofilter and sand filter samples and subsequently throughout the experiment in the three compartments. Nannocystis DNA sequences were first identified in the biofilter sample at day 84 and at later time points in the two other compartments. The proportions of Sorangium and Nannocystis peaked (1.3% and 1.7%, respectively) in the fish tank at day 140 in unit 2, and then decreased afterwards.

Table 3. Detection of putative geosmin producers by Ion Torrent sequencing in RAS units
 Sampling time (days) geoA a GeosminbPutative geosmin producers (genus)Relative abundance (%)c
  1. N.De., not detected.

  2. a

    geoA: Detected by qPCR.

  3. b

    The geosmin concentration in fish tissue. Low: < 1000 ng kg−1 fish flesh. High: more than 2000 ng kg−1 fish flesh.

  4. c

    Proportion of sequences associated with either Sorangium or Nannocystis in the corresponding sample.

Unit 1
Biofilter140+Low Sorangium 0.31
Biofilter224+Low Sorangium 0.96
Nannocystis 0.06
Fish tank168+Low Sorangium 0.04
Fish tank224+Low Sorangium 0.08
Sand filter224+Low Sorangium 0.17
Sand filter252+LowN.De.N.De.
Unit 2
Biofilter56+Low Sorangium 0.85
Sand filter56+Low Sorangium 0.30
Biofilter84+High Sorangium 0.91
Nannocystis 0.32
Biofilter112-High Sorangium 0.71
Nannocystis 0.71
Fish tank140+High Sorangium 1.3
Nannocystis 1.7
Biofilter140+HighN.De.N.De.
Fish tank168+High Sorangium 0.16
Nannocystis 0.99
Sand filter196+High Nannocystis 1.5
Biofilter224+High Sorangium 0.11
Nannocystis 0.11
Sand filter224+High Nannocystis 0.66
Fish tank252+High Sorangium 0.3
Biofilter252+High Sorangium 0.29
Sand filter252+High Sorangium 0.17

We wanted to determine whether there was a positive correlation between the presence of these two putative geosmin producers and specific environmental factors, the occurrence of geosmin in fish tissue, and the detection of the geoA gene. The identification of Sorangium at different sampling times in unit 1, especially in the fish tank, positively correlated (P ≤ 0.05) with the presence of phosphate (Table 2) and the detection of geoA (Table 4). No correlations were found between the presence of Nannocystis in the biofilter and any factors analysed in unit 1. In unit 2, the presence of Sorangium showed no correlation with any factors. In contrast, the presence of Nannocystis in the fish tank positively correlated with the detection of geoA. There was a positive correlation between Nannocystis in both the fish tank and sand filter and the detection of geosmin in fish tissue (Table 4).

Table 4. Spearman correlation between Sorangium and Nannocystis and the detection of geoA and geosmin
ConditionRAS compartmentUnit 1 Spearman factors (r/P)Unit 2 Spearman factors (r/P)
  1. r is the correlation factor with a value between −1 and +1.

  2. P is a second correlation factor that was either equivalent or below 0.05 for significant analysis.

  3. Bold value corresponds to significant results based on the r and P values.

  4. N.A., not applicable.

geoA per genus
Sorangium Fish tank 0.840/0.004 0.661/0.052
 Biofilter0.603/0.0850.159/0.682
 Sand filter0.5/0.1700.036/0.886
Nannocystis Fish tankN.A. 0.688/0.040
 Biofilter0.05/0.898−0.305/0.424
 Sand filterN.A.0.196/0.433
Geosmin2 per genus
Sorangium Fish tank0.279/0.4670.529/0.142
 Biofilter0.085/0.8270.008/0.982
 Sand filter0.091/0.815−0.052/0.886
Nannocystis Fish tankN.A. 0.673/0.046
 Biofilter0.303/0.4280.201/0.604
 Sand filterN.A. 0.797/0.005

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

In RAS, the accumulation of nitrogen compounds (ammonium and nitrite) has a deleterious effect on water quality and fish growth (Colt & Armstrong, 1979; Wickins, 1980; Russo & Thurston, 1991); it has also been shown to have a negative effect on ammonia oxidizers (Cua & Stein, 2011). The biofilter is the activity centre for nitrification carried out by Proteobacteria (Nitrobacter sp., Nitrosomonas sp., and Nitrosospira sp.) and Nitrospira, which keep ammonium and nitrite concentrations below the levels that are toxic to fish (Foesel et al., 2008). We used Ion Torrent sequencing to detect minor bacterial populations (< 1%) such as the nitrifying bacteria, which was not feasible utilizing traditional 16S rRNA gene libraries. Despite the production of short reads (100–200 nucleotides) provided by this technology, it was shown that the V5 region of the 16S rRNA gene sequences targeted in this study allowed for the highest classification accuracy of short reads (Liu et al., 2007; Claesson et al., 2010). This method was also validated in the context of environmental studies by comparing Ion Torrent and 454 data sets (Yergeau et al., 2012). Based on the Unifrac analyses and high-throughput sequencing, the bacterial structure of nitrifying populations and other bacterial populations present in the biofilter was significantly different between the two RAS units. This disparity can be partially explained by the diet used in unit 2 that generates friable faeces, which caused an increase in the suspended organic matter content of the water and higher turbidity.

Ammonia oxidation in RAS is generally associated with two specific bacterial genera identified as Nitrosomonas and Nitrosospira (Sakano & Kerkhof, 1998; Foesel et al., 2008). The elevated proportion of Nitrosospira compared with other nitrifying bacteria populations in unit 2 early during operation may be due to the higher concentrations of TOC and increased turbidity that inhibited the growth of Nitrosomonas (Krümmel & Harms, 1982) and supported the growth of heterotrophic bacteria in the biofilter (Nogueira et al., 2002; Michaud et al., 2006). However, this did not impact the ammonium concentration for the first 42 days, as both units displayed similar values. In unit 1, high nitrite concentrations detected at the beginning of this experiment potentially affected the abundance of Nitrosospira. A switch between the two ammonium oxidizers was observed after 112 days with the presence of Nitrosomonas early in the experiment, while the nitrite concentration was decreasing. During nitrification, Nitrospira is generally considered as the dominant nitrite oxidizer in RAS (Hovanec et al., 1998; Itoi et al., 2007; Keuter et al., 2011); however, it appeared very late in both units (112 and 224 days) and was only predominantly detected in the fish tank of unit 2 among the nitrifiers. The toxic effects of high nitrite concentrations on Nitrospira may explain this delay (Blackburne et al., 2007). However, no significant differences in nitrate concentrations were observed. We showed the coexistence of various nitrifiers in the two RAS units that is evidence of functional redundancy, contributing to the stability and performance of the system for nitrification (Siripong & Rittmann, 2007).

Although the presence of off-flavour compounds in RAS fish farming has been reported, we confirm the detection of the geosmin-synthesis gene prior to the release of geosmin in RAS. The geosmin-synthesis gene (geoA) was detected in the biofilter and in the sand filter in unit 2 at day 56 by qPCR, whereas SPME–GC–MS results showed a higher concentration of geosmin in the trout fillets after 84 days. The biofilter and the sand filter are two compartments that are known to concentrate geosmin producers (Guttman & van Rijn, 2008) and to accumulate phosphate (Shnel et al., 2002). A similar gap of 28 days was found for the detection of geoA in unit 1 by qPCR (140 days), and the synthesis product of this gene determined using SPME–GC–MS on the trout fillets after 168 days. Although qPCR primers were designed specifically to target Streptomyces geoA, Auffret et al. (2011) showed that the detection and quantification of non-Streptomyces geoA was possible. Our results are in accordance with a high diversity that has already been shown among Streptomyces strains (Auffret et al., 2011). However, the geoA qPCR fragments generated from the different units and compartments were more related in the amino acid sequences to GeoA from Myxococcus xanthus DK 1622 (YP_634376) and Sorangium cellulosum ‘So ce 56’ (YP_001612078) than GeoA detected in Saccharopolyspora erythraea ATCC 11635 (YP_001105919) (data not shown). GeoA from S. erythraea served to validate our qPCR method (Auffret et al., 2011). These results concur with the absence of 16S rRNA gene sequences related to Streptomyces in the samples and any attempt to isolate Streptomyces on specific media (data not shown).

All samples in which Myxococales, Sorangium and Nannocystis were detected coincided with the detection of geosmin in the fish filet. Sorangium are known to possess a putative geoA, and Nannocystis have been shown to produce geosmin (Dickschat et al., 2007; Schneiker et al., 2007). The abundance of geoA and the relative abundance of Nannocystis and Sorangium did not always correlate with the geosmin levels observed in the trout fillets. A gap of 28 days between the detection of geoA in samples and the presence of geosmin in trout fillets could explain this observation. Geosmin is also a secondary metabolite produced by Actinomycetes spp. during the stationary growth phase (Seto et al., 1998). It is highly probable that the detection of geoA preceded the production and/or release of geosmin in the RAS units. Moreover, higher concentrations are generally detected in trout fillets (Petersen et al., 2011) due to the accumulation of these compounds in the gills, skin or gastrointestinal tract by lipid-rich fish tissues (Klausen et al., 2005). Attempts to isolate geosmin producers, such as actinomycetes or cyanobacteria (Robin et al., 2006; Guttman & van Rijn, 2008; Schrader & Summerfelt, 2010), on different selective media or to detect them by Ion Torrent sequencing were unsuccessful (data not shown). The presence of other geosmin-producing bacteria that have not been characterized as producers of this secondary metabolite cannot be excluded.

The presence of phosphate could have facilitated the growth of Sorangium and promote geoA expression, thereby providing conditions that enhance geosmin production. The phosphate concentration was nearly 1 mg L−1 in unit 2 (twice the concentration in unit 1) early during the operation prior to the marked rise of geosmin in fish flesh. An amount of 1 mg L−1 (equivalent to 10 μM) was previously correlated with the occurrence of geosmin production in aquaculture systems (Robertson et al., 2006). This amount has also been shown to control the biosynthesis of secondary metabolites such as geosmin in Streptomyces sp. (Schrader & Blevins, 2001; Martín, 2004). This level of phosphate occurred much later in unit 1 (day 154) during the increase in geosmin concentration in the fish flesh.

A deeper understanding of the interaction between the nitrifying populations and environmental factors affected by the water quality may help producers ensure nitrification performance and high water quality. Moreover, our results showed that the detection of geoA was observed before the occurrence of geosmin in fish flesh in both units. We believe that our approach can be used to indicate the establishment of off-flavour producers prior to the release of geosmin in the RAS. Furthermore, problems of off-flavours is a major concern in different water treatment systems like drinking water obtained from surface water (Suffet et al., 1996) and have also been associated with off-flavour problems in fish and seafood products (Smith et al., 2008). This problem affects some other industrial processes like the production of wine or other alcohols (Du & Xu, 2012).

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

This manuscript was supported by a Strategic Projects grant from the Natural Sciences and Engineering Research Council of Canada, the Réseau Aquaculture Québec and the Société de recherche et de développement en aquaculture continentale inc. We are very grateful to Sylvie Sanschagrin for the Ion Torrent sequencing analysis and to Nathalie Fortin for performing the MacVector analysis. We also thank Jean-Christophe Therrien and Karla Vazquez for technical support. This manuscript was revised by Elsevier Language Editing Services.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
fem12053-sup-0001-FigureS1.pdfapplication/PDF85KFig. S1. Unifrac analysis of the total bacterial community found in both RAS units.
fem12053-sup-0002-TableS1.docxWord document12KTable S1. Ion Torrent primer sequences.
fem12053-sup-0003-TablesS2-S4.docxWord document23K

Table S2. Relative abundance of the major bacterial families identified in the fish tanks of both units based on 16S rRNA gene sequencing.

Table S3. Relative abundance of the major bacterial families identified in the sand filters of both units based on 16S rRNA gene sequencing.

Table S4. Relative abundance of the major bacterial families identified in the biofilters of both units based on 16S rRNA gene sequencing.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.