A transdisciplinary approach to the initial validation of a single cell protein as an alternative protein source for use in aquafeeds

Single cell protein biomass and feed pellet formulation

Methylobacterium extorquens (strain KB203) was produced via standard aerobic fermentation processes (Bélanger et al., 2004) and de-watered to form a flour referred to as KnipBio Meal (KBM). KB203 was incubated at 30 °C at 200 RPM on CHOI4 liquid medium (Supplemental Table 1) with 0.5% methanol for 24 hr in 50 mL liquid medium (in a 250 mL baffled flask). To determine purity, the suspension was streaked onto tryptic soy agar and incubated at 30 °C for 96 hr. Only colonies of a single morphology were considered pure and fit for further use in scale-up.

CHOI4-defined medium and trace metals stock solution recipes were used for growing M. extorquens to high cell densities. The trace metals solution was prepared separately as a concentrate and autoclaved for 30 min at 121 °C. A 30 mL trace metals solution was added to the medium before sterilization. Although precipitation is often observed in these solutions, they have been used repeatedly with success for growing M. extorquens to high cell densities (Bélanger et al., 2004).

A 20 L fermenter (equipped with two Rushton-type impellers; Chemap, Uster, Switzerland) was used for growing the inoculum for the main fermenter. Two pH probes (Mettler, Toledo), two pO₂ probes (Ingold) and one methanol probe (volatile organic compound (VOC) probe; NRC, Montreal, Canada) were prepared and fit into the 20L fermenter before sterilization. Ten liters of CHOI4 medium was prepared and sterilized in the fermenter for 45 min. After cooling to room temperature, a two-point calibration was conducted on the methanol probe by aseptically adding two defined volumes of methanol to reach a final concentration of 0.18% in the fermenter (2 × 9 mL). Sterile ammonium hydroxide and methanol were connected to the fermenter to control pH (at 7.0) and methanol concentration. The pO₂ and methanol probes were calibrated under standard minimal positive pressure (0.05 bar). Final fermentation occurred in a 1500 L fermenter (equipped with three Rushton-type impellers and a mechanical foam breaker; Chemap, Uster, Switzerland) with similar specifications to seed fermentation, prepared with two pH probes (Mettler Toledo, Columbus, OH, USA), two pO₂ probes (Ingold and Hamilton, Reno, NV, USA) and two methanol probes (NRC, Montreal, Canada). The probes were prepared and fit into the fermenter before sterilization. 

For bench top fermentations, biomass was collected through centrifugation (Beckman-Coulter, Indianapolis, IN, USA) and freeze dried for 48 hr at −80 °C, and moisture content was verified to be below 10% (New Jersey Feed Labs, Ewing Township, NJ, USA). At larger scale, the biomass was harvested by cooling to 20 °C and pressure was increased to 0.8 bar before feeding to the BTPX 205 disc stack centrifuge (Alfa-Laval, Lund, Sweden) at 100L/hr (discharged every 2 min). The slurry was sent directly into 50 L polypropylene carboys and stored at 4 °C until further treatment. After approximately 24 hr, the slurry was spray dried at 182 °C/70 °C (inlet/outlet).

Diet formulation

Experimental diets used for all animal trials were produced using commercial manufacturing methods. Specifically, for both Pacific white shrimp (Litopenaeus vannamei, Table 1) and Atlantic salmon (Salmo salar, Table 2) trials, ingredients were ground to a particle size of <200 µm using an air-swept pulverizer (Model 18H; Jacobsen, Minneapolis, MN, USA). The diets were processed using a twin-screw cooking extruder (DNDL-44, Buhler AG, Uzwil, Switzerland) with a 25 sec exposure to 127 °C in the extruder barrel (average across five sections). Pellets were dried with a pulse bed drier (Buhler AG, Uzwil, Switzerland) for 20 min at 102 °C with a 10 min cooling period, resulting in final moisture levels less than 10%. All oil was top-coated after the pellets were cooled using a vacuum-coater (AJ Mixing, Ontario, CA, USA). Yttrium oxide was added to the reference diet at 0.1% of dry weight to serve as an inert, indigestible markerand was diluted to 0.07% when the test ingredients were added. Diets were stored in polypropylene plastic bags at room temperature until fed. All diets were fed within four months of manufacture.

For smallmouth grunt (Haemulon chrysargyreum) trials (Table 3), all dry ingredients were mixed using a feed mixer (Model KSMS, Kitchen Aid Inc.; St. Joseph, Michigan, USA). Pollock liver oil and lecithin were added to the mixture followed by about 45% distilled water to aid the pelleting process via meat chopper (Royal, Tokyo, Japan, type 22VR-1500). After pelleting, all diets were dried at 70 °C in a constant temperature oven (DK 400; Yamato Scientific Co., Ltd., Tokyo, Japan). The dried pellets were steamed at 100 °C for 1 min in a cylindrical steamer to improve water stability. Pellets were stored in plastics bags at 30 °C until used. All diets were fed within four months of manufacture.

The effect of KBM on growth, survival, and feed efficiency of the Pacific white shrimp

Hatchery-raised Pacific white shrimp (Litopenaeus vannamei) were acquired from SKY8 Shrimp Farm, LLC (Stoughton, MA, USA) and stocked at 60 shrimp/tank (shrimp average weight was 4.52 ± 0.21 g (1.S.D.)) into twelve 110 L glass aquaria (0.228 m³) comprising a 1,675 L single clear water recirculating saltwater aquaculture system with mechanical and biological filtration. Experimental systems were maintained at 27.5–28.5 °C, and 29.5–32.5 ppt salinity, and ammonia, nitrite, and nitrate were maintained at ≤0.25, ≤0.25, and ≤80.00 ppm, respectively. This density was greater for the initial stocking given that destructive sampling would take place. The final stocking density is on par with that practices by intensive land-based systems. Animal care and procedures used in this trial were approved by Roger Williams University Animal Care and Use Committee (IACUC protocol R-13-12-20).

To determine the effect of KBM on shrimp growth and survival, three diets of varying KBM inclusion were formulated (Table 1). Each of the 12 experimental tanks was randomly assigned one of the three diets, totaling four replicates per treatment. Each tank was fed to apparent satiation four times/day (800, 1100, 1400, 1600 hours). Uneaten food from the previous feed, as well as any excrement or molts, was manually siphoned from each tank prior to the next feeding. Water chemistry was tested and corrected daily throughout the duration of the experiment.

The gross wet weight (g) of all shrimp per tank was measured at days 0, 60, and 150. All shrimp were measured on day 0 (n = 60), and day 60 (n = 45–55, depending on survivorship). At day 60, 20 shrimp from each tank (n = 80 per treatment) were randomly selected and returned to their original tank (a density of 87 m⁻²) and maintained according to the above experimental design for an additional 90 days, as which point each individuals was enumerated for wet weight (g) and carapace length (mm). The shrimp not selected for the second 90-day trial (n = 25–35, depending on tank) were euthanized, placed on ice, and wet weight (g), and carapace length (mm) were measured for each individual.

A blind taste test was conducted immediately following the conclusion of the second 90-day trial. After final data were collected, the shrimp were grouped by treatment, placed on ice, and transported to JR Bean Saloon in Bristol, RI, USA for preparation for human consumption. Shrimp were aggregated by treatment and were subsequently shelled, de-veined, boiled, and immediately served to 39 test participants. Diet treatment identity was concealed from test participants using colored plastic forks, one color per treatment. To randomize the order of shrimp consumption per participant, each fork was randomly labeled with a number (1, 2, or 3), indicating the order in which each participant should consume his/her shrimps. Upon completing the tasting, each participant ranked the three shrimp based on overall taste from 1–3 by placing the corresponding fork from each shrimp into one of three buckets. Participants were given the option to vote for a “tie”, however, these votes only accounted for 5% of total votes, and were not included in the analysis.

The effect of KBM on smallmouth grunt growth, proximate composition, and gut microbiome

Hatchery-raised smallmouth grunts (Haemulon chrysargyreum; N = 120; 1.37 ± 0.27 g wet wt) were stocked at 10 fish/tank into twelve 110 L glass aquaria (0.113 m³) comprising a 1,675 L recirculating saltwater aquaculture system with mechanical and biological filtration. Each of the 12 experimental tanks was randomly assigned one of four experimental diets, totaling three replicates per treatment (Table 2). Each tank was fed until apparent satiation four times/day (800, 1100, 1400, 1600 hours). Uneaten food from the previous feed, as well as any excrement, was manually siphoned from each tank prior to the next feeding. Water chemistry was tested and corrected daily throughout the duration of the experiment. Experimental systems were maintained at 27.5–28.7 °C, 31.0–33.8 ppt salinity. Ammonia, nitrite, and nitrate were maintained at ≤0.25, ≤0.25, and ≤80.00 ppm, respectively. Animal care and procedures used in this trial were approved by Roger Williams University Animal Care and Use Committee (IACUC protocol R-13-12-20).

Wet weight (g) and standard length (mm) was determined for each fish on day 0 and 41 (N = 120 and 112, respectively). Fish were individually collected using a dip net and blotted dry with a towel prior to measurements. Wet weight was measured by placing each fish on the center of a digital balance, and standard length per fish was measured from photographs of each individual using ImageJ digital imaging software (Schneider, Rasband & Eliceiri, 2012). Specific growth rate (SGR as g ⁻¹) was calculated as ((lnW_f − lnW_i × 100)∕t) where lnW_f = the natural logarithm of the final weight, lnW_i = the natural logarithm of the initial weight, and t = time (days) between the two measures.

On day 41, three fish were randomly selected from each treatment (N = 12 total), freeze-dried for 48 hr, homogenized using a mortar and pestle, and stored in borosilicate vials. Processed samples were then shipped to the New Jersey Feed Laboratory for whole body proximate, amino acid, and fatty acid analyses (% composition).

The foregut of three additional fish from each treatment (N = 12 total) was removed via dissection and immediately frozen on dry ice for gut microbial DNA analysis. DNA was extracted from the entire foregut of each animal using MoBio PowerSoil^® DNA Isolation Kits (Carlsbad, CA, USA). DNA was amplified using nested PCR, with initial amplification using primers 27F and 1525R that targeted the V1–V6 section of the 16S rRNA gene (Coates et al., 1999). A second amplification using uniquely barcoded primers 515F and 808R (Caporaso et al., 2012) were used to generate final amplicons for sequencing. The amplicons were gel purified using the Qiagen QIAquick^® Gel Extraction Kit following the manufacturer’s instructions (Valencia, CA, USA), and then were sequenced using an Illumina MiSeq high throughput DNA sequencer. The resulting sequences were quality filtered and analyzed in QIIME using default parameters (Caporaso et al., 2010).

Digestibility of KBM using Atlantic salmon

Atlantic salmon (Salmo salar; N = 96; 635 ± 97 g wet wt) were stocked at 16 fish/tank into six 417 L fiberglass tanks (0.265 m³) comprising a 7,500 L flow-through recirculating system with a drum filter, bio-filter, and 1200 L sump. Brackish well water (∼2 ppt salinity) was supplied to the system (19 L min⁻¹) and water quality was monitored weekly to ensure that a healthy environment was maintained during the trial. Dissolved oxygen (90–125%) and temperature (11.5–12.2 °C) were monitored daily. Animal care and procedures used in this trial were approved per USDA-ARS Animal Care and Use Committee (IACUC FY2014-001).

To determine the digestibility of KBM, a reference diet was formulated and then KBM was added in a standard 70%/30% (Table 3). Each of the six experimental tanks was randomly assigned to one experimental treatment, totaling three replicates per treatment. All tanks were fed the control diet (C; Table 3) for 7-days prior to initiating the experimental treatments. The basal diet contained 40% protein and 25% lipid with an estimated digestible energy of 19.6 kJ g⁻¹. Fish were fed three times/day; at 800 and 1200 hours using automatic feeders (Arvo-Tec Oy, Huutokoski, Finland) and at 1600 hours by hand to apparent satiation. The feeding software was developed from experimental growth models validated from commercial data and different genetic stocks (From & Rasmussen, 1984; Ruohonen & Mäkinen, 1992; Ursin, 1967).

Fecal material was collected from each tank 18 hr post feeding on day 2 and day 4 by manual stripping (Austreng, 1978; Hajen et al., 1993). All fish in each tank were sedated with tricaine methylsulfonate (MS 222, 0.1 g L⁻¹) and physically restrained. Pressure was applied to the abdomen to initiate defecation into clean stainless steel pans. Fish were returned to their respective tanks and allowed to recover from handling. Fecal samples collected from all fish (N = 6) in each tank were pooled as one composite sample/tank, averaging ≥5 g of dried feces. The fecal material was dried at 60 °C for 24 hr, placed into plastic bags and stored at −20 °C until analysis.

The methods of Cho, Slinger & Bayley (1982); Cho, Slinger & Bayley (1982)) and Bureau, Harris & Cho (1999) were used to estimate apparent digestibility coefficients. Yttrium oxide served as the inert maker.

The diets and fecal material were analyzed for organic matter by drying the samples at 120 °C for 2 hr and ashing the dried samples at 550 °C for 3 hr (AOAC, 1990). Organic matter was calculated as 100 minus ash content. The diets and fecal material were analyzed for lipid content by ether extraction using an Ankom lipid extraction instrument (XT-10; Ankom Technologies, Macedon, NY, USA). Crude protein was determined by the Dumas method (Ebeling, 1968) using a Leco Nitrogen Determinator (FP 528; Leco Corporation, St. Joseph, MI, USA). Energy was determined by using a Parr Instruments calorimeter (Model 1281; Moline, IL, USA). Yttrium oxide determination was by inductively coupled plasma atomic emission spectroscopy (University of Idaho Analytical Laboratory, Moscow, ID, USA). Amino acid analysis was performed using a Beckman 7300 Amino Acid Analyzer (University of Missouri Analytical Lab, Columbia, MO, USA).

Apparent digestibility coefficients of each nutrient in the experimental diets were calculated according to the following equations (Kleiber, 1961; Forster, 1999): ${\displaystyle {\mathrm{ADCN}}_{\mathrm{diet}}=100-100\left\{\text{\%}\mathrm{Y\; d}\mathrm{X}\mathrm{Nf}\right\}∕\left\{\text{\%}\mathrm{Y\; f}\mathrm{infecesXNd}\right\}}$ ${\displaystyle {\mathrm{ADCN}}_{\mathrm{ingredient}}=\left\{\left(a+b\right)\mathrm{X}{\mathrm{ADCN}}_{\mathrm{t}}-\left(a\mathrm{X}{\mathrm{ADCN}}_{\mathrm{r}}\right)\right\}{b}^{-1}}$ where, ADCN = apparent digestibility coefficient of nutrient; Yd = % Yttrium oxide in diet; Nf = % nutrient in feces; Yf = % Yttrium oxide in feces; Nd = % nutrient in diet; p = proportion of test ingredient in the test diet; a = (1 − p) × nutrient content of the reference diet; b = p × nutrient content of the test ingredient; t = apparent digestibility coefficients of the nutrient in the test diets; r = apparent digestibility coefficients of the nutrient in the reference diet.

Statistical analyses

Statistical analyses were conducted using JMP 8.0, SAS, Inc. (Cary, NC, USA). The effect of experimental diets on shrimp growth rate and feed conversion ratio, and the effect of experimental diets on animal growth rate, was assessed using a one-factor analytical design. Normal data (indicated by a Jarque-Barre (JB) test for normality; Zaiontz, 2015) were tested with a one-way analysis of variance (ANOVA). If the JB test indicated non-normal data, the analyses were conducted nonparametrically using a Kruskal-Wallis test. Post-hoc comparisons of normal data were made and post hoc comparisons of orthogonal contrasts from ANOVA tests were examined using the Real Statistics Resource Pack software (Release 4.9, Zaiontz, 2015). The effect of experimental diet on shrimp taste was determined using a two-factor (treatment and order of preference) chi-square goodness-of-fit test. The effect of experimental diets on grunt microbiome was determined using adonis, a permutational multivariate analysis of variance, (PERMANOVA) implemented in R, and the similarities among treatments were calculated using the Bray-Curtis similarity metric and were visualized using a principal coordinates analysis in QIIME (Caporaso et al., 2012). The significance level for all analyses was set at P ≤ 0.05. All values are presented as mean ± standard error.

Pacific white shrimp

Diet had no effect on shrimp survival (one-way ANOVA, F_2,9 = 2.4, p > 0.1, combined average =84.7 ± 5.6%); however, diet did influence shrimp growth (one-way ANOVA, % weight gain, F_2,9 = 5.4, p < 0.05; SGR, F_2,9 = 8.6 g d⁻¹, p < 0.01). Shrimp fed diet with 100% FM replacement (SHR-KH) grew less than those fed the control diet (SHR-C), and shrimp fed diet with 50% FM replacement (SHR-KL) showed growth intermediate to, and not statistically different from either SHR-C or SHR-KH (Table 4). Diet influenced shrimp feed efficiency (one-way ANOVA, F_2,9 = 5.27, p < 0.05, Table 4). The food conversion ratio (FCR) of shrimp fed diets containing KBM (SHR-KL and SHR-KH) were not statistically different than those fed the control diet (SHR-C, 1.70 ± 0.12, orthogonal contrast, Q Test = − 0.99, p > 0.7). The shrimp fed the SHR-KH had a statistically greater FCR than those fed SHR-KL (orthogonal contrast, Q Test = − 4.48, p < 0.05).

In the consumer taste trial, diet did influence shrimp preference (df = 4, χ² = 9.8, p < 0.05). This was largely because shrimp fed diet SHR-C received more votes as “most preferred” (50% of all first place votes). The shrimp fed diets SHR-KL and SHR-KH each received 25% of the “most preferred” votes. However, there was no diet difference in the shrimp that was voted “least favorite”; each diet received 1/3 of the last place votes.

Smallmouth grunt

Diet had no effect on grunt mortality, length, weight, condition factor, or specific growth rate (SGR) (for all variables, one-way ANOVA F_3,8 > 2.85). Only 12 of the 120 grunts died during the experiment, and no more than one fish was lost per tank. For all fish, the average weight increase was 353.2 ± 45.9%, length increase was 48.4 ± 6.2%, condition factor was 2.3 ± 0.2, and SGR was 3.8 ± 0.3 g d⁻¹. Average feed conversion ratio (FCR) per treatment ranged from 1.09–1.24. Diet did affect fish proximate composition (one-way ANOVA F_3,8 = 4.3, p < 0.05), where fish fed the control diet (GRU-C1) had the greatest protein content (dry matter, 56.3 ± 1.9%), while those fed diet with 50% FM replacement (GRU-KH) had the lowest (52.93 ± 1.2), while the control diet with pigment (GRU-C2) and the diet with 10% FM replacement (GRU-KL) were intermediate and not statistically different from the extremes (R30: 53.6 ± 1.0%; C+: 53.76 ± 1.1%. The percent fat on a dry matter basis, while not significant, exhibited the opposite trend to protein with the fish fed GRU-C having 22.9 ± 2.2% fat, while those fed diet GRU-KL had 27.3 ± 1.3% fat.

Diet did not significantly affect grunt gut microbial community (adonis, p > 0.05). Regardless of diet, ∼60% of the fish gut microbiome was composed of three types of bacteria: Halomonas, Oxalobacteracaea, and Shewanella. The composition of the remaining microbial community varied among individuals, but differences in the total number of operational taxonomic units among treatments was not statistically significant (Fig. 1), and there was no clear treatment grouping on the principal coordinates analysis.

Atlantic salmon

For salmon, KBM inclusion had no measurable effect on the Apparent Digestibility Coefficient (ADC) value. The ADC for protein was slightly greater for the animals fed the control diet compared to the 30% inclusion KBM diet (67.8 ± 2.8 and 63.0 ± 3.1 (averages ±1 S.D.) respectively). Diet did have a positive although insignificant influence amino acid digestibility, as the KBM diet (SAL-K) resulted in better digestibility for seven of the eight essential amino acids and six of the 11 non-essential amino acids (Table 5) than the control (SAL-C) diet (for all amino acids, binomial probability, ∏ = 0.5p < 0.08).