By Ivan Yu. Evdokimova, Angelina V. Malkova, Alena N. Irkitova, Maksim V. Shirmanova, Altai state University and Dmitrii V. Dementyev, Arsal Limited Liability Company (LLC), Russia
One of the main problems of the 21st century is the lack of food. In addition, aquaculture is an important source of nutrition for many people. Moreover, aquaculture today is one of the most promising sectors of agriculture. Breeding and rearing of aquatic animals is widespread in coastal countries, such as India, Thailand, Mexico, etc. [1, 2, 3]. The shrimp aquaculture industry can be distinguished as one of the most in demand. Its production exceeds millions of tons per year. But in the first place is still fish aquaculture
In the industrial breeding of crustaceans, farmers face problems such as infections and mass mortality due to them. Pathogens in the aquatic environment include representatives of the genus Vibrio Salmonella, Escherichia coli etc. Probiotics actively studied and used to improve the sanitary background of aquaculture enterprises.
A group of Indian scientists in a series of experiments carried out work on enriching Artemia with a Lactobacillus sporogenes supplement. Probiotic was used with different concentrations for further feeding to freshwater shrimp M. rosenbergii. According to their data, M. rosenbergii shrimp showed better survival, rapid growth and a greater content of protein, amino acids, carbohydrates, and lipids in the body after enrichment of mobile feed (Artemia crustaceans). They also proved that the concentration of L. sporogenes 6 × 108 CFU/g was optimal for the use of these microorganisms in foraging.
Spanish scientists conducted an experiment to analyze the effect of probiotic using as a means of combating potentially pathogenic members of the family Vibrionaceae. For this, they introduced a probiotic based on L. rhamnosus in the medium to A. metanauplii. As a result, they show a positive effect of L. rhamnosus on crustaceans. It was expressed in the level of pathogens decrease by a whole order of magnitude compared to control.
Scientists from India conducted a similar study. They studied the effect of a probiotic supplement based on Saccharomyces boulardii on a pathogenic microorganism. Researchers found that yeast-enriched Artemia nauplii, after artificially infected by representatives of Vibrio, have improved resistance to infection. This proved the positive effect of S. boulardii microorganisms on Artemia.
Scientists from Iran carried out work to assess the impact of the Artemia enrichment with a Bacillus subtilis probiotic supplement on the growth, reproduction, microflora and resistance to Aeromonas hydrophila for ornamental fish Poecilia latipinna. They found that the enrichment of Artemia positively affects the parameters of the test group of fish. It also was found that there are no significant differences in the ontogenesis of enriched and unenriched Artemia groups.
A group of Indian scientists conducted research on the enrichment of nauplii A. parthenogenetica by various microorganisms: L. rhamnosus, B. coagulans. During the experiment, the intestine load and the retention time in the intestine of the Artemia nauplii of probiotic supplements were studied. As a result, scientists revealed that the nauplii of Artemia enriched with L. rhamnosus reached the filled state of the intestine after 39 ± 1.41 minutes, and the enriched nauplii of B. coagulans – after 39.5 ± 0.71 minutes. The load on the intestine and the retention time in the intestine varied in experimental groups.
From the above data, it can be see that well-studied microorganisms of the genera Lactobacillus and Bacillus are the most often appear in the composition of probiotic preparations. For the sustainable aquaculture development, it is necessary to increase the pool of microorganisms used as probiotics for hydrobionts. It is also necessary to expand the range of preparations based on a consortium of microbes, rather than monocultures.
The purpose of the work was to study the use effectiveness of a new probiotic based on the strains B. toyonensis B-13249 and B. pumilus B-13250 consortium in incubating Artemia franciscana cysts using mechanical counting and automatic calculation by dint of a convolutional neural network.
Microorganisms and nutrients
A consortium based on B. toyonensis B-13249 and B. pumilus B-13250 strains (Figure 1) from the collection [19, 20] was selected as the active component of the biopreparation for research.
The following nutrients were use in the work:
L-medium (g/l): yeast extract – 5.0; peptone – 15.0; NaCl – 5.0; agar – 15.0; distilled water – 1.0 l. Vegetative nutrient medium (g/l): yeast extract – 5; peptone – 3; NaH2PO4 – 1; K2HPO4 – 4; MgSO4 – 0.25; MnSO4 – 0.03; NaCl – 1; CaCl2 – 1; FeSO4 – 0.01; glucose – 15; H2O – up to 1 l. (pH 6.8-7.0). Fermentative nutrient medium (g/l): molasses – 25; corn extract – 12.5; yeast extract – 1; tripton – 0.5; MgSO4 – 0.25; MnSO4 – 0.03; CoCl2 – 1; saline runoff – 10 μl; H2O – 1 l. Cryoprotective medium (g/l): gelatin – 25, sucrose – 100 [21].
Production of an experimental probiotic lot
The development of an experimental-industrial lot of probiotic for testing also carried out based on the laboratory EC Prombiotech of Altai State University (Barnaul, Russia). The finished product includes a composition of B. toyonensis B-13249 and B. pumilus B-13250 strains, which were cultivated separately and mixed after drying.
To prepare the mother culture the strains inoculated by loop in 500 ml flasks with 150 ml of medium. Bacteria were cultured in “Innova 44” shaker-incubator at 37 °C about 24 hours at 250 rpm (eccentricity 5 cm). The number of bacilli, the check of the fermentation process and the finished preparation purity were carried out using spread plate method. Petri dishes were incubated in a “Binder BD 115” thermostat at 37 °C for 24 hours.
Deep cultivation of each strain carried out in bioreactors with a volume of 15 and 250 liters. To obtain more B. pumilus and B. toyonensis biomass, fermentation carried out in a 250 l bioreactor with a Fermentative culture medium. Having previously grown microbial cultures of bacilli to the stage of active division on a Vegetative nutrient medium in a bioreactor by 15 liters. Before loading the nutrient medium, bioreactor and individual air filter previously sterilized with sharp steam for 1 hour at 121-125 ° C and a pressure of 0.13-0.15 MPa.
After the fermenter was inoculated, the temperature of the culture liquid maintained at 37 ± 1 °C. At the beginning of the process, sterile air continuously supplied to the fermenter in an amount of 0.5 l/min per 1 l of medium and the pressure maintained at 0.02-0.03 MPa, stirring at 250 rpm. Sterile sampling carried out every 2 hours to observe the development of bacterial culture, morphological state and absence of extraneous microflora, as well as measurements of the optical density of the culture fluid (at 490 nm).
The target product (biomass) separated by centrifugation on the flow GTGQ-1251 Tubular Centrifuge at 15000 rpm. After concentration, the biomass was mixed with the Cryoprotective medium, frozen and sent to the chamber of the freeze-dryer “SP Scientific 25l Genesis SQ Super ES-55”. The drying time was about 40 hours, after which the dried concentrate tested for moisture content on a Shimadzu MOC63u humidity analyzer. The optimum moisture content of the finished concentrate is less than 5%.
The obtained concentrates mixed with each over and with a filler – maltodextrin until the final number of bacteria in the preparation (at least 1 × 1010 CFU/g). The ratio of B. toyonensis to B. pumilus was always 1:3, since the B. pumilus strain develops a higher abundance. Therefore, its predominance in the preparation is economically justified.
Incubation and analysis of Artemia cysts
The study was carried out in laboratory of Arsal LLC in Yarovoye, Russia. The experiment used the A. franciscana cysts of two lots of two different lakes: Z29.04 (Lake Big Yarovoye) and C9 (Lake Kuchuk). Cysts incubated in cones (Figure 2) during 48 h at a constant bubbling air and standard conditions: the volume of incubatory solution – 1 l, salinity – 30 ‰, temperature – 25-30 °C, рН – 8.0-8.5, the oxidation-reduction potential (ORP) – 148-220 mV, illumination – 1500-2000 lux., quantity of the brought cysts – 2.0 g, the activator – 0.2 ml of 3% H2O2 solution.
The experiment consisted of four variants for each eggs lot (1 – control, 2, 3, 4 – experimental variants) in 24 repetitions. Incubation of both lots nauplii was carried out using probiotic doses (g): 0 (control), 0.05, 0.1, 0.2 respectively. The probiotic directly introduced into the cone with the incubation solution simultaneously with the start of the incubation process. After 48 hours of incubation, the number of crustaceans or nauplii (N), cysts (C) and embryos (U – umbrellas) counted in each cone. After that, the hatching rate (HR) calculated using the standard method and the biomass coefficient corresponding to the multiplicity of the increase in the total crude mass compared to the mass of the dry egg originally introduced.
The principle of the neural network
Until now, industrial manufacturers of Artemia crustaceans carried out the counting process by visual method – using a binocular and microscope. It takes a large amount of time, measurement accuracy, due to fatigue and constant attention is not always correct. In addition, to save at least any data you have to use additional devices or gadgets.
To reduce statistical error, at least 8 times repetition is required. To reduce the time of counting, increase the accuracy and save data, in our work an Artemia robot counter with an algorithm based on a convolutional neural network (CNN) was used for the first time in the Russian Federation. The CNN used the algorithm YOLO (You Only Look Once).
A dataset of 1200 hand-labeled photomicrographs of brine shrimp samples was used to train the neural network (Figure 3). Each sample was 0.1 ml of water from an incubator with a random number of non-hatched cysts, embryos and hatched nauplii.
All objects were marked on each sample; their calculation is the task of the neural network. The total number of marked images (dataset) was divided into two unequal parts – a training sample (about 800 images), from which the neural network learns, and a test sample (400 images), on which the neural network checks itself. Figure 4 shows photos in the sizes and qualities on which the neural network trained.
The images enlarged from their original sizes. Photo resolutions: the number of pixels per unit area is sufficient for the operation of the neural network, since the natural dimensions of objects in the visual visibility of the detector are commensurate with the samples specified for training the CNN.
Neural network training took place in the cloud service GoogleColab with connection to the Nvidia K80s, T4s, P4s and P100s graphics processors. Upon completion of training, the so-called neural network weights obtained – settings of parameters by which the neural network most accurately determines the objects sought in the image. Subsequently, weights were loaded from the cloud service for the possibility of applying them already offline.
The mechanical part of the robot was also developed and built in the Russian Federation, and the entire complex of automatic counting of nauplii called «Artemetr-1» (Figure 5).
Effect of probiotic on incubation rates of A. franciscana cysts
The results of the study on the lot of cysts Z29.04 presented in Table 1, and on the lot of cysts C9 – in Table 2. Moreover, with both lots the best results on biomass yield obtained by adding 0.1 g of probiotic per 2 g of dry cysts.
Mean differences between the control and each of the experimental cones are significant at P<0.05
The hatching percentage when adding a probiotic to a Z29.04 lot changes not so significantly compared to the control but improves by 1.4% maximum. CV from 3 to 5% indicates a slight spread of values and allows us to talk about the reliability of the result.
Mean differences between the control and each of the experimental cones are significant at P<0.05
When adding a biopreparation to a C9 lot, an increase in HR of about 10% is noted. The coefficient of variation when working with the second experimental lot was significantly higher (6.4-11.8%), but remained at an acceptable level to speak of the reliability of the result. In this case, the lowest variation observed in the experimental group with the addition of 0.1 g of probiotic.
The biomass yield shows what mass of Artemia crustaceans turned out after incubation. The graph shows the results of the correlation between the applied dose of the probiotic and the biomass yield of Artemia franciscana from two independent batches taken from two different lakes (Fig. 6).
Various samples of A. franciscana cysts (C9 and Z29.04) have different hatching percentage and biomass yield in control cones. However, in both batches, there was a decrease in these indicators with the addition of 0.2 g of probiotic compared to the best result in cone 3 (with added 0,1 g of probiotic). With sample Z29.04 excessive dosage of the biological product contributed to significant indicators decrease to a level below the control values. The effect of the probiotic is most pronounced at initially low rates of hatching percentage and biomass yield, as in the case of the C9 lot.
The biomass coefficient shows how many times the mass of brine shrimp increases after incubation. The tables also present the results of the correlation between the applied dose of probiotic and the biomass coefficient of the studied lots of Artemia.
Validation of counting by convolutional neural network
To verify the validity of the results of automatic counting, the entire lot of 192 images additionally was checked by the classical (mechanical) method (Table 3).
* N – nauplii, C – cysts, U – embryos. Mean differences between the control and each of the experimental cones are significant at P<0.05
According to the results of the inspection, no reliable differences between the counting methods were found. The entire set of obtained images (example – Figure 7) was saved, and it can be used for further research.
Shrimps are demanding on food mobility during feeding in the early stages of development. In this regard, the technology of feeding shrimp prelarvae with live, freshly hatched brine shrimp nauplii is widely used in aquaculture. The incubation solution for A. franciscana is a very favorable environment for the development of any, including unfavorable, microflora that affects the quality of the final product. The number of pathogenic bacteria in Artemia crustaceans increases exponentially during their hatching and nutrient enrichment processes. For the same reason, brine shrimp producers are interested in adding probiotic and prebiotic preparations to their product to transfer them to shrimp prelarvae and reduce the possible development of pathogens.
A large number of works in recent years have been devoted to the enrichment of Artemia with probiotic microorganisms for further feeding by crustaceans of other aquaculture objects. Therefore, this area of research is quite relevant. Since enrichment with beneficial non-pathogenic bacteria has a positive effect on cultivated species of various aquaculture objects by improving the intestinal microflora, eliminating pathogens and increasing nutritional value [32]. In addition, a number of recent scientific works aimed at fortifying the brine shrimp themselves to combat Vibrio bacteria as well as to generally increase the impact on the intestinal flora, the immune system and increasing resistance to many other pathogenic bacteria.
It known that the Artemia nauplii massively begin to hatch from cysts after 20 hours of incubation and after 6-8 hours after birth they begin actively feed. Therefore, during 48 hours of incubation, Artemia cysts not only inseminate the surface of the nauplii, but also populate their gastrointestinal tract with probiotic bacilli. The study revealed that with both lots of Artemia cysts the probiotic adding give a good effect. It is expressed as an increase in biomass compared to the control. In addition, a consortium of B. toyonensis B-13249 and B. pumilus B-13250 strains in the course of its vital activity probably favorably change the Artemia environmental conditions. This manifested in an increase of A. franciscana cysts hatching percentage in the trial compared to control. This may be due to the antagonistic properties of the used bacilli to pathogens. In our previous studies, we have found that B. toyonensis B-13249 and B. pumilus B-13250 had a pronounced antimicrobial effect on E. coli, Staphylococcus aureus, S. atyphimurium, Shigella sonnei, Pseudomonas aeruginosa.
In the first experimental group Z29.04 (samples of cysts from Lake Big Yarovoye) a noticeably higher percentage of hatching is observed. It is the norm of this Artemia lot and is not associated with the use or non-use of probiotic. The minimum spread of the Kuchuk lot (C9) indicators observed precisely in the experimental group with the addition of 0.1 g of probiotic. This may indirectly indicate a stabilizing effect of the probiotic on the hatching. Distinction in the hatching percentage in different samples of Artemia cysts (Z29.04 and C9) are due to the fact that they were obtained from divers in salt lakes. Hence, they can characterize by various properties, because they evolved in different ecosystems.
Shrimp are demanding on the mobility of the feed during feeding in the early stages of development. In this regard, the technology of feeding shrimp prelarvae with live, newly hatched brine shrimp nauplii is widely used in aquaculture. The A. franciscana incubation solution is a very favorable environment for the any (including unfavorable) microflora development. Therefore, this can affect the quality of the final product. For this reason, the Artemia manufacturers are interested in probiotic and prebiotic adding to their products to reduce the possible pathogens development and transfer it to shrimp prelarvae.
The advantages of the developed Artemia robot counter with an algorithm based on a CNN repeatedly cover various complexities. These include the use of several devices at the same time, reducing the processing time of the received data, the ability to save data for future use, and reducing the time the operator is present.
The main difference between the YOLO algorithm and other neural network algorithms used to detect objects is instant detection of objects in real time. The principle of YOLO operation implies the introduction of a full-size image, which completely passes through the convolutional neural network once. In other algorithms, this process occurs many times.
At the time of the start of work, one of the most productive neural networks for determining objects was the fourth version of this neural network – YOLOv4 (developed in 2020). Testing several neural networks from the MS COCO test sample (2017) of 127287 images showed the maximum performance of exactly YOLOv4. The YOLOv4 algorithm works 2 times faster than one of the most accurate and perfect EfficientDet. Compared to the older version, YOLOv3, the AP indicator (Average Precision) is improved by 10%, and the performance (FPS) is increased by 12% [36]. This led to its use as a platform for creating the “Artemeter-1” robot for the organisms counting.
The lack of reliable differences between mechanical and automatic counts confirms the high accuracy of the developed robotic counter. Its mass use will optimize the counting of nauplii, cysts and Artemia embryos during scientific research.
Thus, a new probiotic preparation based on B. toyonensis B-13249 and B. pumilus B-13250 strains has a positive effect on the hatching cysts percentage and the biomass yield of A. franciscana. The recommended dose is 0.1 g of biopreparation per 2 g of cysts. The increase in the hatching percentage when the probiotic added at a given concentration to the Z29.04 lot was 1.4% and when the biopreparation added to the C9 lot – 10%. The biomass yield in the control cones for lots Z29.04 and C9 was 5.30 ± 0.60 g and 4.60 ± 0.50 g, respectively. When a probiotic added to these lots at a concentration of 0.1 g, the biomass yield was 7.40 ± 0.69 g and 6.80 ± 0.43 g, respectively.
A new robotic system for counting nauplii using machine-learning technologies has been successfully tested for the first time. “Artemetr-1” significantly reduces the processing time of samples (15 times per sample), the amount of labor and allows saving data for future use.
Published in the March 2024 issue of International Aquafeed.
