doi
stringlengths 19
19
| pmcid
stringlengths 10
11
| plain_text
stringlengths 201
3.26k
| technical_text
stringlengths 536
3.67k
| full_text
stringlengths 7.12k
136k
| journal
stringclasses 1
value | topics
sequencelengths 1
1
| keywords
sequencelengths 2
17
|
|---|---|---|---|---|---|---|---|
10.3390/ani11113294 | PMC8614317 | Mulberry leaf is widely used in ruminants feeding, such as sheep, beef cattle, and dairy calves. Due to the high content of crude fiber in mature mulberry leaves and branches and the presence of anti-nutritional factors such as tannin, excessive addition will affect the production performance and health of livestock and poultry, and limit its large-scale application in animal production to a certain extent. The disadvantages of woody plants can be improved by microbial fermentation, which can reduce the content of anti-nutritional factors, and increase the content of peptides and amino acids, probiotics, and bioactive components. In this study, Lactobacillus, Saccharomycetes, and Bacillus subtilis were used to make mixed strains to ferment mulberry leaf powder, and different proportions were added to the diet of yellow feathered chicken broilers. The results showed that the addition of fermented mulberry leaf in the diet could improve the digestion and absorption of nutrients, and then improve its growth performance, and increase the contents of inosine monophosphate (IMP), total amino acids, essential amino acids, and delicious amino acids in breast and thigh muscle, and improved polyunsaturated fatty acids and essential fatty acids in breast muscle; this also has a positive effect on improving meat quality. | This study was conducted to investigate the effects of feeding fermented mulberry leaf powder (FMLP) on growth performance, slaughter performance, and meat quality of broilers. A total of 360 1-day-old chickens were randomly divided into 5 groups. The control group was fed basal diet (CON), 3% FMLP, 6% FMLP, 9% FMLP, and 3% unfermented mulberry leaf powder. The (MLP) group was fed basal diet supplemented with 3%, 6%, 9% fermented mulberry leaf powder, and 3% MLP, respectively. The experiment lasted for 56 days, with 1–28 days as the starter phase and 29–56 days as the grower phase. The results on the growth performance showed that diets supplemented with 3% FMLP significantly increased the ratio of villus height to crypt depth in the duodenum, jejunum, and ileum of broilers, enhanced the activity of intestinal amylase and digestibility of dry matter and crude protein, improved the average daily gain (ADG), and decreased the feed to gain ratio (F/G) (p < 0.05). Compared with the control group diet, the 3% FMLP group diet significantly increased the breast muscle yield (p < 0.05), reduced the abdominal fat ratio (0.1 < p < 0.05), and improved the slaughter performance of broilers. The 3% MLP group diet increased the shear force of breast muscle (p < 0.05) and thigh muscle of broilers compared to the control group, and adding FMLP could reverse the above results. Additionally, relative to the control group, FMLP supplementation improved the contents of inosine monophosphate (IMP), total amino acids (TAA), essential amino acids (EAA), and delicious amino acids (DAA) in breast and thigh muscle, and improved polyunsaturated fatty acids (PUFA) and essential fatty acids (EFA) in breast muscle; the 6% and 9% FMLP groups showed preferably such effects (p < 0.05). In conclusion, dietary supplementation of FMLP can improve the digestion and absorption of nutrients, and then improve the growth performance of broilers; it also has a positive effect on improving slaughter performance and meat quality. | 1. IntroductionWith the development of population and the improvement of people’s living standards, the demand for livestock and poultry production and the conventional feed resources is increasing. The shortage of feedstuff has become increasingly prominent and the price of conventional feedstuff with large consumption has gradually risen. Therefore, finding cheap and reasonable feed resources to replace conventional feedstuff has become a research hotspot and strategic future after the COVID-19 outbreak [1].Mulberry is a deciduous perennial woody plant, belonging to Morus of Moraceae. Its leaves are considered as a high-quality forage plant resource because of its rich crude protein content (22~29.8%), balanced amino acid composition, rich in vitamins, trace elements, phytosterols, flavonoids, alkaloids, polysaccharides and other bioactive substances [2,3], and so on. However, due to the high content of crude fiber in mulberry leaves and branches and the presence of anti-nutritional factors such as tannin, the excessive addition of mulberry leaves and branches would affect the production performance and health of livestock and poultry, which, to a certain extent, limits its large-scale use in animal production [4]. The related disadvantages of woody plants could be improved by a microbial fermentation treatment, which reduced the content of anti-nutritional factors, increased the content of polypeptides and amino acids, and contains a variety of beneficial products such as probiotics bioactive ingredients [5]. Studies have found that adding fermented mulberry leaves into feeds could enhance immunity [6], regulated lipid metabolism [7], and improved the quality of animal products [8]. In conclusion, fermented mulberry leaf, as a new protein feed resource, has a broad application prospect in animal husbandry production.Mulberry leaves as unconventional feed resources are mainly used in ruminants, such as sheep [9], beef cattle [10], and dairy calves [11]. Fermented feed can be used to improve the intestinal health of broilers [12,13] and growth performance [14,15] has been reported. However, there are few studies on the application of mulberry leaves in poultry production after fermentation.In recent years, probiotic fermentation technology has become a powerful tool to reduce anti-nutritional factors in feed, and improve nutritional quality and the bioavailability of nutrients [16,17]. Therefore, in this study, the mixed strains of Lactobacillus, Saccharomycetes, and Bacillus subtilis were used to ferment mulberry leaf powder to investigate the effects of fermented mulberry leaf powder on production performance, slaughter performance, and meat quality of broilers, so as to provide a theoretical basis for the application of fermented mulberry leaf in livestock production, especially in areas where mulberry leaves are widely planted.2. Materials and Methods2.1. Preparation of FMLP SampleMulberry leaf powder (MLP) and Fermented mulberry leaf powder (FMLP), which were made from the leaves of hybrid feed mulberry, also known as Yajin protein mulberry, were provided by Hunan Institute of Sericulture Science. Fermentation strains (Lactobacillus, Saccharomycetes and Bacillus subtilis = 1:2:1, viable count ≥ 3 × 109 cfu/g), provided by Shandong Kangdien Biotechnology Co., Ltd. (Linyi, China). FMLP was prepared by solid-state fermentation for one week. After laboratory testing, the routine nutrients of MLP and FMLP were obtained and are shown in Table 1.2.2. Experimental Birds and FeedingAll of the experimental procedures were approved by the Animal Care and Use Committee of Hunan Agricultural University. In total, 360 one-day-old male yellow-feathered broilers provided by Hunan Xiangjia Animal Husbandry Co., Ltd. (Hunan, China) were randomly divided into five groups, consisting of 6 replicates of 12 birds each, which was then denoted as CON group (basal diet), 3% MLP group (basal diet supplemented with 3% mulberry leaf powder), 3%, 6%, and 9% FMLP group (basal diet supplemented with 3%, 6%, and 9% fermented mulberry leaf powder). The addition dosage of FMLP was adjusted accordingly, on the basis of the study of Has et al. [18]. All birds were raised in wire cages with 3-level battery following standard temperature regimens, which gradually decreased from 32 to 25 °C. The lighting scheme was all day lighting, throughout the test. Meanwhile, birds were offered basal diet and diet supplemented with mulberry leaf powder and different doses of fermented mulberry leaf powder and provided ad libitum access to water and diet in crumbled (1–28 d) and pelleted form (29–56 d). The experiment lasted for 56 days. The basal diets of the starter (1–28 d) and grower phase (29–56 d) formulated according to the feeding standard of chicken (NY/T 33-2004) are shown in Table 2.2.3. Sample CollectionAt 56 d of age, after 8 h of starvation, 6 birds (1 bird per replicate) were randomly selected from each treatment group. The weight of broilers after plucking and bloodletting was taken as dressed weight (DW) and after removal of head, foot, and viscera was taken as eviscerated weight (EW). Dressing percentages was calculated by DW/BW. Eviscerated yield was calculated as the percentages of BW. Breast muscle, thigh muscle, and abdominal fat pad including leaf fat surrounding the cloaca and gizzard were separated and weighed. Breast and thigh muscle yields were calculated as the percentages of EW. Abdominal fat percentage was calculated by abdominal fat weight/(abdominal fat weight + EW). Subsequently, within 10 min postmortem, all the right entire pectoralis majors and thigh muscle were collected for the determination of meat quality. Parts of the pectoralis major and thigh muscle samples were cut from the same location, quickly frozen in liquid nitrogen, and then kept at −80 °C for further analysis.2.4. Growth PerformanceFeed intake was recorded weekly, and total feed consumption in each replicate were recorded at 1, 28, and 56 d to determine average daily feed intake (ADFI), average daily gain (ADG), and feed to gain ratio (F/G).2.5. Apparent Nutrient DigestibilityDuring the experiment, 0.5% titanium dioxide (TiO2) was added to the diet as an exogenous indicator. The basic diet and four experimental diets were fed to the different treatment groups respectively. The first 3 days were used to adapt the birds while in the last 3 days, about 300 g of representative fecal samples were selected from each replicate every day, pooled, weighed, oven-dried (55 °C), milled, and stored pending chemical analyses. Apparent digestibility values for crude fibre, crude protein were calculated according to the following formula:
AD (%) = 100 − [(G1 × F2)/(G2 × F1)] × 100
AD: apparent digestibility of dietary nutrients, G1: titanium content in diet, F1: nutrient content in the diet, G2: titanium content in feces, F2: nutrient content in feces.2.6. Intestinal Digestive Enzyme ActivityAfter slaughtering, the intestines of the experimental chickens were taken out, and the middle part of the jejunum about 10 cm was separated with a scalpel. The contents of the jejunum were put into a centrifuge tube, frozen in liquid nitrogen and stored at −80 °C. Amylase (Amy), lipase (LIP), and protease (PT) in jejunum contents were determined by commercial kits (Nanjing Jiancheng Bioengineering Institute, Nanjing, China), according to the manufacturer’s recommendations.2.7. Intestinal HistomorphologyBriefly, the intestinal samples were dehydrated with increasing concentrations of ethanol, cleared with xylene (Surgipath Medical Industries, Richmond, IL, USA), and embedded with paraffin wax (Thermo fisher scientific, Kalamazoo, MC, USA), and cut into 4-μm thick histological sections for hematoxylin and eosin staining. The tissue sections were measured under a microscope using a 40 × combined magnification, and an image processing and analysis system (Version 1, Leica Imaging Systems Ltd., Cambridge, UK). Villus height (VH); villus width (VW); crypt depth (CD); and VH/CD ratio (VH:CD) of the small intestine were determined by Program Image-pro Plus 6.0.2.8. Meat QualityThe meat color was measured at 60 min postmortem from a mean of three random readings made with a portable chromameter (CR-300, Minolta, Japan), which was calibrated with a white tile according to the manufacturer’s manual. At 45 min and 24 h after slaughtering, the pH of breast and thigh muscles were measured with a pH meter (pH-STAR, SFK technology, Denmark), previously calibrated with pH 4.6 and 7.0 buffers. The drip loss of breast and thigh meat was determined as described by Zhang et al. [19]. In brief, take a 3 × 2 × 1 cm piece from position of each sample of breast and thigh meat to determine drip loss. This sample was weighed and the mass was recorded as W1, and then suspended from a hook and placed in an inflatable zip–lock bag with the direction of the muscle fiber parallel to the gravity direction and hung for 24 h at 4 °C. After 24 h, the sample was removed and cleaned of moisture using filter paper, then weighed to obtain W2. Drip loss was then calculated as a percentage, where drip loss (%) = (W1 − W2)/W1 × 100%. L* (lightness), a* (redness), and b* (yellowness) of five random locations surface of the chicken breast and thigh meat were measured using a colorimeter (Konica Minolta Sensing Inc., Osaka, Japan) 1 h postmortem [20]. Cooked breast and thigh meat were cooled to room temperature and then rectangular-shaped samples (1 × 1 × 2 cm) at the same location were removed to measure tenderness using a TA-XT2 texture analyzer (Stable Micro Systems, Godalming, UK) with a Warner-Bratzler blade (code HDP/BS, Stable Micro Systems). Shear force was measured perpendicular to the axis of muscle fibers in 6 replicates for each treatment.2.9. Muscle Chemical AnalysisAbout 50 g breast and thigh muscles samples were sliced up, weighed, placed in a weighing bottle, and reweighed. The weighing bottle was placed into a freeze dryer at 50 °C for 48 h, and then reweighed. The weight difference between the initial sample and the dried sample was used to calculate the moisture percentage. Then, the dried samples were powdered with Muller CS-700 (Wuyi Haina Electric Appliance Co., Ltd., Zhejiang, China) and used for the analysis of crude protein (CP), amino acid and fatty acid composition. The crude protein, ether extract (EE), and crude fiber (CF) content were analyzed according to the method of the Association of Official Analytical Chemists.2.10. Inosine Monophosphate Content MeasurementAbout 5 g fresh muscle samples were weighed into a 15 mL centrifuge tube and homogenized in ice bath at 10,000 rpm for 30 s with t-25 ultra turrax homogenizer (IKA, Staufen, Germany). Then, we weighed 2.5 g of homogenate into a 50 mL centrifuge tube and added 25 mL of 5% perchloric acid. After shaking, it was centrifuged at 3500 rpm for 10 min in 4 °C refrigerated centrifuge, and then filtered into a 100 mL beaker. Then, 15 mL of 5% perchloric acid was added into the centrifuge tube. This was shaken well for 5 min, then centrifuged again and the two filtrates were mixed. After adjusting the pH to 6.5 with 5 mol/L and 0.5 mol/L NaOH, the filtrate was transferred into a 100 mL volumetric flask and diluted to the calibration tail with ultrapure water. The samples were filtered into the automatic vial and then used for HPLC.2.11. Amino Acid Composition of MuscleAbout 150 mg dried breast and thigh muscle were weighed into a glass bottle and 15 mL of 6 mol HCl were added. After nitrogen filling, the mixture was hydrolyzed for 22–24 h at 110 °C. Next, the hydrolysate was transferred to a 50 mL volumetric flask and diluted to calibration tail with ultrapure water. The solution was filtered using a 0.45 μm membrane filter into an autosampler vial, and then analyzed by L-8900 amino acid analyzer (HITACHI, Japan).2.12. Fatty Acid Composition of MuscleLipid extraction from breast muscle samples was performed by the Folch et al. method [17]. The extracted lipid was hydrolyzed in 2 mL KOH–methanol (C = 0.5 mol/L). After shaking for 1 min, the mixture was reacted in 95 °C water for 10 min to obtain a mixture of free fatty acids. The free fatty acid mixture was esterified in 2 mL BF3–methanol solution (W = 10%). After shaking for 10 s, the mixture was reacted in 80 °C water for 20 min. Subsequently, adding 1 mL n-hexane and 5 mL saturated NaCl solution, mixed for 1 min, then centrifuged for 15 min at 3000 rpm. Next, a volume of 800 μL fatty acid methyl esters was separated and analyzed with a GC-2010 plus gas chromatograph (Shimadzu, Japan). The injector and detector temperatures were maintained at 250 °C and 260 °C, respectively. Nitrogen was used as carrier gas, and the flow rate was 2.5 mL/min. The column temperature profile was as follows: maintained at 100 °C for 5 min, increased to 180 °C at 8 °C/min, increased to 210 °C at 4 °C/min, and maintained at 210 °C for 5 min. Next, the temperature was raised to 230 °C at 10 °C/min and then kept unchanged for 10 min. Fatty acids could be identified by comparing the retention time of the peaks with known standards (Sigma, St. Louis, MO, USA).2.13. Statistical AnalysesData are expressed as the mean ± standard deviation. Statistical analysis of the index was carried out according to the replicate of each group. The differences among the groups were analyzed by One-Way Analysis of Variance (ANOVA) followed by Tukey’s test using the SPSS 22.0 software (SPSS, Chicago, IL, USA). Significance was set at p < 0.05.3. Results3.1. Growth PerformanceThe effects of dietary supplementation of FMLP on growth performance are presented in Table 3. In the starter phase, ADG increased by 11.44%, 10.46% (p < 0.01); F/G decreased by 15.88%, 10.59% (p < 0.05), respectively, in chicken receiving 3% mulberry leaf powder and fermented mulberry leaf powder meal compared to those given basal diet group. Moreover, the ADFI in 3% fermented mulberry leaf powder group was significantly increased by 7.52% (p < 0.05) compared to the control group. In the grower phase, compared with the 3% mulberry leaf powder group, the ADG of the 3% fermented mulberry leaf powder group was significantly increased by 14.24% (p < 0.05); the ADFI and F/G of broilers among all groups had no significant differences (p > 0.05); adding low dose of fermented mulberry leaf powder had a trend to increase the ADFI of broilers (0.05 < p < 0.10). In the entire experimental phase, compared to the control group, the ADG of broilers in the 3% fermented mulberry leaf powder group was dramatically increased by 18.39% (p < 0.05), and the F/G of broilers in the 3% fermented mulberry leaf powder group was sharply decreased by 10.88% (p < 0.05).3.2. Apparent Nutrient DigestibilityThe dry matter (DM), CP digestibility of broilers in FMLP group were improved, and the 3% FMLP group were markedly increased by 6.98%, 10.36%, respectively (p < 0.05, Table 4) compared to the control group. Ether extract (EE) digestibility improved by 5.98%, 3.84%, and 9.89% (p < 0.05) in chicken receiving 3%, 6%, 9% FMLP meal compared to those given basal diet group. The digestibility of EE and ASH of broilers in all experimental groups were increased compared to the control group, but there was no significant difference between them (p > 0.05).3.3. Intestinal Digestive Enzyme ActivityThe digestive enzyme activity of amylase, lipase, protease in jejunum of broilers are shown in Table 5. Compared with the control group, the activities of amylase in jejunum of broilers in the 3%, 6%, 9% FMLP groups and 3% MLP group were significantly increased by 15.33%, 13.63%, 11.93%, and 19.46%, respectively (p < 0.01, Table 5). The lipase activity in jejunum of broilers in the 9% FMLP group was significantly increased by 24.44% (p < 0.05), and in other experimental groups it was increased, but there was no significant difference (p > 0.05) compared to the control group.3.4. Intestinal HistomorphologyThe normal function and structure of the intestinal tract were indicated by the villus height, crypt depth, and villus length/crypt depth (V/C), as shown in Table 6. The duodenal villus height of broilers in each experimental group was increased (p > 0.05), and the duodenal crypt depth was lower than that in the control group (p > 0.05). Compared with the control group, the duodenal V/C ratio of broilers in each dose of FMLP group was markedly increased by 25.69%, 19.71%, and 33.72% (p < 0.05). Compared to the control group, the villus height of jejunum in the 9% FMLP group was sharply decreased by 31.07% (p < 0.05), and the V/C ratio of jejunum in the 3% FMLP group was significantly increased (p < 0.05). The ileal crypt depth of broilers in each group decreased (p > 0.05), and the ileal V/C value of broilers in the 3% and 9% FMLP groups and in the 3% MLP group increased by 23.06%, 16.71%, and 18.27%, respectively (p < 0.05) compared to the control group.3.5. Slaughter PerformanceAs shown in Table 7, the breast muscle yield of broilers in the 3% FMLP group was markedly increased by 8.38% (p < 0.05), compared with the control group. In addition, the abdominal fat percentage of broilers in the 3% MLP group was sharply decreased by 29.68% (p < 0.05), and the abdominal fat percentage of broilers in other treatment groups had a decreasing trend (0.1 < p < 0.05) compared to the control group. There were no significant differences in dressing percentage, eviscerated carcass yield, thigh muscle yield among the treatment groups (p > 0.05).3.6. Meat QualityThe meat quality of the breast and thigh muscle fed with diets containing different doses of FMLP and MLP are summarized in Table 8. The L* value of thigh muscle was decreased (p > 0.05) and that of breast muscle was increased by adding FMLP to broiler diet, and in the 9% FMLP group, it was significantly increased by 16.45% (p < 0.05). Adding FMLP to broiler diet could reduce the a* value of thigh muscle (p > 0.05), but increased the a* value of breast muscle. Compared with the control group, the b* value of breast muscle in broiler diets supplemented with FMLP had no significant effect (p > 0.05), but it could reduce the b* value of thigh muscle, and in the 6% and 9% FMLP groups, it decreased by 27.80% and 25.65%, respectively (p < 0.05). Compared to the control group, the 3% MLP group increased in muscle shear force of breast muscle (p < 0.05) and thigh muscle (p > 0.05). However, the shear force of breast muscle and thigh muscle in the FMLP group decreased, especially in the 6% and 9% FMLP groups, in which it markedly (p < 0.05) decreased, compared with the 3% MLP group. Adding FMLP and MLP to the broiler diet could reduce the drip loss of breast and thigh muscles (p > 0.05). In addition, there were no significant effect on the ph45min value and ph24h value of broiler muscles (p > 0.05).3.7. Muscle Chemical Composition and Inosine Monophosphate ContentThe chemical composition including the moisture, EE, CP, and IMP content in breast and thigh muscle is presented in Table 9. The IMP of breast muscle increased (p < 0.05) by 31.11, 35.55, and 32.92%, the IMP of thigh muscle increased by 16.77, 19.88 (p < 0.05), and 24.84% (p < 0.05) in chickens fed 3, 6, and 9% FMLP diet compared to those given basal diet. There were no significant differences in the contents of moisture, EE, and CP in the muscle of broilers among the groups (p > 0.05).3.8. Amino Acid ProfileDietary supplementation of FMLP increased the contents of total TAA, EAA, and DAA in breast muscle and thigh muscle (Table 10). Concentrations of TAA in the 6% and 9% FMLP groups were significantly increased by 2.84% and 3.23%, and in the thigh muscle were remarkably increased by 4.16% and 4.39% compared with those in the control group, respectively (p < 0.05, Table 10). Compared with the control group, seven EAA (Try was not detected) in breast and thigh muscles of the diet treatment group were increased (Table 10 and Table S1). Lys and Val in breast muscles of the 3%, 6%, and 9% FMLP groups were significantly increased by 4.46%, 5.94%, 7.92% and 13.27%, and 14.16% and 15.04%, compared to the control group, respectively (p < 0.05, Table 10 and Table S1). Lys, Met, Phe, and Val in the 6% and 9% FMLP groups were significantly increased by 4.5% and 7.66%, 16.33% and 30.61%, 5.98% and 10.26%, an 11.38% and 13.82%, respectively (p < 0.05, Table 10 and Table S1). Compared to the control group, supplementing FMLP in the diet could significantly increase the content of DAA in the muscle of experimental chickens (p < 0.05, Table 10). Except Gly, Tyr, and Phe, the DAA concentrations in breast muscle of each diet treatment group were remarkably higher than those of the control group (p < 0.05, Table 10 and Table S1). Concentrations of DAA (except Val and Gly) in thigh muscles of the 6% and 9% FMLP groups were remarkably higher than those of the control group, in which Asp, Glu, Ala, and Phe were markedly increased by 6.78% and 9.60%, 3.00% and 11.97%, 16.52% and 25%, and 5.98% and 10.26%, respectively (p < 0.05, Table 10 and Table S1).3.9. Fatty Acid ProfileThe fatty acids detected in the breast and thigh muscles of broilers were basically the same, mainly composed of C16:0, C18:0, C18:1n-9t, and C18: 2n-6*, as shown in Table 11 and Table S2. Compared to the control group, the C18:1n-9t in the breast muscle of MLP group and each dose FMLP group increased 5.22%, 5.93%, 7.36%, and 12.54% (p < 0.05, Table S2). Concentrations of C22: 6n-3*, C18: 2n-6*, and C20: 4n-6* in the breast muscle of the control group were remarkably increased in each group (p < 0.01, Table S2), among which the C22: 6n-3* increased by 13.15%, 15.79%, 18.42%, and 23.68%, and C18: 2n-6* increased by 15.32%, 21.07%, 13.79%, and 24.77%, and C20: 4n-6* increased by 28.67%, 21.67%, 20.97%, and 23.08%, respectively. In addition, the concentrations of PUFA and EFA in the dietary treatment group was significantly higher than that in the control group (p < 0.05, Table 11), and the content of MUFA in the breast muscle had a tendency to increase (0.05 < p < 0.1, Table 11). Compared with the control group, the C16:1n-7 content in broiler thigh muscle was increased by adding FMLP in the diet, and the C16:1n-7 content in the 6% and 9% FMLP groups was significantly increased by 22.71% and 18.98% (p < 0.05, Table S2). Supplementation of 3%, 6%, and 9% FMLP significantly improved the content of C18:3n-6* in thigh muscle of broilers compared to the control group (p < 0.05, Table S2).4. DiscussionMulberry leaf, as a new type of feed resource, has a great potential for development and utilization in animal production due to its characteristics of large yield and balanced nutrition. However, its large-scale use is limited due to the fact that its mature leaves and stems contain anti-nutritional factors such as tannin. Fermentation, especially probiotic fermentation, has attracted more and more attention, because of its potential to reduce dietary anti-nutritional factors, improve feed nutritional quality, and promote animal growth performance [13,14]. Our results showed that, relative to the control group, dietary supplementation of low-dose FMLP markedly increased the ADG and sharply decreased the F/G in the starter phase, remarkably increased the ADFI, and significantly decreased the F/G in the whole experiment period, but the effect was weakened with the increase of the supplemental dose. Nutrient digestion and absorption may play an important role in improving growth performance. Previous studies have also found that the addition of 10% and 20% fermented and unfermented mulberry leaves in the diet of broilers significantly reduced the final body weight and dietary dry matter and crude protein digestibility of broilers with the increase of mulberry leaves supplemental dose [18]. This study showed that dietary supplemented with 3% FMLP could improve the digestibility of dry matter and crude protein nutrients of broilers, and the digestibility decreased with the increase of supplemental dose compared to the control group, which was consistent with the findings of Has et al. [18]. The digestibility decreased with the increase of supplemental dose, which may be attributed to the accelerated digestion rate caused by the increase of dietary fiber content, thus reducing the digestion time of nutrients and the digestion and absorption of nutrients by the gastrointestinal tract [21], and affecting the retention of nutrients (dry matter, organic matter, nitrogen) [22], because the fiber content of MLP and FMLP is higher, with 12.30% and 8.67%, respectively. In addition, in this study, relative to 3% MLP group, the F/G of broilers in the 3% FMLP group markedly decreased from 1 to 56 days. The reason for this result was related to the reduction of antinutritional factors in the diet and the degradation of macromolecular organic matter into small molecular by fermentation which is easy to be absorbed and utilized and the presence of probiotics in the diet, so as to improve the nutrient absorption and animal growth performance [23,24]. Other factors affecting nutrient absorption, such as digestive enzyme activity and intestinal morphology, were also examined. The results showed that dietary supplementation of FMLP can remarkably increase the activity of the intestinal amylase. Relevant studies have demonstrated that adding fermented feed and probiotics to broiler diet can improve the activity of intestinal digestive enzymes, which may be connected to the metabolism of probiotics in the intestine to produce part of digestive enzymes and improve the activity of related digestive enzymes [24,25]. Normal intestinal function and structure are the biological basis for growth and nutrient digestion and absorption of animals [26]. Villus height, crypt depth, and ratio of villus height to crypt depth (V/C) are important indexes to evaluate intestinal digestion and absorption in animals. The higher the villi height, the better the intestinal digestion and absorption function [27]. In this study, dietary supplementation MLP markedly increased ileum V/C value, and supplementation of FMLP significantly increased duodenum, jejunum, and ileum V/C value of broilers, with the 3% FMLP group having the most significant effect. These results were consistent with the study by Feng et al. [28], which revealed improved intestinal tissue morphological structure and increased intestinal digestive enzyme activities by adding fermented feed to broilers’ diets. According to the present results, it is suggested that dietary supplementation of FMLP promotes the growth performance of broilers by improving intestinal tissue structure, digestive enzyme activity, and nutrient digestibility.Slaughter performance is an important index to measure the carcass quality of meat livestock and poultry. It can not only directly reflect the percentage of the mass of different tissue parts in the total mass, but also reflect the difference of the deposition amount of nutrients in different tissue parts. High abdominal fat in broiler chickens will directly affect the processing of meat products, reduce slaughter rate and consumers’ purchase desire, and affect economic benefits [8]. In this study, it was found that the supplementation of MLP and FMLP in the diet of broilers can reduce the abdominal fat percentage of broilers, which may be relevant to the regulation of active substances in mulberry leaves on fat metabolism of broilers. Previous studies have demonstrated that 1-deoxynojirimycin (DNJ), the main alkaloid in mulberry leaves, had the effect of lowering blood glucose and blood lipid [29]. In our study, diets supplementing with FMLP had no effect on dressing percentage, eviscerated carcass yield, and thigh muscle yield of broilers, which was consistent with the finding of Semjon et al. [30]. In addition, diets supplementing with 3% FMLP may significantly increase the breast muscle yield of broilers, which may be related to the improvement of the digestibility of crude protein.Meat color is an important appearance index of meat quality, which directly affects consumers’ purchasing desire [31]. Indicators reflecting meat color are L*, a*, and b* [32]. Within a certain range, higher the a* value, the better the quality and freshness of the meat; the higher the L* value of meat color, the higher the gloss of the meat and the paler the color. The a* value is directly proportional to meat quality, while the b* and L* values are inversely proportional to meat quality [33]. Our results showed that the addition of 9% FMLP significantly reduced the b* of thigh muscle and increased the L* value of breast muscle, which indicated that the different types of muscle fibers might be the reason for the different effects of FMLP on different muscle tissues of broilers [34]. Probiotics and active substances may play an important role in the effect of dietary FMLP on meat color of broilers. This conjecture is consistent with the findings of Yu et al. [35], who reported that dietary supplementation of high concentrations had the most significant effect on meat color [35,36,37], and the findings of Shen et al. [38], who revealed that the effect of adding bamboo leaf extract in the diets with different concentrations on meat color was linearly increased [38]. Tenderness (shear force) may be the most important edible quality parameter that determines consumer acceptance [39]. Shear force is an intuitive indicator of muscle tenderness [40]. In the present study, the addition of FMLP could reduce muscle shear force, especially at medium and high doses. Probiotics in fermented mulberry leaf powder may play an important role in it. Previous studies similarly have found that dietary supplementation probiotic can reduce muscle shear force [35,41]. Relative to the control group, the muscle shear force of MLP group was significantly increased, which may be related to the increase of ADG, resulting in the increase of muscle fiber diameter, which in turn led to the increase of shear force, because the smaller the muscle fiber diameter is, the more tender the muscle is [42].The composition and content of amino acids, fatty acids, and nucleotides flavor substance in animal and poultry meat are important factors that affect the nutritional value and flavor. In the present study, compared to the control group, diets supplementing with FMLP remarkably improved the content of IMP, increased the contents of EAA, TAA, and DAA, and the effect of medium and high dosage FMLP were most significant, but dietary supplementation MLP had little effect on the IMP content in breast and thigh muscle of broilers, suggesting that probiotics may play an important role in the FMLP group. Previous studies similarly have demonstrated that diet supplementing with 5% alfalfa (similar to MLP, it can be used as unconventional protein feed) meal did not affect IMP of breast and thigh muscles; dietary supplementation probiotics increased the content of IMP, DAA, EAA, and DAA in breast muscle [35]. In addition, in the present study, fatty acids in breast muscle and thigh muscle of broilers were mainly C16:0, C18:0, C18:1n-9t, and C18:2n-6. Their total contents accounted for a significant proportion in the total fatty acid composition, and they were the main component of muscle fatty acids in broilers, and unsaturated fatty acids were the main component, which was consistent with the findings of Semjon et al. [30]. Farmer et al. [43] illuminated PUFA are more likely to form volatile flavor substances during lipid oxidation, which makes meat more delicious. C18:2n-6, C18:3n-6, and C20:4n-6 are EFA indispensable to the human body in PUFA, which play a very important role in maintaining normal development and health, and can effectively prevent atherosclerosis and myocardial infarction [44]. C18:3n-6 and C22:6n-3 are important raw materials for the formation of biofilms, which promote the development of the nervous system and brain [45]. Our result showed that dietary supplementation of MLP and FMLP markedly increased the content of PUFA (C18:1n-9t, C18:2n-6, C20:4n-6, C22:6n-3) and EAA in breast muscle of broilers, indicating that active substances and probiotics may play an important role in MLP and FMLP groups. Previous studies similarly have reported that diets supplementing fermented ginkgo biloba leaves increased the contents of flavonoids and polysaccharides in diets, and then increased the content of total PUFA in breast muscles [46]; diets supplementing with probiotics improved the content of PUFA and SFA in breast muscles [47]. Additionally, dietary supplementation with MLP and FMLP had little effect on thigh muscle, only improved the content of C18:3n-6 in muscle, which indicated that the effects of MLP and FMLP on fatty acid content of muscle in different parts of broilers were different, which might be caused by potential differences in nutrient absorption and distribution among different tissues [48]. According to the present results, it is suggested that dietary supplementation with MLP and FMLP can improve the nutritional value and flavor of meat by affecting the composition and content of PUFA in muscle of broilers.5. ConclusionsDiet supplementing with FMLP at a dosage of 3% could improve the digestion and absorption of nutrients, such as the digestibility of dry matter, CP and EE, amylase and V/C ratio, and then improve growth performance. Adding 6% and 9% FMLP could improve the meat quality of breast and thigh muscles without affecting the growth performance, such as increasing the concentration of IMP, TAA, EAA, DAA, PUFA, and EFA, reducing the shear force, and the 9% FMLP group showed preferably such effects. | animals : an open access journal from mdpi | [
"Article"
] | [
"fermented mulberry leaf powder",
"broiler chicken",
"meat quality",
"growth",
"slaughter performance"
] |
10.3390/ani13050785 | PMC10000113 | This study was conducted to investigate the effects of dietary supplementation with Bacillus licheniformis and a combination of probiotics and enzymes on the growth and blood parameters of grazing yak calves. The body weight, body size, serum biochemical parameters, and growth hormone levels of grazing yaks were assessed. We found that supplementation with probiotics alone or with a combination of probiotics and enzymes significantly increased the average daily gain, compared to the controls, and the combination of probiotics and enzymes showed a better performance. Supplementation with the complex of probiotics and enzymes significantly increased the concentration of serum growth hormone, insulin-like growth factor-1, and epidermal growth factor, which may be the main reason for the higher daily weight gain. The findings of this study may help improve the growth efficiency of yak calves on the Qinghai–Tibetan Plateau. | Early weaning is an effective strategy to improve cow feed utilization and shorten postpartum intervals in cows; however, this may lead to poor performance of the weaned calves. This study was conducted to test the effects of supplementing milk replacer with Bacillus licheniformis and a complex of probiotics and enzyme preparations on body weight (BW), size, and serum biochemical parameters and hormones in early-weaned grazing yak calves. Thirty two-month-old male grazing yaks (38.89 ± 1.45 kg body weight) were fed milk replacer at 3% of their BW and were randomly assigned to three treatments (n = 10, each): T1 (supplementation with 0.15 g/kg Bacillus licheniformis), T2 (supplementation with a 2.4 g/kg combination of probiotics and enzymes), and a control (without supplementation). Compared to the controls, the average daily gain (ADG) from 0 to 60 d was significantly higher in calves administered the T1 and T2 treatments, and that from 30 to 60 d was significantly higher in calves administered the T2 treatment. The ADG from 0 to 60 d was significantly higher in the T2- than in the T1-treated yaks. The concentration of serum growth hormone, insulin growth factor-1, and epidermal growth factor was significantly higher in the T2-treated calves than in the controls. The concentration of serum cortisol was significantly lower in the T1 treatment than in the controls. We concluded that supplementation with probiotics alone or a combination of probiotics and enzymes can improve the ADG of early-weaned grazing yak calves. Supplementation with the combination of probiotics and enzymes had a stronger positive effect on growth and serum hormone levels, compared to the single-probiotic treatment with Bacillus licheniformis, providing a basis for the application of a combination of probiotics and enzymes. | 1. IntroductionYaks (Bos grunniens) occur on the Qinghai–Tibet Plateau at high altitudes and with long cold seasons and limited pasture resources. This species is a unique product of long-term natural selection, providing local herders with the most basic living materials and livelihood resources, such as meat, milk, shelter (hides and furs), and fuel (dung), and is an indispensable part of the ecology and economy of the Qinghai–Tibetan Plateau [1]. However, the low reproductive rate of yaks seriously restricts their production and utilization. The cold season on the Tibetan Plateau lasts for eight months (October to the following May), during which time the quantity and quality of pasture decrease below the nutritional requirements of lactating yaks [2]. The deficiency of feed intake results in a negative body energy balance and metabolic stress [3]. On the other hand, under traditional grazing management, plateau-grazing yak calves are weaned naturally or artificially under various conditions at an age of 18–24 months [4], rather than the weaning age of domestic beef cattle (<6 months). The slow recovery itself and the late weaning of yak calves, which result in a poor postnatal physical condition, severely affect the onset of the next estrous cycle in the cow. Most yaks exhibit a long postpartum anestrous period and calve twice every 3 years or once every 2 years [5]. Therefore, the early weaning of yak calves may help mitigate these adverse effects.Early weaning has become more popular in recent years for various reasons, including the better use of limited feed resources and alleviating grazing pressure on pastures by reducing the nutritional needs of cows [6]. Weaning calves before the start of the breeding season improves the reproductive performance of cows [7,8] because the cows can regain their weight faster, thus accelerating the onset of postpartum estrus. The use of milk replacer in early weaning is common in livestock production [9,10]. The milk replacer has demonstrated positive benefits in animal experiments, such as improved immunity and relieved weaning stress response [11]. Increasing evidence suggests that enhanced milk replacer feeding is beneficial for improving gut microbial development and growth performance in early-weaned lambs [12,13].Over the past few decades, probiotics have been widely used in livestock and poultry production for their ability to enhance animal disease resistance, improve feed utilization, and improve growth performance [14]. In ruminants, yeasts and bacteria, including Lactobacillus, Bifidobacterium, Bacillus, Propionibacterium, and Enterococcus, alone or in combination, are used as additives in diets [15,16]. Probiotics can decrease diarrhea, improve production and feed utilization efficiency, and strengthen the immunity system in young ruminants [17,18,19]. Moreover, supplementation with probiotics improves the rumen and intestinal epithelial cell growth, which enhances the gastrointestinal tract development and health status of calves [17,20,21]. Oral administration of Bacillus licheniformis can increase ruminal digestibility and total volatile fatty acid concentrations in Holstein cows [22] and growth performance in Holstein calves [23]. In vitro inoculation with Bacillus licheniformis also improves ruminal fermentation efficiency of forage of various qualities [24]. However, no information is currently available on the effect of Bacillus licheniformis on the growth performance of yak calves.Compound enzyme preparations are produced from one or more preparations containing a single enzyme as the main entity, which is mixed or fermented with other single enzyme preparations to form one or more microbial products [25], including saccharylases, amylases, cellulases, proteases, phytases, hemicellulases, and pectinases. Depending on the differences in digestive characteristics and diet composition, specific enzyme preparations can be used for livestock [26]. Specific enzyme complex preparations can degrade multiple feed substrates (antinutrients or nutrients), and different types of enzymes can work synergistically to maximize the nutritional value of feed [27]. In buffalo calves, cellulase and xylanase are more effective with regard to average daily weight gain (ADG) and feed efficiency [28]. Further, the addition of exogenous fibrolytic enzymes to wheat straw has no effect on starter feed intake and increases nutrient digestibility and recumbency, but decreases the ADG of weaned Holstein dairy calves [29].The effects of probiotics or compound enzyme preparations on the production performance and biochemical blood indexes of calves are not consistent [29,30,31,32,33]. The respective discrepancies may be due to differences in the amounts of added probiotics and exogenous enzymes, the strains of probiotics, diets, and animal management strategies. Therefore, this study was conducted to compare the effects of Bacillus licheniformis and a combination of probiotics and enzymes on the growth performance and serum parameters in yak calves, so as to provide a theoretical basis for the application of probiotics in grazing yak calves.2. Materials and Methods2.1. Animals and TreatmentThis study was performed in accordance with the Chinese Animal Welfare Guidelines, and the experimental protocols were approved by the Animal Care and Ethics Committee of the Institute of Animal Husbandry and Veterinary Medicine, Tibet Academyof Agriculture and Animal Husbandry Science (No. #TAAAHS-2016–27).The feeding trial was conducted at Damxung Co., (Lhasa, China; 30.5° N, 91.1° E) from July to October. The average altitude was 4200 m, the average annual temperature was 1.3 °C, and the average annual precipitation was 456.8 mm. Thirty two-month-old male yaks (38.89 ± 1.45 kg body weight (BW)) were fed milk replacer solution at 3% of their BW every day and were randomly assigned to three dietary supplementation treatments (n = 10, each), according to BW and age, as follows: T1, supplemented with 0.15 g/kg Bacillus licheniformis (2 × 1010 CFU/g); T2, supplemented with a 2.4 g/kg combination of probiotics and enzymes (containing 0.4 g/kg Bacillus licheniformis, 2 × 1010 CFU/g; 1.0 g/kg yeast, 1 × 1010 CFU/g; 1.0 g/kg mixture of xylanase, cellulase, and glucanase in a 1:1:1 ratio, xylanase, 20,000 U/g, cellulase, 1500 U/g, glucanase, 6000 U/g); and a control treatment. The milk replacer, probiotics, and enzyme preparations were provided by the Chinese Academy of Agricultural Sciences (Beijing, China). All yak calves were allowed to graze on an alpine meadow during daytime for the 60-day trial, and they were individually fed milk replacer before and after grazing (0800 and 2000 h, respectively). The forage of the alpine meadow was mainly composed of Kobresia tibetica, and the nutrient composition (dry matter basis) was analyzed in our previous study [34], i.e., 10.4% crude protein, 2.1% ether extract, 67.8% neutral detergent fiber, 34.2% acid detergent fiber, and 4.6% ash. The powdered milk replacer was weighed and mixed with warm water (approximately 40 °C) at a ratio of 1:7 (w/v) to obtain milk replacer solution, according to our previous study [35]. Based on preliminary assessments, the feeding amount of milk replacer was calculated so that all yak calves were able to feed without surplus [35]. The nutrient composition of the milk replacer is shown in Table 1.2.2. Sample Collection and AnalysisThe BW of each yak calf was recorded before morning feeding on d 0, 30, and 60 using a platform scale, and the ADG was calculated accordingly. The body size indexes of all yak calves were determined using a linen tape at the beginning (d 0) and end (d 60) of the experiment, as previously described [36].Blood samples (approximately 10 mL) were collected from the jugular vein of the yak calves using a vacuum tube before morning feeding on d 0 and 60. The blood samples were centrifuged at 1100× g for 10 min to obtain serum, which was then aliquoted in 1.5 mL centrifuge tubes and stored at −20 °C.The serum biochemical parameters, including blood urea nitrogen (BUN), globulin (GLB), blood glucose (GLU), and non-esterified fatty acids (NEFAs), were analyzed using an automatic biochemical analyzer 7020 (Hitachi, Tokyo, Japan). Metabolic hormones in the serum, including insulin-like growth factor-1 (IGF-1), epidermal growth factor (EGF), cortisol, insulin (INS), and growth hormone (GH), were determined using commercial ELISA kits (Jiahong Technology Co., Ltd., Beijing, China) according to the manufacturer’s instructions. Briefly, 50 μL of each five-fold diluted serum sample was added to each well of a 96-well ELISA plate. After 30 min of incubation at 37 °C, the plate was washed five times using PBS (Servicebio, Wuhan, China) to remove unbound proteins. Then, 50 μL of HRP-conjugated antibodies was added to allow them to bind with their corresponding antigens. The 3,3′,5,5′-tetramethylbenzidine working solution was added to each well, followed by stop solution. Absorbance was measured using a multi-plate reader (Varioskan LUX, Thermo Fisher Scientific, Waltham, MA, USA) at a wavelength of 450 nm.2.3. Statistical AnalysisAll experimental data of this study were statistically analyzed using a one-way analysis of variance followed by Duncan’s post hoc test with SPSS 26.0 software (SPSS Inc., Chicago, IL, USA). Each yak calf was considered an experimental unit. Data are expressed as means ± standard error. p < 0.05 was considered statistically significant.3. Results3.1. Body WeightThe three treatments did not differ significantly in terms of BW on d 0, 30, and 60 (Table 2). The ADG was higher (p < 0.05) in the calves under T2 treatment than those under the control treatment, from d 0 to 30, d 30 to 60, and d 0 to 60, and higher (p < 0.05) than that of those calves under the T1 treatment from d 0 to 60, indicating that the supplementation of Bacillus licheniformis and the combination of probiotics and enzymes could improve the growth performance of early-weaned grazing yak calves. The ADG of calves under T1 treatment was higher (p < 0.05) than that of those under the control treatment from d 0 to 60.3.2. Body SizeThe body size parameters did not differ significantly among the three treatments on d 0 and 60 (Table 3), indicating that the supplementation of Bacillus licheniformis and the combination of probiotics and enzymes did not affect the body size of yak calves within 60 d.3.3. Serum Biochemical ParametersThe concentrations of serum GLB, BUN, GLU, and NEFAs did not differ significantly among the three treatments on d 0 and 60 (Table 4).3.4. Serum HormoneAs shown in Table 5, the concentrations of serum IGF-1 on d 60 were higher in T2-treated calves than in the T1- and control-treated calves (p < 0.05, each). The concentrations of serum EGF and GH on d 60 were higher in the T2-treated calves than in the controls (p < 0.05). The concentration of serum COR on d 60 was higher in the control calves than those under the T1 treatment (p < 0.05).4. DiscussionEarly weaning may have various benefits for cows; however, early weaned calves generally perform poorly compared to naturally weaned calves [37]. Early weaned calves without breastfeeding grew at a lower rate and subsequently took longer to reach their target weight than breastfed calves [38]. To improve the growth performance of early-weaned calves, several improvements were made to the composition of milk replacer or additional feeds were added [39,40,41]. Moreover, the addition of probiotics to the diets of calves significantly improved the ADG [29,30,33]. Dietary supplementation with compound enzyme preparations also improved growth performance in weaned piglets [42,43] and growing-finishing pigs [44]. However, previous studies also reported that supplementation with probiotics, yeast cultures or enzymes had no effect on the growth performance of calves [31,32,45]. In the current study, the addition of Bacillus licheniformis alone or a complex of probiotics and compound enzyme preparations to the milk replacer significantly improved the performance of grazing yaks and calves compared with milk replacer alone. Further, the addition of probiotics is beneficial for the regulation of the intestinal microbiota community structure, improving intestinal health and fecal consistency, and reducing diarrhea prevalence [19,31,46,47,48]. The supplementation of fibrolytic enzyme to the diet of crossbred calves improved their nutrient digestibility with a positive effect on daily gain [49]. Calves typically exhibit high metabolism and fast growth; however, their growth performance is susceptible to environmental stress and nutrient absorption and digestive problems, especially in the period after weaning [50]. Under natural grazing conditions on the Qinghai–Tibet Plateau, due to the long-term lack of pasture and harsh environmental conditions, the normal growth of yak calves is severely restricted [48]. In the present study, none of the study animals died, which may be attributed to the supplementation with milk replacer. Therefore, the addition of probiotics and compound enzyme preparations was beneficial for the growth of grazing yak calves.In most cases, calf weight is positively correlated with body length, and body length can be used to predict calf live weight [51,52]. Supplementation with Bacillus subtilis results in an increased body length and BW in Barki lambs at the third and fourth week, as observed in a four-week continuous feeding trial [53]. In the present study, neither body size nor BW differed among the treatments, which may be due to insufficient trial duration and individual differences in animals. Therefore, more time may be required to elucidate whether the probiotic and compound enzyme preparations affected the calves’ body size.To a certain extent, blood biochemical parameters reflect the metabolism and the acid–base balance of the animal body, and they vary within a certain range [54,55]. The results of the current study revealed that supplementation with Bacillus licheniformis and the complex of probiotics and enzyme preparations had no effect on the blood biochemical parameters of grazing yak calves, which is consistent with previously reported results in crossbred and Holstein calves [56,57]. The blood biochemical values of calves vary with the growing stage and are strongly influenced by weaning [58,59], and these possible factors may be stronger than the influence of diet on blood biochemical indicators.Insulin-like growth factors (IGFs) are small polypeptide hormones mainly synthesized and secreted from the liver, and they are structural homologs of insulin, with similar activities. These consist in binding to specific carrier proteins in the blood to form a composite factor that stimulates systemic body growth and has growth-promoting effects on almost every cell in the body [60,61]. As mediators of GH action, the synthesis of IGFs is also affected by the blood level of GH [62]. EGF is a member of the growth factor family, a single polypeptide of 53 amino acid residues that is involved in regulating cell proliferation [63]. We found that the addition of probiotics and a combination of probiotics and enzymes significantly increased the concentration of serum IGF-1, EGF, and GH, whereas supplementation with Bacillus licheniformis alone did not achieve this effect. These results are consistent with the ADG results. GH and IGF-1 are important controllers in regulating amino acid metabolism in calves, where GH promotes the entry of amino acids in muscle tissue into cells and increases protein synthesis, and IGF-1 increases protein deposition by promoting protein synthesis [63,64]. Cortisol is commonly used as a marker of stress responses (such as weanling stress) in animals, and it occurs at high serum levels for a period of time after calves are weaned [65]. In line with our results, oral supplementation with probiotics markedly decreases the concentrations of serum cortisol in neonatal and weaned calves [66,67]. Interestingly, we found that the concentrations of serum cortisol were lower in the T1 than in the T2 group, which was, however, not statistically significant. This suggested that the addition of Bacillus licheniformis alone may better alleviate weaning stress in grazing yak calves. However, the respective mechanisms remain to be resolved in more detail.A limitation of this study is that the T2 group did not strictly control a single variable compared to the T1 group, and the factors (yeast or xylanase, cellulase and glucanase) that contributed to the difference were unclear. This was due to the initial intention of this study to improve the milk replacer by adding probiotics or compound enzyme preparations, and ultimately promote the growth performance of yak calves on the Qinghai–Tibet Plateau. Further, we were unable to collect data on diarrhea and determine nutrient digestibility in grazing calves, which would have further improved our understanding of the weight gain of yaks under the various treatments.5. ConclusionsOur results suggest that supplementation with Bacillus licheniformis alone or with a complex of probiotics (Bacillus licheniformis and yeast) and compound enzyme preparations (xylanase, cellulase, and glucanase) can improve the ADG of grazing yak calves, and the complex had a better effect on the ADG. The addition of the complexes of probiotics and complex enzyme preparations also increased the concentrations of serum GH, IGF-1, and EGF, which may have led to a higher ADG. Thus, the addition of a combination of probiotics and enzymes to milk replacer may serve as an effective strategy to improve the production of yak calves. | animals : an open access journal from mdpi | [
"Article"
] | [
"early weaning",
"probiotics",
"enzyme preparations",
"yak calves",
"growth performance"
] |
10.3390/ani11051397 | PMC8156027 | Automated training devices are commonly used for investigating learning, memory, and other cognitive functions in warm-blood vertebrates, whereas manual training procedures are the standard in fish and other lower vertebrates, thus limiting comparison among species. Here, we directly compared the two different approaches to training in guppies (Poecilia reticulata) by administering numerical discrimination tasks of increasing difficulty. The automated device group showed a much lower performance compared to the traditionally-trained group. We modified some features of the automated device in order to improve its efficiency. Increasing the decision time or inter-trial interval was ineffective, while reducing the cognitive load and allowing subjects to reside in the test tank improved numerical performance. Yet, in no case did subjects match the performance of traditionally-trained subjects, suggesting that small teleosts may be limited in their capacity to cope with operant conditioning devices. | The growing use of teleosts in comparative cognition and in neurobiological research has prompted many researchers to develop automated conditioning devices for fish. These techniques can make research less expensive and fully comparable with research on warm-blooded species, in which automated devices have been used for more than a century. Tested with a recently developed automated device, guppies (Poecilia reticulata) easily performed 80 reinforced trials per session, exceeding 80% accuracy in color or shape discrimination tasks after only 3–4 training session, though they exhibit unexpectedly poor performance in numerical discrimination tasks. As several pieces of evidence indicate, guppies possess excellent numerical abilities. In the first part of this study, we benchmarked the automated training device with a standard manual training procedure by administering the same set of tasks, which consisted of numerical discriminations of increasing difficulty. All manually-trained guppies quickly learned the easiest discriminations and a substantial percentage learned the more difficult ones, such as 4 vs. 5 items. No fish trained with the automated conditioning device reached the learning criterion for even the easiest discriminations. In the second part of the study, we introduced a series of modifications to the conditioning chamber and to the procedure in an attempt to improve its efficiency. Increasing the decision time, inter-trial interval, or visibility of the stimuli did not produce an appreciable improvement. Reducing the cognitive load of the task by training subjects first to use the device with shape and color discriminations, significantly improved their numerical performance. Allowing the subjects to reside in the test chamber, which likely reduced the amount of attentional resources subtracted to task execution, also led to an improvement, although in no case did subjects match the performance of fish trained with the standard procedure. Our results highlight limitations in the capacity of small laboratory teleosts to cope with operant conditioning automation that was not observed in laboratory mammals and birds and that currently prevent an easy and straightforward comparison with other vertebrates. | 1. IntroductionThe study of learning, memory and perception in animals has, since its inception, benefited from the use of automated training equipment. The use of these methods offers a two-fold advantage. First, they reduce the time needed for training and the related human labor required. Some experiments, especially those in discrimination learning, may require thousands of training trials [1,2]. With manual execution, many months and hundreds of hours of work are required to train each subject [3,4,5]. The second advantage is that automated equipment allows for the control of every detail of the experiment, standardizing procedures across different studies and laboratories, while minimizing the need for human intervention and reducing the possible influence from researchers’ expectations [6,7].While automated training devices are frequently employed to study mammals and birds, they are rarely used outside these two vertebrate classes [8,9,10]. In the last two decades, there has been an increasing interest in studying cognition in other animal taxa, such as fish, reptiles, and arthropods. Teleost fish in particular have been thoroughly studied, and in some cognitive domains, they show abilities surprisingly close to those seen in mammals and birds [11,12,13] as well as high degree of genetic homology to humans [14]. Therefore, some small tropical fish, in particular zebrafish (Danio rerio), medaka (Oryzias latipes), and guppy (Poecilia reticulata), have become important models in neurobiological research [15,16,17], and several laboratories have tried to develop Skinner-box apparatuses for these species [18,19,20,21].Numerical cognition is one of the cognitive domains most widely investigated. On average, fish show capabilities comparable to those of mammals and birds [22]. Guppies (Poecilia reticulata), in particular, can be quickly trained to discriminate up to four from five objects, even if continuous perceptual cues (i.e., cumulative surface area or density) are made irrelevant [23]. As for primates, task difficulty increases with decreasing distance between the quantities to be discriminated [23,24].The numerical capacities of guppies exceed those observed in various warm-blooded vertebrates such as dogs, horses, and domestic chicken, though they appear lower than those exhibited by humans, apes and some other mammalian and by avian species [22]. However, these discrepancies are difficult to interpret, due to marked differences in methods. Monkeys, pigeons, and rats were studied with automated conditioning devices, whereas guppies were studied using traditional, manually operated, training methods or with spontaneous preference procedures. Studies on the former species commonly involve hundreds of training trials per day and a few thousand trials per experiment [1,2]. In contrast, guppies usually receive 12–15 daily trials, and the number of trials per experiment rarely exceeds 100 [23,25,26].In recent years there have been some attempts to use automated operant conditioning devices for studying cognitive abilities in fish [18,19,20,21,27]. Our laboratory has developed a device based on the Skinner tank built by Manabe and colleagues [18]. The device is controlled by a microcomputer, which displays stimuli on a monitor, tracks the movements of the fish, and delivers a very small amount of food when the subject makes a correct response. Guppies tested with this device can easily perform 80 trials in each daily session. If required to discriminate between two stimuli with different colors, guppies showed excellent performance, reaching 90–95% accuracy in two–three sessions. The performance in shape discrimination tasks was only slightly lower, with subjects reaching a maximum 80–85% accuracy in approximately one dozen sessions [28]. This is comparable or even superior to the performance obtained by guppies and other fish in color or shape discrimination in other studies using more traditional non-automated approaches to training [29]. Unexpectedly, when asked to discriminate between two sets of items with different numbers of elements (3 vs. 5 and 3 vs. 4), the guppies showed very poor performance [30]. This clearly conflicts with capacities demonstrated by guppy studies that used different training procedures [23,24,31]. It is worth noting that other studies using automated devices have reported unexpectedly low performance in other species and with different tasks, which calls into question the reliability of this approach of operant conditioning in fish [27,32,33]. For example, in one study zebrafish demonstrated good accuracy in color discrimination, but no evidence of learning shape discrimination [32]. No reference data are available in this case, though zebrafish have shown significant capability for recognizing familiar from unfamiliar shapes [34,35].There are two critical aspects concerning this research area. The first is that it is presently difficult to have a precise estimation of the effectiveness of the automated approach compared with the traditional ones. None of the above-mentioned studies included control groups tested in the same task with a reference standard procedure, and many differences with prior studies (i.e., type of subjects, stimuli, operant conditioning protocol) could potentially explain discrepancy in results. For example, the numerical stimuli in Gatto and colleagues [30] differed in the format and in the numerical ratios from those used in previous studies [23,26]. Furthermore, Gatto and colleagues [30] presented a mixture of two numerical discriminations simultaneously, whereas the best performing test [23] used a particular protocol in which subjects were first trained on a very easy numerical discrimination and then, upon achieving the criterion, were given progressively more difficult tasks until they reach the discrimination threshold.The second critical aspect is that even if the difference in certain tasks would be confirmed, it will not be easy to identify the causes of the efficiency gap between automated devices and traditional methods to train fish. Automated operant conditioning represents a fundamentally different approach to training which largely differs from traditional training methods in its objectives, and which is subjected to distinct constraints. The two approaches differ in a larger number of features, many of which are not directly related to automation itself. Historically, Skinner boxes have been developed to reduce human intervention and to speed up training of laboratory animals, such as rats or pigeons. The inner space of a Skinner box is extremely reduced so that stimuli, operandums, and the devices for delivering reinforcers are fitted in a small area and close to the subject. This feature permits a subject to run trials in rapid succession, usually many trials per minute, which, combined with the possibility to administer infinitesimal rewards, allows an animal to perform hundreds of trials in a brief session. Thanks to these characteristics, a single Skinner box is used to train many subjects in rotation.In the traditional, manually-operated, operant conditioning methods, the need to visually assess the subject’s choice, has made necessary testing subjects in large apparatuses. In this way, the two stimuli are kept far apart, and the subject makes its choice by moving in the direction of one of the two targets or by choosing an arm in a T-maze. The need for the experimenter to prepare the stimuli and the reward for the subsequent trial usually determine long inter-trial intervals. This, in turn, implies that the number of daily trials cannot usually exceed one or two dozen. The way stimuli are presented is also very different in the two approaches. The fully-computerized control of a Skinner box requires that stimuli are displayed in some digital format. On the other hand, since all operations of manual training are usually performed by a single experimenter, in most situations adding the extra task of operating a computer can be impractical. For this reason, solid objects or printed stimuli are usually preferred with traditional training approaches.The presence of a large number of differences could make the search for causal factors very long and complex. However, some factors are better candidates than others to account for the low efficiency of automated training approach in fish. To explain the low performance of guppies in numerical tasks, Gatto and coworkers [30,32] argued that automated devices may force fish to perform additional cognitive tasks (related to device functioning), increasing the overall cognitive load (i.e., the amount of cognitive resources that are devoted for dealing with one specific task). This factor is expected to be especially important when a subject is performing complex tasks [36,37]. Indeed, it was argued that numerical discriminations are more cognitively demanding than other types of discrimination and certainly they imply a greater number of steps [38]. A numerical discrimination requires that, in each trial, subjects estimate the first quantity, then the second, and then compare the two amounts. In contrast, color discrimination requires only learning to avoid one color or to choose the other, with no comparison needed after the first trials. A second hypothesis regards the time allowed by the two approaches to subjects for collecting information and issue the correct response. In traditional manual training experiments the subject slowly approaches the stimuli, with plenty of time to make a decision; in Skinner tanks, subjects performed trials in quick succession, with less time to process information, take a decision and, if necessary, to correct a wrong decision [30]. Finally, the automated device for fish was designed, as the original Skinner box, to allow testing several subjects, which need consequently to be moved daily from their home tank to the apparatus. While rats and pigeons adapt very quickly to being manipulated and introduced into the Skinner box periodically, fish may perceive a threat and, therefore, reduce the attention paid to the task. In addition, in fish frequent manipulation may produce chronic stress with a consequent negative impact on cognitive performance [39,40,41].The first experiment of this study aimed to evaluate the influence of the training approach on numerical tasks performance under controlled conditions. Training done with our automated device was benchmarked against a standardized manual training procedure [31,42,43] using the same strain, the same stimuli, the same numerical ratios and adopting in both cases the protocol used by Bisazza and colleagues [23].The fact that automated devices appear very effective in certain types of tasks but unpredictably inefficient in others, makes this approach of operant conditioning unreliable for fish cognition research. In the second part of the study, we tested five variants of the automated procedure, in the attempt to improve its effectiveness (experiments 2–5). We modified some features of this approach to training that we hypothesized to be responsible for its reduced effectiveness in numerical tasks. In particular, we elongated the decision time, increased inter-trial interval, removed internal partitions to increase visibility of stimuli, allowed subjects to reside in the conditioning chamber, and attempted to reduce the cognitive load by uncoupling learning how to use the conditioning chamber from learning the numerical discrimination.2. Materials and Methods2.1. Animal HousingWe used adult female guppies derived from the ornamental strain known as ‘snakeskin cobra green’ that is regularly bred in our laboratory at the Department of General Psychology (University of Padova). We tested subjects of the same sex to avoid the additional confounding effect of sex differences and to have a more homogeneous sample. The guppies were maintained in mixed-sex groups of 30 individuals in 150 L tanks. Each tank was enriched with natural vegetation and a gravel bottom. The water temperature was kept at 27 ± 1 °C and was aerated via a biomechanical filter. A 30 W fluorescent lamp provided illumination for 12 h per day. The guppies were fed live nauplii brine shrimp (Artemia salina) and commercial flakes (AquaTropical, Isola Vicentina, Italy) twice per day. All fish used in the experiments were naïve to the experimental protocol. After completion of the experiment, all fish were released in other maintenance tanks and kept for reproductive purposes only.2.2. Determination of Sample SizeTo determine the sample size for the experiments, we calculated the number of subjects necessary to achieve 90% power at a two-tail significance level of p = 0.05. For the manual conditioning experiments, we fitted parameters derived from a previous study that investigated numerical discrimination using a very similar protocol [31] and for automated conditioning Experiment the parameters from a similar recent experiment [30]. The estimated sample size was n = 6 for the former and n = 9 for the latter. We conservatively used slightly larger sample size (8 and 11 respectively). In Experiment 2 we dropped sample size at six subjects as a very large number of fish were discarded in the pre-training phase, due to failure to learn to detour the transparent barrier. In Experiment 3, the number of subjects was increased to 10 since we had two slightly different methods of presenting stimuli as a between factor.2.3. Experiment 1: Comparison of Manual and Automated TrainingIn this experiment, we directly compared the recently developed automated operant conditioning procedure with the manual training procedure routinely used in our laboratory [42,43]. In both cases, the task consisted of numerical discriminations of progressively increasing difficulty, following the protocol of Bisazza and colleagues [23].The subjects of this experiment were 19 adult females, 11 trained with manual conditioning and eight with the automated conditioning procedure. Stimuli were made with Adobe Illustrator CC 2019 and consisted of sets of black dots of differing numerosity on a white background: 3 vs. 12, 2 vs. 3, 3 vs. 4, 4 vs. 5, and 5 vs. 6. Like other vertebrates, guppies can discriminate two quantities of objects using non-numerical attributes of the stimuli, such as cumulative surface area, density, and convex hull (the convex polygon that circumscribes all items) [44,45]. To prevent fish from using this alternative strategy, we controlled the stimuli for all the above-mentioned non-numerical variables. To control for cumulative surface area, we used dots of different diameters (range 0.75–0.95 cm): in one-third of the stimuli used, the ratio between the cumulative surface area of the smaller over the larger set was between 76 and 85%; one-third, between 86 and 95%; and in one-third between 96 and 105%. Furthermore, we varied the position of the dots in order to equate the convex hull in half of the trials and the density of the dots in the other half of the trials [23]. We used 18 pairs of stimuli for the 3 vs. 12 numerical discrimination and 24 for all others numerical discrimination. The stimuli used for 3 vs. 12 were controlled for density and convex hull, but not for the cumulative surface area.Stimuli were printed on laminated white cards (3 × 3 cm) for the manual conditioning procedure and were displayed on a monitor for the automated conditioning procedure.2.3.1. Apparatus and ProcedureManual ConditioningWe used an apparatus and procedure recently developed for other investigations on guppy cognition [42,43]. Subjects were tested individually in a 20 × 50 × 32 cm glass tank filled with 28 cm of water (Figure 1A).Eleven identical apparatuses were used at the same time and placed in a dark room. Each tank was in olfactory communication with an aquarium placed beneath the apparatuses that housed approximately 20 conspecifics; the water in all tanks was filtered by a silent filtering system. Two trapezoidal lateral compartments (10 × 6 × 32 cm) internally shaped each tank into an hourglass. These lateral compartments, made of transparent plastic, housed natural plants to provide an enriched environment for the subjects. Externally each wall of each tank was covered with green plastic to avoid any possible external influence during the experiment. One 15 W fluorescent lamp illuminated two adjacent apparatuses on a 12:12 h light:dark photoperiod schedule. All trials were recorded with video cameras placed above each tank. To present the stimuli to the subjects, each card was affixed to the end of a transparent panel (3.5 × 15 cm) with an L-shaped blocker that allowed us to fix the panel to the wall of the tank. During each trial of the training phase (see below for details), two transparent panels were simultaneously inserted on the same short wall of the tank.Following the protocol of previous studies [42,43], each subject underwent a habituation phase, a pre-training phase and a training phase. The habituation phase lasted two days, during which the guppies could familiarize themselves with the experimental apparatus. During this phase, fish were fed four times per day using a Pasteur pipette placed in alternating positions near the two short sides of the tank.The pre-training phase also lasted two days, during which guppies could familiarize themselves with the experimental procedure. During the first day of this phase, a white card was presented to the subjects eight times. Guppies performed four trials in the morning session and four trials in the afternoon session; the two sessions were divided by a 90-min interval, and consecutive trials were divided by 15-min intervals. When a subject approached the card, a Pasteur pipette was used to deliver a small quantity of food reward (i.e., reinforcement) consisting of a drop of live brine shrimp (A. salina). During the second day of this phase, the subjects performed a total of 12 trials (six in the morning session and six in the afternoon session) identical to the trials of the first day. To maintain a high level of motivation during the sessions, no other food was provided during the course of the experiment. The number of times the card was presented on each of the two short sides of the tank was evenly distributed in the trials and counterbalanced across each session on both days. We admitted to the training phase only those subjects that approached the card all 12 times in the second day of the pre-training phase. One subject that failed to achieve this criterion was not admitted to the experimental phase and was replaced with a new subject.In the training phase, each subject performed 12 trials per day for a maximum of 12 consecutive days for the 3 vs. 12 numerical discrimination and 10 days for all other discriminations. Specifically, the subjects performed six trials in the morning session and six in the afternoon session, with a 90-min interval between the sessions and a 15-min interval between each consecutive trial. The guppies were presented up to five different discriminations that corresponded to five different difficulty levels. All subjects were trained to select the larger numerosity. A choice was considered when the subject approached (swam at less than one body length) one of the two stimuli. To assess reliability of this measure, a subsample of the video-recorded trials of each subject was analyzed by a second observer who was blind to the experimental hypotheses. In all discriminations, when the subjects approached the correct stimulus first, they were given a food reward. If they approached the smaller numerosity first, no food reward was given and both stimuli were removed simultaneously.Each subject started with the 3 vs. 12 discrimination. Subjects were presented a more difficult discrimination (i.e., 2 vs. 3) if they met one of the two learning criteria. The primary learning criterion was defined as a rate of at least 75% correct choices (18/24) of all trials over two consecutive days (statistically significant using a binomial test). The secondary learning criterion was defined as a frequency of at least 60% correct choices (87/144 for the 3 vs. 12 discrimination and 72/120 for all others) over all trials (statistically significant using a binomial test). If they failed to reach one of the two criteria within 120 trials, the experiment ended. In the case of success, the same procedure with the same criteria was used to present the discrimination tasks for 3 vs. 4, 4 vs. 5 and finally, 5 vs. 6. The left/right position of the larger numerosity line and the short side of the tank on which the cards were presented over the trials were counterbalanced.Automated ConditioningFish were tested in a 12 × 16 × 10 cm conditioning chamber made of semi-transparent white plastic (0.3 cm in thickness). The internal compartment was uniformly illuminated by room light. The bottom of the chamber was made of transparent plastic and a camera was positioned 12 cm below the chamber to track the subjects during the experiment. The chamber (Figure 1B) was internally divided in a starting area (12 × 5.5 cm) connected through a corridor (4 × 2.5 cm) to a V-shaped decision area (12 × 3.5 cm), and two choice areas (6 × 4.5 cm). The distance between the exit of the starting area and the entrance to each choice area was 6. The two choice areas were separated by a suspended semi-transparent white plastic sheet. Each choice area presented a 6 × 5 cm window that allowed projection of stimuli via an LCD computer monitor (Samsung S19C450, Suwon, South Korea). An automatic feeder was placed above and between the two choices areas. The feeder consisted of a servomotor (Futaba S3305) connected to a transparent cylinder filled with A. salina decapsulated eggs (Shg Srl, Alessandria, Italy) held by two metallic rods. When appropriate (see details below), activation of the servomotor caused the vibration of metallic rods, and the release of 3–4 eggs as a positive food reward (i.e., reinforcement). The presentation of the stimuli, the tracking of the fish using the camera, and the activation of the feeder were simultaneously controlled through a Raspberry Pi (Raspberry Pi 3 Model B V1.2, 2015, Sony UK Technology Centre, Pencoed, UK) running custom-made Python software.One week before the experiment started, two subjects were randomly selected and moved into a 30-L aquarium provided with the same conditions as the maintenance tank. Each aquarium housed three immature guppies as social companions. The individual subjects were placed in the conditioning chamber only during the experimental sessions. Fish were identified based on their individual characteristics, such as coloration and tail shape. A pump connected to the housing tank and to the conditioning chamber served to ensure water exchange and to provide social odors to the subjects during the daily training sessions.Pre-training phase. During the pre-training phase, subjects were habituated to the conditioning chamber and learned how operate the automated system. We conducted two daily pre-training sessions between 10:00 h and 14:00 h, lasting 30-min each. Subjects were individually inserted in the conditioning chamber while the monitor projected a white background. When the subject spontaneously moved into the starting area, the monitor background changed from white to grey and the trial started. Once the subject entered one of the two choice areas, the automatic feeder delivered a small quantity of food reward and the monitor’s background changed from grey to white. A new trial did not start until the subject entered again in the starting area. There was no interval between trials. Each session ended when the subject obtained a maximum of 80 reinforcements or when 30 min had elapsed. The second daily pre-training session was conducted after a 2-h interval.During each pre-training session, the experimenter monitored the subject’s behavior using an LCD monitor connected to the camera. We admitted to the training phase only those subjects that consumed at least 30 food rewards in two pre-training sessions. Thirteen subjects failed to achieve this criterion within 12 pre-training sessions. These subjects were discarded and substituted with new subjects.Training phase. Daily training sessions lasted one hour and were administered between 10:00 h and 14:00 h for a maximum of 12 days. Subjects were individually inserted in the conditioning chamber while the monitor projected a white background. A pair of stimuli that differed in numerosity appeared once the subject entered the starting area. All subjects were trained to select the larger numerosity. The system randomly alternated the position of the correct stimulus between trials and was set to prevent three consecutive presentations of the correct stimulus in the same choice area. Once the subject chose the correct stimulus by entering the corresponding choice area, the system activated the feeder to release the food reward above the two choice areas. The stimuli then disappeared, and a new trial started when the fish spontaneously moved back into the starting area. If the subject entered the choice area corresponding to the incorrect stimulus, the monitor immediately displayed a black background, and no food reward was released. In this case, the same pair of stimuli was presented in the same left–right position (correction trials) and the correction trial was repeated until the subject chose the correct stimulus and received the reward. The interval between an incorrect choice and the subsequent correction trial was set to 10 s. These correction trials were not considered in the analysis, whether correct or incorrect, and only served to prevent the fish from developing left–right biases.The maximum number of reinforcements given per day was 80, and the training session ended automatically when the subject reached the maximum number or when 60 min had elapsed. The subject was then gently moved back in its maintenance aquarium. When a fish obtained fewer than 40 reinforcement, an additional training session was performed on the same day 2 h after the end of the first session. The additional training session was conducted using the same schedule as the first; however, the number of possible reinforcements changed based on the number of rewards the fish had already obtained, up to the 80 daily maximum reinforcements.Each day, we quantified the subjects’ accuracy as the proportion of correct choices by considering the first choice of each trial (correction trials were not considered). We used two learning criteria. The primary learning criterion was defined as 75% correct choices in two consecutive days. As in the manual procedure, a secondary learning criterion was defined as the frequency of at least 60% correct choices over a total of 12 days (statistically significant at the binomial test). Subjects that fulfilled one of the two learning criteria were presented with a series of more difficult numerical discriminations (each one lasted for a maximum of 10 days following the same training protocol).2.3.2. Statistical AnalysisWe performed the statistical analysis in RStudio version 1.2.5019 (RStudio Team, 2019; RStudio: Integrated Development for R. RStudio, Inc., Boston, MA, USA). We used a two-tailed statistical test; the significance threshold was set at p = 0.05, and descriptive analyses were reported as mean ± standard deviation.We analyzed subjects’ accuracy in four steps. The first step concerned the assessment of overall accuracy using evidence of learning the task for each numerical discrimination task. We analyzed the performance at the group level by comparing the percentage of correct choices to one expected at chance level (i.e., percentage of 50%) using a one-sample t-test. Since the subjects’ accuracy could be random in the first training session due to the lack of previous experience with automated conditioning, we analyzed the overall performance in the second half of the training session with the same approach. The second step concerned the assessment of individual accuracy using a binomial test on the number of correct choices and incorrect choices. The third step was focused on accuracy improvement over the training session. We analyzed subject performance as a repeated observation of binomial choices (i.e., correct and incorrect choices) in each training session using a generalized linear mixed-effects model with logit link function and binomial error distribution (GLMM, “glmer” function from the “lme4” R package). We fitted the model with the training session as a fixed effect, and subject (i.e., individual ID) as a random effect. The effect of the parameters was evaluated using the ‘Anova’ function of the ‘car’ R package. In Experiment 1, manual conditioning, the model was also fitted with numerical discrimination (i.e., 3 vs. 12, 2 vs. 3, 3 vs. 4, 4 vs. 5) as a fixed factor to evaluate difference in subjects’ accuracy among tasks.The fourth step concerned the comparison of subjects’ accuracy showed in the manual and automated conditioning procedures. We compared the overall performance by using a two-sample t-test. The learning performance was then evaluated using a GLMM, fitted with the training session and type of procedures as fixed factors, and the subject as a random factor.2.3.3. ResultsManual conditioning. Interrater reliability calculated for 936 trials was found to be very high (Cohen’s kappa coefficient κ = 1, p < 0.001). All 11 subjects reached the primary learning criterion of 18/24 correct choices in two consecutive days on the 3 vs. 12 discrimination (Figure 2). Table 1 shows the individual performance on each discrimination task with statistics. On average, the fish needed 55.64 ± 35.30 trials to meet the criterion (average number of days: 4.63 ± 2.94). Overall, subjects’ accuracy was greater than that expected at chance level (70.25 ± 6.99%; one-sample t-test: t10 = 9.604, p < 0.001).Ten out of 11 subjects reached the primary learning criterion in the 2 vs. 3 discrimination. On average, these fish needed 38.40 ± 21.01 trials to meet criterion (average number of days: 3.20 ± 1.75). The remaining subject (N10) stopped participating after 72 trials when its performance was statistically significant (Table 1). Overall, subjects’ accuracy was greater than that expected at chance level (73.96 ± 7.67%; t10 = 10.367, p < 0.001).Eight out of 10 subjects reached the primary learning criterion in the 3 vs. 4 discrimination. On average, these eight fish needed 60.00 ± 37.09 trials to meet the criterion (average number of days: 3.75 ± 1.83). Overall, subjects’ accuracy was greater than that expected at chance level (68.07 ± 9.66%; t9 = 5.918, p < 0.001).Two out of eight subjects reached the primary learning criterion and they needed on average 30.00 ± 8.49 trials (average number of days: 2.50 ± 0.71). One additional subject reached the secondary learning criterion. Overall, subjects’ accuracy was greater than that expected at chance level (60.76 ± 7.71%; one-sample t-test: t7 = 3.947, p = 0.006).None of these three subjects achieved the 5 vs. 6 discrimination according to the learning criteria. Overall, subjects’ accuracy in the 5 vs. 6 numerical discrimination was not greater than that expected at chance level (51.67 ± 0.83%; t3 = 3.464, p = 0.742).The overall analysis revealed a significant improvement in subjects’ accuracy over training session (χ21 = 5.371, p = 0.020), and a significant difference in accuracy among the four numerical discriminations (χ24 = 26.912, p < 0.001). The interaction was significant (χ24 = 10.512, p = 0.033), suggesting a different learning curve among tasks.Automated conditioning. No subjects achieved the 3 vs. 12 numerical discrimination according to the learning criteria. Overall, subjects’ accuracy was not greater than that expected at chance level performing on average 532.88 ± 180.91 trials (51.55 ± 4.38%; t7 = 0.998, p = 0.351). Even when the last six training sessions were considered, the subjects’ accuracy was no greater than that expected at chance level (258.00 ± 114.80 trials; 53.65 ± 5.01%; t7 = 2.061, p = 0.078). Table 2 shows the individual performance with statistics for all subjects.The GLMM revealed a significant improvement in the subjects’ accuracy over training sessions (χ21 = 8.448, p = 0.004).Manual vs. Automated conditioning. Overall, subjects’ accuracy was significantly different between Manual and Automated conditioning (two-sample t-test: t17 = 6.653, p < 0.001). The GLMM revealed a statistically significant improvement in the subject’s accuracy over training sessions (χ21 = 7.516, p = 0.006), a significant effect of the procedure (χ21 = 44.408, p < 0.001), and no session × procedure interaction (χ21 = 1.049, p = 0.306).2.4. Experiment 2: Elongation of Decision TimeIn this experiment, we increased the size of the conditioning chamber, increasing the distance between the corridor and the choice areas. Additionally, a transparent barrier was placed between the corridor and the choice areas so that the animal could not rush into them. These changes aimed to force the subject to take more time before the decision.2.4.1. Subjects, Apparatus and ProcedureFor this experiment, we used six adult females. The conditioning chamber was increased in size to 16 × 32 cm (Figure 1C). As previously described, it was internally divided in a starting area (16 × 16 cm) connected through a corridor (4 × 3 cm) to a V-shaped decision area (16 × 8.5 cm), and two choice areas (8 × 4.5 cm). As a consequence, the distance between the exit of the starting area and the entrance to each choice area was increased from 6 to 11.5 cm. To further increase the travel time necessary to reach the choice areas, a transparent barrier was placed between the V-shaped decision area and the choice areas so that the subject had to detour it in order to enter one of the two choice areas. The two choice areas were separated by a white plastic divider to prevent the fish from seeing the stimulus projected in the other choice area. Two automatic feeders were placed in correspondence with both areas. When the fish entered the choice area associated with the correct stimulus, the system activated the corresponding feeder to release a food reward. In this experiment, we modified the method to provide water exchange and social odors to the subjects. The conditioning chamber was placed inside a larger tank (20 × 50 × 32 cm) filled with 28 cm of water. The tank was provided with a gravel bottom, natural vegetation, a bio-mechanical filter, and it housed three immature guppies. As in Experiment 1, subjects were housed in a 30 L aquarium provided with the same condition as the maintenance tank and placed in the conditioning chamber only during the experimental sessions.During the pre-training phase, we habituated the subjects to detour a 7 × 10 cm transparent barrier. When the subject obtained at least 30 reinforcements in a pre-training session, we changed the transparent barrier for a wider one (9 × 10 cm) that was used during the training phase. Other details of the procedure were identical to those of Experiment 1 (automated conditioning). Stimuli were the same as those used in Experiment 1. Thirty-one fish did not pass the pre-training phase and were replaced with new subjects. The exceedingly large number of fish discarded in this experiment was due to a difficulty of most individuals to learn to efficiently detour the transparent barrier.Subjects’ performance was analyzed using the same statistical approach described in Experiment 1. We finally compared subjects’ performance between Experiment 1 (both manual and automated conditioning procedures) using a GLMM. The model was fitted with training session and type of procedure as fixed factors, and subject as a random factor.2.4.2. ResultsOne subject reached the primary learning criterion in the 3 vs. 12 numerical discrimination after 339 trials in 8 days. The remaining subjects did not achieve this discrimination after performing on average 502 ± 204.62 trials. Overall, subjects’ accuracy was not greater than that expected at chance level (51.47 ± 4.38%; t5 = 0.820, p = 0.449). Even when the last six training sessions were considered, subjects’ accuracy was not greater than that expected at chance level (52.87 ± 12.66%; t5 = 0.555, p = 0.603). Table 3 shows the individual performance with statistics for all subjects.The GLMM did not reveal an improvement in the subjects’ accuracy over the training sessions (χ21 = 0.608, p = 0.435). The performance of this discrimination did not significantly differ from Experiment 1 with an automated device (GLMM; procedure: χ21 = 0.673, p = 0.412; training session: χ21 = 3.092, p = 0.079; training session × procedure: χ21 = 5.909, p = 0.015), but was significantly lower than in the manual training experiment (procedure: χ21 = 20.239, p < 0.001; training session: χ21 = 0.638, p = 0.425; training session × procedure: χ21 = 0.005, p = 0.945).The subject tested in 2 vs. 3 discrimination, failed to achieve the task (overall: 51.94 ± 6.40%; t9 = 0.960, p = 0.362).2.5. Experiment 3: Removal of Internal PartitionsIn Experiment 3, we built a new conditioning chamber from which all internal dividers had been removed, so to increase the visibility of the stimuli. (Figure 1D). In this experiment we also compared two different spatial arrangement of the stimuli presented to the subjects.2.5.1. Subjects, Apparatus, and ProcedureFor this experiment, we used 10 adult females. Stimuli were the same as those of Experiment 1 (automated conditioning) but were presented in two different ways. To half of the subjects, we presented stimuli as in previous experiments, always arranged in the same spatial position except for left-right alternation (Condition A). To the remaining subjects (Condition B), we varied the horizontal and vertical position of the two stimuli and the distance between them, obtaining 18 different positions, the same used in two previous studies [23,46]. Stimuli were presented in these positions in rotation. This procedure is thought to minimize the development of left-right side bias and to favor discrimination learning [23,47].We used a modified version of the apparatus adopted in Experiment 2, consisting of a 16 × 32 × 10 cm conditioning chamber (Figure 1D). Internally, two white plastic walls divided a 16 × 10 cm starting area from the rest of the chamber. A feeder was positioned in correspondence with each choice area (8 × 4.5 cm) and a window allowed the presentation of the stimuli. The distance between the exit to the starting area and the entrance to each choice areas was increased to 17.5 cm. When the subject entered the choice area associated with the correct stimulus, the system activated the corresponding feeder to release a food reward. Other details of the procedure were identical to those of Experiment 1 (automated conditioning). Seventeen subjects did not pass the pre-training phase and were replaced with new subjects.Statistical analysis was the same as that for to previous experiments. Learning performance was evaluated using a GLMM fitted with training session and experimental condition (i.e., stimuli presentation) as fixed effects, and subject as a random effect. We compared subjects’ performance between Experiment 1 and Experiment 3 using a GLMM. In this analysis the subjects of the two conditions in Experiment 3 were considered as a single group. The model was fitted with training session and type of procedure as fixed factors, and subject as a random factor.2.5.2. ResultsNo subjects achieved the 3 vs. 12 numerical discrimination according to the criteria (average number of trials: 390.00 ± 74.48; Table 4). There was no difference between the two conditions (t8 = 0.755, p = 0.472), and the two conditions were pooled for further analysis. Accuracy was significantly greater than that expected at chance level (51.707 ± 2.18%; t-test: t9 = 2.478, p = 0.035). Table 4 shows the individual performance with statistics for all subjects.The GLMM did not reveal an improvement in subjects’ accuracy over training sessions (χ21 = 0.807, p = 0.369), nor effect of condition (χ21 = 0.456, p = 0.499), nor interaction (χ21 = 0.938, p = 0.333).The performance of this experiment did not significantly differ from that of Experiment 1 with automated device (GLMM; procedure: χ21 = 0.010, p = 0.921; training session: χ21 = 1.921, p = 0.166; training session × procedure: χ21 = 7.574, p = 0.006), but was significantly inferior to the performance in the manual training experiment (procedure: χ21 = 43.374, p < 0.001; training session: χ21 = 1.067, p = 0.302; training session × procedure: χ21 = 0.039, p = 0.844).2.6. Experiment 4: Subjects Resident in the Conditioning Chamber and Manipulation of the Inter-Trial IntervalIn all previous experiments, fish were housed in a maintenance tank and individually transferred to the conditioning chamber for the training session. The experimental manipulation may have stressed the subjects and, consequently, affected their learning abilities. In this experiment, guppies were tested in their home tank. The conditioning chamber was built inside a large aquarium and each subject was maintained in the conditioning chamber for the whole experiment.2.6.1. Subjects, Apparatus, and ProcedureFor this experiment, we used 16 adult females. We used the same set of stimuli adopted for Experiment 3 arranged in different vertical and horizontal positions. For this experiment, we built a 20 × 32 × 32 cm conditioning chamber inside a 20 × 50 × 32 cm glass tank, the same type of tank used for the manual training (Figure 1E). The chamber was internally divided into a starting area (20 × 10 cm), a corridor (5 × 2 cm), a V-shaped decision area (20 × 16.5 cm), and two choice areas (4.5 × 10 cm) so that, in this version the shape and the size of the various areas were very similar to the corresponding areas in the manual training apparatus. A feeder was positioned in correspondence with each choice area, and a window allowed the presentation of the stimuli. The distance between the exit of the starting area and the entrance to each choice area was increased to 18.5 cm. The remaining part of the glass tank was supplied with natural vegetation and a gravel bottom, a bio-mechanical filter, and it housed, for the whole experiment, three immature guppies as social companions. A transparent plastic partition furnished with a series of holes separated the conditioning chamber from the area containing social companions. At the end of each experimental session, a male was inserted as a social companion into the conditioning chamber and removed the following day, one hour before the beginning of the daily session.Experiment 4 was, in fact, composed of two distinct experiments which shared the same apparatus. In Experiment 4A the temporal schedule of the trials was the same as that described for Experiment 1 and each fish daily underwent a one-hour session with a maximum of 80 trials. In Experiment 4B we modified inter-trial interval and the overall temporal schedule of the experiment. Each daily training session lasted 8 h and the interval between the two correction trials was increased from 10 s to 10 min. The session terminated if the subject reached 80 reinforced trials. This modification aimed to prevent subjects from developing the strategy of producing a rapid series of random responses to get a 50% on rewarded trials. The change also had the effect of spacing out trials over time, making the temporal schedule more similar to that of manual training procedure. To habituate the fish to an 8-h training session, one day of pre-training was added, in which subjects underwent four pre-training sessions of increasing duration, 15, 30, 45, and then 60 m, separated by 90-min intervals. Other details of the procedure were identical to that of Experiment 1 (automated conditioning). Five subjects in Experiment 4A and two subjects in Experiment 4B did not pass the pre-training phase and were replaced with new subjects.2.6.2. ResultsIn Experiment 4A, six out of eight subjects achieved the 3 vs. 12 numerical discrimination according to the primary learning criterion (average number of trials: 291.00 ± 82.34; average number of days: 5.17 ± 1.47) and one additional subject reached the secondary learning criterion (Table 5). A subject did not achieve this discrimination after performing 597 trials. No subject achieved the 2 vs. 3 discrimination after performing on average 495.14 ± 135.10 trials.In Experiment 4B, five out of eight subjects achieved the 3 vs. 12 numerical discrimination according to the primary learning criterion (average number of trials 184.00 ± 33.96; average number of days: 5.20 ± 1.64; Table 6). Three subjects did not achieve this discrimination after performing on average 329.67 ± 5.69 trials. Only one out of five subjects reached the primary learning criterion in the 2 vs. 3 discrimination, after 160 trials in two days. This subject failed to achieve the 3 vs. 4 discrimination (47.59 ± 6.64%; t9 = 1.147, p = 0.281). The remaining four subjects did not achieve the 2 vs. 3 discrimination after performing, on average, 336.25 ± 110.07 trials.To examine the influence of inter-trial interval (10 s vs. 10 min) we compared accuracy between Experiment 4A and Experiment 4B. In the 3 vs. 12 discrimination, we found no statistically evidence of an effect of time elongation (t10 = 0.285, p = 0.780) and the subjects of the two experiments were pooled for further analysis. Subjects’ accuracy was significantly greater than that expected at chance level (59.88 ± 6.95%; t15 = 5.689, p < 0.001). The GLMM revealed a significant improvement in the subjects’ accuracy over training sessions (χ21 = 20.948, p < 0.001), no effect of condition (χ21 = 0.030, p = 0.863). The session × condition interaction was significant (χ21 = 6.844, p = 0.009), due to the fact that subject with the standard procedure (Experiment 4A) had a steeper learning curve than subjects which had a 10 min delay after each incorrect trial (Experiment 4B). The performance was significantly higher than that of Experiment 1 with automated device (GLMM; procedure: χ21 = 18.474, p < 0.001; training session χ21 = 29.835, p < 0.001; training session × procedure χ22 = 8.775, p = 0.003), but did not reach the performance of manual training (χ21 = 12.190, p < 0.001; session χ21 = 25.220, p < 0.001; session × procedure χ21 = 4.817, p = 0.028).In the 2 vs. 3 discrimination, we found no statistically evidence of an effect of time elongation (t4.455 = 0.747, p = 0.493) and the subjects of the two experiments were pooled for further analysis. Accuracy was not greater than that expected at chance level (54.48 ± 10.03%; t11 = 1.546, p = 0.150). When the last six training sessions were considered, subjects’ accuracy was not greater than that expected at chance level (53.28 ± 9.62%; t11 = 1.179, p = 0.263).2.7. Experiment 5: Reducing Cognitive LoadThe automated procedure may increase the cognitive load of the task, thereby reducing fish learning performance, by requiring the fish to first attain a set of other skills (i.e., learn how to swim through the tank sectors, to approach the stimuli to cause food delivery, swim to the back chamber to prepare a new trial, etc.). In the present experiment, before subjecting guppies to the numerical task, we allowed them to become acquainted with these additional skills by giving them a series of three different non-numerical (shape, color, and size) discrimination tasks.2.7.1. Subjects, Apparatus, and ProcedureFor this experiment, we used eight adult females. The conditioning chamber was identical to those used in Experiment 1. Subjects initially underwent a series of three visual discriminations: the first task required fish to discriminate between a circle (diameter 3 cm) and a 3 × 3 cm triangle on a white background; the second task was a color discrimination between red (red-green-blue color model: 255, 0, 0) and green (RGB: 0, 255, 0); and the third task was a size discrimination between a 3 × 3 cm square and a 1.5 × 1.5 cm square on a white background (ratio: 0.25). Fish that successfully completed this sequence underwent a fourth discrimination that was the same numerical discrimination 3 vs. 12 presented first in Experiment 1. We originally planned to continue the experiment by increasing the difficulty of the numerical task, as in previous experiments. However, as subjects appeared less motivated to complete the task, we decided to skip the last part the experiment and perform instead a second shape discrimination (different from the first one) to compare learning at the beginning of the experiment and after two months. The stimuli were a cross (3 cm in length and 0.75 cm in width) and a 3 × 1.5 cm horizontal bar. The learning criteria and other details of the procedure were identical to those adopted in Experiment 1. Thirteen subjects did not pass the pre-training phase and were replaced with new subjects.We followed the same statistical approach described in previous experiments to analyze subjects’ accuracy. We additionally compared subjects’ performance between the first shape discrimination and the second shape discrimination using a GLMM. The model was fitted with training session and type of discrimination as a fixed factor, and the subject as a random factor.2.7.2. ResultsFive out of eight subjects achieved the shape discrimination according to the primary criterion after performing on average 225.60 ± 156.19 trials (average number of days: 3.20 ± 2.17), whereas the remaining subject did not achieve the first discrimination (average number of trials: 435.67 ± 94.94; Table 7). Overall, the group of subjects did not show an accuracy greater than that expected at chance level (62.43 ± 15.88%; t7 =2.215, p = 0.062). Table 7 shows the individual performance with statistics for all subjects. The GLMM did not reveal a significant improvement in the subjects’ accuracy over training sessions (χ21 = 0.370, p = 0.543).All five subjects achieved the color discrimination according to the primary criterion after performing on average 416.00 ± 266.64 trials (average number of days: 5.60 ± 3.78). Overall, subjects’ accuracy was significantly greater than that expected at chance level (68.52 ± 13.29%; t7 = 3.115, p = 0.004). The GLMM revealed a significant improvement in the subjects’ accuracy over training sessions (χ21 = 43.416, p < 0.001).These 5 subjects also achieved the size discrimination according to the primary criterion after performing on average 373.20 ± 251.21 trials (average number of days: 4.60 ± 3.21). Overall subjects’ accuracy was significantly greater than expected by chance level (75.32 ± 10.15%; t4 = 5.580, p = 0.005, Table 7). The GLMM revealed a significant improvement in the subjects’ accuracy over training sessions (χ21 = 10.594, p < 0.001).Two out of five subjects achieved the 3 vs. 12 numerical discrimination according to the primary criterion after performing on average 564.00 ± 137.18 trials (average number of days: 7.5 ± 2.12), and two additional subjects reached the second learning criterion after performing on average 795.00 ± 7.07 trials. The remaining subject did not achieve it after 596 trials. Overall, subjects’ accuracy was significantly greater than that expected at chance level (60.29 ± 4.80%; t4 = 4.790, p = 0.009, Table 7). The GLMM revealed a significant improvement in the subjects’ accuracy over training sessions (χ21 = 30.408, p < 0.001). The performance on this discrimination was significantly higher than that found in Experiment 1 with automated device (GLMM; procedure: χ21 = 18.474, p < 0.001; training session: χ21 = 29.835, p < 0.001; training session × procedure: χ21 = 8.775, p = 0.003), but did not reach the performance of manual training (procedure: χ21 = 12.190, p < 0.001; session: χ21 = 25.220, p < 0.001; session × procedure χ21 = 4.817, p = 0.028).At this point in the experiment, one subject died, and the others did not appear very motivated to complete the test. We decided to subject the four remaining guppies to a new shape discrimination to determine whether, after nearly two months of experiments and more than two thousand trials, the performance of the subjects had declined. Only one out of four subjects achieved the second shape discrimination according to the primary criterion after 640 trials in 8 days, whereas the remaining subject did not (average number of trials: 746.67 ± 94.94). Subjects’ accuracy was not greater than that expected at chance level (60.20 ± 6.89%; t3 = 2.961, p = 0.060, Table 7). The GLMM did not reveal a significant improvement in the subjects’ accuracy over training sessions (χ21 = 0.800, p = 0.371). A comparison of this task and the shape discrimination at the beginning of this experiment revealed a significant effect of the treatment (GLMM, task: χ21= 10.187, p = 0.001), indicated that subjects showed a higher performance in the first shape discrimination task (Figure 2). There was no significant effect of the session (χ21 = 0.016, p = 0.899), nor the interaction (χ21 = 1.624, p = 0.203).3. DiscussionIn recent years, there have been numerous attempts to produce automated operant conditioning devices for small teleosts. A comparison of studies employing traditional training methods and studies employing automated devices seems to suggest that the latter enhance performance on some tasks [21,28] but worsen it on others [27,30,32]. Using a direct comparison of the two methods, we demonstrated that, when all the parameters of the tasks are controlled for, automated methods reduce the subjects’ performance on a numerical discrimination task. However, we showed that residency and reduction of cognitive load moderately improved the subjects’ performance.3.1. Comparison of Manual and Automated TrainingIn Experiment 1, we compared in the same numerical task subjects trained with a recently developed automated device [28,30] and subjects trained with a traditional approach, a manual operant conditioning protocol that was uses in many previous studies [31,42,43]. We used the protocol that was previously found to be the most efficient in training fish on a numerical discrimination [23]. The subjects were first tested on an easy numerical contrast and the difficulty of the task progressively increased until the fish failed to reach the criterion. This protocol is ideal for making comparisons between different conditions because it allows comparing both learning performance within a single numerical task and the numerical acuity estimated by administering task of increasing difficulty. All of the subjects manually trained achieved the easiest discriminations, 3 vs. 12 and 2 vs. 3 items. Additionally, the two following, more difficult, discriminations were achieved by several subjects (3 vs. 4: 9 out of 11 subjects; 4 vs. 5: 3 out of 8 subjects). The 5 vs. 6 discrimination was not achieved by any subject. This result completely overlaps that obtained in the previous study with a different manual training protocol [23] and could not be attributed to an observer bias effect, as showed by a very high interrater reliability of scoring method. Conversely, no subject trained with automated device reached the criterion in 3 vs. 12 discrimination and only 3 out of 8 fish had a percentage of correct responses significantly above chance. It is worth noting that, with traditional training procedure, guppies were allowed a maximum of 120 trials for each discrimination, and, in most cases, they reached the criterion after a few dozen trials, whereas with the automated device they were allowed to perform up to a thousand trials. The poor numerical performance with automated training procedures observed in a previous study [30] is, thus, confirmed even with the adoption of a most efficient protocol, which starts from simple discriminations to gradually increase the difficulty. We can exclude that the low performance in the Skinner box is merely due to the inadequacy of the apparatus or the procedure since with the same equipment and procedure guppies rapidly reach up to 90% correct response in other types of discrimination, such as color and shape discrimination [28]. This performance is actually higher than previously reported in fish and also superior to that of many warm-blooded vertebrates. Since the proposed numerical tasks are easily discriminated by guppies trained with various non-automated protocols ([23,31], the present study), the cause of the poor performance must be sought in one or more characteristics that differentiate the two approaches to training.One important difference regards the time available to the subjects to gather information about the stimuli, analyze them and make a decision. Indeed, a trade-off between decision speed and accuracy was shown for a variety of species [48,49,50]. Skinner box are typically very reduced in size compared with manual training apparatuses [21,27,51,52]. Specifically, in our study, the distance between the exit from the corridor and the line of choice was 22 cm for the manual training and only 6 cm (approx. one body length) in the automated training. This resulted in a choice time on average three times longer for the manual procedure. The short time interval (~1 s) of the Skinner box may be enough for perceiving the color or the shape of two objects but not sufficient to make an accurate numerical discrimination. In a numerical discrimination, stimuli change for every trial and the subject must independently estimate the number of items in the two stimuli, compare these two quantities and thereby decide based on a learned rule. In addition, if number discrimination requires a longer processing time, the possibility of inhibiting and correcting initially wrong choices may be very reduced in the Skinner box due to the much shorter time allowed to the subject before it reaches the choice areas [30].An additional difference between color or shape discriminations and numerical discriminations is that the latter requires the fish to observe both stimuli in their entirety. A subject, for example, can perform a correct color discrimination even if one or both stimuli are partly hidden from view. Conversely, a larger of two numerosities may become the smaller one if some of its items are not visible. The tiny space of the automated conditioning chamber may cause a problem in visibility of the stimuli that was not present in the large tank used for manual training.Another possibility for explaining the different performances is that automated procedures increase the cognitive load of the task by simultaneously requiring that the fish learn to solve the discrimination, to swim fluidly through the sectors, to associate its approach to the stimuli with food delivery, and to swim back to the starting chamber to launch a new trial. This hypothesis is corroborated by the observation that in experiments with automated training procedures many subjects are discarded before the training phase starts due to their inability to learn how to operate the device ([28,32,53], the present study). In many vertebrates, including fish, the addition of concurrent tasks increases cognitive load and decreases the performance on the primary task [54,55]. This is expected to occur, particularly when subjects are performing complex tasks [36,37]. Although it is not clear whether numerical discriminations are more cognitively demanding for fish than color or shape discriminations [56], they certainly imply a larger number of cognitive operations. In a shape or color discrimination, a subject sees the same pair of stimuli throughout the experiment and can solve the task simply by learning to approach the positive stimulus or move away from the negative one without necessarily making a comparison of the two. Conversely, in a numerical discrimination task, the two stimuli change at each trial because position, size and density of the items vary systematically. In each trial, the subject must make a new estimate of the two quantities, compare the two numerosities and, hence, apply the learned rule (e.g., choose the larger one). It is plausible that the additional cognitive load related to the use of the automated device is particularly detrimental when a fish is required to perform such a task.Finally, the two procedures differ in the amount of manipulation received by the subjects during the experiment. As usual for conditioning protocols of fish (e.g., [43,57,58]), in the manual training experiment, guppies resided for the whole experiment in their experimental tank. In the automated training experiment, by contrast, they resided for most of the time in their home tank, and once or twice a day, they were netted and transferred to the conditioning chamber for the duration of the test. Netting and change in the environment, even for a short period, may cause severe and long-term stress in teleosts [59,60,61,62], and even chronic stress in case of repeated events [63]. Given the well-known negative impact of stress on cognitive performance [39,40,41], the manipulation required for the automated procedure may cause fish reduced discrimination learning success. Being in a new environment is also expected to cause short term increase in vigilance against predators, which typically reduces the attention devoted to foraging activities and decreases food-finding efficiency [54,55,64].Other differences between the two methods of training could potentially be involved. For example, the number of trials per sessions or the presence of correction trials have been sometimes found to affect learning rate in other organisms [65,66]. However, they do not normally have dramatic effects on task achievement. In addition, an automated procedure with all these features led to performance equal or even superior to the traditional methods with color or shape discrimination [28] and with the discrimination of the size of stimulus (see Experiment 5) and it is difficult to devise hypotheses that easily explain why these features should selectively compromise learning in numerical tasks.3.2. Can Automated Training Devices Be Improved?Following the above hypotheses, in the second part of this study we tested four modified versions of the setup to see whether we could detect the key factors and reduce the gap in numerical discrimination performance between the automated training approach and the other methods.Elongation of decision time. The first variant was intended to force guppies to take a longer time interval before making their choice (Experiment 2). The size of the conditioning chamber was increased, almost doubling the distance between the corridor and choice areas, and a transparent barrier was placed between the corridor and the choice areas to further lengthen the distance to reach the stimuli. Only one out of the six subjects learned to discriminate the 3 vs. 12 numerical discrimination, but this subject failed the subsequent 2 vs. 3 discrimination. These results suggest that the lengthening of the decision time, at least as it was induced here, is not enough to improve performance.Removal of internal partitions. In Experiment 3, we further modified the conditioning chamber of the previous experiment by removing all internal dividers so that the animal had a complete view of the two stimuli from afar, as happened in the manual training tanks. None of the ten subjects tested in this experiment learned the 3 vs. 12 discrimination, suggesting that the visibility of the stimuli was probably not a crucial factor in determining the difference between the two training approaches of Experiment 1.Subjects resident in the conditioning chamber. In Experiment 4, we modified the chamber so that we could keep the experimental subject there for the entire duration of the experiment, as was the case in manual training experiment. With this modification, 12 out of the 16 subjects achieved the 3 vs. 12 numerical discrimination, a considerable improvement on the setup of Experiment 1. Fish likely adapt less well to being moved daily between the home tank and the experimental chamber, relative to domestic species of mammals and birds, such as rats and pigeons for which the original Skinner box was developed. In the rainbowfish, Melanotaenia duboulayi, familiarity with the testing apparatus was found to increases task performance probably because familiarity decreases stress, allowing subjects to pay more attention to the task [67]. This hypothesis may also explain the unexpectedly low performance in numerical tasks found in a study in which guppies undergo a traditional manual training but were moved in turn to the conditioning chamber for the trial [44].In order to allow the subject to reside in the apparatus, in this experiment we had to increase the size of the conditioning chamber. Therefore, in this experiment the size of the chamber is a potential confounding factor. One might for example argue that guppies are not at ease in tiny spaces and would be stressed by being tested in small apparatuses. However, in their natural habitat, guppies avoid open spaces and are generally found in the proximity of the river’s margins, in small pools that contain a fraction of the volume of our Skinner box and they often forage in small gaps of thick vegetation that can host just one fish [68,69,70]. As a further confirmation of the fact that the size of the chambers per se is unlikely to represents a key factor, excellent performance was obtained with the smallest version of the conditioning chamber in other tasks [28], and some studies have found good numerical performances using very small manual training apparatuses (e.g., [71]). It is worth noting the global performance of fish resident in the conditioning chamber was still inferior to the manually trained group of Experiment 1. Being resident in the test apparatus is not the only factor that determines the difference between the two approaches of training used here.One of the advantages of the Skinner box was that a single small-sized equipment could be used to train several rats or pigeons in rotation. Maintaining the fish in the conditioning chamber improves performance but suffers the major drawback that it implies the use of a large tank for each experimental subject and therefore experiments require much more space and equipment. A compromise may be the one adopted in a previous study [30], in which the subjects are semi-residential, i.e., they live in a tank that houses the conditioning chamber to which they are moved in turn at the time of the test. In this way, the relocation is rapid and the subject stays in identical physical, chemical and olfactory conditions for the duration of the experiment. A rough comparison between the results of the two studies suggests that this solution is sufficient to achieve some improvement. In our Experiment 1, only 3 of the 8 subjects had a performance greater than that expected by chance; this happened for all subjects but one in the previous study, though in the latter, the ratios to discriminate were more difficult. Indeed, the performance significantly differed between the two studies (Mann–Whitney U test; p = 0.045). It is also possible to devise a conditioning chamber in which the subjects are accustomed to leave their group in turn, and swim to the testing area at the time of test, a common practice in primate studies (e.g., [72,73,74]).The results of Experiment 4 have potential implications for the welfare of the laboratory studies in fish. Moving repeatedly the subject forth and back between home cage and the testing apparatus is a common practice of cognition studied in fish, as it is in mammals and birds (e.g., [4,18,75]). However, this practice could induce stress in fish more than in other vertebrates, especially if the experiment is prolonged, as also suggested by the outcome of Experiment 5 (see below).In this experiment we also manipulated inter-trial interval, another factor that frequently differentiates automated conditioning approach from traditional ones and that also differed considerably in Experiment 1. We increased the inter-trial interval in half of the subjects from 10 s to 10 min, making it very similar to the inter-trial interval adopted in the manual training experiment, but we found no evidence that this treatment affected discrimination learning or numerical acuity of the subjects.Reducing cognitive load. The aim of Experiment 5 was to reduce the cognitive load during the execution of the task by uncoupling learning the use of the conditioning chamber from learning the numerical discrimination. Before being admitted to the numerical discriminations, guppies underwent a shape, a color, and a size discrimination. The performance on these three tasks was substantially similar to that observed in previous studies employing different training procedures [23,29]. When subjected to the 3 vs. 12 numerical discrimination, guppies showed an evident improvement compared to the automated training of Experiment 1, although their performance was significantly lower than in the experiment with manual training.At the end of this series of tasks, after nearly two months of experiments and more than two thousand trials, one animal died, and others showed signs of stress. Therefore, we decided to stop the experiment and we do not know the quality of the performance of the fish of this treatment on the more challenging numerical discriminations. This outcome tells us that, unlike monkeys and mice, fish cannot be tested for unlimited time durations, at least not if it is necessary to move them frequently from their home tank to an experimental chamber. A solution to the problem of moving fish may be to have the subject reside in the experimental chamber as in Experiment 4, in which this solution seems to have brought benefits in terms of performance. It is necessary to verify whether guppies that are resident or semi-resident in the conditioning chamber can do a thousand trials without suffering or diminishing their performance.The first part of Experiment 5 confirms previous finding about the efficiency of our automated training device for discrimination learning [28,32]. Considering the first days of training in each discrimination, before the best performing subjects were admitted to the next discrimination, guppies evidenced an excellent performance in color, shape and size discriminations reaching 75–80% correct choices in few sessions. It is interesting to note that the first part of this experiment also shows that guppies, like mammals and birds [76,77], can learn a series of different discriminations in rapid sequence without apparent interference of one task on the subsequent.3.3. A Comparison with Other Studies Using Automated Conditioning DevicesAs mentioned in the Introduction, learning, memory, and many other cognitive functions are investigated in warm-blooded vertebrates using primarily automated procedures whereas manual training is the standard procedure used with teleost fish, with other lower vertebrates and with invertebrates. Some laboratories have attempted overcome this limitation by developing a Skinner-box apparatus for fish. In the 1960s and 1970s, many studies investigated learning and memory in the goldfish using fully automated operant conditioning devices with the aim to compare a fish species with the classical mammalian and avian model species [78,79,80]. When discrimination learning experiments were performed, goldfish appeared to master well color [79,80,81] as well as shape discrimination [82,83].In recent years, some small tropical fish, in particular zebrafish, medaka and guppy, have become important models in neurobiological research. Taking advantage of the emerging digital technologies, there have been half a dozen independent attempts to implement automated devices, obtaining mixed results. Manabe and colleagues [18] set up a computer-controlled operant apparatus for small fish, in which stimuli were presented by means of LEDs and the approach to response key was sensed through an optical fiber sensor, which triggered an automated feeder delivering small amount of food. With the same apparatus, Kuroda and colleagues [19], using a color discrimination, demonstrated that zebrafish can learn a reversal learning task.Color discrimination is, by far, the most frequently investigated discrimination task in the remaining studies that employed automated training approach. Another device allowed for the simultaneous training of six subjects, each resident in an adjacent chamber. A computer tracked the position of each fish, controlled stimulus presentation on a LCD monitor and commanded the delivery of the food reward [21]. The apparatus was tested by training zebrafish on a blue-green color discrimination. Subjects slowly improved their performance over 30 daily sessions of 20 trials each, reaching 80% of correct choices at the end of experiment. Conversely, Miletto Petrazzini and colleagues [27], who tested zebrafish with a commercially available operant conditioning apparatus, found a much lower performance on two color discriminations. With the exception of the latter study, the others were in agreement with the results obtained with our device in the present study or in previous ones [28,32].Two studies conducted in our laboratory investigated shape discrimination. Guppies showed good learning performance, fully comparable with the results of manual training experiments [28]. Conversely, with zebrafish, no subject reached the criterion on the same shape discrimination task, even after extended training [32], despite this species showing 75% accuracy in color discrimination tasks with the same device. There is scarce information about the discriminative abilities of this species except for color discrimination. Zebrafish proved able to discriminate novel shapes from familiar ones [34,35]. However, the sole study that examined shape discrimination in zebrafish using appetitive conditioning was an interspecific study in which zebrafish showed a rather poor performance compared to other teleosts [25].To date, only another laboratory investigated numerical discrimination in a fish, Gambusia affinis, using an automated system [33]. The device and the procedure differed from those used in the other studies. During the training phase, the quantities to discriminate were presented on two monitors placed at the opposite ends of a rectangular tank and a computer-operated system delivered a food reward when the subject approached the correct stimulus. Learning was measured in a probe trial in which the subject was exposed to the stimuli without reinforcement, and the percentage of time spent near the positive stimulus was calculated from video-recordings. This experiment replicated a study on the same species with an identical procedure, but in which the training phase was done manually [84]. A rough comparison of the two studies shows that, even in this case, the performance on numerical discrimination tasks was better when fish were trained manually (p = 0.016; Welch t-test).4. ConclusionsIn summary, the results of this study confirm that the automated training device we developed modelling the classical Skinner boxes can satisfactorily be used to train guppies in some tasks (i.e., color, shape, and size discriminations) but are totally inadequate for other tasks, such as a numerical discrimination. Similar inconstancy in performance have been reported with other automated devices in different fish species and with different tasks which suggests that small laboratory teleosts may be limited in their capacity to cope with some undetermined aspects of the automated approach to training. These limitations need to be overcome, as they presently prevent an easy and straightforward comparison of teleosts with the other vertebrates.In a series of experiments, we introduced modifications to the automated training apparatus and procedure in an attempt to fill the performance gap with the best performing manual training procedure available for guppies. The difference did not appear to be related to a longer decision time in manual procedure nor to the different visibility of the stimuli in the two procedures. Being resident in the test tank improved the performance of the subjects but this factor alone did not lead to the same performance of the manual training experiment. A similar effect was obtained by reducing the cognitive load through the temporal dissociation of the familiarization with the functioning of the automatic equipment from the numerical discrimination task.It, therefore, seems that there is no single factor that explains the different efficiency of the two procedures tested in this study, but rather several factors acting synergistically to determine the different performance. There are many other differences between the two methods compared in this study, and in future, it will be necessary to investigate other factors that may be important. One interesting difference that was not addressed concerns the way the stimuli are presented in the two approaches, generated on a computer screen or stuck onto real objects introduced into the water. Computer-generated stimuli have already been used successfully in fish for visual discrimination tasks [28,85,86] and a vast literature shows that fish react to stimuli presented on the monitor as they react to the real objects [87,88,89,90,91]. However, it is possible that stimuli introduced into the tank are much more salient and focus the subject’s attention on the task. Indeed, in several species the salience of the stimulus was shown to influence the performance on discrimination tasks (capuchin monkeys [92]; pigeons [93]; keas [94]) and there is also partial support for this effect in the guppy [46]. It will be interesting to verify whether an automated procedure that commands the presentation of solid stimuli introduced into the subject’s tank allows for a performance comparable to that we obtained in our study with the reference manual training procedure. | animals : an open access journal from mdpi | [
"Article"
] | [
"automated conditioning",
"fish cognition",
"learning constraints",
"numerical discrimination",
"Poecilia reticulata",
"Skinner box"
] |
10.3390/ani13101602 | PMC10215146 | The aim of this study was to characterize the RHU competencies according to the distance (short vs. long), causes of deaths, and associated risk factors. The studied population comprised 16,856 horses that participated in RHU rides from 2007 to 2018. During the entire period, there were 99 fatalities. The percentage of inexperienced horses and those who completed the ride was greater in short races than in long races. In both types of rides, more horses died during than after the ride, and inexperienced horses were more likely to be dead than horses with prior experience in the sport. Short rides were associated with increased risk of sudden death, while long rides were associated with increased risk of death due to metabolic alterations. | RHU is the oldest endurance sport in Uruguay. However, despite 80 years of racing, there are no studies to characterize this type of competition, explore rates and causes of death, and identify the associated risk factors. The aim was to characterize the Raid Hipico Uruguayo (RHU) competencies according to the distance (short (SR, 60 km) vs. long (LR, 80–115 km)), the causes of deaths, and the associated risk factors. The study population comprised horses (n = 16,856) that participated in RHU rides from 2007 to 2018. LR were more frequent than SR (p < 0.001). The average speed of winners was higher in SR (32.12 km/h) than in LR (28.14 km/h) (p < 0.001). There were 99 fatalities (5.9 per 1000 starts). SR had greater frequency of high comfort index (CI = Temp [°F] + Humidity [%]) than LR, and LR had greater frequency of low CI than SR (p < 0.001). The percentage of inexperienced horses and those who completed the ride was greater in SR than in LR (p < 0.001). In both types of rides, more horses died during than after the ride, and inexperienced horses were more likely to suffer fatalities than horses with prior experience in the sport (p < 0.05). SR were associated with increased risk of sudden death, while LR were associated with increased risk of death due to metabolic alterations. The high fatality index shown in this work warrants urgent investigation in this sport to minimize mortality associated with RHU-specific diseases. | 1. IntroductionEndurance equestrian sports have a long history, but it has experienced great growth in recent decades, mainly FEI endurance. As described by the FEI, “Horsemanship and Horse welfare are the core of endurance riding. Endurance is a test of the Athlete’s ability to manage the Horse safely over an Endurance course. It is designed to test the stamina and fitness of the Athlete and Horse against the track, distance, terrain, climate, and clock, without compromising the welfare of the Horse” [1]. Because metabolic disturbances and deaths occur more frequently than in any other type of equestrian discipline, all equine resistance sports have a strict veterinary control that ensures the health of equine competitors [1,2]. A series of veterinary inspections and examinations are established in the interest of the health, safety, and welfare of the horse in the competition. Only competitors whose horses have passed all the inspections and examinations are entitled to be classified in the final list of results [1,3].Elimination rates appear to have increased over recent years, which is a source of concern for the sport’s ethics and image. Main reasons for elimination are lameness and metabolic disturbances, associated with dehydration and electrolyte disturbances, and with substrate depletion in active muscle fibers. Moreover, there are severe consequences of this metabolic derangement, including heat stroke, rhabdomyolysis, colic, kidney and liver insufficiency, laminitis, and disseminated intravascular coagulation [2,4].The Raid Hipico Uruguayo (RHU) was the first endurance sport in Uruguay. This sport dates from 1944 and since then, rides have been regulated by the Uruguayan Equestrian Federation (FEU). It is the most popular and typical equestrian discipline of Uruguay, with some unique characteristics. RHU is considered as an endurance discipline. Research into veterinary problems in endurance horses is increasing, but there is almost no information available on endurance races not regulated by FEI. In theory, most of the information generated from FEI endurance races could be applicable to RHU horses. However, many of these practices were not useful in RHU horses, and most management, training and nutrition techniques come from empiric experience without any scientific basis. Something similar occurs with many other equestrian sports that are scientifically overshadowed by disciplines with common characteristics. However, small differences in the sports can produce large differences in metabolic and locomotor behavior, with strong impact on the athletic horse.Ride distances vary from 60 to 115 km and according to this, FEU has determined 2 categories: short rides (60 km) and long rides (80–115 km), all of which are divided in only 2 phases, being the first phase of long duration (2/3 of the total distance). Winning horses average speeds from 25 to 37 km/h, reaching top speeds of 50 km/h. The winning horse is the first to cross the finish line and meets subsequent veterinary requirements. For the rest of riders, the cut-off time of crossing the finish line is 45 min after the winning horse arrives [3].The breed of the horses competing in RHU was not officially registered until 2019. In an unpublished study from Brito et al., of 305 horses that raced in 1 season of RHU, 237 were crossbreeds, 39 thoroughbreds, 15 Anglo-Arabians, and 14 Arabians horses. Horses called crossbreeds mostly had more than 75% thoroughbred blood.The competitions take place on flat terrain with mostly hard surfaces, and the minimum weight of the riders should be 85 kg. Horses and riders can compete in any ride, regardless of their previous experience. Horses are examined by official veterinarians before the ride, after the first phase, and at the end. Veterinary control after the first phase is performed 20 min after the arrival of the horse, and once passed, horses must wait a 40-min compulsory rest period before starting the last phase. Horses are eliminated from the ride if veterinarians consider their metabolic status or orthopedic condition not to be adequate to enable them to continue the ride [3]. Sixty-one percent of the participants do not finish the ride due to lameness or metabolic reasons [5], which can sometimes lead to fatalities.From the point of view of health and welfare, the death of equines during or after competition is a major concern for vets, riders, organizers, and for spectators. Studies on the causes of death in sports horses are scarce and mostly refer to racehorses, being the main causes: sudden death [6,7] and catastrophic musculoskeletal injury [8,9].Fatalities during endurance exercise are recognized as a consequence of prolonged exercise, but data documenting incidence and causes are very limited. Balch et al. (2019) reported 127 fatalities out of 335,456 starts (0.28 fatalities per 1000 starts) during the period 2002 to 2018 [10]. According to Balch et al. (2019), 77% of deaths were attributed to the high demands of endurance exercise (leading to severe muscle cramping and exhaustion, mostly attributable to acute abdominal pain) and 33% due to injuries not associated with the metabolic demands of endurance exercise (such as falling off a cliff or the trail, kick injury) [10]. In addition, the risk of death increased with the distance traveled (0.12, 0.35, and 1.48 fatalities per 1000 starts in rides of 48, 80, and 160 km, respectively) [10].Although this sport (RHU) has been carried out in Uruguay for several decades, there are no reports characterizing the ride conditions, as well as the causes of death and their risk factors. We hypothesize that although this sport has similarities with other endurance disciplines, it has very different characteristics that affect the causes of death during races. Therefore, the aim of this study was to characterize the RHU competencies according to the distance (SR vs. LR), the causes of deaths, and the associated risk factors.2. Materials and MethodsThe study was performed with the endorsement of the FEU, guaranteeing confidentiality regarding the names of the horses and owners.2.1. Data Collection and Studied VariablesThis was a retrospective cohort study. Data from all RHU rides from 2007 to 2018 were collected from FEU archives. All rides were contested annually between the months of March to December. The information obtained was entered into a computerized database, and each RHU ride was assigned a unique identification number. The database included the temperature and humidity, ride type and length, number of horses that started, number of eliminated and retired horses (total, by phases and reason), number of inexperienced horses (no RHU racing experience), number of horses that completed the ride and average speed (km/h) of each phase (phases 1 and 2), and average speed of the winning horse. If a variable was not recorded, the variable was assigned a value of not available. Reports with incomplete data for more than two variables were not included in the study. When fatalities occurred, the cause of death was recorded. To establish the cause of death, necropsy examinations by official veterinarians were performed on all dead equines. All data were obtained from FEU reports.2.2. Variable CategorizationComfort index (CI) was calculated by the sum of the temperature in Fahrenheit degrees and the relative humidity as a percentage [11]. CI was classified into three categories: low (CI < 130), medium (CI 130–150), and high (CI > 150). Ride types were classified by the length into two categories according to the FEU designation: short ride (SR: 60 km) and long ride (LR: 80–115 km). Causes of death were classified in four categories [9]: metabolic conditions (colic, exhausted horse syndrome, disseminated intravascular coagulation), catastrophic musculoskeletal injuries (defined as horses that died or were euthanized due to severe acute bone fractures that carry a poor clinical prognosis), sudden death (defined as acute death in a closely observed and previously apparently healthy animal), and accidental.Additionally, the part of the race where the death occurred was classified as during the ride (phase 1 or phase 2) or after the ride (24 h after the finish of the ride) to know if the horses that suffered fatalities completed the course or not.2.3. Statistical AnalysisDescriptive analyses (mean and standard deviations or percentage values) were calculated for the variables CI, horse experience, completed ride, speed (phase 1, phase 2, and average), overall fatalities, cause of death, and ride type. Statistical differences by ride type were calculated using a Chi-squared test, Fisher’s test, or Wilcoxon rank sum test. Screening of all exposure variables for overall fatalities (Live/Death) were performed separately using univariable Logistic Mixed Model analysis. Only variables with p < 0.2 were considered for inclusion in the multivariate Logistic Mixed Model analysis. In both analyses, all variables were considered fixed effects and the ride was considered as a random effect. The multivariate models were built using a forward selection procedure whereby variables with a Wald-test p < 0.05 were retained in the model. p-values of less than 0.05 were considered statistically significant. All analyses were performed in R (Version 4.2.2, 2022) and RStudio (version 2022.12.0 Build 353) [12].3. ResultsFrom a total of 702 RHU competitions taking place between 2007 and 2018, there was a 3-fold greater frequency (p < 0.0001) of LR (509, 42 rides per year) than SR (193, 16 rides per year). There were 16,856 horse starts, of which the number of horses was also greater in LR than in SR (Table 1, p < 0.0001). The average of horses competing per ride was 22 in SR and 29 in LR. Overall, 40.5% of the horses completed the ride, 43.1% were not able to complete the course due to metabolic reasons, 12.3% did not complete because of lameness, while 4.1% of the horses were retired from the ride due to rider-decision. The average speed of the winning horses was 28.1 km/h and 32.1 km/h (Table 1), with maximum and minimum speeds of 32.6 km/h and 20.3 km/h for LR and 35.9 km/h and 25.5 km/h for SR. The highest average speed in each phase and the average speed of the winning horse were recorded in SR (Table 1).Over the 12-year study period, there were 99 fatalities, and 68 of these horses were euthanized. All the euthanasias were performed with a rigorous criterion evaluating the future life and the suffering of the horse. The risk of fatality over the entire period was 5.87 per 1000 starts. The average number of deaths per year were 8.25 and did not differ over the years studied. The risk of fatality was significantly greater (p = 0.05, odd ratio = 1.52) for participation in SR (7.9 fatalities per 1000 start) than in LR (5.2 fatalities per 1000 start) (Table 2). There were significant differences in causes of death by ride type (Table 1). Short rides were associated with a greater proportion of sudden death, and LR were associated with a greater proportion of deaths due to metabolic alterations (Table 1). Catastrophic injuries occurred in a high proportion in both ride types (Table 1). Of the total deaths for metabolic reasons, 24 (69%) of the fatalities developed acute abdominal pain, 7 (20%) equine exhausted syndrome, and 4 (11%) disseminated intravascular coagulation.CI varied with the type of ride, SR had a greater frequency of high CI in comparison to LR; and LR had a greater frequency of low CI than SR, while there were no differences between ride type for the medium CI (Table 1). However, when CI was evaluated separately for each ride type, CI did not represent a significant risk factor for death in either ride type (Table 2).Inexperienced horses were more likely to suffer fatalities than experienced horses in both ride types (Table 2 and Table 3).There was no association between experience and the probability of dying in LR (Table 2), but under a multivariate analysis (Experience and Completed ride), a significant association was found between these variables in LR (Table 3).Horses that participated in SR completed the ride in a greater proportion than those that participated in LR (Table 1). Besides, regardless of ride type, more horses died during the ride than after them, so most of them did not complete the ride (Table 2 and Table 3).4. DiscussionThis is the first study to characterize RHU rides according to ride-type (short vs. long), as well as the causes of death and their risk factors. This was a retrospective cohort study, where many variables, such as sample population and variability associated with horse background and the environment of the races, could not be recorded. The extent of the limitation should be considered for the interpretation of the results.Equestrian endurance sports require the greatest physiological demand for the athlete horse FEI [2,4,13,14]. The FEI endurance is the most widely described [2,4,15,16,17]. The racing distances were classified as short and long according to the FEU regulations. Long FEI endurance races would be longer than 140 km. Compared to FEI endurance races, RHU has higher speeds, greater weight load, fewer stages, a lower proportion of horses that finish the race, and a higher fatality rate. According to our data, RHU is most likely one of the most demanding events for horses [5].During the period 2007–2018, LR were more frequent than SR, which follows the same pattern as FEI endurance 14. However, unlike endurance, where the speeds of the races were below 25 km/h [14,18,19,20], RHU races were faster, with speed averages between 28 km/h and 32 km/h, reaching maximum speeds close to 36 km/h. In addition, horses completing RHU races were 10 to 40% less than those reported for endurance [20,21,22]. Another clear difference with endurance is that while in RHU the main cause of elimination was due to metabolic condition (43%), in endurance the highest percentage of elimination (25 to 40%) was due to lameness [14,21,22,23].Among the main RHU characteristics, the following are briefly highlighted: LR are more frequent than SR; SR are faster than LR ones; SR had a higher frequency of high CI than LR, and LR had a higher frequency of low CI than SR; and the percentage of inexperienced horses and those who completed the ride was greater in SR than in LR.Regarding fatalities, there were 99 deaths, of which 68 were euthanasia. In both types of rides, more horses died during than after the ride. Most causes of fatalities are incompatible with the endurance exercise. The horses that suffered fatalities and finished the course were due to metabolic causes, except for one animal that presented sudden death. The higher rate of occurrence of metabolic fatalities in LR most likely induced the higher course completion in horses that died during LR (Table 1 and Table 2).The probability of suffering fatalities was higher in inexperienced horses. In addition, SR were associated with increased risk of sudden death, while LR were associated with increased risk of death due to metabolic alterations. It is interesting to note that although equine resistance sports are very popular throughout the world and that their popularity has been growing [24], according to our knowledge, except for one study [10], there are no reports on equine fatalities in endurance equestrian sports. During the 12-year study period, the fatality rate in RHU was 5.87 per 1000 starts.This result is much greater than that reported by Balch et al. (2019) in endurance horses under AERC rules (0.28 fatalities per 1000 starts) [10]. Horse deaths in RHU competitions attract the attention of public opinion and negatively affect the sensitivity of the public towards these sports; they also generate controversial opinions regarding the intensity of the exercise carried out by the animals that participate.The higher fatality rate of the RHU compared to the endurance competition once again highlights the high metabolic and locomotor demand that this sport represents for the horse. The metabolic requirements demanded by high speeds over long distances, with only two stages, create highly challenging conditions for the RHU horse. Additionally, horses run mostly on hard ground, with a high rider weight. Another characteristic of the sport is the important prize money, which adds excitement to the already exciting competition, and can distract competitors from the horse’s state of health and well-being. Since the death of equines during competitions represents a major welfare concern, it should be a priority to know the frequencies and causes of death in all equine sports worldwide. It is a great challenge for veterinarians to try to minimize the frequency of equine deaths during sports. In this sense, the fatalities in the RHU show the need for greater controls and strictness in the limits to which equines are exposed.In this study, the type of the races (SR vs. LR) in RHU was influenced by the comfort index, the percentage of inexperienced horses, whether they completed the race, the speeds, and the number and cause of fatalities. The SR had higher speed, higher frequency of high comfort index, greater percentage of inexperienced horses, and those who completed the ride in comparison with LR. The comfort index was used as an indicator of thermal stress in this work. It is widely used due to its simplicity and low cost, but there are significant weaknesses due to misleading for many combinations of temperature and humidity. Many indices have been evaluated to assess heat stress, even in horses, showing a better predictive capacity. This limitation should be considered for the interpretation of the results.The risk of death of horses was higher in SR than in LR, unlike what was reported in endurance by Balch et al. (2019), in which the risk of death increased with the distance traveled (0.12, 0.35, and 1.48 fatalities per 1000 stars in rides of 48, 80, and 160 km, respectively) [10]. It is possible to speculate that, although less distance is covered, the big locomotor and metabolic demand imposed by a greater speed in SR than in LR determined a greater proportion of equine deaths—mainly catastrophic and sudden deaths. According to the type of death, in relation to the type of race, SR had some similarities with racehorses, since deaths due to catastrophic injuries and sudden death predominate [2]. On the other hand, LR agreed with what has been reported for endurance events where the highest proportion of causes of death were metabolic alterations and catastrophic injuries [6,7]. In previous studies [24,25] no associations were found between the speed of individual horses and elimination for lameness or metabolic reasons. Although in our study, the speed of the races in which horses died did not differ statistically from those in which they survived, we cannot conclude that speed does not influence the deaths of horses. Recorded speed was the average speed of the stage in which they died, or the average speed of the race if they managed to finish it. Therefore, the individual speed of the horse was not considered, nor was the accumulated distance—which requires another study design for its analysis.The highest proportion of deaths in SR was due to sudden death, defined as acute, exercise-associated death in a closely observed and previously apparently healthy animal [26]. Cardiovascular disease is often confirmed or suspected [27]. Most of the studies on sudden death are in thoroughbred racehorses and eventing horses [7,28]. According to Lyle et al. (2011), the prevalence of sudden death in racehorses in the UK between 2000 and 2007 was 0.28 deaths per 1000 starts [6]. Comyn et al. (2017), found a prevalence of 0.14 death per 1000 starts in FEI eventing horses between 2008 and 2014 [29]. In the present study, the mortality rate from this cause was 1.02 deaths per 1000 starts, which is higher than any previous report. Intense exercise requires high oxygen consumption such as horse racing or three-day events, producing large increases in cardiac output and blood pressure, increasing propensity for major cardiovascular events [6,29].Navas de Solis et al. (2018) studied cases of sudden death in many types of equestrian sports and riding horses, and reported that 71.9% of the horses that suffered sudden death collapsed during exercise [27]. These results are in contrast to the work of Lyle et al. (2011) on racehorses, where most sudden deaths occurred after the race [6]. Exercise time during competitions in racehorses is much less than in three-day eventing or RHU. The intensity and duration of the exercise most likely play an important role in the incidence and moment of presentation of sudden deaths in sport horses. Horses collapsing during exercise may suffer catastrophic lesions, and riders may fall and be injured during these episodes and while riding a horse that presented sudden death. Therefore, the study of the causes of sudden death during RHU, and its prevention may be imperative, not only for increasing the horse’s welfare, but also for human safety concerns.The most frequent cause of death in LR was metabolic condition. This cause of death occurs mainly in horses after long duration submaximal exercises [2,4]. The marked increase in metabolism for such a long period of time is accompanied by intense energy consumption and loss of body water and electrolyte stores, as a consequence of thermoregulation [2,4,10,14,15]. Mild dehydration is compatible with competitive performance, however, serious medical problems may develop in horses that compete in endurance events—even death [2,15]. The distance, due to the longer time in exercise, most likely has an influence on the presentation of metabolic alterations, since it represents a risk factor for its presentation in endurance horses [14]. In our work, the proportion of deaths due to metabolic alterations was higher in LR.The main cause of death due to metabolic disorders was Acute Abdomen Syndrome (AAS). Balch (2019) reported AAS as being the most common clinical presentation (85%) as a prelude to death in AERC endurance races [10]. The management of the horse before and during the race can lead to episodes of AAS, such as gastric dilatation due to the effect of fasting before the race and peristalsis problems due to dehydration and electrolyte imbalances [2,15].Catastrophic injuries occurred in a high proportion in both ride types. In this study, the animals with catastrophic injuries did not complete the ride (43/44), being euthanized when an irreparable injury was diagnosed. Of the total catastrophic injuries, 11 were in SR and 33 in LR, although no differences were found. In the present work, the mortality rate due to catastrophic injuries was 2.6 dead horses per 1000 starters, which is even higher than those reported in racehorses or in three-day eventing horses.According to Misheff et al. (2010), the risk of suffering from bone pathologies during endurance races increases with the distance covered and the increasing speeds [30]. Most epidemiological studies show a positive association between the risk of catastrophic injuries and distance covered, although not all findings have been consistent [31].In our study, we also show that experience is a risk factor for fatalities. Therefore, another element that could have contributed to the higher risk of death in SR was the greater proportion of inexperienced horses in comparison to LR. Previous studies in sporting horses as endurance [25] or thoroughbred racehorses [32,33] have found experienced horses to be associated with reduced odds of deleterious outcomes compared to horses that were inexperienced (in endurance for the distance) or less experienced thoroughbred racehorses. However, the concept of inexperience is not the same, since in FEI races there is a qualification system that does not allow horses to compete if they have not completed shorter distance races. Therefore, they are only inexperienced at that distance. In our case, although many animals had competed in other sports, none of the horses considered inexperienced had competed in RHU races. The fatality rate for inexperienced horses was 13.2 and 8 fatalities per 1000 starts in SR and LR, respectively.Based on this set of results, it is clear that although LR are more frequent than SR in RHU, the risk of death is higher in SR than in LR. Therefore, the need to control racing conditions is evident, especially in SR, where speeds are higher.The differences and characteristics of the RHU as an equestrian endurance sport shown in this study, with notable consequences for equine athletes, warn about the importance of independent evaluation of each discipline to ensure the well-being of our equine and human athletes in all equestrian sports. Further studies are needed to assess more specific risk factors, in order to provide objective data that can help to plan racing schedules and serve as a basis for regulations, ultimately improving both the welfare of RHU horses and the public perception of this discipline.5. ConclusionsSR and LR have important differences that are manifested in the characteristics of the race, the causes of death, and the risk of fatality. This must be considered to understand the physical impact of equine participation in this type of event.The high fatality index shown in this work, especially in inexperienced horses, warrants urgent investigation in this sport to minimize the mortality associated with RHU-specific diseases and to improve the welfare of the RHU horses. | animals : an open access journal from mdpi | [
"Article"
] | [
"horse",
"fatalities",
"comfort index",
"experience",
"equine",
"animal welfare"
] |
10.3390/ani11092614 | PMC8465983 | "The red deer (Cervus elaphus) de novo genome assembly (CerEla1.0) has provided a great resource for(...TRUNCATED) | "The family Cervidae groups a range of species with an increasing economic significance. Their karyo(...TRUNCATED) | "1. IntroductionThe family Cervidae (Ruminantia) groups more than fifty extant deer species, includi(...TRUNCATED) | animals : an open access journal from mdpi | [
"Article"
] | ["BAC mapping","comparative cytogenetics","chromosome fission","chromosome fusion","FISH","genome as(...TRUNCATED) |
10.3390/ani13061115 | PMC10044701 | "Recently, increasing the efficiency of porcine embryo cultures by promoting oocyte maturation in vi(...TRUNCATED) | "As a member of the neurotrophic family, brain-derived neurotrophic factor (BDNF) provides a key lin(...TRUNCATED) | "1. IntroductionOvarian follicle development is a fundamental process of reproductive physiology in (...TRUNCATED) | animals : an open access journal from mdpi | [
"Article"
] | [
"BDNF",
"porcine",
"proliferation",
"microRNA",
"CCND1",
"ERK"
] |
10.3390/ani12010052 | PMC8749932 | "Transport stress (TS) can impact the physiology and psychology of broilers, and this can be an impo(...TRUNCATED) | "Abnormal iron metabolism can cause oxidative stress in broilers, and transport stress (TS) may pote(...TRUNCATED) | "1. IntroductionThe pre-slaughter transport process is an important integral part of poultry managem(...TRUNCATED) | animals : an open access journal from mdpi | [
"Article"
] | [
"transport stress",
"broilers",
"iron homeostasis",
"TMT proteomics"
] |
10.3390/ani12030387 | PMC8833775 | "The European eel is a species with high commercial value for aquaculture, and it has suffered a dra(...TRUNCATED) | "The European eel (Anguilla anguilla) is a commercially valued species for aquaculture. Over the pas(...TRUNCATED) | "1. IntroductionThe European eel (Anguilla anguilla) is a commercially valued species, especially fo(...TRUNCATED) | animals : an open access journal from mdpi | [
"Article"
] | ["European eel","spermiation","ovulation","human chorionic gonadotropin hormone","carp pituitary ext(...TRUNCATED) |
10.3390/ani11082164 | PMC8388502 | "Sea cage farms dominate European aquaculture production of seabass (Dicentrarchus labrax) and gilth(...TRUNCATED) | "The behavioural responses of fish to a stressful situation must be considered an adverse reaction c(...TRUNCATED) | "1. IntroductionAnimal welfare evaluation should be promoted so that decisions are made based on sci(...TRUNCATED) | animals : an open access journal from mdpi | [
"Review"
] | [
"seabass",
"seabream",
"welfare",
"stunning",
"slaughtering"
] |
10.3390/ani12030309 | PMC8833531 | "In Aotearoa–New Zealand, the helicopter application of the toxin sodium fluoroacetate (1080) is a(...TRUNCATED) | "Aotearoa–New Zealand has embarked on an ambitious goal: to completely eradicate key invasive mamm(...TRUNCATED) | "1. IntroductionInvasive mammals present a global threat to biodiversity, especially in island ecosy(...TRUNCATED) | animals : an open access journal from mdpi | [
"Article"
] | ["density","eradication","invasive species","predator-free","Rattus rattus","rodent","sodium fluoroa(...TRUNCATED) |
End of preview. Expand
in Dataset Viewer.
Dataset Card for SciLay
Dataset Summary
SciLay comprises 43,790 instances, each representing a scientific article in the biomedical domain. Each instance in the dataset includes the following components:
- plain_text: Containing a plain language summary of the scientific article. This section is written in a simple and accessible language, and is intended to be understandable by a wide audience.
- technical_text: This section contains the abstract of the scientific article. It provides a detailed and technical description of the research conducted in the article.
- full_text: This section contains the complete article of the scientific research.
In addition to the textual content, each instance is associated with the following metadata: - Keywords: Keywords that capture the main topics and themes addressed in the article. - Journal: The journal in which the article is published, providing context about the source of the research. - DOI (Digital Object Identifier): A unique identifier for the article, facilitating easy referencing.
The main objective of the SciLay dataset is to support the development and evaluation of text summarization models that can effectively simplify complex scientific language while retaining the essential information. Each article is published by a scientific journal. There are 15 journal sources:
- NC: Nature Communications
- A: Animals : an Open Access Journal from MDPI
- PLGEN: PLoS Genetics
- PLPAT: PLoS Pathogens
- PLCB: PLoS Computational Biology
- PLNTD: PLoS Neglected Tropical Diseases
- B: Biology
- I: Insects
- PLB: PLoS Biology
- CB: Communications Biology
- SD: Scientific Data
- MBIO: mBio
- C: Cancers
- OTHER: which includes additional journals that taken individually would not have contributed sufficient instances
Current defaults are 1.0.0 version (cased raw strings) and 'all' journals:
from datasets import load_dataset
ds = load_dataset("disi-unibo-nlp/SciLay") # default is 'all' journals
ds = load_dataset("disi-unibo-nlp/SciLay", "all") # the same as above
ds = load_dataset("disi-unibo-nlp/SciLay", "NC") # only 'NC' journal (Nature Communications)
ds = load_dataset("disi-unibo-nlp/SciLay", journals=["NC", "A"])
Languages
English
Dataset Structure
Data Instances
Each instance contains a set of doi, pmcid, plain_text, technical_text, journal, topics, keywords. Each of which was extracted by scraping articles in XML and HTML format.
{
'doi': '10.3390/ani12040445',
'pmcid': 'PMC8868321',
'plain_text': 'PPP3CA is one of the candidate genes for goat reproduction, but no studies have been carried out yet. Therefore, the purpose of this study was to determine the associations between copy number variations in the goat PPP3CA gene and litter size and semen quality in goats, including Shaanbei white cashmere goats (SBWC) (n = 353) and Guizhou Heima (GZHM) goats (n = 64). Based on the association analysis, the results showed that only CNV1 (copy number variation 1) and CNV2 (copy number variation 2) were distinctly related to the first-birth litter size in female goats (p = 7.6802 × 10−11; p = 5.0895 × 10−9), and they were also significantly associated with the semen quality of SBWC goats (p < 0.05). These findings prove that the PPP3CA gene plays an important role in reproduction traits in goats.',
'technical_text': 'Copy number variations (CNVs) have many forms of variation structure, and they play an important role in the research of variety diversity, biological evolution and disease correlation. Since CNVs have a greater impact on gene regulation and expression, more studies are being finalized on CNVs in important livestock and poultry species. The protein phosphatase 3 catalytic subunit alpha (PPP3CA) is a key candidate gene involved in the goat fecundity trait, and has important effects on precocious puberty, estrogen signal transduction pathways and oocyte meiosis. Additionally, PPP3CA also has a dephosphorylation effect in the process of spermatogonial stem cell meiosis and spermatogenesis. So far, there is no research on the relationship between the copy number variations of the PPP3CA gene and reproduction traits. Therefore, the purpose of this study was to determine the association between copy number variations in the goat PPP3CA gene and litter size and semen quality in Shaanbei white cashmere goats (SBWC) (n = 353) and Guizhou Heima goats (n = 64). Based on the association analysis, the results showed that only CNV1 and CNV2 within the PPP3CA gene were distinctly related to the first-birth litter size in female goats (p = 7.6802 × 10−11; p = 5.0895 × 10−9, respectively) and they were also significantly associated with the semen quality of SBWC goats (p < 0.05). In addition, individuals with Loss genotypes demonstrated better phenotypic performance compared to those with other types. Therefore, CNV1 and CNV2 of the PPP3CA gene are potentially useful for breeding, as they are linked to important goat reproduction traits.',
'full_text': '...'
'journal': 'Animals : an Open Access Journal from MDPI',
'topics': [ 'Article' ],
'keywords': [ 'goat', 'PPP3CA', 'copy number variation (CNV)', 'litter size', 'semen quality' ]
}
Data Fields
doi: (Digital Object Identifier). It is a unique alphanumeric string assigned to a digital document, such as a research paper, article, or dataset. Not all istances have it.pmcid: A unique identifier in the PubMed Central library database. Not all istances have it.plain_text: The summary of the article in plain english.technical_text: The abstract of the article.full_text: The complete article.journal: The journal which published the article.topics: An object containing the types in which the article is classified (i.e. Research Article, Review, ecc.). Not all istances have it.keywords: An object containing the keywords of the article. Not all istances have it.
Data Splits
| train | validation | test | |
|---|---|---|---|
| all | 35026 | 4380 | 4384 |
| NC | 5549 | 694 | 694 |
| A | 3909 | 489 | 489 |
| PLGEN | 3087 | 386 | 386 |
| PLPAT | 2920 | 365 | 365 |
| PLCB | 2589 | 324 | 324 |
| PLNTD | 2289 | 286 | 287 |
| B | 1617 | 202 | 203 |
| I | 1181 | 148 | 148 |
| PLB | 896 | 112 | 113 |
| CB | 867 | 108 | 109 |
| SD | 725 | 91 | 91 |
| MBIO | 607 | 76 | 76 |
| C | 6782 | 848 | 848 |
| OTHER | 2008 | 251 | 251 |
- Downloads last month
- 131