VisEgg: a robust phenotyping tool to assess rainbow trout egg features and viability

Assessing female fish reproductive success requires a thorough evaluation of egg characteristics, including egg number, size, and variability as well as egg developmental potential through the monitoring of embryo survival after fertilization. While embryonic success relies, at least in part, on paternal contribution, some parameters are strictly related to egg characteristics, one of the main ones being the viability of the egg when released into the water at spawning. It is however not necessarily possible, at least in salmonid fish that lay nontransparent eggs, to separate the different causes of egg/embryo failure. In this context, our aim was (i) to develop a simple and rapid system to capture images of rainbow trout eggs combined with computerized processing of these images to perform a fully automatic individual characterization of egg features including number and size (ii) to estimate unfertilized egg viability through the monitoring of the percentage of eggs that will not survive to water hydration. To evaluate the VisEgg system, unfertilized eggs (approximatively 400 eggs per batch) originating from 105 different females were hydrated in water. After 24 h, a picture of the eggs was obtained using a dedicated shooting system consisting of a light source and a digital single-lens reflex (SLR) camera. An image processing algorithm was developed to allow the automatic detection and separation of the eggs and to perform automatic measurements of egg number and individual egg size. The presence of white egg was used as an indirect measure of egg integrity, the “whitening” being the result of water entry into the egg through the vitelline membrane. These white eggs were therefore considered nonviable, as a result of their lack of physical integrity. Fertilization assays were performed in parallel using a subsample of the same egg batch. Embryonic development was monitored and hatching rate was calculated. A significant correlation between white egg percentage after hydration and hatching rate was observed (Spearman coefficient = −0.557, p < 0.001), in consistency with the fact that nonviable egg will not allow successful embryonic development. In contrast, the percentage of eggs that do not successfully hatch includes egg/embryo failures of different nature including reduced egg viability. Using the VisEgg, we were able to quantify the lack of viability of the eggs separately from the different other events that may occur during fertilization and incubation. the VisEgg is a convenient and reliable tool to obtain individual measures on trout eggs. It can be used to assess not only egg size and egg number but also unfertilized egg viability before fertilization.

Abstract Assessing female fish reproductive success requires a thorough evaluation of egg characteristics, including egg number, size, and variability as well as egg developmental potential through the monitoring of embryo survival after fertilization. While embryonic success relies, at least in part, on paternal contribution, some parameters are strictly related to egg characteristics, one of the main ones being the viability of the egg when released into the water at spawning. It is however not necessarily possible, at least in salmonid fish that lay nontransparent eggs, to separate the different causes of egg/embryo failure. In this context, our aim was (i) to develop a simple and rapid system to capture images of rainbow trout eggs combined with computerized processing of these images to perform a fully automatic individual characterization of egg features including number and size (ii) to estimate unfertilized egg viability through the monitoring of the percentage of eggs that will not survive to water hydration. To evaluate the VisEgg system, unfertilized eggs (approximatively 400 eggs per batch) originating from 105 different females were hydrated in water. After 24 h, a picture of the eggs was obtained using a dedicated shooting system consisting of a light source and a digital single-lens reflex (SLR) camera. An image processing algorithm was developed to allow the automatic detection and separation of the eggs and to perform automatic measurements of egg number and individual egg size. The presence of white egg was used as an indirect measure of egg integrity, the "whitening" being the result of water entry into the egg through the vitelline membrane. These white eggs were therefore considered nonviable, as a result of their lack of physical integrity. Fertilization assays were performed in parallel using a subsample of the same egg batch. Embryonic development was monitored and hatching rate was calculated. A significant correlation between white egg percentage after hydration and hatching rate was observed (Spearman coefficient = −0.557, p < 0.001), in consistency with the fact that nonviable egg will not allow successful embryonic development. In contrast, the percentage of eggs that do not successfully hatch includes egg/embryo failures of different nature including reduced egg viability. Using the VisEgg, we were able to quantify the lack of viability of the eggs separately from the different other events that may occur during fertilization and incubation. the VisEgg is a convenient and reliable tool to obtain individual measures on trout eggs. It can be used to assess not only egg size and egg number but also unfertilized egg viability before fertilization.

Introduction
The control of egg quality (i.e., the ability of the egg to be fertilized and subsequently develop into a normal embryo) is of major importance for many, if not all, aquaculture fish species (Migaud et al. 2013). A regular supply of high-quality eggs is mandatory for rainbow trout hatchery sustainability (Bromage et al. 1992). Given the high cost of broodstock rearing, large variations in the quantity or quality of gametes can significantly impact the competitiveness and sustainability of fish farms and aquaculture companies (Bromage et al. 1992;Bobe 2015). Hence, controlling egg quality is a major issue in aquaculture with important economic consequences.
Teleost eggs consist of an inner ooplasm comprising vitellus (or yolk) surrounded by the vitelline membrane also known as plasma membrane. The outer membrane called the chorion is a thick layer with a single micropyle. Spermatozoa must pass across the micropyle to penetrate the egg and allow fertilization (Kuchnow and Scott 1977). Teleost eggs are soft in the ovary and when the egg is freshly stripped from the female, the chorion is limp and the egg is flaccid, the egg is not really round. A thin space between the chorion and the vitelline membrane exists and corresponds to the perivitelline space. When the eggs are transferred into the water, a contractile activity of the yolk mass occurs as a result of subsequent discharge of the cortical vesicle contents in the perivitelline space (Kobayashy 1985). Colloid substances, with water taking up property, are thus discharged into the perivitelline space. Water is osmotically drawn through the fine pores of the chorion into the perivitelline space. This process creates a turgid pressure in the egg and the chorion is stretched until a steady state between the internal and external environment is achieved, resulting in the hardening of the chorion (Alderdice 1988). The stage of equilibrium is reached in about 20 min. This phenomenon is called egg water hardening process and egg from rainbow trout absorbs approximatively 20% of its initial volume in the water (Yamamoto 1962;Blaxter 1969;Lahnsteiner et al. 1999). The vitelline membrane is the only significant barrier to the diffusion of water and ions between the vitellus and the external medium. When eggs are of good quality, the vitelline membrane is highly impermeable to water and electrolytes and it is not easy to cause a rupture of the membrane (Gray 1932). Inversely, when eggs are of bad quality, the vitelline membrane is weakened and the water pressure is sufficient to break it. The vitellus is rich in vitellogenin and protein, specifically in globulin. Globulins and vitellogenins are soluble due to the salts contained in the vitellus. But when the membrane breaks, the water enters and salts are diluted, resulting in the precipitation of globulins (Gray 1932;Van Heerden et al. 1996) and vitellogenins (Engelmann et al. 1976;Fremont and Riazi 1988). In this case, the egg turns white. The presence of white egg was thus interpreted as an indirect measure of vitelline membrane integrity, the "whitening" reflecting the result of water entry into the ooplasm. These white eggs were thus considered nonviable.
The improvement of female reproductive traits (egg diameter, absolute fecundity, number of viable eggs) through changes in their rearing conditions (feed, temperature, light cycle, etc.) or genetic selection implies the ability to measure the quality of the spawn in a very large number of individuals. The assessment of fish breeding performance requires, in particular, a precise evaluation of the number of spawned eggs but also information on the egg ability to be fertilized and subsequently develop into a viable embryo. Prior to fertilization, it is extremely difficult to estimate differences between good-and bad-quality eggs in rainbow trout, despite the fact that the characterization of predictive estimators or egg quality markers would have major applications for research and industry (Migaud et al. 2013). For the above reasons, enabling a comprehensive evaluation of egg quality before fertilization is highly desirable in aquaculture. Moreover, during fertilization and incubation, poor egg quality can lead to several problems including insufficient egg hydration, lack of fertilization, developmental arrests, embryonic mortalities, and deformities (Brooks et al. 1997;Bobe and Labbé 2010;Migaud et al. 2013). Given the wide variety of the problems observed, it is usually very difficult to determine the causes of embryonic/developmental failure.
In the present study, an automatic phenotyping system based on image analysis was developed to characterize different egg features (number, size) and due to the occurrence of nonviable eggs in order to separate this cause from other events that may occur later during fertilization or embryonic development. The phenotype measured here is based on the egg ability, specifically the vitelline membrane capacity, to resist to water pressure during the hydration process. This method is complementary, yet much more rapid, to a full evaluation of developmental success that is performed on fertilized eggs to separate the origin of egg quality defects (viable eggs vs. nonviable eggs that do not allow developmental success).

Ethical statements
Experimentations were conducted in the INRAE PEIMA experimental facility (Sizun, France-Agreement number B29-277-02). All fish were reared and handled in strict accordance with French and European policies and guidelines of the INRAE PEIMA Institutional Animal Care and Use Ethical Committee, which specifically approved this study. Fish were monitored daily during the experiment. If any clinical symptoms (i.e., morphological abnormality, restlessness, or uncoordinated movements) were observed, fish were sedated by immersion in MS-222 solution at a concentration of 50 mg L −1 and then euthanized by immersion in a MS-222 solution at a concentration of 400 mg L −1 (anesthetic overdose) in 3 min.

Broodstock breeding and experimental design
Female rainbow trouts from an autumn-spawning strain were held under natural photoperiod until their first reproduction (2 years) in the INRAE PEIMA experimental facility. After spawning, fish (initial weight: 1693 ± 306 g) were transferred into 6 outdoor 2-m 3 tanks. Fish were fed on a commercial trout broodstock diet (Le Gouessant, Lamballe, France) for a period of 5 months. During this 5-month period, an artificial photoperiod regime was applied to obtain a second reproduction during summer: a 2.5-month-long photoperiod (20 h light, 4 h dark from December to March 2017) followed by a 2.5-month-short photoperiod (8 h light, 16 h dark from March to June 2017). The water temperature during the experiment ranged from 6 to 12°C.

Sample acquisition
During spawning season (June 21 to August 3, 2017), females were checked for ovulation once a week by applying a manual pressure onto the abdomen. When ovulation was detected, fish were given an intraperitoneal injection of an antibiotic (Vetrimoxin) to prevent Flavobacterium infection. Two days after the detection of ovulation, females were manually stripped. Eggs were weighted and two samples were taken for fertilization assays and for VisEgg phenotyping analysis. A total of 105 spawns were individually analyzed.

Fertilization assays
For each spawn (i.e., each female, n = 105), approximately 400 eggs were fertilized with a pool of sperm collected from males fed ad libitum on a commercial diet and embryonic success was monitored. Fifteen microliters of a pool of semen obtained from five males presenting the highest sperm motility were used, and 15 mL of Actifish solution (sperm motility activating saline solution, IMV technologies, L'Aigle, France) were added onto the eggs. Five minutes later, the sperm motility activating solution was drained and egg batches were transferred into individual incubators in a recirculated water unit. Water temperature (12.0-12.6°C ) was monitored daily. Dead eggs and embryos were periodically manually counted and removed. Survival at eyeing, hatching, and completion of yolk-sac resorption (YSR) were monitored and calculated as a percentage of the initial number of eggs used for fertilization. The occurrence of noticeable morphological malformations (spinal cord torsion, head or caudal fin malformations, etc.) at YSR was also recorded (Supplementary Table 1).

Phenotyping analysis
For VisEgg phenotyping, unfertilized eggs (20 to 25 g corresponding to approximatively 400 eggs) were stored in a 100-mL container and then water from the incubation system was added to hydrate the eggs (80 mL). After a 24-h hydration at 4°C, a picture of the egg sample was taken using a dedicated commercially available shooting system. This system consisted in a light tablet and a digital single-lens reflex (SLR) camera (canon EOS 1000D, resolution: 10.1 M pixels) (Fig.  1). Eggs were rinsed with water and delicately poured into a petri dish (diameter = 14 cm). The image calibration was performed with an image of a 1-euro coin with a known area to calculate the pixel size. The petri dish was then placed on the light tablet. It was verified that the eggs were not superimposed and a picture was taken. An image processing algorithm was developed and coded using a Visual Basic for Applications (VBA) macro with Visilog 7.3 software (Thermo Scientific). This software allows the automatic detection and separation of the eggs in order to make fully automatic measurements of egg number and specific measurement for each egg (size, shape, color, white egg occurrence). The total number of ovulated eggs per females could be back calculated using the number of measured eggs and the total weight of the spawn. The workflow of the VisEgg macro-command is presented in Fig. 2. All details and relevant information about the VisEgg image analysis workflow are available here: https://github. com/LpgpImage/VisEgg/wiki.

Statistical analysis
Statistical analysis of the data was carried out using the R software. The percentage data were transformed using an arcsine transformation before analysis. Simple linear regressions were performed to test the correlation between survival rates at different stages. Multiple linear regressions (MLR) were performed to test the correlation between different egg quality parameters measured with the VisEgg (i.e., egg diameter, variability of egg diameter, and white egg percentage) and hatching rate. MLR was simplified to the minimal adequate model by stepwise removing of nonsignificant elements. The adequacy of the final linear regression was assessed by making residual plots to check the normality. This showed that nothing was amiss; the analysis was thus validated. The Spearman coefficients were calculated. For statistical analysis performed to compare the two calculating methods of hatching rate (using all eggs or on only viable eggs, see Fig. 5), a Kruskal-Wallis test was used followed by Dunn test's post hoc analysis.

Results and discussion
Automatic assessment of egg features Evaluating the reproductive performance traits requires counting the eggs and monitoring developmental success. This time-consuming task, performed manually in hatcheries, is particularly tedious. In most cases, eggs are weighed and not counted, as the weight is easier to determine. Furthermore, measuring the mean weight of a spawn does not provide information on individual variability within each spawn, which is also an indicator of spawn quality. Using the VisEgg automatic phenotyping tool presented here, we were able to observe differences that would have been overlooked using standard simpler approaches.
The total egg number is an indicator of the reproductive performance variability between females as illustrated in Fig. 3 for two different females: the total egg number can be similar despite the different weight of the spawn. When only the total weight of the spawn is used, the size of the eggs remains unknown. Using the VisEgg, it is now possible to measure a large number of eggs very easily. In this study, a very large number of eggs (n = 52,000 eggs originating from 105 spawns) were automatically and individually measured. This significantly increased the accuracy of the measurement. Using the VisEgg, we were able to measure individual egg size of all sampled eggs and therefore to estimate intra-spawn variability (i.e., individual egg variability) and inter-spawn variability (i.e., individual female variability).
The distribution of egg diameter obtained with 105 spawns individually analyzed showed values ranging from 3.8 to 5.2 mm with a mean of 4.6 mm. The distribution of the coefficient of variation of egg diameter for all spawns ranged from 2.2 to 9.2% with a mean of 4.6%. These values show a high variability between the spawns analyzed in this study that could not have been observed with a standard technique. The egg size remains an interesting parameter to measure. Although the possible benefit effect of egg size on egg survival and alevins is still debated, especially under aquaculture conditions (Bobe and Labbé 2010;Bromage et al. 1992;Campbell et al. 1992;Jastrebski and Morbey 2009;Migaud et al. 2013;Springate and Bromage 1985b), it is established that larger eggs produce larger alevins (Springate and Bromage, 1985) with substantial fitness advantages over small alevins (Heath et al. 2003). While egg size is very often measured in studies on fish reproductive performance, it is unusual to measure egg size variability within the spawn. This is certainly due to the difficulty of assessing this parameter. Female rainbow trout (Oncorhynchus mykiss) is a group-synchronous species which produces a single spawn each year where all oocytes develop and ovulate at the same time (Lubzens et al. 2010). In normal conditions, eggs from a single spawn are of homogeneous size; the presence of large egg size intra-spawn variability in salmonid can be the consequence of a disruption of the physiological processes underlying oogenesis (Jastrebski and Morbey 2009). The measurement of this parameter therefore appears to be relevant for rainbow trout reproduction studies. Thus, the VisEgg is a convenient and reliable tool to obtain accurate individual measures such as egg size and intra-spawn size variability.

Assessment of egg viability regardless of fertilization and subsequent developmental success
Unfertilized eggs rather than fertilized eggs were used to evaluate our phenotyping tool because the hardening process is independent of fertilization (Lahnsteiner et al. 1999). Two distinct phenotypes can be observed after the hydration process: either the egg hydrates normally and grows slightly or the egg turns white. This phenotype is easily observable on images (Fig. 3D). Moreover, the VisEgg can separate normal (i.e., viable) eggs from white (i.e., nonviable) eggs and then calculate the percentage of white egg for each analyzed egg batch (Fig. 3C).
In this study, multiple linear regression was performed to test the correlation between hatching rate and the different parameters measured using the VisEgg (i.e., egg diameter, variability of egg diameter, and white egg percentage). Only white egg percentage was significantly correlated with hatching rate. In contrast, egg diameter and variability of egg diameter appear to have no link with egg quality and survival of alevins (Spearman coefficient = 0.39 and −0.30, respectively, for egg diameter and variability of egg diameter). Linear Fig. 2 The VisEgg workflow regression between white egg percentage and hatching rate shows a significant negative correlation (p-value = 2.6 x10 -10 ; Spearman coefficient = −0.56). It should be noted that we observed highly significant correlations between surviving rates at eyeing and hatching as well as between surviving rates at hatching and yolk-sac resorption (YSR) (p-value = < 2 × 10 -16 , Spearman coefficient = −0.99 for both correlations). Surviving rates at eyeing and YSR are therefore also significantly negatively correlated with the percentage of nonviable eggs (p-value = 9.9 × 10 -11 , Spearman coefficient = −0.58 and p-value = 4.7 x10 −10 , Spearman coefficient = −0.55 for eyeing and YSR, respectively).
The negative correlation between white egg percentage and developmental success was expected because nonviable eggs will not allow successful embryonic development. In addition, the modest coefficient of correlation can be explained by the composite nature of survival rate that includes the capacity of the egg to survive in the water (i.e., egg viability), to be fertilized and to develop into an embryo (or to die at different times during development). Figure 4 illustrates the successive different origins of egg/embryo failure that can occur between spawning and yolk-sac resorption. The graph shows the mean values obtained using the 105 different spawns analyzed in the present study. Figure 5A shows the hatching rate calculated using all eggs (in blue) or only using viable eggs (in red), the batches being ranked based on the overall hatching rate (blue dots). The blue and red dots do not follow exactly the same pattern due to the differences in the origins of egg/embryo failure up to hatching, with major differences between the two groups for some individuals. In some cases, when the red and blue dots are far apart, there is a high incidence of egg viability on the overall developmental success. In contrast, when the red and blue dots overlap (or are close), there is a low incidence of egg viability on the overall developmental success. Figure 5 (A and B) shows that a high incidence of egg viability is observed for a significant number of females. It should be Fig. 3 Images of trout eggs as processed by the VisEgg. (A) Initial image obtained with the VisEgg system equipment presented in Fig. 1. (B) Image as processed by VisEgg for an automatic detection and separation of the eggs in order to make fully automatic measurements of egg number and size. (C) Image as processed by the VisEgg for an automatic detection of white eggs and normal eggs (white eggs and normal eggs are represented in red and blue colors, respectively). (D) Pictures with macro objective illustrating normal and white eggs after hydration noted that the importance of egg viability on the overall success of development is highly variable as illustrated by the significant differences between red and blue dots that can be observed regardless of the overall development success with the exception of high-quality spawns (> 80% hatching rate; Fig. 5B). In 15.2% of the cases, there is more than 10% of difference between red and blue dots. The percentage of batches over 10% variation between the two calculating methods of hatching rate can reach 50% of the batches for hatching rates ranging from The data represent the average percentage values calculated from 105 egg batches followed individually. Viable and nonviable egg data were obtained with the VisEgg system. Fertilized eggs and survival at eyeing, hatching, and yolk-sac resorption (YSR) are expressed as a percentage of the initial number of eggs used for fertilization. Non-malformed alevins with full completion of yolksac resorption are considered viable alevins. Alevins with noticeable morphological malformations (spinal cord torsion, head or caudal fin malformations, etc.) and dead alevins between hatching and resorption are considered nonviable alevins (*nonviable or malformed alevins) Fig. 5 (A) Hatching rate calculated on the number of all eggs (blue) or on only viable eggs (red). (B) Percentage of batches with ≥ 10% variation between the two calculating methods of hatching rate. Statistical differences were evaluated by the Kruskal-Wallis test followed by the Dunn test post hoc analysis. Different causes of egg quality defects for FISH A, B, C, and D were represented in Fig. 6 21 to 40%. While it is not necessarily easy to decipher the causes of egg/embryo failure up to hatching, the VisEgg offers the possibility to quantify the percentage of viable eggs. As just an example, four particular cases are shown in Fig. 6 to illustrate the different causes that can lead to a similar developmental success. Fish A and Fish B present low and equivalent hatching rates (23.0 and 24.3%, respectively). However, Fish A exhibits a high incidence of egg viability (39.6% of nonviable eggs) on the overall developmental success, whereas Fish B exhibits a low incidence of egg viability (0.2% of nonviable eggs) on the overall developmental success. Similarly, Fish C and Fish D exhibit high hatching rates (65.2 and 69.2%, respectively) but for different reasons: Fish C exhibits a high incidence of egg viability (22.1% of nonviable eggs) on the overall developmental success, whereas Fish D exhibits a low incidence of egg viability (1.1% of nonviable eggs). These two examples clearly illustrate that similar developmental success (either low or high) can be observed even though the phenotype, and most likely the biological underlying cause, is completely different. The VisEgg tool offers the possibility to separate and quantify the incidence of nonviable eggs on the overall developmental success rather than loosing this information when only developmental success is monitored. This approach is thus complementary to a full evaluation of developmental success that is performed on fertilized eggs to separate the origin of egg quality defects (viable eggs vs. nonviable eggs that do not allow developmental success).

Conclusions
In summary, the VisEgg is a convenient, fast, automatic, and reliable tool to obtain individual measures of trout eggs. It can be used to assess not only egg size and number but also to assess unfertilized egg viability before fertilization. The VisEgg is the first automatic phenotyping tool for egg viability, a key component of egg quality. This tool, developed in rainbow trout, is now being used routinely on trout eggs but it could be adapted to the eggs of other species. In the future, the VisEgg will be adapted to measure other egg characteristics such as color, shape (eccentricity value), and postovulatory aging.
Funding information This study was supported by the Europea n M a r i t i m e a n d F i s h e r i e s F u n d G r a n t N o . PFEA470016FA1000002.
Compliance with ethical standards Experimentations were conducted in the INRAE PEIMA experimental facility (Sizun, France-Agreement number B29-277-02). All fish were reared and handled in strict accordance with French and European policies and guidelines of the INRAE PEIMA Institutional Animal Care and Use Ethical Committee, which specifically approved this study. Fish were monitored daily during the experiment. If any clinical symptoms (i.e., morphological abnormality, restlessness, or uncoordinated movements) were observed, fish were sedated by immersion in MS-222 solution at a concentration of 50 mg L −1 and Fig. 6 Different origins of egg quality defects that can occur between spawning and yolk-sac resorption in four fish are displayed in Fig. 5. See legend from Fig. 4 for the details on calculation methods