Fertilized eggs and newly hatched larvae are developmental stages outside the scope of the Guidelines of the European Union on the protection of animals used for scientific purposes (Directive 2010/63/EU), and approval from the Norwegian Food Safety Authority was not needed. Broodstock handling was performed as part of common hatchery practice.

Egg batches were collected from commercial broodstock at Nordic Halibut AS (Midsund, Norway) during routine production in two spawning seasons (summer 2020 and spring 2021; three females per season). Hatchery conditions and stripping procedures followed standard farm practice and are described in detail in Niepagen et al. (2025a), water temperature was kept at 6 °C two month prior and during the stripping season at 34.4 ppt salinity. Individual metadata such as age, size, first time spawner status, and pedigree were not available from hatchery records for the sampled females, and domestication generation could therefore not be resolved reliably. Once the first egg batch had been obtained from a female via hand stripping, the hatchery routinely attempted stripping at fixed 82 h intervals. This interval reflects established operational practice and approximates the typical 3 to 4 day batch cycle under farm conditions (expressed as a degree-day schedule in Niepagen et al., 2025a). Because the precise time of ovulation cannot be determined directly, a fixed interval can result in variable in vivo retention time at stripping among females. For this experiment, the planned stripping time for the experimental batch was advanced by 10 to 13 hours relative to the routine hatchery schedule (that is, stripped earlier), after at least two previous batches had been obtained from each female. This was done to increase the likelihood that the experimental batch was collected closer to ovulation and to minimize in vivo retention prior to the start of the ex vivo storage time series.

Once an egg batch was collected for the experiment, the eggs were first transferred into a dry bowl. To reduce potential inhomogeneity within the batch, the eggs were gently mixed and then evenly distributed into six ceramic containers (150 mL nominal volume; manufacturer not recorded). Each container was filled to near capacity to minimize headspace and was sealed with food grade polyethylene cling film to reduce air exposure and evaporation. One fraction (t0) was fertilized immediately, while the remaining five fractions were stored at 6 °C and fertilized after 1 (t1), 2 (t2), 4 (t4), 6 (t6), and 12 (t12) hours of storage. Batch 1 was excluded from subsequent analyses due to minor protocol adjustments implemented from Batch 2 onward; therefore, only Batches 2 through 6 are reported in this study. Batches 2 and 3 were fertilized using the same cryopreserved milt from SquarePacks (single male; Cryogenetics AS, Hamar, Norway), while Batches 4, 5, and 6 were fertilized using straws of cryopreserved milt from a different single male (Cryogenetics AS, Hamar, Norway). For each fertilization, approximately 50–150 mL of eggs was mixed in 1 L of seawater with either one-third of a SquarePack, or one straw of cryopreserved milt. Following fertilization, eggs were left to harden for 30 minutes, then rinsed in 5 L of seawater. Subsequently, 5–10 mL of fertilized eggs (150–300 eggs) was transferred into 800 mL cell culture flasks containing seawater with 0.25 mg/L of terramycin solution (Terramycin 100 mg/mL, Zoetis, USA) (Skaalsvik et al., 2015). This procedure was repeated for each time point. To avoid repeated handling of the same embryos across developmental endpoints, separate incubation units were prepared for each stage and were not reused across sampling points. For each batch and storage time, three replicate Petri dishes were prepared for early assessments (fertilization success and 8-cell morphology), and nine replicate cell culture flasks were prepared for longer-term incubation (three flasks allocated to each of the BL, 50dd, and hatching endpoints). These replicates are technical (within-batch) replicates derived from the same female’s egg batch; the biological replicate in the study is the female (batch). Fertilized eggs were incubated in Petri dishes or 800 mL cell culture flasks (incubation units). Petri dishes were placed in a temperature-controlled refrigerator at 6 °C. Cell culture flasks were placed standing upright in lidded polystyrene boxes to maintain darkness and reduce temperature fluctuations. The boxes were supplied with chilled seawater (34.4 ppt) to keep the surrounding bath at 6 °C; an overflow outlet maintained a constant water level to ensure consistent immersion of the flasks. Water inside Petri dishes and flasks was not exchanged during incubation. An overview of the experimental workflow and incubation setup is provided in Supplementary Figure S1.

During incubation of Batch 3, signs of microbial (bacterial/fungal) growth were observed in some replicates, manifesting as elevated early mortality in affected incubators. However, contamination was not uniform across replicates: at least one replicate from each time point remained relatively unaffected, and mortality patterns stabilized after the early developmental period. Importantly, samples for RNA extraction were collected from incubators that did not show visible signs of contamination at the time of sampling. Because microbial growth is unlikely to systematically upregulate vertebrate developmental pathways, and because the transcriptomic patterns observed (e.g., downregulation of organizer genes and patterning pathways with increasing storage time) are consistent with known effects of post-ovulatory aging in other species, we concluded that the contamination did not substantially confound the gene expression analysis. Nonetheless, this limitation is acknowledged, and validation of key findings in additional batches would strengthen future mechanistic studies.

To ensure randomization and observer blindness, incubators were labelled with numeric codes, and treatments (batch number and storage duration) were randomly assigned. At each sampling point, the contents of a given incubator were poured into a Petri dish, photographed using dark field illumination and a mirrorless digital camera (Canon EOS M100) for later analysis, and recorded without revealing the batch number or storage time to the observer. Each developmental endpoint was assessed from three independent incubation replicates per batch and storage time (technical replicates).

Early assessments (fertilization success and 8C morphology) were performed using the three Petri dish replicates and were not returned to longer-term incubation after handling and imaging. For all egg batches, Petri dishes were incubated for 14 hours to assess fertilization success and the percentage of normally developing embryos at the 8C stage (14 hpf). Fertilization success was assessed from the Petri dish replicates (three per batch and storage time) by counting the number of cleaving embryos relative to the total number of eggs in the dish (typically 150 to 300 eggs per replicate). The resulting fertilization proportion was then used to estimate the initial number of fertilized eggs in the corresponding incubation flasks from the same batch and storage time (estimated fertilized eggs = fertilization proportion × total eggs stocked), which served as the denominator for survival and hatching metrics. Because fertilized and unfertilized eggs were not separated prior to stocking flasks, this approach was used to account for the unfertilized fraction when calculating survival and hatching outcomes. Normal blastomere morphology was evaluated according to the criteria described by Kjørsvik and Lønning (1983); Kjørsvik et al. (1984). The percentage of eggs with normal 8C blastomeres was calculated from the total number of fertilized eggs. At 48 hours post-fertilization, samples were taken at the blastula stage (BL stage). However, image quality at this stage was insufficient to reliably identify morphological abnormalities or deformities. At 200 hours post-fertilization, corresponding to the 50 degree-day (50dd) stage, the percentage of embryos displaying normal morphology was recorded (Niepagen et al., 2025b). Survival at hatching was assessed at 330 hours post-fertilization. The percentage of live eggs and larvae was calculated relative to the number of fertilized eggs initially incubated. Hatching success was defined as the proportion of larvae that successfully hatched from the total number of fertilized eggs. Dead eggs were manually removed and counted throughout the incubation period. For Batches 2 and 3, removal occurred only 2–3 times. For Batches 4–6, dead eggs were removed a total of nine times during incubation.

Batch 3 was selected for differential gene expression analysis. Because the precise time of ovulation in Atlantic halibut cannot be determined, time point zero (t0) is not directly comparable across different batches. Therefore, only a single batch (Batch 3) was used for transcriptomic analysis to ensure internal consistency. This longitudinal design allowed us to follow the same female’s eggs across storage times and developmental stages, thereby maximizing internal consistency and minimizing confounding by among-female variation in baseline egg quality or ovulatory timing (Antonsson and Melsted, 2025). Following completion of all developmental stage sampling, the remaining eggs from Batch 3 were collected in triplicate into 1.8 mL cryotubes and immediately flash-frozen in liquid nitrogen for RNA extraction. Total RNA was extracted and sequenced independently for each subsample.

The statistical analysis was conducted using RStudio version 2022.7.2.576 (RStudio Team, 2021). Egg quality parameters, expressed as proportions, were analyzed using mixed-effects beta regression models implemented in the glmmTMB package (Brooks et al., 2017). Storage time (t0–t12) was included as a fixed effect, and batch (female) was included as a random intercept to account for individual variation among batches. Estimated marginal means were obtained using the emmeans package. For comparisons against the control (t0), Dunnett-adjusted contrasts were used. To assess differences among all storage times, Tukey-adjusted pairwise comparisons were performed and summarized using compact letter display, where groups sharing the same letter are not significantly different (p > 0.05). Model estimates were back-transformed to the response scale (percentage) for visualization. The observational unit for the beta regression models was the incubation unit (Petri dish for fertilization and 8C, or flask for later endpoints). For each female batch and storage time, each endpoint was assessed from three technical replicate incubation units; female (batch) was treated as the biological replicate and included as a random intercept.

For exploratory visualization of batch-specific patterns, within-batch correlations between storage time and egg quality parameters were assessed using Spearman’s rank correlation on arcsine square-root transformed data. Statistical significance was set at p < 0.05 for all analyses.

Kaplan–Meier survival analysis was based on the counts of dead eggs removed from each incubator during development. Each removal represented a mortality event recorded at that observation day. The embryos that were still alive in the incubators that were terminated at sampling of BL stage and 50dd stage before the experiment was terminated (at hatching) were counted as censored (Dudley et al., 2016). Statistical significance was established for p ≤ 0.05.

To quantify the effect of storage time after stripping, unfertilized eggs from five females were stored at 6 °C for up to 12 h (t0, t1, t2, t4, t6, t12) prior to fertilization, and egg quality endpoints were tracked through hatching. At t0 (fertilized immediately after stripping), fertilization success ranged from 77% to 87%, normal 8-cell (8C) morphology from 80% to 98%, and normal development at 50 degree-days (50dd) from 55% to 96%. Survival to hatching at t0 ranged from 18% to 58%, and hatching success from 15% to 53% (Table 1).

Average values of egg quality parameters across all batches and storage times are summarized in Table 1, while detailed results for each batch are presented in Supplementary Table S1. Across females (n = 5), mean fertilization success declined from 81 ± 4% at t0 to 60 ± 13% at t12, and mean normal 8C morphology declined from 88 ± 7% to 64 ± 16% (Table 1). Normal 50dd development showed the clearest reduction at extended storage, decreasing from 74 ± 16% at t0 to 45 ± 13% at t12 (Table 1). Survival until hatching and hatching success were more variable among females, but both were lowest at t12 (17 ± 10% and 22 ± 10%, respectively; Table 1).

Mixed-effects beta regression identified significant storage-time effects for fertilization success, normal 8C morphology, and normal 50dd development (Table 1, Figure 1). Fertilization success declined progressively with storage time, with model-estimated means decreasing from approximately 82% at t0 to approximately 30% at t12. Normal 8C morphology remained relatively stable through t2 but was significantly reduced at t6 and t12 (Dunnett-adjusted p < 0.05). Normal 50dd development declined after t4 and was significantly reduced at t6 (p = 0.023) and t12 (p < 0.001). For hatching survival and hatching success, model estimates indicated a decline toward t12, but effects were weaker and more variable among females (Table 1).

Batch-level values (Supplementary Table S1; Supplementary Figure S2) showed that the timing and magnitude of decline differed among females. Fertilization and early cleavage (8C) remained relatively high through t4 to t6 in some batches (e.g., B3 and B5) but decreased earlier or more steeply in others (e.g., B6). Similar between-batch variability was observed for 50dd development and hatching-related endpoints. Cumulative survival patterns during incubation mirrored the endpoint metrics, with the lowest survival typically observed in t12 groups across batches.

Principal component analysis (PCA) was used to summarize global transcriptome-wide variation among samples within each developmental stage (Figure 1) and to evaluate whether storage time is associated with consistent shifts in overall expression profiles. At the unfertilized (UF) stage, eggs stored for 12 hours (t12) formed a loose yet distinct cluster, while other time points showed considerable overlap. At the 8C stage, no clear clustering patterns were observed among the storage times. In contrast, BL samples showed the clearest separation by storage time, with a progressive shift in ordination space from t0 toward later storage times and the greatest separation at t12. A similar, but less distinct, trend was observed at 50dd (Figure 1).

Differential gene expression was assessed by comparing eggs fertilized immediately after stripping (t0) to eggs stored for 1, 2, 4, 6, and 12 hours. In total, 20,844 genes were downregulated and 19,095 were upregulated across all stages and storage times (Figure 2A). DEG counts were strongly stage-dependent, with the largest response at the BL stage, followed by 50dd and UF, while the 8C stage showed comparatively few changes (Figure 2A). Within each stage, t12 consistently produced the highest number of DEGs. Gene Ontology (GO) enrichment analysis revealed 1,612 significantly upregulated and 1,623 downregulated biological processes in stored versus immediately fertilized eggs across all stages (Figure 2B; Supplementary Table S2). Across stages, the number of enriched GO terms broadly tracked DEG counts, with the strongest enrichment at BL and at longer storage durations (Figure 2B).

GO overrepresentation analysis identified stage-dependent transcriptional responses to increasing ex vivo post-stripping storage duration. A compact overview of downregulated biological process enrichment across stages and storage times is provided in Figure 3, while the complete enrichment outputs for both upregulated and downregulated DEGs (including statistics for all terms, stages, and time points) are reported in Supplementary Table S3.

At the UF stage, enriched biological processes were exclusively upregulated from 4 to 12 h of storage and were dominated by RNA processing and transcript turnover, including regulation of RNA stability and mRNA metabolic and catabolic processes. Additional upregulated terms were associated with regulation of catabolic processes and ubiquitin-dependent protein catabolic activity, consistent with increased protein turnover, alongside ribonucleoprotein complex biogenesis and other development-related terms (Supplementary Table S3). No significant enrichment of downregulated biological process terms was detected at this stage.

At the 8C stage, enrichment was again primarily upregulated and centered on protein and ribonucleoprotein complex assembly, ribosome and ribonucleoprotein biogenesis (including rRNA processing), ncRNA metabolism, chromatin organization, and mitotic cell cycle-associated processes (Supplementary Table S3). In contrast, downregulated enrichment at 8C was limited and restricted largely to glycosaminoglycan and aminoglycan metabolic and biosynthetic processes.

At the BL stage, relatively few terms were consistently enriched among upregulated DEGs, whereas a marked expansion of downregulated enrichment emerged from 4 h onward (Supplementary Table S3). The downregulated response encompassed key developmental and regulatory programs, including germ layer and system development, embryonic organ and skeletal morphogenesis, heart morphogenesis, determination of left-right symmetry, ameboidal-type cell migration, and regulation of Wnt signaling. The structure of the BL-stage downregulated term landscape is further visualized in the treemap summary (Figure 4).

At the 50dd stage, consistently enriched upregulated terms included rRNA modification and ribosomal assembly, mitochondrion organization, protein targeting, cytoplasmic translation, and related RNA-processing functions (Supplementary Table S3). Downregulated enrichment was modest at early time points but broadened from 4 h onward and was dominated by morphogenesis- and muscle-related programs and associated regulatory processes, including muscle development and differentiation, cellular component assembly involved in morphogenesis, regulation of transmembrane transport (including calcium ion transport), and transcriptional regulation (Supplementary Table S3).

To assess how storage time affects key developmental regulators, we examined the expression of genes involved in dorso-ventral and anterior-posterior axis patterning across storage times (Figure 5). Maternal transcripts at the UF and 8C stages showed relatively stable expression across storage times, with inter-gene variation exceeding time-dependent changes (Figure 5A). This stability suggests that maternal mRNA pools are largely resistant to short-term aging under the storage conditions used here. In contrast, genes expressed at the BL stage displayed marked sensitivity to storage time. Organizer and patterning genes including gsc, ved, chrd, fgf24, and bmp2b declined progressively with increasing storage duration, particularly at t6 and t12 (Figure 5). Most BL-stage genes showed peak relative expression at t0-t1, followed by progressive downregulation through t12, indicating that extended storage specifically disrupts gene activation at the mid-blastula transition. At 50dd, expression patterns varied by gene cluster (Figure 5B). Some genes (gdf6a, fgfr1b, bmpr1bb) showed delayed upregulation in t12 embryos, while others (bmp7b, fgfr3, cav1) exhibited progressive downregulation with extended storage. These divergent patterns likely reflect downstream consequences of disrupted axis specification at earlier stages.

A critical consideration when interpreting these results is that our experimental design employed in vitro storage of stripped eggs at 6 °C, whereas post-ovulatory aging in commercial hatcheries occurs in vivo within the body cavity. However, several aspects of our design minimized differences between these conditions. Eggs were stored in the ovarian fluid that was stripped along with them, and the storage temperature of 6 °C matched the broodstock holding water exactly, as eggs were maintained in the same water source and temperature-controlled system as the fish themselves. Thus, the primary distinction between our in vitro storage and in vivo retention is anatomical location rather than the immediate biochemical environment.

Nevertheless, this distinction may have biological relevance. Ovarian fluid in vivo is not static: during in vivo over-ripening of rainbow trout eggs, ovarian fluid pH decreases while protein levels, esterified and non-esterified fatty acids, and enzyme activities (aspartate aminotransferase and acid phosphatase) increase, and these ovarian-fluid parameters correlate with egg viability (Lahnsteiner, 2000). Baseline compositional analyses further demonstrate that salmonid ovarian fluid contains substantial inorganic and organic constituents, consistent with a buffered physiological milieu (Lahnsteiner et al., 1995). In contrast, ovarian fluid stored ex vivo may undergo passive chemical change without physiological regulation by the female. Following ovulation, unfertilized salmonid eggs retained in vivo undergo progressive changes, with vitellogenin fragments and other breakdown products accumulating in the coelomic fluid (Rime et al., 2004); whether similar accumulation occurs at comparable rates ex vivo is unknown. Studies directly comparing in vitro and in vivo post-ovulatory aging in the same species are scarce (Mohagheghi Samarin et al., 2015), and this represents an important methodological gap in the field.

The transcriptomic analysis revealed that post-ovulatory aging exerts its most pronounced effects at the blastula stage rather than in unfertilized eggs or early cleavage stages. While maternal transcript pools at the unfertilized and 8-cell stages remained largely stable across storage durations, the blastula stage showed massive transcriptomic disruption beginning at 4 hours post-stripping. This pattern suggests that the primary defect may not be widespread degradation of maternal mRNAs, as has sometimes been assumed (Ma et al., 2019), but rather a failure to properly execute developmental programs that depend on zygotic genome activation.

This interpretation aligns with proteomic observations in Atlantic halibut, where poor-quality eggs exhibit mitochondrial damage and structural breakdown rather than gross differences in transcript abundance (Yilmaz et al., 2022). Similarly, in other teleosts, post-ovulatory aging has been linked to cytoskeletal disintegration, yolk degradation, and mitochondrial dysfunction, with oxidative stress implicated as a key mediating factor (Lord and Aitken, 2013; Mohagheghi Samarin et al., 2015; Takahashi et al., 2013). Increased reactive oxygen species have been implicated in triggering mitochondrial oxidative stress, ATP depletion, calcium imbalance, and apoptosis in aging oocytes, mechanisms conserved across vertebrates and likely relevant to fish (Lord and Aitken, 2013; Takahashi et al., 2013). The subcellular damage accumulated during post-ovulatory storage may compromise the oocyte’s capacity to support the massive increase in transcriptional and translational activity required at the mid-blastula transition, even if the maternal mRNA pools themselves appear intact.

The specific pathways downregulated at the blastula stage, ribosome biogenesis, RNA processing, germ layer formation, Wnt signaling, and cell migration, are precisely those required for axis formation and organogenesis. The progressive decline of organizer genes including gsc, ved, vox, and chrd with increasing storage time provides a direct molecular link to the truncated body axes and absent eye structures observed in abnormal 50dd embryos. Goosecoid plays a central role in axial mesoderm development and is normally repressed by ventralizing signals such as vox and ved, which also declined in parallel (Imai et al., 2001; Ramel and Lekven, 2004; Shimizu et al., 2002). Disruptions in FGF signaling also paralleled the observed morphological phenotypes; fgf24, fgfr1a, and fgfr2 were strongly expressed at early time points but declined steeply with prolonged storage, likely contributing to the loss of normal body axis formation, as these genes are essential for mesoderm induction, posterior patterning, and elongation of the embryonic axis (Challa and Chatti, 2013; Fürthauer et al., 2004; Rohner et al., 2009; White et al., 2017). These findings parallel observations in mammals, where post-ovulatory aging has been linked to long-term developmental defects (Lord and Aitken, 2013), suggesting conserved underlying mechanisms potentially involving mitochondrial dysfunction and oxidative damage to cellular machinery required for zygotic genome activation.

It is important to acknowledge that the transcriptomic analysis was conducted on a single batch (Batch 3), which limits inference about among-female variation in transcriptional responses. Additionally, this batch experienced microbial contamination during incubation that affected some replicates, though RNA samples were collected from unaffected incubators. While the observed transcriptomic patterns are biologically coherent with known post-ovulatory aging mechanisms, validation in additional batches would strengthen these conclusions. Nevertheless, the concordance between the transcriptomic disruption at the blastula stage in Batch 3 and the phenotypic decline across all five batches supports the interpretation that post-ovulatory aging compromises developmental competence through stage-specific failure at the mid-blastula transition.

The practical implications of these findings for Atlantic halibut hatcheries are substantial but require careful consideration of the differences between our experimental design and commercial practice. By advancing stripping time by 10–13 hours relative to the standard 82-hour interval, we obtained eggs with fertilization rates of 77–87% and normal 50dd development of 55–96%, performance substantially better than what is often achieved in commercial operations. However, this success was partly fortuitous, as the optimal advancement depends on knowing the actual ovulation time for each female, which cannot be precisely determined with current methods.

The finding that egg quality declines within 4–6 hours post-stripping, combined with the observation that a 10–13 hour advancement improved quality, suggests that standard 82-hour stripping intervals may frequently miss the optimal window for many females. Individual variation in ovulatory timing means that fixed-interval protocols inevitably result in some females being stripped too early (pre-ovulation, yielding no eggs) and others too late (extended post-ovulatory aging). The batch-specific variation we observed, with some females showing earlier or steeper quality declines than others, reinforces this point. This is consistent with findings in other teleosts showing rapid post-ovulatory quality decline (Clarkson et al., 2024; Konar et al., 2024; Mohagheghi Samarin et al., 2015).

Determining peak ovulation timing in Atlantic halibut remains challenging. Fertilization rate, while useful as a retrospective indicator, cannot guide timing for the current batch. The more intensive monitoring protocol described by Norberg et al. (1991), involving more frequent stripping attempts based on individual tracking, provides more reliable timing by minimizing the interval between ovulation and stripping. Hormonal markers, such as plasma levels of maturation-inducing steroids or prostaglandins, have been used to predict ovulation timing in other species (Scott et al., 1983; Zohar et al., 1986; Springate et al., 1984; Sorensen et al., 2018; Takahashi et al., 2018) but are to our knowledge not commonly implemented in halibut hatcheries. Non-invasive monitoring approaches, including ultrasound imaging of ovarian development, might allow more precise tracking of individual females’ reproductive status (Næve et al., 2018; Mlingi et al., 2023; Loher and Stephens, 2011; Novelo and Tiersch, 2012; Guitreau et al., 2012). The economic value of Atlantic halibut juveniles may justify the labor investment required for such intensive monitoring in select broodstock populations.

Rather than fixed-interval stripping, a feedback-driven approach could improve synchronization with individual ovulation cycles. If fertilization rates for a given female’s batch fall below approximately 60% (as a practical trigger in our dataset), this signals that the next stripping attempt should occur earlier than the typical 72–96 hour ovulatory cycle, perhaps at 60 hours rather than 82. Implementing such protocols requires sufficient labor to monitor each female individually throughout the spawning season, suggesting that hatcheries may benefit from limiting the number of actively managed females to those they can monitor adequately, rather than attempting to manage large numbers with generic protocols. The economic implications are significant: improving fertilization from approximately 60% to 80% and normal 50dd development from 45% to 75%, as observed between t12 and t0 eggs, could more than double viable juvenile production per spawning female.