A transcriptional program underlying the circannual rhythms of gonadal development in medaka

Significance

Circannual rhythms are ~1-y biological rhythms driven by the endogenous circannual clock. The existence of circannual rhythms has been demonstrated in various organisms. Thus, circannual rhythms are a fundamental property of living organisms and are crucial for life on Earth, from the equator to the poles. However, the genes involved in these rhythms and the mechanisms underlying long-term circannual oscillators remain unknown. The Japanese medaka fish is a powerful and well-established animal model for studying seasonal rhythms. Here, we show the existence of ~6-mo endogenous circannual rhythms in medaka. Genome-wide gene expression analysis of medaka hypothalamus and pituitary identified the circannual genes and an analysis of these genes highlighted the involvement of tissue remodeling in the circannual time-keeping.

Abstract

To cope with seasonal environmental changes, organisms have evolved approximately 1-y endogenous circannual clocks. These circannual clocks regulate various physiological properties and behaviors such as reproduction, hibernation, migration, and molting, thus providing organisms with adaptive advantages. Although several hypotheses have been proposed, the genes that regulate circannual rhythms and the underlying mechanisms controlling long-term circannual clocks remain unknown in any organism. Here, we show a transcriptional program underlying the circannual clock in medaka fish (Oryzias latipes). We monitored the seasonal reproductive rhythms of medaka kept under natural outdoor conditions for 2 y. Linear regression analysis suggested that seasonal changes in reproductive activity were predominantly determined by an endogenous program. Medaka hypothalamic and pituitary transcriptomes were obtained monthly over 2 y and daily on all equinoxes and solstices. Analysis identified 3,341 seasonally oscillating genes and 1,381 daily oscillating genes. We then examined the existence of circannual rhythms in medaka via maintaining them under constant photoperiodic conditions. Medaka exhibited approximately 6-mo free-running circannual rhythms under constant conditions, and monthly transcriptomes under constant conditions identified 518 circannual genes. Gene ontology analysis of circannual genes highlighted the enrichment of genes related to cell proliferation and differentiation. Altogether, our findings support the “histogenesis hypothesis” that postulates the involvement of tissue remodeling in circannual time-keeping.

Circannual rhythms are approximately 1-y biological rhythms driven by the endogenous circannual clock (1, 2). The circannual clock entrained to seasonal environmental cycles is crucial for life on Earth, from the equator to the poles, as it regulates various physiological and behavioral characteristics and provides organisms with adaptive advantages. In temperate zones, for instance, organisms are exposed to pronounced seasonal environmental changes, such as harsh winters and intense heat. The circannual clock allows them to better adapt to predictable changes in cyclic environments and determines the timings of reproduction, diapause, hibernation, and molting. Seasonal environmental changes are less generalizable in tropical zones, but organisms in these zones still use the circannual clock to synchronize the timings of reproduction and migration. The existence of the circannual rhythm under constant photoperiod and temperature conditions has been demonstrated in various species, including dinoflagellate (3), plants (4), mollusks (5), insects (6, 7), fish (8, 9), reptiles (10), birds (11, 12), and mammals (13, 14), strongly indicating its functional significance during the course of evolution.

Several hypotheses have been suggested regarding the mechanisms underlying circannual rhythms. Gwinner proposed frequency demultiplication hypothesis regarding the interaction with the circadian clock (15). However, this hypothesis is not supported by experimental data. Lincoln and Hazlerigg proposed the histogenesis hypothesis, which postulates that the circannual clock is driven by cell birth and death as well as tissue regeneration (16). The binary switching mechanism, in which seasonal calendar cells may record photoperiodic history as a binary code, rapidly flips from the winter or summer state (17). Loudon and his colleagues proposed that this mechanism may generate long-term circannual rhythms. The involvement of epigenetic modification such as histone modification and DNA methylation has also been proposed (2, 18). Since an analysis of F₁ hybrids of two populations with different circannual patterns showed intermediate patterns, circannual rhythms are believed to be genetically controlled (19, 20). However, the genes that regulate circannual rhythms and the underlying mechanisms controlling long-term circannual oscillators remain unclear in any organism.

Results

Seasonal Changes in Gonadal Size under Natural Outdoor Conditions.

Among various vertebrate species, medaka fish (Oryzias latipes) show robust seasonal dependency in reproduction and behavior (21–23) and hence provide an excellent model to investigate the molecular basis of the circannual clock. To characterize annual reproductive rhythms in medaka, we first kept male and female medaka under natural outdoor conditions at the National Institute for Basic Biology, Okazaki, Japan (35.0°N, 137.2°E) (SI Appendix, Fig. S1A). The gonadosomatic index (GSI: gonad weight as a percentage of body weight), which is an indicator of gonadal maturation, was measured every 2 wk for 2 y, from October 2015 to October 2017. This showed that both female and male medaka exhibit clear annual reproductive rhythms (SI Appendix, Fig. S2). Since the magnitude of gonadal changes was larger in females than in males, we focused mainly on females in further analyses.

Regression Analysis Predicts Changes in Gonadal Size.

To quantitatively investigate when and to what extent input stimuli such as solar radiation, water temperature, and day length contribute to seasonal gonadal development in medaka, we applied a multiple linear regression model (Fig. 1 A–C). We constructed the following regression model to explain the temporal differentiation of GSI at a given time point using a weighted sum of past input information (Eq. 1):

$$GS{I}_{{}_t}^{\text{'}}={{\displaystyle \sum }}_i^4{{\displaystyle \sum }}_{j=-T}^0{w}_{i,j}{\text{input}}_{i,t+j}+{\epsilon }_t$$ [1]

Fig. 1.

Predicting gonad development of female medaka using a multiple linear regression model. (A) Schematic representation of the multiple linear regression model. (B) Inputs used for multiple linear regression. Left: Solar radiation (SR; pink), water temperature (WT; blue), and day length (DL; orange). Right: Each dot and line represent the individual gonad mass (gonadosomatic index, GSI) and mean, respectively, of female medaka (n = 10). (C) The time derivative of female GSI (GSI′) indicates whether the gonad becomes larger or smaller at that point in time. (D and E) Multiple linear regression with SR, WT, DL, and GSI. The gray lines are GSI′ (D) and GSI (E) obtained by experiments, while the green dotted line and solid line show the training results and test results, respectively, obtained by linear regression analysis. (F) The response functions of SR, WT, DL, and GSI. (G) Prediction error is shown as a bar graph, with bars ordered by increasing prediction accuracy of linear regression when input information varies.

where GSI′_t denotes the temporal differentiation of GSI at time = t, and input_{i, t} indicates the four types of inputs, specifically solar radiation (SR, i = 1), water temperature (WT, i = 2), day length (DL, i = 3), and GSI (i = 4), at time = t. Utilizing own past values as input in an autoregressive model is widely used to predict the output in machine learning. Therefore, we included GSI values at time = t as an input. w_{i, j} is a regression parameter for the input i and delay time j, and ε_{n, t} is white Gaussian noise. w_{i, j} (i = 1, 2, 3, 4; j = 0, …, T) is a so-called response function for each input i, where T is the maximum delay time. Given that an impulse stimulation of the input i is applied, the response of GSI′ can be simply obtained as w_{i, t}.

Using the Ridge regression method, which prefers smooth response functions, we analyzed the first half of the 2 y of female GSI data as a training dataset to estimate the response functions that fit the experimental GSI′ dataset (Materials and Methods) (Fig. 1D). The penalty strength of the Ridge regression and the time delay value of the response function were optimized to minimize the error in predicting data in the second half of the 2 y, i.e., a test dataset (SI Appendix, Fig. S3A). The estimated response functions using the input data predicted the latter half of the GSI′ time series (Fig. 1D). The predicted GSI′ data also succeeded in reconstructing the time series of GSI, especially the transient drop of GSI in Japan’s rainy season, around June (Fig. 1E). Next, we focused on the estimated response function of SR, WT, DL, and GSI (Fig. 1F). We found that the absolute value of the weight of the response function for GSI was much larger than that for the other inputs, suggesting that GSI itself contributes to the change in the magnitude of GSI. In addition, the response function for GSI yielded a positive value within a few days before the sampling day, but a negative value as time elapsed (Fig. 1F). Thus, the shape of the response function is a differentiator, with the most recent GSI having a positive impact on the GSI change and the GSI of a few days prior having a negative impact. Furthermore, we examined the predictive accuracy of GSI′ with fewer input types and found that it was higher when GSI itself was used as an input (Fig. 1G). A similar analysis of the GSI of male medaka yielded nearly identical results (SI Appendix, Fig. S3 B–H). This is unsurprising considering the feedback mechanism in the hypothalamus–pituitary–gonadal axis.

Identification of Broad-Sense Seasonally Oscillating Genes (bsSOGs).

In order to uncover how the circannual clock coordinates physiological and behavioral seasonal rhythms, we next performed RNA-sequencing (RNA-seq) analysis of brain regions containing the hypothalamus and pituitary (SI Appendix, Fig. S4A). Since genes critical for the photoperiodic response are induced around dusk (22), brain samples were collected every month at around 8 pm (Japan Standard Time) for 2 y, from October 2015 to October 2017, from medaka kept under natural outdoor conditions (SI Appendix, Fig. S1A). Sunset in Nagoya is ~19:10 in June and ~16:40 in December. Among 26,835 genes, 21,286 genes were expressed. To extract genes that exhibit annual rhythms, a nonlinear regression analysis was performed by fitting a cosine curve to the expression data to calculate the amplitude, phase, and coefficient of determination for each gene. This analysis identified 3,341 genes, which we defined as broad-sense seasonally oscillating genes (bsSOGs) (Fig. 2A and Dataset S1). These bsSOGs comprised 15.7% of the expressed genes (3,341/21,286 genes). The peak expression of most of these genes was in summer (from June to August) or winter (from December to February), with very few exhibiting peak expression in spring or autumn (SI Appendix, Fig. S4B). Of note, many seasonally oscillating genes with peak expression in summer or winter have also been reported in plants (24). Importantly, the bsSOGs included several genes encoding hormones involved in the regulation of various physiological functions, such as reproduction [glycoprotein hormones, α polypeptide (cga), follicle-stimulating hormone β subunit (fshb), luteinizing hormone subunit β (lhb), inhibin subunit β A (inhba)], osmoregulation [prolactin (prl)], stress response [corticotropin-releasing hormone b (crh), pro-opiomelanocortin-like (pomcl)], and metabolism and seasonal reproduction [thyroid-stimulating hormone β subunit a (tshba)] (Fig. 2B). Most of these exhibited peak expression in summer (except for pomcl in winter), when medaka are reproductively active. This temporal organization appears to optimize reproductive success in this species.

Fig. 2.

Transcriptional landscape of seasonal adaptation under natural conditions. (A) Heatmap showing gene expression of broad-sense seasonally oscillating genes (bsSOGs) under natural conditions in brain regions containing the hypothalamus and pituitary. The color scale represents normalized signal intensity. (B) Expression profiles of hormone genes and circadian clock genes under natural conditions. Each point represents the expression level in a single sample. Red lines represent trend lines determined by cosine curves. Orange indicates summer (June to August) and blue indicates winter (December to February). (C) Heatmap showing gene expression of daily oscillating genes (DOGs) in each season. The color scale represents normalized signal intensity. (D) Expression profiles of circadian clock genes under natural conditions. Each point represents the expression level in a single sample. Red lines represent trend lines determined by a cosine curve. White bars represent daytime (i.e., from sunrise to sunset), whereas the black bars indicate nighttime for each season. (E) Venn diagram showing the overlap of DOGs in each season. (F) Venn diagram showing the overlap of bsSOGs and DOGs.

We then examined the associations between seasonal expression and gene functions. Genes in bsSOGs with peak expressions in specific seasons were extracted (window size, 30 d; step, 10 d; adjusted P < 0.01), and gene ontology (GO) enrichment analysis was performed. Thirty-three GO terms pertaining to biological processes were found to be significantly enriched in at least one window (SI Appendix, Fig. S4C). GO terms related to ribosome function such as “rRNA processing,” “rRNA metabolic process,” “ribonucleoprotein complex biogenesis,” “ribosome biogenesis,” “ribosomal large subunit biogenesis,” and “ribosome assembly” were significantly enriched in winter. On the other hand, the GO terms “nucleosome assembly,” “chromatin assembly,” and “DNA replication” were enriched in summer.

Identification of Daily Oscillating Genes (DOGs) and Narrow-Sense SOGs (nsSOGs).

Many circadian clock genes, such as bmal2 (bmal2a and bmal2b), cryptochrome (cry1-1, cry1-2, cry1-3, and cry2), period (per1, per2a, and per3), rev-erbb, and RAR-related orphan receptor (rora and rorb), were also identified as bsSOGs (Fig. 2B), suggesting that bsSOGs may include genes that exhibit seasonal changes in daily oscillation patterns. The circannual system consists of an input pathway, an oscillator, and an output pathway, similar to the circadian system; photoperiodism plays an important role in the input pathway of the circannual system (1). Because seasonal modulation of the circadian clock gene expression has been proposed to act as a circadian coincidence timer for photoperiodic time measurements (25), waveforms of the circadian clock gene expression at different times of the year are of interest. We next investigated the daily expression profile of each gene in each season. Samples were collected every 4-h for a 24-h period on the spring equinox, summer solstice, autumn equinox, and winter solstice in 2018 and 2019 (SI Appendix, Fig. S1B). Nonlinear regression analysis revealed 1,381 genes, which we defined as daily oscillating genes (DOGs) (Fig. 2C and Datasets S2–S5). There was seasonal variation in the number of DOGs detected, with most identified at the summer solstice and autumn equinox (spring equinox: 138 genes; summer solstice: 696 genes; autumn equinox: 566 genes; winter solstice: 237 genes). DOGs included many circadian clock genes, such as clock, bmal1, per1, per2b, cry1-1, and rorb, whose expression showed clear daily oscillation at the summer solstice and autumn equinox but obscure daily oscillation at the spring equinox and winter solstice (Fig. 2D). Season-specific GO analysis of DOGs using the web-based portal Metascape 3.5 (26) also highlighted robust circadian oscillation at the summer solstice and autumn equinox (SI Appendix, Fig. S5). Most DOGs showed daily oscillation only in a specific season. Indeed, although 211 genes were identified as DOGs at both the summer solstice and autumn equinox, no genes were shown to be DOGs in all seasons (Fig. 2E).

Seasonal changes in the daily oscillating gene expression waveforms may play an important role in the input pathway (photoperiodism) of the circannual system; however, circannual rhythms were generated even under constant photoperiodic conditions. Because the expression profiles of daily oscillating genes (e.g., phase and amplitude) are generally considered stable under constant photoperiodic conditions, we consider the expression changes in daily oscillating genes to not be the heart of the circannual clock mechanism. Therefore, to identify genes with “true” seasonal oscillation, we further identified 2,948 narrow-sense seasonally oscillating genes (nsSOGs), defined as seasonally oscillating genes that did not exhibit daily oscillation in any season (Fig. 2F and Dataset S6).

Changes in Biological Processes during Seasonal Transitions.

In a principal component analysis of transcriptomes from different seasons, there was clear differentiation between those from the winter solstice and spring equinox and those from the summer solstice and autumn equinox (Fig. 3A). Accordingly, the number of differentially expressed genes (DEGs) was very high when comparing “Spring vs. Summer” and “Autumn vs. Winter,” but low when comparing “Summer vs. Autumn” and “Winter vs. Spring” (Fig. 3 B and C). These results suggest that transcriptomes from the summer solstice and autumn equinox are similar, as are those from the winter solstice and spring equinox. Pathway analysis of the DEGs highlighted the enrichment of cellular and metabolic processes during the transition from both spring to summer and autumn to winter (Fig. 3 D and E).

Fig. 3.

Comparison of transcriptomes in each season. (A) Principal component analysis (PCA) plot of transcriptome data for individuals in each season. Each color indicates the season during which samples were obtained. (B) Volcano plot of differentially expressed genes (DEGs) in each comparison. Red, blue, and gray dots indicate up-regulated, down-regulated, and unchanged genes, respectively (Padj < 0.05 and log2 fold change > 1). (C) Venn diagram showing the overlap of genes expressed in each season. (D) GO enrichment analysis of DEGs in different seasons. Red, blue, and green bars indicate biological processes, cellular components, and molecular functions, respectively. (E) Expression profiles of genes related to lipid metabolism, phototransduction, immune function, and pigmentation under natural conditions. Each point represents the expression level in a single sample. Red lines represent trend lines determined by a cosine curve. Orange indicates summer (June to August), and blue indicates winter (December to February).

Existence of a Circannual Clock in Medaka.

Recently, Kumar et al. demonstrated clear circannual rhythms in spotted munia under various light–dark conditions such as 12-h light/12-h dark (LD12:12), 24-h light/24-h dark (24L24D), and constant light (LL) conditions (12). In many animals, however, circannual rhythms are expressed only within a narrow range of environmental conditions. For example, clear circannual rhythms are observed under LD12:12 conditions, but not under 18-h light/6-h dark (LD18:6) conditions in the willow warbler (Phylloscopus trochilus) (27). Therefore, to test the presence of an endogenous circannual clock, medaka kept under short-day and warm (SW: 10-h light/14-h dark, 26 °C) conditions for 1 mo were transitioned to long-day and warm (LW: 14-h light/10-h dark, 26 °C) conditions or equatorial (EQ: 12-h light/12-h dark, 26 °C) conditions, or kept under SW conditions (SI Appendix, Fig. S1C). The numbers of egg-laying individuals were counted every morning from Monday to Friday for ~500 d (Fig. 4A). The amplitude of the egg-laying rhythmicity under the LW condition was larger than under the EQ and SW conditions (Fig. 4A). Similarly decreased amplitude of circannual rhythms is often observed under decreased photoperiods due to a reduced amplitude of gonadal change (1). Although the free-running period of circannual rhythms is usually relatively close to 12 mo, it varies from about 6 to 16 mo depending on the species and experimental conditions (1). When we performed a Lomb–Scargle periodogram analysis (28, 29), the free-running circannual periods of medaka were approximately 6 mo long under all the conditions examined (Fig. 4 B and C). Interestingly, this period tended to become shorter as the photoperiod became shorter.

Fig. 4.

Free-running circannual rhythms and circannual genes in medaka. (A) The number of egg-laying individuals under long-day and warm (LW: 14-h light/10-h dark, 26 °C) conditions, equatorial (EQ: 12-h light/12-h dark, 26 °C) conditions, and short-day and warm (SW: 10-h light/14-h dark, 26 °C) conditions. (B) Actograms of medaka kept under LW, EQ, and SW conditions generated by ActogramJ from (A). Red triangles and dotted lines indicate the onsets of reproductive activity. (C) Free-running circannual period of medaka kept under LW, EQ, and SW conditions. Red lines indicate statistical significance (P < 0.01). (D) Effects of summer-like stimulus (16-h light/8-h dark, 26 °C) on gonadosomatic index (GSI) changes at different times of year. (E) Changes in GSI under constant LW conditions. These individuals were used for RNA-seq analysis. The GSI value represents the mean ± SEM (one-way ANOVA, F_(6,14) = 26.6, P < 0.001, Dunnett’s test, **P < 0.01 vs. day 0, n = 3). (F) Heatmap showing gene expression of circannual genes under constant LW conditions. The color scale represents normalized signal intensity. (G) Expression profiles of hormone genes under constant LW conditions. Each point represents the expression level in a single individual. Red lines represent trend lines determined by a cosine curve. (H) Venn diagram showing the overlap of nsSOGs and circannual genes. (I) Gene ontology analysis of 98 genes characterized as both nsSOGs and circannual genes.

In the European starling (Sturnus vulgaris) and some other species, time-of-year dependent effects of external stimuli on gonadal size have been reported, which are considered to reflect the different phase shifts of circannual rhythmicity (15). Next, at intervals of 2 mo, we examined the effects of summer-like stimuli (16-h light/8-h dark and 26 °C for 2 wk) on GSI changes in medaka kept under natural outdoor conditions (SI Appendix, Fig. S1D) and showed that these effects varied throughout the year (Fig. 4D). Together, these results suggest the existence of an endogenous circannual clock in medaka.

The Transcriptional Program of the Circannual Clock.

Because circannual rhythms were most obvious under LW conditions, we transferred medaka from SW conditions to LW conditions, and samples of brain regions containing the hypothalamus and pituitary gland (SI Appendix, Fig. S4A) were collected in an effort to identify the transcriptional program of the circannual clock (Fig. 4E and SI Appendix, Fig. S1E). Because we observed a free-running circannual period of approximately 6 mo, we collected samples every month, covering six time points. Although the existence of individual variation was expected in the free-running period, GSI change under constant conditions reflected a bimodal pattern in the natural outdoor environment (i.e., a transient drop of GSI was observed between the two peaks in spring and summer) (Figs. 1B and 4E). A nonlinear regression analysis identified 518 “circannual genes” (Fig. 4 F–H and Dataset S7), of which 98 overlapped with the previously identified nsSOGs. GO analysis of these genes highlighted enrichment of genes related to “cell proliferation,” “cell differentiation,” “development,” “neurogenesis,” and “morphogenesis,” suggesting the involvement of tissue remodeling in circannual time-keeping (Fig. 4I).

Discussion

To cope with dynamic seasonal changes in the environment, animals have evolved endogenous circannual clock and show robust annual changes in their physiology and behaviors. In winter, medaka rarely eat and have greatly reduced metabolic activity, and hence they become less active and stay on the riverbed. By contrast, they become active and reproduce during summer (21–23). In the present study, we first demonstrated clear seasonal changes in gonadal size in medaka kept under natural outdoor conditions. We applied a multiple linear regression model to estimate the contribution of input stimuli to the generation of annual reproductive rhythms under these conditions. Gonadal size had a stronger influence than external environmental factors such as photoperiod, WT, and SR, suggesting that medaka behavior is not dependent simply on external environmental factors. This result appears to support the notion that an endogenous program plays a significant role in the seasonal rhythms of animals in the field, even though environmental factors modify this program (1).

Our analysis of monthly transcriptomes revealed temporal dynamics in hormonal changes that appeared to optimize reproductive success and survival. We found that many genes involved in ribosome function exhibited an expression peak in winter. Ribosomes are the site of protein synthesis, which is fundamental to the maintenance of vital functions. It has been reported that in plants, the expression of ribosome-related genes increases in winter under natural conditions (24), and that ribosome remodeling plays an important role in cold acclimation (30–32). Hence, medaka may adapt to the cold winter environment by increasing the expression of these genes in winter months, suggesting that ribosomal remodeling is important to low-temperature adaptation in animals as well. In addition, many genes involved in chromatin assembly and DNA replication show expression peaks in summer. Both of these processes are tightly linked to cell proliferation since proliferating cells must duplicate their DNA and replicated DNA must be assembled as chromatin; therefore, these gene expression peaks suggest that cell proliferation occurs in summer. Of note, seasonal changes in hypothalamic cell proliferation have been reported in mammals (33, 34).

Seasonal modulation of the circadian clock gene expression has been proposed to act as a circadian coincidence timer for photoperiodic time measurements (25). In the present study, we found seasonal and daily oscillation of circadian clock genes (Fig. 2 B and D). The analysis of transcriptomes obtained on the equinoxes and solstices demonstrated a decreased amplitude of daily oscillation in winter and spring. It is interesting to note that a similarly reduced oscillation of circadian clock genes in winter has been reported not only in fish (23) but also in plants (24). As the circadian clock is important for all seasons, posttranslational regulation of the circadian clock protein may be important during winter.

The endogenous period of the circannual rhythm is often close to 10 mo (1, 35). In this study, however, medaka exhibited a free-running circannual period of about 6 mo. Compared with the overall range of free-running circadian periods, that of circannual rhythms is rather large, and several species, which include fish, birds, and mammals, exhibit much shorter free-running periods (5 to 6 mo) (1). This is not surprising because the ranges of entrainment of circannual rhythms are obviously much larger than those of circadian rhythms. For example, the range of entrainment of circadian rhythms normally extends from 25% shorter to 25% longer than 24 h (i.e., about 18 to 30 h). By contrast, reduction or expansion of the zeitgeber period by 50% or more or by 200% or more, respectively (where zeitgebers are environmental cues that are capable of entraining rhythms), are usually followed by corresponding reduction and expansion of the circannual period (1). It should also be noted that some organisms that exhibit “typical” circannual cycles under some conditions demonstrate much shorter rhythms under other conditions. For example, the common dormouse (Glis glis) tends to have highly variable circannual rhythms under 5 °C, with an average period of about 5 mo (range 28 to 425 d) (36), but shows a rather consistent and typical circannual cycle under 12 °C, with a period of 12 to 13 mo (37). In addition, golden-mantled squirrels kept for 34 mo under 9.5 °C had a considerably longer circannual period (~2 mo) than those kept at 21 °C (38). Although “temperature compensation” is an important characteristic of circadian rhythms (39–41), resulting in a stable free-running period under different temperature conditions, only a certain degree of temperature compensation has been reported in a limited number of organisms for annual cycles (1, 2), and the circannual rhythms of several species are indeed temperature dependent (36–38). In the present study, we monitored free-running circannual rhythms under 26 °C. In their natural habitat, however, medaka can live under a wide range of water temperatures, from 4 °C to 35 °C (42). Therefore, it is possible that medaka may exhibit a much longer free-running circannual period under certain water temperature conditions.

The studies of circannual rhythms are time consuming and challenging. As a result, the mechanism underlying the circannual clock remains a mystery, although several hypotheses have been proposed. Our monthly transcriptome analysis under constant photoperiod and temperature conditions finally identified genes related to circannual rhythm, and our GO analysis suggested the involvement of tissue remodeling in circannual time-keeping, supporting the “histogenesis hypothesis.” However, our data do not rule out the involvement of other mechanisms, such as binary switching and epigenetic mechanisms, in circannual time-keeping. For instance, we identified epigenetic factors (e.g., kdm6b and kdm7a) in circannual genes.

The pars tuberalis (PT) of the pituitary gland is a key regulator of seasonal rhythms in birds and mammals (43–45) and has been proposed to be the center of the circannual rhythm (17). However, fish do not possess anatomically distinct PT. In masu salmon, the saccus vasculosus (SV) acts as a seasonal sensor (46). Interestingly, medaka neither have SV nor PT. Further investigation of the identified circannual genes could provide a broad avenue for understanding the circannual clock localization in fish and the mechanism by which cyclic histogenesis results in long-term circannual rhythms.

Materials and Methods

Animals.

Medaka fish (Oryzias latipes) were obtained from a local dealer (Fuji 3A Project, Nagoya, Japan). In the experiments under natural conditions, medaka were kept in aquarium tanks installed at the National Institute for Basic Biology (35.0N, 137.2E) or Nagoya University (35.2N, 137.0E). In the experiments under artificial conditions, medaka were kept under short-day and warm (SW: 10-h light/14-h dark, 26 °C) conditions, long-day and warm (LW: 14-h light/10-h dark, 26 °C) conditions, or equatorial (EQ: 12-h light/12-h dark, 26 °C) conditions in a housing system (MEITO system, Meito Suien; LP-30LED-8CTAR, NK system). Fish at each time point were transferred into a small tank prior to sampling. Medaka were anesthetized with 0.05% 3-aminobenzoic acid ethyl ester methanesulfonate salt (MS-222) for sample collection. For sampling during the dark phase, the medaka were anesthetized and decapitated in the dark. All animal studies were carried out in accordance with the ARRIVE guidelines, and all methods complied with relevant guidelines and regulations and were approved by the Animal Experiment Committee of Nagoya University and the National Institutes of Natural Sciences.

Meteorological Data Acquisition and GSI Measurement.

Day length and solar radiation data were obtained from the meteorological station at Nagoya by the Japan Meteorological Agency. Water temperature data were obtained using a water thermometer (TR-52i, T&D Corporation) installed in the aquarium tank. Medaka raised in aquarium tanks were killed at the time of sampling, and body and gonad weights were measured using a microbalance (Practum213-1S, Sartorius or XSR105DUV, Mettler Toledo). Gonadosomatic index (GSI) was calculated by dividing gonad weight by body weight and multiplying by 100.

Regression Analysis.

Regression analysis was performed according to a previous report (47). First, time-series data of average GSI values sampled approximately every 2 wk were interpolated and converted to daily data. Time-series data for the four types of inputs, specifically solar radiation (SR, i = 1), water temperature (WT, i = 2), day length (DL, i = 3), and GSI (i = 4), were normalized to have a mean of 0 and variance of 1. The regression weights w_{i, j} (i = 1, …, 4; j = −T, …, 0) in Eq. 1 were estimated via ridge regression by minimizing the following cost function E:

$$E={\displaystyle \sum }_{t=0}^F{\left({\displaystyle \sum }_i^4{\displaystyle \sum }_{j=0}^T{w}_{i,j}{\text{input}}_{i,t-j}-GS{I}_t^{\text{'}}\right)}^2+\lambda {\displaystyle \sum }_i^4{\displaystyle \sum }_{j=0}^T{w}_{i,j}^2$$ [2]

where λ is a positive constant that controls the strength of the regularization term. In our regression analysis, λ was tuned to minimize the regression error in the test dataset.

RNA-Seq Analysis.

In the experiments under natural conditions, brains were collected from 10 fish every month from October 2015 to October 2017 or every 4-h at four seasons (June 2018, September 2018, December 2018, and March 2019). Five individuals were pooled as one biological replicate, and two replicates were used for the RNA-seq analysis at each time point. In the experiments under artificial conditions, brains were collected from two to three fish every month for 6 mo. Brains were immersed in RNAprotect Tissue Reagent (QIAGEN) for 1 d, and the portion of the brain containing the hypothalamus and pituitary was dissected (SI Appendix, Fig. S4A). Total RNA was prepared using the RNeasy Micro Kit (QIAGEN) and treated with DNase I (QIAGEN) to remove genomic DNA. For library construction, mRNA was first purified from the total RNA using poly-T oligo-attached magnetic beads and then fragmented, after which cDNA synthesis was performed. The synthesized cDNA was used for the preparation of PE100 strand-specific sequencing libraries, which were subsequently sequenced using the Illumina HiSeq platform to generate 50 to 60 million reads for each library (BGI). Clean reads were mapped on the Oryzias latipes reference assembly using bowtie2 (version 2.2.5), and the fragments per kilobase of exon per million fragments mapped (FPKM) were calculated using RSEM (version 1.2.12).

Curve Fitting with Nonlinear Regression.

R 3.5.2 (https://www.r-project.org) was used to analyze the time-series expression data obtained by RNA-seq. First, FPKM values were log-transformed by log₂(FPKM + 0.001), and genes with an average value > −6 across all sampling time points were designated as expressed. Then, a cosine curve was fitted to the expression data of each expressed gene using the Nonlinear Least Squares (nls) function in R. The coefficient of determination (R-squared, RS) was also obtained from the sample values to evaluate the goodness of fit. The amplitude (alpha) of the fitted curve was defined as the difference between the maximum and minimum of the seasonal oscillation, and the time or day at the maximum of the fitted curve was defined as the phase (phi), which represents the expression peak of the gene. For the analysis of 2-y samples under natural conditions, genes with RS higher than 0.2 and an amplitude greater than 2 were extracted. The threshold for RS was determined by examining the coefficient for the correlation between the distributions of the seasonal oscillation phase in the first and second years, and adopted RS > 0.2, which resulted in correlation coefficient greater than 0.8. For the analysis of DOGs, genes with RS higher than 0.7 and an amplitude greater than 1.5 were extracted because only one data cycle was available for this analysis. For the analysis of circannual genes, genes with RS higher than 0.3 and an amplitude greater than 1.5 were extracted. In this analysis, the threshold for RS was set lower than that for the analysis of DOGs because individual variation in the free-running period of circannual clock is much larger than that of circadian rhythms (1).

GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) Enrichment Analyses.

In the GO enrichment analysis of bsSOGs, we prepared original GO annotations for the medaka transcriptome. We searched for homologues of every medaka gene using BLAST searches (blastp) for all coding sequences in mice (Mus musculus). We used the gene with the best hit and with an e value < 1e−5. We reconstructed ancestral GO terms from the present GO terms using the GO.db package in the bioconductor and assigned them to corresponding genes. Statistical tests of enrichment were performed using the Fisher’s exact test function provided by R statistical software. P values for the results were corrected using false discovery rate (FDR) correction (48). GO analyses of circannual genes and of DOGs in each season were performed using Metascape 3.5 (26). From the ortholog table of medaka genes and mouse genes obtained by BLAST searches (blastp), medaka gene symbols were converted to mouse gene symbols and analyzed using the Express Analysis pipeline with default parameters. In the GO analysis of DEGs from different seasons, the reference sequence was aligned against the NCBI nonredundant protein database using the blastx function in the DIAMOND program (49), and the associated GO annotations for the best homologues were extracted. The same DIAMOND blastx function was used to search for the annotated proteins in the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Next, the identified DEGs were classified based on these GO and KEGG annotations, after which enrichment analysis was performed using the phyper function in R, and the FDR was calculated for each P value. We defined pathways with an adjusted P < 0.01 as significantly enriched.

Analysis of Circannual Rhythms under Constant Photoperiod and Temperature Conditions.

Before the experiment, medaka were habituated under SW conditions for at least 1 mo and then transferred to LW conditions. After the transition, the number of egg-laying individuals was counted every weekday morning from 09:00 to 11:00. ActogramJ was used to draw actograms, perform periodogram analysis, and calculate reproductive activity onset (50). The Lomb–Scargle method was applied in the periodogram analysis.

Effects of Long-Day and Warm Conditions on GSI Changes.

Medaka kept under natural outdoor conditions were transferred to artificial summer-like (16-h light/8-h dark, 26 °C) conditions every 2 mo in 2022. Before and 2 wk after the transition, 3 to 10 female medaka were killed, and their body and gonad weights were measured on a microbalance to calculate GSI. The change in GSI was calculated by subtracting GSI before the condition transition from that after the condition transition.

Statistical Analysis.

One-way ANOVA with Dunnett’s post hoc test was used to analyze GSI.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Dataset S06 (XLSX)

Dataset S07 (XLSX)

Data, Materials, and Software Availability

The RNA-seq data generated in this study have been deposited in the NCBI’s Gene Expression Omnibus (GEO) (accession numbers GSE234400, GSE234401, and GSE234565) (51–53). All codes used in the analysis are available at https://github.com/animalphysiology/script_medaka2023.git (54).