Deep learning to assess erythritol in zebrafish development, circadian rhythm, and cardiovascular disease risk

Abstract

Erythritol is one of the most widely used artificial sweeteners, yet the potential risks remain a subject of debate. In this experiment, the concentration of erythritol was determined according to pre-experiments (3, 5, 7 mg/mL), and exposure was initiated at 2 h post-fertilization of embryonic development until the end of 120 hpf. Here, we employed a deep learning approach to uncover the time and tempo aspects of embryonic development. Erythritol may contribute to motor behavioral impairments in larvae by delaying the development of the swim bladder. Furthermore, erythritol causes circadian rhythm disorder and affects the expression of the rev-erba, which leads to thrombus formation. The formation of thrombi is accompanied by oxidative stress and inflammation. The above findings contribute to the understanding of the effects of erythritol on zebrafish development, neurobehavior, and cardiovascular disease and provide a basis for the safety of erythritol as an artificial sweetener.

Introduction

The disruption of circadian rhythms is a pervasive phenomenon of contemporary lifestyles¹. Circadian rhythm disorder leads to alterations in the biological clock. A distinctive feature of biological clock disorders and circadian rhythm disruption is the sleep-wake disorder. Sleep-wake disorder is a pervasive phenomenon in contemporary society. The main characteristics of this condition are decreased sleep and increased activity. Sleep-wake abnormality triggers numerous pathological processes, including metabolic disorders, reduced immunity, and the development, metastasis, and spread of cancers^2,3,4,5,6. Moreover, sleep-wake disorder impairs insulin sensitivity, raising the likelihood of getting type 2 diabetes⁷. Nevertheless, further research is required to elucidate the relationship between sleep-wake abnormalities and cardiovascular disease, particularly thrombosis.

Cardiovascular disease (CVD) is one of the most common diseases worldwide⁸. One of the main pathogenic factors contributing to morbidity and death in a range of CVDs is thrombosis⁹. Generally, thrombosis is mainly divided into arterial thrombosis and venous thrombosis^10,11. Platelet activation is a key step in arterial thrombosis^12,13. Circadian rhythm disruption is an important factor leading to platelet abnormalities that increase thrombosis⁹. In a recent investigation, the consumption of artificial sweeteners was shown to be associated with thrombosis¹⁴. The finding has led to increased vigilance in the consumption of artificial sweeteners.

In an effort to cut back on calories and sugar intake, artificial sweeteners have been added to a wider range of foods^15,16. With the global epidemic of obesity and metabolic diseases such as diabetes, people are often advised to use artificial sweeteners as sugar substitutes to lose weight and improve blood glucose levels^17,18. As a result, artificial sweeteners are becoming increasingly common in confectionery, beverages, and processed foods^19,20. Indeed, research has revealed that tap and groundwater contain artificial sweeteners²¹. Data from the 2013–2014 National Health and Nutrition Examination Survey (NHANES) and the US FDA show that the average daily intake of erythritol per person is 30 g/day^14,22. An FDA document entitled ‘GRAS Determination of Erythritol for Use in Human Food’ set the 90th percentile intake of erythritol for all users at 63.0 g/person/day²². There is growing evidence that artificial sweeteners are not safe and have been linked to the development of many metabolic diseases, such as insulin sensitivity²³, weight gain^24,25, increased urine output²⁶, psychiatric disorders^27,28, type 2 diabetes²⁹, cardiovascular disease^30,31,32, and even cancer³³. Therefore, research on artificial sweeteners still needs to be explored.

Four-carbon polyol erythritol is frequently used as a sugar substitute³⁴. It is widely used in food processing because of its low caloric content^35,36. One study has shown that erythritol is associated with adverse cardiovascular risk events¹⁴. This study showed that erythritol leads to platelet aggregation, which in turn leads to the development of thrombi¹⁴. However, the reason why erythritol causes platelet aggregation is not clear. To determine how erythritol affects other factors, further investigation is needed.

The difference in this study compared to previous methods is the use of deep learning models instead of traditional statistical methods. In this study, a pretrained Twin Network model was used to infer and analyze the similarity between embryos’ images³⁷. Twin Network is a computational method based on Twin Network, primarily used to calculate similarities between complex input vectors. Recently, researchers employed the Twin Network to create a dataset. They were able to precisely ascertain the stage of development of zebrafish through utilizing this dataset to build the Twin Network model utilizing image sequences. After a series of tests, the Twin Network model also proved effective in characterizing natural variability in zebrafish development, robustly identifying embryos that developed abnormally, detecting small-molecule-induced modifications to development’s phenotype, and automatically delineating major developmental stages and transitions³⁷. The improvement in this study is the higher accuracy achieved in predicting the effects of artificial sweeteners through the deep learning approach. The deep learning model demonstrated enhanced capacity for staging embryos with greater precision and for objectively quantifying the timing and rhythm of embryo development. This results in more reliable predictions.

The zebrafish has emerged as a valuable model organism for toxicological evaluations in the field of food science³⁸. Here, we used deep learning to explore the effects of erythritol on embryonic development. Furthermore, the impact of erythritol on larval locomotor behavior, circadian rhythms, and cardiovascular disease was examined. The objective of this study was to assess the potential risks associated with erythritol and to raise awareness about the consumption of artificial sweeteners.

Results

Using deep learning to reveal the time and tempo of embryonic development

The objective of this study was to analyze the impact of varying concentrations of erythritol on the developmental process of zebrafish embryos (Fig. 1A). Embryos were treated with erythritol at 2 hpf, and their development was observed at 3, 6, 10, and 24 hpf. Using a deep learning approach, we can stage zebrafish embryos in an unsupervised manner (Fig. 1B). Twin Networks was applied to stage the embryos exposed to different concentrations of erythritol groups at the blastula stage, gastrula stage, and segmentation stage. The results indicated that the embryos developed differently between the treatment group and the control group (Fig. 1C). Next, we used Twin Networks to reveal the rhythm of embryo development. We found that the experimental time was 6 h before and tested for the existence of a delay in the development time of embryos in the treatment group compared to the control group. The result indicates that there is a delay in the development of zebrafish embryos by erythritol exposure during this time period. However, it’s noteworthy to observe that at 24 h, the embryonic development of the erythritol-treated group became faster compared to the control group (Fig. 1D, E).

Fig. 1: Utilizing deep learning Twin Networks to delineate embryonic developmental periods and reveal embryonic developmental rhythms.

A Flowchart of the entire experiment. Some of the material in (A) has been taken from Figdraw, with permission from Figdraw (license code: zzSSS0bb64). The image is created by the author, and there is no copyright or conflict of interest. B Schematic diagram of embryo age prediction. The test image (top panel) is compared with a sequence of reference images (middle panel) of known chronological order. The age of the test image corresponds to the age of the embryo image with the highest similarity (light blue dashed line) to the input data; relative (dark blue curve) similarity. The blue shading indicates the width of the peak. Some of the material in (B) has been taken from Figdraw, with permission from Figdraw (license code: zzSSS0bb64). The image is created by the author, and there is no copyright or conflict of interest. C Similarity plots of test embryos (top) versus reference images (bottom). Each test embryo was compared to three sets of reference images. The average of the cosine similarity with these reference image sets was plotted as a data point for each reference image time point. The box plots are based on distributions with similarity values above 0.8. The center of the graph represents the median, the box boundaries represent the upper and lower quartiles, the whiskers represent the 1.5-fold interquartile range, and the red dot represents the curve maximum. The three images in the figure are from a single embryo collection and represent three independent experiments. D, E Automated analysis of embryo development at different concentrations using Twin Networks.

Validating embryonic developmental self-similarity

We validated the developmental self-similarity of embryos using Twin Networks³⁷. We tested the similarity of control and treated embryos at different developmental times (Fig. 2A). At 3 hpf, there was no significant difference in embryo similarity between the control and treatment groups (H = 1.4368, p = 0.6969, Fig. 2B). At 6 hpf, there was a significant difference between the control and erythritol-treated groups, and the different concentrations of erythritol-treated groups showed a decrease in similarity, with a highly significant decrease in similarity in the 5 mg/mL erythritol-treated group (H = 22.206, p < 0.0001, Fig. 2C). As the embryos developed to 10 hpf, the cosine similarity of 3 mg/mL and 7 mg/mL embryos was significantly reduced compared to the control group (H = 23.125, p < 0.0001, Fig. 2D). At 24 hpf, cosine similarity was significantly decreased in 5 mg/mL and 7 mg/mL embryos compared to controls (H = 27.771, p < 0.0001, Fig. 2E). We then performed a heat map analysis of the different concentration treatment groups based on the embryo similarity results (Fig. 2F).

Fig. 2: Validation of embryo developmental self-similarity using deep learning Twin Networks.

A Schematic diagram of similarity calculation among embryos at the same developmental stage under different concentrations of erythritol treatment. B–E A comparison of cosine similarity between control and treated embryos at 3, 6, 10, and 24 hpf, respectively. F A heat map analysis of embryo similarity under different concentrations of erythritol treatment. Some of the material in (F) has been taken from Figdraw, with permission from Figdraw (license code: zzSSS0bb64). The image is created by the author, and there is no copyright or conflict of interest. N = 3. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Effects of erythritol on embryo development

We also examined and analyzed the morphology of 96 hpf larvae in order to gather more information regarding the larval toxicity of erythritol (Fig. 3A). We found that morphological abnormalities could be clearly seen after treatment with different concentrations of erythritol (3, 5, and 7 mg/mL) compared with the control group. These abnormalities included shorter body length [F (3, 8) = 11.31, p = 0.0030, Fig. 3B], smaller eye area [F (3, 8) = 37.24, p < 0.0001, Fig. 3C], smaller head area [F (3, 8) = 88.94, p < 0.0001, Fig. 3D], and delay in the swim bladder [F (3, 8) = 75.44, p < 0.0001, Fig. 3E, F]. According to the above results, erythritol may delay the development of zebrafish larvae.

Fig. 3: The impact of erythritol on embryonic development.

A Morphological characteristics of different concentrations of erythritol exposed to 96 hpf. B–F Comparison of body length, eye area, head area, and swim bladder area of 96 hpf zebrafish larvae exposed to different concentrations of erythritol. Data are represented as mean ± SD. N = 3; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Effect of erythritol on the motility of zebrafish larvae

The swim bladder is an important organ for diffusive movement in zebrafish³⁹. Therefore, we further investigated the locomotor behavior of zebrafish larvae. We investigated the effect of erythritol exposure on the thigmotaxis behavior of 5 dpf zebrafish larvae in both light and dark environments (Fig. 4A). The results showed that erythritol exposure resulted in increased thigmotaxis of larvae in both light and dark situations [F (3, 8) = 11.94, p = 0.0025, Fig. 4B; F (3, 8) = 152.7, p < 0.0001, Fig. 4D]. The movement trajectories of the larvae at different time points were then analyzed (Fig. 4F, G). We further analyzed the parameters of locomotor behavior. Compared with the control group, the movement distance of the larvae in the erythritol-exposed group decreased significantly [F (3, 8) = 16.38, p = 0.0009, Fig. 4C; F (3, 8) = 20.12, p = 0.0004, Fig. 4E]. Furthermore, the results showed that the transcript levels of th, mao, and dbh were significantly reduced in the treated group compared with the control group [F (3, 8) = 203.1, p < 0.0001, Fig. 4H; F (3, 8) = 33.17, p < 0.0001, Fig. 4I; F (3, 8) = 165.3, p < 0.0001, Fig. 4J].

Fig. 4: The effect of erythritol on the locomotion of zebrafish larvae.

A Flowchart of locomotion of zebrafish larvae. The image is created by the author and there is no copyright or conflict of interest. B–E Quantify the percentage of 5 dpf zebrafish thigmotaxis behavior (B, D) and distance (C, E) under light and dark conditions. F The colored two-dimensional (2D) and three-dimensional (3D) images represent the movement trajectories of zebrafish larvae during the day. G The red images represent the movement trajectories during the night. H–J The mRNA levels of genes related to diffusive motor behavior, including th (H), mao (I), and dbh (J). Data are represented as mean ± SD. N = 3; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Erythritol interferes with circadian rhythms in larval zebrafish

To further investigate the effect of erythritol exposure on neurobehavior, we monitored the circadian behavior of the larvae for 48 h (Fig. 5A). The results showed that erythritol exposure significantly increased activity and decreased the resting time (Fig. 5B, C). On this basis, K-means cluster analysis (K = 2) was used to evaluate the correlation between the changes in larval behavior under different concentrations of erythritol exposure (Fig. 5D). It was found that total activity and waking activity were significantly increased in the erythritol-treated group compared with the control group (Fig. 5H, I). In addition, the total resting time of the treated group was shortened (Fig. 5E), the number of resting bouts at night was increased (Fig. 5F), the number of resting bouts length during the day were decreased (Fig. 5F), and the length of resting rounds was decreased (Fig. 5G). Collectively, this suggests that the circadian rhythms of zebrafish are abnormal after erythritol exposure.

Fig. 5: Erythritol interferes with circadian rhythms in larval zebrafish.

A The sleep/wake experiment flowchart. The image is created by the author, and there is no copyright or conflict of interest. B, C Normalized waking activity (B) and normalized total rest (C). The exposed groups’ and the controls’ mean values were shown by the red and blue lines, accordingly. The on and off lighting was represented by the black and white bars. D The result of K-means clustering analysis (K = 2). The black bar signifies the night, the white bar symbolizes the day, and the rows reflect various sleep/wake factors. The various erythritol treatment concentrations are shown in columns. Red denotes higher than the control group, and green shows lower than the control group for these metrics, which are standardized with the control group. E–I Significance analysis of rest total (E), the number of rest bouts (F), rest bouts length (G), total activity (H), and waking activity (I), in the treated and the control groups. Each value represents an average of 16 larvae. The lighting was represented by the black and white bars, respectively. J The five behavior parameters’ fingerprints. Rows represented the various erythritol-treated groups. Various behavior parameters were displayed in columns. The standard deviation from the control group was used to standardize the data. They were either greater or lesser than the controls, as represented by the colors red and blue. The lighting was represented by the black and white bars, respectively. K Schematic diagram of clock genes. The image is created by the author, and there is no copyright or conflict of interest. L–N Core clock genes bmal1b, clock1a, and rev-erba: relative mRNA expression in the daily rhythm. N = 3; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Since circadian rhythms are controlled by endogenous biological clocks synchronized with external time signals related to biological clock factors, we further investigated the clock genes bmal1b, clock1a, and rev-erba⁴⁰. The circadian changes in the transcript levels of the important clock genes bmal1b, clock1a, and rev-erba were detected (Fig. 5K–N). We found that bmal1b expression induced by erythritol 3 mg/mL was significantly decreased at ZT17, ZT23, and ZT5 (Fig. 5L). Erythritol 5 mg/mL and 7 mg/mL induced a significant decrease in bmal1b expression at ZT23, ZT5 (Fig. 5L). Interestingly, bmal1b expression induced by 7 mg/mL erythritol was significantly upregulated at ZT11 (Fig. 5L). Different concentrations of erythritol treatment groups induced a significant decrease in clock1a expression both at night and during the day at ZT5 (Fig. 5M). Interestingly, clock1a expression induced by erythritol 3 mg/mL and 5 mg/mL was significantly downregulated at ZT11, and clock1a expression induced by 7 mg/mL was significantly upregulated (Fig. 5M). Erythritol 5 mg/mL-induced rev-erba expression was significantly increased at night (Fig. 5N). Erythritol 7 mg/mL induced significantly increased rev-erba expression at ZT17 and during the day, but decreased at ZT23 (Fig. 5N). Thus, erythritol may disrupt the rhythmic oscillatory expression of the clock genes bmal1b, clock1a, and rev-erba, leading to disrupted circadian rhythms.

Erythritol exposure causes thrombus

The research found that increased expression of rev-erba leads to platelet aggregation by reviewing the published literature⁹. Consequently, larvae were stained, and the results indicated that erythritol reduced red blood cells in the heart [F (3, 8) = 87.17, p < 0.0001, Fig. 6A, B]. Next, we applied qPCR arrays to identify the expression of a specific panel of platelet activation and coagulation factors, such as fgb and f2. Erythritol (5 mg/mL and 7 mg/mL) induced significantly higher levels of f2 and fgb transcripts compared to control [F (3, 8) = 5.571, p = 0.0233, Fig. 5C; F (3, 8) = 13.32, p = 0.0018, Fig. 6D]. Next, we performed analysis of the genes ophn1, rhoa, and ezrin downstream of rev-erba (Fig. 6E). It was discovered that erythritol caused a notable decrease in ophn1 expression [F (3, 8) = 138.6, p < 0.0001, Fig. 6F]. Erythritol 3 mg/mL induced a significant decrease in rhoa, but erythritol 5 mg/mL and 7 mg/mL induced a significant increase in rhoa expression [F (3, 8) = 32.98, p < 0.0001, Fig. 6G]. In addition, erythritol induced a significant decrease in ezrin expression [F (3, 8) = 57.93, p < 0.0001, Fig. 6H]. The above experiments indicate that exposure to erythritol leads to abnormal expression of thrombosis-related genes, which may contribute to the occurrence of thrombosis. This is also consistent with previously published studies¹⁴.

Fig. 6: Erythritol exposure causes thrombosis in zebrafish larvae.

A Representative images of o-dianisidine staining in control and different concentrations of erythritol groups. The heart regions and tail regions are delineated by a red dotted line. B Quantitative analysis of the intensity of cardiac erythrocytes. C, D Relative mRNA expression of coagulation factor II (f2) and fibrinogen β chain (fgb). E Schematic diagram of platelet aggregation due to increased expression of the rev-erba gene. Some of the material in (E) has been taken from Figdraw, with permission from Figdraw (license code: zzSSS0bb64). The image is created by the author, and there is no copyright or conflict of interest. F Relative mRNA expression of oligophrenin-1 (ophn1). G Relative mRNA expression of rhoa. H Relative mRNA expression of ezrin. Data are represented as mean ± SD. N = 3; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Erythritol induces oxidative stress and inflammation

Oxidative stress and inflammation are one of the triggers of cardiovascular disease and a common trigger of metabolic abnormalities in the body⁴¹. We examined ROS production in zebrafish treated with different concentrations and analyzed the expression of genes related to oxidative stress and inflammation (Fig. 7). We used the DCFH-DA to monitor oxidative stress. The DCFH-DA signal was elevated in zebrafish in the erythritol-treated group compared with the control group, suggesting that it may lead to oxidative stress [F (3, 8) = 60.32, p < 0.0001, Fig. 7A, B]. The transcript levels of oxidative stress-related genes (sod1, cat) were also changed by erythritol [F (3, 8) = 83.13, p < 0.0001, Fig. 7C; F (3, 8) = 57.70, p < 0.0001, Fig. 7D; F (3, 8) = 8.597, p = 0.0070, Fig. 7E]. In the 7 mg/mL treatment group, the transcript levels of sod1 were significantly suppressed (Fig. 7C).

Fig. 7: Erythritol induces oxidative stress and inflammation.

Representative pictures (A) and quantifications (B) of the DCFH-DA stained fluorescence signals in various exposure groups. C–E The relative mRNA expression of oxidative stress-related genes in larval zebrafish, such as sod1 (C), cat (D), gpx1 (E). F–I The zebrafish larvae’s relative mRNA expression of genes related to inflammation, such as the pro-inflammatory factors IL-6 (F) and IL1-β (G), as well as the anti-inflammatory factors IL-10 (H), and IL-4 (I). Data are represented as mean ± SD. N = 3; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

The generation of thrombi is typically related to an inflammatory response⁴². Therefore, we further analyzed the transcript levels of pro-inflammatory factors (IL-6, IL1-β) and anti-inflammatory factors (IL-10, IL-4) (Fig. 7F–I). It was found that erythritol 7 mg/mL induced a significant increase in the expression of IL-6, IL1-β compared with the control group [F (3, 8) = 9.632, p = 0.0049, Fig. 7F; F (3, 8) = 2085, p < 0.0001, Fig. 7G]. In comparison to the control group, the mRNA level of IL-10 was found to be reduced in the 5 mg/mL erythritol group [F (3, 8) = 27.83, p = 0.0001, Fig. 7H]. The expression of IL-4 was significantly decreased in erythritol groups of 3 mg/mL and 5 mg/mL [F (3, 8) = 84.40, p < 0.0001, Fig. 7I].

Discussion

Here, we conducted the assessment of the safety of erythritol as an artificial sweetener. The results demonstrated that erythritol exposure at varying concentrations had a detrimental impact on zebrafish embryonic development, resulting in reduced head and eye area, shortened body length, and delayed swim bladder development. Erythritol exposure resulted in alterations to both thigmotaxis behavior and sleep-wake behavior in zebrafish larvae, indicating that erythritol may exert neurotoxic effects on zebrafish larvae. Furthermore, we conducted an assay of biological clock genes and observed significant effects on the crucial clock genes bmal1b, clock1a, and rev-erba. The gene of rev-erba is associated with platelet formation. It is hypothesized that erythritol may cause platelet aggregation through abnormal sleep arousal and disruption of the clock gene rev-erba, which may lead to thrombosis with oxidative stress and inflammation (Fig. 8).

Fig. 8: Schematic representation of the effects of erythritol on development, neurobehavioral, and cardiovascular risk in zebrafish larvae.

Some of the material in this picture is taken from Figdraw, with the permission of Figdraw (license code: zzSSS0bb64). The image is created by the author, and there is no copyright or conflict of interest.

Erythritol is a natural, artificial sweetener⁴³. It is employed in the production of zero-carbohydrate food products⁴⁴. Nevertheless, a recently published study indicates that erythritol is associated with adverse cardiovascular events¹⁴. This association is attributed to erythritol’s ability to cause platelet aggregation. Additionally, previous studies have demonstrated that erythritol can exacerbate intestinal inflammation and induce anxiety-like behavior⁴⁵. Consequently, further research on erythritol is a crucial undertaking. The paucity of erythritol reporting in the literature may be attributed, in part, to the challenges associated with its quantification, similar to that of other polyols, due to its structural isomers¹⁴. Although the amount of erythritol that can be consumed per person per day is specified in FDA documents, an appropriate daily intake of ‘not specified’ was awarded by the World Health Organization/Food and Agriculture Organization of the United Nations Expert Committee on Food Additives⁴⁶. The FDA does not mandate that the amount of erythritol in food items be disclosed, which makes it challenging to ascertain the levels of erythritol in foods as an additive¹⁴. Furthermore, due to confidentiality policies on food labels, ‘artificial sweeteners’ are often reported in aggregate. This makes it more difficult to identify certain sweeteners on labels and to keep track of the unfavorable long-term effects of specific sweeteners in clinical trials. Additionally, it makes it difficult to relate dietary use of artificial sweeteners to circulating levels. The current findings underscore the need to study the effects of erythritol on the body as the prevalence of consumption continues to rise.

The Twin Network Model is a machine-learning technique that can be used for multimodal tasks in the research of animal development with the goal of automatically and objectively defining developmental processes, and it demonstrates superior precision compared to traditional vector diffusion maps used for image registration. The core idea of it involves the unsupervised calculation of similarities between different states.

The Twin Network Model offers a standardized method for staging and comparing embryos, which is crucial because the accuracy in estimating the age of an embryo greatly impacts developmental biology research. Embryonic stages can affect experimental outcomes as they vary significantly, yet matching an observed embryo to standardized stages in an atlas is challenging due to fluid phenotypic transitions.

The Twin Network incorporates these gradual changes by calculating systematic similarities between a test image and a reference sequence of images. This produces a similarity chart that allows for the precise placement of an embryo within a developmental timeline in the reference. Notably, the Twin Network dynamically integrates phenotypic traits for similarity assessments across various developmental stages rather than relying on static phenotype classes. Additionally, it suggests a directional arrow-of-time, indicating the progression of development.

A deep learning approach was employed in an experimental study, which found that embryos exposed to erythritol exhibited delayed development in the early stages (zygote period, cleavage period, blastula period) and accelerated development in the later stages (gastrula period, segmentation period) compared to controls. The zygote stage, which lasts approximately 0–0.75 h, encompasses the period from fertilization to the first cleavage⁴⁷. The cleavage stage (approximately 0.75–2.25 h) is characterized by multiple consecutive rapid divisions, resulting in the formation of a multicellular embryo⁴⁷. The number of cells increases during this stage, yet the total size of the embryo remains essentially unchanged. The blastula stage (approximately 2.25–5.25 h) is characterized by the transition of the embryo into the mid-embryo sac, the formation of the yolk syncytium layer, and the initiation of the outer movement⁴⁷. In brief, the addition of erythritol acts as a disruptor during the initial stages of embryonic development (zygote period, cleavage period, blastula period), and therefore, erythritol exposure results in delayed embryonic development. With the formation of the yolk syncytium layer at the blastula stage, the outer movement begins. Furthermore, zebrafish larvae are capable of absorbing nutrients through the yolk⁴⁸. Erythritol is a metabolite of the pentose-phosphate pathway⁴⁹. It is hypothesized that in the later stages of embryonic development (gastrula period, segmentation period), embryos may ingest erythritol through the yolk to provide nutrients for the organism.

The results of our experiments indicated that the head area, eye area, body length, and swim bladder development of zebrafish larvae were all found to be delayed in comparison to the control group. The swim bladder is an organ that inflates the body of the larvae³⁹. It has been demonstrated that larvae elevate their buoyancy by enlarging their swim bladders. Furthermore, the swim bladder is also related to the locomotor behavior of the larvae³⁹. The preference of animals to remain or move close to a region’s border is known as thigmotaxis⁵⁰. Our findings demonstrated that the larvae’s thigmotaxis behavior increased and tended to remain in the marginal area in both light and dark circumstances, suggesting that erythritol treatment had an impact on the zebrafish nervous system. TH is necessary for the production of dopamine⁵¹. The enzyme monoamine oxidase (MAO) is responsible for metabolizing neurotransmitters. A reduction in MAO levels indicates that the body is inhibiting the inactivation of neurotransmitters⁵². Dopamine (DA) can be hydroxylated into norepinephrine by dbh⁵³. A review of the literature revealed that the genes of clock1a and rev-erba can regulate TH, and mice lacking either clock1a or rev-erba exhibited elevated midbrain dopaminergic activities and mania-like behavior^54,55. Moreover, another study demonstrated that the circadian clock exerts direct control over mao and dbh, which are involved in DA catabolism and the conversion of DA into NE, respectively⁵⁶. The present experimental study showed that the erythritol-exposed group had increased levels of Rev-erba gene expression and lower levels of mao, the, and dbh transcripts compared to the control group. These results suggest that the transcript levels of mao, the, and dbh are likely co-regulated by the clock genes clock1a, bmal1b, and rev-erba. Furthermore, the reduced transcript levels of mao, the, and dbh in the erythritol-exposed group likely lead to a reduction in DA levels, which may contribute to the development of ADHD.

Behavior is the primary method for assessing the effects of external substances on neural development⁵⁷. As a diurnally active fish, the zebrafish exhibits a distinct sleep/wake behavior, which represents an advanced form of neural activity compared to simple motor behavior⁵⁸. The results of the study demonstrated that the zebrafish exposed to erythritol exhibited a significant increase in locomotor activity, accompanied by a corresponding decrease in resting time and an increase in resting rounds. This indicates that sleep is frequently disrupted in larvae, and it is hypothesized that such disruptions may be associated with an increased risk of cardiovascular disease⁵⁹. Previous research has established a correlation between sleep disturbances and an elevated risk of cardiovascular disease⁶⁰. It is hypothesized that the observed abnormalities in sleep arousal are a consequence of erythritol-induced adverse cardiovascular events¹⁴, which in turn result in circadian rhythm disruption. The biological clock, an endogenous circadian rhythm regulator, is responsible for maintaining circadian rhythms. We examined the expression of key biological clock genes. Bmal1b and Clock1a have been demonstrated to regulate the expression of rev-erbα, while rev-erbα has been shown to inhibit the expression of bmal1b. Our findings indicate that the expression of clock genes is abnormal.

Erythritol has been demonstrated to cause platelet aggregation, which is associated with adverse cardiovascular events¹⁴. Furthermore, erythritol can be synthesized endogenously from glucose⁴⁹. It has been demonstrated that high levels of sugar can enhance the activation of platelets⁶¹. The precise mechanism underlying erythritol-induced platelet aggregation remains elusive. The results of our experiments indicated that the clock gene rev-erba was expressed in an aberrant manner. A related report indicated that increased rev-erba is associated with platelet aggregation⁹. Subsequently, we examined the downstream gene expression of rev-erbα, including that of oligophrenin-1, rhoa, and ezrin⁹. It was determined that all three genes exhibited aberrant expression in the erythritol-exposed group in comparison to the control group. From these observations, we postulate that erythritol-induced platelet aggregation may be related to the aberrant expression of the clock gene rev-erba.

Oxidative stress represents a risk factor for abnormal blood coagulation and thrombosis^62,63. The pathogenic process of thrombotic disorders, such as atherosclerosis, stroke, and myocardial infarction, is influenced by oxidative stress⁶⁴. Therefore, we examined the reactive oxygen species (ROS) and the expression of the sod1, cat, and gpx1 genes in zebrafish. Antioxidant enzymes such as sod1, cat, and gpx1 can help scavenge excess ROS in the body and maintain the dynamic balance of oxidative and antioxidant systems⁶⁵. The results demonstrated that the high erythritol exposure group exhibited a notable reduction in the transcriptional level of sod1 and gpx1, suggesting the generation of oxidative stress in the larvae.

The generation of thrombi is typically accompanied by an inflammatory response⁴². The interplay between factors that promote and inhibit inflammation is a crucial determinant of thrombosis⁶⁶. For instance, inflammatory cytokines, including interleukin-6 (IL-6) and IL1-β, have been linked to an elevated risk of thrombosis. Our analysis of the results indicates that erythritol may contribute to the development of an inflammatory response, which may occur as a consequence of thrombosis.

However, this study has several limitations. Firstly, erythritol, a food-grade artificial sweetener, remains to be studied for other aspects of cardiovascular disease and type 2 diabetes. Secondly, erythritol is challenging to quantify in vivo. Thirdly, the present study only examined the impact of short-term erythritol exposure on zebrafish larvae. In reality, however, individuals are exposed to artificial sweeteners for a considerable amount of time; thus, further investigation into the effects of long-term exposure to artificial sweeteners on zebrafish is warranted.

In summary, our research suggests that erythritol might have an impact on zebrafish embryo development. It has been demonstrated that erythritol may contribute to motor behavioral impairments in zebrafish larvae by delaying the development of the swim bladder. It is possible that erythritol may contribute to platelet aggregation by affecting sleep-wake and biological clocks, which in turn leads to oxidative stress and inflammation. The above findings contribute to the understanding of the impact of erythritol on zebrafish development, neurobehavior, and cardiovascular disease. This is a cause for concern, given that subjects (diabetics, obese patients, and patients with impaired renal function) who are marketed by artificial sweeteners generally have a greater risk of cardiovascular problems in the future. It is imperative to assess the long-term safety of erythritol.

Methods

Animal care ethics

The NIH Guide for the Care and Use of Laboratory Animals (no. 8023, revised 1996) was followed in all experimental methods and procedures involving zebrafish, which were approved by the Animal Experimentation Committee of the College of Life Sciences, Nankai University (no. 2008). This experiment strictly follows ARRIVE guidelines 2.0 on reporting animal studies.

Zebrafish maintenance

The adult zebrafish wild-type AB strain was procured from Nanjing EzeRinka Biotechnology Co., Ltd. A zebrafish recirculating aquaculture system with 0.05 g/L KCl, 0.025 g/L NaHCO₃, 3.5 g/L NaCl, and 0.1 g/L CaCl₂ (pH 7.0–7.2, 28.5 ◦C) was used to raise adult zebrafish⁶⁷. The zebrafish were fed at nine in the morning and seven at night⁶⁷. Standard laboratory conditions were used to maintain the strains of zebrafish⁶⁷. The wild-type AB strain of adult zebrafish was used to obtain wild-type embryos, which were maintained in E3 medium at 28.5 °C and staged as previously reported⁶⁷.

Drug exposure

Erythritol (CAS# 149-32-6, 99% purity) was purchased from Tianjin Hynes Optronics Co. (Tianjin, China). According to the 2013–2014 National Health and Nutrition Examination Survey data and FDA standards (USA FDA 2001)²², human intake of erythritol is up to 30 g/day. Our study also references the concentration of erythritol used in mouse studies by setting up a zebrafish exposure erythritol concentration gradient^45,68,69. We also conducted a pre-experiment at the beginning of this study to determine the experimental dose of erythritol (3 mg/mL, 5 mg/mL, 7 mg/mL) (Supplementary Figs. S2–S4). The powdered erythritol was meticulously weighed and dissolved in E3 medium to yield a 10 mg/mL stock solution. In accordance with the experimental design, the stock solution was subsequently diluted with E3 medium to obtain a series of experimental solutions, each containing distinct concentrations of erythritol (3 mg/mL, 5 mg/mL, and 7 mg/mL). These solutions were then utilized in the subsequent experiments, which were conducted in 6-well plates. At 2 h post-fertilization (hpf), a random number generator was utilized to randomly assign zebrafish embryos to distinct groups. Thirty zebrafish embryos per group were gathered and moved to 6-well plates^70,71,72. The control group exposed to E3 medium (0.037 g/L CaCl₂, 0.013 g/L KCl, 0.081 mg/L MgSO₄⋅7H2O, 0.292 g/L NaCl)⁷³. In contrast to the control group, the erythritol-treated group was exposed to varying concentrations of erythritol solution. This process should be repeated three times for each group. The exposed embryos were kept at 28.5 °C for further incubation. The light-dark cycle was 14 h light: 10 h dark. The exposure solution is renewed daily, and dead embryos are removed. The exposure regime was strictly based on the methods in the laboratory⁶⁷. During the experimental phase, a number of larvae were randomly selected from different groups for the experiment, once more employing random sampling.

Various stages of zebrafish embryonic development

After different erythritol concentration treatments, we observed the morphological traits at 3, 6, 10, and 24 hpf with a stereomicroscope (Olympus SZX10, Tokyo, Japan) and photographed with a digital camera (Canon, Tokyo, Japan).

Deep learning approach

Briefly, we utilized the model architecture and corresponding pretrained weights provided by Toulany et al.³⁷ to evaluate the developmental similarity between two embryonic images. Specifically, the model takes two images (which have undergone standardized preprocessing) as input and outputs a similarity score between them, ranging from 0 to 1. During our study, we used the standard temporal sequence images provided by Toulany et al.³⁷ to predict the developmental time of our experimental images. Also, we conducted pairwise comparisons of the similarity among our own experimental images, with the aim of conducting automatic detection of developmental epoch.

Calculation of embryo similarity

Calculation of the similarity between two embryos begins by employing the ResNet50 model, a deep convolutional neural network known for its efficacy in image recognition tasks, to generate image embeddings. These embeddings are high-dimensional vector representations extracted from the trained network, capturing the essential visual features of each image.

Once the embeddings are obtained, the next step involves calculating the cosine similarity between the vectors representing each pair of images. Cosine similarity is a measure used in vector space that evaluates the cosine of the angle between two vectors. Mathematically, it is defined by the formula³⁷:

$$\phi =\frac{a\,\cdot \,b}{\left|\left|a\right|\right|\,\cdot \,\left|\left|b\right|\right|}=\frac{{\sum }_{i=1}^{n}{a}_{i}{b}_{i}}{\sqrt{{\sum }_{i=1}^{n}{a}_{i}^{2}}\sqrt{{\sum }_{i=1}^{n}{b}_{i}^{2}}}$$

(1)

Equation (2) is explained in detail as follows:

• (1) $a$ and $b$: These represent the vector embeddings of two different images. Specifically, both $a$ and $b$ are vectors of length $n$, where $n$ is the number of features or dimensions in the embedding space.

• (2) ${a\cdot b}$ (Dot Product): This is the dot product between the two vectors $a$ and $b$, which computes the sum of the element-wise products of the corresponding components of the vectors. Mathematically, it is expressed as:

  $$a\cdot b=\mathop{\sum }\limits_{i=1}^{n}{a}_{i}{b}_{i}$$

  (2)

  where ${a}_{i}$ and ${b}_{i}$ are the individual elements of vectors $a$ and $b$, respectively.

• (3) $\parallel a\parallel$ and $\parallel b\parallel$ (Magnitudes of $a$ and $b$): These represent the magnitudes (or norms) of vectors $a$ and $b$, respectively. The magnitude of a vector is a measure of its length, and it is computed as the square root of the sum of the squares of its components:

  $$\parallel a\parallel =\sqrt{\mathop{\sum }\limits_{i=1}^{n}{a}_{i}^{2}},\,\parallel b\parallel =\sqrt{\mathop{\sum }\limits_{i=1}^{n}{b}_{i}^{2}}$$

  (3)

  This normalization ensures that the similarity measure is independent of the magnitudes of the vectors, focusing instead on the direction of the vectors.

• (4) Cosine Similarity: The cosine similarity is the result of dividing the dot product of the two vectors by the product of their magnitudes:

$$\frac{a\cdot b}{\parallel a\parallel \cdot \parallel b\parallel }.$$

(4)

The value of cosine similarity ranges from −1 to 1:

• (4.1) A value of 1 indicates that the two vectors point in the same direction, meaning the two images are very similar.

• (4.2) A value of 0 means the vectors are orthogonal, suggesting no similarity between the images.

• (4.3) A value of −1 indicates that the vectors point in opposite directions, meaning the images are completely dissimilar.

Morphological observation of zebrafish larvae

Control and experimental larvae were sedated with 0.168 g/L tricaine (MS-222) at 96 hpf as well and stereomicroscope photos were taken. Each group contained about 30 larvae. ImageJ (National Institute of Mental Health, Bethesda, MD, USA) software was used to quantify the head area, body length, eye area, and swim bladder area.

Determination of locomotor behavior in larvae

At 5 days post-fertilization (dpf), zebrafish larvae from the control and each treatment group were transferred to 24-well plates with one larva per well^67,74. The E3 solution was added to the level of the liquid in order to ensure that the detection camera captured a clear image. The 24-well plates were placed within a behavioral detection box, equipped with a camera positioned at the top (MV-VS078FM, Micro Vision, Japan). The larvae were first allowed to acclimate for a period of five minutes. Subsequently, the larvae’s locomotor behavior was observed under illumination (80 lx) for a period of five minutes. Thereafter, the light was extinguished (0 lx), and their locomotor behavior was recorded in the absence of light for a further five minutes. Behavioral tests were conducted between the hours of 9 a.m. and 4 p.m. Hatchling movements were recorded at a frame rate of 15 frames per second. An infrared light-emitting diode array was employed as a backlight. Each group contained 24 larvae. The inactive larvae statistics were removed to reduce errors⁷⁵. This experiment was repeated three times.

Sleep/wake behaviors

The experimental design was altered from an earlier investigation^76,77. In brief, the sleep/wake behaviors of zebrafish larvae were monitored for 48 h from 108 to 156 hpf. The dark cycle in the system commenced at 9:00 pm and concluded at 7:00 am the following day, while the light cycle commenced at 7:00 am and concluded at 9:00 pm. The temperature was maintained at 28.5 °C throughout the experiment. Prior to the commencement of the experiment, the zebrafish larvae were transferred to 96-well plates. Each group contained 16 larvae. The well plates were placed in an acrylic box, coated with a clear film, and ventilation holes were left around the borders to limit evaporation in order to create distinct images. The liquid levels were adjusted until flat. A video camera (MV-VS078FM, Micro Vision, Japan) was used to monitor sleep/wake behaviors. The real-time data was derived via an algorithm that removed the previous neighboring frames using Microsoft Visual Studio 2010 (Microsoft Corporation, Redmond, WA, USA) and OpenCV 2.4.3 (Intel Corporation, Santa Clara, CA, USA).

O-Dianisidine staining

The zebrafish possesses a fundamental circulatory system that extends from the heart to the caudal venous and dorsal aorta, and subsequently returns to the heart⁷⁸. This makes zebrafish an ideal model organism for the study of thrombus formation⁷⁸. One of the hallmarks of thrombus formation in the body of a zebrafish is a decrease in the intensity of red blood cells (RBCs) in the heart⁷⁸. The concentration of o-dianisidine solution utilized in this experiment was 0.6 g/L. The preparation of the solution was conducted in accordance with the established methodology⁷⁹. A total of 0.03 g of o-dianisidine and 0.041 g of sodium acetate were weighed into a 50 mL volumetric flask. Subsequently, 1.1 mL of 35% hydrogen peroxide was added to the flask, and the solution was diluted with 40% ethanol and shaken well. At 72 hpf, the larvae from each group were stained with o-dianisidine to detect the level of RBCs in the heart. There were 10 larvae in each group. In this experiment, ImageJ was employed for the purpose of quantifying the fluorescence intensity of cardiac erythrocytes. Subsequently, the data were subjected to statistical analysis using Prism 8.0 (GraphPad, USA)^80,81.

Oxidative stress analysis

A fluorescent indicator of H₂O₂ or other ROS, DCFH-DA, was employed as an oxidative stress monitor⁸². At 72 hpf, the larvae were treated for one hour at 28.5 °C in the dark with 10 μM diluted DCFH-DA⁵⁸. There were 10 larvae in each group. Subsequently, the specimens were imaged using a Leica DM3000 stereo fluorescence microscope in Germany. The ImageJ software (NIH, USA) was employed to quantify the intensity of the ROS staining.

qRT-PCR

A total of 30 larvae were collected at 72 hpf and 120 hpf from the control group and the treated groups. The 72 hpf larvae were employed for the analysis of the expression levels of genes associated with thrombus, inflammatory and oxidative stress, while the 120 hpf larvae were utilized for the analysis of the mRNA levels of genes correlated with locomotion behavior. Furthermore, a previous method^58,67 and a time-based analysis of sleep/wake behaviors during the day were employed, with 30 larvae from different treatment groups. Samples were collected at specific time points (ZT17, ZT23, ZT5, and ZT11) at six-hour intervals throughout the day⁸³ for the purpose of detecting biological clock genes. Following the manufacturer’s instructions, Prime Script RT regent kit (Takara, China) was used to synthesize cDNA by reverse transcription after total RNA was isolated from zebrafish larvae using Trizol (LEAGENE, China). FastStart Universal SYBR Green Master was used to do quantitative real-time PCR (qRT-PCR) (Roche, Germany). The following were the reaction conditions: 40 cycles of 50 °C for two minutes, 95 °C for ten minutes, 95 °C for fifteen seconds, and 60 °C for one minute were performed. β-Actin was used as a reference gene. The expression of these genes in zebrafish larvae was monitored using the 2^–ΔΔCt method⁸⁴. Primer sequences are in Supplementary Materials (Table S1).

Data analysis

Each experiment was run through three times in duplicate. The Shapiro–Wilk test was used to check for data normality for more than two groups. In case the data sets did not meet the normality assumption, a nonparametric Kruskal–Wallis’s test was used, followed by Dunn’s multiple comparison test. When normally distributed data met the homoscedasticity assumption (tested by Brown–Forsythe), statistical significance between experimental and control groups was determined using one-way ANOVA with Tukey’s post hoc test. If the homoscedasticity assumption was violated, Welch’s ANOVA followed by Dunnett’s T3 test was applied instead. In addition, three-dimensional trajectory plots were generated using RapidMiner Studio v7.0.1. The following variables were analyzed using the ImageJ software (NIH, USA): body length, eye area, head area, swim bladder area, cardiac erythrocyte intensity quantification, and ROS staining fluorescence intensity analysis. Statistical graphs were generated using Prism 8.0 (GraphPad, USA).