As the major organ of drug metabolism, liver is more likely to suffer from drug injury than other organs. Drug-induced liver injury (DILI) always leads to severe clinical adverse events, including hepatitis, liver fibrosis, liver failure, and even death [1
]. Currently, more than 1100 medicines have been reported to cause various degrees of liver toxicity [4
]. Data from the United States indicated that DILI has become the leading cause of acute liver injury events in clinical [5
]. In addition, among 14 toxicity factors of drug withdrawal from markets, DILI ranking first by a percentage of 27% [6
]. Therefore, it is of significance to study DILI.
The use of herbs in treating diseases has a history of thousands of years in many Asian countries. During the past decades, herbs or their extracts have gained increasing attention worldwide due to their significant efficacy in treating and preventing some diseases [7
]. The most typical example is artemisinin, which was first discovered by the Chinese scientist Prof. You-you Tu and her team from the plant Artemisia annua
(a herb which was commonly prescribed for hemorrhoids and malaria in Chinese). For its significant efficacy in the prevention and treatment of malaria, artemisinin was widely recognized by the world. Due to this important discovery, Prof. You-you Tu was awarded the 2015 Nobel Prize in Physiology or Medicine [12
]. In recent years, artemisinin was found to also treat schizophrenic disorders for which the Phase III clinical trial has been completed [13
]. Currently, it was reported that 4 billion people, about 80% of the world’s population consumed herbs to treat various diseases [14
]. In contrast to synthetic drugs, herbs were commonly thought to be safe and harmless, as they come from nature. It was roughly estimated that more than 500 herbal products were consumed as harmless health products worldwide [15
]. However, like synthetic drugs, herbs also possess adverse effects. In recent years, more and more clinical cases and laboratory data have demonstrated that some herbal products may cause varying degrees of liver damage. In a recent report, the etiology of DILI in Mainland China was investigated based on 25927 DILI cases from 308 medical centers. As a result, herbal and dietary supplements ranking first among 11 single classes of implicated drugs (26.81%) of DILI [16
]. In another research from Korea in 2019, twenty-five percent of the DILI cases were attributed to herbals [17
]. Herb-induced liver injury (HILI) is not restricted to China, Japan, and Korea but commonly observed in all countries that consume herbs. Cases mentioned above showed that HILI is a serious public health problem and deserves more attention.
Several factors as follows were documented to be associated with HILI. (1) The gender, age, organism state, and genetic background of patients [18
], (2) the dose and course of treatment of herbs [22
], (3) the metabolism, misuse, and abuse of herbs [24
], and (4) the quality of herbs, including adulterated products, bacterial contamination, the presence of heavy metals, pesticides, or solvents [26
]. Additionally, drug interaction may be another risk factor of HILI [30
]. Depending on the type of target cells suffered from damage, DILI can be generally divided into three types: cholestatic liver injury, hepatocellular liver injury, and mixed liver injury (hepatocellular and cholestatic) [32
]. According to the detail pathological features, some academics even categorized DILI into 21 types [33
]. In the light of the pathogenesis of liver injury, hepatotoxic agents can be classified into two groups: direct and indirect hepatotoxic agents. Direct hepatotoxic agents are substances that cause cholestasis by intervening the bile secretion process. In contrast to direct hepatotoxic agents, indirect hepatotoxic agents induced liver injury via producing selective metabolic blocks or physiologic lesions. These lesions always lead to the disruption of some essential metabolic pathways and cellular biochemical processes [34
]. Numerous influential risk factors and extensive disease spectrum increased the complexity and diversity of the mechanisms of DILI. Currently, hepatotoxic mechanisms for most herbs are far from elucidation. Herb is not a single chemical entity but a complex mixture of both beneficial and toxic ingredients. Therefore, identifying toxic ingredients may be the first step to understanding the exact mechanisms of HILI. During the past decades, hundreds of herbs/herbal extracts were reported to cause varying degrees of liver damage. However, until now, a systemic review focused on these hepatotoxic substances was still unavailable, which prevented academics and pharmacologists from understanding the occurrence of HILI comprehensively. Therefore, in this work, we summarized and analyzed the hepatotoxic herbs/ingredients published in the literature by which we attempted to find some clues about the occurrence of HILI from the perspectives of the phylogenetic relationship and structure-toxicity relationship.
As the natural source of drugs and dietary supplements, herbs play an essential role in drug discovery and development. During the past decades, hundreds of novel drugs and health care products derived from herbs directly or indirectly were approved by China Food and Drug Administration (CFDA) or Food and Drug Administration (FDA) [105
]. An investigation into the distribution of drug-productive species against the phylogenetic tree indicated that most novel drugs are derived from preexisting drug-productive families [108
]. In another research, the authors reported that species with close taxonomic relationship always tend to possess similar medicinal properties [109
]. These instances laid foundation for assessing the hepatotoxic risk of medicinal plants via phylogenetic tree analysis. Theoretically, medicinal plants with potential hepatotoxicity should be concentrated in some families rather than distantly distributed in phylogenetic tree. In this work, the distribution of 335 hepatotoxic medicinal plants was investigated against the phylogenetic tree of the mesangiospermae clade. As a result, a significant cluster pattern was observed in order, family, and genus scale, respectively. Medicinal plants with potential hepatotoxicity tended to have closer taxonomic relationship rather than distributed in phylogenetic tree irregularly. Therefore, we can claim that there was certain association relationship between the hepatotoxicity of medicinal plants and their taxonomic positions. Medicinal plants close to hepatotoxic medicinal plants may be faced with higher hepatotoxic risk. Summarily, the hepatotoxic medicinal plants list provided in this work will help to assess the hepatotoxic risk of medicinal plants through phylogenetic tree analysis.
In some Asian countries, many people always believed that herbal medicine is safe and harmless. However, several commonly used herbs have been reported to cause a series of adverse effects in recent years [112
]. Among the toxic effects of herbs, hepatotoxicity is one of the leading concerns. Hepatotoxicity always leaded to a series of clinical adverse events, including hepatitis, hepatic fibrosis, liver failure, and even death [1
]. Currently, DILI has been deemed as one of the major factors of the failure of drug discovery [5
]. Therefore, risk assessment focused on hepatotoxicity in the early stage of drug discovery is indispensable. However, it is always time-consuming and highly cost to detect the hepatotoxicity of drug candidates by experiment. With the steady accumulation of data related to DILI and the rapid development of computational chemistry, predicting the hepatotoxic risk of drug candidates by in silico methods has gained increasing attention in recent years. Compared to the experimental methods, in silico methods are much faster and cheaper. Currently, there have been many in silico models focused on predicting the hepatotoxic risk of drug candidates in the literature [117
]. However, almost all of these models were developed solely based on the synthetic drugs. As we all know, the chemical space of natural products is quite different from that of the synthetic drugs. The distribution of the synthetic drugs in the chemical space is relatively compact, while the distribution of natural products is more dispersed [119
]. In addition, the elementary composition of natural products is also different from that of the synthetic drugs. The synthetic drugs are rich in N atoms. However, natural products tend to contain more O atoms [121
]. Therefore, in silico models developed solely based on the synthetic drugs may have limited predictive capabilities for HILI. By incorporating the use of three machine learning algorithms and twelve types of molecular fingerprints, Ai et al. developed an ensemble in silico model for predicting DILI based on 1241 diverse compounds. For ease of access, the authors also provided a user-friendly web server [122
]. To the best of our knowledge, this is the latest publicly accessible prediction model for DILI. Based on this in silico model, we evaluated the hepatotoxic risk of each compound collected in this work (296 hepatotoxic ingredients and 584 non-hepatotoxic ingredients). As a result, the overall concordance between the predicted results and the results reported in the literature was only 45.23%. It indicated that in silico models developed solely based on the synthetic drugs is not applicable for natural products. In a recent study, it was reported that adding natural products data into the modeling dataset can improve the performance of in silico model when predicting the hepatotoxicity of natural products [123
]. Currently, an in silico model for predicting HILI was still unavailable. The lack of a large scale dataset related to HILI is one of the leading causes. Herein, we provided a large scale dataset of HILI which will lay foundation for the construction of in silico models for predicting HILI.
The distribution of hepatotoxic and non-hepatotoxic ingredients in the chemical space was investigated. Most hepatotoxic ingredients gathered together. Significant difference of the physicochemical properties between hepatotoxic and non-hepatotoxic ingredients was also detected. These results indicated that it is feasible to assess the hepatotoxic risk of natural products based on the QSAR methods.
For each compound collected in this work, we annotated its structural category and sub-category. The structures of hepatotoxic ingredients were diverse and complex, which involved in a series of sub-categories. Alkaloids and terpenoids were found to be the two major structural categories with frequency of 127 (42.91%) and 88 (29.73%), respectively. Taken several major sub-categories as cases, we summarized and analyzed their hepatotoxicity from the perspectives of structure, pathological characteristic, and mechanism, respectively. There were some chemical bonds/groups significantly associated with the hepatotoxicity of phytochemicals, including the 1–2 unsaturation in PAs, the MDP structure of safrole, the iminiun bond within sanguinarine, and the α-isopropylidene ketone unit within pulegone analogues. The pathological characteristics of HILI were complex and diverse, including morphology changes and apoptosis of hepatocytes, hepatic congestion, inflammation, fat vacuole, and centrilobular necrosis of hepatocytes. Actually, HILI cannot be viewed as a single disease entity, but a spectrum of liver diseases. Generally, it can be roughly categorized into two classes, “intrinsic” and “idiosyncratic”. Intrinsic HILI is predictable, reproducible, and dose-dependent. Idiosyncratic HILI is not clearly dose related but host dependent. Therefore, it is often difficult to detect it by the regulatory animal toxicity studies. Many different mechanisms were reported to be associated with intrinsic HILI. Some drugs were activated by the drug-metabolizing enzymes and metabolized to minor electrophilic metabolites. These highly reactive metabolites depleted glutathione and initiated covalent binding to cellular proteins/DNAs, resulting in lipid peroxidation, disrupting the mitochondrial functions, activating oxidative stress and endoplasmic reticulum stress, and eventually triggering cell death pathways. Drugs decreasing the activity of drug-metabolizing enzymes always resulted in the accumulation of toxic metabolites and leaded to liver injury. The disruption of bile acid homeostasis is another mechanism of HILI. By regulating the expression or localization of bile acids transporters, some herbs or their reactive metabolites inhibited the clearance of bile acids and resulted in hepatocyte damage and cholestasis. In addition, the activation of Fas-, P53-, and mitochondrial-apoptosis pathways also play an essential role in the development of HILI. Idiosyncratic HILI is a rare disease which mainly depends on the host genetic, immunologic, and metabolic factors. Many drug-metabolizing enzymes are highly polymorphic and there are interindividual variability in their metabolic activity. Relatively low metabolic activity always resulted in the accumulation of drugs or their metabolites in the body and leaded to liver toxicity. In addition, some drugs or their reactive metabolites generated autoantibodies by binding to the endogenous proteins. The formation of autoantibody led to hepatocyte death. In summary, the mechanism of HILI is rather complicated. Currently, the hepatotoxic mechanisms for most herbs are far from elucidation. Scientific studies focused on uncovering the molecular basis and mechanism of HILI deserve more attention.
Based on SARpy software, we identified 15 SAs for hepatotoxicity. Two of these structural alerts, ID 1 and ID 12, have been proved to be significantly associated with the occurrence of hepatotoxicity. These SAs cannot only assist to decipher the hepatotoxic mechanisms of herbs but also can help to construct in silico models for predicting HILI.
Finally, a herb-ingredient network was constructed for the first time against the “herb-ingredient” pair data included in TCMSP, TCMID, and ETCM databases. This herb-ingredient network described the relationship between hepatotoxic herbs and hepatotoxic ingredients. It can be used to identify the hepatotoxic molecular basis of herbal/herbal formula quickly and comprehensively.
Although some herbs or their ingredients were reported to possess potential hepatotoxicity, one cannot deny their significant efficacy in treating many complex diseases [124
]. In many cases, their toxicity can be alleviated or eliminated through modifying structure, changing the dosage form, or combining with other herbs/ingredients [127
]. Nevertheless, herbs or their ingredients with hepatotoxicity should be carefully administered in clinical application. Generally, damage to the liver will be brought under control after stopping the herbal medication. Avoiding high doses and long term use of herbals may be an effective strategy to reduce the incidence of HILI [53
]. In the future, a comprehensive and systematic database for HILI may help physicians and pharmacologists to reduce HILI.
Nevertheless, we must acknowledge limitations of this work as follows: (1) although a comprehensive literature retrieval was conducted. There may still exist some hepatotoxic ingredients and hepatotoxic medicinal plants escaped from our vision. (2) The completeness of the results provided by our herb-ingredient network is highly depended on the “herb-ingredient” pair data included in the databases. In this work, the herb-ingredient network was constructed against three typical herb databases. With the introduction of more comprehensive “herb-ingredient” pair data, the results provided by our network will be more completeness. (3) The majority of the data used in the current study were not derived from clinical case reports or observational studies but from animal/cell experiments. In the future, with the accumulation of clinical data of HILI, we believe studies focused on HILI will make great progress.