What have we learned from animal models of idiosyncratic, drug-induced Liver Injury?

Abstract

Introduction

Idiosyncratic, drug-induced liver injury (IDILI) continues to plague patients and restrict the use of drugs that are pharmacologically effective. Mechanisms of IDILI are incompletely understood, and a better understanding would reduce speculation and could help to identify safer drug candidates preclinically. Animal models have the potential to enhance knowledge of mechanisms of IDILI.

Areas covered

Numerous hypotheses have emerged to explain IDILI pathogenesis, many of which center on the roles of the innate and/or adaptive immune systems. Animal models based on these hypotheses are reviewed in the context of their contributions to understanding of IDILI and their limitations.

Expert opinion

Animal models of IDILI based on an activated adaptive immune system have to date failed to reproduce major liver injury that is of most concern clinically. The only models that have so far resulted in pronounced liver injury are based on the multiple determinant hypothesis or the inflammatory stress hypothesis. The liver pathogenesis in IDILI animal models involves various leukocytes and immune mediators such as cytokines. Insights from animal models are changing the way we view IDILI pathogenesis and are leading to better approaches to preclinical prediction of IDILI potential of new drug candidates.

1. Introduction

1.1. Intrinsic vs. idiosyncratic reactions

Paracelsus was a clinician-scientist who lived in the 1500s and is considered by toxicologists to be the “Father of Toxicology.” He is credited with saying that all things are toxic, and that it is only the dose that separates a toxic agent from a nontoxic one. With respect to the categories of toxicity, i.e., “intrinsic” and “idiosyncratic,” this statement clearly refers to intrinsic toxicity. Intrinsically hepatotoxic responses are demonstrably dose-dependent and occur in all individuals at toxic doses [1]. For acute hepatotoxicity, the onset of injury occurs after a usually short latent period, and the pathogenesis follows a predictable timecourse. Chemicals that are intrinsically hepatotoxic display characteristic lesions, and, importantly, the toxicity is reproducible in experimental animals. For drug candidates, one important goal is to identify a dose that causes a therapeutic effect and that is much smaller than the dose at which toxicity begins to be seen (ie, a wide therapeutic window). An example of an intrinsically hepatotoxic drug is acetaminophen. At therapeutic doses, acetaminophen is both safe and effective; however, as dose is increased, dose-related hepatotoxicity occurs both in humans and in animals. Liver injury begins after 24 – 48 hours in humans and 2 to 4 hours in mice, and in both species it is associated with centrilobular hepatocellular necrosis.

Even though intrinsically hepatotoxic responses are dose-dependent, individuals vary in sensitivity to toxic effects. There are numerous determinants of sensitivity, including life stage (age), sex, metabolic activity, immunologic reactions, tissue reserve capacity, absorption/distribution differences, coexisting disease, coexisting inflammation, coexposure to other xenobiotic agents, and nutritional status. Most of these determinants can be influenced by genetic and/or environmental factors. Because many sensitivity determinants can change within an individual over time, so too can the threshold for toxicity change. For example, an acute inflammatory episode can increase the sensitivity of mice to the hepatotoxic effects of acetaminophen, as evidenced by a marked leftward shift in the dose response relationship for liver injury [2]. Accordingly, factors within the host can influence intrinsically hepatotoxic responses, and these influences can blur the lines between intrinsic and idiosyncratic hepatotoxicity [3,4].

Idiosyncratic, drug-induced liver injury (IDILI) can be defined as an hepatotoxic response that occurs in a minority of patients during drug therapy [1]. Unlike intrinsically toxic reactions, which usually occur clinically in overdose situations, IDILI responses occur at therapeutic doses and typically in a small, susceptible fraction of people. Thus, IDILI is highly influenced by host sensitivity. The relationship of exposure to the onset of toxicity is variable within and among drugs, and liver pathology can vary even for a single drug. Importantly, idiosyncratic reactions are not reproducible in typical preclinical animal tests done during drug development, and this presents a long-standing challenge in identifying drugs with idiosyncrasy liability prior to widespread exposure to humans.

1.2. Impact of idiosyncratic hepatotoxicity

Understanding and dealing with IDILI is important for at least three reasons. First is the impact on human health. IDILI reactions represent a significant fraction (approx. 10%) of acute liver failure cases [5]. Recovery from liver insult often occurs upon cessation of drug exposure; however, IDILI responses can be severe, resulting in death or a need for liver transplant. It is these potentially life-threatening reactions that are of clinical significance and most concerning from a public health perspective. Secondly, IDILI reactions can force restricted use or market withdrawal of drugs that are pharmacologically effective. This can result in fewer and potentially less effective pharmaceutical options for treating disease. Thirdly, these reactions present an economic issue for pharmaceutical companies, insofar as withdrawal from market means loss of financial investment in the development of a drug and also in costly lawsuits from affected patients or their families.

Clinical trials represent the most costly stage of drug development. Accordingly, approaches to identifying, early in the drug development process, those drug candidates that have IDILI liability could lead to better decision-making as to which candidates to move forward in development. Thus, there is a need for animal and in vitro models that will enhance understanding of IDILI modes of action and that can identify culprits preclinically.

1.3. What do we know about the etiology of IDILI?

It is obvious that these reactions depend upon (1) the properties of the drug and (2) the susceptibility of the patient. Beyond this, much remains unknown, and consequently there continues to be much speculation about the underpinnings of IDILI. The uncertainties have resulted in numerous theories about IDILI modes of action. These surround reactive metabolites, drug transporters, mitochondrial toxicity, failure to adapt to modest injury, immune system activation, multiple intersecting determinants, etc. Although both basic scientists and clinician-scientists tend to cling to one hypothesis or another, it should be kept in mind that proof of mechanistic underpinnings of IDILI reactions for most drugs remains minimal or lacking. It is also important to note that many current theories are not mutually exclusive but rather tend to overlap.

2. Scope of the work

As in many fields of biomedical science, understanding of modes of action and mechanisms depends on animal models and on in vitro, cell–based studies. In the remainder of this commentary, animal models that have been developed consistent with selected mode-of-action hypotheses will be described and critically reviewed. Inasmuch as hepatocellular death leading to liver dysfunction is of paramount clinical concern in humans, only animal models in which overt liver injury occurs from drugs associated with human IDILI will be discussed. All of the models to be discussed involve the immune system in some fashion. Although some factors such as drug transporters and metabolic bioactivation will not be emphasized in this discussion, they are important contributors to some of the models.

3. The multiple determinant hypothesis of IDILI

As noted above, numerous factors can influence the sensitivity of individuals to the intrinsically toxic effects of drugs. So too might these factors determine the propensity of an individual to experience an IDILI reaction. The multiple determinant hypothesis suggests that the intersection of various risk factors within an individual results in a toxic reaction upon drug exposure [6,7] (Fig. 1). Each factor would be associated with a probability (0–1) of occurring within a given individual. If several risk factors were required to precipitate a toxic reaction, then the probability of occurrence would be the product of the probabilities of the risk factors. This could be a very small number and explain why IDILI reactions are rare.

Fig. 1. The multiple determinant hypothesis of IDILI.

Several risk factors must be present to precipitate hepatotoxicity during drug exposure.

An animal model of halothane hepatitis consistent with this hypothesis has been reported [8,9]. Halothane is a volatile anesthetic, the hepatotoxicity of which has been extensively documented in humans. Halothane exposure causes modest increases in serum markers of liver injury in a substantial fraction of patients (approx. 20%), but it also causes severe liver injury, termed “halothane hepatitis,” in a much smaller fraction of people who are anesthetized with this drug [10,11]. Although halothane is less widely used today, toxicity from it remains a clinical concern in some countries [12].

Clinical studies have revealed risk factors for halothane hepatitis, and these include female sex, genetic predisposition and age, i.e., post-adolescent individuals are more susceptible [13,14]. Dugan et al. [8] reported a halothane hepatitis model in mice that employed these three risk factors. Fed, female, BalbC mice eight weeks of age responded to a single, IP injection of halothane with pronounced centrilobular necrosis and plasma ALT activity of approximately 10,000 U/L (Fig. 2A and B). Younger BalbC mice (four weeks old) and C57Bl6 mice were completely resistant to the hepatotoxicity, indicating that age and genetics were two critical susceptibility factors (Fig. 2C and D). In addition, male mice were far less sensitive (Fig. 2D), and this sensitivity difference was not due to a difference in trifluoroacetylated proteins that result from halothane bioactivation. Since all patients are routinely fasted prior to being anesthetized, overnight fasting was included as an additional, potential risk factor, and liver injury was greater in fasted mice (Fig. 2E).

Fig. 2. Multiple determinant hypothesis of IDILI: Halothane hepatitis animal model.

Female sex, genetic predisposition and age are known risk factors for human halothane hepatitis. A. Female BalbC mice given halothane i.p. respond with greater hepatotoxicity than male counterparts. B. Centrilobular lesion in halothane-treated female mouse. C. Pronounced difference in toxicity in two mouse strains given the same dose of halothane. D. Female mice 8 weeks of age are far more susceptible to halothane hepatotoxicity than younger mice. E. Fasting enhances hepatotoxicity of halothane in female, BalbC mice. From [8].

Interestingly, the severity of halothane–induced liver injury was dependent on stage of the estrus cycle in the female mice, and ovariectomy rendered mice much less sensitive [15]. Progesterone appears to contribute to the female sensitivity by enhancing the inflammatory response [16,17]. Thus, the female sensitivity, genetic predisposition and insensitivity of the young, i.e., risk factors well known in human halothane hepatitis, were recapitulated in the mouse model. In addition, steatosis and centrilobular necrosis that typically characterize liver histopathology in human halothane hepatitis also defined the mouse lesions. Also, isoflurane, an anesthetic known to be less associated with liver injury in humans, was not toxic in the mouse model. Accordingly, these results in mice in many ways resemble halothane hepatitis that occurs in human patients inasmuch as the liver lesions in both are characterized by centrilobular necrosis, inflammatory infiltrates and steatosis. Furthermore, in both people and mice, females are more sensitive, the young are less sensitive and genetics are a sensitivity factor.

Additional investigation revealed that mRNA expression of several inflammatory cytokines was elevated in liver tissue of susceptible, halothane-treated mice [9]. Tumor necrosis factor-alpha (TNF) and interferon-gamma (IFNγ) were elevated in the plasma, and liver injury was markedly reduced in IFNγ-knockout mice [15]. Moreover, halothane exposure elevated HMGB-1 in the plasma, and animals with a dysfunction mutation in toll-like receptor 4 (TLR4) had less severe liver injury and less IFNγ in plasma in response to halothane; these results suggested a contribution of innate immune system activation to the hepatotoxicity [15]. Interestingly, depletion of Kupffer cells with clodronate-encapsulated liposomes affected neither liver injury nor the IFNγ elevation [18], but natural killer (NK) cells, which produce and are activated by IFNγ, were activated by exposure of mice to halothane, and reduction of NK activity with an anti-AsGM1 antibody decreased both liver injury and appearance of IFNγ [15]. Activated NK cells can release perforin and granzyme B, cytotoxic proteins that form pores in the membranes of target cells and mediate cell death; mice with defective perforin (BALB^Prf1) released less IFNγ and were protected from liver injury [15]. Natural killer T cells played a role as well in the liver injury, at least in part by recruiting neutrophils, which contributed to the hepatocellular damage [8,15,18]. IL-17 released by Th17 cells appear to be involved in the infiltration of neutrophils into the liver [19].

Based on these results, Dugan et al. [15] proposed that halothane bioactivated to trifluoroacetic acid causes a stress response in hepatocytes that leads to mobilization of receptors (eg, Rae-1) that activate NK cells, which release cytotoxic perforin/granzyme and IFNγ, causing initial damage to hepatocytes. Acting as a danger signal, HMGB1 released from these compromised cells activates TLR4 on NK cells, thereby initiating a positive feedback loop resulting in enhanced activation of NK cells and other leukocytes, such as neutrophils (PMNs) and eosinophils. Eosinophil accumulation in liver has been associated with IDILI from numerous drugs [20]. Expression of CCL11 and CCL24, chemotaxins for eosinophils, was elevated in livers of halothane-treated mice, and reduction of hepatic eosinophil accumulation (with anti-Siglec-F antibody) reduced liver injury [21]. A subsequent study implicated hepatic epithelial-derived cytokine thymic stromal lymphopoietin (TSLP) and interleukin-4 (IL-4) as contributors to the accumulation of eosinophils in the liver [22]. As noted above, neutrophils accumulate in livers of halothane-treated mice, and prior depletion of neutrophils with an anti-CD18 antibody reduced liver injury [8], but this antibody might have also reduced eosinophil accumulation [21]. Similarly, hepatic accumulation of both neutrophils and eosinophils was reduced in ΔdblGata−/− mice, and liver injury from halothane exposure was less severe in these mice. Together, these studies implicate leukocytes such as NK cells and eosinophils and mediators that they produce (and to which they respond) in the pathogenesis of liver injury in this halothane hepatitis model. It is noteworthy that activation of the damaging innate immune response in this model does not involve sensitization followed by subsequent challenge with halothane—ie, a classical adaptive immune response—but does entail cells and soluble mediators that are common to innate and adaptive immune responses (see below).

4. The failure-to-adapt hypothesis of IDILI

This hypothesis arose from a study of responses to acetaminophen in healthy human volunteers [23]. Acetaminophen was given daily at the maximum recommended dose for 2 weeks. A majority of the subjects experienced a modest increase in serum ALT which resolved over time even with continued drug administration. This result suggested that liver stress/modest hepatotoxicity is common in people taking therapeutic doses of acetaminophen and that people typically adapt to it during maintenance therapy. The failure-to-adapt hypothesis suggests that modest liver stress is common from exposure to pharmacological doses of drugs and that patients who fail to adapt to it progress to overt liver injury, which appears as an idiosyncratic reaction. Indeed, such modest liver injury and adaptation appear to occur with other drugs, some of which have been associated with idiosyncratic hepatotoxicity. For example, isoniazid causes modest and clinically insignificant increases in liver enzymes in serum in about 10–20% of patients; however, in a much smaller fraction of isoniazid treated patients, progression to severe liver injury occurs [24]. Similar responses occur with halothane, amodiaquine and other, IDILI-associated drugs [10,11,25,26].

Mechanisms underlying failure to adapt are not well understood. It has been proposed that immune tolerance is responsible for the adaptation and prevents progression to pronounced liver injury [27], but so far no robust animal models have been forthcoming that support this idea. Amodiaquine is a drug that has caused severe liver injury in human patients. In female, C57Bl/6 mice, daily amodiaquine administration resulted in modest liver injury from which the mice recovered despite continued drug treatment [28]. In some studies, immune tolerance was reduced by using mice deficient in programmed cell death 1 protein (PD‐1) given an antibody to cytotoxic T‐lymphocyte‐associated protein 4 (CTLA4). In male PD-1(−/−) mice given anti-CTLA4 antibody, treatment with amodiaquine caused a modest increase in serum ALT activity that was also transient, but in their female counterparts the modest injury was longer lasting [29]. Although no pronounced liver injury developed in this model, the results suggest that escape from immune tolerance might reduce the adaptation to modest liver injury in female mice. However, no evidence currently exists to support a role for escape from immune tolerance in clinically significant liver damage resulting from failure to adapt to modest injury.

Another plausible explanation for failure to adapt relates to inhibition of liver repair. Following liver injury, proliferation of hepatocytes replaces injured tissue. Indeed, proliferative repair is the major way in which the liver responds to injury, and loss of the proliferative response can magnify liver injury [30]; such loss could appear as failure to adapt to damage, permitting injury progression. Proliferative stimuli are numerous and include cytokines and growth factors, interference with which can halt proliferation. Thus, an impaired ability to undergo replicative repair could arise from a number of endogenous and exogenous factors, including a contribution from the offending drug itself. Several drugs associated with IDILI have been shown in vitro to inhibit cell proliferation; these include diclofenac, sulindac, trovafloxacin, halothane, doxorubicin and chlorpromazine [31–36]. To our knowledge, no animal models based on this hypothesis and employing IDILI-associated drugs have emerged. However, Mehendale and coworkers have shown for several xenobiotic agents which cause intrinsic liver injury that hepatotoxicity is magnified in animals when proliferative repair is compromised [30].

5. The adaptive immunity hypothesis of IDILI

A long-standing hypothesis and one to which clinicians and clinician-scientists subscribe heavily is the adaptive immunity hypothesis. Indeed, some clinician-scientists believe that an activated adaptive immune system is responsible for all IDILI reactions [37]. The classical version of this theory is summarized in blue lettering in Fig. 3. The initiating event is metabolic bioactivation of the drug to a reactive metabolite that binds covalently to liver proteins. One or more of these protein conjugates acts as a hapten that is recognized by antigen-presenting cells, such as Kupffer cells. These antigen-presenting cells modify the hapten and present it to T-lymphocytes, which proliferate and become activated and, in turn, activate other leukocytes. The activated immune cells then kill hepatocytes, leading to liver dysfunction. It is now thought that additional events are required for an adaptive immune response to become injurious. One is an additional signal that might be provided by modest cell damage caused directly by the reactive drug metabolite. This results in the release from compromised cells of damage-associated molecular pattern molecules (DAMPs) such as HMGB1, RNA, DNA, etc., that promote the activation of T-cells [38].

Fig. 3. A unified hypothesis of IDILI pathogenesis.

Activation of either the adaptive immune system (blue lettering) or the innate immune system (red lettering) results ultimately in the activation of leukocytes and production of cytokines and other mediators that have the potential to effect cell killing. Cytokines such as TNF do not kill healthy cells but can activate cell death pathways in drug-compromised cells. An offending drug can have several actions that contribute ultimately to hepatocyte death. It can be metabolized to a reactive metabolite that can injure cells directly or form an adduct that acts as a hapten to activate the adaptive immune system. A drug can also enhance the production of damaging mediators (eg, TNF) by leukocytes or stress hepatocytes, rendering them susceptible to killing by cytokines. Drugs might also interfere with immune tolerance or with hepatocellular adaptation to injury (violet Xs).

The liver normally exists in a state of immune tolerance that involves myeloid derived suppressor cells as well as immune checkpoint receptors such as PD‐1 and CTLA4 [39]. Current thinking is that, not only is the additional (danger) signal required, but immune tolerance must also be overcome for the immune system to kill host cells.

Evidence certainly exists in humans for an adaptive immune underpinning for IDILI, based on human leukocyte antigen (HLA) associations in patients, presence of antidrug antibodies, positive lymphocyte transformation tests and, sometimes, presence of lymphocytes in liver biopsies from affected patients. These observations support a weight-of-the-evidence argument for an activated adaptive immune system as initiator of injury, but they are purely associative and do not conclusively prove a cause of hepatic damage. Perhaps the most compelling evidence for this hypothesis is the identification of specific polymorphisms in HLA genes that are associated with IDILI from several drugs (Table 1; for review, see[40]).

Table 1.

HLA polymorphisms associated with IDILI from specific drugs

Drug	HLA allele	Reference
Amoxicillin-clavulanate	DRB*15:01-DQB1*06:02	[105]
Fenofibrate	HLA-A*33:01	[106]
Flucloxacillin	HLA-B*57:01; HLA-B*57:03	[107,108]
Lapatinib	HLA-DRB1*07:01 HLA-DQA1*02:01	[40]
Lumiracoxib	DRB1*15:01	[109]
Minocycline	HLA-B*35:02	[110]
Terbinafine	HLA-A*33:01	[106]
Ticlopidine	HLA-A*33:03, A*33:01	[106]
Ximelagatran	DRB1*07 and DQA1*02	[40,52]
Carbamazepine	HLA-A*31:01	[111]
Polygonum multiflorum	HLA-B*35:01	[112]

Attempts to develop animal models based on the adaptive immunity hypothesis have appeared in recent years (Table 2). Chakraborty and coworkers [41] developed a model of halothane-induced injury in which mice were depleted of myeloid-derived suppressor cells (MDSCs) with an anti-GR-1 antibody to eliminate immune tolerance. Mice were given an initial, markedly hepatotoxic dose of halothane, and then challenged 14 days later with the same dose of halothane in an attempt to elicit immune-mediated injury (Fig. 4). Twenty-four hours after both the initial exposure and the challenge, mice responded with pronounced liver injury, similar to the acute model described above (see section on Multiple Determinant Hypothesis), and there was no difference in liver injury between MDSC-depleted mice and isotype controls (Fig. 4). Additionally, at 7 days post-challenge there was no serum ALT increase in either group. However, 9 days after the halothane challenge, MDSC-depleted mice had mild liver injury (serum ALT, 500 U/L) that was not present in halothane-challenged controls. The lesions were characterized by pericentral lymphocyte accumulation and scattered apoptotic hepatocytes. At this post-challenge time, serum immunoglobulin against protein adducts from bioactivated halothane were elevated in the MDSC-depleted mice, serum IL-4 and chemotactic CCL11 concentrations were elevated, and eosinophils, CD8+ T-cells and IFNγ-producing CD4+ T-cells accumulated in liver. An anti-CD4 antibody eliminated the modest increase in serum ALT at 9 days post-challenge.

Table 2.

Adaptive Immunity Hypothesis: Animal Models

Drug with human IDILI liability	Mouse Strain	Cotreatment	Liver injury	Reference
Halothane	BalbC	Anti-GR-1	Modest	[41]
Amodiaquine	PD-1 KO	Anti-CTLA-4	Modest	[29]
Nevirapine	PD-1 KO	Anti-CTLA-4	Modest	[46]
Isoniazid	PD-1 KO	Anti-CTLA-4	Modest	[46]
Troglitazone	PD-1 KO	Anti-CTLA-4	Modest	[45]
Tolcapone	PD-1 KO	Anti-CTLA-4	Modest	[45]
Ximelagatran	DR7xhCD4 DQ2xhCD4	None None	None None	[52]
Abacavir	HLA-B*57:01 HLA-B*57:01	None CpGODN	None Modest	[53]

DR7 mice express HLA-DRB1*0701 and HLA-DRA*0102; DQ2 mice express HLA-DQB1*0202 and HLA-DQA1*0201; Abbreviations: PD-1 KO, Protein cell death protein 1 knockout; CTLA-4, cytotoxic T-lymphocyte-associated protein 4; CpGODN, CpG oligodeoxynucleotides

Fig. 4. Adaptive immunity hypothesis: Halothane animal model.

A. Experimental design: Mice were depleted of myeloid-derived suppressor cells (MDSCs) with antiGR-1, then given an hepatotoxic dose of halothane; 14 days later they were challenged with the same halothane dose, and their livers were examined 7 or 9 days later. B. Pronounced hepatotoxicity resulted from both the sensitizing and challenge administrations of halothane. Seven days postchallenge, no serum ALT increase was observed, but there was a modest ALT increase by 9 days (C). (From [41]) D. Summary of responses in this model.

To summarize, the control mice that received the halothane challenge 9 days earlier recovered from the liver injury, but the livers of MDSC-depleted mice had accumulation of eosinophils and T-cells and mild hepatocellular injury that depended on CD4+ T-cells. This modest liver injury mediated by an adaptive immune response depended on creating severe injury from both the initial and challenge doses; this severe injury required an innate, not an adaptive immune response (see above). Thus, the adaptive immune-mediated aspect of the injury in this model did not mimic the severe halothane hepatitis that is of clinical concern in human patients. Another caveat regarding this approach is that the GR-1 epitope is not entirely selective for MDSCs, so that the antibody used to deplete these cells might have affected other cell types (eg, neutrophils).

Amodiaquine is an antimalarial drug that causes mild increases in serum ALT activity in some patients and severe liver injury in others and has been withdrawn from the market partly for that reason. The clinically significant IDILI is characterized by markedly increased serum ALT activity and jaundice that persists for several months [26]. Amodiaquine is bioactivated in the liver to a reactive quinoneimine metabolite capable of binding to liver proteins and thus might act as a hapten [42]. In female, wild type mice (C57Bl/6), amodiaquine caused a very modest increase in serum ALT (40 U/L) that was transient and not associated with any histologic evidence of liver necrosis, a result indicating no significant liver injury [28]. Uetrecht and coworkers have attempted to develop a model of amodiaquine-induced IDILI using female, PD-1(−/−) mice that were cotreated with an antibody to CTLA4 to compromise immune tolerance. When these mice were fed amodiaquine in the diet, a modest increase in serum ALT activity (approx. 200 U/L) developed after 2–3 weeks that was maintained for several weeks [29](Fig. 5). This was associated with large areas of leukocyte accumulation in the liver that included macrophages and CD8+ T-cells and with modest hepatocellular necrosis near the areas of infiltrate. The CD8+ T-cells produced perforin and granzyme B in the PD‐1(−/−) mice cotreated with amodiaquine and anti‐CTLA4. Depletion of CD8+ T-cells with an anti-CD8 antibody eliminated the modest ALT increase, suggesting a critical role for these cells in the mild injury that occurred (Fig. 5).

Fig. 5. Adaptive immunity hypothesis: Amodiaquine animal model.

PD-1 knockout mice that were treated with an anti-CTLA-4 antibody to eliminate immune tolerance were fed amodiaquine in the diet for several weeks. A. A very modest but sustained increase in serum ALT occurred after 1 week. This increase was associated with few necrotic hepatocytes. B. Depletion of CD8+ T-cells eliminated the ALT increase, suggesting that the modest liver injury was adaptive-immune mediated. (From [104] https://pubs.acs.org/doi/10.1021/acs.chemrestox.5b00137; further permissions related to the material excerpted should be directed to the ACS). C. Summary of findings.

Histologically, the infiltrates in this model resembled tertiary lymphoid structures (aka, perivascular ectopic lymphoid structures) that develop in nonlymphoid organs during persistent graft rejection, infections, cancer and autoimmune conditions [43]. Such structures are not necessarily associated with tissue damage, and their role in disease is poorly understood [44]. Although it has been claimed that “decreased liver function” occurred in this amodiaquine model [27], no decrement in hepatocellular function has been reported. Indeed, given the very mild parenchymal changes histologically, none would be expected.

Similarly, very modest liver injury was observed using the same strategy with either isoniazid, nevirapine, troglitazone or tolcapone (Table 2), four other drugs that are associated with severe IDILI in humans [45,46]. Numerous immune modifiers have been explored in an attempt to enhance the liver injury, but none has thus far been successful in doing so [45,47]. As noted above in the section on Failure-to-Adapt, many drugs that cause severe IDILI of low incidence also cause, in a larger fraction of people, mild and usually reversible serum ALT increases that are of less clinical significance. Mild injury occurs in human patients with each of the drugs mentioned above for which animal models based on an adaptive immune response have been developed (i.e., halothane, isoniazid, nevirapine, troglitazone, tolcapone) [24,48–51].

Another approach to evaluating the role of the adaptive immune response in IDILI has been to develop animal models based on expression of specific HLA alleles. These models have in general failed to produce liver injury following treatment with a drug. For example, no liver injury was seen when ximelagatran was given to mice expressing HLA-DRB1*0701 and HLA-DRA*0102 or HLA-DQB1*0202 and HLA-DQA1*0201 [52]. Perhaps this is not surprising because even in humans, carrying an HLA polymorphism was insufficient to provoke injury, since the vast majority of people who were exposed to the drug and had the same polymorphism did not experience liver injury. This suggests that other factors in addition to HLA polymorphisms are critical. Interestingly, in transgenic mice expressing HLA-B*57:01, an allele associated with abacavir-induced IDILI in people, treatment with abacavir alone did not cause liver damage, but cotreatment with abacavir and a toll-like receptor agonist that stimulates the innate immune system resulted in mild increases in ALT activity in the plasma as well as necro-inflammatory lesions in the liver [53].

Thus, all of the models developed to date that depend on an activated adaptive immune system have resulted in either no injury or injury that resembles the modest, subclinical alteration in liver homeostasis that is now recognized to occur during maintenance therapy with numerous drugs and to which people typically adapt. Accordingly, based on the results from animal models to date, it appears that the adaptive immune system is either (1) responsible only for mild, subclinical cases of drug-induced liver injury or (2) is a necessary but insufficient contributor to severe IDILI from some drugs, requiring an as yet unknown factor(s) to progress from minor to major liver damage. As mentioned above, inhibition of replicative repair of liver might underlie failure to adapt to modest injury; it would be interesting to determine if inhibited cell replication could turn the modest injury seen in the amodiaquine and other adaptive immunity animal models into pronounced liver damage.

6. The inflammatory stress hypothesis of IDILI

We are all subject to bouts of acute inflammation as we go through life. The inflammatory stress hypothesis proposes that some IDILI reactions result from the intersection of drug-exposure with an acute inflammatory stress that by itself is noninjurious [54]. Modest inflammation can be caused by infection with viruses or bacteria, a compromised intestinal barrier that leaks microbial products into the portal circulation, and other factors. Inflammation is initiated by pathogen–associated molecular pattern molecules (PAMPS), such as bacterial lipopolysaccharide (LPS), that stimulate TLRs on various cells of the innate immune system, such as macrophages (eg, Kupffer cells) and other leukocytes (see red lettering in Fig. 3). This results in leukocyte activation and downstream inflammatory events that include production of various proinflammatory mediators, some of which can damage host tissue. As proposed, interaction of a drug with this innate immune response leads to death of hepatic parenchymal cells in IDILI. One can envision a patient on maintenance therapy with a drug doing well until the occurrence of an acute inflammatory stress of sufficient magnitude to interact with the drug precipitates a liver injury response.

This hypothesis can explain why IDILI reactions are typically rare and occur at variable times after the onset of drug therapy, since acute inflammatory episodes are uncommon, occur sporadically and presumably not always with sufficient magnitude to interact with a drug. As with other IDILI hypotheses, evidence supporting the inflammatory stress hypothesis is associative and incomplete. It is of interest that the classes of drugs most commonly associated with IDILI, i.e., antibiotics and nonsteroidal anti-inflammatory drugs (NSAIDs), are used to treat conditions characterized by inflammation. Indeed, rheumatoid arthritis, a disease characterized by flares of inflammation, has been identified as a risk factor for IDILI caused by NSAIDs [55].

Animal models based on the inflammatory stress hypothesis have been developed for more than a dozen drugs (Table 3). All of these models involve treating rodents with a noninjurious dose of an inflammatory PAMP (most often LPS) and a nontoxic dose of a drug that causes IDILI in people. Inflammatory stress models have been developed with drugs that cause clinical patterns of liver injury that are predominately hepatocellular (isoniazid, halothane), cholestatic (chlorpromazine, dicloxacillin, diclofenac) or both (amiodarone, sulindac, methimazole) (https://www.ncbi.nlm.nih.gov/books/NBK547852/, livertox.NIH.gov [Jan. 31, 2020]). Most of these models involve administration of LPS at a dose that causes a modest inflammatory response but no serum ALT elevation or histologic evidence of hepatocellular necrosis. For many of the models, liver injury produced by interaction of drug with LPS is pronounced.

Table 3.

Animal models based on the inflammatory stress hypothesis

Drug with human IDILI liability	Inflammatory stimulus (PAMP)	Species	Reference
Chlorpromazine	LPS	rat	[113]
Sulindac	LPS	rat	[68,71,72]
Amiodarone	LPS	rat	[57]
Ranitidine	LPS	rat	[114]
Halothane	LPS Poly I:C	mouse	[8,18]
Trovafloxacin	LPS, TNF, PGN/LTA	mouse, rat	[60,66,115]
Amodiaquine	LPS	rat	[116]
Dicloxacillin	IL-4, DK-PGD2	mouse	[117]
Diclofenac	LPS	rat	[74,75]
Doxorubicin	LPS	mouse	[86]
Isoniazid	LPS	rat	[118,119]
Propylthiouracil	LPS	mouse	[120]
Methimazole	LPS	mouse	[120]
cis-Stilbene glucoside	LPS	rat	[121]
Chloroquine	LPS	rat	[116]
Pevonedistat	TNF	rat	[122]

Abbreviations: LPS, lipopolysaccharide; Poly I:C, polyinosinic:polycytidylic acid; TNF, tumor necrosis factor-α; PGN/LTA, peptidoglycan/lipoteichoic acid; DK-PGD2, 13,14-Dihydro-15-keto-PGD2

One example is a model that was developed with amiodarone, an antiarrhythmic drug used in patients after myocardial infarction or in congestive heart failure. Amiodarone was approved in the United States in 1985 and is associated with adverse side effects in several organs, including lungs, thyroid, eyes and liver. With regard to liver, amiodarone administration is associated with relatively common but asymptomatic elevations in serum ALT activity but also with liver reactions that are of greater clinical concern (ALT activity greater than three times the upper limit of normal [ULN]) in 1% – 3% of patients. These occur both in long-term oral therapy and after acute, intravenous administration [56].

Lu et al. [57] developed a liver injury model in which Sprague-Dawley rats were given a nontoxic dose of amiodarone 16 hours before a nontoxic dose of LPS (Fig. 6A). Animals treated with LPS alone or the drug alone developed no increase in serum ALT activity, but rats treated with amiodarone and LPS developed dose-dependent liver injury that was detectable within a few hours after LPS administration [57] (Fig. 6BCD). The injury was characterized by large areas of midzonal necrosis and serum ALT values exceeding 1000 U/L. Amiodarone is metabolized predominantly to mono–N–desethylamiodarone (DEA), but production of this metabolite was not altered by LPS exposure. The liver injury from amiodarone/LPS was highly dependent on the timing of exposures; amiodarone given 20 or 16 hours before LPS produced pronounced liver injury, but when the interval between drug and LPS administration was shortened, no injury occurred (Fig. 6E).

Fig. 6. Inflammatory stress hypothesis: Amiodarone animal model.

A. Rats were treated with amiodarone (AMD) and lipopolysaccharide (LPS) as shown. B. Timecourse of ALT activity increase in serum. C. Histopathology showing areas of midzonal necrosis only in AMD/LPS-cotreated rats. D. AMD dose-response relationship in LPS-cotreated rats. E. Influence of altering the interval between AMD and LPS administrations. (Modified from [57])

A single administration of LPS to rodents is well known to cause a rapid increase in plasma concentration of TNF that peaks about 1.5 hours after LPS administration. Amiodarone caused a slight prolongation of the LPS-induced appearance of TNF, and pretreatment with etanercept to neutralize TNF reduced liver injury [57], suggesting that this cytokine contributed to the hepatotoxicity. In addition, amiodarone potentiated activation of the coagulation system and the impairment of fibrinolysis caused by LPS [58]. Fibrin deposition in liver and tissue hypoxia occurred in amiodarone/LPS–treated animals before the onset of liver injury. Anticoagulant heparin eliminated the deposition of fibrin and reduced liver injury. Hepatic accumulation of neutrophils was observed in rats treated with either LPS or amiodarone/LPS, but activation of these cells occurred only in cotreated rats. Reduction of neutrophil accumulation with an anti-rat neutrophil serum prevented neutrophil activation and attenuated liver injury, and anticoagulation with heparin prevented neutrophil activation. Thus, the LPS-induced activation of inflammatory cells, the coagulation system and proinflammatory cytokine response are enhanced by amiodarone and contribute to the amiodarone/LPS interaction in rats that is markedly hepatotoxic.

Another drug/LPS interaction model was developed with trovafloxacin. This broad-spectrum fluoroquinolone antibiotic was released to the market in 1998 after no findings of significant liver injury in preclinical safety evaluation or clinical trials. In 1999, trovafloxacin was linked to several cases of liver failure, and its use was curtailed [59]. Trovafloxacin by itself was not hepatotoxic in mice, even at large doses. However, oral administration of trovafloxacin three hours prior to a nontoxic dose of LPS resulted in a large increase in serum ALT activity (5000 U/L) that began within five hours after LPS administration and peaked within 24 hours [60] (Fig. 7AB). This was associated with pronounced midzonal and centrilobular hepatocellular necrosis that appeared both apoptotic and oncotic histologically [60]. Trovafloxacin enhanced and prolonged the LPS-induced appearance of TNF in plasma [60] (Fig. 7C). Results of subsequent studies suggested that this increase was due to both an increase in TNF production and a decrease in its clearance [61,62]. Liver injury was eliminated by pretreatment of mice with etanercept, a soluble TNF receptor that neutralizes TNF (Fig. 7D). Knockout of either TNF receptor 1 or receptor 2 also protected against trovafloxacin/LPS–induced liver injury, suggesting a major role for this cytokine in the pathogenesis [63]. IFNγ in plasma was also increased markedly by trovafloxacin in LPS-treated mice prior to the onset of hepatotoxicity, and liver injury did not occur in IFNγ knockout mice [64]. Moreover, liver injury was less severe in mice with knockout of the gene encoding IL-18, a cytokine that stimulates IFNγ release from immune cells. As with the amiodarone/LPS model, an activated coagulation system and neutrophils contributed to the injury [65,66]. Levofloxacin, a drug in the same pharmacologic class but with far less propensity to cause IDILI, did not affect appearance of the cytokines and failed to cause liver injury upon LPS coadministration [60].

Fig. 7. Inflammatory stress hypothesis: Trovafloxacin animal model.

A. Timecourse of administrations of trovafloxacin (TVX) and lipopolysaccharide (LPS) to mice. B. Development of liver injury after TVX/LPS administration. C. Timecourse of appearance of tumor necrosis factor-α (TNF) in plasma of mice treated with TVX/LPS or levofloxacin (LVX)/LPS. D. Etanercept eliminated liver injury from TVX/LPS cotreatment. (Modified from [60])

Nonsteroidal anti-inflammatory drugs (NSAIDs) also interact with LPS to cause liver injury in rodents. Sulindac is a nonselective inhibitor of cyclooxygenases 1 and 2 and was associated with liver injury and deaths in some patients, in whom liver injury was characterized variously as hepatocellular, cholestatic, or mixed [67]. As for other IDILI-associated drugs, sulindac interacts with LPS to cause liver injury in rodents. In this model, male rats were given two oral administrations of sulindac (50mg/kg) separated by 16 hours; LPS was administered 0.5 hours before second sulindac administration (Fig. 8A). This coadministration resulted in liver injury characterized by large areas of midzonal necrosis [68]. In addition to a large increase in serum ALT activity (1500 U/L serum ALT) (Fig. 8B), increases in gamma-glutamyl transferase, alkaline phosphatase, bilirubin and bile acids occurred, consistent with the clinical presentation in humans.

Fig. 8. Inflammatory stress hypothesis: Sulindac animal model.

A. Treatment protocol for sulindac (SLD) and lipopolysaccharide (LPS) administration to rats. B. Timecourse of ALT increase in plasma from SLD/LPS cotreatment. C. Enhancement of LPS-induced tumor necrosis factor-α (TNF) appearance in plasma by SLD. D. Neutralization of TNF by etanercept reduces liver injury. E. Eglin C, a protease inhibitor, reduces liver injury from SLD/LPS administration. Similar results were obtained with DMSO, a free radical scavenger. (Modified from [68,70,72]).

Sulindac is a sulfoxide that is metabolically oxidized to the corresponding sulfone and also reduced to sulindac sulfide, which is toxic to hepatocytes [69]. The possibility that toxicity was caused by a LPS-induced enhancement of internal exposure to sulindac or its sulfide was explored. However, LPS cotreatment decreased concentrations of both sulindac and sulindac sulfide in liver tissue [70]. In contrast, the concentrations of these two were increased in feces and intestinal tissue, suggesting that LPS cotreatment decreased absorption of sulindac and did not enhance exposure of the liver to sulindac or its toxic metabolite.

As with the two drugs mentioned above, sulindac enhanced the appearance of TNF in plasma of LPS-cotreated rats, and etanercept markedly reduced the liver injury, suggesting a central role for this cytokine [70] (Fig. 8CD). In addition, oxidative stress occurred in the livers of sulindac/LPS treated rats [71] ( Fig. 8). Expression of several genes located in the mitochondria and involved in fatty acid metabolism were downregulated in livers of sulindac/LPS-cotreated rats, suggesting a decrement in mitochondrial function. Expression of glutathione peroxidase also decreased before the onset of liver injury, suggesting impaired cellular antioxidant defense. Sulindac sulfide increased the production of reactive oxygen species in HepG2 cells and decreased mitochondrial membrane potential in isolated mitochondria, raising the possibility that mitochondria were a source of the hepatic oxidative stress [71]. Leukocytes are another likely source of reactive oxygen species in this model. Neutrophils accumulated in the livers of LPS-treated rats, and sulindac prompted the activation of these cells, which was eliminated by neutralizing TNF. Moreover, liver injury was markedly reduced by neutrophil depletion and attenuated by pretreatment with either antioxidants [71] or an inhibitor of neutrophil proteases (i.e., eglin C) [72] (Fig. 8E). As in the trovafloxacin model, sulindac enhanced the LPS-induced activation of the hemostatic system, fibrin(ogen) deposition and liver tissue hypoxia, and anticoagulant heparin or an inhibitor of plasminogen activator inhibitor–1 reduced liver injury. Together, these results suggest that the pathogenesis involves a complex interplay of various inflammatory factors for which TNF plays a critical initiating role.

Diclofenac is another NSAID that remains widely used in some countries, even though it causes rare liver and intestinal injury in patients. Interestingly, osteoarthritis is a susceptibility factor for clinical diclofenac hepatotoxicity, suggesting a role for inflammation in the pathogenesis [73]. Similar to sulindac, a nonhepatotoxic dose (20mg/kg) of diclofenac resulted in liver injury in rats only when they were exposed to a noninjurious dose of LPS two hours earlier [74] (Fig. 9A). Hepatic gene expression of the neutrophil chemotactic factor, MIP-2, was elevated in diclofenac/LPS-treated rats, as was the appearance of MIP-2 in serum. Reduction of blood neutrophil numbers with an antibody reduced liver injury caused by diclofenac/LPS-cotreatment. Subsequently, Ramm and Mally [75] found that a smaller dose of diclofenac (7.5mg/kg) given daily for seven days prior to LPS challenge produced similar liver injury and was associated with hepatic expression of inflammatory cytokines.

Fig. 9. Inflammatory stress hypothesis: Diclofenac animal model.

A. Rats treated with a nontoxic dose (20mg/kg) of diclofenac (DCLF) 2 hr before lipopolysaccharide (LPS) exposure experienced liver injury. A larger, hepatotoxic dose (100mg/kg) of DCLF caused liver injury without LPS coexposure (C) and also resulted in bacteria accumulating in the liver (B). Prior oral administration of nonabsorbable antibiotics decreased the hepatic bacterial load and markedly reduced DCLF-induced liver injury (C). (From [74])

Unlike other IDILI-associated drugs, diclofenac causes liver injury in rats by itself at large doses. Since diclofenac is known to increase intestinal permeability, we explored whether gut bacteria contributed to liver injury caused by a large dose of diclofenac [73]. Rats given 100mg/kg of the drug experienced liver injury (Fig. 9C). Large numbers of gram-negative bacteria were present in livers from these rats, and pretreatment with a cocktail of nonabsorbable antibiotics markedly reduced the hepatic bacteria and the liver injury (Fig. 9BC). Together, these results suggested that diclofenac causes subclinical stress of the liver, rendering it susceptible to injury from intestinal bacteria or their PAMP products (e.g., LPS). Recent studies in vitro support the idea that diclofenac and other IDILI-associated drugs stress hepatocytes, rendering them susceptible to injury from cytokines like TNF that are produced in response to bacterial products [76,77].

In addition to diclofenac, both intestinal inflammation/injury and hepatitis occur as adverse effects from other drugs. These drugs include other NSAIDs as well as checkpoint inhibitors used in cancer chemotherapy (ipilimumab, nivolumab and others) [78–81]. In a mouse model of graft vs. host disease in which the mice had xenografted tumors, ipilimumab and nivolumab treatment exacerbated intestinal injury and modestly increased serum ALT and AST [82]. Furthermore, TNF neutralization with etanercept reduced both of these adverse drug effects. Another drug that has caused both intestinal and liver injury is vemurafenib, a kinase inhibitor used to treat late stage melanoma [83–85]. A causal relationship between intestinal injury and drug hepatotoxicity remains unclear for these and other drugs, but such associations are consistent with drug-induced intestinal disruption and consequent increased gut permeability to PAMPs as contributors to IDILI.

In summary, for the examples of drug/LPS interaction models reviewed above as well as others, TNF was a critical contributor to liver injury (Table 4). LPS administration causes rapid appearance of TNF that is enhanced and/or prolonged by several IDILI-associated drugs. TNF precipitates downstream events such as production of other cytokines, leukocyte activation, oxidant production, coagulation system activation, etc. It might also interact with drugs in hepatocytes to activate cell death pathways directly [3,31,77]. IFNγ appears to be important as well in halothane/LPS and trovafloxacin/LPS models [3]. In a murine model of doxorubicin/LPS-induced liver injury, IFNγ but not TNF appeared to mediate the liver damage [86].

Table 4.

Cytokine involvement in inflammatory stress animal models

Drug with human IDILI liability	Inflammatory stimulus	Species	TNF and/or IFNg Involvement	Reference
Sulindac	LPS	rat	Yes	[72]
Amiodarone	LPS	rat	Yes	[57]
Halothane	Poly I:C	mouse	Yes	[18]
Trovafloxacin	LPS, TNF, PGN/LTA	mouse, rat	Yes	[60,66,115]
Diclofenac	LPS	rat	Yes	[75]
Doxorubicin	LPS	mouse	Yes	[86]

Abbreviations: LPS, lipopolysaccharide; Poly I:C, polyinosinic:polycyttidylic acid; TNF, tumor necrosis factor-α; PGN/LTA, peptidoglycan/lipoteichoic acid

The inflammatory stress hypothesis of IDILI has been criticized because “inflammation such as ulcerative colitis is not associated with an increased risk of IDILI…” [27]. This criticism is based on reports of endotoxemia due to increased intestinal permeability in some patients with inflammatory bowel disease; however, this is not a consistent finding [87]. Furthermore, in human patients with inflammatory bowel disease, LPS concentrations in plasma appear to be modestly and chronically elevated, irrespective of whether their disease was active or inactive [88,89]. This chronic elevation could lead to tolerance to LPS, resulting in an attenuated inflammatory response that would not be expected to interact with drug exposure [90]. Results from inflammatory stress animal models suggest that an acute “flare” of inflammation is needed to precipitate a damaging drug-inflammation interaction and that drug administration must occur in a narrow temporal window relative to the inflammatory stimulus. In fact, IDILI has not been reported to occur in animals with chronic underlying inflammation or endotoxemia. Interestingly, flares of inflammation are known to occur in patients with arthritis, and IDILI reactions are more prevalent in patients with this condition [55,73]. Accordingly, if some drugs do cause IDILI through an inflammatory stress mode of action, then an acute episode of inflammation might be required to occur near in time to peak plasma concentration of drug. This requirement could explain not only why IDILI reactions are rare but also why they tend to occur at variable times after maintenance drug therapy has been initiated.

It has been suggested that the dependence on the window of time between LPS exposure and drug dosing for production of liver injury is unrealistic in humans [91]. For some drugs, the onset of liver injury tends to occur within a certain timeframe after the onset of drug exposure. Since LPS exposure is likely a random event, it would be expected to lead to liver toxicity at random times. However, it is not unreasonable to hypothesize that a bout of acute inflammation with the requisite qualities to interact with a specific drug occurs with a frequency that results in a tendency to precipitate liver injury with a characteristic onset time. Moreover, such a scenario could explain why IDILI reactions in humans occur infrequently.

Another criticism of the inflammatory stress hypothesis is that the inflammatory infiltrate in the liver is predominately neutrophilic in the drug/LPS interaction models, whereas lymphocytic infiltrate is more common in livers of human IDILI patients. Although this is true, it should be borne in mind that the histopathological evidence for lymphocytic infiltrate commonly seen in human IDILI is based on samples taken postmortem or from biopsies taken long after the initiation of injury. It remains possible that some drugs cause an initial, damaging, neutrophilic influx in human patients that is followed by accumulation of other types of immune cells as the injury progresses. Liver lesions in the concanavalin A murine model of autoimmune hepatitis are characterized by lymphocytic infiltration and T cell-dependent hepatocellular injury. Early hepatic accumulation of neutrophils occurs, and neutrophil depletion protects against the liver injury and the influx of T cells [92]. A similar scenario occurs in other organs. For example, pulmonary exposure to particulate silica (SiO2) triggers in lungs of rodents an early influx of neutrophils and alveolar macrophages (the counterpart to Kupffer cells in liver) as the first responders. Continued silica exposure prompts a vicious cycle of continued phagocytosis of silica, cell death and autoantigen release that overrides immune tolerance and leads ultimately to accumulation of lymphocytes and progression of injury [93,94]. These results indicate clearly that neutrophilic influx into tissue can precede and even contribute to later accumulation of lymphocytes. Indeed, they raise the possibility that some drugs prompt a neutrophil-dependent injury early that progresses ultimately into injury mediated by infiltrating lymphocytes.

7. Commonalities in hypotheses

As noted in the Introduction, various hypotheses put forth to explain IDILI pathogenesis are not necessarily mutually exclusive. For example, inflammatory stress can be considered one of the susceptibility factors in the multiple determinant hypothesis. Failure-to-adapt could be a determining factor as to whether liver injury occurs in adaptive or innate immune modes of action (Fig. 3). Similarly, effects of damaging cytokines such as TNF could be a final common pathway to hepatocellular death in hypotheses centered around adaptive immunity or inflammatory stress (Fig. 3).

8. Conclusions and expert opinion

8.1. The adaptive immunity hypothesis remains the most popular theory to explain IDILI pathogenesis, being widely embraced especially by clinicians and clinician-scientists. It is evident that exposure to some drugs can initiate adaptive immune responses in people, but whether such responses are responsible for severe IDILI is less clear. As with other theories about IDILI pathogenesis, evidence in humans that an activated adaptive immune system is responsible for most IDILI reactions is associative and does not prove cause and effect. Associations among drug exposure, liver lesions with infiltrated lymphocytes and liver injury exist for several drugs. However, as noted above, drugs and other chemicals can activate the adaptive immune system without causing hepatocellular injury; conversely, liver histopathology that includes an immune cell infiltrate can result from other initiating mechanisms [21].

To date, attempts to develop robust animal models based on an activated adaptive immune response and in which pronounced liver injury occurs have failed [95]. Indeed, in some animal models activation of the adaptive immune system reduced liver injury [42]. Although it has been claimed that adaptation due to immune tolerance must be responsible for this failure [96], evidence for this assertion is lacking. Moreover, despite claims that “impairment of immune tolerance has led to a model with characteristics very similar to IDILI in humans” [97], none of the animal models that involve sensitization and challenge with a drug and consequent activation of the adaptive immune system have led to the sort of pronounced increases in ALT and decreased liver function that characterize human IDILI of most clinical concern. Thus, that an adaptive immune response underlies clinically significant IDILI is not supported by available animal models. Indeed, considering only the results from presently available animal models would lead an objective observer to consider that an adaptive immune system can evoke modest liver injury after drug exposure, but that severe IDILI results from other, or additional causes. The results of the currently available animal models and the circumstantial nature of the evidence in humans should raise questions about the contribution of sensitization, challenge and adaptive immune system activation in the pathogenesis of severe liver injury from drug exposure.

8.2. The only animal models that have so far resulted in pronounced liver injury are based on either the multiple determinant hypothesis or the inflammatory stress hypothesis. Although these hypotheses can account for characteristics of human IDILI, evidence in humans is circumstantial and not very extensive. Animal models that support the inflammatory stress hypothesis have been developed for several IDILI-associated drugs and result in pronounced necrosis histologically, large elevations of ALT and, in some cases, elevations in other serum biomarkers of liver injury. Additional investigation in people is needed to support the results in animal models.

8.3. In animal models in which pronounced liver injury occurred, the pathogenesis has involved immune cells such as NK cells, neutrophils, Th-17 cells and eosinophils, as well as cytokines such as TNF and IFNγ. Involvement of these cells and soluble immune mediators does not necessarily require sensitization and subsequent challenge with drug that typifies a classical adaptive immune response. Available evidence in animal models suggests that TNF is an apical mediator that, under the appropriate conditions, can prompt activation of the coagulation system, stimulation of downstream immune cells and the consequent release of mediators such as other cytokines and chemokines, reactive oxygen species, toxic proteases, etc. All of these factors can contribute to liver pathogenesis. The identification in animal studies of such critical factors has informed the design of in vitro studies to explore pathways that result in death of hepatocytes. Predictive assays based on results of these studies are currently being explored for use in preclinical drug development.

In addition to the ability of TNF to activate downstream inflammatory events that effect hepatocellular killing, this cytokine can act directly on hepatocytes stressed by drug exposure to cause their demise. This assertion has been supported by studies in vitro [70,98–100]. In cultured hepatocytes, some IDILI-associated drugs cause sublethal stress, rendering the cells susceptible to direct killing by cytokines like TNF and IFNγ [3]. The drug-cytokine interaction involves MAPK activation and caspases. In addition, results in drug/LPS interaction models in mice suggest that some IDILI-associated drugs enhance cytokine release from stimulated leukocytes, a hypothesis subsequently supported by studies in vitro [1,61,62,101]. Thus, animal models of IDILI have contributed to the understanding of potential cellular and molecular mechanisms by which drugs cause hepatocellular necrosis in IDILI.

8.4. Whether IDILI pathogenesis is initiated by an adaptive or an innate immune response, the distal events have common features, such as production of cytokines that can kill hepatocytes (Fig. 3). For example, TNF is produced by cells of both the adaptive (eg, CD8+ T-cells) and the innate (eg, Kupffer cells) immune systems. Results from animal models and cultured cells suggest that TNF and IFNγ are likely candidates as killers of hepatocytes, interacting with offending drugs to do so [3]. Other candidates, such as granzyme, perforin and Fas/FasL, have been activated in some animal models of IDILI [15,18,29] and in drug-treated T-cells in vitro [102,103], but a role for them in the demise of hepatocytes from IDILI-associated drugs has not been established.

Animal and in vitro studies suggest that offending drugs might have several roles in the pathway to liver injury, including becoming bioactivated to a reactive metabolite, enhancing production of immune mediators, prompting failure-to-adapt, inhibiting immune tolerance and sensitizing hepatocytes to the cell killing potential of cytokines (Fig. 3). Identifying cell death signaling events in IDILI models might lead to strategies to treat patients suffering from IDILI, but more research is needed to understand molecular events that result in hepatocellular death in IDILI. The need to speculate about IDILI pathogenesis will be reduced only by greater understanding derived from studies in humans as well as animal and in vitro models.

Article highlights.

• Idiosyncratic, drug-induced liver injury (IDILI) remains an intractable problem in human patients

• Because the pathogenesis remains poorly understood, several theories to explain modes of action and associated animal models have emerged

• These hypotheses potentially have common elements

• Rodent models based on the multiple determinant hypothesis and the inflammatory stress hypothesis are associated with pronounced liver injury, whereas only very modest injury has resulted from models based on the adaptive immunity hypothesis

• Several immune cells and soluble mediators contribute to hepatotoxicity in these models

• Uncertainty and speculation surrounding IDILI pathogenesis will only be reduced by evidence in humans coupled with animal and in vitro models.