Mitochondrial toxicity assessment in industry – a decade of technology development and insight

Keywords::

• drugs
• liver injury
• mitochondria
• toxicity

In almost all eukaryotic cells, mitochondria provide the vast majority of the ATP required to maintain viability and to execute all cellular functions such as muscle contraction, movement, replication and neuronal action potentials. ATP is generated by the coupling of two independent, but closely interdependent processes; the mitochondrial electron transport system (ETS) that generates an electrochemical potential across the mitochondrial inner membrane, and ATP synthase that phosphorylates ADP using the energy of the potential. The ETS consists of five large protein complexes that transfer electrons down an energy gradient until they ultimately reduce molecular oxygen to water which drives ATP synthesis, hence the name for the overall process is oxidative phosphorylation (OXPHOS). If the function of any of these proteins is impaired, ATP is generated less efficiently imperiling cell viability, and depending on the severity, potentially leading to cell death. Moreover, if the impermeability of the inner membrane is undermined, the membrane potential can bypass ATP synthase and is dissipated as heat. Under this circumstance, ETS is said to be uncoupled from ATP production. Decline in mitochondrial function has been associated with many age-related diseases such as diabetes, cardiovascular impairment, and a host of neurodegenerative diseases such as Alzheimer’s and Parkinson’s. Mitochondria contain their own DNA (mtDNA) and both inherited and acquired mutations in mtDNA have been linked to well-over a 100 mitochondrial syndromes such as Leber’s hereditary optic neuropathy, mitochondrial myopathy, encephalomyopathy, lactic acidosis, stroke-like symptoms and myoclonic epilepsy with ragged red fibers [1].

Drug-induced mitochondrial toxicity has been studied in academic settings for well over 50 years. Indeed, many such studies relied on xenobiotic inhibitors, such as rotenone, antimycin and oligomycin, to deduce the function of the ETS, and uncouplers such as dinitrophenol, to understand how mitochondria generate ATP. As a result, we should not be surprised that pharmaceutical xenobiotics intended to be therapeutics could also have deleterious ‘side effects’ on mitochondrial function. However, most of these side effects were not detected by the preclinical cell and animal studies in use at the time, and our intention here is to provide a brief overview of assays that have been developed to detect off-target mitochondrial impairment. For example, the blockbusters troglitazone and cerivastatin were withdrawn from the market in 1997 and 2001 respectively, because of severe liver injury in the case of troglitazone, and rhabdomyolysis of cerivastatin. Both were shown to impair mitochondrial function leading to the reasonable inference that this could contribute to the etiology of the toxicities [2,3]. Such data have prompted industry to respond by developing early mechanistic-based cell and animal screens to avoid such toxicities and the resulting late-stage attrition in the future.

Drugs can impair mitochondrial function in a variety of ways such as inhibition of the ETS proteins or uncoupling ATP production, as noted above. However, drugs can also inhibit mitochondrial protein synthesis (some antibiotics), or impede mtDNA replication (some antivirals), which undermines mitochondrial viability over longer time frames, and correspondingly erodes cellular bioenergetic status. In this regard, many drugs still on the market, but which carry black box warnings, have been shown to cause organ toxicity at least in part through mitochondrial impairment [4]. Interestingly, although such toxicity is seen with an increasingly wide variety of drug classes, each class contains compounds that impair mitochondrial function, but others that don’t, or do so with substantially less potency. In this way, early preclinical assessments of mitochondrial liabilities will foster development of compounds with the desired efficacy profile while reducing the probability of organ toxicity.

Industry typically assesses compounds in large quantities to determine, for example, ADME properties such as solubility and permeability. However, high-throughput methods to detect mitochondrial toxicity in vitro were unavailable prior to 2004 when one of the authors (Y.W.) joined Pfizer to focus on mitochondrial toxicity. Traditional assessment of mitochondrial function had previously been done using a Clark-type electrode to monitor oxygen consumption using isolated mitochondria or intact cells [5,6].

Meanwhile, traditional cuvette approaches with either isolated mitochondria or tissue homogenates were used to conduct structure activity relationship (SAR) measurements of the five OXPHOS proteins. Although elegant, all of these techniques have insufficient throughput for an industrial setting, and also require substantial amounts of sample and compounds, not to mention technician time.

In response, the first HTS assay developed was in collaboration with Luxcel, an Irish start-up [7] and is a 96-well plate (now 384-well) assay that measures the effects of compounds on mitochondrial oxygen consumption using soluble oxygen sensors and time-resolved fluorescence. This respiratory screening technology (RST assay) allows facile dose–response curves and distinguishes between inhibition and uncoupling [8-10]. Over 300 compounds can be easily analyzed per day at single dose, or 56 compounds in a six-point dose–response curve. This new technology allowed, for the first time, the analysis of entire drug classes rather than a single compound so that the first publications focused on the analysis of all members of a particular drug class. Such studies revealed that the compound with the most severe effect on mitochondrial respiration was usually the one reported to have the most adverse effects in humans. For example, the thiazolidinedione troglitazone is a much more potent mitochondrial toxin than pioglitazone, which is still on the market and well-tolerated in the majority of patients [11]. The same was true when the class of fibrates was examined [11]. Interestingly, concentration of statins where toxicity was found using this in vitro approach was a 1000-fold higher than the reported actual maximum plasma drug exposure reported in humans (C_max) [12]. However, in the case of statins, Westwood et al. [3] had shown that fast twitch muscle fibers are preferentially susceptible to necrosis, and this tissue distribution reflects the distribution of the plasma membrane monocarboxylate transporter isoform 4 for which the statins are a substrate [13]. This suggests that bioaccumulation increases the cytoplasmic concentration compared to plasma levels, although it remains to be determined if statins further bioaccumulate into mitochondria [4].

This notion of accumulation is underscored by data from the biguanides. Although metformin remains on the market and is generally well-tolerated, the other two members of the class buformin and phenformin were withdrawn because of fatal lactic acidosis. Interestingly, metformin initially showed no effect in the RST assay, while buformin and phenformin were readily resolved, with potencies in accord with their EC₅₀s for lactic acidosis in rats [14]. However, when the isolated mitochondria were preincubated with the drugs for 40 min before analysis, mitochondrial function was further impaired by buformin and phenformin, and the mitotoxicity of metformin was obvious [15].

The majority of drugs that have been reported to cause mitochondrial toxicity cause liver injury, in some measure because the liver typically sees the highest organ levels of oral drugs. It is known that liver injury can also entail reactive metabolite formation, inhibition of the bile salt efflux pump (BSEP), formation of reactive oxygen species (ROS oxidative stress) and ensuing glutathione depletion. To parse these effects, we studied the antidepressants nefazodone, trazodone and buspirone using primary rat liver hepatocytes and high content imaging [16]. Nefazodone, which was withdrawn because of idiosyncratic liver injury, showed the most detrimental effect on mitochondrial function, as well as on BSEP, ROS formation and glutathione depletion. However, both, trazodone and buspirone also showed some effects on oxygen consumption and OXPHOS activities. Trazodone is also reported to cause liver injury, whereas no liver injury has been reported for buspirone. When C_max was taken into consideration, the margin of toxicity over efficacy for buspirone was approximately 5 times higher than for nefazodone, and for buspirone it was > 1000. These data led to the conclusion that in vitro data become more meaningful in the context of human exposure, and that isolated organelles potentially can overestimate, or underestimate, a drug effect due to the lack of major metabolism [16].

For decades, nascent drugs were evaluated for toxicity in cells grown in media that contains high glucose levels, typically 25 mm, 5 times physiological levels. Under these conditions, most cells maintain a mitochondria population, but derive most of their ATP from glycolysis, not from OXPHOS. Under these conditions, mitochondrial inhibitors do not kill the cells, with the result that many drugs with mitochondrially mediated toxicity were incorrectly deemed safe. This prompted the development of a cell-based assay, today widely known as the ‘glucose–galactose’ assay [17,18]. This assay utilizes HepG2 cells that are either grown in high glucose or in the absence of glucose, but in the presence of galactose. Unlike cells grown in glucose, which fosters glycolytic ATP production, cells grown in galactose obtain 80% less ATP from glycolysis because of slower kinetics of the reactions necessary to get galactose into the pathway [19]. Under these slower glycolytic conditions, cells rely on OXPHOS rendering them more susceptible to mitochondrial insult thereby revealing toxicity that is primarily mediated by mitochondrial impairment. For drugs with multiple off-target effects, toxicity is likely to be equal in both cell lines [17]. As demonstrated by Hynes et al. [20], the glucose–galactose assay only detects about 2 – 5% of all mitotoxicants, which underscores the contention that most compounds that cause organ toxicity do so via multiple off-target mechanisms. The above-mentioned RST assay is often used as a follow-up to assess potential mitochondrial toxicity, and a modified version of glucose–galactose [21] has been commercialized [22].

Not every cell line is amenable to culture in galactose, so that the 2007 launch of a metabolic flux analyzer [23] permitted analysis of drug effects in any cell type by monitoring simultaneous oxygen consumption rate, and media acidification, tantamount to glycolytic lactate efflux. Drugs that impair mitochondrial function force the cells to accelerate glycolysis, which is corroborated by an increase in the extracellular acidification rate [24]. The flux analyzer is available as a 24- or 96-well instrument. Whereas the glucose–galactose assay has become industry standard for early assessment of hundreds or even thousands of compounds, the Seahorse platform is preferred for follow-up investigation if compounds are advanced in the drug-development process based on an anticipated favorable safety margin. In recent years, a variety of improvements have been made to the Seahorse platform. For example, different plate coatings are available to accommodate hepatocyte and neuronal cultures, and in 2014, inserts were introduced that allow three-dimensional spheroids to be analyzed. This is not just important for the assessment of potential mitochondrial toxicity, but also for any drug discovery screening projects that involve a mitochondrial target. The development of a membrane permeabilization kit and use of different substrates also illuminates which complexes are affected by the drug. In this technique, the plasma membrane is selectively permeabilized while mitochondria remain intact.

It was the availability of all these high-throughput assays that helped to define appropriate therapeutic windows, at least for liver injury. For example, Porceddu et al. [25] tested over 200 drugs known to cause liver injury and found > 90% specificity (ability to correctly predict liver injury) at 100-fold human C_max. To the best of our knowledge, no other retrospective study of drugs causing other organ toxicities has been published, in part because fewer drugs are reported to cause cardiac or kidney injury. However, it must be noted, the Porceddu et al. study [25] did not take into account other possible mechanisms that could contribute to liver injury such as reactive metabolite formation, ROS formation, glutathione depletion, BSEP inhibition and other plasma membrane transporters, as mentioned above. Recently, Aleo et al. [26] have shown that there is a very strong correlation with severe liver injury, when both BSEP and mitochondrial function are inhibited, compared to when only one is impaired.

Since mitochondrial toxicity has only recently been identified as a contributor to organ toxicity, acceptance of such liabilities is still slow; often industrial development teams want to see how their compound performs in animal models and whether an acceptable therapeutic index can be reached. The dilemma is twofold. First, severity of mitochondrial toxicity appears to be yet another idiosyncratic variable. Second, in most instances, animal models fail to reveal such toxicity, as is the case for troglitazone, nefazodone and cerivastatin, all of which passed the requisite regulatory animal studies without any signs of liver or muscle injury. Accordingly, efforts have been directed toward creating animal models that better detect mitochondrial toxicity, such as Boelsterlie’s work on the MnSOD +/− mouse model where mitochondrial superoxide dismutase is repressed. However, even in this model, only some hepatotoxic compounds cause histopathologically overt liver necrosis [27], while others showed effects on enzyme activities or gene expression, neither of which is routinely examined in conventional animal studies in industry [28]. We suspect that the lack of response is due to several factors: in vivo toxicity studies are usually done in drug-naïve, young animals that have robust mitochondrial reserves; lack of sufficient genetic diversity to allow for idiosyncratic responses; absence of environmental factors; co-medication; insensitivity of histopathology for revealing mitochondrial failure. For example, it was recently reported that mouse strains harboring different mtDNA sequences respond differently to mitochondrial toxicants [29]. In the future, we envision that an individual’s mtDNA could be analyzed to inform drug selection, similar to what is already done for many cancer treatments.

Histopathology with the usual H&E stains fails to detect mitochondrial dysfunction until it has become catastrophic. When performed, electron microscopy typically reveals swollen mitochondria with loss of cristae structure. However, this is the typical appearance of mitochondria in many pathologies with diverse etiologies, and does not illuminate whether mitochondrial pathology contributed to etiology, or resulted from toxicities from other pathways.

ATP is often measured as an index of mitochondrial function, but this has problems based on cellular bioenergetics. The adenylate charge is heavily defended via the creatine phosphate pool and creatine kinase, so the energy charge of the cell has to be profoundly repressed before ATP is depleted. A better index of mitochondrial function is the rate of the phosphagen pool recharge after depletion via exercise or ischemia. ³¹P-NMR is an excellent tool for monitoring ATP and the creatine phosphate pool, plus pH and inorganic phosphate, in real time. A small biomass, such as a mouse limb, can be put into the magnet to obtain baseline readings, and then monitor the phosphogens during ischemia, and after reperfusion. Simultaneously measuring myoglobin and hemoglobin oxygenation levels can provide insight into O₂ consumption [30]. Noninvasive assessment of lactate by NMR can also serve as surrogate for mitochondrial dysfunction, but care must be taken because stress and exercise will also raise lactate, and the Cori cycle is confounded when liver function is impaired. Animal core temperature, lethargy or exercise intolerance, and oxygen consumption can also be used as noninvasive reflections of mitochondrial dysfunction.

In addition to the above-mentioned RST, glucose–galactose and seahorse assays, another high-throughput technology was developed to identify specific targets of toxicity. For example, instead of the aforementioned cuvette assays, it is now possible to immunocapture isolated OXPHOS complexes that retain activity allowing individual complex assessments in 96-well formats [8,31]. To our surprise, we found that many drugs with mitochondrial liabilities inhibit multiple complexes. For example, troglitazone most potently inhibits complex IV, but also inhibits complexes II/III and complex V [11,12].

We have screened more than 2000 commercially available or proprietary compounds in the RST assay, and it has become clear that unfavorable physical–chemical properties, such as high lipophilicity (clogP > 3) frequently yields unspecific binding to many targets in addition to mitochondrial targets. As demonstrated in Figure 1, a recent screen of > 530 commercially available drugs confirmed that those with low IC₅₀ values in the RST assay (potent drugs) and low IC₅₀ cytotoxicity values when grown in galactose-containing media, showed predominately clogP > 3. Using the RST assay, Naven et al. [32] were able to develop an SAR approach for the uncouplers, although many of inhibitors require further investigation.

Figure 1. Compounds that inhibit mitochondrial function and cause cytotoxicity most often have clogP > 3.

The final assay to mention was specifically developed to support antiviral and antibiotics programs. Because of widespread experience with use of antivirals for HIV/AIDS treatments, it is generally accepted that such drugs inhibit mtDNA transcription. Moreover, because of the mitochondrial endosymbiotic evolution from bacteria and the corresponding similarity of mitochondrial and bacterial ribosomes, some antibiotics, such as the oxazolidines, also inhibit mitochondrial protein synthesis. These effects do not become apparent until several days after initiating drug treatment, so that neither the short-term RST assay, nor the glucose–galactose viability, is suitable as screening tools for mitochondrial effects. In response, an assay was developed using an antibody approach where mtDNA- and nuclear-encoded mitochondrial protein amounts were measured using high-content imaging screening [33]. If mitochondrial transcription or translation was impaired by the drug, then the amount of mtDNA-encoded protein would diminish, whereas the nuclear DNA encoded would remain unchanged.

Although, many of the advancements in mitochondrial assessments described here were developed or beta tested in a small subset of industrial settings, most companies now have one or more preclinical mitochondrial assays in place. For example, a recent survey of the European mechanism based integrated systems for the prediction of drug induced liver injury (MIP-DILI) initiative revealed that of the 11 industry members, 9 of them are now conducting mitochondrial toxicity screening. MIP-DILI is an innovative Medicines Initiative/European Federation of Pharmaceutical Industries and Associations (IMI/EPFIA)-funded consortium of 26 participants from the pharmaceutical industry, subject matter experts and academic institutions. The 5-year project has total budget of €32.4 million and was launched in February 2012.

In summary, early data began emerging from academia and industry indicating that off-target mitochondrial impairment contributed to the etiology of drug-induced organ toxicity. Indeed, several late-stage drug failures, such as troglitazone and cerivastatin with previously obscure pathways to toxicity, were shown to have direct mitochondrial liabilities. In response, industry started to develop high-throughput assays for mitochondrial impairment, which resulted in the technologies described here. Such assays have revealed that, although mitochondrial impairment may be a characteristic shared by all members of a drug class, the potency varies among the members. Such diversity suggests that chemical motifs responsible for toxicity can be identified and hence avoided, but also encourages optimism that at some point we’ll be able to predict mitochondrial toxicity based on structure alone, as is the case now for many uncouplers. In the interim, even if industry has to determine mitochondrial toxicity empirically, technologies like those described here provide the high-throughput and specificity needed to make drugs of the future safer.

The future

The authors believe that early preclinical screening for mitochondrial impairment can now be implemented in any pharmaceutical setting, and the earlier it is conducted, the more chemical diversity is available to circumvent mitochondrial liabilities. Knowing that poor physical–chemical characteristics are prime determinants of off-target potency will inform medicinal chemistry as hits are identified. If compounds with potential mitochondrial liabilities are moved forward, projected exposure will determine a possible safety margin; a mitochondrial liability at 100 μM is a moot point for a compound with a 1-nM efficacy (unless bioaccumulation is afoot). The dilemma of preclinical species not revealing mitochondrial impairment as frank organ toxicity via histopathology requires more refined approaches such as gene expression analysis [34]. The data with diverse mice strains [29] cautions that mitochondrial susceptibility differs which would contribute to idiosyncratic responses. Whereas haplotype association to disease is well established, no similar associations have been made to toxicity outcomes. New assays, such as analysis of mtDNA mutation load might be helpful to reveal an individual’s genetic make-up. Recent efforts have been focusing on potential biomarkers to identify mitochondrial damage even in the absence of frank lesions. For example, McGill and Jaeschke’s efforts are on liver injury [35,36], while Schnellman’s group is working on kidney toxicity [37,38]. As reporter mice become more advanced [39], one could imagine studying mitochondrial toxicity through noninvasive imaging. However, the above-mentioned difficulties with animal models in general could also jeopardize the outcomes here, but the more we understand, the better we can optimize the animal models to predict physiological reality and so serve as viable models of human responses.

Declaration of interest

Y Will and J Dykens were employees of Pfizer when the studies reviewed here were conducted. Y Will is still an employee of Pfizer, J Dykens is now self-employed. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.