Building a tiered approach to in vitro predictive toxicity screening: a focus on assays with in vivo relevance

Abstract

One of the greatest challenges facing the pharmaceutical industry today is the failure of promising new drug candidates due to unanticipated adverse effects discovered during preclinical animal safety studies and clinical trials. Late stage attrition increases the time required to bring a new drug to market, inflates development costs, and represents a major source of inefficiency in the drug discovery/development process. It is generally recognized that early evaluation of new drug candidates is necessary to improve the process. Building in vitro data sets that can accurately predict adverse effects in vivo would allow compounds with high risk profiles to be deprioritized, while those that possess the requisite drug attributes and a lower risk profile are brought forward. In vitro cytotoxicity assays have been used for decades as a tool to understand hypotheses driven questions regarding mechanisms of toxicity. However, when used in a prospective manner, they have not been highly predictive of in vivo toxicity. Therefore, the issue may not be how to collect in vitro toxicity data, but rather how to translate in vitro toxicity data into meaningful in vivo effects. This review will focus on the development of an in vitro toxicity screening strategy that is based on a tiered approach to data collection combined with data interpretation.

Fig. (1)

Diagrammatic representation of the major steps in the drug discovery process. A typical path for drug discovery is presented. Compounds in a chemical library are screened to identify those molecules that interact with the intended target. Molecules that are positive in this assay “Hits” begin the process of lead identification (Hit-to-Lead) and Lead optimization. In Vitro toxicity screening as well as screens designed to identify ADME, genotoxicity, and cardiac toxicity should be done early in this process in order to identify high risk molecules early.

Fig. (2)

Reasons for drug failure in preclincial and clinical studies.

Fig. (3)

A tiered approach to early toxicity screening. A single in vitro screening platform is unlikely to provide the diverse set of data required to evaluate risk and predict in vivo toxicity. Therefore, a tiered systematic approach to in vitro screening should be employed. Solubility, protein binding, an understanding of the relationship between potency and toxicity, and the identification of severe toxicity should be an early consideration. This is followed by detailed information on the mechanism of toxicity and prediction of toxicity in rodents. The identification of compounds that would cause gene toxicity or have a high risk for producing cardiac toxicity should also be determined as early as possible. Potential issues related to drug-drug interactions (DDIs) and species-specific toxicity round out the toxicity profile in late discovery. Prior to selecting a candidate for preclinical development, the compound should be reviewed in terms of therapeutic area, risk/benefit scenarios, and anticipated duration of exposure.

Fig. (4)

Diagrammatic representation of tissue architecture with precision cut tissue slice. The ability to assess whether a new drug candidate could produce hepatobiliary toxicity, or toxicity resulting from activation of Kupffer cells and subsequent release of chemical mediators, can only be tested in cell models that maintain a cellular architecture similar to the that in vivo. A key advantage of PCTS over primary hepatocytes and co-culture models is that the relative abundance of each cell type is maintained.

Fig. (5)

Importance of multiple endpoints, time and concentration-response curves. The toxicity of rotenone was evaluated with the rat hepatoma (H4IIE) cell line. Cells were seeded into a 96-well culture plate at a density of 10,000 cells per 100 µL of culture medium containing 20% serum. Following a 48 h equilibration period, rotenone was added at concentrations ranging from 0 to 100 µM and allowed to incubate at 37°C with 5% CO₂ for 6 h. These data illustrate the importance of time, concentration-response, and multiple endpoint analysis for interpreting in vitro toxicity data. After 24 h, all of the biochemical endpoints respond in a similar manner, and resolution between response profiles is difficult. The addition of a shorter exposure time allows separation between the different endpoints. It is clear that mitochondrial markers (ATP and MTT) are most sensitive to rotenone. Each point on the graph represents a mean of 4-5 wells. The coefficient of variation ranged from 10-15%.

Fig. (6)

In vitro biochemical toxicity profiles that identify cytostatic drugs. Velcade is a first in class proteasome inhibitor prescribed for the treatment of cancer. The drug provides a good example of the importance of reference points for the interpretation of in vitro data. The drug was tested in the rat hepatoma (H4IIE) cell line as described under Fig. (5). In panel A, the biochemical cell based markers (ATP, MTT, and cell proliferation) appear maximally affected at the lowest exposure concentration. Note that the marker for cell viability (GST leakage) is essentially unaffected. In vitro data should be normalized to cell number, thus the fact that all cell based endpoints follow the loss of cell number indicates a cytostatic effect. In order to confirm this interpretation, exposure concentrations are reduced in order to define the full response curve. In panel B, a classical exposure-response curve was produced. Again, all cell-based markers are dependent on cell number. This means that the only true effect on the cells under the conditions tested was on cell proliferation, which is a desirable effect for an anticancer drug. If a single marker, such as ATP or MTT had been used, the interpretation would have been that the compound was toxic. If a membrane leakage marker had been used, the conclusion would have been that the drug had no toxicity. When cytostatic drugs are identified by in vitro tests, it is important to evaluate the drug in a normal non-proliferating cell model. In panel C, velcade was evaluated in rat primary hepatocytes. There was no change in cell viability or cell number, but there was a reduction in ATP and MTT indicating effects on mitochondrial function.

Fig. (7)

In vitro toxicity profiles produced by methotrexate. Rat hepatoma (H4IIE) cells were used as the test system under the conditions described under Fig. (5). Concentration response data, measured with several markers of cell health, show a pronounced reduction in cell proliferation. A 50% response is a maximum response given that the exposure time was 24 h and the doubling time for these cells is 22 h. Cell viability remains high in the face of inhibited cell proliferation. MTT is dependent on cell number; however, the observed reduction in cellular ATP was independent of cell number. If MTT had been the only assay used to assess toxicity, the compound would have appeared toxic. If membrane leakage had been used to measure viability, the compound would have shown low toxicity. By combining several key endpoints that measure cell health, it is possible to determine the primary effect, most sensitive subcellular target, and intended effect versus unintended toxicity.

Fig. (8)

Chemical structures of three antifungal drugs.

Fig. (9)

In vitro toxicity screening differentiates toxicity between drugs in the same class. The antifungal drugs ketoconazole, itraconazole, and fluconazole ll have a similar mode of action but were not developed at the same time. If the three molecules had been part of a single discovery program, in vitro toxicity screening could have provided important information regarding the relative safety of these drugs. To demonstrate this, all three drugs were evaluated in the rat hepatoma (H4IIE) cell line according to conditions described under Fig. (5). Under the conditions tested, ketoconazole showed the highest potential to produce toxicity, while fluconazole showed the lowest potential to produce toxicity. This interpretation is consistent with clinical observations. Values represent the mean of 4-5 wells. The coefficient of variation was between 10 and 15% across the assays. Standard error of the mean bars are not shown for clarity. C_tox = estimated plasma concentration at steady state where toxicity would be expected to occur in liver, kidney, bone marrow, or heart. ND = not determined.

Fig. (10)

In vitro toxicity profile for terfenadine, an example of a false positive. Terfenadine produced significant toxicity at both 6 and 24 h. If this compound had been evaluated in an early discovery program, the in vitro data would have indicated toxicity. This compound was successful in all IND enabling studies. The reason for the low levels of toxicity measured in vivo is first-pass metabolism. Terfenadine has a bioavailability of <1% and undergoes first-pass metabolism. Thus, in vivo plasma concentrations of parent drug are below toxic levels. In vitro, where metabolism is low, parent compound can reach concentrations that can produce toxicity.

Fig. (11)

Metabolic stability of terfenadine (Seldane). Rat microsomes were used to evaluate the metabolic stability of terfenadine (Seldane). Microsomes were diluted to a final concentration of 0.5 mg/mL in phosphate buffered saline (PBS), pH 7.3, with NADPH (100 µM), and test drug at 1.0 and 10 µM. Following an incubation of 30 min, the amount of parent drug remaining was determined by using LC-MS. The data are expressed as the percent of parent remaining (%R). Seldane and midazolam are examples of drugs that have low metabolic stability. They are subject to first-pass metabolism and typically have low bioavailability. In vitro systems with low metabolic capacity evaluate the cytotoxic effects of parent drug. In order to reduce misinterpretation of these data, it is important to determine metabolic stability.

Fig. (12)

The Interaction of a drug with plasma protein is dynamic.

Fig. (13)

Effects of protein binding on `in vitro` toxicity. The toxicity of drugs with a high affinity for proteins (low Kd) is affected by the amount of protein in the in vitro test system. The response profiles depicted in the graphs demonstrate how the toxicity of a drug can be changed by the amount of protein in the test system. Compounds evaluated early in discovery that show low in vitro toxicity in the presence of high protein concentrations (>10%) should be tested for protein binding. In the example shown, the test compound was evaluated in the presence of 20, 10, and 5% serum. Values represent the mean of 4-5 wells. The coefficient of variation was between 10% and 15% across the assays. Standard error of the mean bars are not shown for clarity.

Fig. (14)

In vitro cytotoxicity of azathioprine. Rat hepatoma (H4IIE) cells were exposed to azathioprine as described under Fig. (5). The most sensitive subcellular endpoint was depletion of ATP and this is consistent with reports from other laboratories. This compound is highly toxic and is known to cause toxicity in patients during therapy. Values represent the mean of 4-5 wells. The coefficient of variation was 10-15% across the assays. Standard error of the mean bars were omitted for clarity.

Fig. (15)

In vitro toxicity screening data combined with clinical data show good concordance. In this experiment, 150 approved drugs were selected and analyzed in a blinded manner. The aim of the study was to determine the relationship between the estimated plasma concentration where toxicity would be expected to occur in rat 14-d repeat dose studies (Ctox), and the maximum therapeutic plasma concentration (MTPC) achieved during the course of therapy in humans. The diagonal red line represents the threshold of toxicity where the estimated plasma concentration for toxicity is equal to the maximum plasma concentration measured in humans. If the Ctox value has any relationship to MTPC, most approved drugs should not exceed this threshold value. The data above indicate that 97% of the approved drugs do not achieve plasma concentrations equal to or greater than the predicted level of toxicity. If the drugs tested had been screened for toxicity early in their discovery life-cycle, those that had estimated plasma concentrations for toxicity that fell below 20 µM would have been flagged as toxic. The horizontal red line depicts this point on the graph. Most of the approved drugs are above this line, however some fall below, and therefore might be considered false positives in the in vitro screen. Upon closer inspection, these compounds all have been associated with toxicity in human patients. The key to why these drugs are still used in the clinic is related to their potency, in vitro margin of safety, and risk/benefit analysis. It is clear that as drug plasma concentrations approach the threshold value (Ctox) of toxicity the probability of an adverse event increases. This graph was developed by Dr. Georgor Zlokarnik and the work was part of a collaborative research project between CeeTox, Inc. and Vertex Pharmaceuticals.