A Predictive Model of Human Myelotoxicity Using Five Camptothecin Derivatives and the In vitro Colony-Forming Unit Granulocyte/Macrophage Assay

Most anticancer and many anti-HIV compounds produce severe myelotoxicity that limits their clinical usefulness (1, 2, 3, 4, 5). Potential toxicity to hematopoietic tissue must be evaluated early in the development of such compounds. However, results from commonly used cancer models in animals (i.e., murine or human tumor xenografts in nude mice) have provided little predictive value for clinical pharmacodynamics and pharmacokinetics in humans. Consequently, many promising anticancer drugs fail during Phase I clinical trials because of myelosuppression effects in patients. Selection of the starting dose for a Phase I trial [typically based on the maximum-tolerated dose (MTD) of the most sensitive species], is critical. The goal is to decrease the risk of a lethal overdose in the first cohort of patients, whereas effectively determining the MTD to achieve the maximum therapeutic effect of an anticancer drug. To achieve this, it is desirable to predict the human MTD for safe and efficient clinical dose escalation studies, while reducing the number of patients treated with an ineffective dose. When combined with efficacy data, preclinical evaluation of hematopoietic toxicity may identify the least toxic analogue with the best therapeutic index in humans. For almost all of the cytotoxic agents, clinical testing starts at a dose that kills 10% of the tested animals (LD₁₀). The dose is then gradually increased in modified Fibonacci steps. In vitro hematotoxicity assessment using a colony-forming unit, granulocyte-macrophage (CFU-GM) as a surrogate marker of myelosuppression could play a key role in linking animal toxicology studies to clinical investigations. However, the predictive bone marrow MTD for the neutrophil lineage will equal the actual human MTD only when the myelopoietic tissue of bone marrow is a primary target of toxicity. Several methods of predicting the human MTD have been proposed for a number of anticancer drugs using in vitro myelotoxicity data (6, 7, 8, 9). These predictions were made on the basis of the concentration that inhibited CFU-GM by 90% (IC₉₀) as a more predictive end point for the MTD in animals and humans. However, because they did not consider the pharmacokinetics of the compounds, the in vitro–in vivo correlations were not evaluated quantitatively. Erickson-Miller et al. (7) showed the differential sensitivity between murine and human myelosuppression in three camptothecin analogues (i.e., topotecan, 9-aminocamptothecin, and camptothecin) and tried to find in vitro–in vivo correlations. However, a correct prediction of human MTD was difficult except for topotecan. Pessina et al. (9) developed a standard operating procedure for optimizing the CFU-GM assay. Whereas the purpose of the CFU-GM assay is to predict the human MTD, it is also necessary to develop a mechanism-based model for accurate pharmacokinetic/pharmacodynamic analysis. Our objective was to validate a human MTD prediction model using five camptothecin derivatives.

CPT-11, 7-ethyl-10-[4-(1-piperidino)-1-piperidino]carbonyloxycamptothecin hydrochloride trihydrate, is metabolized to SN-38 (Fig. 1) in mice and humans. This active metabolite plays an important role during the antitumor effect of CPT-11 (10), which was approved by the United States Food and Drug Administration in 2000 as a first-line treatment for advanced colorectal carcinoma (11). CPT-11 is now widely used as a promising anticancer agent. The principal dose-limiting toxicity of CPT-11 is myelosuppression and diarrhea (2, 12, 13). SN-38 contributes to both the efficacy and toxicity of CPT-11. Consequently, SN-38 was used in the current in vitro study. DX-8951f, (1S,9S)-1-amino-9-ethyl-5-fluoro-1,2,3,9,12,15-hexahydro-9-hydroxy-4-methyl-10H,13H-benzo [de]pyrano[3′,4′:6,7]indolizino[1,2-b]-quinoline-10,13-dione, monomethanesulfonate dihydrate is also shown in Fig. 1. In vitro, DX-8951f has demonstrated stronger antitumor activity than other clinically relevant camptothecin analogues, such as SN-38, topotecan, or camptothecin itself (14). This increased antitumor activity of DX-8951f has been observed in nude mice, using gastric, pancreatic, colon, breast, ovarian, and lung tumors (14). DX-8951f is being evaluated currently in Phase III clinical trials. Myelosuppression, especially neutropenia, is the principal dose-limiting toxicity of the camptothecin derivatives, such as DX-8951f, topotecan (3, 15, 16), 9-aminocamptothecin (4, 17, 18), and camptothecin (19, 20). In clinical studies, camptothecin derivative-induced myelosuppression can be readily managed by administering granulocyte colony-stimulating factor. But if it were possible to predict the hematotoxicity of an anticancer agent and its maximum tolerated level, a predictive model would play an important role in the discovery and development of the new drug.

Thus, we investigated the myelotoxicity of SN-38 and DX-8951f to bone marrow samples from three species (i.e., mouse, dog, and human). Subsequently, we correlated these in vitro data with available in vivo myelosuppression data of SN-38, DX-8951f, topotecan, 9-aminocamptothecin, and camptothecin. On the basis of these findings, we have proposed a mechanism-based predictive model of the human MTD for camptothecin derivatives. This model may extend to other anticancer drugs.

Seven to 10-week-old B6C3F1 mice were obtained from Taconic (Germantown, NY) and Charles River Laboratories (Raleigh, NC). Mice received standard laboratory diet and filtered tap water ad libitum. Animals were sacrificed by CO₂ asphyxiation, marrow was aseptically flushed from the femurs, and single-cell suspensions were prepared by gentle disruption. Cells were washed with medium and adjusted to the appropriate concentration. Medium consisted of Iscove’s Modified Dulbecco’s Medium (IMDM) containing 25 mmol/L HEPES buffer and 5% (v/v) fetal bovine serum, referred to as IMDM/5. The femoral marrow from 4 animals was pooled for each of the three experiments conducted in mice. Canine marrow samples were obtained from the University of Auburn Scott-Ritchey Research Center (Auburn, AL). The 3 dogs used in these studies were from 4 to 6 months old when their marrow was aspirated. Upon receipt, marrow was centrifuged over a Ficoll-Hypaque gradient (ρ = 1.083) to obtain the mononuclear fraction. Cells were then washed in IMDM/5 and adjusted to the appropriate cell concentration. Human marrow samples were obtained from AllCells, LLC (Berkeley, CA) from donors that had given informed consent. The age of the 3 donors ranged from 20 to 45 years. Upon receipt, marrow was centrifuged over a Ficoll-Hypaque gradient (ρ = 1.077) to obtain the mononuclear fraction. Cells were then washed in IMDM/5 and adjusted to the appropriate cell concentration.

The negative control for SN-38 and DX-8951f was 0.5% DMSO (as these compounds were dissolved in DMSO) and was obtained from Fisher Scientific (Fair Lawn, NJ). SN-38 was synthesized by Yakult Honsha (Tokyo, Japan). DX-8951f was synthesized by Daiichi Pharmaceutical Co., Ltd. (Tokyo, Japan). The compounds were weighed and dissolved in DMSO. Serial dilutions were made in DMSO for subsequent addition to tubes containing bone marrow cells and MethoCult hemopoietic culture mix. The final DMSO concentration in all of the cultures was 0.5%. Test articles were tested at final concentrations ranging from 16 pmol/L to 500 nmol/L (SN-38) and 3.2 pmol/L to 500 nmol/L (DX-8951f).

CFU-GM were assayed in duplicate as described by Kuwata et al. (21), dishes containing bone marrow cells (5 × 10⁴ for mouse, 2 × 10⁵ for dog, and 1 × 10⁵ for human). MethoCult GF M3534, GF+ H4535, and GF H4534 methylcellulose media (StemCell Technologies, Vancouver, British Columbia, Canada) were used for mouse, dog, and human, respectively. The test compounds remained in contact with the cells for the entire culture period (14 days). CFU-GM colonies were scored using an inverted microscope with phase contrast. All of the studies were performed in accordance with United States Food and Drug Administration Good Laboratory Practice Regulations (21 CFR Part 58).

The 50%, 75%, and 90% inhibitory concentration values (IC₅₀, IC₇₅, and IC₉₀, respectively) were estimated from log-linear regression relating drug concentrations and CFU-GM inhibition data from individual experiments for these species.

To predict the human MTD for acute neutropenia, three predictive models were applied using in vitro IC₉₀ values for myelotoxicity, the murine MTD (LD₁₀), area under the concentration-curve (AUC) of total (lactone + carboxylate) drug at the MTD, and the plasma unbound ratio for each compound. The actual murine MTD and in vivo pharmacokinetic parameters were taken from the following literature sources: SN-38 (2, 12, 13, 22, 23, 24, 25), DX-8951f (26, 27, 28, 29, 30), topotecan (3, 15, 16, 31, 32, 33, 34, 35), 9-aminocamptothecin (4, 18, 36, 37, 38, 39, 40), and camptothecin (19, 20, 41, 42, 43).

where free fraction of AUC = plasma unbound ratio × AUC.

The plasma unbound ratios are referred to in Table 1.

The observed and predicted MTDs on the basis of the above models were independently subjected to log-linear regression analysis (SAS version 8.02, SAS Institute Inc., Cary, NC) and led to the r² values and Ps reported for the predictive modeling.

Representative results from the experiments in the three species are shown in Fig. 2. The data were normalized to “percent of control” values, using the number of CFU-GM in the 0.5% DMSO control group as 100% for both compounds. A summary of the inhibition caused by these compounds, showing IC₅₀, IC₇₅, and IC₉₀ values calculated, is shown in Table 2.

For mice, IC₉₀ values with SN-38 ranged from 25.3 to 368 nmol/L with a mean ± SD of 191 ± 172 nmol/L, as shown in Fig. 2 and Table 2; the IC₉₀ values with DX-8951f ranged from 9.5 to 16.8 nmol/L with a mean ± SD of 13.6 ± 3.8 nmol/L. For dogs, IC₉₀ values with SN-38 ranged from 9.6 to 15.6 nmol/L with a mean ± SD of 13.4 ± 3.3 nmol/L; IC₉₀ values with DX-8951f ranged from 1.5 to 8.3 nmol/L with a mean ± SD of 5.0 ± 3.4 nmol/L. For humans, IC₉₀ values with SN-38 ranged from 15.8 to 18.7 nmol/L with a mean ± SD of 17.5 ± 1.5 nmol/L; IC₉₀ values with DX-8951f ranged from 1.6 to 12.7 nmol/L with a mean ± SD of 5.9 ± 6.0 nmol/L. Table 2 also shows the IC₅₀ and IC₇₅ values of all three of the species, all of which showed the same general trends that DX-8951f was more myelotoxic than SN-38 and that both compounds exhibited greater toxicity in dogs and humans than in mice. The IC₉₀ ratios of mouse/human for DX-8951f and SN-38 were 2.3 and 10.9, respectively (Table 2).

To predict acute neutropenia from in vitro data and in vivo pharmacokinetic and toxicological data for the five camptothecin derivatives, DX-8951f, SN-38, topotecan, 9-aminocamptothecin, and camptothecin, predictive models A, B, and C (described in Materials and Methods) were applied. Model A predicts in vivo MTD in humans using in vitro CFU-GM data from mice and humans. Model B expands upon Model A incorporating AUC data. Model C builds on Model B by taking into account the free fraction of AUC at the in vivo MTD.

The human MTD values of the five camptothecin derivatives were estimated using the three models (Tables 3,4,5,6,7). For all of the compounds, several Phase I studies were conducted separately with different dosing schedules in several patient groups. Therefore, several MTD values were obtained, depending on the dosing schedule for each compound. These estimates in the human for predicted and observed MTD values of the compounds were plotted for each model (Fig. 3). The predicted MTDs from models A, B, and C were then subjected to log-linear regression analysis individually against the observed MTD using SAS version 8.02 software. As a result, model C (r² = 0.887, P < 0.001) was a more predictive model of human MTD than model A (r² = 0.628, P < 0.001) or model B (r² = 0.463, P < 0.001) for all of the dosing schedules of these camptothecin derivatives.

Upon comparing the five camptothecin derivatives, the in vitro inhibitory values of hematotoxicity suggested that DX-8951f, 9-aminocamptothecin, and camptothecin were comparable in their myelotoxicity in all three of the species, whereas SN-38 and topotecan were less myelotoxic, particularly in mice (Table 8). All of the compounds exhibited greater toxicity in the dog and human than they did in mice, as mentioned for topotecan, 9-aminocamptothecin, and camptothecin by Erickson-Miller et al. (7). We used the IC₉₀ ratios of mouse/human to determine the relative toxicity coefficients for SN-38, DX-8951f, topotecan, 9-aminocamptothecin, and camptothecin. This yielded ratios of 10.9, 2.3, 9.8, 10.6, and 2.3, respectively; those of dog/human were 0.8, 0.9, 0.2, 1.2, and 0.2, respectively (Table 8). SN-38, topotecan, and 9-aminocamptothecin appeared to show greater susceptibility in human and mouse than did DX-8951f and camptothecin. A similar sensitivity between canine and human CFU-GM was observed for all of the compounds, as opposed to murine and human CFU-GM. The IC₅₀ and IC₉₀ values of all of the compounds (except 9-aminocamptothecin) were lowest in dog, followed by those in human, and then in mouse, suggesting that the level of sensitivity to these drugs, in terms of in vitro myelotoxicity, was dog > human > mouse. For the 9-aminocamptothecin, this hierarchy of sensitivity was human > dog > mouse.

With regards to the prediction models, model A requires the murine MTD and the IC₉₀ ratio of human/mouse from in vitro CFU-GM assay data. It is possible to apply this model when a drug shows similar pharmacokinetic properties between mouse and human, as was the case for topotecan. However, this model should be applied cautiously, because species differences in pharmacokinetic properties may unexpectedly be observed, such as for DX-8951f and camptothecin [mean total body clearance: DX-8951f: 12 L/hour/m² in mice³ and 1.8 L/hour/m² in humans (26, 27, 28, 29, 30); camptothecin: 20 L/hour/m² in mice (41, 42) and 1.1 L/hour/m² in humans (19)]. For such compounds, it may be difficult to predict the human MTD on the basis of the murine MTD and in vitro CFU-GM data alone (model A). For 9-aminocamptothecin and camptothecin, species differences of the plasma-free fraction were observed [9-aminocamptothecin: 12% in mice³ and 0.30% in humans (40); camptothecin: 26% in mice (43) and 0.36% in humans (19)]. For such compounds, plasma-free fraction values should be considered in the prediction process. Therefore, models A and B would be inadvised, favoring model C on the basis of the free fraction of AUC at the MTD, which is useful in many cases, because model C integrates the plasma unbound fraction, as well as drug exposure and in vivo toxicological information. For models A and B, interspecies differences of pharmacokinetic parameters are not adequately considered; with these models the human MTD was sometimes predicted to be greater or extremely less than the actual MTD. Model C also supports individualized dose escalation, because the target AUC at the MTD can be determined for each patient. Furthermore, model C also contributes to accelerated dose-escalation. As observed with topotecan and 9-aminocamptothecin, model C is also appropriate for special populations, such as pediatric patients whose pharmacokinetics are different, in many cases, from those of adults. Pessina et al. (9) applied model A as the predictive model of human MTD on the basis of a prevalidation study. In doing so, they mentioned that the pharmacokinetic differences across species can contribute to as much as a 4-fold difference in the MTD, because their prediction model did not incorporate the interspecies differences of total body clearance. This model is valuable when a prediction is needed before a clinical study and human pharmacokinetic information is not yet available. However, considering the accuracy of its prediction, the model would be better served by including pharmacokinetic factors, such as clearance and the plasma protein-binding ratio in each species. Our present retrospective analysis with each population group and several dosing schedules suggested that a prediction model requires both the free fraction of the target AUC at the MTD in terms of pharmacokinetic/pharmacodynamic analysis associated with the interspecies differences. The incorporation of model C may be beneficial to the pharmacokinetically guided drug development program, which is proposed by Collins et al. (44). After a careful collection and fitting of Model C to the preclinical pharmacokinetic/pharmacodynamic information obtained from in vitro and in vivo studies, the human MTD could be estimated using the preliminary pharmacokinetic data obtained from the first group of patients in the starting dose level of a Phase I study. Subsequent dose escalations could be made in an attempt to target the predicted AUC in a series of graded steps in subsequent patient cohorts. Therefore, the construction of a mechanism-based model plays an important role in reducing patient numbers to reach an MTD rather than the use of the traditional dose-escalation system. An acceleration of dose escalation is also available, based on this theoretically accurate model.

For SN-38 (the active metabolite of CPT-11, which contributes to bone marrow suppression as the dose-limiting toxicity in Phase I studies), the model allowed for a successful prediction of the human MTD. Therefore, the prediction of the MTD with model C is applicable to the prodrug of an anticancer agent such as CPT-11. As seen in these studies, it is important to evaluate pharmacokinetic parameters such as AUC at the MTD and the plasma protein-binding ratio of the corresponding active metabolite. The applicability of the approach is limited to agents that have myelosuppression as their dose limiting toxicity and for those that have relatively well-behaved pharmacokinetic/pharmacodynamic relationships. In other words, the pharmacokinetics have to be generally predictable, and the pharmacodynamic response has to be correlated with the plasma drug exposures (or free fraction drug exposure). Further improvements in these correlations might be achieved in the case of the camptothecins by accounting for the degree of plasma lactone stability that can differ for various camptothecin derivatives (45). However, increasing the complexity of the predictive model may or may not additionally improve things. The approach of the construction of the predicting model should be broadly applicable throughout drug development in cancer chemotherapy.

In conclusion, for this family of camptothecin derivatives, model C appeared to be superior to the other models of human MTD prediction, because it takes into account in vivo toxicity and pharmacokinetic data such as AUC at the MTD and plasma protein binding in human and experimental animals, as well as in vitro data of hematopoietic progenitor cells. By identifying the safety margins of these compounds in humans at an earlier stage, it should be possible to conduct dose escalation safely and efficiently in clinical studies of new anticancer drugs and, thus, reduce the number of patients treated at significantly lower doses than potentially therapeutic doses.

The authors would like to thank John-Thomas Farmer, Parrish L. Payne, and Stacy E. Powell (Southern Research Institute) for their technical assistance.