The first in vivo tumor models were developed in the mid-1960s. These models were mouse leukemia models grown as ascites. The growth pattern was like that of bacteria in vivo and therefore it was possible to apply similar mathematics of growth and response to these tumors as had been worked out for bacteria. Since the development of the murine leukemia models, investigators have devoted a large effort to modeling solid tumors in mice. There are now a variety of models including syngeneic mouse tumors and human tumor xenografts grown as s.c. nodules, syngeneic mouse tumors and human tumor xenografts grown in orthotopic sites, models of disseminated disease, “labeled” tumor models that can be visualized using varied technologies, and transgenic tumor models. Each of these types of models has advantages and disadvantages to the “drug hunter” searching for improved treatments. [Mol Cancer Ther 2006;5(10):2435–43]
The science of preclinical modeling of anticancer therapies began in the 1950s; however, the establishment of guidelines for experimental quality and end point rigor can be attributed in large part to the group headed by Howard Skipper at the Kettering-Meyer Laboratory affiliated with Sloan-Kettering Institute, Southern Research Institute in Birmingham, Alabama. In the mid-1960s, this group published a series of reports on the criteria of “curability” and on the kinetic behavior of leukemic cells in mice and the effects of anticancer chemotherapy. The principles put forward in these reports were derived directly from the behavior of bacterial cell populations exposed to antibacterial agents and were based on experimental findings in mice bearing i.p. implanted L1210 or P388 leukemia (1–11).
The initial assumptions were that (a) one living leukemic cell can be lethal to the host; therefore, to cure experimental leukemia, it will be necessary to kill every leukemic cell in the mouse, regardless of the number, anatomic distribution, or metabolic heterogeneity, with treatment that spares the host. (b) The percentage, not the absolute number, of in vivo leukemic cell populations of various sizes killed by a given dose of a given drug is reasonably constant. The phenomenon of a constant fractional (or percentage) drug kill of a cell population, regardless of the population size, has been observed repeatedly. (c) The percentage of experimental leukemic cell populations of any size killed by single-dose treatment of drug to the host is directly proportional to the dose level of the drug (i.e., the higher the dose, the higher the percentage cell kill). Thus, it is necessary to kill malignant cells faster than they are replaced by proliferation of the cells surviving the therapy if “cure” is to be approached (6–8).
Skipper and the group at the Kettering-Meyer Laboratory went on to develop the murine L1210 leukemia (11), as well as the murine P388 leukemia (12), into very sensitive and reasonably quantitative in vivo bioassay systems to study anatomic distribution and rate of leukemic cell proliferation and the effects of chemotherapy in tumor-bearing mice. These studies were based on the notion that the principle mechanism by which drug-induced increase in host life span is achieved is through leukemic cell kill, not “inhibition of growth” of the leukemic cell population (13). Leukemic cells that gained access to the brain and other areas of the central nervous system were not markedly affected by certain drugs; therefore, if there were leukemic cells in the central nervous system when treatment was initiated, it was necessary to employ a drug that passes the blood-brain barrier to achieve “cure" (14, 15). A key observation was that there was a close relationship between the number of L1210 leukemic cells inoculated into BDF1 mice and the life span of these mice (13).
Antitumor activity in these early murine ascitic leukemia models was assessed on the basis of percent mean or median increase in life span, net log10 cell kill, and long-term survivors (16, 17). The percent mean or median increase in life span was the ratio of the survival time of treated mice (days) compared with the survival time of untreated control mice (days). Calculations of net log10 cell kill were made from the tumor doubling time determined from an internal tumor titration consisting of implants from serial 10-fold dilutions (Fig. 1; ref. 18). Long-term survivors were excluded from calculations of percent mean or median increase in life span and net log10 tumor cell kill. To assess net log10 tumor cell kill at the end of treatment, the survival time (days) difference between treated and control groups was adjusted to account for regrowth of tumor cell populations that may occur between individual treatments.
As solid tumor models were developed, the appropriate end points devised were tumor growth delay or tumor control. These assays required that drugs be administered at doses producing tolerable normal tissue toxicity, so that the response of the tumor to the treatment can be observed for a relatively long period of time. Treatment with test agents was initiated either before tumor development or after a tumor nodule appeared. If treatment began the day after or the day of tumor cell implant, the experiment was a tumor growth inhibition study. If treatment began when an established tumor nodule (50–200 mm3) was present, the experiment was a tumor growth delay study. Activity in a tumor growth delay is a stronger evidence of clinical potential than activity in a tumor growth inhibition (19, 20).
Experiments where tumor volume measurements are made over a relatively long time until the tumors reach 1.5 to 2 cm3 allow the calculation of the tumor growth delay and percent treated versus control at multiple time points and the exponential tumor volume doubling time. Tumor growth delay is the difference in days for treated versus control tumors to reach a specified volume, usually 1 cm3 (Fig. 2). This value is a critically important measure of antitumor effectiveness because it most closely mimics clinical end points that required observation of the mice through the time of disease progression (21, 22).
The end points applied to solid tumors require several assumptions: (a) the mass of a tumor is directly proportional to the number of malignant cells in the mass; (b) the cells killed by the test agent become nonviable promptly; and (c) the cells remaining viable despite treatment begin to grow again, after a relatively short lag, and proliferate at the same average rate as tumor cells in untreated control mice. Many therapies such as radiation therapy do not kill cells “destined to die” promptly but kill cells over several generations of proliferation, and these methods cannot be applied to such agents with accuracy. The conversion of tumor growth delay values to log10 cell kill is possible if the tumor maintains a log-linear growth pattern and if the volume doubling time of the tumors regrowing posttreatment approximates that of the tumors in untreated control mice (19, 22). For many tumor/treatment combinations, tumor volume provides an inaccurate measure of tumor cell kill (23). Many human tumor xenograft models do not conform to the requirement of log-linear growth and many anticancer therapies do not kill promptly with resumption of log-linear growth in the treated groups.
Response to therapies in solid tumor models can also be assessed with excision assays (23–25). Tumor excision or cell survival assays can be done if the tumor cells grow in mice and in cell culture with a good plating efficiency (1–20% or higher). The ability to measure cell survival (cell killing) from in vivo treatment directly is important because it gives basic information about the definitive cellular effect (Fig. 3). Tumor excision assays also allow greater accuracy and finer resolution between various therapeutic regimens than do in situ assays. Perhaps the greatest disadvantage of excision assays is that extended treatment regimens cannot be used due to tumor cell loss and tumor cell proliferation over the treatment time (26–31). Thus, an excision assay provides a static picture of tumor response a short time after treatment.
Syngeneic tumor models, a mouse tumor growing in mice of the strain in which the tumor originated, offer several undeniable advantages (32, 33). They are relatively low cost; reproducible; grow in immunocompetent hosts; come in a wide variety of tumor types; are generally nonimmunogenic; have a long history of use and therefore a strong baseline of drug response data; the hosts are readily available; and studies are easily conducted with statistically meaningful numbers of mice per group. The main disadvantages of syngeneic tumor models are that the tumor cells are rodent, and therefore express the mouse/rat homologues of the desired targets, and the tumors tend to grow fast. As more targeted therapies, especially antibody therapeutics, are being discovered, the homology between the human and mouse versions of the target can become a serious limiting factor for syngeneic models.
Carcinogen-induced tumors, usually in rats, have proved to be especially useful for the study of chemoprevention agents (32–34). These models, such as azoxymethane-, 3,2′-dimethyl-4-aminobiphenyl–, or N-methyl-N′-nitro-N-nitrosoguanidine–induced colon tumors, have enabled studies of potential chemopreventive agents for colorectal cancer (33). The N-nitroso-N-methyl urea–induced mammary carcinoma model has been instrumental in the elucidation of selective estrogen receptor modulators for the prevention of breast cancer (34). These models have also allowed the understanding of processes of mutation, oxidative stress, and inflammation that occur during carcinogenesis and the identification of new molecular targets for potential chemoprevention. These models are rarely used as cancer treatment models because the tumors are internal; there is some variability; and because the hosts are rats, a larger animal.
Human Tumor Xenografts
The development of human tumor xenograft models was a big step in moving toward more clinically relevant tumor models (35–40). The advantages of using human tumor xenografts are that the malignant cells are human; many of these models are quite reproducible; there are a wide variety of tumor lines available; many of these lines have a long history and therefore a strong baseline of drug response data; the hosts are readily available; and statistically valid numbers of mice can be used in studies. The disadvantages are that these models are more costly to run than syngeneic models; the stromal component of the tumors is rodent; the hosts are immunodeficient; most of the tumor lines were developed using early technology; and most of the time, the tumors are grown in a nonnatural site (s.c.).
Several investigators have moved away from s.c. implants and toward implanting human tumor xenografts (or syngeneic tumors) in the orthotopic site of the primary tumors from which the tumor lines were derived (41, 42). The advantages of this more refined model are that the tumor growth is in the tissue of origin of the primary tumor type; there is some variety in the tumor lines; and the hosts are readily available. The disadvantages of the orthotopic human tumor xenograft model are that the surgeries are often complex, leading to the use of low numbers of mice per study; the models are more costly; the stroma is still rodent and the tumor lines are old; the hosts are often immunodeficient; the tumor growth and response are difficult to follow and therefore survival is often the end point; and statistics are difficult. Both syngeneic and human xenograft tumors can be grown as disseminated disease either from a primary tumor implant or by intracardiac injection of tumor cells. The advantages of a disseminated disease model are that the disease pattern may more closely resemble the clinical problem because the tumor cells home to metastatic sites; there is a good variety of tumor types that can be applied in this model; and the hosts are readily available. The disadvantages are that the tumor implant requiring either a resection of a primary tumor or intracardiac injection of tumor cells is difficult and, therefore, low numbers of mice are used per study; the stroma is rodent; the growth and response of the disease are difficult to follow, often leading to a survival end point; and statistics are difficult.
“Labeled” Tumor Models
To enable orthotopic and metastatic/disseminated malignant disease to be followed, several investigators have developed “labeled” tumor models (43–51). These models involve the development of tumor lines that carry a foreign protein such as green fluorescence protein (GFP), red fluorescence protein (RFP), or firefly luciferase (52–55). The advantages of labeled tumor models are that the progress/response of the tumor can be followed with some form of imaging; metastasis can be visualized; internal tumor nodules can be measured; there is some variety in the tumor lines available; and the hosts are readily available. The disadvantages are that most of the tumor lines are clonal, representing a single modified cell from the original tumor line, and the clonal sublines may be a poor representation of the original disease; visualizing the disease requires costly equipment; the stroma is rodent; and statistics are difficult. It was not long ago that Hoffman (43, 44) announced the development of human and rodent cancer cell lines that stably express GFP and produce metastatic cancer when grown in rodent. The fluorescence from the tumor cell was visualized externally by use of quantitative whole-body fluorescence imaging. The development of GFP-expressing transgenic mice followed shortly thereafter (47). The implantation of RFP-expressing B16 melanoma into GFP-expressing transgenic mice allowed the visualization of tumor-induced host processes with dual-color fluorescence imaging. Tumor cells could be definitively distinguished from host vascular cells, infiltrating cells, and stromal cells, allowing a previously unattainable look at tumor-host interactions (47). To noninvasively examine the significance of hypoxia and hypoxia-inducible factor-1 activation in incipient tumor angiogenesis, Cao et al. (52) genetically engineered HCT116 human colon carcinoma cells and 4T1 mouse mammary carcinoma cells to constitutively express RFP as a tumor marker and GFP as a reporter for hypoxia and hypoxia-inducible factor-1 activation. The mouse dorsal skin-fold window chamber was used to provide a tumor angiogenesis model that was easily visualized. The results showed that incipient tumor angiogenesis preceded the occurrence of detectable levels of hypoxia, and thus initiation of tumor angiogenesis may not depend on hypoxia or hypoxia-inducible factor-1 activation. In another application of a window chamber model, Kashiwagi et al. (53) established an in vivo tissue-engineered blood vessel model consisting of eGFP-HUVECs and DsRed-10T1/2 cells. This model showed that nitric oxide mediates endothelial-mural cell interaction before vessel perfusion and induces recruitment of mural cells to angiogenic vessels, vessel branching and longitudinal extension, and subsequent stabilization of vessels.
Clonal or highly selected sublines of more than 50 human tumor cells lines have been developed that express GFP, more than 25 have been developed that express RFP, and about 20 luciferase-transfected clonal sublines have been developed. In addition, clonal or highly selected sublines derived from several mouse and rat tumor lines and the hamster CHOK-1 line expressing GFP or RFP have been developed (Hoffman website). The molecular characterization of many of the GFP- and RFP-expressing sublines has been minimal. The nomenclature of the lines indicating the fluorescence protein and the tumor line from which the clone was derived may lead to the assumption that these new lines have the molecular and biological characteristics of the tumor line from which they were derived. However, these sublines should be regarded as “new” models and evaluated for the goodness with which they represent the human disease and must be examined individually for molecular characteristics. Using a mouse model carrying primary, genetically modified myc-driven lymphomas (murine stem cell virus–infected lymphoma cells expressing GFP, GFP/p53 null, or GFP/bcl2), Schmitt et al. (45, 46) showed that disruption of apoptosis downstream of p53 by bcl2 or a dominant-negative caspase-9 confers a selective advantage and completely alleviates pressure to inactivate p53 during lymphomagenesis. Apoptosis was the only p53 function selected against during lymphoma development. Defective cell cycle checkpoints and aneuploidy seemed to be by-products of p53 loss. To further elucidate the role of hypoxia-inducible factor in adaptation to hypoxia, Safran et al. (49) developed a genetically engineered mouse that ubiquitously expresses a bioluminescent reporter consisting of firefly luciferase fused to a region of hypoxia-inducible factor that is sufficient for oxygen-dependent degradation. This engineered mouse can be useful for monitoring hypoxic tissues and evaluating therapeutic agents that stabilize hypoxia-inducible factor. To define the role of matrix metalloproteinase-9 in tumor growth and dissemination, Acuff et al. (50) implanted luciferase-expressing human lung cancer–derived A549 cells into mice that were controls or matrix metalloproteinase-9 null. Temporal effects of stromal matrix metalloproteinase-9 expression were followed using bioluminescence imaging. As early as 19 hours after intratrachea injection, there were fewer tumor cells in the lungs of matrix metalloproteinase-9 null mice compared with control mice; however, there was no difference in subsequent tumor growth rate in the mice. Thus, matrix metalloproteinase-9 from the bone marrow seems to contribute to the early survival and establishment of lung tumors but not to the subsequent growth in the lungs. Ramesh et al. (51) developed a clonal cell line, SW780-Luc, by stably transducing the human bladder transitional cell carcinoma with a lentiviral vector coding for the luciferase gene product under control of the cytomegalovirus early gene promoter/enhancer. This tumor line was implanted orthotopically on the luminal surface of the bladder by intravesically instilling the cells in nude mice. The model was used to determine the efficacy of intravesical instillation of a conditionally replicating granulocyte macrophage colony-stimulating factor–armed oncolytic adenovirus for the treatment of bladder cancer using weekly bioluminescence imaging.
Transgenic Tumor Models
Transgenic tumor models have been available for a number of years and are growing in sophistication (56–70). These models involve tumors that arise in genetically engineered hosts. The advantages of transgenic models are that the tumor arises “naturally” in the host and can frequently be followed over a long time course. The disadvantages include that breeding and maintaining a colony large enough to generate sufficient numbers of mice at the same age/gender can be very costly; tumors arise at variable stages animal-to-animal and nodule-to-nodule; the stroma is rodent; the tumors are difficult to follow; the end point is frequently survival; and statistics are difficult (64–79). Most current transgenic models may be more useful for studying the biology of hyperproliferative/malignant cells than for drug discovery.
Two transgenic models, transgenic adenocarcinoma of the mouse prostate (TRAMP) and RIP/Tag2, have been applied to therapeutic studies, albeit of limited scope. Now, several transplantable tumor lines have been developed from TRAMP tumors, making a more readily accessible mouse prostate cancer model. The RIP-Tag2 mouse is a single transgenic that is a model of islet cell carcinogenesis (64). The RIP-Tag2 model has a well-documented and relatively synchronous angiogenic switching and multistage tumorigenesis pattern that allows for testing experimental therapies in three modes: (a) prevention of progression of progenitor lesions to angiogenesis; (b) interventional treatment during the rapid phase of tumor growth; and (c) regression of established tumor burden (65, 66). Recently, Casanovas et al. (67) explored the therapeutic potential of antibodies to vascular endothelial growth factor (VEGF) receptor (VEGFR)-1 and VEGFR2 in controlling angiogenesis in the RIP-Tag2 model. Late-stage tumors were relatively resistant to VEGFR2 blockade because the tumor resumed growth after an initial period of growth delay. There seemed to be induction of fibroblast growth factor and other proangiogenic factors in these mice that bypassed the VEGF pathway.
In the TRAMP model, the minimal regulatory elements of the rat probasin promoter enforce prostate-specific expression of the SV40 T and t oncoproteins (68). In these mice, prostatic intraepithelial neoplasia lesions appear by 8 to 12 weeks of age, moderately differentiated carcinomas by 18 weeks of age, and poorly differentiated tumors appear by 18 to 24 weeks of age (69). Lymph node metastasis can occur as early as 12 weeks of age. The tumors begin as androgen-dependent disease at 12 weeks of age but most are androgen independent by 24 weeks of age (70, 71). The TRAMP model has been used to study tumor angiogenesis (72–74), chemoprevention (75, 76), and immunotherapy and gene therapy (77–79). Huss et al. (72) found that treatment of TRAMP mice between 10 and 16 weeks of age with the antiangiogenic kinase inhibitor SU5416 did not influence angiogenesis or tumor progression. However, treatment of TRAMP mice between 16 and 22 weeks of age with SU5416 decreased tumor-associated vessel density and increased apoptotic index and pronounced regions of cell death. In a chemoprevention study with all-trans retinoic acid, TRAMP mice were treated for 6 or 8 weeks at low, medium, and high dose. Although the mice developed lower grade, more differentiated tumors, there was no change in the incidence of tumors or frequency of metastasis at any dose (75).
With the development of conditional alleles that precisely mimic the mutations found in human cancers and that can be spatially and tissue-specifically controlled using “smart” systems, such as the tetracycline system and Cre-Lox technology, more refined engineered mouse models of the neoplastic disease process are being developed (80, 81). Although these mouse models may allow insights into the fundamental processes underlying normal and malignant cell physiology, the main purpose for developing genetically engineered mice is to model the human disease (80). To test the hypothesis of immunosurveillance, Willimsky and Blankenstein (81) generated a mouse model of sporadic cancer based on rare spontaneous activation of a dormant oncogene, SV40 T antigen. These mice spontaneously developed tumor in a variety of anatomic sites and the tumors grew progressively despite the presence of a tumor specific immune response by the host. The mice had undiminished specific B-cell responses to the tumor; however, the host T cells seemed to be anergic. Serum levels of transforming growth factor-β, which suppresses T-cell function and induces local inflammation, were increased in the tumor-bearing mice (81). Transforming growth factor-β seemed to originate from both tumor and host cells as a component of the malignant process.
Dose, Schedule, and Combination Regimens
Dose and schedule are critical determinants of drug activity (37–41). In a recent study, Graham et al. (40) evaluated the depsipeptide FK228, a histone deacetylase inhibitor, in 39 pediatric xenografts in tumor growth delay and evaluated pharmacokinetic and pharmacodynamic variables with tumor sensitivity. The results showed that FK228 inhibits its target in vivo; however, this effect did not correlate with sensitivity to the compound. For comparisons of time to treatment failure, survival distributions of each treatment group were compared with the survival distribution for the controls using the exact log-rank test. The relatively low objective response rate [3 of 39 (8%) tumor lines showing greater than or equal to partial response and 4 (10%) stable disease] at doses that give clinically achievable drug exposures suggests that FK228 may have limited single-agent activity in pediatric solid tumors. Keyes et al. (82, 83) measured the plasma levels of VEGF, basic fibroblast growth factor, and transforming growth factor-β in mice bearing 12 human tumor xenografts over time. Keyes et al. (83) went on to determine the effects of LY317615 (enzastaurin) on the circulating VEGF levels in mice bearing SW2 small-cell lung cancer xenografts. VEGF levels were markedly decreased in mice receiving enzastaurin; however, this change in VEGF did not correlate with tumor response. In addition to the search for better models, there remains an effort to improve methods used for determination of the additivity/synergy of drug combinations, an understanding of which is vital to developing improved cancer therapeutic regimens (84–88).
In the study of multimodality therapy or combined chemotherapy, it is of interest to determine whether the combined effects of two agents are additive or whether their combination is substantially different than the sum of their parts. Combination index methods, isobologram analysis, and uniform measures response surface analysis each provide a rigorous basis for defining regions of additivity, supra-additivity, and sub-additivity, and protection. It has been recognized that the schedule and sequence of drugs in combination can affect therapeutic outcome. Over the last 30 years, the definition of additivity and therapeutic synergism has evolved with increasing stringency. In early therapeutic modeling, therapeutic synergism between two drugs was defined to mean that “the effect of the two drugs in combination was significantly greater than that which could be obtained when either drug was used alone under identical conditions of treatment” (89–92). Using this definition, the combination of cyclophosphamide and melphalan administered simultaneously by i.p. injection every 2 weeks was reported to be therapeutically synergistic in the Ridgeway osteosarcoma growth delay assay (89).
Conceptual foundations for the use of isobologram analysis of two agent combinations were based on the construction of an envelope of additivity derived from multiple isoeffect plots (isobologram; refs. 84–88). Overall, combinations that produce the desired effect and are within the boundaries of Mode I and Mode II are considered additive, those displaced to the left are supra-additive, whereas those displaced to the right are sub-additive. Combinations that produce effect outside the rectangle defined by the intersections of Ae and Be are protective. Application of isobologram analysis requires dose (or concentration) response data for each agent alone from which the envelope (range) of additive results for the combination is calculated.
As an example, the combination of gemcitabine and navelbine was studied in the syngeneic Lewis lung carcinoma model. Gemcitabine was an active anticancer agent in mice bearing the Lewis lung carcinoma. Gemcitabine was well tolerated by the mice over the dosage range from 40 mg/kg × 3 to 80 mg/kg × 3 (Fig. 4; ref. 93). Navelbine was administered in three different well-tolerated regimens with total doses of 10, 15, and 22.5 mg/kg. Both gemcitabine and navelbine produced increasing tumor growth delay with increasing dose of the drug. To assess the efficacy of the drug combination, the intermediate dosage regimen of navelbine was combined with each dosage level of gemcitabine. These combination regimens were tolerated and the tumor growth delay increased with increasing dose of gemcitabine. The isobologram method was used to determine whether the combinations of gemcitabine and navelbine achieved additive antitumor activity. At gemcitabine doses of 40 and 60 mg/kg, the combination regimens achieved additivity, with the experimental tumor growth delay falling within the calculated envelope of additivity. At the highest dose of gemcitabine, the combination regimen produced less than additive tumor growth delay (93).
The untreated control mice in this study had a mean number of 35 lung metastases on day 20. Gemcitabine was highly effective against disease metastatic to the lungs such that the mean number of lung metastases on day 20 was decreased to 1.0 to 1.5 or 3% to 4% of the number found in the untreated controls. Each of the navelbine regimens decreased the number of lung metastases on day 20 to 10 or 11 or about 30% of the number found in the untreated control mice. The combination regimens were highly effective against Lewis lung carcinoma metastatic to the lungs, with a mean number of <1 to 0 metastases found on day 20. These results support the notion that gemcitabine and navelbine may be an effective anticancer drug combination against non–small-cell lung cancer (93).
Response surface modeling provides an effective statistical approach to modeling complex patterns of synergy, additivity, and antagonism in the same data set (39, 88). Tan et al. (39) developed a class of repeated-measures models to analyze the dose-response relationship in xenograft experiments while accounting for incomplete (missing at random and informative censoring) and variable constraints. The method is based on the expectation/conditional maximization algorithm to estimate the dose-response relationship. Dose ranges of temozolomide and irinotecan and combinations of temozolomide and irinotecan were tested in the Rh18 rhabdomyosarcoma. The results showed that irinotecan was highly active whereas temozolomide had little activity in the Rh18 xenograft. The data analysis showed some evidence for synergism but it fell short of statistical significance (39).
The tumor response end points described in this chapter are well established and reflective of clinical outcomes in the human disease. Whereas the value of the particular tumor lines and the value of the mouse as a host remain the subject of discussion, response end points, survival end points, and tumor cell killing end points are well coordinated with clinical investigations. The s.c. implanted tumor nodule remains the mainstay of tumor models because it is easy to follow and because it has yet to be shown that orthotopic tumor models (unlabeled or labeled) would discover drugs missed by s.c. tumors. Although all of the types of tumor models discussed are being applied to drug discovery, the study of drug combinations, and to the discovery of biomarkers, transgenic tumor models primarily remain as tools to study tumor biology. Most malignant diseases are still without useful biomarkers of response to therapies. Both imaging modalities and plasma/serum markers are being explored in the preclinical setting with hopes of developing clinically useful assays; however, examples of biomarkers/surrogate markers of tumor response remain quite rare. It is important to apply quantitative methods to the assessment of tumor growth and response to therapy and to the determination of the additivity/synergy of combination regimens.