Breast Cancer Research and Treatment 46: 2550.78, £997.
`(9 1997 Klnwer Academic Publishers. Printed in the Netherlands.
`
`Issues in experimental design and endpoint analysis in the study of
`experimental cytotoxic agents in vivo in breast cancer and other models
`
`Robert Clarke
`
`Vincent T. Lombardi Cancer Center, Georgetown University Medical School, Washington DC, USA
`
`Key words: xencgrafts, breast cancer, cell lines, resistance, cytotoxic drugs, synergy
`
`Summary
`
`Considerable effort has been placed into the identification of new antineoplastic agents to treat breast
`cancer and other malignant diseases. The basic approaches, in terms of model selection, endpoints, and
`data analysis, have changed in the previous few decades. This article deals with many of the issues
`associated with designing in vivo studies to investigate the activity of experimental and established
`compounds and their potential interactions. Endpoints for both in situ and excision assays are described,
`including approaches for determining cell kill, tumor growth delay, survival, and other estimates of
`activity. Suggestions for approaches that may limit the number of animals also are included, as are
`possible alternatives for death as an experimental endpoint. Other concerns, such routes for drug
`administration, drug dosage, and preliminary assessments of toxicity also are addressed.
`Statistical
`considerations are only briefly discussed, since these are addressed in detail in the accompanying article
`by Hanfelt (Hanfelt J], Breast Cancer Res Treat 46:279—302, 1997). The approaches suggested within this
`article are presented to draw attention to many of the key issues in experimental design and are not
`intended to exclude other approaches.
`
`Introduction
`
`The disseminated nature of breast cancer and the
`
`tumors are the
`development of crossresistant
`primary causes of failure of current therapies. By
`the time many tumors are detected, there is a high
`probability that metastatic lesions will be present,
`many of which may already contain resistant sub-
`populations [1]. Not surprisingly, there is sub-
`stantial
`interest
`in the identification of novel
`
`cytotoxic and endocrine agents for the treatment
`of breast cancer.
`
`A systemic approach is required to eradicate
`
`the majority of metastatic breast disease, and this
`remains primarily in the form of cytotoxic chemo-
`
`therapy or endocrine manipulation. The latter
`began with the initial studies on oophorectomy by
`Beatson [2] and was followed by the administra—
`tion of high dose estrogens. These were largely
`replaced by the development of antiestrogens,
`
`Address for correspondence and ofiprints: Robert Clarke PhD, Vincent T. Lombardi Cancer Center and Department of Physi—
`ology & Biophysics, W405A The Research Building, Georgetown University Medical School, 3970 Reservoir Road NW,
`Washington DC 20007, USA; Telephone: (202) 687-3755; Fax: (202) 687-7505; email: Clarker@gunet.georgetown.edu
`
`Genentech 2052
`
`Celltrion v. Genentech
`
`|PR2017-01122
`
`Genentech 2052
`Celltrion v. Genentech
`IPR2017-01122
`
`

`

`256
`
`R Clarke
`
`Table 1. Several of the human breast tumor cell lines used
`as xenografts in the current NCI drug screening program*
` Cell Line ER status Cltation
`
`
`
`BT7549
`H5578T
`MCF-7
`
`MCF7/ADR
`(multidrug resistant)
`
`MDA-MBw231
`(metastatic)
`MDAiMB—ABS
`(metastatic)
`
`negative
`negative
`positive
`
`negative
`
`**
`(106)
`(107)
`
`(39,40,106)
`
`negative
`
`(108,109)
`
`negative
`
`(108 110)
`
`
`
`positiveT47D ( l l l)
`
`
`
`* Kindly provided by Dr. Joseph Mayo.
`** There is no primary citation for this cell line listed by
`the provider (ATCC, Roekville, MD), who indicate in their
`“Catalogue of Cell Lines and Ilybridomas“ that the cells
`were derived from a papillary invasive ductal tumor in a
`72-year old patient. Metastatic disease was found in three
`of seven regional lymph nodes. The originators were Dre
`W.G. Coutinho and E.Y. Last'argues. A discussion of other
`human breast cancer cell
`lines can be found elsewhere
`[112].
`
`initially the triphenylethylene Tamoxifen [3] and,
`more recently, introduction of the steroidal anti-
`estrogen ICI 182,780 [4]. Other endocrine strate-
`gies include the use of inhibitors of estrogen
`biosynthesis, primarily aromatase inhibitors, and
`the use of LHRH agonists and antagonists.
`The use of chemotherapeutic agents in the
`management of neoplastic disease began with
`Rhoad’s description of the use of nitrogen mus—
`tard for the treatment of Hodgkin’s lymphoma.
`[5]. The number of cytotoxic agents available has
`increased substantially over the intervening years,
`with breast
`tumors generally exhibiting good
`overall response rates to several cytotoxic drug
`combinations. While chemotherapy can produce
`gains
`in overall survival, most patients with
`metastatic breast cancer will eventually recur.
`Many reasons may account
`for
`this
`failure,
`including poor dose scheduling,
`inappropriate
`combinations of drugs and the emergence of cell
`populations resistant to the antineoplastic agents.
`Animal models provide one approach for the
`
`the
`optimization of drug scheduling and for
`identification of novel compounds that exhibit
`promise in such in vitro prescreens as
`those
`currently utilized by the National Cancer Institute
`[6]. For many years a primary in vivo screen,
`utilizing the L1210 and P388 mun'ne leukemias,
`was an integral part of the NCI‘s preclinical drug
`development program. However, the lack of a
`significant representation of solid human tumors
`drew criticism, and partly explained the weakness
`of the screen to identify new agents active against
`the more common solid tumors, including breast
`cancer.
`Perhaps not surprisingly,
`the screen
`appears to have been more successful in identify
`ing agents active against hematologic malignan—
`cies. Some drugs with well established clinical
`efficacy (busulfan, hexamethylmelamine) fail to
`demonstrate substantial activity in the L1210/P388
`screen [7,8].
`
`The P388 and L1210 in viva screen has largely
`been replaced by panels of disease specific human
`tumor cell
`lines
`(prescreen)
`and xenografts
`(primary screen). While there are relatively few
`ideal in vivo models for breast cancer, several of
`the available human cell lines fulfil some of the
`
`requirements for screening, 6.3., stable phenotype,
`high tumor take rate, predictable and reproducible
`kinetic properties. Several of those currently used
`by NCI are shown in Table 1.
`
`Some considerations for choice of model,
`scheduling, dosage, and endpoint will be dis—
`cussed below. The discussions and issues raised
`
`are to draw attention to different approaches to
`experimental design. AS such, these Should be
`
`considered in the light of the other articles on
`experimental design and data analysis in this issue
`and elsewhere. Some issues are generic, while
`others are more directly applicable to screening
`cytotoxic activities, and may be of lesser rele-
`vance to testing chemopreventive and endocrine
`agents.
`It is hoped that the topics discussed will
`assist
`investigators to consider key issues in
`experimental design. However,
`these are pro-
`vided only as suggestions. There are many ways
`to identify potentially active compounds and drug
`combinations, and these are constantly being
`
`

`

`modified by improvements in both cellular models
`and approaches to the use of animals in biomed-
`ical research.
`
`Log cell kill and tumor kinetics in cytotoxic
`
`cancer chemotherapy
`
`Early attempts to modify cytotoxic therapies were
`
`The initial criteria for de—
`largely empirical.
`signing cytotoxic therapies were based on the
`observations of Skipper, who demonstrated that
`
`cytotoxic drug-induced cell kill follows a similar
`kinetic pattern to that established by Arrhenius at
`the turn of this century for the killing of bacteria.
`The log cell kill hypothesis states that cytotoxic
`drugs kill cells by first order kinetics [9]. Thus.
`a constant proportion of the cells will be killed
`regardless of the size of the cell population. For
`many cytotoxic drugs a 4—log cell kill is achiev—
`able and would eradicate a tumor population of
`103 cells and have a one in ten chance of elimin—
`ating a population of 104 cells [10]. However,
`clinically detectable primary tumors have a cell
`population frequently in excess of 109 cells.
`Many metastases also could contain cell popula—
`tions greater than 104 cells. While theoretically
`sound,
`log cell kill can be affected by several
`biological parameters, including the presence of
`de novo resistant cells,
`the degree of tumor
`vascularity, and other factors that may affect drug
`perfusion and metabolism.
`The principles elucidated by Skipper were
`further modified by Norton & Simon, who ap-
`
`plied Gompertzian growth kinetics to tumor cell
`populations [11]. This clarified the inverse re-
`lationship between growth fraction and tumor
`size. The ability of a cytotoxic drug to inhibit
`tumor cell growth was determined to be directly
`related to the tumor’s growth rate, which also is
`related to tumor volume [12]. The tumor mass
`
`killed by a cytotoxic treatment is proportional to
`the growth fraction multiplied by the total tumor
`volume [12].
`
`A GOmpertzian kinetic growth pattern pro
`duces a growth fraction that ultimately decreases
`
`Breast cancer models for drug screening
`
`257
`
`exponentially with time. This inverse relationship
`between tumor size and growth fraction implies
`that micrometastases should be more kinetically
`sensitive to cytotoxic chemotherapy than larger
`
`tumor masses [13,14]. Thus, early intervention
`when the tumor mass is small should provide the
`
`greatest opportunity for induction of remission.
`Metastases tend to exhibit a more rapid tumor
`doubling time (TD) than primary tumors, particu—
`larly for the common solid tumors like those of
`the breast [15].
`There are a number of factors that contribute
`
`to the apparent TD of any tumor. These include
`the rate of cell production, the size of the growth
`fraction, cell recruitment from G0, and the rate of
`cell loss. Cell loss includes shedding of cells into
`other compartments. e.g.. metastasis. differenti-
`ation to a non—proliferating cell type, or entry into
`
`prolonged G0 and apoptotic cell death. The high
`proportion of nonwproliferating or normal cells
`
`present in many solid tumors, and the frequently
`high rate of cell loss, generally produce tumors
`with a long TD. The rate of cell loss may be as
`high as 80% of the rate of cell production [16].
`
`Growth kinetics in human tumors and animal
`models
`
`The kinetic parameters of tumor growth represent
`one of the major differences between animal
`models and the human disease. While Gompertz~
`ian kinetics apply to experimental
`tumors and
`
`those in patients, the TDs and growth fractions are
`frequently quite different. For example, many
`human breast tumors exhibit long TDs and often
`small relative growth fractions.
`In marked con—
`trast, human breast tumor xenografts generally
`have short TDs and high growth fractions. Estro—
`gen-treated MCF~7 tumors (ER—positive) have TDs
`of approximately 1042 days [17-19] compared
`with over 100 days for many tumors in patients
`
`[20]. We have generated several estrogen-in-
`dependent MCF-7 variants. These have TDs as
`long as 100 days when grown in the absence of
`
`

`

`258
`
`R Clarke
`
`estrogen, but grow as rapidly as parental MCF—7
`tumors in estrogen-supplemented mice [17-19].
`MDA435/LC‘C‘6 cells (ER-negative, ascites variant
`
`of MDA-MB-435) growing as solid tumors have
`much shorter TDs of 3-5 days [21]. Most of the
`breast cancer xcnografts used in the current NCI
`screen have mean TDs of 210 days.
`The differences in kinetic properties between
`xenografts and tumors in patients would tend to
`make the xenografts more sensitive to agents with
`a strong cell cycle/cell phase specificity. For the
`purposes of a primary in viva screen for novel
`agents, a limited overestimate of activity may not
`be a major concern.
`For studies to optimize
`scheduling or combinations of established drugs,
`the relative sensitivity of the in viva screen is a
`concern only if the model is either too sensitive
`or too resistant to a combination, when it will
`become difficult to assess interactions. Most of
`
`these concerns are readily addressed by a careful
`choice of in. viva model(s). Various schedules
`
`have already been identified for the established
`drugs, examples are provided in Table 2.
`
`the dose response relationship and dose
`intensity
`
`The relationship between treatment and response
`is described by:
`
`k—Cxt
`
`res—
`t = time. Thus,
`where C m concentration;
`ponse should be approximately equivalent where
`C x t values (area under the concentration time
`
`curve) are equivalent. This can enable the design
`of clinically relevant in vitro analyses of estab«
`lished drug combinations based on pharmaeokin-
`etic measurements previously obtained in patients
`or animals.
`Clinical studies can utilize the
`
`reasonable across-species dosage relationship of
`mg/m2 to estimate dose from data obtained in
`preclinical animal screening. The doses may re—
`
`quire some further modification, since the serum
`half-life of some drugs can be longer in man than
`in rodents [22]. One approach is to use 1/ 10th
`
`the maximum tolerated dose (MTD) in rodents as
`
`the approximate starting dose for a Phase I trial in
`humans [93]. Where possible, it may he better to
`
`use a dose that produces comparable pharmaco-
`kinetics, since this can increase the predictability
`of the mOuse xenograft to human tumor model
`[24].
`
`The pharmacokinetics for cytotoxic drugs are
`frequently similar in mice and men [25]. For
`many drugs, the mouse LD10 also approximates
`the MTD in humans when expressed as mgl'm2
`[26]. However, there are exceptions. The C x t
`
`values at the LD10 are higher for mitomycin C,
`vincristine, and cyCIOphosphamide, and lower for
`methotrexate and 5-fluorouracil,
`in mice when
`
`compared with humans (reviewed in [26]).
`In clinical practice. a narrow therapeutic index
`is frequently responsible for the reduction of
`dosage due to side effects. However, the steep
`dose response curve for most cytotoxic drugs im—
`
`plies that even a small perturbation in dosage may
`produce a significant change in response. The
`effects of alterations in dosage on clinical res—
`
`It has
`ponse has been widely reviewed [22,27].
`been suggested that a major contributing factor to
`the failure of many treatments is the ad hoc re—
`
`duction of drug dosage [22].
`It has been widely acknowledged that the most
`effective treatments involve a high dose intensity
`chemotherapeutic regimen. This is partly based
`on the steep dose response relationship for most
`cytotoxic
`drugs
`and
`various other
`clinical
`observations. Dodwell er al. [27] have reviewed
`the published data, and reexamined the role of
`dose intensity in reSponse, for a number of the
`more common malignancies. While they con-
`clude that high intensity regimens can produce
`significant advantages in disease free survival,
`clear demonstrations of increased overall survival
`
`are obtained much less frequently. The relation-
`
`ship between dose intensity and response often
`varies with both drug, tumor model, and disease.
`Animal models provide a safe and logical means
`
`to explore this and related issues, rather than
`attempting to identify appropriate or potentially
`dangerous schedules directly in patients.
`
`

`

`Table 2. Examples of dosages and schedules for several established cytotoxic agents. The tumor models against which the
`drugs were tested were of various origins, and not exclusively breast. Some drugs that are not widely used in breast cancer
`are included, since these may be useful as controls for establishing the extent to which a tumor model reflects the human
`disease. Toxicity can vary with strain, sex, age and other parameters, and the information in this Table reflects that diversity.
`
`Breast cancer models for drug screening
`
`259
`
`Drug
`
`Adriumycin
`
`Ara-C6
`
`BCNU
`
`Methyl-CCNU
`
`Cyclophosphamide
`
`S-Fluorouracil
`
`lfosfamide
`
`Melphalan
`
`Methorrexate
`
`Mitomycin C
`
`Ctr-Platinum
`
`Taxol
`
`Taxotere
`
`Vinblastine
`
`Dosel
`
`6 nag/kg
`6 mg/kg
`6.8 mg/kg
`8.5 mg/kg
`24 Inglm2
`
`40 rug/kg
`50 mgfkg
`
`18 rug/kg
`
`18 mgfkg
`
`35 mgfkg
`60 mglkg
`100 rug/kg
`143 rug/kg
`200 mgfkg
`286 rug/kg
`290 mgfkg
`
`32 mglkg
`40 mg/kg
`50 mgfkg
`60 rng/kg
`180 rnglrn2
`
`150 mgfkg
`300 rug/kg
`
`12 mgfkg
`12 mg/kg
`
`4 mg/kg
`15 mg/kg
`20 mg/kg
`
`2 mg/kg
`2.5 rag/kg
`4.5 mg/kg
`5 mg/kg
`18 rngfrn2
`
`4 mgfkg
`4 rug/kg
`7.5 mglkg
`10 mgfkg
`2’? mgi'm2
`
`20 rug/kg
`
`15 mg/kg
`
`3 mg/kg
`
`Route
`
`i.v.
`i.v.
`i.p.
`i.p.
`i.v.
`
`so.
`i.p.
`
`i.p.
`
`i.p.
`
`i.p.
`i.p.
`LV.
`i.p.
`i.p.
`i.p.
`i.p.
`
`i.p.
`i.p.
`i.p.
`i.v.
`i.p.
`
`i.v.
`i.v.
`
`i.p.
`i.p.
`
`i.p.
`i.v.
`i.p.
`
`i.v.
`i.v.
`i.p.
`i.p.
`i.p.
`
`i.p.
`i.v.
`i.p.
`i.p.
`i.p.
`
`i.p.
`
`i.v.
`
`i.v.
`
`Schedule2
`
`4,5,9
`1,5,9
`single dose
`single dose
`single dose
`
`24 hr infusion
`1-7
`
`single dose
`
`single dose
`
`single dose
`0-4, '7—11
`1,5,9
`single dose
`1,15
`single dose
`single dose
`
`5,6,7,8
`1-4, 15-18
`0-4
`1,5,9
`single dose
`
`single dose
`single dose
`
`single
`0,4
`
`1,4,8,11,15,18
`1-5
`03
`
`1.15
`single dose
`single dose
`single dose
`single dose
`
`1,5,]0
`1,5,10
`single dose
`single dose
`single dose
`
`single dose
`
`4,6,8
`
`1,15
`
`Toxicity
`
`N133
`ND
`none4
`LD5
`MTD
`
`none
`alopecia
`
`LD
`
`MTD7
`
`alopecia
`ND
`ND
`none
`LD
`LD
`none/LD
`
`ND
`LD
`ND
`ND
`MTD
`
`none
`none
`
`ND
`ND
`
`LD
`ND
`ND
`
`LD
`LD
`none
`ND
`MTD
`
`ND
`ND
`none
`ND
`MTD
`
`none
`
`MTD
`
`LD
`
`Citation
`
`[67,113]
`[1 14]
`[21]
`[21]
`[25]
`
`[115]
`[1 16]
`
`[21]
`
`[71]
`
`[1 1.6]
`[117]
`[114]
`[118]
`[119]
`[118]
`[81]
`
`[72]
`[I 19]
`[11.7]
`[114]
`[25]
`
`[7'7]
`[77]
`
`[120]
`[120]
`
`[119]
`[119]
`[117‘]
`
`[119]
`[121]
`[21]
`[122]
`[25]
`
`[123]
`[123]
`[21]
`[123]
`[25]
`
`[2]]
`
`[67]
`
`[119]
`
`1 = Dose/injection; 2 = Schedule is given as days unless otherwise indicated; 3 = Not defined or not described; 4 = None reported
`or no deaths; 5 : Lethal dose (one or more deaths attributed to drug~induced toxicity): 6 = 1-l3—D-arabinofuranosylcytosine: 7 =
`Approximate MTD as defined by the investigators.
`
`

`

`260
`
`R Clarke
`
`General considerations in experimental
`design
`
`Choice of host for xenografis
`
`There are several considerations relating to the
`choice of immunocompromised host,
`including
`the degree of immune-competence and the ability
`of the strain to support the growth of the tumor.
`For syngeneic tumors the choice of host should be
`obvious. However, the choice of host for xeno-
`
`grafts is more complex. The host not only should
`facilitate the growth of the tumor but also should
`enable the tumor to maintain a biologically rele-
`vant phenotype.
`It also is useful if the selected
`model enables some estimation of the selectivity
`of drug action and the determination of potentially
`lethal toxicities. For most of the breast cancer
`
`cell lines/xenografts available, the nil/nu mouse is
`sufficient, irrespective of the background strain
`[28].
`
`The potential for immunological modulation
`to contribute to tumor response to combination
`chemotherapy is an important consideration. This
`is particularly relevant when the combination
`includes biological response modifiers that can
`influence the activation of cells associated with
`
`interferons and
`cell mediated immunity, e.g.,
`interleukins.
`In some cases, it may be desirable
`to attempt to either eliminate or exclude effects
`on the immune competence of the host. Thus, the
`choice of an appropriate immune—deficient strain
`may become paramount. Since the different im—
`munobiologies of available rodent hosts have been
`recently reviewed [28], they will not be further
`discussed.
`
`Choice of appropriate tumor model
`
`The model used for the screening of a drug sus-
`pected of activity against a particular
`tumor
`should reflect
`the biological properties of that
`tumor as closely as can reasonably be achieved.
`For example, a drug active against a leukemia
`with a short TD, high growth fraction, and
`
`relatively short cell cycle time would be less
`likely to demonstrate activity in a screen against
`a breast tumor with a longer TD, small growth
`fraction, and long cell cycle time. Thus, choice
`of an appropriate model to screen combinations of
`agents with known pharmacological and kinetic
`requirements
`should (where possible) closely
`reflect
`the major biological properties of the
`human disease. To some extent the model also
`
`should reflect the requirements of the drug. For
`example, there would be limited value assessing
`a drug expected to show no crossresistance to P—
`
`glycoprotein without including a model that ex—
`presses P-glycoprotein.
`Selecting a breast tumor model for screening
`cytotoxic compounds can be problematic. Rela-
`tive to the murine ascites models, most solid
`
`breast tumors exhibit a relatively slow TD. For
`example, MCF-7 tumors have a TD of approxi-
`mately 10-12 days when growing in appropriately
`estrogen supplemented animals [19]. While this
`may be acceptable for screening antiestrogenic or
`chemopreventive agents, cstrogenie supplementa—
`tion can alter the activity of some cytotoxic drugs
`[29,30]. In general, the most rapidly proliferating
`human breast tumor xenografts do not express
`estrogen receptors. MDA435/LCC6 tumors have
`TDs of 2-3 days.
`Since the parental cell
`line
`(MDA—MB-435) was obtained from a patient who
`had not received chemotherapy, these models may
`be useful in screening cytotoxic agents for activity
`against breast cancer [21]. The MDA-MB—231
`cell line also has a comparably rapid TD in vivo
`(R. Clarke, unpublished observations). The more
`rapid TDs of ER-negative xenografts broadly
`reflect
`the characteristics of such tumors in
`
`patients, where ER—negative tumors tend to have
`shorter TDs [31].
`From a purely practical viewpoint, relatively
`slowly proliferating breasr tumors can produce
`significant problems in logistics and experimental
`design. The slower growing tumors frequently
`exhibit significant intertumor variability and can
`require substantial numbers of animals to enable
`meaningful statistical analysis of data.
`Deter—
`mining the period at which a tumor is considered
`
`

`

`"cured" also can become problematic. Even the
`more rapidly proliferating solid tumors with TD 2
`48 hr may require up to four months of post treat-
`ment'observation to establish "cure" [32].
`
`It is unlikely that any one tumor model will
`adequately represent the maior biological charac-
`teristics of a particular malignancy. Thus, the use
`of a series of tumors (where appropriate/available)
`may be. required to determine the sensitivity of a
`particular neoplastic disease to a either a single or
`a combination chemotherapy regimen. However,
`this must be considered in the context of reducing
`
`animal usage, cost, and the value of the additional
`data obtained.
`
`For breast cancer, there are several potential
`models available for screening (Table 1). Most of
`
`the ER—positive models require estrogenic supple—
`mentation for tumorigenicity or maximal growth.
`We have generated ER—positive models that will
`grow without supplementation, but the respective
`TDs are relatively long [19]. While this may be
`more representative of breast tumors in general,
`this characteristic is inappropriate for screening
`cell cycle or cell phase specific cytotoxic com-
`pounds. We also have developed an ascites mod-
`el based on a variant of the MDA—MBABS cell
`
`line (MDA435/LCC6). The pattern of response to
`a variety of cytotoxic drugs appears to reflect
`closely that seen in breast cancer patients [21].
`For example, breast cancers in general respond
`poorly to nitrosoureas [33], as do MDA435/LCC6
`ascites to BCNU. Etoposide also does not pro—
`
`duce long term Survivors, and this drug generally
`has been ineffective as a single agent in breast
`cancer [34,35]. Adriamycin [34], mitomycin C
`[36], and taxo] [37] are among the most effective
`single agents
`in previously untreated breast
`cancer, and all of these drugs produced long term
`survivors in mice bearing the MDA435/LCC6
`ascites.
`The characteristics of several breast
`
`cancer xenografts have been reviewed elsewhere
`[28].
`Another example of the choice of tumor
`model
`is in studies to evaluate P—glycoprotein
`reversing agents, compounds which may have sig-
`nificant potential in some breast cancer patients
`
`Breast cancer models for drug screening
`
`261
`
`[38]. For these types of analyses the choice of
`tumor model
`is critical. Cells to be used as
`
`xenografts should be transfectants rather than
`selected for
`resistance,
`since
`selection can
`
`produce multiple unrelated resistance mechanisms.
`For example, MCF—7ADR (selected for resistance
`to adriamycin [39]), but not MDRlutransdnced
`MCF-7 cells (CL 10.3), are cross resistant to
`Tumor Necrosis Factor [40]. Since both adria—
`
`mycin and Tumor Necrosis Factor can inhibit
`cells by the generation of free radicals [41,42],
`this cross resistance in MCFFJ'ADR
`cells suggests
`
`the presence of functional adriamycin resistance
`mechanisms in addition to P—glycoprotein, includ-
`ing altered expression Uf ilianganous superoxide
`dismutase [40]. These cells also exhibit increased
`
`glutathione transferase and topoisomerase 11
`activities [43,44]. The use of transfected cells
`
`allows for a clearer interpretation of the data.
`Cells concurrently expressing multiple resis~
`tance mechanisms may more closely reflect the
`drug resistance that occurs in patients [38], but
`interpreting responses in a mechanistic light may
`be difficult. This does not invalidate their use
`
`where the purpose is simply to sereen compounds
`for potential antineoplastie activity.
`indeed, the
`choice of a series of models that are too sensitive
`
`will likely identify compounds with limited activ—
`ity in patients, whereas active compounds identi—
`fied in otherwise resistant models may have a
`higher probability of being active in patients [45].
`This is likely to be true in principle, but
`the
`extent
`to which it applies will depend upon
`whether the resistance mechanisms operating in
`the tumor model contribute significantly to the
`
`resistance phenotype in patients.
`It
`is apparent that
`there are two potential
`types of screening approaches, each with different
`objectives that will result in different choices of
`models. Where a broad based. non-mechanism
`
`required, a disease—specific
`oriented screen is
`panel of xenografts with widely differing but
`biologically relevant phenotypes is likely to be
`optimal. Knowledge of the pattern of response to
`a series of established drugs, for each component
`of the panel, is required. Such a panel might be
`
`

`

`262
`
`R Clarke
`
`expected to contain both sensitive and resistant
`models (see Table 1 for examples included in one
`possible panel for breast cancer). For a mechan—
`ism or structure/function based screen, the choice
`
`of Components is likely to depend upon assump-
`tions inherent in the mechanism. For example,
`
`the panel
`where a specific target is identified,
`may contain several models with different levels
`of expression of the target, e.g., P-glycoprorein,
`multidrug resistance related protein [21,38,46,47].
`In many cases,
`this requirement may be most
`effectively met by a series of transfected cell lines
`and their respective control populations.
`
`Phenotypic stability
`
`The stability of the phenotype is a critical deter-
`minant for tumor model selection. Some tumor
`
`xenografts may require periodic Cycles of in vivo/
`in vitro growth in order to maintain the ease of
`reestablishment in vitro for some excision cyto-
`toxieity assays.
`Prolonged in viva growth of
`some established cell lines can re5ult in signif-
`icant phenotypic alterations. We have described
`the isolation of hormone-independent sublines of
`
`the estrogen-dependent MCF—7 human breast can-
`cer cell line following prolonged selection in vivo
`[17,18]. While reSponses to antiestrogens remain
`unaltered [18,48], there are significant changes in
`their responsiveness to estrogens [17,18,481 For
`
`many cell lines, this problem can be overcome by
`using cells within a limited number of passages
`(£10) from a single frozen stock of cells. The
`frozen stock should be from a single passage of
`cells with a well characterized phenotype.
`The intertumor stability of the growth patterns
`
`and cell cycle profiles also may be important
`considerations. A high degree of variability in
`intertumor growrh fractions could significantly
`influence the reliability or case of data interpre-
`tation. The stability of the metastatic potential
`must be well defined. Cells with an unpredictable
`
`metastatic capacity may alter tumor burden and
`affect survival and/or the host's sensitivity to the
`toxicity of cytotoxic treatments.
`
`Therapeutic index and dose scheduling
`
`The difference in the dose response curves of
`
`normal and neoplastic tissues, often referred to as
`the therapeutic window or therapeutic index,
`is
`widely applied in the clinical pharmacology of
`
`cytotoxic drugs [22,49-51]. This difference in
`drug sensitivity enables the administration of
`sufficient drug to produce cytotoxic effects in the
`tumor, but not to induce significant and irrevers-
`
`ible toxicity to normal cells. Unfortunately, many
`cytotoxic drugs exhibit a steep dose response
`curve with a small therapeutic index.
`In many
`cases, the development of unacceptable toxicity is
`the dose limiting factor for cytotoxic chemo-
`therapy.
`The sensitivity of normal cell populations
`
`reflects the rapid growth rate and high growth
`fraction of some cells.
`For example, cells in
`normal bone marrow and the intestinal crypts
`have respective thymidine labeling indices of
`30%—70% and 12%-18% [15]. These values are
`generally higher than observed in most solid
`tumor cell populations. This accounts for the
`high incidence of hematopoietic and gastro-
`intestinal
`side effects
`associated with many
`
`chemotherapeutic regimens. Granulocytes are
`highly susceptible to cytotoxic agents because of
`their high growth fraction and short lifespan, but
`a pool of non-cycling stem cells ensure repopula—
`tion [15]. Generally, the purpose of dose schedul-
`ing is to administer the second and subsequent
`treatments at times that will allow some normal
`
`stem cells to evade the drug and enable repopula—
`tion to occur without permanent damage to bone
`marrow. Examples of some drug schedules for
`rodents are provided in Table 2.
`
`Drug administration in vivo
`
`The route of administration can influence drug
`pharmacokinetics, toxicity. and antitumor activity.
`
`introduces the drug
`injection iv.
`For example,
`directly into the blood. with a relatively rapid
`exposure of hematopoietic stem cells. Adminis-
`
`

`

`tration so. or i.[9. would be expected to produce
`a slower drug equilibration with this compart-
`ment, and produce a delayed or reduced myelo~
`
`suppression. However, local tissue damage could
`be increased relative to the iv.
`route.
`The
`
`toxic/pharmacokinetic differences as—
`potential
`sociated with routes of administration can be
`
`For example,
`controlled by careful planning.
`While an i.v. bolus of an agent may produce
`
`the same dose in mg/kg
`unacceptable toxicity,
`body weight can occasionally be administered
`either by infusion, s.c,. 19.0., or multiple lower
`doses given i.v., with significant alterations in
`host toxicity.
`For the majority of solid tumors, an iv.
`administration most closely reflects the clinical
`administration of the drug. However, for experi—
`mental drugs, the route of administration will vary
`primarily with the physico—chemical properties of
`the drug. The iv. route is generally limited to
`water soluble compounds with a pH >4.0 and
`<85. Water~insoluble drugs can be administered
`either S.c., provided they do not produce un-
`acceptable local damage, or p.0.
`if they have
`sufficient chemical stability.
`For steroid hor—
`rnones and untillorinones that can require S115“
`
`tained delivery, i.p. or 3.0. depots in peanut oil,
`S. c. Alzet mini pumps (Alza Scientific, Palo Alto
`CA), cholesterol—based slow release pellets (In—
`novative Research of America, Sarasota, FL) or
`
`silastic pellets all can produce appropriate plasma
`levels of drug for sustained periods of time.
`The timing of administration also is an im-
`portant consideration.
`Initiation of treatment
`within a few days of tumor cell inoculation may
`produce evidence of activity, when administration
`to established tumors indicates inactivity, For
`
`cytotoxic agents, administration within a few days
`of cell inoculation is often inappropriate. Treat-
`mcnt of established tumors, where these are
`
`clearly palpable and fall within a predefined size
`range,
`is generally more appropriate and allows
`for assessment of the most widely used endpoints.
`
`This size should not be so large as to influence
`response to the drug. Early administration of
`drug is usually appropriate when the tumor is
`
`Breast cancer models for drug screenng
`
`263
`
`directly xenografted from another animal and has
`a rapid TD of only a few days, or perhaps when
`sensitivity is
`required in a primary screen.
`Another exception is for the testing of chemo-
`preventive agents, cg. antiestrogens or retinoids,
`which could be given to high risk women without
`evidence of clinically detectable disease. With
`these compounds, chemopreventiveichemosup-
`pressive activity against low tumor burde