Cancer Therapy Vol 2, 297-304, 2004

Cancer Center Karolinska, Department of Oncology-Pathology, Karolinska Institute and Hospital, S-171 76 Stockholm.

__________________________________________________________________________________

*Correspondence: Maria C. Shoshan, Cancer Center Karolinska, CCK R8:03, Karolinska Institute, S-171 76 Stockholm, Sweden; Tel.: 46-8-51775460; Fax: 46-8-339031; E-mail: mimmi.shoshan@onkpat.ki.se

Abbreviations: cyclin-dependent kinases, (CDKs); daunorubicin, (DNR); doxorubicin, (DXR); farnesylation transferase inhibitors, (FTIs); National Cancer Institute, (NCI); protein kinase C, (PKC); reactive oxygen species, (ROS)

Sophisticated molecular-level knowledge of cancer cell biology has proven difficult to translate into drugs that are clinically effective and the prognosis of patients with metastatic carcinoma is generally as poor today as 30 years ago. Although this is sometimes put down to preclinical researchers not understanding oncology and oncologists not understanding cell biology, a biological explanation appears more adequate, i.e., the insight that the lack of single "wonder drugs" is due to cancer being a complex set of diseases, and to each cancer cell phenotype being defined by combinations of severe functional disturbances.

Some single drugs of course do work wonders. Despite its serendipitous discovery as a toxic compound formed at platinum electrodes during electrolysis of bacterial cultures, the widely used drug cisplatin does cure testicular cancer, even at advanced stages and is also rather effective in ovarian cancer. It is perhaps ironic that cisplatin is clinically widely effective, while a sophisticated VEGF inhibitor may be less so. One explanation might be the ability of cisplatin to rapidly induce pro-apoptotic reactive oxygen species (ROS) — an effect which has long been known to underlie its oto- and nephrotoxicity. There is no reason why similar "side effects" would not be involved in its antitumoral activities, and it has indeed been shown that scavengers of ROS will block cisplatin-induced apoptosis in cultured tumor cells. Again, it is also ironic that had the antiproliferative effect of cisplatin not been discovered in the sixties (Rosenberg et al, 1965), this drug would not have been considered for evaluation by current chemogenomics, due to its chemical reactivity and lack of drug-likeness ("druggability") as generally defined by lipophilicity, H-bonding properties etc (Lipinski, 2000).

Cisplatin and its various derivative forms are efficient against a wide range of solid tumors. All of them are classified as DNA-damaging agents which form bifunctional DNA adducts. However, it is only about 1% of the cellular uptake of cisplatin that actually forms DNA adducts (Eastman, 1991). The activated cisplatin molecule is electrophilic and highly reactive also towards, e.g., cysteine and methionine residues in polypeptide chains (Peleg-Shulman et al, 2002). Nephrotoxicity is a major dose-limiting side effect of cisplatin which depends on induction of antioxidant-sensitive lipid peroxidation in renal cortical cells (Hannemann and Baumann, 1988). Cisplatin-induced apoptosis as well as necrosis in renal cortical cells were shown to involve peroxidation and calcium release from intracellular stores (Kim et al, 1997; Baek et al, 2003). Ototoxicity is another serious side effect of cisplatin, and there are indeed numerous reports on the protective in vivo and in vitro effects of antioxidants on cisplatin-induced toxicity (Kim et al, 1997; Baliga et al, 1999; Kalkanis et al, 2004; Ekborn et al, 2003; Schaaf et al, 2002) and many more. The rapidity (minutes — hours) of ROS induction, peroxidation and lethal cell injury in these - and other nonproliferating - systems suggests the existence of cisplatin targets other than DNA. There is no reason to believe that these targets would not be affected also in tumor cells.

To test the possibility that cisplatin has other targets besides DNA, we treated tumor cell cytoplasts, i.e., cells from which nuclei have been removed, with cisplatin at clinically relevant concentrations and found that apoptosis was induced also in the absence of nuclear DNA, conclusively demonstrating a non-nuclear effect in the presence as well as absence of p53 (Mandic et al, 2003). This finding may furthermore help explain why p53 is not always a determinant of the anticancer action of cisplatin, despite induction of DNA damage.

Indisputably, DNA damage and p53 are involved in the total efficiency of cisplatin, and are of particular importance under conditions of low intracellular cisplatin levels leading not to apoptosis, but to an irreversible damage-induced senescent state. The role of senescence as a major response determinant (Roninson, 2004) is of increasing interest and in agreement with the hypothesis of multiple, dose-dependent effects. In our hands, cisplatin at low doses (up to 2-5 m M depending on cell line) does induce DNA damage, inhibition of proliferation by senescence, but not apoptosis, whereas more than five times higher doses (15-30 m M) are required for induction of apoptosis (manuscript in preparation). The total antitumor effect of cisplatin is therefore the sum of nuclear and non-nuclear effects, or, differently put, the sum of two types of proliferation inhibition.

Anthracyclines, such as doxorubicin (DXR), are classified as inhibitors of topoisomerase-II, and have, like cisplatin, been in widespread clinical use for decades as some of the most efficient and wide-spectrum anticancer drugs ever developed. Despite the classification, topoII-mediated DNA lesions do not correlate well with DXR-mediated tumor cell killing (Minotti et al, 2004). The toxic side effects, notably cardiotoxicity, of DXR have long been known to involve oxidative events. These were recently shown to be pro-apoptotic and to involve oxidization of DNA bases leading to a varied spectrum of DNA lesions (Minotti et al, 2004). Significant increases in at least nine different oxidized, cytotoxic DNA bases were identified in vivo in peripheral blood cells from patients undergoing DXR infusions (Doroshow et al, 2001). The complex chemistry of anthracyclines allows the intracellular formation of a range of free radicals and other reactive or toxic metabolites, and oxidative damage is indeed now considered important also for the pro-apoptotic and antitumoral effect of anthracyclines (Minotti et al, 2004). Daunorubicin (DNR) differs from DXR only in that the side chain of DXR ends with a primary alcohol, whereas in DNR it ends with a methyl. Nevertheless, DXR and DNR are used in different spectra of disease: DXR in breast cancer, childhood solid tumors, soft tissue sarcomas and in lymphomas, whereas DNR is used in acute lymphoblastic or myeloblastic leukemias (Minotti et al, 2004). The structural difference between DXR and DNR may alter lipophilicity and thence cellular uptake. However, the possibility that these drugs induce in part different signaling is supported by the fact that DNR and DXR appear in different mechanistic clusters in the response pattern map of growth inhibitory drugs created by the National Cancer Institute (see section V).

Figure 1. General model of multiple drug effects Many standard drugs induce various growth inhibitory effects via different types of DNA damage. However, concentration-dependent drug effects on non-nuclear targets may also contribute to the net effect, either by directly inducing cell death and/or via lipid peroxidation which in turn can lead to secondary DNA damage. In non-proliferating non-tumor cells, effects on non-nuclear targets probably dominate, and subsequent apoptosis in sensitive organs leads to undesired side effects, e.g., neuro- and nephrotoxicity.

Figure 1 illustrates that the total antitumoral effect of many drugs, such as cisplatin and DXR, is the sum of DNA damage plus additional multifunctional damage originating in non-nuclear compartments. The toxic side effects in differentiated, low-proliferative normal cells are primarily caused by these non-nuclear effects.

The plant alkaloid ellipticine is a topoisomerase-II inhibitor, and is used in the treatment of, e.g., gliomas. Similar to cisplatin, it was found to induce DNA-damage independent apoptosis in cells lacking nuclei (Hägg et al, 2004). Etoposide, a topoisomerase-II inhibitor, classically induces DNA-damage and ensuing apoptosis but can at higher doses also induce a distinct effect on mitochondria (Robertson et al, 2000, 2002). In accordance with effects on mitochondria rather than primarily on DNA, etoposide treatment of HL60 cells induces ROS already after 15 min, protein carbonylation within one hour, and cell death at 4 h posttreatment is reduced by half by a ROS scavenger (England et al, 2004).

Methotrexate is an antimetabolite which prevents DNA/RNA synthesis, but which also has anti-inflammatory and immunosuppressive effects. Depending on cell type it can induce reactive oxygen within 4 h and massive apoptosis within 16 h, or growth arrest. Both responses were inhibitable by scavengers of ROS or by GSH (Phillips et al, 2003).

Paclitaxel exerts an antiproliferative effect via mitosis-inhibiting stabilization of microtubuli. When added to purified mitochondria, this drug induced loss of mitochondrial membrane potential and formation of reactive oxygen (Andre et al, 2000); paclitaxel-induced release of cytochrome c subsequent to these same events has also been reported (Varbiro et al, 2001). Added to intact cells at 10 m M, paclitaxel induces release of mitochondrial Ca²⁺ within seconds probably via an effect on the mitochondrial permeability transition pore (Kidd et al, 2002). Although 10 m M would appear an unphysiologically high concentration, similar levels have indeed been observed in plasma (Hajnoczky et al, 1994). Long-term incubation with paclitaxel at low concentrations is also known to lead to its accumulation to 40 m M in cells and tissues (Jordan et al, 1993).

Together, these and other examples (Table 1) suggest to us that although the generally accepted classification of drugs according to basic mechanisms of action is correct, the very labels may prevent viewing the drugs in terms of other possibilities.

Table 1. Additional targets or effects of some standard chemotherapeutic drugs

Signal transduction pathways specific for tumor cells have long been principal targets for drug development. Because activation of protein kinase C (PKC) promotes tumor formation, inhibitors of PKC were early on investigated as potential new drugs. Several of these, e.g., UCN-01 (7-hydroxystaurosporine) and bryostatin-1, were indeed found to have antitumor activity in clinical trials, although it remains unclear which PKC isoforms are critical to the these effects. However, the major effect of UCN-01 is mediated by inhibition of cyclin-dependent kinases (CDKs), inhibition of AKT and PKC-independent induction of apoptosis (Dai and Grant, 2004). Similar to the tumor promoters phorbol ester and mezerein, bryostatin-1 can either activate or downregulate PKC, depending on the duration of exposure. Unlike the tumor promoters, however, bryostatin-1 presents a range of antiproliferative effects in tumor cell lines, e.g., proliferation inhibition, differentiation and apoptosis, suggesting a degree of promiscuity in its mode of action in vitro. Nevertheless, in clinical trials, the use of PKC inhibitors appears to lie more in combinations than as single agents (Swannie and Kaye, 2002; Kortmansky and Schwartz, 2003). Similar to PKC inhibitors, the antisense oligonucleotide ISIS3521 (LY900003) directed towards PKCalpha has modest effects in clinical trials. Since ISIS3521 by nature is more specific than the inhibitors, this suggests that at best PKCalpha is crucial only to certain types or subtypes of cancer cells.

Several small molecule inhibitors of receptor and non-receptor tyrosine kinases have recently been developed for anticancer therapy (Smith et al, 2004). These drugs fall into three categories: inhibitors of the EGFR tyr kinase family (e.g., gefitinib (ZD1839; Iressa)), inhibitors of the split kinase domain tyrosine kinases (e.g., SU11248), and inhibitors of tyrosine kinases from multiple subgroups (e.g., STI571 (Gleevec/Glivec)) (Laird and Cherrington, 2003). Gefitinib was originally intended to target lung cancers due to their often high levels of EGFR. However, as a single agent it yielded few responders in clinical trials; neither was there any correlation between response and EGFR levels. This was later explained by the finding that gefitinib blocks only such EGFR that contain an activating mutation (Lynch et al, 2004; Paez et al, 2004), probably making gefitinib a highly specific drug both in terms of target and of disease. Other tyr kinase inhibitors may be fortuituously or purposely designed to be promiscuous, i.e., to inhibit a certain group of kinases rather than one specific. SU11248 is a multi-target tyrosine kinase inhibitor affecting FLT3 which is often activated in acute myeloid leukemia (O'Farrell et al, 2003). It also blocks c-Kit and PDGFR (Abrams et al, 2003). STI571 was originally found to inhibit the Bcr-Abl chimeric kinase in CML patients, but is now known to inhibit also the c-Kit tyrosine kinase. A perhaps unexpected effect of STI571 in a murine in vivo system was its tumor-specific enhancement of uptake of another drug, epothilone B (Pietras et al, 2003).

Ras is the most frequently mutated protooncogene in human cancers. Because farnesylation of the Ras protein provides the plasma membrane association required for its full activity, farnesylation transferase inhibitors (FTIs) have been developed. For those which have been successful in clinical studies, it has been clear for some time, however, that this is not the key mode of action. For instance, FTIs have shown activity in leukemias and some solid tumors regardless of the Ras mutational status (Russo et al, 2004). The antioncogenic protein RhoB, which is structurally and functionally related to Ras, was early on believed to be a candidate alternative target of FTIs, and it was recently suggested that FTIs act by recruiting RhoB to interfere with pro-oncogenic signaling (Zeng et al, 2003). FTIs may also inhibit farnesylation of other proteins, e.g., prelamins A and B and centromere proteins (Russo et al, 2004). R115777 is one of the most promising FTIs in clinical trials and shows activity in, e.g., myelomas, breast, lung and gastrointestinal carcinomas. In a study on advanced myeloma it induced stable disease in more than 60% of the patients, although this effect did not correlate with inhibition of farnesyltransferase (Santucci et al, 2003). In an in vitro myeloma system, it induced apoptosis also when Ras remained farnesylated, and this apoptosis appeared to involve both the mitochondria and the ER (Beaupré et al, 2004).

The above examples suggest that cornerstone conventional drugs are promiscuous and therefore efficient, but have toxic side effects, whereas new drugs may be more specific and less toxic, but act primarily as sensitizers in combinations with conventional drugs, including various antibody therapies. Novel drugs have usually not been fully evaluated yet, but when potentially efficient as single agents, e.g., the FTIs, they appear to have several targets.

This is entirely consistent with the biological insight that the numerous functional disturbances in cancer cells are determined by likewise numerous combinations of molecular signals, i.e., that no single oncogenic mutation can on its own lead to, or maintain, cancer. It is also in accordance with empirical as well as recent experimental experience showing that combination therapies of many kinds are generally more efficient. This concept of promiscuous drug efficiency is at some variance with "classical" medicinal chemistry, which, however, is mainly aimed at diseases with a biologically less complex etiology than cancer.

Combinations of several different single-target treatments are illustrated, e.g., by a murine xenograft model where the complete, but not the partial, combination of antisense Bcl2/Bcl-xL, antisense protein kinase A and ZD1839 (Iressa) resulted in 50% growth inhibition and 50% tumor-free mice after 5 weeks (Tortora et al, 2003). Combinations of novel inhibitors of cyclin-dependent kinases with other single-target agents have also shown preclinical promise (Dai and Grant, 2004).

It is reasonable to believe that future scenarios will involve rationally developed combinations of low-dose cytotoxic, promiscuous agents plus targeted sensitizers. This allows individual adjustment of treatment according to cancer type, and possibly also to each individual patient. Alternatively, a great number of targeted drugs, one each for a large number of well-defined sets of signal transduction disturbances, might be envisaged to constitute a "building-block" set from which the individual treatment is to be rationally created. In either case, adequate diagnostic tools for identifying the specific combinations of dysregulated signal transduction pathways of each individual tumor is a prerequisite.

These scenarios thus require detailed knowledge of signal transduction pathways and how they interact. This knowledge must be not only on a molecular and intracellular level, but should incorporate also micro-environmental, pharmacological and pharmacokinetic parameters. The order, timing and uptake of drugs in a certain combination are important, as are the drug effects on and influence of surrounding tissue. The type, concentration-dependence and kinetics of cell death are crucial, since these parameters affect the choice of therapy protocol as well as the induction of secondary effects and feedback loops which affect the net outcome. Some of these aspects are summarized in Figure 2. It should also be kept in mind that cell death type and kinetics are to a great extent determined also by factors such as the oxygenation and energy status of each tumor cell.

Many current drug discovery projects use vast drug libraries screened for molecules showing in vitro binding to and inhibition of a specific target, often a protein shown to be overexpressed in tumors or to be otherwise implicated in the phenotype. With this approach, parameters such as cell permeability, general propensity for macromolecule interactions ("stickiness") and the tumor cell killing potential of the drugs are investigated only at later steps, making late attrition of impossible leads an expensive problem. Computational design of molecules to fit a specific target molecule structure is another approach to drug discovery which also requires that the actual cell kill of each candidate be assessed after its synthesis. Cell-based screening is therefore on the increase, its obvious advantage being that the desired effect — proliferation inhibition or cell death — is also the endpoint. Thus, all non-permeable compounds are immediately excluded, and, importantly, cell-based screening also directly confronts the complexity of tumor cell signaling. However, as indicated in Figure 2, any experimental protocol and interpretations should take into careful consideration the time-scale, the concentration range and the type of endpoint determination.

Using the simple colorimetric MTT assay, the National Cancer Institute (NCI) has screened more than 100,000 compounds for growth inhibition over three days in 60 different human tumor cell lines representing leukemia, melanoma and cancers of the lung, colon, brain, ovary, breast, prostate and kidney. This has led to, for instance, the use of ellipticine in glioma treatment, and in identification of the kinase inhibitor UCN-01 as a lead molecule. This type of project also contributes to tumor biology knowledge, by revealing patterns of cellular responsiveness summarized in "maps" which can be

Figure 2. General model of concentration- and time-dependent anti-proliferauve effects of many drugs on cancer cells. Low doses can induce states of arrest and senescence during which the cells remain viable as seen by, e.g., MTT assays. Over-lapping or higher concentrations may induce expression of various signaling factors such as c-Jun, NFk B, TRAIL and Fas ligand etc, in the entire cell population, or parts thereof. Such factors may over time have either anti-proliferative or protective effects, either intra-cellularly or as part of extracellular signaling loops. Apoptosis which develops over several days time can thus be a secondary effect of such signaling. Higher drug concentrations can lead to acute apoptosis within, e.g., 24 h. The apoptotic "concentration window" before induction of necrosis may vary depending on drug and cell type.

Using a caspase-/apoptosis-specific assay, we have screened the mechanistic subset of the NCI library in order to find drugs showing apoptosis synergy within 24 h in combination with cisplatin, and identified ellipticine as one such drug (Hägg et al, 2004). This assay system was used also to assess the role of p53 in acute apoptosis induction by drugs in the same drug library. Most of the proapoptotic drugs induced apoptosis also in p53-null cells, and this apoptosis was markedly dependent on early permeabilization of the lysosomes (Erdal et al, submitted).

Necrosis is another type of cell death. Since agents with unspecific toxic effects may induce necrosis and hence present a very small therapeutic window, assessment of necrosis is important for evaluating drug candidates. However, like apoptosis, necrosis can be a regulated process and is not necessarily to be avoided. For many drugs in vivo, the balance between apoptosis and necrosis likely depends not only on the signal transduction "wiring" of the tumor cell, but also on oxygenation and energy status. Necrosis can even be the expected outcome: it has been reported that cell death induced by alkylating agents such as nitrogen mustard is mainly necrotic due to the NAD-/ATP-depleting effect of PARP activity induced by DNA damage (Zong et al, 2004). Studies on drug responses in vivo should therefore include necrosis as well as apoptosis. We have developed a method to assess the relative contributions of apoptosis and necrosis in the same sample in vitro and in post-treatment serum samples from cancer patients, and found that induction of necrosis may be a quite common response to many drugs (Kramer et al, 2004; Erdal et al, submitted).

As exemplified above in the section on cisplatin, treatment-induced senescence (which in some aspects differs from replicative senescence) is yet another possible growth inhibitory outcome (Roninson, 2004), but it will not be detected by apoptosis/necrosis assays and may confuse the interpretation of viability assays such as the MTT. Therefore, because they show a net anti-proliferative result, clonal outgrowth assays may be appropriate for assessing total drug effects. However, the difficulties in adapting them to high-throughput systems are a major drawback; other disadvantages include cell culture conditions which may be even more non-physiological than in shorter-term monolayer cultures (e.g., extremely low cell density and the subsequent alterations in gene expression, decay of nutrients during the long incubation; also, the unphysiological global stimulation when media are changed after a period of starvation can per se induce death).

As the NCI screening projects and our own results show, screening systems are useful not only for drug discovery, but also for investigating signal transduction. In this context, clonal outgrowth assays are unsuitable, precisely because they reflect the net influence of many pathways, feedback loops and types of cell death. In view of the roles of microenvironmental factors, cell-based screening systems for specific types of cell death must, however, be improved. A first step is to introduce three-dimensional cell cultures in high-throughput screens; later steps involve development of tissue-like and monitorable reconstructions of in vivo situations. Systems reflecting the differing responses of normal and tumor cells, respectively, are also required. The future screening systems will thus be able to determine the net outcome of multiple drug effects in combination with biological conditions, i.e., they will resolve many questions in very few steps.

To use a drastic metaphor, the anticancer drugs that are promiscuous, but toxic, provide us with a variety of deadly bombs, whereas the single-target drugs constitute carefully aimed guns. While the bombs will efficiently remove several species of deadly prey but cause a great deal of collateral damage, the smaller guns are efficient only at close range, but will therefore leave other species of the forest unharmed. To continue this analogy, future improvements in cancer therapy require knowledge of forest ecology and the habits of the quarry, but also of weapons technology. Because tumor cell signaling complexity must be countered with complex combinations of medicinal weapons, future drugs will range from wide-spectrum multi-target compounds to specially designed molecules targetting tumor-specific, mutated forms of signaling proteins.

We would like to acknowledge the support from Cancerföreningen in Stockholm, and from the Swedish Cancer Society.