Elsevier

Biochemical Pharmacology

Volume 69, Issue 7, 1 April 2005, Pages 1009-1039
Biochemical Pharmacology

Anticancer metal compounds in NCI's tumor-screening database: putative mode of action

https://doi.org/10.1016/j.bcp.2005.01.001Get rights and content

Abstract

Clustering analysis of tumor cell cytotoxicity profiles for the National Cancer Institute (NCI)'s open compound repository has been used to catalog over 1100 metal or metalloid containing compounds with potential anticancer activity. The molecular features and corresponding reactivity of these compounds have been analyzed in terms of properties of their metals, their associated organic components (ligands) and their capacity to inhibit tumor cell growth. Cytotoxic responses are influenced by both the identity of the metal and the properties of its coordination ligand, with clear associations between structural similarities and cytotoxicity. Assignments of mechanisms of action (MOAs) for these compounds could be segregated into four broad response classes according to preference for binding to biological sulfhydryl groups, chelation, generation of reactive oxygen species (ROS), and production of lipophilic ions. Correlations between specific cytotoxic responses and differential gene expression profiles within the NCI's tumor cell panel serve as a validation for candidate biological targets and putative MOA classes. In addition, specific sensitivity toward subsets of metal containing agents has been found for certain tumor cell panels. Taken together, our results expand the knowledge base available for evaluating, designing and developing new metal-based anticancer drugs that may provide the basis for target-specific therapeutics.

Introduction

The importance of metal compounds in medicine dates back to the 16th century [1] with reports on the therapeutic use of metals or metal containing compounds in the treatment of cancer. Now the list of therapeutically prescribed metal containing compounds includes platinum (anticancer), silver (antimicrobial), gold (antiarthritic), bismuth (antiulcer), antimony (antiprotozoal), vanadium (antidiabetic) and iron (antimalarial). Metal ions are electron deficient, whereas most biological molecules (proteins and DNA) are electron rich; consequently, there is a general tendency for metal ions to bind to and interact with many important biological molecules. Metal ions also have a high affinity for many small molecules, e.g. O2, that are crucial to life. These considerations alone have fueled much of the past and current interest in developing novel means to use metals or metal containing agents to modulate biological systems.

Biologically essential life processes requiring metals usually involve enzymatic, structural or reactive roles. Catalytic activities for an estimated 12% of all enzymes can be ascribed to metal centers. Metals act to bridge substrate to enzyme such that electrons are withdrawn from the metal and the excess local positive charge lowers the free energy of enzyme activation. Alternatively, structural metals function much like disulfide bonds to participate in the correct tertiary folding of proteins. Transition metals, on the other hand, are known to generate potentially harmful, but in some cases useful, reactive intermediates. General classes of metals can be loosely assigned according to their enzymatic, structural or reactive roles. Iron (Fe) and manganese (Mn) act as enzyme cofactors and catalyze redox reactions. Zinc (Zn) and calcium (Ca) provide, respectively, structural integrity, such as in zinc finger proteins, or flexibility, as in calmodulin, for many proteins. Fe plays a critical role in oxygen transportation and electron shuttling. Sodium (Na) and potassium (K) function as charge carriers. Magnesium (Mg) and manganese (Mn) function in hydrolysis and group-transfer. These diverse roles for metals provide many opportunities for biological modulation.

Metal binding substances1 have furnished many useful drugs. These compounds mainly exploit metals’ role in enzymatic activation or molecular structure. Organo-metallic compounds have achieved importance as enzyme inhibitors, partly due to their capacity to alter enzyme function by, for example, binding biological targets more strongly than metal free organic substrates [2]. Based on this feature, inhibition of metalloenzymes can be achieved by metal coordination to exogenous ligands or by chelation substitution or exchange. Precedence for chelation affecting biological activity existed over 50 years ago with the demonstration that the biological properties of oxine were linked to its ability to bind metal ions [3]. Metal complexes can also inhibit non-metalloenzymes by coordinating to their active site residues and sterically blocking substrate interaction, or coordinating to structurally important residues near the active site. A secondary effect of transition metals is to catalyze the generation of reactive oxygen species (ROS), the presence of which is believed to play important but poorly understood roles that can modulate drug-induced cytotoxic responses and affect cancer pathogenesis [1], [4]. Metal-coordination is also one of the most efficient strategies in the design of repository, slow-release or long-acting drugs. Often overlooked, the property of metals to release therapeutic ligands has found utility, especially when ligands are not strongly bound, as exchange vehicles for alternative ligands within a targeted biological path [5]. While many metals are essential for all forms of life, their levels in normal homeostasis or therapeutic intervention must be strictly regulated because most are toxic in excess. As with all drugs, the use of metals in drug development will depend largely on understanding their MOA and selectively controlling their toxicity.

Historically, the antitumor potential of metal-containing agents, from early transition metals to main group elements, has been extensively evaluated. The specific case of organo-platinum compounds, such as cisplatin and its analogs [6], has generated high interest in discovering alternative platinum-containing complexes. To date, considerable evidence points to platinum's therapeutic efficacy being severely attenuated depending, for example, on its molecular configuration, as some trans-configured ligands exhibit greater cytotoxicity when compared to their cis counterparts. Of particular interest is that trans-platinum complexes show cytotoxicities in cisplatin resistant tumor cell lines [7].

In addition to selectively controlling their toxicity, metal-containing compounds must also have appropriate pharmacological properties. Many interesting antitumor active compounds have failed to reach clinical use due to poor physio-chemical properties, including insufficient water solubility, hydrolytic instability and the tendency to readily decompose when exposed to solvents, humidity, light or air. Difficulties in controlling selective toxicity or devising appropriate pharmacological properties have contributed to a general reluctance in the development of metal-based drugs. Consequently, despite the fact that metal complexes are often cytotoxic in vitro at a significantly lower dose than organic drugs, metal compounds are often not investigated further in tumor model systems. Another critical factor when determining therapeutic potential is the metal's oxidation state. Increased oxidation potential for most transition metals occurs with decreased ionic radius associated with higher valence. Toxicity, however, does not appear to relate simply to a higher or lower oxidation state, since chromium(VI) compounds are highly toxic when compared to chromium(III) compounds, whereas arsenite (As(III)) is more toxic than arsenate (As(V)). This brief introduction underscores the much-storied history of metals in clinical applications. The primary goal of the following analysis is to provide a comprehensive study of tumor cell cytotoxicity in light of many of the considerations discussed above as important for the discovery of novel metal-containing antitumor agents.

Our analysis focuses on the tumor cell line cytotoxicity data generated at the NCI for in vitro anticancer drug screening. These tumor cell lines reflect diverse cell lineages (lung, renal, colorectal, ovarian, breast, prostate, central nervous system, melanoma and hematological malignancies, referred to collectively as the NCI60). Since its inception in 1990, cytotoxicity measures for over 40,000 compounds have been obtained that are publicly available. Using this database, our group has developed a suite of computational tools (http://spheroid.ncifcrf.gov), incorporating structural chemotypes as well as biological data into self-organizing maps (SOMs) [8]. SOM is a neural network-based, unsupervised learning algorithm [9] that has been widely used to organize high-dimensional data into lower dimensional space. This lower dimensional space can be organized into a two-dimensional, hexagonal lattice in such a way that the initial data points lying near each other are mapped to nearby locations on the SOM. This level of organization can be used as a tool to visualize complex dependencies between and within different data sets. Here we apply the SOM method specifically to the tumor cell growth inhibition (GI50) data available at the NCI. GI50 growth patterns have been found to be an information rich source for establishing a compound's MOA [8], [10]. For each compound the SOM procedure uses the GI50 data for all cell lines to cluster similar response profiles into nodes. Each node is developed as the best characterization of the data vectors belonging to that node. The distinguishing feature of SOM with respect to hierarchical clustering methods is that, during the generation of the SOM, the developing cluster is established in the context of all clusters, rendering the division of the data into a multi-dimensional dendrogram where neighboring clusters can share common features. The visualization of the resultant clustering can be rendered as a two-dimensional map to emphasize connectedness between clustered nodes. A further characterization of the two-dimensional SOM is to establish larger regions of similarity between groups of nodes. In this case the information contained in the node representation is utilized to cluster similar nodes into a hierarchy of clades, the region is thus defined as a collection of similar clades, which represents the major response categories of the GI50 data. SOM clustering of the NCI60 data previously segregated compounds into six major response categories: mitosis (M), membrane function (N), nucleic acid metabolism (S), metabolic stress and cell survival (Q), and two unexplored regions P and R [8]. Each of these regions is further divided into a total of 51 clades (sub-regions; M1–M2, N1–N12, P1–P14, Q1–Q7, R1–R9, S1–S7). The compounds analyzed by Rabow et al. [8] included over 1100 compounds containing transition metals, main group metals, or metalloids (a total of 55 different elements). SOM clustering, based solely on differential cytotoxicity, found metal compounds exhibiting diverse cytotoxic response profiles (Fig. 1). Special cases are found where clusters with similar cytotoxic response profiles are comprised of structurally similar compounds.

Our analysis inspects metal containing compounds that have been clustered according to similarities in their GI50 response profiles. Compounds in these clusters are subsequently examined according to their reactivity and structural similarities (including properties of the metal alone, the ligands coordinated to the metal, and the coordination mode). The differential cytotoxic response profiles for selected compound sets are in turn correlated with the gene expression profiles generated from these same tumor cell lines. Based on these results the potential targets and possible MOAs of these metal compounds are proposed and discussed. Our results reveal a clear association between certain types of metals and cytotoxicity, as well as a perspective on the importance of the organic component in defining cytotoxicity, and the underlying mechanistic origins of the diverse cytotoxic responses associated with metal compounds. This perspective can also be used to help assign MOAs for non-metal-containing agents that show similar cytotoxic response profiles to those of metal-containing compounds. Taken together, our results expand the knowledge base needed for evaluating, designing and developing new metal-based anticancer drugs.

The subsequent analysis and results will be presented in two major sections. The first section surveys the chemical structural features of the metal compounds according to their appearance in the previously defined four SOM subsections (Q, S, N/P and M). Similarities in chemical features and consequent similarities in reactivity within each SOM region, together with literature support, form the basis for proposing putative MOAs. Metal type seems to play a predominant role in defining the cytotoxic response for some SOM regions (e.g., Q), whereas ligand properties appear to be the determining factor in others (e.g., S1 and S2). Cases also exist where cytotoxicity is dictated by the combined effect of the metal–ligand entity (e.g., N/P and M). These results will reveal a wide variety of metal complex chemotypes that affect a diverse array of molecular targets and biological pathways. The second section analyzes the genomic features uniquely associated with each SOM region, which implicate classes of genes representing various biological pathways that may be affected by drug insults. The examination of region specific gene-drug relationship in this section serves as additional support for the putative MOAs proposed in the first section and may also reveal novel targets and MOAs for metal-based drug molecules. The present study represents a comprehensive survey of metal compounds possessing anticancer activity viewed in terms of their chemical structural features, putative MOAs and correlated genomic features.

Section snippets

Compound structural features and putative MOA: heavy metal compounds preferably bind to sulfhydryl (SH) groups (Region Q)

The greatest number of metal containing compounds is found in the lower left corner of the SOM, designated as the Q-region (see Fig. 2, Panel A). The most predominant feature of these metal complexes is their high affinity for sulfur-containing ligands. Approximately 90% of these metal complexes consist of second- or third-row transition metals (Rh, Pd, Ag, Ir, Pt, Au, Hg, etc.), or heavy post-transition metals (Sn, Pb, Sb, Bi, etc.). Heavy metals are generally known as “soft acids”; they tend

Compound structural features and putative MOA: chelators and metal complexes of chelators (sub-regions S1 and S2)

The metal compounds in the S-region fall into several major clusters. The compounds in sub-regions S3 and S5 are mainly clusters of platinum complexes. As mentioned earlier in the Q-region discussion, these compounds are either cis-configured or dinuclear platinum complexes with their primary MOA being the formation of intra- or inter-strand cross-links between DNA base pairs. S1 and S2 are the other two S sub-regions enriched with metal compounds. The key feature of these compounds is that

Compound structural features and putative MOA: reactive oxygen species and oxidative stress (regions N and P)

The metal compounds clustered in regions N and P are the most diverse throughout the SOM in both structural types and reactivity. The cytotoxicity response profiles of the compounds in these two regions are the most dissimilar when compared to other SOM regions. Regions N and P are adjacent to each other on the SOM and many metal complex structural motifs are shared between these two regions (see Table 3; Fig. 2, Panel C). Nevertheless, metal compounds having the same structural motif still

Compound structural features and putative MOA: lipophilic ions (region M)

The organometallic complexes in the M region cluster almost exclusively in the lower half of sub-region M2. A distinct feature of these complexes is their capability to produce ionic/charged species. The most representative class of complexes is a series of cyclopentadienylrhenium complexes that are either ionic compounds composed of the cationic [(C5H5)(NO)(Ph3P)ReL]+ moiety and one of the negatively charged counter ions, BF4 or PF6 (Table 4, Entry 1); or carboxylates of the type [(C5H5

Correlations between cytotoxicity and gene expression

Approximately 10,000 constitutive gene expression measurements have been made available for each of the NCI60 tumor panel cells. Gene expression data complements the small molecule cytotoxicity data used previously to assign structure–activity relations, MOAs, and role of the cellular environment in cytotoxicity. The goal of a global drug–gene correlation analysis is to identify SOM regions having significant correlations (p < 0.05) with gene expression profiles (see Supplementary Information for

Role of metal versus coordination ligand in dictating cytotoxicity response

SOM clustering analysis of the differential cytotoxic response profiles of over 1100 metal/metalloid compounds segregates them into four broad classes, each consisting of specific structural motifs and possessing a variety of MOAs: heavy metal complexes that have high affinity for sulfhydryl groups (Q), metal chelates (S1/S2), metal/metalloid compounds that can induce oxidative stress (N/P), and metal/metalloid compounds that can produce lipophilic ions (M2). In addition, the global

Known drugs containing metal ions

A wide variety of metal-based drugs have been approved by the FDA for clinical use or evaluated in clinical trials. The Known Drugs database compiled by Leadscope® contains ∼13,000 compounds reported in the period of January 1982 to December 2002, 1080 of which contain metal ions or metalloids. Forty-four different metal/metalloid elements, ranging from alkali (Li, Na, K) and alkaline earth metals (Mg, Ca) to various transition metals (Ag, Au, Cd, Co, Cr, Cu, Fe, Hg, Ir, Mn, Mo, Nb, Ni, Pd, Pt,

Cellular sensitivity

The metal/metalloid compounds in the NCI60 cancer cell screen database display unique cytotoxicity response profiles compared to non-metal compounds. The intra-tumor panel average cytotoxicity profiles for GI50 measurements of all screened compounds, all metal compounds, and screened compounds that also appear as antineoplastic or immunosuppressive agents in the Leadscope® Known Drugs database are shown in Panel A of Fig. 6. These histograms are ordered from left to right to describe tumor

Summary

SOM clustering of the metal/metalloid compounds in the NCI60 cancer cell screen segregates these agents into four broad proposed MOA classes. Global drug–gene correlation analysis provides additional support for these classes. The cytotoxic activity of a metal complex is found to be dictated by both the identity of the metal and the organic components (ligands) that are bound to the metal and, in many cases, one could be the dominating factor over the other. Target specificity can be achieved

Acknowledgements

The authors would like to thank the members of the STB staff and the Laboratory of Computational Technologies, especially Drs. John Cardellina, Narmada Thanki and Xiang-Jun Lu, for valuable contributions during the preparation of this manuscript. We would also like to give special thanks to Dr. Alfred A. Rabow, whose encouragements, helpful comments and discussions during the initial stages of this project are sincerely appreciated. This project has been funded in whole or in part with Federal

References (156)

  • A.D. Smith et al.

    Aurothioglucose inhibits murine thioredoxin reductase activity in vivo

    J Nutr

    (1999)
  • N. Katsaros et al.

    Rhodium and its compounds as potential agents in cancer treatment

    Crit Rev Oncol Hematol

    (2002)
  • E.S.J. Arner et al.

    Analysis of the inhibition of mammalian thioredoxin, thioredoxin reductase, and glutaredoxin by cis-diamminedichloroplatinum (II) and its major metabolite, the glutathione–platinum complex

    Free Radic Biol Med

    (2001)
  • K. Munter et al.

    Inhibition of cellular activities by triethyllead. Role of glutathione and accumulation of triethyllead in vitro

    Biochem Pharmacol

    (1989)
  • J. Adams

    Potential for proteasome inhibition in the treatment of cancer

    Drug Discov Today

    (2003)
  • T. Audesirk et al.

    Low levels of inorganic lead noncompetitively inhibit calpain

    Toxicology

    (1998)
  • K.C. Agrawal et al.

    The chemistry and biological activity of alpha-(N)-heterocyclic carboxaldehyde thiosemicarbazones

    Prog Med Chem

    (1978)
  • F.J. Giles et al.

    Phase I and pharmacodynamic study of Triapine, a novel ribonucleotide reductase inhibitor, in patients with advanced leukemia

    Leuk Res

    (2003)
  • G. Nocentini et al.

    Antitumor activity of 2,2′-bipyridyl-6-carbothioamide: a ribonucleotide reductase inhibitor

    Gen Pharmacol

    (1997)
  • J.G. Cory et al.

    Structure–function relationships for a new series of pyridine-2-carboxaldehyde thiosemicarbazones on ribonucleotide reductase activity and tumor cell growth in culture and in vivo

    Adv Enzyme Regul

    (1995)
  • D.B. Lovejoy et al.

    Novel “hybrid” iron chelators derived from aroylhydrazones and thiosemicarbazones demonstrate selective antiproliferative activity against tumor cells

    Blood

    (2002)
  • E.W. Ainscough et al.

    Cytotoxicity of salicylaldehyde benzoylhydrazone analogs and their transition metal complexes: quantitative structure–activity relationships

    J Inorg Biochem

    (1999)
  • L.L. Koh et al.

    Complexes of salicylaldehyde acylhydrazones: cytotoxicity QSAR and crystal structure of the sterically hindered t-butyl dimer

    J Inorg Biochem

    (1998)
  • B. Bottari et al.

    Nickel(II) 2,6-diacetylpyridine bis(isonicotinoylhydrazonate) and bis(benzoylhydrazonate) complexes: structure and antimycobacterial evaluation. Part XI

    Bioorg Med Chem

    (2001)
  • Z.D. Liu et al.

    Design of iron chelators with therapeutic application

    Coord Chem Rev

    (2002)
  • C. Gocmen et al.

    Effect of neocuproine, a selective Cu(I) chelator, on nitrergic relaxations in the mouse corpus cavernosum

    Eur J Pharmacol

    (2000)
  • L.X. Zhao et al.

    Synthesis, topoisomerase I inhibition and antitumor cytotoxicity of 2,2′:6′,2″-, 2,2′:6′,3″- and 2,2′:6′,4″-terpyridine derivatives

    Bioorg Med Chem Lett

    (2001)
  • J. Nordenberg et al.

    Anti-proliferative effects and phenotypic alterations induced by 8-hydroxyquinoline in melanoma cell lines

    Eur J Cancer

    (1990)
  • M.E. Vol’pin et al.

    Transition metal complexes as catalysts in biochemical systems. Interaction with electron transfer processes

    Inorg Chim Acta

    (1981)
  • K.R. Huber et al.

    Superoxide dismutase-like activities of copper(II) complexes tested in serum

    Biochim Biophys Acta

    (1987)
  • M. Patel et al.

    Metalloporphyrin class of therapeutic catalytic antioxidants

    Trends Pharmacol Sci

    (1999)
  • N. Kasugai et al.

    Selective cell death by water-soluble Fe-porphyrins with superoxide dismutase (SOD) activity

    J Inorg Biochem

    (2002)
  • M.D. Maines et al.

    Expression of heme oxygenase-1 (HSP32) in human prostate: normal, hyperplastic, and tumor tissue distribution

    Urology

    (1996)
  • Q. Kong et al.

    A threshold concept for cancer therapy

    Med Hypotheses

    (2000)
  • P. Amstad et al.

    Glutathione peroxidase compensates for the hypersensitivity of Cu, Zn-superoxide dismutase overproducers to oxidant stress

    J Biol Chem

    (1994)
  • G.D. Mao et al.

    Superoxide dismutase (SOD)-catalase conjugates Role of hydrogen peroxide and the Fenton reaction in SOD toxicity

    J Biol Chem

    (1993)
  • B.A. Omar et al.

    The cardioprotective effect of Mn-superoxide dismutase is lost at high doses in the postischemic isolated rabbit heart

    Free Radic Biol Med

    (1990)
  • A.Y. Louie et al.

    Metal complexes as enzyme inhibitors

    Chem Rev

    (1999)
  • A. Albert et al.

    The influence of chemical constitution on antibacterial activity III. A study of 8-hydroxyquinoline (oxine) and related compounds

    Br J Exp Pathol

    (1947)
  • F. Caruso et al.

    Anticancer titanium agents

    Expert Opin Ther Pat

    (2001)
  • E. Wong et al.

    Current status of platinum-based antitumor drugs

    Chem Rev

    (1999)
  • A.A. Rabow et al.

    Mining the National Cancer Institute's tumor-screening database: identification of compounds with similar cellular activities

    J Med Chem

    (2002)
  • T. Kohonen

    Self-organizing maps

    (1995)
  • K.D. Paull et al.

    Display and analysis of patterns of differential activity of drugs against human tumor cell lines: development of mean graph and COMPARE algorithm

    J Natl Cancer Inst

    (1989)
  • R.G. Pearson

    Acids and bases

    Science

    (1966)
  • N. Ballatori

    Mechanisms of metal transport across liver cell plasma membranes

    Drug Metab Rev

    (1991)
  • A. Naganuma et al.

    Cellular glutathione as a determinant of sensitivity to mercuric chloride toxicity. Prevention of toxicity by giving glutathione monoester

    Biochem Pharmacol

    (1990)
  • G. Hennighausen et al.

    Nonprotein sulfhydryl groups and glutathione S-transferases in acute poisoning by di-n-alkyltin dichlorides in mice and rats

    Pharmazie

    (1985)
  • R.A. Henry et al.

    Inhibition of glutathione-S-aryltransferase from rat liver by organogermanium, lead and tin compounds

    Biochem Pharmacol

    (1976)
  • K.R. Siebenlist et al.

    Inactivation of yeast hexokinase B by triethyltin bromide and reactivation by dithiothreitol and glucose

    Biochemistry

    (1983)
  • Cited by (0)

    View full text