Using Magic Targets, not Magic Bullets, to treat cancer

To edit collaborative proposals, please register by clicking 'Join this Space' on the left,

then click the 'Edit This Page' button on the right!!!

To make comments without becoming a member, use the 'discussion' tab.

Posted 30 December 2007 by Noam Y. Harel


Cancer-prone tissues can be prophylactically transduced with a library of lentiviral constructs encoding a death gene under control of unique siRNA-targeting sequences. By sequencing DNA from subsequently arising tumors, matched siRNAs may be used to induce death gene expression. The death gene will allow selective killing of all cells clonally derived from the cell containing the targeted genetic tag. This system provides a potent weapon against clonally derived cancers.


Cancers can and will arise from any cell in the body. In most cases, tumors derive from a clonal cell population. In a few cases, tumors express distinct markers, such as the chromosomal translocation leading to the bcr-abl fusion kinase in chronic myelogenous leukemia (Daley et al., 1990; Nowell and Hungerford, 1960). Such tumor-specific tags, when identifiable, provide a target for so-called ‘magic bullet’ anti-cancer drugs – imatinib, for instance (Druker et al., 2001).

But the great majority of tumors do not express known targets for magic bullets. To generate these targets, is there a way to tag cells with unique genetic codes before they become cancers? What if these unique codes were used to control expression of an inducible death gene that would allow selective killing of all tumor cells clonally derived from cells with that code? This proposal outlines a feasible approach to these questions, integrating several key technologies: RNA interference, bioinformatics, and lentiviral gene therapy.

Inducible expression with RNA interference

Traditional inducible expression constructs rely on drugs such as doxycycline or tamoxifen to mediate increased transcription of the gene of interest (Goverdhana et al., 2005; Szulc et al., 2006). At the desired time, one drug induces gene expression in all cells transduced with the construct. But what if we desire to stimulate expression in only a subset of the transduced cells? This feature requires a mechanism more complex than traditional drug-induced constructs.

The crux of the RIDD (RNAi-Inducible Death Dimer) system to be used in this proposal is the ability to use customizable short interfering RNAs (siRNAs) to induce gene expression only in targeted cells. siRNAs induce the endogenous RNA interference (RNAi) pathway, leading to degradation of mRNAs that harbor complementary sequences to the siRNA (Elbashir et al., 2001). In the RIDD system, degradation of the repressor mRNA will lead to induction of the death dimer. By varying the sequence at the siRNA-targeting site, a library of RIDD constructs may be composed, each selectively inducible by the appropriately matched siRNA. At the desired time (for instance, when a RIDD-transduced cell becomes cancerous), the death gene may be specifically induced in the responsible cell and its progeny, allowing selective killing of tumor cells.

Bioinformatics to optimize a library of unique siRNA-targeting sites

Sequences composing the siRNA-targeting sites that regulate the RIDD system need to satisfy two requirements: 1) mediate effective knockdown of the targeted RNA; 2) avoid similarity with endogenous sequences that could lead to harmful off-target effects. Genomic bioinformatics approaches can drastically reduce the empiric testing that would be required to meet these requirements. Freely available online siRNA optimization tools will help with the first requirement. To meet the second requirement, part of each targeting sequence will consist of an 11 nucleotide “nullomer” that is absent from the human genome (Acquisti et al., 2007; Hampikian and Andersen, 2007).

Lentiviral gene therapy

The Magic Target approach necessitates insertion of a genetic construct into a cell’s genome such that the DNA is passed on to all of that cell’s progeny. Retroviral vectors provide the most efficient method for mediating integration of exogenous DNA into host chromosomes (Thomas et al., 2003). Lentiviruses comprise a class of retroviruses that are able to integrate into the chromosomes of non-dividing cells as well as dividing cells. Even though this proposal plans on tagging cells that may become cancerous (i.e. dividing out of control), the tags will be administered one or more times before cells become cancerous, at times when they may not be dividing. As with other types of retroviral gene therapy vectors, many measures have been taken to ensure that lentiviral vectors have almost no risk of pathogenicity (Thomas et al., 2003).


  • Design and test a prototype RIGG construct (RNAi-Inducible-GFP-G418). Confirm specific and robust induction of GFP expression with appropriately matched siRNAs.
  • Test a library of RIDD constructs for specific siRNA-inducible death dimer induction and killing of one clonal line of cells among a mixed population in culture.
  • Test in a rodent cancer model: Infect cancer-susceptible tissue with a lentiviral RIDD vector library. Once a tumor forms, determine the siRNA target sequence of the responsible clonal cell population. Then administer the matched siRNA systemically to induce death gene expression in all clonally derived tumor and metastatic cells. Determine percentage of clonally derived cancer cells that express the death gene, and percentage of death gene-positive cells killed by the drug.
  • Determine which patients and cancer types are most amenable to prophylactic RIDD vector transduction. Begin clinical studies to determine frequency and dosages of RIDD vector required to effectively tag all potential tumors to arise from susceptible tissues. Optimize dosages and toxicity of systemic siRNA and death dimer-inducing drug.


RIDD construct overview

The RIDD construct (schematic to be posted) is adapted from the KRAB-based repressor construct described by Szulc et al. (Szulc et al., 2006). Two transcription cassettes will be inserted between lentiviral long terminal repeats. One will encode a fusion of the GAL4 DNA-binding domain with the KRAB transcriptional repression domain. Through GAL4 DNA-recognition motifs in the promoter of the repression cassette, GAL4-KRAB will silence death dimer expression.

The death dimer cassette encodes a fusion of caspase-3 to FKBP12. The FKBP12 domain binds to AP1903, a non-toxic analog of the FDA-approved drug FK506 (MacCorkle et al., 1998). When expressed in the absence of AP1903, the FKBP12-caspase-3 fusion has no intracellular activity. Administration of AP1903 results in FKBP12 dimerization, activating the apoptotic function of caspase-3, leading to cell death. Thus, two levels of regulation (and safety) are built into RIDD vectors: The KRAB domain represses death dimer transcription. Once the death dimer is induced by the appropriate siRNA (see below), no toxicity will occur unless AP1903 is administered.

The key regulatory element of the RIDD system consists of a short sequence introduced into the repressor cassette’s 5’ untranslated region. When transcribed, this randomly generated 25nt sequence will serve as a targeting site for a matching exogenously administered siRNA. Knockdown of GAL4-KRAB expression will allow expression of the death dimer cassette. To prevent overlap of siRNA-targeting sequences with endogenous human mRNA sequences, part of each targeting sequence will consist of an 11 nucleotide “nullomer” that is absent from the human genome (Acquisti et al., 2007; Hampikian and Andersen, 2007). The combination of the 11nt nullomer plus a random 14nt sequence should allow generation of a library of over 2.5x10e8 vectors, each containing different sequences at the siRNA-targeting site.

Aim 1: Design and test a prototype RIGG construct (RNAi-Inducible-GFP-G418).

For testing purposes, a modified RIDD construct (RIGG) will be employed in which GFP replaces the death dimer, and the repression cassette includes an IRES followed by the G418 resistance gene. Human peripheral blood mononuclear cells (PBMCs) will be infected with this construct (with one prototype siRNA-targeting site), followed by G418 selection for successfully integrated cells and their progeny.

Before and after adding the matching siRNA (or mismatched controls) to the culture, cells will be monitored for GFP expression by fluorescence-activated cell sorting (FACS). Note that successful RNAi-mediated GFP induction would coincide with knockdown of both GAL4-KRAB and the neomycin resistance gene, so G418 will be removed from the medium at the time of siRNA administration. The percentage of initially G418-resistant cells that become GFP-positive upon siRNA addition will indicate the efficiency of RNAi induction. Any GFP expression before siRNA addition, or after mismatched control siRNA addition, would indicate “leakiness” of the GAL4-KRAB repression cassette. Leakiness could be further reduced by adding multiple GAL4-binding sites to the death dimer (or GFP) promoter region, or by other means. Doses of lentiviral vector and siRNA will be titrated for optimal results.

PBMCs will then be transduced with a library of RIGG lentiviral vectors harboring variable siRNA-targeting sequences. G418 selection will result in a mixed population of RIGG-transduced clones. An siRNA corresponding to one of the RIGG clones will then be administered, followed by FACS. GFP-positive cells will represent either specifically-induced cells containing the targeted RIGG sequence, or non-specifically-induced cells containing other sequences. Conversely, GFP-negative cells will comprise appropriately uninduced off-target cells, or on-target cells that failed to induce. To differentiate these populations, DNA from FAC-sorted cells will be hybridized to microarray chips lined with the library of siRNA-targeting sequences. This will allow quantitative assessment for the presence of non-specific siRNA-targeting sequences within the GFP-positive population. GFP-negative cells will be subjected to similar analysis, to determine the percentage of cells containing the targeted RIGG sequence that were not successfully induced. siRNA dosage and sequence characteristics will be titrated to optimize the specificity and efficacy of clonal induction.

Aim 2: Test a library of RIDD constructs for specific siRNA-inducible death dimer induction and killing of one clonal line of cells among a mixed population in culture.

Once parameters have been optimized using RIGG vectors, they will be applied to RIDD vector testing. Assessment methods will be adapted for RIDD vectors’ lack of GFP and G418 sequences. Rather than FAC-sorting GFP-positive cells, overall PBMC transfection and integration efficiency will be determined by quantitative PCR using RIDD-specific primers. Serial Analysis of Gene Expression (SAGE) will quantify the distribution of unique siRNA-targeting sites among expressed mRNAs (Velculescu et al., 1995). After administering experimental or control siRNAs, SAGE will again be used to assess the knockdown efficiency of both specific and off-target RNAs. Immunoblotting for death dimer expression will confirm effective induction, though it will not differentiate between specific and non-specific induction.

Finally, efficiency and selectivity of cell death mediated by AP1903 will be tested. Titrated doses of AP1903 will be administered to parallel transduced cultures at different times following siRNA administration, followed by FACS analysis to sort apoptotic from unaffected cells. DNA microarray analyses on sorted cells will determine the percentage of cells containing targeted and non-targeted RIDD clones that have been killed.

Several important controls will be tested during this step. As negative controls (aside from testing various mismatched siRNAs), cells will be either mock-transduced, or transduced with RIGG (lacking the death dimer) rather than RIDD vectors. As a positive control for AP1903-mediated killing, AP1903 will also be administered to cultures constitutively expressing the death dimer. Using this combination of experimental and control conditions, multiple parameters will be optimized for maximal efficiency and selectivity with minimal toxicity: RIDD vector dosage; siRNA dosage and sequence characteristics; timing of AP1903 administration after siRNA induction; and dosage of AP1903.

Aim 3: Test in a rodent cancer model.

For in vivo RIDD testing, multiple chemically- or genetically-inducible mouse cancer models may be used (Boutwell et al., 1982; Hutchinson and Muller, 2000; Kemp, 2005). Cancer-susceptible tissue (for example, skin or breast) will be infected with a lentiviral RIDD vector library (or vehicle or RIGG control) prior to tumor formation. Once tumors form, cells will be obtained via needle biopsy to determine the siRNA-targeting sequence(s) of the responsible clonal cell population(s). The matched siRNA(s) (or mismatched controls) will be systemically administered to induce death gene expression in all clonally derived tumor and metastatic cells. Systemic siRNA administration using lipid-modified siRNAs has been successfully applied in both mice and non-human primates (Soutschek et al., 2004; Zimmermann et al., 2006).

Appropriate timing and dosages for RIDD vector, siRNA and AP1903 will be titrated to optimize several outcomes: the percentage of clonally derived cancer cells that show death gene induction (via SAGE and caspase-3 immunohistochemistry), the percentage of death gene-positive cells killed by AP1903, the number of off-target cells killed by AP1903, and the effect on tumor shrinkage and animal survival.

Aim 4: Determine which patients and cancer types are most amenable to prophylactic RIDD vector transduction.

After testing in mouse cancer models and establishing safety in normal human volunteers, the next steps would be to determine which patients and cancer types are most amenable to prophylactic RIDD transduction. The optimal patients for initial study would likely be those with hereditary predisposition to clonally-derived cancer of a tissue that is relatively accessible to the lentiviral, siRNA, and AP1903 portions of the Magic Target regimen. For instance, patients with BRCA1 mutations predisposing to breast or ovarian cancer; patients with patched mutations predisposing to basal cell carcinoma; or patients with ATM mutations predisposing to leukemia and lymphoma (OMIM 113705, 109400, and 607585, respectively).

Selected patients with one of the above cancer predispositions would receive prophylactic doses of RIDD lentiviral libraries (or placebo) in varying doses and schedules. Any tumors that arise from RIDD-transduced tissue would be needle-biopsied to determine the siRNA-targeting sequence(s) of the responsible clonal cell population(s). Then, matched siRNA(s) will be systemically administered as in Aim 3. Outcomes will be closely monitored as in Aim 3, especially patient survival, tumor and metastasis shrinkage, and any adverse events. Furthermore, the frequency and dosages of RIDD vector required to safely and effectively tag all potential tumors to arise from susceptible tissues will be determined.


Though the Magic Target approach is not expected to be applicable to all cancers, it is expected to provide a potent weapon against multiple types of clonal cancers, if the RIDD vector is administered prophylactically to tumor-free patients. Tumors (and their metastatic progeny) transduced with RIDD vectors will harbor a ‘death dimer’ gene that is inducible by systemically-administered siRNA. The death dimer gene in itself will be inert unless the dimer-inducing drug, AP1903 is co-administered. In effect, AP1903 will act as a chemotherapeutic drug, but with much higher selectivity and thus lower toxicity than most chemotherapeutics currently in use. The inert nature of the death dimer in the absence of AP1903 provides an extra layer of safety in case there is a low level of ‘leaky’ baseline death dimer expression.

Some potential questions, hurdles and possible solutions are listed below:
What is the best location for the siRNA-targeting site? The 5’ UTR? 3’UTR? Coding region?
Any of these sites should be feasible. Another option would be to design siRNA-targeting sites to mediate transcriptional silencing (rather than RNA degradation). This mechanism of siRNA function was recently described by Han et al. (Han et al., 2007).
Inducible-expression systems are notorious for ‘leaky’ expression.
True. Attempts to limit leakiness will include optimizing the number and location of GAL4 DNA recognition motifs in the RIDD construct, and experimenting with different types and combinations of repressor domains. However, even if some leaky death dimer expression does occur, the death dimer in itself is non-toxic unless the dimer-inducing AP1903 is administered. It is hoped (and expected) that the levels of induced death dimer in targeted cells is several orders of magnitude higher than levels in non-targeted cells, allowing the use of a small enough dose of AP1903 that would not affect cells with low-level leaky expression.
Will the RIDD construct remain stable and targetable despite the genetic instability of most tumor cells and their metastatic progeny?
This is a potential weakness. However, RIDD test-doses will range to multiplicities of infection greater than one – if a tumor stem cell harbors multiple RIDD constructs, then it is expected that (at least) one of them will provide a stable target for inducing the death dimer when desired.
Will knocking down most, but not all of the KRAB repressor cassette allow sufficient expression of the death dimer for effective cell killing?
This is a question that will be answered during the early stages of Aim 1. The distance and number of GAL4 DNA recognition motifs from the death dimer cassette, the type of repressor domain used, the number of siRNA-targeting sites used, and the siRNA characteristics used will all be varied empirically to optimize siRNA-targeted induction.
The number of siRNA-targeting sequences that mediate effective knockdown without overlapping with endogenous mRNAs may be smaller than the number of cells that would need to be prophylactically transduced with RIDD vectors. Either not all cancer-prone cells can be transduced, or multiple cells will receive RIDD vectors with identical siRNA-targeting motifs.
Given the use of an 11-nucleotide ‘nullomer’ (Acquisti et al., 2007; Hampikian and Andersen, 2007) as the siRNA-targeting backbone, there are 4e14 to 4e16 unique siRNA-targeting motifs available (depending on whether 25nt or 27nt sequences are used). Of course, not all of these may serve as efficient siRNAs, thus reducing that number. However, it is expected that even if there are fewer unique RIDD vectors than cells to transduce, the effects will be negligible. Suppose 10 different cells receive identical RIDD vectors. If one becomes cancerous and divides out of control whereas the other 9 maintain regulated division, then even if the progeny of all 10 originally-transduced cells are killed, the benefit of killing the tumorous progeny will far outweigh the cost of killing what would be a tiny fraction of the non-cancerous cells in the targeted tissue.
Are there other potential uses for RIDD technology?
One obvious application would be for developmentally tagging cells for fate-mapping, at almost any stage, in almost any tissue. For fate-mapping, one could use a RIGG library, inducing GFP (or any other easily visualized gene product) in all cells clonally derived from a cell with a selected siRNA-targeting sequence. Or, one could combine fate-mapping with selective ablation of clonally derived cells by using a RIDD library, and tracking cell fate with in situ hybridization. Essentially, the death dimer cassette can be replaced by any gene of interest, allowing induction of that gene in all clonally derived cells in a mixed population.


This Magic Target proposal is a more detailed version of a proposal submitted to The Gotham Prize Foundation (Gotham Prize) on 12/27/2007 to obtain eligibility for a one million dollar award.
The Gotham Prize Foundation is the best example SCIEnCE has found so far of an organization effectively promoting the OPEN sharing of ideas and collaboration, without going against natural individual human motivations. ALL granting agencies/institutions/foundations should adopt this approach!


  • Acquisti, C., Poste, G., Curtiss, D., and Kumar, S. (2007). Nullomers: really a matter of natural selection? PLoS ONE 2, e1022.
  • Boutwell, R.K., Verma, A.K., Ashendel, C.L., and Astrup, E. (1982). Mouse skin: a useful model system for studying the mechanism of chemical carcinogenesis. Carcinogenesis; a comprehensive survey 7, 1-12.
  • Daley, G.Q., Van Etten, R.A., and Baltimore, D. (1990). Induction of chronic myelogenous leukemia in mice by the P210bcr/abl gene of the Philadelphia chromosome. Science 247, 824-830.
  • Druker, B.J., Talpaz, M., Resta, D.J., Peng, B., Buchdunger, E., Ford, J.M., Lydon, N.B., Kantarjian, H., Capdeville, R., Ohno-Jones, S., et al. (2001). Efficacy and safety of a specific inhibitor of the BCR-ABL tyrosine kinase in chronic myeloid leukemia. N Engl J Med 344, 1031-1037.
  • Elbashir, S.M., Harborth, J., Lendeckel, W., Yalcin, A., Weber, K., and Tuschl, T. (2001). Duplexes of 21-nucleotide RNAs mediate RNA interference in cultured mammalian cells. Nature 411, 494-498.
  • Goverdhana, S., Puntel, M., Xiong, W., Zirger, J.M., Barcia, C., Curtin, J.F., Soffer, E.B., Mondkar, S., King, G.D., Hu, J., et al. (2005). Regulatable gene expression systems for gene therapy applications: progress and future challenges. Mol Ther 12, 189-211.
  • Hampikian, G., and Andersen, T. (2007). Absent sequences: nullomers and primes. Pacific Symposium on Biocomputing, 355-366.
  • Han, J., Kim, D., and Morris, K.V. (2007). Promoter-associated RNA is required for RNA-directed transcriptional gene silencing in human cells. Proc Natl Acad Sci U S A 104, 12422-12427.
  • Hutchinson, J.N., and Muller, W.J. (2000). Transgenic mouse models of human breast cancer. Oncogene 19, 6130-6137.
  • Kemp, C.J. (2005). Multistep skin cancer in mice as a model to study the evolution of cancer cells. Seminars in cancer biology 15, 460-473.
  • MacCorkle, R.A., Freeman, K.W., and Spencer, D.M. (1998). Synthetic activation of caspases: artificial death switches. Proc Natl Acad Sci U S A 95, 3655-3660.
  • Nowell, P., and Hungerford, D. (1960). A minute chromosome in human chronic granulocytic leukemia. Science 132, 1497.
  • Soutschek, J., Akinc, A., Bramlage, B., Charisse, K., Constien, R., Donoghue, M., Elbashir, S., Geick, A., Hadwiger, P., Harborth, J., et al. (2004). Therapeutic silencing of an endogenous gene by systemic administration of modified siRNAs. Nature 432, 173-178.
  • Szulc, J., Wiznerowicz, M., Sauvain, M.O., Trono, D., and Aebischer, P. (2006). A versatile tool for conditional gene expression and knockdown. Nat Methods 3, 109-116.
  • Thomas, C.E., Ehrhardt, A., and Kay, M.A. (2003). Progress and problems with the use of viral vectors for gene therapy. Nature reviews 4, 346-358.
  • Velculescu, V.E., Zhang, L., Vogelstein, B., and Kinzler, K.W. (1995). Serial analysis of gene expression. Science 270, 484-487.
  • Zimmermann, T.S., Lee, A.C., Akinc, A., Bramlage, B., Bumcrot, D., Fedoruk, M.N., Harborth, J., Heyes, J.A., Jeffs, L.B., John, M., et al. (2006). RNAi-mediated gene silencing in non-human primates. Nature 441, 111-114.