Strand breakage by decay of DNA-bound 124I provides a basis for combined PET imaging and Auger endoradiotherapy.

Abstract Purpose DNA ligands labelled with 125I induce cytotoxic DNA double-strand breaks (DSB), suggesting a potential for Auger endoradiotherapy. Since the 60-day half-life of 125I is suboptimal for therapy, we have investigated another Auger-emitter 124I, with shorter half-life (4.18 days), and the additional feature of positron-emission, enabling positron emission tomography (PET) imaging. The purpose of this study was to compare the two radionuclides on the basis of DNA DSB per decay. Materials and methods Using a 124I- (or 125I)-labelled minor groove binding DNA ligand, we investigated DNA breakage using the plasmid DNA assay. Biodistribution of the conjugate of the labelled ligand with transferrin was investigated in nude mice bearing a K562 human lymphoma xenograft. Results The probability of DSB per decay was 0.58 and 0.85 for 124I and 125I, respectively, confirming the therapeutic potential of the former. The crystal structure of the ligand DNA complex shows the iodine atom deep within the minor groove, consistent with the high efficiency of induced damage. Biodistribution studies, including PET imaging, showed distinctive results for the conjugate, compared to the free ligand and transferrin, consistent with receptor-mediated delivery of the ligand. Conclusions Conjugation of 124I-labelled DNA ligands to tumor targeting peptides provides a feasible strategy for Auger endoradiotherapy, with the advantage of monitoring tumor targeting by PET imaging.

Introduction 125 I is the best known member of a family of radionuclides referred to as Auger emitters. The decay of these radionuclides is characterized by the emission of multiple low energy electrons (Charlton & Booz 1981), and the consequent highly focused radiochemical damage at the site of decay, which is particularly evident when the decay is closely associated with DNA (Martin & Haseltine 1981). For 125 I covalently incorporated into DNA as a labelled pyrimidine, the majority of DNA single-stranded breaks (SSB) are clustered within a few base pairs of the decaying atom (Lobachevsky & Martin 2000a). On average, each such decay induces a DNA double-strand break (DSB) (Schmidt & Hotz 1973), and the accumulation of 50-100 such events is sufficient to kill mammalian cells (Burki et al. 1973), prompting consideration of the potential of the Auger effect for cancer therapy (Adelstein et al. 1991(Adelstein et al. , 2003. In this context, the requirement of cell cycle activity for incorporation of labelled DNA precursors has the potential to provide a basis for preferential uptake by tumors (Bloomer & Adelstein 1977), but this has not yet proved useful clinically.
The demonstration that decay of 125 I that is non-covalently associated with DNA can induce DNA DSB and cause consequent cytotoxicity, initially using a labelled intercalator (Martin et al. 1979;Kassis et al. 1989), and later minor groove binding ligands (Adelstein & Kassis 1996;Walicka et al. 1999;Lobachevsky & Martin 2004a;Balagurumoorthy et al. 2006;Lobachevsky et al. 2008;Balagurumoorthy et al. 2012) opened up the possibility of more sophisticated targeting of the Auger effect. In a further example, decay of 125 I incorporated into one of the strands of a DNA triplex induces breaks in the unlabelled strands (Panyutin & Neumann 1996;Sedelnikova et al. 1998), thus prompting strategies to target 125 I-induced breaks to specific DNA sequences. However, delivery of triplex-forming labelled oligodeoxynucleotide to the nucleus has proved challenging.
Receptor-mediated targeting of labelled DNA ligand-protein conjugates is another possible strategy to target the Auger effect to the DNA of tumor cells, supported by demonstration of targeted delivery of DNA-binding cytotoxics (Fenton & Perry 2005) and photosensitizers (Karagiannis et al. 2006) by receptor-mediated endocytosis of the respective protein conjugates. This approach is an extension of the early use of specific nuclear receptors to target 125 I-tamoxifen (Bloomer et al. 1980), or 125 I-triiodothyronine (Sundell-Bergman & Johanson 1982) to the DNA of certain tumor cells.
In considering the use of 125 I DNA ligand-protein conjugates in Auger endoradiotherapy, the 60-day half-life of 125 I is a severe limitation for therapeutic applications. Clinical endoradiotherapy studies using 125 I in large amounts pose considerable radiation protection problems (Welt et al. 1996). These considerations have generated an interest in 123 I (Makrigiorgos et al. 1992), an Auger emitter with a much shorter physical half-life (13.2 hours), and which is also better suited to in vivo imaging of biodistribution. Although 123 I is a 'weaker' comparing to 125 I Auger emitter, the efficient induction of DSB by DNA associated decay of 123 I has been reported in several studies (Lobachevsky & Martin 2005;Balagurumoorthy et al. 2008), and incorporation of 5-[ 123 I]Iodo-2 0 -deoxyuridine into DNA resulted in radiotoxicity (Makrigiorgos et al. 1989(Makrigiorgos et al. , 1992. The short half-life of 123 I obviates radiation protection problems, however it is actually a little too short, given the experience of many radioimmunotherapy studies which show that optimal tumor/blood ratios are achieved only 1-2 days after administration of radioimmunoconjugates (Goldenberg 2002). Other Auger electron emitting radionuclides, such as for example 99m Tc (6.0 h half-life) and 111 In (2.8 d half-life), that are widely used for nuclear medicine imaging, have also been evaluated for the decay-induced DNA breakage (Sahu et al. 1995;Karamychev et al. 2000;Haefliger et al. 2005;Kotzerke et al. 2014) and the use in Auger endoradiotheraphy (Tavares & Tavares 2010;Cornelissen et al. 2012). They however are less efficient in inducing DNA DSB, especially considering that chelating agents incorporating these metal radionuclides have to be conjugated to DNA ligand which imposes restrictions on positioning of the decay close to DNA. In this context, another halogen Auger emitter 124 I (4.18 d half-life) incorporated into a DNA minor groove binder, represent a good opportunity. Moreover, this radionuclide is also a positron emitter, so in the context of Auger endoradiotherapy, it has the important complementary potential for assessment of tumor targeting by positron emission tomography (PET) imaging. However, this promise relies on the assumption that 124 I, like the prototype Auger emitter 125 I, induces DNA double-stranded breaks upon decay in the vicinity of the DNA molecule. This assumption may not be valid since 124 I, like 123 I, is known to have a 'weaker' Auger decay compared to 125 I (Pomplun et al. 1996;Iimura et al. 1997). We report here for the first time, the results of a direct comparison of 124 I and 125 I, both on the basis of Monte Carlo simulation of the decay events, and plasmid DNA breakage experiments. For the latter, we used a 124 Ilabelled ligand, para-iodoHoechst, that positions the iodine atom in the DNA minor groove, as verified in the crystal structure of the ligand-DNA complex. Our results show that 124 I, although somewhat less efficacious than 125 I, nevertheless induces DNA double-stranded breaks with a robust efficiency. Finally, we took advantage of the availability of the stannylated precursor of the transferrin conjugate of the meta-isomer of iodoHoechst, prepared for another project (Karagiannis et al. 2006), to demonstrate PET imaging of receptor-mediated delivery of the 124 I-labelled DNA ligand.

Organic synthesis and preparation of para-[ 124 I]/[ 125 I]-iodoHoechst
The synthesis of para-iodoHoechst (1 in Figure 1) is described in detail in the Supplementary Information (available online). Briefly, the linear synthesis approach, as described for Figure 1. Crystal structure of the bibenzimidazole para-iodoHoechst molecule (1) bound in the minor groove of the DNA dodecamer d(CGCAAATTTGCG)2. Carbon atoms are coloured black, nitrogen atoms are blue, and the iodine atom is purple. The ligand spans the 5 0 -ATTTGC site, with the iodine atom positioned centrally in the minor groove. The diagram also depicts the highly focused radiochemical damage associated with the decay of 124 I and other Auger-emitting isotopes, and the positron emission which would provide the basis for the use of PET imaging to monitor tumor-targeting of such 124 I-labelled DNA ligands. Receptor-mediated targeting of protein-DNA ligand conjugates provides a potential strategy for tumor-targeting. methylproamine , was adopted. Thus piodobenzaldehyde was reacted with a substituted orthophenylenediamine (generated in situ from the nitro-aniline) to form the second benzimidazole ring. A similar approach was used to produce the substrate required to prepare the 124 I-and 125 I-ligands, using p-tri-methylstannylbenzaldehyde. The synthesis of the ligand derivatives required for preparation of the amide-linked ligand-transferrin conjugates, with a succinyl moiety introduced onto the aliphatic nitrogen of the piperazine ring, is also described in detail in the Supplementary Information.

Conjugation of iodoHoechst to transferrin
To prepare conjugates for investigation of receptor-mediated targeting, a carboxylic acid analogue of meta-tri-methylstannyl Hoechst was synthesized (1, Figure S4) as described in the Supplementary Information and used as a precursor for conjugation. At the first stage the precursor was iodinated with 125 I or 124 I by iododestannylation, purified on HPLC and dried in a vacuum concentrator as described in the previous section for para-iodoHoechst. The pellet of the radiolabelled intermediate  Figure S4). Then the solutions of N-hydroxysuccinimide (Sigma, 500 mM in DMSO, 3-4 ll) and N, N 0 -dicyclohexyl carbodiimide (Sigma, 290 mM in DMSO, 5-6 ll) were added, and the reaction mixture was incubated at 60 C for 4-6 h. The reaction resulted in formation of the N-hydroxy-succinimidyl-ester of meta-iodoHoechst (3, Figure S4). Following incubation, an aliquot of the reaction mixture (15-20 ll) was added to a solution of transferrin (Sigma, 15 mg/ml) in 50 mM borate buffer at pH 8.5, and the reaction mixture was incubated for 2 h. The reaction resulted in formation of an amide linkage between aliphatic amino groups on the protein and the N-hydroxy-succinimidyl-ester of meta-iodoHoechst (4, Figure S4). Following incubation, 350 ll of phosphate buffered saline (PBS) containing 0.6 mM FeCl 3 (Sigma) were added to the reaction mixture. The conjugate was separated from the reaction mixture by gel filtration chromatography on a Sephadex G-25 NAP-5 cartridge (Pharmacia Biotech, Piscataway, NJ) equilibrated with PBS.

DNA binding and X-ray crystallography
The binding affinity of para-iodoHoechst for DNA was investigated using spectrophotometric titration with 12-mer synthetic self-complementary oligodeoxynucleotide d(CGCAA ATTTGCG) 2 (GeneWorks, Thebarton, SA, Australia) or calf thymus DNA (Sigma). The titration of the ligand was performed in a spectrophotometric cuvette with the absorbance spectrum in the range from 300-400 nm recorded following addition of each subsequent aliquot of DNA. Spectrophotometric measurements were performed on a Cary 300 UV-Vis spectrophotometer (Varian Australia, Mulgrave, VIC, Australia). A set of recorded spectra was used to calculate the fractions of DNA bound ligand (f). The fraction of bound ligand at various DNA concentrations was then subjected to non-linear regression analysis to calculate binding parameters K d (binding dissociation constant) and b (frequency of binding sites). As an alternative approach, binding parameters were derived from non-linear regression analysis of the plasmid breakage data at different DNA concentration and ligand/plasmid ratio. Details of non-linear regression analysis are described in the online Supplementary Information (Data Analysis section).
For the X-ray crystallography studies, preparation of crystals of the complex formed between para-iodoHoechst and d(CGCAAATTTGCG) 2 , and the subsequent structural analysis, generally followed the procedures used for the counterpart methylproamine complex ). The procedures are also described in detail in the Supplementary Information (X-ray crystallography section and Table S1).

Monte Carlo simulation
The Auger-electron and conversion electron spectra were calculated using the BrIccEmis code (Lee et al. 2012) and nuclear structure data obtained from (Katakura & Wu 2008;Katakura 2011). The calculations were based on the atom in the condensed phase approximation, i.e., assuming fast neutralization. The ray tracing calculations were carried out using the Penelope 2008 code system (Salvat et al. 2008).

Plasmid/DNA-ligand incubation and quantitation of plasmid forms
We varied the conditions in the ligand-plasmid solution, such as DNA concentration A (plasmid/lm 3 ) and molar ligand/plasmid ratio P, in order to better evaluate the contribution of the remote component of DNA breakage and the effect of the fraction of bound ligand. To control the ligand/plasmid ratio, non-radioactive para-iodoHoechst was added to some samples as a solution in methanol and dried in a vacuum concentrator, prior to addition of the radio-labelled ligand.
Plasmid pBR322 DNA (Roche Diagnostics, Mannheim, Germany, 4361 bp, molecular weight 2.83 Â 10 6 dalton, 250 lg/ml) was diluted to the required concentration (10-40 lg/ml) in 20 mM Tris, 1 mM ethylene-diamine tetraacetate (EDTA, Sigma), 100 mM NaCl (Sigma) pH 7.3 (TE), or the same buffer containing 2M DMSO (TE þ DMSO). Aliquots of the DNA solution (70-80 ll) were added to incubation tubes containing iodinated ligand, and the ligand was dissolved in the DNA solution. Given that a fraction of the ligand remained adsorbed to the tube wall, in some experiments an aliquot of 90-95% of the initial volume was transferred to another tube. Tubes were incubated at 0 C for accumulation of decay events. Aliquots of the incubation samples (typically 4 ll volume) were taken after various incubation times and mixed with 16-20 ll of the gel loading buffer (Sigma) containing 0.03% bromophenol blue, 0.03% xylene cyanol and 3% glycerol in electrophoresis buffer. One part of this volume (6-12 ll, $50 ng of pBR322) was used for analysis by agarose gel electrophoresis, and the remaining part was used for radioactivity measurement.
Supercoiled, relaxed and linear form of plasmid were separated by electrophoresis on 0.8% agarose gel containing 0.005% of Vistra Green fluorescent dye solution (Amersham Biosciences, Piscataway, NJ) for DNA detection, in 42 mM Trisborate-acetate/1 mM EDTA buffer for 2.5-3 h at 70 V. Gel images were obtained on a Molecular Imager FX scanner (Bio-Rad Laboratories, Hercules, CA). Fluorescence intensities of gel bands corresponding to supercoiled, relaxed and linear plasmid were calculated using image analysis software Quantity One (BioRad Laboratories). To calculate the relative fractions of supercoiled, relaxed and linear form, fluorescence intensities were multiplied by correction coefficients reflecting variation in binding of the dye to different plasmid forms. These coefficients were determined in a separate experiment in which equal amounts of supercoiled, relaxed and linear forms were analysed on the same gel. The values of the coefficients were 1.0, 0.69 and 0.70 for supercoiled, relaxed and linear form, respectively.

Evaluation of radionuclide activity in samples
Radioactivity was counted on a Wallac 1470 Gamma Counter (Perkin Elmer Australia, Glen Waverley, VIC, Australia). For each sample analyzed by gel electrophoresis, radioactivity was normalized to the initial incubation sample volume and the start of incubation date and time. The average was calculated for each incubation sample.
Given the limited solubility of para-iodoHoechst in aqueous solutions and its ability to adsorb to the incubation tubes, a fraction of the radiolabelled ligand remained insoluble and was not recovered in the solution. At the end of incubation, the ligand-DNA solution was removed and the insoluble activity remaining in the tube was counted and normalized to the start incubation time. Based on the amount of the soluble and insoluble activity, we calculated the recovery ratio R, which is the fraction of the activity in solution (soluble activity) to the total activity in a sample. Values of R for each sample are presented in Tables S3-6.

Plasmid breakage analysis
The relationship between the number of radionuclide decay events per plasmid, n, and the fraction of supercoiled S(n), relaxed R(n) and linear L(n) plasmid forms is described by the following expressions (Lobachevsky & Martin 2004b): SðnÞ ¼ e Àsan 1 þ d a n LðnÞ ¼ d a n 1 þ d a n RðnÞ ¼ 1 À SðnÞ À LðnÞ where s a and d a are the apparent (observed) yield per plasmid per decay of SSB and DSB respectively. Values of s a and d a were obtained from the non-linear regression analysis of the experimental fractions of supercoiled S(n) and linear L(n) plasmid form for various numbers of accumulated decay events n using expressions 1 and 2. As discussed in detail in earlier publications (Lobachevsky & Martin 2004a, 2004b, breakage of a particular ('target') plasmid molecule can result from two quite different mechanisms. The first of these, and the one which is the focus of attention for the current study, we refer to as internal breakage; damage arising from a decay event involving a labelled ligand bound to the target plasmid. The second mechanism or component concerns damage to the target plasmid arising from decays in labelled ligands bound to neighbouring plasmid molecules or in unbound ligand molecules, and thus referred to as remote breakage. The two components are reflected in the following expressions for the apparent breakage yields (expressions S10-11 in the Supplementary Information): where the first term represents internal breakage with d and s being the probabilities per decay of DSB and SSB respectively; f is the fraction of bound ligand (fraction of DNA-associated decay events). The second term represents remote breakage determined by the yield of DSB, d and SSB, r per unit of absorbed dose (decay/lm 3 for convenience); A is the DNA concentration, R accounts for the presence of non-soluble radioligand that contributes to the remote breakage, as described in the previous section, but not to the evaluation of activity in solution. The dimensions of r and d in Equations 4-5 are breaks per plasmid per decays per lm 3 , and A is expressed as plasmids per lm 3 . The objective of the analysis is to determine the efficiency of internal breakage, that can be calculated from the apparent yield of breaks (s a and d a ) by subtraction of the remote component (rA and dA) and correcting for the fraction of DNA associated decay events (f): Non-linear regression analysis of the dynamics of plasmid forms was performed using SigmaPlot for Windows Version 11.0 software (Systat Software Inc, San Jose, CA). The regression tool of SigmaPlot was used to obtain values and standard errors of curve fitting parameters.
In vivo biodistribution studies in mice bearing K562 cell xenografts Experimental protocols that involved animals were approved by the Peter MacCallum Cancer Centre Animal Experimentation Ethics Committee. The biodistribution of the amide linked 125 Ilabelled transferrin conjugate was investigated in BALB/c nu/nu mice bearing a K562 human lymphoma (100-200 mm 3 ) subcutaneously implanted into the right flank. Mice were injected (100 ll intraperitoneal injection) with 125 I-labelled transferrin (130 kBq, 0.33 mg protein), the amide linked meta-[ 125 I]-iodoHoechst-transferrin conjugate (12 kBq, 0.5 mg protein) or meta-[ 125 I]-iodoHoechst carboxylic acid (2 in Figure S5; 5.6 kBq, 60 lM). Mice were euthanized 24 h after injection and various tissue samples were weighed and assayed for 125 I-activity on a Wallac 1470 Gamma Counter.
For PET imaging, the mice bearing xenograft tumor described above were injected intravenously with meta-[ 124 I]-iodoHoechst-transferrin conjugate (3.3 MBq) or directlylabelled 124 I-transferrin (2.6 MBq). The biodistribution of the radionuclide was investigated at various time intervals after administration by small animal PET imaging and tissue autoradiography. PET images were obtained from 15-min scans on a Phillips MOSAIC small animal PET scanner (Philips Medical Systems, Eindhoven, Netherlands), while mice were anaesthetized by inhalation of 2.5% isoflurane (Abbott Laboratories, Macquarie Park, NSW, Australia)/50% O 2 in air. PET images were reconstructed using the three-dimensional row action maximum likelihood algorithm and displayed using standard image software available on the scanner workstation. Tissue samples were dissected from mice euthanized after the 24-h PET imaging and autoradiograph images taken with a high resolution autoradiography imager (Biospace, Paris, France), and subsequent counting was done as described for the 125 I-samples.

DNA binding and structure of the ligand-DNA complex
The crystal structure of para-iodoHoechst-DNA dodecamer complex is shown in Figure 1. The critical feature of the crystal structure is the location of the iodine atom, in the middle of the minor groove, with the closest contacts to the walls of the groove being 3.64 Å to C5' of Thy 7 on one side of the groove, and 3.81 Å to C4' of Thy 21 on the other side. The DNA adopts a classical B-type conformation. The binding site covers approximately six base pairs over the 5 0 -ATTTGC sequence, with the aromatic rings of the ligand lying edge-in to the narrow AT tract of the minor groove of the helix. The planar benzimidazole groups and phenyl ring twist to follow the three-dimensional curvature of the groove, with the torsion angles of benzimidazole-benzimidazole and benzimidazole-phenyl being 16.9 and 23.2 , respectively. An unusual feature of the structure is the inclusion of a hexacoordinate Mg ion in the major groove. This and other features of the crystal structure are described in more detail in the Supplementary Information (X-ray Crystallography section and Table S2).
The following values of binding parameters were obtained: K d ¼ 88 6 47 nM, b ¼ 0.030 6 0.005 bp À1 for TE buffer and K d ¼ 220 6 84 nM, b ¼ 0.047 6 0.006 bp À1 for TE þ DMSO. These values were used to calculate fraction of bound ligand and to correct the breakage probabilities for each individual sample in Tables S3-6.

Monte Carlo simulation results
The complete list of the emitted radiations (c-rays, conversion electrons, Auger electrons and X-rays) are given in the Supplementary Information and in Lee et al. (2016). The energy spectra of the Auger and conversion electrons are shown in Figure S5 and S6 for 124 I and 125 I decay, respectively. These spectra were generated using 2.5 eV wide energy bins. The probability distributions of the absorbed energy for different radius water spheres are also shown in Figures S5  and S6. In the simulations, 10 8 electrons were ejected from the centre of the spheres. The results of Monte Carlo simulations are summarized in Table 1 that presents average per decay values for a range of parameters characterizing electron and electromagnetic radiations from Auger decay of 124 I and 125 I. There is a good agreement in terms of radiation yields between the present results and Howell's calculations (Howell 1992) up to the N-shell. The total number of Auger electrons is about 25% higher in the later publication. Most of the extra yield in 125 I can be attributed to low energy OOX transitions. Our calculations (Lee et al. 2016) show that these transitions are energetically forbidden even in a single ionized atom. While the shape of the emitted Auger and conversion electron spectrum is very similar for 124 I and 125 I, the average absorbed electron energy is different, due to the difference in the number of electrons per decay. The absorbed electron  Figures 2 and 3, respectively. The properties of the representative samples are listed in Table 2. Dynamics of the change in the relative proportions of the plasmid forms are shown as a function of the accumulated decay events per plasmid molecule for the case of iodinated ligands, or per lm 3 for the case of freely distributed radionuclide (sodium iodide). The use of these two chemical forms of the radionuclide is a reflection of two major mechanisms of DNA breakageinternal and remote for these two cases respectively that determine the appropriate variables and units for each mechanism (decay/plasmid or decay/lm 3 ) as described in more detail previously Lobachevsky & Martin 2004a). The curves shown in Figures 2 and 3 were generated by non-linear regression analysis of the experimental data. The best fit values for the apparent break yields, s a and d a , obtained from the regression analysis are given in Tables 3 (for SSB) and 4 (for DSB). The diagrams in panels A and B (Figure 2) demonstrate that the production of the linear form exceeds that of the relaxed form for incubation of plasmid with both [ 125 I]-and [ 124 I]-para-iodoHoechst. By contrast, in the case of incubation with freely distributed radionuclide, formation of the relaxed plasmid, that reflects the induction of SSB, predominates as demonstrated in panels C and D (Figure 2). Closer inspection of the results in panels C and D reveals that the yield of SSB damage is somewhat higher for 124 I decays compared to 125 I decays. For example for 125 I-sodium iodide, about half the plasmid molecules are relaxed after the accumulation of about 400 decays/lm 3 , whereas for 124 I-sodium iodide, a similar amount of damage is induced with only about 100 decays/lm 3 .
Whilst these comparisons relate to differences between the two radionuclides in relation to the minor non-radical  From data such as that shown in Figures 2 and 3, the probabilities of DSB and SSB per decay were calculated according Expressions 6 and 7, thus considering the contribution of remote breakage (rA and dA) and fraction of bound ligand (f). For this calculation, values for r and d were obtained from the regression analysis of the data from the experiments involving only freely distributed radionuclide (panels C and D in both Figures 2 and 3) assuming f ¼ 0 in Expressions 4 and 5. The results of calculation are shown in Table 3 (for SSB/relaxation events) and Table 4 (for DSB/linearization events).
The two salient features of comparison of the values for r and d given in Tables 3 and 4 are the dominance of SSB over DSB induction by remote breakage and the higher efficiencies of remote breakage for 124 I compared to 125 I. The SSB/DSB ratio values in TE þ DMSO are 54 and 94, and in TE 74 and 133, for 125 I and 124 I, respectively. Decay of 124 I is more efficient than 125 I by a factor of 4.1 and 6.3 for SSB induction, and by a factor of 2.3 and 3.5 for DSB induction, in TE þ DMSO and TE, respectively. Examination of the results in Table 4 confirms that the contribution of remote DSB is  Table 2. Properties of representative samples. The results and data analysis for these samples are shown in Figures 2 and 3 and Tables 2 and 4. The sample ID identifies samples in the complete set of results from an extensive series of experiments (more than 80 entries in total) performed under various conditions and listed in Tables S3, S4, S5 and S6 in the Supplementary Information, available online. Apart from parallel experiments done in TE in the presence and absence of 2M DMSO, the ligand/plasmid ratio was varied over two orders of magnitude (2-400), and the DNA concentration varied within the range of 2-8 plasmids per lm 3 . negligible for both radionuclides in the presence of the radical scavenger. However for SSB under scavenging conditions (Table 3), the contribution of the remote component is significant for 125 I ($5%) and substantial (>10%) for 124 I. Under non-scavenging conditions the contribution of remote breakage is still modest for DSB, but much higher for SSB, as reflected in the values of rA and dA in Tables 3 and 4, respectively. For DSB induction, the contribution of remote damage is low ($ 0.2%) for 125 I, but somewhat higher ($1.4%) for 124 I decays. For para-[ 125 I]-iodoHoechst in the absence of DMSO, and at the prevailing DNA concentrations, the relative contributions of remote and internal SSB breakage are similar, but for para-[ 124 I]-iodoHoechst, the majority ($ 80%) of apparent damage is due to the remote mechanism. The results and data analysis shown in Figures 2 and 3 and Tables 3 and 4, are representative of the complete set of Table 3. Analysis of SSB (relaxation events). The results are obtained from the data shown in Figures 2 (TE þ DMSO) and 3 (TE) for representative incubation samples. The yield of breaks r (per decay/lm 3 of freely distributed radionuclide) and the apparent yield s a (per decay/plasmid) are the best fit parameters and their standard errors obtained from non-linear regression analysis using Expressions 1, 2 and 4. Values of s are calculated using Expression 6.   results detailed in the Supplementary Information. A distillation of the results obtained from samples (listed in Tables 3-6 in the Supplementary Information, online) for internal breakage is summarized in Table 5, as probabilities of SSB and DSB formation following decay of a single DNA-associated radionuclide. These results demonstrate that DNA associated decay of a 'weaker' Auger electron emitter 124 I induces a DNA DSB with a probability of 0.58 in comparison to 0.85 DSB per decay of 125 I.

PET imaging and biodistribution studies
The biodistribution of radionuclide following injection of 124 Ilabelled DNA ligand-transferrin conjugate into mice bearing tumor xenograft was investigated by small animal PET imaging and autoradiography of tissue sections. In some experiments, 125 I was used instead of 124 I with sampling of tissues and measurement of 125 I-activity.
PET images of mice at various time intervals following injection of meta-[ 124 I]-iodoHoechst-transferrin conjugate or directly-labelled 124 I-transferrin are shown in Figure 4A, and autoradiographs of tumor, liver and muscle are shown in Figure 4B. Although the high liver uptake precluded imaging of the tumor xenograft implanted on the right flank of the mice, the outstanding feature of the results is the persistence of the liver image for the labelled conjugate, compared to the directly labelled transferrin, which was rapidly cleared, with some retention in the bladder.
The results of the uptake of radionuclide in a range of tissues 24 h after administration of meta-[ 125 I]-iodoHoechsttransferrin conjugate to tumor-bearing mice are shown in Figure 5. Directly labelled 125 I-transferrin and meta-[ 125 I]-iodoHoechst were used as controls in this experiment. Analysis of tissue samples collected demonstrated for liver 5.7, 0.8 and 1.2% ID/g (injected dose per gram) for conjugate, transferrin and unconjugated ligand respectively. The corresponding figures for tumor were 1.3, 1.1 and 0.13%. The findings demonstrate that in contrast to the labelled ligand which is rapidly metabolized and eliminated, the directly labelled transferrin persists in the circulation for 24 h after intraperitoneal injection. The results indicate good uptake of the conjugate into splenocytes and hepatocytes. The uptake of the conjugate in the tumor compared to directly labelled transferrin and ligand is encouraging.
Another key feature of the results in Figure 5 is the low 125 I-activity in the neck following injection of the conjugate, or the ligand, as compared to the directly labelled transferrin.

Discussion
Given the characteristic feature of Auger decay, the emission of multiple low-range electrons, the location of the radionuclide decay relative to the DNA is critical in generating strand breaks. The crystal structure of the ligand-DNA dodecamer complex obtained in our study confirmed that para-iodo Hoechst binds to the minor groove of DNA. The consensus binding site for bibenzimidazole ligands is 3-4 consecutive AT base pairs (Pjura et al. 1987;Murray & Martin 1988), so the binding site in the crystal structure, AAATTT is one of a large family of sites that occur in native DNA. The general features of the structure are comparable to published structures of Hoechst ligands with the same or similar DNA dodecamers (Squire et al. 2000), and to that for Hoechst 33258 bound to the same DNA as used in this study (Spink et al. 1994).
It is known from previous studies with decay of 125 I Lobachevsky et al. 2004;Lobachevsky & Martin 2005) that the majority of the remote DNA breakage (> 95%) is radical-mediated (i.e., indirect DNA damage), while internal breakage is induced mainly by direct ionization of atoms in the DNA molecule. Accordingly, inclusion of a radical scavenger such as DMSO minimizes the contribution of remote Our results of Monte Carlo simulation confirmed that decay of 124 I in the condensed condition, that produces on average 8.38 Auger electrons, is a weaker Auger emitter compared to 125 I, decay of which generates 20.0 Auger electrons. Considering the energy deposited by the decay event in a 1 nm sphere as a parameter for the prediction of DNA damage, there is a good correlation between values of this energy for 124 I and 125 I obtained from Monte Carlo simulation (155.3 and 364.3 eV, respectively) and the probabilities of DSB induction (0.58 and 0.85, respectively). The observation that the ratio of energy depositions for 125 I/ 124 I (2.35) is higher than the ratio of DSB probabilities (1.47) is consistent with the concept that for high energy deposition situations, associated with the induction of multiple DSB, would still be recorded as a single linearization event in the plasmid. Thus, the probability of linearization is expected to approach 1 as the energy deposition increases. Also, we have not considered in our analysis the potential contribution of the charge neutralization effect (Lobachevsky & Martin 2000b).
We demonstrated that use of the DNA ligand para-iodoHoechst, that positions radionuclide in close vicinity to DNA, enables induction of a DSB by decay of 124 I with a relatively high probability (0.58 per decay). This efficiency can be regarded as being appropriate to exploit DNA-targeted 124 I as a potential radionuclide for Auger endoradiotherapy, especially considering the additional feature, namely positron emission, that makes 124 I suitable for PET imaging. The potential for a combined therapy and PET imaging strategy is depicted in Figure 1.
The capability of PET imaging to monitor receptor-mediated delivery of DNA-targeted 124 I is demonstrated by the results of the proof-of-concept experiment summarized in Figure 4. We attribute the persistence of the liver image to accumulation of the labelled DNA ligand bound to nuclear DNA, with successive rounds of receptor-mediated endocytosis and intracellular degradation of the conjugate and consequent release of the labelled DNA ligand. Although this interpretation remains conjecture until nuclear uptake is demonstrated, it is consistent with high expression of transferrin receptors on liver cells (hepatocytes) that are involved in the transferrin bound iron metabolism (Gkouvatsos et al. 2012;Tandara & Salamunic 2012). Indeed, the number of transferrin receptors on liver endothelial cells was reported to be higher than that for K562 cells (Soda & Tavassoli 1984;Tavassoli et al. 1986), although this is controversial (Vogel et al. 1987).
The results of the 125 I-biodistribution experiments are consistent with PET imaging and autoradiography observations. These results indicate good uptake into hepatocytes and splenocytes, (the latter are also known to express relatively high levels of transferrin receptors), which can be attributed to sequestration of the 125 I-labelled ligand in the cells following injection of the conjugate, internalization and subsequent intracellular cleavage of the amide bond and nuclear localization of the 125 I-labelled ligand. However the observation that the uptake of the conjugate relative to that of transferrin is higher in the liver than in the tumor requires consideration of other explanations for the high liver uptake of the conjugate. For example, the 125 I-labelled ligand could be prematurely cleaved from the conjugate, and then accumulated in liver. This is consistent with relatively high uptake in the liver, compared to other tissues, following injection of the non-conjugated ligand, although overall recovery was low. Finally, the conjugate, in contrast to transferrin, could be accumulated in the liver as a part of its metabolism and elimination process, consistent with the general observation of high liver uptake of radiolabelled bioconjugates (Goldenberg 2002). It is relevant to note the low 125 I-activity in the neck following injection of the conjugate or the ligand, compared to the directly labelled transferrin (through 125 I-tyrosine), indicates that the labelled ligand is stable in vivo, in contrast to the directly labelled transferrin, which is known to be de-iodinated with subsequent accumulation of radioiodine in the thyroid.
In conclusion, we suggest that conjugates of 124 I-labelled DNA ligands with tumor-targeting proteins that are internalized by receptor-mediated endocytosis have significant potential as a general therapy and PET imaging platform. The strategy invokes two levels of targeting. The specificity of receptor-mediated targeting to tumor cells, with evaluation of the efficacy of the chosen receptor system by PET imaging, and the DNA ligand targets the highly focused radiochemical damage associated with Auger emitters to the radiosensitive molecular target.