“One Ring to Bind Them All”—Part II: Identification of Promising G-Quadruplex Ligands by Screening of Cyclophane-Type Macrocycles

A collection of 26 polyammonium cyclophane-type macrocycles with a large structural diversity has been screened for G-quadruplex recognition. A two-step selection procedure based on the FRET-melting assay was carried out enabling identification of macrocycles of high affinity (ΔT1/2 up to 30°C) and high selectivity for the human telomeric G-quadruplex. The four selected hits possess sophisticated architectures, more particularly the presence of a pendant side-arm as well as the existence of a particular topological arrangement appear to be strong determinants of quadruplex binding. These compounds are thus likely to create multiple contacts with the target that may be at the origin of their high selectivity, thereby suggesting that this class of macrocycles offers unique advantages for targeting G-quadruplex-DNA.


Introduction
The macrocyclic scaffold has proven to be highly valuable for the design of efficient and selective G-quadruplex ligands (see the review article in the present issue, "One Ring to Bind Them All"-Part I, by Monchaud et al.) [1][2][3][4]. Along these lines, two families of macrocyclic ligands have been particularly and thoroughly studied for their ability to specifically interact with quadruplex-DNA, namely, the families of telomestatin- [5] and porphyrin (TMPyP4)-related compounds [6][7][8]. Recently, a third family has emerged, namely the polyammonium cyclophane-type macrocycles, generally referred to as "CBI macrocycles" (for Cyclo-Bis-Intercalators), that adopt a non-planar conformation and therefore are structurally highly different from the two aforementioned classes. The efficiency of this family of ligands with regard to quadruplex DNA recognition is particularly illustrated by the compound BOQ 1 (Figure 1(a)) [9][10][11] as discussed in "One Ring to Bind Them All"-Part I. This compound is comprised of two large aromatic quinacridine units [12][13][14] able to interact with DNA bases through ππ interactions, and linked together via two short polyamine side chains, which ensure a global tetracationic charge at physiological pH affording both high water solubility and strong electrostatic attraction to DNA.
We have previously reported that BOQ 1 exhibits a high affinity for the human telomeric G-quadruplex DNA, along with a poor ability to associate with double-stranded DNA [9][10][11]. Consequently, this compound not only rivals the performances of telomestatin and TMPyP4 in terms of quadruplex-affinity (Figure 1(c)) [16], but elicits a quadruplex-over duplex-DNA selectivity comparable to that of telomestatin and highly superior to that of TMPyP4. One key to explain the very good quadruplex-binding properties (It is worth noting that the FRET-melting values obtained herein for BOQ 1 differ from that of the initial publication (see [9]). This discrepancy originates in the different experimental conditions used, given that FRET-melting is performed herein as an HTS unlike the initial report (single-cell experiment); the disparities between the two methods are extensively detailed in [15] (among which can be cited the final volume (25 versus 600 μL), the buffer conditions (10 mM lithium cacodylate + 100 mM NaCl versus 10 mM sodium cacodylate, 100 mM LiCl), etc.).) of CBIs stands in their peculiar structural organization, which represents a non-planar conformation with a significant degree of molecular flexibility [11], thus clearly distinct from the planar rigid macrocycles telomestatin and TMPyP4. Indeed, previous works (NMR, X-rays and molecular modelling studies, vide infra) have shown that the CBI scaffold exists in a semiclosed conformation (Figure 1(b)) that results both from internal hydrophobic interactions between the two aromatic moieties and from a constrained distance between them imposed by the linkers [17]. This particular spatial organization is an essential determinant that governs the interaction of CBIs with secondary structures of DNA. In the case of quadruplex-DNA, BOQ 1 was initially thought to interact in an open-state conformation [9], but subsequent modelling studies [11] as well as recent molecular dynamics studies [18], suggested that the CBI scaffold never adopts the fully open conformation. In fact the semi-closed conformation of BOQ 1 has been shown to be not only the most stable conformation after in silico geometry optimization (Figure 1(b)) but also to be the favoured one when interacting with quadruplex structures (see "One Ring to bind them all"-Part I) [11]. It is thus hypothesized that this particular structure is at the origin of the efficient G-quadruplex recognition performed by BOQ 1 , which is supposed to occur via a mixed binding mode combining quartet and loop interactions. Finally, given that the length of the polyammonium linker does not enable intercalation of both aromatic residues between contiguous base pairs in duplex-DNA due to the neighbour-exclusion principle [19,20], the macrocyclic scaffold of BOQ 1 impedes binding to duplex-DNA, and consequently is responsible for the good selectivity for quadruplex-versus duplex-DNA. The promising results obtained with BOQ 1 in terms of quadruplex recognition and selectivity prompted us to screen a collection of 26 CBI-type macrocycles previously synthesized [17,18,21,22]. The members of this series (represented schematically in Figure 2, in extenso in Figure 3) differ by the nature of the aromatic units, the topology of the macrocyclic framework that contains either two different aromatic units or the same units, but with different connectivities; the nature and the derivatization of the linking chains; and finally the number of the constitutive "intercalator" units (two for CBI, three for CTI (Cyclo-Tris-Intercalators), Figure 2). In detail, the macrocycles are comprised of either polyaromatic cycles (naphthalene, anthracene, and biphenyl moieties), or heterocycles (such as acridine, quinacridine, phenanthroline, phenazine and bipyridine moieties) or even an organometallic unit (ferrocene), along with the possibility of various connectivities that can be present in a given series. Moreover, oxygen or sulphur atoms were introduced in the linking chains in place of the secondary amino groups present in BOQ 1 (X = Y = NH, O or S, Figure 2), and in some cases the linkers bear tertiary amino groups due to substitution with pendant arms (X = Y = NR). All these structural variations allow to divide the present collection of macrocycles into four categories ( (c) pendant-arm macrocycles containing a bisnaphthalene scaffold (A = B = 2, 6-naphthylene) with one (monobrachial) or two (bibrachial) side-chains with additional functional groups; and (d) homotrimeric macrocycles (cryptand-type) containing three identical aromatic units. With this series in hand, we systematically studied the influence of the various structural components of the macrocyclic scaffold on both the quadruplex-affinity and selectivity.

Results and Discussion
To evaluate the 26 compounds shown in Figure 3, we used the FRET-melting assay, which has been recently developed to a high-throughput screening format and provides reliable information concerning quadruplex-affinity and -selectivity in a straightforward manner [15]. This method is based on monitoring the ligand-induced stabilization of a fluorescently labelled quadruplex-forming structure (F21T, FAM-G 3 [T 2 AG 3 T] 3 -Tamra). The stabilization, measured via a fluorescence resonance energy transfer (FRET) between the two fluorescent partners (6-carboxyfluorescein, FAM, and 6-carboxytetramehtylrhodamine, Tamra), is expressed as an increase in melting temperature of F21T (ΔT 1/2 ) in the presence of the ligand. This method also enables to gain insight into the quadruplex-over duplex-DNA selectivity if the melting experiments are carried out in the presence of various amounts of a 26-bp doublestranded DNA competitor (ds26, the self-complementary [5 -CA 2 TCG 2 ATCGA 2 T 2 CGATC 2 GAT 2 G-3 ], up to 50 equiv.) Last but not least, FRET-melting enables also a direct comparison with recently reported ligands, given that this technique is now widely used by the quadruplex DNA community and often, as it is the case herein, as a test for the primary selection of ligands.

Selection of the High-Affinity Ligands
In the first round of experiments, we searched for the ligands with high affinity for the human telomeric quadruplex-DNA. FRET-melting experiments were thus carried out in the standard conditions previously calibrated for selection of high-affinity ligands [15], that is, with 0.2 μM F21T and 1 μM ligand. This first round of selection was performed both in sodium (10 mM lithium cacodylate, 100 mM NaCl) and potassium conditions (10 mM lithium cacodylate, 10 mM KCl, 90 mM LiCl) of identical ionic strength. This systematic comparison was necessary since the telomeric quadruplex is known to fold into different conformations depending on the cation present in the medium, displaying different loop arrangements that might result in differences in ligand binding [23][24][25]. The two conformations currently admitted to correspond to the predominant species in solution and in each are represented in Figure 1 of "One Ring to Bind Them All"-Part I. The results of the FRET-melting experiments with the 26 macrocycles are presented as a bar graph in Figure 4.
This series of experiment enabled us to classify the compounds in two categories, namely the high-affinity ligands (ΔT 1/2 > 10 • C) and the ligands of moderate to low affinity (ΔT 1/2 < 10 • C). The selection criteria, that is, the affinity threshold of 10 • C, was motivated by the experience acquired over the past years when developing the FRETmelting assay.
Altogether the results of this primary screen confirm that the association between the macrocycles and the quadruplex-DNA is strongly dependent on both the rigidity and the topology of the macrocyclic framework. Finally, as depicted in Figure 6, the behaviour of the majority of compounds shows a poor dependency on the cation present in the medium, thereby confirming that the quadruplexrecognition is primarily governed by tetrad-recognition.

Selection of Highly Selective Ligands
The first selection based on the amplitude of the F21T stabilisation induced by the ligand resulted in the selection of 16 compounds. For entering the second round of experiments, we further decided to arbitrary remove two candidates (3, 3 -BisBP and 2,7-NP/9,10-AN), since they do not reach the affinity threshold (ΔT 1/2 = 10 • C) in the two cationic conditions (Figure 4). The 14 remaining compounds were thus subjected to the second selection step that consists in reproducing the melting experiment of F21T in the presence of the nonlabelled duplex-DNA competitor ds26. As shown in previous works, the presence of the duplex-DNA competitor diminishes the quadruplex-binding of nonselective compounds (thus resulting in decreased ΔT 1/2 values) and poorly affects the binding of selective compounds (thus resulting in unchanged ΔT 1/2 values). The competitive FRET-melting assay enables thus a rapid evaluation of quadruplex-over duplex-DNA selectivity, a criterion that is of utmost importance for further developments of quadruplex ligands in cell-based assays.
Competitive FRET-melting experiments were carried out in the conditions used in the first experiment [15], that is, with 0.2 μM F21T, 1 μM ligand without ds26 or with two different concentrations of ds26 (3 and 10 μM, i.e., 15-and 50-molar excess relative to F21T). These experiments have been performed with the 14 selected macrocycles both in Na + and K + conditions; the results (ΔT 1/2 values) are summarized as bar graphs shown in Figure 7.
It is worth mentioning here that the selectivity ranking is not globally affected by the nature of the cation (Na + versus K + , Figures 7(a) and 7(b), resp., Table 1); however, the values are significantly higher in presence of K + , thereby indicating a higher resistance of the ligands to the competition of the duplex-DNA in these conditions. For this reason and also for the sake of clarity, we used only results obtained in K + (Figure 7(b)) for rationalizing the data: a close examination of the results enables us to identify three structural parameters that influence the selectivity of the quadruplex-recognition by the macrocycles, namely, (a) the presence of pendant arms, (b) the existence of a specific three-dimensional conformation, and (c) the rigidity of the macrocyclic framework. Remarkably, pendant side-arms terminated by a functional group (pyrene (PY) or ferrocene (FC)) lead to ligands that exhibit higher quadruplexaffinity and -selectivity than their amino-terminated counterpart (i.e., BisNP-1NH 2 , Figure 7 and Table 1; e.g., in presence of 10 μM ds26, ΔT 1/2 = 29.7 versus 20.7 • C and FRET S = 0.72 versus 0.38 for BisNP-1PY and BisNP-1NH 2 , resp.). This clearly indicates that these functional groups are responsible for additional interactions with quadruplex-DNA, eventually with loops and/or grooves, which may result in the multiple anchorage of the ligand in its binding site. Finally, the better performances of monobrachial compounds as compared to bibrachial ones, both in terms of affinity (in presence of 10 μM ds26, ΔT 1/2 = 29.7/25.  These two cases indicate that subtle modifications of the aromatic units that compose the CBI scaffold significantly increase the selectivity; these modifications may contribute to determine a particular topology of the CBI, thus disfavoring the association with duplex-DNA. Interestingly, the crystal structure of 2,9-BisPN has recently described [31] evidence of a "horseshoe" conformation that may be prone to fit into the quadruplex-DNA bonding site. The importance of the 3D topology suggested here is in line with the role of the connectivity observed during the primary screen (vide supra).
(c) Finally, in the NP series, the formal introduction of a third NP branch in 2,7-BisNP scaffold significantly improves the binding properties of the resulting compounds (i.e., 2,7-TrisNP, in presence of 10 μM ds26, ΔT 1/2 = 22.7, and FRET S = 0.57); we can thus conclude that, given that 2,7-TrisNP is structurally more constrained than 2,7-BisNP, the rigidity of the ligand is a valuable parameter to control when searching to improve the quality of the recognition of the DNA target. The same trend is observed in the bipyridine (BP) series, since 3, 3 -TrisBP displays a higher quadruplex-affinity than 3, 3 -BisBP (ΔT 1/2 = 14.9 versus 10.2 • C); interestingly, 3, 3 -TrisBP shows a quite exceptional selectivity (with FRET S >0.98), although its stabilizing effect is modest as compared to 2,7-TrisNP. These two examples demonstrate the interest of the CTI design for improving the quadruplex-selectivity, supporting the observation that the rigidity of the CBI is not only favorable for the quadruplex-binding but also for decreasing the duplex-binding. Last but not least, these results also indicate that electrostatic interactions are not critically determinant for the binding of CTI to both quadruplex-and duplex-DNA. Thus, the two-step selection process described herein (step 1: affinity, step 2: selectivity) enables the rapid identification of six candidates (2,9-BisPN, 2,7-NP/2,7-A, BisNP-1PY, BisNP-1FC, BisNP-2FC, and 3, 3 -TrisBP) with remarkable quadruplex-interacting properties, that is, high affinity and selectivity. To further increase the selection pressure and select hits for subsequent developments, we decided to focus our attention on compounds that display ΔT 1/2 ≥ 15 • C in presence of 10 μM ds26 (upper red dashed line in Figure 7(b)). Four compounds fulfill this requirement, namely, 2,7-NP/2,7-A, BisNP-1PY, BisNP-1FC, and 3, 3 -TrisBP (Figures 7 and 8, it is worth noting that this selection has been extended to 3, 3 -TrisBP, since its ΔT 1/2 in presence of 10 μM ds26 (14.7 • C) is an average value resulting from experiments with ΔT 1/2 ≥ 15 • C). Amongst these four compounds, two were thus elected as the most promising compounds over the 26 macrocycles screened: (i) BisNP-1PY, since this CBI combines the best affinity and selectivity on the basis of the three values used for the selection procedure (in K + conditions: ΔT 1/2 = 29.7 and 21.5 • C in absence and presence of 10 μM ds26, resp. FRET S = 0.72; in Na + conditions: ΔT 1/2 = 31.9 and 22.9 • C in absence and presence of 10 μM ds26, resp., FRET S = 0.71), and (ii)-3, 3 -TrisBP, since, while displaying a decent affinity, this CTI reaches an unprecedented level of quadruplex-over duplex-DNA selectivity ( FRET S = 1.01 and 0.98 in Na + and K + conditions, resp.).
The two-step selection procedure followed herein as well as the structural diversity of the collection studied enabled us to identify several structural features that govern the efficiency of the quadruplex-recognition by the CBI ligands. Among these, the presence of a pendant side-arm and the existence of a particular topological arrangement appear to be strong determinants. The selected hits possess sophisticated architectures and are thus likely to create multiple contacts with the quadruplex-DNA structure. This should result from a complex interplay of π-stacking interactions, H-bonding, and electrostatic forces both with the accessible G-tetrad of the quadruplex-DNA and with the quadruplexelements that surround the G-tetrad (loops and/or grooves). These secondary interactions may be at the origin of the high selectivity observed.
In conclusion, we have performed the first step toward the identification of two molecules that are highly promising ligands of the human telomeric quadruplex in vitro and opened the way towards new molecular designs of high efficiency. We are thus now performing further studies with these ligands to investigate their peculiar binding mode and in vivo properties.