Optimization of G‐Quadruplex Ligands through a SAR Study Combining Parallel Synthesis and Screening of Cationic Bis(acylhydrazones)

Abstract

G‐quadruplexes (G4s), secondary structures adopted by guanine‐rich DNA and RNA sequences, are implicated in numerous biological processes and have been suggested as potential drug targets. Accordingly, there is an increasing interest in developing high‐throughput methods that allow the generation of congeneric series of G4‐targeting molecules (“ligands”) and investigating their interactions with the targets. We have developed an operationally simple method of parallel synthesis to generate “ready‐to‐screen” libraries of cationic acylhydrazones, a motif that we have previously identified as a promising scaffold for potent, biologically active G4 ligands. Combined with well‐established screening techniques, such as fluorescence melting, this method enables the rapid synthesis and screening of combinatorial libraries of potential G4 ligands. Following this protocol, we synthesized a combinatorial library of 90 bis(acylhydrazones) and screened it against five different nucleic acid structures. This way, we were able to analyze the structure–activity relationships within this series of G4 ligands, and identified three novel promising ligands whose interactions with G4‐DNAs of different topologies were studied in detail by a combination of several biophysical techniques, including native mass spectrometry, and molecular modeling.

Ready to screen. A library of cationic bis(acylhydrazones) as putative G4‐DNA ligands was prepared by a straightforward condensation of the corresponding building blocks. Screening the resulting samples (without any isolation or purification steps) by fluorescence‐melting experiments identified three novel hits, whose interactions with G4‐DNA targets were studied in detail by several biophysical methods and molecular modeling.

Introduction

G‐quadruplex (G4) DNA and RNA structures play important roles in regulating DNA‐ and RNA‐related cellular transactions such as replication, transcription, telomere maintenance, epigenetic regulation, DNA damage, and translation of mRNA. ^[1] The effects of G4 structures might stem from the formation of physical roadblocks interfering with the progression of DNA or RNA polymerases,[ ² , ³ , ⁴ , ⁵ ] promotion of R‐loops formation, ^[6] recruitment of G4‐interacting proteins, ^[7] or other mechanisms. ^[8] Small molecules (ligands) that specifically bind to G4 structures may interfere with G4‐mediated regulation, acting through excessive stabilization of these structures, ^[9] by competing with G4‐interacting proteins, ^[10] or through the formation of ternary complexes. ^[11] These effects may be exploited to gain control over G4‐regulated biological processes with therapeutic purposes.[ ⁹ , ¹² , ¹³ ] Indeed, a number of G4 ligands demonstrated enticing biological properties, and a few have been admitted to clinical trials.[ ¹⁴ , ¹⁵ , ¹⁶ ] However, considering the large number of putative G4‐forming loci in the genome, ^[17] the potential off‐target effects of G4 ligands might be disastrous. ^[18] For these reasons, ligands with a specificity towards a given G4, or at least a given subset of G4 structures, are highly desirable but still extremely scarce.[ ¹⁹ , ²⁰ ]

While certain progress in the design of more specific G4 ligands has been achieved during the last years, ^[21] one of the obstacles is the lack of high‐throughput technologies combining the synthesis and biochemical evaluation (profiling) of G4‐tagerted libraries of putative ligands. In this regard, emerging technologies such as G4‐targeted fragment‐based screening,[ ²² , ²³ ] DNA‐encoded chemical libraries, ^[24] affinity chromatography[ ²⁵ , ²⁶ ] and affinity‐selection mass spectrometry,[ ²⁷ , ²⁸ ] or rapid synthesis and evaluation of G4 ligands on a microfluidic platform ^[29] represent promising alternatives to the classical, robust but time‐consuming combinatorial chemistry approaches that rely on the preparative synthesis of putative ligands, followed by their biophysical or biochemical assessment on a one‐by‐one basis.[ ³⁰ , ³¹ , ³² ]

Within the chemical space of G4 ligands, quaternary heterocyclic bis(carboxamides) represent one of the oldest and most efficient families. The prototype of this series, bisquinolinium PDC (360A, Figure 1) was shown to localize to telomeres and exhibit anticancer activity.[ ³³ , ³⁴ ] Subsequently, this compound, along with a few others, has been extensively exploited as a scaffold for the development of G4‐targeted photo‐crosslinking, ^[35] photocleaving, ^[36] fluorescent,[ ³⁷ , ³⁸ ] and immunotagged probes. ^[39] The phenanthroline derivative PhenDC3 ^[40] currently represents the “gold standard” in terms of G4 affinity (with K _d values down to 2.1–2.4 nM, according to SPR studies) and selectivity with respect to duplex DNA, ^[41] and is being widely used as a reference G4 probe in biological models.[ ⁴² , ⁴³ ] Likewise, naphthyridine analogues (e. g., 3AQN) ^[44] also demonstrate efficient G4‐binding properties and anti‐parasitic (Plasmodium falciparum) activity, ^[45] showing evidence of the potential of this molecular scaffold.

Figure 1.

Structures of quinolinium bis(carboxamide) G4 ligands PDC (360A), PhenDC3, and bis(acylhydrazone) analogues PyDH2 and PhenDH2. Counter‐ions are omitted for clarity.

Along these lines, we demonstrated that cationic bis(acylhydrazones) PyDH2 and PhenDH2 (Figure 1) bind to G4‐DNA and G4‐RNA at least as efficiently as their bis(carboxamide) counterparts. Both compounds were proposed for therapeutic targeting of the G4‐RNA‐modulated immune evasion of the oncogenic Epstein–Barr virus, especially because of their significantly lower host‐cell toxicity as compared to PhenDC3.[ ⁴⁶ , ⁴⁷ ] More recently, PhenDH2 was shown to bind to G4‐RNA structures in LANA1 mRNA of Kaposi's sarcoma‐associated herpesvirus and interfere with nuclear export, nucleolin binding, and translation of this mRNA more efficiently than PhenDC3. ^[48] With the aim of further exploration of this promising scaffold, we elaborated the generation of G4‐targeted dynamic combinatorial libraries (DCLs) of bis(acylhydrazones) in conditions compatible with native G4‐DNA structures. ^[49] However, the identification of ligands from DCLs turned out to be tedious and not free of bias due to nonspecific interactions of ligands with the oligonucleotide support (i. e., streptavidin‐coated magnetic beads). Therefore, novel methods for high‐throughput synthesis and evaluation of potential ligands are still needed. Towards this end, in this work, we designed a combinatorial chemistry strategy based on the straightforward preparation of “ready‐to‐screen” libraries of cationic bis(acylhydrazones), coupled with a screening of their interaction with a panel of G4‐DNA targets. We demonstrate that this technically simple approach provides valuable information about structure–activity relationships of G4 ligands and results in a successful identification of several novel ligands, that combine an unprecedented level of affinity to G4‐DNA with preferential binding to antiparallel structures.

Results and Discussion

Library design and synthesis

We designed a combinatorial library of 90 cationic bis(acylhydrazones) 3Aa–Ji (Scheme 1) that could be obtained through a one‐step condensation of a set of ten bis(acylhydrazides) (1A–J) with nine cationic aldehydes (2 a–i). The set of bis(acylhydrazides) was constructed to include heterocyclic derivatives with a varied number of condensed rings (1A–E), a benzene counterpart (1F), three aliphatic derivatives with a varied length of the spacer between the acylhydrazide groups (1G–I), as well as carbohydrazide (1J) as the minimal dihydrazide. Chelidamic acid dihydrazide (1K) was also considered but discarded after the initial trials due to the formation of undesired side products during the reaction. The set of cationic aldehydes included the quaternary derivatives of pyridine (2 a), quinoline (substituted at different positions with alkyl or benzyl residues, 2 b–h), and a pyridinium‐substituted benzaldehyde derivative (2 i); all derivatives were readily obtained by the quaternization of the corresponding precursors (Supporting Information). Of note, several of the resulting bis(acylhydrazones), including 3Ab (i. e., PyDH2) and 3Eb (i. e., PhenDH2), were previously obtained by conventional synthesis. ^[46]

Scheme 1.

Library design and one‐step synthesis of “ready‐to‐screen” solutions of bis(acylhydrazones).

In order to generate “ready‐to‐screen” samples of the products, stock solutions of 1A–J were mixed with solutions of 2 a–i in individual vials, to give reaction mixtures containing 2.0 and 4.4 mM (2.2 molar equivalents) of dihydrazide and aldehyde components, respectively, in DMSO, and supplemented with AcOH (final concentration: 1 M). All 90 reaction vials were simultaneously heated at 60 °C for 48 h, to ensure complete conversion of the dihydrazides. The use of DMSO as reaction solvent allowed to avoid the precipitation of final products, which facilitated the subsequent workflow; moreover, DMSO is the standard solvent for preparation and handling of compound libraries, compatible with most biochemical and even cellular screening protocols. Of note, higher reaction temperatures (e. g., 100 °C) and shorter times were disadvantageous, as these conditions led to the formation of significant amounts of dealkylated products resulting from the heat‐induced loss of alkyl substituents from quaternary nitrogens. ^[50] The resulting solutions were analyzed by LC/MS and revealed an average purity of 87 % and a minimum purity of 80 %, as per peak area (Table S1 in the Supporting Information), approaching the standards of commercially available pre‐plated compound libraries (purity threshold of 85–90 %).[ ⁵¹ , ⁵² ] For example, the concentration of products 3Ab and 3Eb in crude reaction mixtures was verified by comparison of LC peak areas with that of 2 mM solutions prepared from previously synthesized, analytically pure compounds, and was found 1.83 and 1.74 mM, respectively (Figure S1). The as‐synthesized sample solutions were subsequently employed for screening by fluorescence‐melting experiments without further treatment.

Screening by fluorescence‐melting experiments

The capacity of the as‐synthesized, “ready‐to‐screen” ligands to stabilize G4‐DNA was assessed by fluorescence‐melting experiments, which monitor thermal denaturation of G4‐forming oligonucleotides labeled with a fluorophore–quencher pair.[ ⁵³ , ⁵⁴ , ⁵⁵ ] Four representative G4‐DNA targets belonging to different topology classes were selected for screening: 25TAG (a variant of human telomeric sequence forming hybrid‐2 structure in K⁺ conditions); 22CTA (another variant of human telomeric sequence forming an antiparallel structure), myc22 (a variant of the MYC promoter sequence forming a parallel G4 structure), and Pu24T (another variant of the MYC promoter sequence forming a parallel G4 structure featuring a unique snap‐back diagonal loop), schematically presented in Figure 2a (for sequences, see Table S2). A hairpin with 8 base pairs in the stem part (hp2) was included as a mimic of double‐stranded DNA with a comparable thermal stability. As the intrinsic thermodynamic stability of different DNA structures varies, the conditions of melting experiments were adjusted by varying the K⁺ content of the buffers, while keeping the overall ionic strength identical, such that the denaturation of all structures (including hp2) was observed at T _m ⁰≈60 °C (Table S3). This way, ligand‐induced stabilization of different G4‐DNA can be compared with a minimal bias. ^[55]

Figure 2.

Results of the fluorescence‐melting assay with 90 “as‐synthesized” bis(acylhydrazone) ligands 3Aa–Ji. a) Schematic depiction of the G4‐DNA structures used in the assay; all oligonucleotides were labeled with a fluorophore (5′‐FAM) and a quencher (3′‐TAMRA). Heatmaps of ligand‐induced stabilization (ΔT _m) of G4‐DNA: b) 25TAG, c) 22CTA, d) myc22, and e) Pu24T. Data are average values from three technical replicates. Conditions: c(G4‐DNA)=0.2 μM, c(ligands) ≈1 μM, buffers: see Table S3. f) Structures of novel bis(acylhydrazone) derivatives selected for detailed studies.

In a preliminary test, fluorescence‐melting experiments were carried out using a representative G4‐DNA (myc22), either in the presence of two as‐synthesized samples (3Ab and 3Eb) or with pure compounds obtained by conventional synthesis. The results revealed no significant differences in the ligand‐induced stabilization effect (Figure S2), confirming that neither the impurities formed during the synthesis nor the catalyst (AcOH) interfered with G4 melting or ligand‐induced stabilization. Indeed, since the standard conditions of this assay employ a relatively large excess of ligands (1 μM, or 5 molar equivalents), the stabilization effect of high‐affinity ligands typically reaches a plateau in the presence of ∼2.5 molar equivalents of ligands. ^[56] Conversely, the effect of poor ligands shows a rather weak concentration dependence in the broad range of concentrations. Therefore, minor differences in ligand concentration have no or very little effect on the ΔT _m values or the shape of melting curves. On these premises, the “ready‐to‐screen” solutions of ligands, after dilutions with the corresponding buffer, were directly employed for fluorescence‐melting screening experiments performed in a 96‐well microplate format, both in the absence and in the presence of a double‐stranded competitor (10 μM ds26, i. e., 50 molar equivalents with respect to the G4 target). The results of this screening, presented as heatmaps of ΔT _m values (Figure 2b–e; for numerical data, see Table S4), allow a direct comparison of the relative capacity of ligands to stabilize different G4 structures and an estimation of their G4‐vs.‐duplex selectivity.

Structure–activity relationships and hit identification

The results of fluorescence‐melting screening display some interesting regularities. With respect to molecular structure, the capacity of bis(acylhydrazones) to stabilize G4‐DNA structures seems to be governed mostly by the combination of two major factors, namely i) the nature of the central unit (L in Scheme 1) and ii) the total number of aryl rings (N _Ar). Regarding the nature of the central unit, the derivatives containing pyridine (A), pyrazine (C), naphthyridine (D), and phenanthroline (E) fragments systematically demonstrated high capacity to stabilize G4‐DNA, with the latter giving rise to the most efficient ligands; conversely, ligands containing pyrimidine (B) and benzene (F) fragments, as well as all aliphatic derivatives (G–J) were significantly less efficient. This fact confirms the assumption that G4‐binding properties are directly related to the ability of the ligand to adopt a V‐ or a U‐shaped conformation, presumably stabilized by water‐mediated hydrogen bonds between the acylhydrazone groups and the central heterocyclic unit, similar to those observed in the solid‐state structures of 3Aa and 3Ea. ^[46] The importance of the second factor, N _Ar, is demonstrated by the strong, statistically significant positive correlation between N _Ar and ligand‐induced stabilization (ΔT _m, Figure S3), as can be seen from the corresponding Pearson's r values (0.62–0.74, p≪0.001, Table S5). Accordingly, the derivatives containing “unfavorable” central units but a large number of aromatic rings in side chains, such as carbohydrazone derivatives 3Jc and 3Jh (N _Ar=6) show significant thermal stabilization effects, with ΔT _m values from 6–9 °C (for myc22 and Pu24T) to 11–14 °C (for 25TAG and 22CTA). Finally, the nature of lateral moieties (X in Scheme 1) appears to play a secondary role in G4‐stabilizing properties. Thus, 1‐methylpyridinium (X=a) and 4‐(pyridinium‐1‐yl)methylphenyl (X=i) substituents, generally, resulted in less efficient G4 stabilizers, compared to quinolinium fragments, unless combined with the phenanthroline unit such as in compounds 3Ea and 3Ei. Conversely, in the case of quinolinium substituents (X=b–h), the G4‐binding properties of the ligands were relatively insensitive to the nature or position of the side chain.

Importantly, most ligands behaved rather uniformly with respect to different G4‐DNA targets, considering the generally lower susceptibility of parallel G4 (myc22 and, particularly, Pu24T) to ligand‐induced thermal stabilization as compared to 25TAG and 22CTA (Figures 2d–e and S3). This is additionally illustrated by a strong mutual correlation of ΔT _m values obtained with different G4 structures (r=0.94–0.98, p≪0.001, see Figure S4 and Table S5) and indicates that selective targeting of one or another G4 structure can hardly be achieved within this series of ligands. However, one notable exception is represented by compound 3Ei which, in addition to anomalously high stabilization of all four G4‐DNA, demonstrated a significant bias towards the antiparallel G4 22CTA (Figure S4a and d). Finally, none of the compounds had any significant effect on the melting of the hairpin hp2 (ΔT _m≤4.0 °C, Table S4 and Figure S5). Moreover, for most ligands the stabilization effect did not decrease by more than 1–2 °C when the experiments were performed in the presence of double‐stranded competitor (Table S4), demonstrating an excellent level of G4‐vs.‐duplex selectivity.

Considering all the data, we selected and synthesized three compounds (Figure 2f) in a preparative fashion, in order to obtain analytically pure samples (Supporting Information). Specifically, in addition to compound 3Ei identified above, we chose its close analogue 3Ef that consistently demonstrated high stabilization of most G4 structures without skewness towards one of them, as well as ligand 3Cb, which was considered as a representative moderate‐affinity G4 binder of the series (Figure S3). The G4‐binding properties of these ligands were studied in more detail and compared with those of PyDH2 (3Ab) and PhenDC3.

Hit validation

As mentioned above, the conditions of fluorescence melting experiments employed in the screening step (i. e., 5 molar equivalents of ligands) were selected in order to compensate for possible minor differences in ligand concentrations in “ready‐to‐screen” samples. However, these conditions are poorly suited for the ranking of high‐affinity binders, since the stabilization effect induced by such ligands (ΔT _m≥36 °C) reaches the experimental limit of the method, and therefore, lower concentrations of ligands (0.2–0.5 μM) have been recommended for the assessment of high‐affinity binders.[ ⁵⁵ , ⁵⁶ ] For the pure ligands, we studied the ligand‐induced stabilization of the four G4 structures a range of ligand concentrations (0.1–1.0 μM). The results (Figure 3) demonstrated that, in all cases, almost no stabilization was observed with 0.1 or 0.2 μM of ligands. However, with 25TAG, compounds 3Ei and PhenDC3 yielded an almost maximal stabilization effect (ΔT _m >30 °C) already at a 0.4 μM concentration; other ligands were slightly less efficient, but also demonstrated a similar “off/on” stabilization effect with a threshold concentration of ∼0.4 μM. Of note, a similar threshold‐like effect was recently observed for PhenDC3 and, to a lesser extent, PDC in fluorescence melting experiments with another G4 target. ^[57] With 22CTA and myc22, the concentration dependence of the stabilization effect was less steep for all ligands. In all cases, the stabilization effect observed with 1 μM of ligand matched the value observed in the screening experiments (“Screen” in Figure 3), thus giving additional proof that this experiment has an excellent tolerance with respect to minor differences in ligand concentration and the presence of potential impurities resulting from the synthesis procedure. Altogether, a clear ranking of ligands could be observed with all four G4‐DNA: 3Ei≥PhenDC3>3Ef>3Ab>3Cb; the superior activity of compound 3Ei, compared to PhenDC3, was most evident in the case of 22CTA (Figure 3).

Figure 3.

Concentration‐dependent stabilization of G4‐DNA observed in fluorescence‐melting experiments: a) 25TAG; b) 22CTA; c) myc22; d) Pu24T. Conditions: c(G4‐DNA)=0.2 μM, buffers: see Table S3. Data are means±s.d. from three technical replicates. The upper limit of ΔT _m detection is 36 °C (T=96 °C). The ΔT _m values observed in screening experiments with “as‐synthesized” ligands (Screen) are plotted for comparison.

In order to exclude the possible interference of ligands with the fluorophores inherent to the fluorescence‐melting assay, we additionally performed UV‐melting experiments by monitoring the G4‐characteristic melting transition at 295 nm ^[58] in the absence and in the presence of 1 and 2 molar equivalents of ligands (Figure S6). Two additional G4‐DNA structures, namely 24TTG (a variant of human telomeric sequence forming a hybrid‐1 structure in K⁺ conditions) and hTel21T₁₈ (a variant forming a stable chair‐like, three‐layer antiparallel structure; see Table S2) were included in this assay. Remarkably, the melting curves observed in the presence of ligands were often biphasic or featured the absence of a transition below 96 °C, in particular in the case of 3Ef and 3Ei, precluding an accurate determination of ligand‐induced stabilization effects (ΔT _m). This may be due to excessive stabilization of quadruplex structures by these ligands, as the concentrations employed in UV‐melting experiments (3 and 6 μM) are higher than the highest concentration used in the fluorescence‐melting assay (1 μM). Nonetheless, whenever ΔT _m values could be calculated (Table S6), they clearly demonstrated a remarkably efficient stabilization of most G4s (e. g., for 22CTA, ΔT _m=37 and 25 °C in the presence of 2 molar equivalents of 3Ab and 3Cb, respectively). None of the ligands had a significant effect on the melting of the hairpin hp2 (Figure S6, Table S6).

Next, the G4‐binding properties of these ligands were evaluated by the fluorescent indicator displacement (FID) assay, which assesses the capacity of ligands to displace Thiazole Orange (TO), a probe that is dark in the unbound state but fluorescent in the complex with G4‐ or duplex DNA. ^[59] In this assay, the capacity of ligands to bind to G4‐DNA is quantified by DC₅₀ values (i. e., the concentration required to decrease the fluorescence of G4‐bound TO by half). Notably, the results (Figure 4) demonstrate that the relative capacity of ligands to displace the TO probe was different with respect to G4‐stabilization effects observed in melting experiments. Specifically, pyridine and pyrazine derivatives (3Ab and 3Cb, respectively) were systematically more active in the FID assay than phenanthroline derivatives (3Ef, 3Ei, and PhenDC3). A similar discrepancy between the results of FRET‐melting and fluorescent probe‐displacement assays was already observed upon binding of several bis(acylhydrazone) ligands to G4‐RNA ^[46] as well as with other ligands. ^[60] This behavior can be attributed to the formation of ternary complexes (TO‐G4‐ligand, evidenced by native mass spectrometry) instead of, or in parallel to the displacement of the probe. ^[60] Interestingly, the results shown in Figure 4 point to an intriguing selectivity of compound 3Ei towards antiparallel G4‐DNA 22CTA (DC₅₀=0.23 μM), followed by hTel21T₁₈ and 24TTG (DC₅₀=0.35 and 0.31 μM, respectively), which is in line with high thermal stabilization of 22CTA by this compound observed in melting experiments (Figure 3b). It should be noted, however, that DC₅₀ values lower or too close to the concentration of the oligonucleotide employed in the FID assay (i. e., 0.25 μM) should be treated with caution and interpreted merely as an evidence of high affinity of the ligand (K _d≤0.25 μM). Finally, none of the five compounds induced significant displacement of the probe from the hairpin hp2 at concentrations <2.5 μM (Figure 4).

Figure 4.

Ligand‐induced displacement of TO (0.5 μM) from G4‐DNA (0.25 μM). Data are presented as DC₅₀ values (i. e., the concentration of the ligand required to displace 50 % of TO, mean±s.d.) obtained from three technical replicates. The experiments were performed in K100 buffer (10 mM LiAsO₂Me₂, 100 mM KCl, 1 % v/v DMSO). Note that color scale is not linear.

Circular dichroism (CD) spectroscopy

The binding of ligands to G‐quadruplexes may induce conformational changes of polymorphic structures at temperatures far below denaturation, most likely due to a particularly high affinity of the ligand towards one or another quadruplex fold (including minor or transient conformations) that drives a conformational selection towards these structures.[ ⁶¹ , ⁶² , ⁶³ , ⁶⁴ ] The best‐known example of this transformation is the shift from hybrid conformations adopted by the human telomeric sequence in K⁺‐rich conditions to an antiparallel form, induced by bis(quinolinium) carboxamide ligands such as PDC (360A), 3AQN, PhenDC3, and their derivatives,[ ³⁸ , ⁴⁴ , ⁵⁶ , ⁶² , ⁶⁵ , ⁶⁶ ] whereas a shift toward a parallel form was observed with several other ligands, including phenanthroline derivatives devoid of carboxamide groups.[ ⁶⁷ , ⁶⁸ ] During the revision of this manuscript, structural details of the PhenDC3‐induced conversion of a hybrid‐1 structure (similar to the one formed by 24TTG) into an antiparallel chair‐like fold, namely ligand insertion between two G‐quartets and ejection of one K⁺ ion, became available and finally shed light on this intriguing phenomenon. ^[64] To evaluate the effect of bis(acylhydrazone) ligands on the conformation of G4‐DNA, we recorded CD spectra of unlabeled oligonucleotides in the absence and the presence of 0.5, 1, and 2 molar equivalents of selected ligands in K⁺‐rich conditions (K100 buffer, 100 mM K⁺). The results demonstrate that all tested ligands induced changes in CD spectra of 25TAG, notably a slight increase of the positive peak at 295 nm and an appearance of a negative peak at 260 nm and a positive peak at 240 nm (Figure 5a), consistent with a conversion from hybrid‐2 structure adopted by this sequence in the absence of ligands into an antiparallel form. Of note, this transformation was complete with compounds 3Ab and 3Cb, but only partial with compounds 3Ef and 3Ei. The situation was different with 24TTG whose CD spectra remained consistent with the hybrid‐1 structure in the presence of 3Ab, 3Cb, and 3Ef; only the presence of compound 3Ei led to changes in CD spectra, similar to the ones described above and consistent with the formation of an antiparallel form (Figure 5b). In the case of antiparallel structures 22CTA (Figure 5c) and, in particular hTel21T₁₈ (Figure S7a), addition of ligands 3Ab and 3Cb led to a further increase of CD signals characteristic of the antiparallel G4‐DNA, namely the negative peak at 260 nm and a positive peak at 295 nm; the effect of compounds 3Ef and 3Ei was much less pronounced. Altogether, these results are in line with the phenomena observed with 25TAG and indicate that the ligands are able, to a various extent, induce and/or stabilize antiparallel quadruplex conformations. Finally, CD spectra of parallel G4 myc22 and Pu24T were essentially unaffected in the presence of the ligands, indicating the absence of topology changes (Figure S7b, c).

Figure 5.

CD spectra of a) 25TAG, b) 24TTG and c) 22CTA in the absence (dark green) and in the presence of 0.5 (yellow), 1 (orange), and 2 (red) molar equivalents of the indicated ligands. The arrows indicate the ligand‐induced changes in CD spectra. c(G4‐DNA)=3 μM in K100 buffer.

The capacity of ligands to induce G4 structures from unfolded oligonucleotides was also investigated. In Bis‐Tris buffer devoid of metal cations, CD spectra of all oligonucleotides showed a weak positive band at 260 nm, characteristic of a random‐coil conformation; in addition, an even weaker positive band at 290 nm, indicative of the presence of a minor amount of an antiparallel G4 structure, was observed in the case of 22CTA and hTel21T₁₈ (Figure S8c, d). Addition of the ligands resulted in drastic, dose‐depended changes in CD spectra, observed at ligand concentrations as low as 1.5 μM (i. e., 0.5 molar equivalents). Specifically, in the case of 25TAG, 24TTG, 22CTA and hTel21T₁₈ , these changes were fully consistent with a ligand‐induced antiparallel G4 conformation, characterized by positive peaks at 295 and 240 nm and a negative peak at 260 nm, and with a relative efficacy 3Ab≈3Cb≈3Ei>3Ef (Figure S8a–d). In the case of myc22, all ligands induced, with roughly the same efficacy, a parallel G4 structure evidenced by a strong positive CD peak at 260 nm, (Figure S8e). However, in the case of Pu24T, the induction of a parallel G4 structure was marginal for all ligands (Figure S8f). Interestingly, this quadruplex is characterized by an exceptionally high cooperativity of K⁺ binding, requiring two cations for the folding. ^[69] This might explain its reluctance to folding in the absence of K⁺ cations: while most other sequences are able to fold in the presence of a single K⁺ cation (which is emulated by the ligand in these conditions), Pu24T can accommodate only one ligand in its binding site (see below) which is not sufficient for proper folding of this quadruplex. Altogether, these results highlight the preferential binding of ligands to antiparallel G4 structures, provided they can be formed by a given DNA strand.

Mass spectrometry (MS) of G4‐DNA‐ligand complexes

Native electrospray (ESI) MS of DNA‐ligand complexes is a sensitive method that allows the monitoring of G4 denaturation ^[70] or conformational changes such as the conversion from a three‐quartet to a two‐quartet structure, evidenced by the ejection of one of the two K⁺ cations.[ ⁶² , ⁶⁴ , ⁶⁵ ] We recorded mass spectra of five G4‐DNA oligonucleotides previously validated in ESI‐MS‐compatible conditions (24TTG, 25TAG, 22CTA, myc22, and Pu24T) ^[71] in the presence of selected compounds (Figures 6 and S9–S13). The spectra showed the formation of 1 : 1 and, in most cases, 1 : 2 (G4 : L) complexes in different ratios. Interestingly, in the case of hybrid G4 structures (24TTG and 25TAG), the formation of complexes with all studied ligands was accompanied by ejection of one of the two K⁺ cations initially present in the predominant structure [G4+2K], as evidenced by the appearance of the peaks of complexes [G4+1L+1K] and [G4+2L+1K], in addition to minor peaks of the complexes that maintained two K⁺ ions, [G4+1L+2K] and [G4+2L+2K] (Figures 6a, b, S9, and S10). Interestingly, in the case of 25TAG, the ligands differ in their capacity to eject the K⁺ cation and thus to induce a conformational change of this structure, a fact that can be illustrated by a comparison of relative intensities of the peaks of 1 : 1 complexes [G4+1L+1K] and [G4+1L+2K], with a following trend: PhenCD3≈3Ab≈3Cb>3Ei>3Ef (Figure 7a). In contrast, no K⁺ ejection was observed with 22CTA; instead, the addition of ligands led to the disappearance of the presumably unfolded oligonucleotide species [G4+0K] and the formation of complexes predominantly harboring one K⁺ ion (Figure 7c). Parallel quadruplexes myc22 and Pu24T maintained two K⁺ ions each upon formation of complexes with all ligands (Figures S11–S13). Altogether, this is reminiscent of the behavior previously observed with PhenDC3, PDC (360A), and pyridostatin (PDS). ^[65] Taking into account the recently disclosed structural details of the conformational change of the hybrid‐1 structure induced by PhenDC3 and accompanied by K⁺ ejection, ^[64] this seems to be a general feature of V‐ or U‐shaped bisquinolinium cationic ligands, capable of an insertion between G‐quartets.

Figure 6.

Representative ESI mass spectra of a) 25TAG, b) 24TTG and c) 22CTA in the absence (top) and presence of equimolar concentrations of ligands 3Ab (middle) and 3Cb (bottom). c(G4)=5 μM in 100 mM TMAA, 1 mM KCl buffer. c.p.=calibration peak (Thermo Tune Mix).

Figure 7.

Relative abundances of K⁺ adducts in mass spectra of a) 25TAG, b) 24TTG and c) 22CTA in the absence of ligands (black) and in 1 : 1 complexes [G4+1L+nK] with 3Ab, 3Cb, 3Ef, 3Ei and PhenDC3 (red). Experimental conditions are identical to those in Figure 6.

The relative intensities of molecular peaks of the unbound and ligand‐bound species were exploited to obtain the values of apparent dissociation constants (K _D) for 1 : 1 and 1 : 2 complexes (Table 1). ^[72] Of note, the strong binding of PhenDC3 to telomeric sequences (25TAG: K _D1<1 nM, 22CTA: K _D1=2.1 nM) observed in MS experiments is consistent with the previously reported results of SPR studies (K _D=2.4 nM). ^[41] Likewise, bis(acylhydrazones) 3Ab, 3Cb, and 3Ei were also found to tightly bind to telomeric G4 sequences, in particular to hybrid‐2 (25TAG) and antiparallel (22CTA) structures with low‐nanomolar K _D1 values. Unfortunately, with these structures, the K _D1 values for several ligands could not be determined with satisfactory accuracy (Table 1) due to essentially complete binding at the employed concentrations (5 μM), resulting in a disappearance of the peak of unbound DNA (Figures S9 and S11). Conversely, binding to the telomeric hybrid‐1 (24TTG) and parallel G4 structures (Pu24T, myc22) was characterized by higher K _D1 values that, however, were all in the sub‐micromolar range. Importantly, this binding preference is in line with the results of CD spectrometry (Figure 5): thus, compounds 3Ab and 3Cb that demonstrated the highest affinity to 25TAG and 22CTA in MS experiments were also most efficient in changing the conformation of 25TAG towards the antiparallel form, whereas compound 3Ef characterized by a lower affinity for these structures (22CTA: 54 nM, 25TAG: 80 nM) was much less efficient in inducing the conformational change. Altogether, these observations confirm the hypothesis that the conformational change observed with 3Ab, 3Cb, PhenDC3, and similar ligands stems from their particularly high affinity to the antiparallel (22CTA‐like) form of the telomeric sequence. Finally, bis(acylhydrazones) 3Ab, 3Cb, and 3Ef were significantly better than PhenDC3 as ligands for parallel G4 structures, which is also consistent with the results of the FID assay (Figure 4).

Table 1.

Apparent dissociation constants for 1 : 1 and 1 : 2 complexes (K _D1 and K _D2 [nM]), obtained from ESI‐MS of G4‐ligand mixtures.^[a]

Ligand		25TAG	24TTG	22CTA	myc22	Pu24T
3Ab	K D1 K D2	n.a.[b]	4.4 7900	<1 4700	53 12 000	81 28 000
3Cb	K D1 K D2	<1 2 100	16 63 000	<1 3200	120 50 000	75 36 000
3Ef	K D1 K D2	80 3 900	420 35 000	54 17 000	210 n.d.[c]	58 n.d.[c]
3Ei	K D1 K D2	n.a.[b]	310 2100	4.9 1600	580 24 000	300 95 000
PhenDC3	K D1 K D2	<1 n.d.[c]	22 7600	2.1 n.d.[c]	570 24 000	570 41 000

[a] Conditions: c(G4‐DNA)=c(ligand)=5 μM in 100 mM TMAA, 1 mM KCl buffer. dT₆ (1 μM) was used as internal reference. [b] Not available: the K _D values could not be determined since the concentration of free ligand [L] (Eq. (4), Experimental Section) yielded a negative value due to experimental artifacts. [c] Not detectable: the peak of [G4+2L] complex was not observed.

Fluorescence titrations

The binding of ligands to fluorophore‐labeled G‐quadruplexes may induce dose‐dependent quenching of the fluorophore's emission, which can be exploited to obtain the values of binding constants in isothermal conditions.[ ⁷³ , ⁷⁴ ] In this assay, we used four Cy5‐labeled G4‐forming sequences: 5′‐Cy5‐25TAG, 5′‐Cy5‐22CTA, 5′‐Cy5‐Pu24T, and 5′‐Cy5‐myc22 (Table S2). Pre‐folded oligonucleotides were mixed with serial dilutions of ligands (c=1.8 nM to 10 μM) and the fluorescence intensity of Cy5 was measured; the corresponding dose‐dependent fluorescence quenching curves are shown in Figure S14. The first transitions in the binding curves were used to obtain the values of dissociation constants through fitting to a 1 : 1 equilibrium (Table 2); of note, several ligands, in particular, PhenDC3 and 3Ef gave two‐stage titration curves, giving evidence of multi‐stage binding processes and/or non‐specific quenching at high concentrations (10 μM). However, none of the ligands induced the quenching of the single‐stranded hairpin oligonucleotide (5′‐Cy5‐hp2) at concentrations lower than 5–10 μM (Figure S14), demonstrating the absence of non‐specific quenching effects at low concentrations.

Table 2.

Dissociation constants (K _d [nM]) of ligands to G4‐DNA determined from fluorimetric titrations with 5′‐Cy5‐labeled oligonucleotides.^[a]

Ligand	25TAG	22CTA	myc22	Pu24T
3Ab	11.3±1.0	8.3±0.6	11.2±2.2	7.5±1.0
3Cb	23.8±1.8	19.1±2.6	13.9±1.4	8.5±1.1
3Ef	9.2±1.0	19.7±3.2	18±26	14.8±3.8
3Ei	>10 000	>10 000	∼5500	∼2700
PhenDC3	3.1±0.2	16.8±9.4	8.7±4.0	6.7±1.6

[a] Conditions: c(5′‐Cy5‐G4)=2 nM, c(ligand)=1.8 nM to 10 μM, buffer: 10 mM LiAsO₂Me₂, 100 mM KCl, 0.5 w/v % CHAPS, 0.05 v/v % Triton X‐100, pH 7.2. Data are means±s.d. from three independent titrations fitted to a 1 : 1 model (Experimental Part, Eq. (7)).

Among the tested ligands, 3Ab, 3Cb, and 3Ef showed high affinity to all G‐quadruplexes (K _d<30 nM), similar to PhenDC3. While found in the same range, these values did not totally match the K _D values obtained from mass‐spectrometric experiments; in addition, this assay was not able to reproduce the preferential binding of 3Ab, 3Cb, and PhenDC3 to 25TAG and 22CTA suggested by CD spectroscopy and ESI‐MS experiments, as the K _d values of these ligands for Pu24T and myc22 were found in the same range (7.5–13.9 nM). Most strikingly, compound 3Ei did not induce fluorescence quenching of any of the labeled G4‐DNA (including 22CTA) at concentrations lower than 3–10 μM, which is surprising considering its high affinity evidenced by all other techniques. To explain this discrepancy, several hypotheses can be considered. First, it can be proposed that, with all G4 structures, 3Ei binds to the 3′ quartet, that is, opposite to the fluorophore attachment site (5′); this precludes fluorescence quenching as originally suggested by the authors of this method.. ^[73] However, in the case of Pu24T, this would imply that the binding occurs with the opening of the “bottom” (3′) G‐quartet that is otherwise hindered by the diagonal snap‐back loop (cf. Figure 2a), which is not consistent with the results of molecular modeling showing that 3Ei and PhenDC3 occupy similar binding sites (vide infra). The second hypothesis suggests that inefficient fluorescence quenching in the case of 3Ei might be due to a difference in redox potentials of the N‐benzylpyridinium groups in 3Ei versus N‐methylquinolinium groups in 3Ab, 3Cb, 3Ef, and PhenDC3. As the quenching of the reporter fluorophore by the ligand is most plausibly due to the oxidative photoinduced electron transfer (PeT) process,[ ³⁸ , ⁷⁵ ] ligands with low reduction potential would perform as poor quenchers in this assay; thus, it is possible that the reduction potential (E _red) of N‐benzylpyridinium substituents in 3Ei is not sufficient to quench the fluorescence of Cy5. Indeed, a simple comparison of reduction potentials of isolated N‐methylpyridinium (E _red=−1.32 V) and N‐methylquinolinium cations (E _red=−0.96 V) ^[76] gives evidence that oxidative PeT is less energetically favorable with N‐methylpyridinium cation as a quencher (by 8.3 kcal mol⁻¹, all other parameters being equal). Finally, the third hypothesis suggests that the structure of 3Ei in the complex might not be favorable for the efficient quenching of Cy5, in particular due to unfavorable distance between the quaternary heterocyclic units of the ligand and the fluorophore; this was further elaborated using molecular modeling.

Molecular modeling

In order to get insight into the high affinity of 3Ei to most G4 structures as well as the inconsistencies observed in fluorimetric titrations described above, we performed molecular docking of selected ligands into two G4 receptors whose structures, in the complex with PhenDC3, were recently determined experimentally. Among these, the telomeric sequence 23TAG (for the sequence, see Table S2) adopts a hybrid‐1 conformation in the absence of ligand, whereas the binding of PhenDC3 induces a conformational switch into a chair‐like antiparallel structure with the insertion of the ligand between two G‐quartets (PDB ID: 7Z9L). ^[64] Conversely, the snap‐back structure adopted by Pu24T maintains its parallel conformation upon binding of PhenDC3 (PBD ID: 2MGN). ^[77] The validity of the docking protocol was initially validated by redocking the original ligand (PhenDC3). In the case of the antiparallel structure, the pose obtained in silico was virtually indistinguishable from the experimentally observed structure (Figure S15a); in the case of Pu24T, the ligand underwent a conformational change due to the rotation of one carboxamide group, but nonetheless adopted a pose globally similar to the original structure (Figure S15b) giving evidence of a valid docking protocol. Next, ligands 3Ab, 3Ef and 3Ei were docked into the two receptors. Both 3Ab and 3Ef adopted the poses similar to the ones observed with the original ligand, i. e., insertion between two G‐quartets in the antiparallel structure and stacking over the top G‐quartet in Pu24T, driven mostly by π‐stacking interactions (Figure 8a, b). However, in the case of 3Ei, the positively charged pyridinium substituents were located in the grooves of G4 structures, almost perpendicular to the G4‐quartet and the ligand's plane (Figure 8c), where they contribute to additional stabilization of the complex due to electrostatic interactions with the phosphate groups. This interaction, particularly efficient in the complex with the antiparallel structure adopted by the telomeric sequence, sheds light on the exceptionally high stabilization of telomeric sequences (25TAG and 22CTA) observed in melting experiments. Finally, the results of molecular modeling indicate that all ligands occupy similar poses with respect to the distance to the 5′ terminus which is labeled with the fluorophore in fluorescence titrations, and the differences in ligand poses are therefore unlikely to explain the poor fluorescence quenching observed with 3Ei. Thus, the origin of this phenomenon is likely to stem from the difference of redox potentials of the ligands as discussed above.

Figure 8.

Representative docking poses of a) 3Ab, b) 3Ef and c) 3Ei bound to (left) the antiparallel G4 structure adopted by telomeric sequence 23TAG in the presence of PhenDC3 (from PDB ID: 7Z9L), ^[64] and (right) the snap‐back parallel structure adopted by Pu24T (from PDB ID: 2MGN). ^[77]

Cytotoxic effects in human cancer cells

The cytotoxic effects of four bis(acylhydrazones) (3Ab, 3Cb, 3Ef, and 3Ei) were tested in two human cancer cell lines, namely multidrug‐resistant glioblastoma line T98G ^[78] and cisplatin‐resistant non‐small‐cell lung carcinoma line A549, as well as noncancerous lung fibroblast cells MRC‐5, and compared with that of PhenDC3. The results (Table 3; for growth inhibition curves, see Figure S16) show that bis(acylhydrazone) derivatives were not toxic in T98G cells at concentrations of up to ∼10 μM, whereas PhenDC3 showed significant cytotoxicity (GI₅₀=2.9 μM upon 96 h of incubation). These results are in line with the earlier data indicating much lower toxicity of compounds 3Ab (PyDH2) and 3Eb (PhenDH2) in H1299 (lung carcinoma) cells, compared with PhenDC3. ^[46] In A549 cells, a clear difference was observed between, on one hand, pyridine (3Ab) and pyrimidine (3Cb) derivatives that showed low cytotoxicity (GI₅₀=57.1 and 64.5 μM, respectively), and on the other hand, phenanthroline derivatives (3Ef, 3Ei, and PhenDC3) that demonstrated GI₅₀ values in the low‐micromolar range, with 3Ef being the most cytotoxic compound (GI₅₀=0.4 μM). Finally, all compounds showed rather low cytotoxicity in non‐cancerous MRC‐5 cells (GI₅₀>25–50 μM). Altogether, the selective toxicity of compound 3Ef in A549 cells (with a selectivity index of ∼67 with respect to MRC‐5 cells) represents a certain therapeutic interest and warrants further investigations, including the profiling of this compound in other cell lines and investigation of the mechanism of cell death. Considering the fact that the G4‐binding properties of bis(acylhydrazones) are quite close to those of PhenDC3, the differential toxicity profiles might also be attributable to non‐G4‐related cellular effects; moreover, they may also arise from the differences in cellular uptake of compounds by various cell lines. Finally, the low toxicity of bis(acylhydrazones) in non‐cancerous cells may be exploited in applications where direct cell killing is not desired, such as G4‐mediated immunotherapy.

Table 3.

Cytotoxicity of selected bis(acylhydrazones) and PhenDC3 in T98G, A549 and MRC‐5 cell lines after 96 h of incubation.^[a]

Compound	GI50 [μM], in a cell line
	T98G	A549	MRC‐5
3Ab (PyDH2)	17.8±0.2	57.1±6.5	24.8±1.9
3Cb	12.0±0.5	64.5±2.9	52.3±15.0
3Ef	11.2±1.4	0.4±0.06	26.7±2.4
3Ei	29.7±2.0	3.2±0.6	59.1±5.5
PhenDC3	2.9±0.2	6.6±0.6	34.8±5.0

[a] GI₅₀ is compound concentration inhibiting 50 % of cell growth, determined using CellTiter Glo® cell viability assay. Data are means±s.d. from three biological replicates. Compound vehicle: 0.5 % (v/v) DMSO.

Conclusions

In this work, we have developed a simple method for the rapid synthesis of “ready‐to‐screen” G4‐targeted libraries of cationic bis(acylhydrazones). Specifically, heating mixtures of aldehyde and N‐acylhydrazide building blocks in the presence of a catalyst (AcOH) gives ligands as DMSO solutions with acceptable purity (87 % on average) and approximately 2 mM concentration. Provided both the catalyst (after neutralization by dilution with a suitable buffer) and the reaction solvent are rather inert in biophysical experiments, these libraries can be used directly, that is, without any purification step, in multiple parallel screening assays to estimate their capacity to stabilize various G‐quadruplexes (or any other nucleic acid structures). Furthermore, given the non‐interfering nature of the catalyst and the solvent, these libraries can potentially be used directly in various biological screening experiments.

Screening a library of 90 bis(acylhydrazones) by fluorescence‐melting experiments performed with four G4 sequences of different topologies (and one hairpin control) allowed us to analyze their structure‐activity relationships and select three promising novel compounds for further investigation. These compounds were re‐synthesized in a preparative manner, and their interaction with targets was studied by several additional techniques. First of all, we confirmed that the ΔT _m values obtained in the fluorescence‐melting experiment performed with “as‐synthesized” compounds did not differ from the values obtained with pure samples. We also performed fluorescence‐melting experiments with lower concentrations of ligands (0.1 to 0.8 μM, compared to standard 1 μM) to study the dose‐dependent thermal stabilization of G4s by ligands in more detail; all these results confirmed the high potential of compound 3Ei because with all four G4‐DNAs and, in particular, with 22CTA, this ligand induced thermal stabilization effects comparable with, or superior to, the benchmark G4 ligand PhenDC3. The results of the FID assay, performed with six G‐quadruplexes of different topologies, also suggested the selectivity of 3Ei for the antiparallel structure adopted by 22CTA. The results of CD spectroscopy point to a conformation change of the hybrid‐2 structure (25TAG) into an antiparallel structure induced by compounds 3Ab, 3Cb, 3Ei, and, to a lesser extent, 3Ef, most likely due to insertion of the ligands between the two G‐quartets, as was recently demonstrated for PhenDC3 ^[64] and corroborated by molecular modeling. In the case of hybrid‐1 structure (24TTG), only compound 3Ei was able to induce a partial conformational change, presumably due to the higher intrinsic stability of this hybrid fold. Native mass‐spectrometric experiments confirmed the high G4 affinity of the novel compounds and corroborated the ligand‐induced conformation changes of the hybrid G4 structures (24TTG and 25TAG) induced by compounds 3Ab, 3Cb, 3Ei, and, to a lesser extent, 3Ef, evidenced in mass spectra by the ejection of one K⁺ cation upon formation of ligand‐G4 complexes. Isothermal fluorimetric titrations with Cy5‐labeled G4‐DNA showed a homogenous response of 3Ab, 3Cb, 3Ef, and PhenDC3 for all the studied G4‐sequences; however, in the case of compound 3Ei, the ligand‐induced fluorescence quenching of Cy5 was dramatically less efficient with all G4‐forming sequences. This observation gives evidence of a potential pitfall of this technique, most likely attributable to the differences in redox potentials of the ligands, resulting in inefficient fluorescence quenching despite the tight binding evidenced by all other techniques and molecular modeling. Finally, cytotoxicity studies in three cell lines demonstrated that the novel compounds have therapeutically interesting properties, in particular low cytotoxicity in noncancerous cells and selective activity of one of the compounds (3Ef) in A549 lung cancer cells, that deserve further mechanistic investigations.

Experimental Section

Synthesis of “ready‐to‐screen” samples of bis(aclylhydrazones): In glass screw‐cap vials, stock solutions of bis(acylhydrazides) 1A–J (9 mM in DMSO, 120 μL) were mixed with stock solutions of aldehydes 2 a–i (9 mM in DMSO, 270 μL) and a mixture of DMSO with glacial acetic acid (4 : 1 v/v, 150 μL), to give total reaction volumes of 540 μL each. The reaction mixtures were sealed and heated at 60 °C (drying oven) for 48 h without stirring. After cooling to room temperature, the solutions were analyzed by LC/MS (Table S1). Samples with acceptable purity (>80 %, as per LC peak area) were diluted with an appropriate buffer (Table S3) for fluorescence‐melting experiments, or frozen at −20 °C for long‐term storage.

Oligonucleotides: All DNA oligonucleotides used in this work are listed in Table S2. HPLC‐purified oligonucleotides were purchased from Eurogentec, dissolved in deionized water at a concentration of 100 μM, and stored at −20 °C when not in use.

Fluorescence‐melting experiments: Fluorimetric thermal denaturation experiments were performed with doubly labeled oligonucleotides (5′‐FAM, 3′‐TAMRA, Table S2) that were annealed (95 °C, 5 min) in the corresponding buffers (Table S3) at a concentration of 0.23 μM prior to analysis. In a 96‐well qPCR plate, the annealed oligonucleotide (22 μL) was mixed with samples of a ligand (3 μL of a 8.3 μM solution, prepared by dilution of a “ready‐to‐screen” sample with an appropriate buffer, or from stock solution of a pure ligand), to give samples with a final concentration of oligonucleotide of 0.2 μM and ligand of 1 μM, in a total volume of 25 μL per well. Thermal denaturation runs were performed with a 7900HT Fast Real‐Time qPCR instrument (Applied Biosystems) using a single heating ramp of 0.5 °C min⁻¹ and monitoring the fluorescence intensity in the FAM channel. T _m values were determined through peak analysis of first‐order derivative plots of fluorescence vs. temperature, and ligand‐induced T _m shifts (ΔT _m) were calculated as a difference of mean denaturation temperatures in the presence and in the absence of ligands, from three independent experiments.

FID assay: The unlabeled oligonucleotides were annealed (95 °C, 5 min) in K100 buffer (10 mM LiAsO₂Me₂, 100 mM KCl, pH 7.2) supplemented with 1 mM EDTA and 1 % v/v DMSO, at a concentration of 5 μM and, after cooling, supplemented with TO (10 μM). Every row of a black‐bottom, 96‐well microplate was filled with K100 buffer (q.s.p. 200 μL per well), pre‐folded oligonucleotide+TO solution (final concentrations: 0.25 and 0.5 μM, respectively), and an extemporaneously prepared ligand solution (5 μM in the same buffer; final ligand concentration: 0 to 2.5 μM). After 5 min of orbital shaking, fluorescence intensity was measured with a Fluostar Omega microplate reader (BMG Labtech) using the following parameters: 20 flashes per well, emission/excitation filters: 485/520 nm, gain adjusted at 80 % of the fluorescence from the most fluorescent well. The experiments were performed in triplicate. The percentage of TO displacement was calculated from the fluorescence intensity (F) as %(TO displacement)=100×(1−F/F ₀), where F ₀ is the fluorescence intensity of TO‐DNA complex in the absence of ligands. The percentage of displacement was plotted against the concentration of added ligand, and ligand affinity was characterized by the concentration required to decrease the fluorescence of the probe by 50 % (DC₅₀) after interpolation of the displacement curve. A control experiment was performed with the hairpin DNA oligonucleotide (hp2) in identical conditions.

CD spectroscopy: Samples for CD spectroscopy contained 3 μM oligonucleotides diluted either in K100 buffer (and subsequently annealed at 95 °C), or in 10 mM Bis‐Tris‐HCl buffer, pH 7.2 (not annealed), and supplemented with 0.5, 1, or 2 molar equiv. of ligands or an equal volume of DMSO (control). CD spectra were recorded with a Jasco J‐1500 spectropolarimeter, in quartz cuvettes with an optical path length of 0.5 cm. Parameters used for CD spectra acquisition: wavelength range, 230–330 nm; scan speed, 50 nm min⁻¹; number of averaged scans, 3; data pitch, 0.5 nm; bandwidth, 2 nm; integration time, 1 s; temperature, 22 °C. Spectra were corrected for the blank.

Mass spectrometry: Native ESI‐MS spectra of DNA and DNA‐ligand complexes were obtained using an Exactive ESI‐Orbitrap mass spectrometer (Thermo Scientific, Bremen, Germany). ESI spray voltage and capillary voltage were set to 3.50 kV and −3.5 kV, respectively. Capillary temperature was set to 275 °C. The syringe injection rate was 200 μL h⁻¹. The presented spectra (Figures S7–S11) result from 3‐min accumulations (1 scan per 1.1 s). The concentrations of the initial stock solutions of DNA (∼1 mM) were measured by UV absorbance at 260 nm. 50 μM solutions of DNA were annealed in buffer that contained 100 mM trimethylammonium acetate (TMAA, Ultra for HPLC, Fluka analytical) and 1 mM KCl (>99.999 %, Sigma) by heating at 85 °C for 10 min, followed by slow cooling to room temperature. Analyzed samples contained 5 μM of G4‐DNA, 0 or 5 μM of ligand, and 1 μM of the oligonucleotide 5′‐d(TTTTTT)‐3′ (dT₆ , internal control) in a 100 mM TMAA, 1 mM KCl buffer, and were incubated 2 h before the injection.

The charge state 6− was examined because it contained fewer nonspecific potassium adducts. Peak areas (A) of the signals of free G4, G4+1L and G4+2L complexes (Table S7) were determined assuming that their response factors are equal; this was supported by the fact that total DNA peak areas compared to the signal of dT₆ did not change significantly without or with ligand. Based on this assumption, the concentration of each species was determined from the mass balance equation and from the peak areas, disregarding any particular distribution of K⁺ adducts [Eqs. (1)–3]:

$${\displaystyle \left[\mathrm{G}4\right]{}_{\mathrm{free}}=c\left(\mathrm{G}4\right)\times \frac{A\left(\mathrm{G}4\right)}{A\left(\mathrm{G}4\right)+A(\mathrm{G}4+1\mathrm{L})+A(\mathrm{G}4+2\mathrm{L})}}$$ (1)

$${\displaystyle [\mathrm{G}4+1\mathrm{L}]=c\left(\mathrm{G}4\right)\times \frac{A(\mathrm{G}4+1\mathrm{L})}{A\left(\mathrm{G}4\right)+A(\mathrm{G}4+1\mathrm{L})+A(\mathrm{G}4+2\mathrm{L})}}$$ (2)

$${\displaystyle [\mathrm{G}4+2\mathrm{L}]=c\left(\mathrm{G}4\right)\times \frac{A(\mathrm{G}4+2\mathrm{L})}{A\left(\mathrm{G}4\right)+A(\mathrm{G}4+1\mathrm{L})+A(\mathrm{G}4+2\mathrm{L})}}$$ (3)

The concentration of free ligand, [L]_free, was determined from the mass balance equation [Eq. 4]:

$${\displaystyle \left[\mathrm{L}\right]{}_{\mathrm{free}}=\left[\mathrm{L}\right]{}_{\mathrm{tot}}-[\mathrm{G}4+1\mathrm{L}]-2[\mathrm{G}4+2\mathrm{L}]}$$ (4)

Apparent consecutive dissociation constants (K _D1 and K _D2) were calculated by using Equations (5) and 6:

$${\displaystyle K{}_{\mathrm{D}1}=\frac{{\left[\mathrm{G}4\right]}_{free}\times {\left[\mathrm{L}\right]}_{free}}{[\mathrm{G}4+1\mathrm{L}]}}$$ (5)

$${\displaystyle K{}_{\mathrm{D}2}=\frac{[\mathrm{G}4+1\mathrm{L}]\times {\left[\mathrm{L}\right]}_{free}}{[\mathrm{G}4+2\mathrm{L}]}}$$ (6)

Fluorescence titrations: Isothermal fluorescence titrations were performed as described elsewhere.[ ⁴⁹ , ⁷³ ] 5′‐Cy5‐labeled oligonucleotides were dissolved at a concentration of 1 μM in K100 buffer (Table S3), annealed at 95 °C for 5 min, slowly cooled to room temperature, and then further diluted with the titration buffer (10 mM LiAsO₂Me₂, 100 mM KCl, 0.5 w/v % CHAPS, 0.05 v/v % Triton X‐100, pH 7.2) to a final concentration of 2.22 nM. Ten serial dilutions (1 : 1) of ligands in H₂O containing 0.1 % v/v DMSO were prepared starting from two ligand solutions with initial concentrations of 100 μM and 18 μM, respectively, resulting in a total of 22 solutions with different ligand concentrations and two no‐ligand controls. The oligonucleotide solution (90 μL) and ligand solutions (10 μL) were transferred into 96‐well black, flat‐bottom polystyrene non‐binding (NBS) microplates (Corning; final oligonucleotide concentration: 2 nM, final ligand concentration: from 10 μM to 1.8 nM). After incubation for 2 h at room temperature, fluorescence intensity was measured on a CLARIOstar Plus microplate reader (BMG) using 590BP50 (excitation), LP639 (dichroic) and 675BP50 (emission) filters and an integration time of 0.5 s per well. The fluorescence intensity was normalized by dividing the raw value by the mean intensity of no‐ligand wells. The experiments were performed in triplicate. The titration curves (Figure S14) were fitted to a 1 : 1 binding model according to Equation (7) in OriginPro 2020b (OriginLab Corporation, Northampton, MA, USA).

$${\displaystyle F=F{}_0-(F{}_0-F{}_{\mathrm{b}})\times \frac{A+x/c+1-\sqrt{{A+x/c+1)}^2-4\times x/c}}{2}}$$ (7)

where F ₀ and F _b are fluorescence intensity in the absence and in the presence of saturating concentration of the ligand, respectively; x is a concentration of the ligand and A=K _d/c, where K _d is the dissociation constant and c is the concentration of labeled oligonucleotide (2 nM).

Molecular docking: The NMR‐solved structures of the TAG23‐PhenDC3 ^[64] and Pu24T‐PhenDC3 ^[77] complexes were downloaded from the Protein Data Bank (https://www.rcsb.org, PDB IDs: 7Z9L and 2MGN, respectively). Only the first frames, which presented the least NMR constraints violations, were considered for the study. The ligand was removed, and one (7Z9L) or two (2MGN) K⁺ ions (missing in the NMR‐derived structures) were positioned at the centers of mass of the eight guanine oxygen atoms of each two consecutive G‐quartets. These G4 structures were used as receptors for docking. The putative binding sites were defined where PhenDC3 was initially positioned in the PDB structures, that is, between the first (G3 ⋅ G11 ⋅ G15 ⋅ G23) and the second (G4 ⋅ G10 ⋅ G16 ⋅ G22) G‐quartets (7Z9L), or above the 5′ (G4 ⋅ G8 ⋅ G13 ⋅ G17) G‐quartet (2MGN). To delimit the binding site, the nucleotides at 5 Å around PhenDC3 were selected, and nucleotides beyond this distance were added if necessary to guarantee the continuity of the site. Based on this selection, two cubic boxes were created using Glide package of the Schrödinger suite (version 12.2.012):[ ⁷⁹ , ⁸⁰ , ⁸¹ ] one of 10 Å to determine the space where the center of the ligand will be created, and a bigger one of 30 Å to determine the space where the ligand would be able to move. The two boxes were placed in the center of the binding site. The structures of the ligands were prepared with the LigPrep package from the same Schrödinger suite using the default force field (OPLS 3e) and keeping the charge state (2+) for all ligands. Then Glide was used for docking, with the default parameters. For each ligand, 10 poses were generated and analyzed. Molecular graphics were generated with UCSF Chimera. ^[82]

Cytotoxicity assay: Cytotoxicity of compound was investigated at CIBI platform (ICSN, Gif‐sur‐Yvette, France). Adherent T98G, A549 and MRC5 cells were grown in Minimum Essential Medium (MEM (1X)+GlutaMAX™, Life Technologies) supplemented with 1 % penicillin/streptomycin (Life Technologies), 10 % fetal bovine serum (Biowest) and 1 % MEM non‐essential amino acids (MEM NEAA (100X), Life Technologies), kept in humidified atmosphere with 5 % CO₂ in air at 37 °C and subcultured twice a week by dispersal with TrypLE™ Express Enzyme (Life Technologies). For growth inhibition experiments, cells were seeded at a 0.125–0.25×10⁵ cells mL⁻¹ in a 25 cm² flask 3–4 days before the experiment in the conditions mentioned above. After reaching this time, media was removed and the cells were washed with 5 mL of sterile PBS. For each flask, 100 μL of TrypLE™ Express Enzyme (Life Technologies) was added to the flask for 5–7 min at 37 °C, followed by 900 μL of medium. Cells were resuspended in media at a concentration of 0.125×10⁵ cells mL⁻¹. 200 μL of this suspension was added per well on a 96 well plate. After 4 h, 1 μL of compound dissolved in DMSO solutions were added to obtain a 25 to 0.015 μM range concentrations (concentration of DMSO per well: 0.5 % v/v). The cells were incubated for 96 h in the conditions mentioned above. Cell viability was assessed by using CellTiter Glo® luminescent cell viability assay as per kit instructions. Cells’ viability (V, %) was determined as the ratio of the luminescent signal of the cells treated with the compound(s) and the untreated control (DMSO, 0.5 % v/v). Each condition was assayed at least three times with two technical replicate per plate. GI₅₀ values were obtained by a nonlinear fitting using a dose response curve by plotting the normalized response vs. log c(ligand) in GraphPad Prism 6 (Figure S16).

Conflict of interest

The authors declare no conflict of interest.

Supporting information

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

Supporting Information

Reznichenko O., Leclercq D., Franco Pinto J., Mouawad L., Gabelica V., Granzhan A., Chem. Eur. J. 2023, 29, e202202427.

Data Availability Statement

The data that support the findings of this study are available in the supplementary material of this article.