Introduction

Nucleic acids preserve genetic information to form canonical duplexes with Watson–Crick base pairings. Alternatively, non-canonical structures exist as other structures via non-Watson–Crick base pairs, such as triplexes and tetraplexes, including the guanine quadruplex (G4) and i-motif1,2,3. I-motif is one of the tetraplexes formed in cytosine (C) -rich sequences and comprises a pair of parallel duplexes with intercalating hemi-protonated C:C+ base pairs4. As the pKa value of deoxycytidine is about 4.3, i-motif formation is usually favoured in mildly acidic conditions, although some i-motif structures can exist at neutral pH5, 6. I-motif formations are also stabilised by molecular crowding conditions7,8,9,10. Because the intracellular state is highly crowded with biomolecules, metabolites11, i-motif stability is higher in cells than in vitro12, 13. Moreover, the i-motif formations in human cells depend on their cell cycles, suggesting that i-motifs might have regulatory roles in cells14. The potential i-motif forming sequences are frequently found within gene promoter regions15. Some i-motifs on cancer and neurodegenerative disease genes, such as BCL2, KRAS, HRAS, and VEGF, were identified to regulate their gene expressions16,17,18,19. Therefore, detecting and regulating a certain i-motif in various crowding conditions is necessary for the theranostics of cancer and neurodegenerative diseases.

The most promising approach for detecting and regulating i-motifs involves proteins or small molecules that specifically bind to a certain i-motif. Although there are several proteins that bind to C-rich sequences20,21,22,23, they do not specifically recognise i-motif DNAs over other nucleic acid structures. Therefore, the antibody for i-motifs, iMab, is the only protein capable of binding to i-motif structures, and can detect various i-motifs with high specificity and affinity over other nucleic acid structures14. The use of antibodies for i-motif detection is valuable in identifying the formation of i-motif DNAs under different intracellular conditions, such as changes in cell-cycle and pH. However, this approach is limited to visualising i-motifs without distinguishing a specific i-motif, as demonstrated by iMab's recognition capability for the i-motif scaffold, regardless of its sequence variations14. Instead, the small molecules can be utilised to detect i-motifs in living cells, and various ligands that bind to i-motifs were also reported24, 25. However, there are only a few specific ligands for certain i-motifs, such as IMC-48 for the BCL2 i-motif16, NSC309874 for the PDGFR-β i-motif26, and an acridone derivative B19 for the c-MYC i-motif27. Detection and regulation of i-motif formation by fisetin, which binds to a hairpin-like structure converted from the VEGF i-motif to emit fluorescence and regulate replication efficiency, was reported only by our group28. To develop a specific i-motif ligand for theranostic application, the ligand must generally recognise the unique structure of the i-motif and other motifs by fluorescence emission. The i-motif core is a common scaffold of C:C+ base pairs, while the loop regions connecting the core are specific for each sequence. In the loop regions, the nucleobases can interact with each other29, creating a specific site for ligand binding30. Thus, a ligand that recognises both core and loop regions should be specific for a certain i-motif. In case of G4s, specific small ligands that discriminate G4 topologies have been reported, which may interact directly with G4 loops and grooves31,32,33,34,35. To facilitate the development of theranostics for a specific i-motif, a platform ligand is required to design a dual-targeting ligand binding to both the core and loops of i-motifs.

To find the platform ligand, we focused on the ligands that exhibit very low fluorescence response in the absence of DNA but reveal significant enhancement in emission intensity upon binding with DNA. The ligand we require should interact with both the core and loop region of i-motifs. Triphenylmethane dyes, which possess propeller structure and various dialkyl amino substitutions could be useful due to their low fluorescence emission in water36. In this study, we explored the loop regions of different i-motifs and studied their interactions with different triphenylmethane dyes. Among the triphenylmethane dyes, crystal violet (CV) showed the highest fluorescence response with the i-motif from the promoter region of the human BCL2 gene, compared to that of other structures, such as other i-motifs, G4s, and hairpin. We found that CV bound to the site involving the first and third loops of the BCL2 i-motif, where the guanine bases in the loops contributed to the specific binding of CV. The complexation of the loop regions and CV repressed BCL2 gene transcription in the human breast cancer cell line of MCF-7. Our approach suggests that the loop region of i-motif can serve as a novel framework for the specific binding of small molecules and can potentially develop novel ligands to monitor and regulate the biological reactions of i-motif DNAs.

Results and discussion

Design of different i-motif-forming DNA sequences to screen the specific binding of triphenylmethane derivatives

As the region of the edge of the i-motif core is smaller than that of G4s, the ligand should be smaller than the size of a G-quartet and have multiple binding motifs to interact with both the core of the i-motif C:C+ base pair and loop regions. Therefore, triphenylmethane molecules such as pararosaniline (PR), malachite green (MG), CV, and brilliant green (BG) were chosen (Fig. 1). These derivatives have a positively charged amino group which possibly interacts with the negatively charged phosphate backbone of DNA. Furthermore, different alkyl substitutions on the amino groups also facilitate DNA binding through favourable van der Waals interactions. These molecules are known to bind to nucleic acids, including i-motif to emit fluorescence, although the fluorescence is not high37. For reference, several i-motif binding ligands, such as neutral red (NR)38, berberine (BBR)39, thioflavin T (ThT)40, and thiazole orange (TO)41 were also tested. For DNA sequences, i-motifs are usually categorised into class I motifs having loop bases less than 4 and class II motifs having loop bases greater than 442. Here, we investigated natural forming i-motif sequences of C9ORF72, human telomere region (hTelo), HRAS, ILPR, and VEGF genes from class I type and BCL2, HIF-1α, JAZF1, mid region of KRAS (KRAS mid), and PDGF-A genes from class II type (Table 1). The formation of i-motifs and/or roles in the gene expression of these sequences were identified in living human cells3, 13. In addition to these natural sequences, some additional i-motif sequences of class II having C3-tracts with varying numbers of only thymidine in loops from five to eight were also tested (Table 1). The control sequences were chosen from G4 DNA sequences having different topologies (parallel; BCL2 G4, MYC G4 and VEGF G4, hybrid; human telomere G4, and anti-parallel; thrombin binding aptamer (TBA)), and hairpin DNA having a 9 mers stem (Table 1). All sequences formed i-motifs, G4s with different topologies, and hairpins, respectively, as expected in the solution containing 10 mM KH2PO4, 50 mM KCl, and 1 mM K2EDTA with pH 5.0 at 25 °C (Fig. S1).

Figure 1
figure 1

Chemical structures of the fluorescent ligands used in this study.

Full size image
Table 1 DNA sequences used in this study (5ʹ–3ʹ).
Full size table

Specific fluorescence emission upon binding of crystal violet to BCL2 iM

The fluorescence response of triphenylmethane derivatives in the presence of different i-motifs (shorter loop: class I, and longer loop: class II), G4s, and hairpin DNA sequences was tested at pH 5.0 using 6.0 µM of each ligand with 15 µM of DNA at 25 °C to ensure 1:1 ligand-DNA binding. In the presence of CV, class I i-motifs showed a slight increase in fluorescence (Fig. 2), and the fluorescence level of CV (F/F0; F = fluorescence intensity of CV in the presence of DNA and F0 = fluorescence intensity of CV without DNA) with these sequences were less than 5, whereas a significant fluorescence increase was observed in the presence of the BCL2 i-motif DNA (BCL2 iM), a class II i-motif (Fig. 2). It has been identified that BCL2 iM has three long loops (5′ loop; GCTCCCGC, central loop; TTCCT and 3′ loop; GCGCCCG, Fig. 3A)16, 43, 44. The F/F0 of CV was about 300-fold with BCL2 iM. Compared with the BCL2 iM, the fluorescence level with other sequences was less significant. In the Class II group, KRAS mid iM, and its core C-rich region (KRAS mid I–IV iM) exhibited relatively high fluorescence (F/F0 = 78.5 and 24.3, respectively). The model sequences also showed relatively high fluorescence and among them, C3T777 iM showed the largest response (F/F0 = 70.5). However, other class II i-motifs (HIF1-α, JAZF1, and PDGF-A) showed negligible fluorescence (F/F0 values were below 6). These results indicate that both the length and sequences of loop regions of the i-motif had an important role in CV binding and its fluorescence emission. For G4 structures, the fluorescence level of CV with hTelo G4 (F/F0 = 53.0) and BCL2 G4 (F/F0 = 51.4) were relatively high, although the fluorescence enhancement was much smaller than that with BCL2 iM (F/F0 = 304.3), demonstrating higher selectivity of CV for BCL2 iM compared with the selectivity of CV for hTelo G4, as we previously reported45. Moreover, the hairpin DNA exhibited almost no change in fluorescence level. The BCL2 iM fluorescence was 130-fold higher than that of the hairpin DNA. Previously, C29 iM, a model i-motif sequence having five C:C+ base pairs and the loops containing three thymidines exhibited CV fluorescence37 and a specific aptamer, CV30S, for CV was developed (Table 1)46. In both the cases, the F/F0 values were 2.0 and 18.2 (Fig. S2), revealing much lower fluorescence enhancement when compared to that of BCL2 iM. Extending the fluorescence assay experiment to a neutral pH condition (pH  7.0 at 25 °C), we observed a significant reduction in the fluorescence of CV in the presence of i-motif forming sequences, particularly with BCL2 iM (Fig. S3). However, no significant change was observed with G4 and hairpin forming sequences, indicating that the fluorescence properties of CV is negligibly affected by the pH of the medium. Therefore, the fluorescence experiment in a neutral pH environment demonstrates that the specific recognition of BCL2 iM by CV is primarily determined by the secondary structure of the DNA rather than its primary sequence. To further demonstrate the significance of DNA secondary structures, we investigated the fluorescence response of CV in the presence of 50 mM LiCl instead of KCl at pH 8.0 at 25 °C. Under these conditions, iM and G4 forming DNA cannot form tetraplexes, and thus they serve as unstructured controls for C-rich non-iM forming DNA and G-rich non-G4 DNA, respectively. The fluorescence response of CV upon binding to the BCL2 iM and G-rich non-G4 sequence (H3, Table 1) significantly decreased at pH 8.0 at 25 °C in the presence of LiCl, as compared to the results shown in Fig. 2 (Fig. S4). Moreover, CV did not exhibit a noteworthy fluorescence response (Fig. S4) in the presence of the randomly composed 22-mer DNA (R1, Table 1) under the same conditions. Thus, we established the significance of the DNA secondary structure in determining the fluorescence response of CV.

Figure 2
figure 2

Fluorescence responses of CV (blue), MG (yellow), BG (green) and PR (red) with different DNA motifs. The data was relative to no DNA as the control. The errors were within 5%.

Full size image
Figure 3
figure 3

(A) Schematic illustration of the BCL2 iM structure. Fluorescence responses of CV with (B) the wild type (WT) BCL2 iM sequence and its abasic loop sequences in the 5′ loop (5′L-A), central loop (CL-A), and 3′ loop (3′L-A). (C) Systematic mutations in the 5′ loop, and (D) systematic mutations in the 3′ loop. CV in only buffer was used as the control.

Full size image

To understand the importance of the structural aspect of CV for its specificity towards BCL2 iM, we checked the fluorescence responses of its derivatives, such as MG, BG, and PR, (Fig. 2) having the same core structure of triphenylmethane but different dialkylamino substitutions. A significant decrease in the relative fluorescence response of the MG-BCL2 iM (F/F0 = 49.7) system relative to the CV-BCL2 iM (F/F0 = 304.3) system, implies that the lack of one dimethyl amino group in MG compared with CV is responsible for the lowered fluorescence response. Besides BCL2 iM, MG revealed considerable fluorescence light-up for KRAS mid iM (F/F0 = 93.3) and C3777 iM (F/F0 = 61.1) belonging to the class II i-motif category. For class I i-motifs, the fluorescence response of MG was very low (F/F0 < 8). Among the G4s, the relative fluorescence response of MG was maximum for hTelo G4 (F/F0 = 362.4), projecting MG as a specific probe for hTelo G4, which is already established47. In the case of BG, F/F0 for BCL2 iM was 3.3, which was significantly lower than that of CV. The steric hindrance owing to the presence of two bulky diethyl amino groups in BG could be responsible for the low fluorescence response of BG towards BCL2 iM as well as other i-motifs. PR containing three aniline groups showed low and comparable fluorescence response towards all DNA motifs studied here. These results indicate that CV interacted with BCL2 iM in a structure-specific manner through all its three dimethyl amino groups, and emitted high fluorescence.

To confirm whether the fluorescence of CV upon binding to BCL2 iM was due to the specificity of the ligand’s structure, we also studied the fluorescence responses of four known i-motif binding ligands NR, BBR, ThT, and TO (Fig. S5) in the presence of our studied library of different DNA motifs in pH 5.0 buffer solution at 25 °C. In the case of NR, almost no fluorescence enhancement was observed in the presence of BCL2 iM (Fig. S5A). However, in the presence of KRAS mid iM, F/F0 of NR was about 20, although those for other natural class II sequences showed only about 2.5 or less. Among the model class II sequences, C3T777 iM showed the highest enhancement (F/F0 = 9.3). However, this fluorescence enhancement was much smaller than that of CV and BCL2 iM. Other ligands did not show clear sequence specificity for fluorescence emission (Fig. S5B–D). Hence, compared with conventional i-motif binders, CV was specific for BCL2 iM, implying its sequence-specific binding nature.

Proposed binding mechanism of CV with BCL2 iM

To investigate how CV is bound to the BCL2 iM, we analysed the binding property of CV with BCL2 iM. Job’s plot analysis revealed that CV formed a 1:1 complex with the i-motif (Fig. S6A). Ensuring 1:1 complexation, the binding constant of CV with BCL2 iM was 8.0 × 104 M−1 at 25 °C from fluorescence titration (Fig. S6B). The binding constant of CV with BCL2 iM was smaller than the reported binding constant of the same with C29 iM (1.2 × 106 M−1)37. Using molecular modelling, Ma et al. demonstrated that the probable binding mode of CV with C29 iM consisting of a short loop region (TTT) was through end-stacking to C:C+ base pairs, resulting in a high binding constant. However, unfavourable entropy contribution (see the following section) due to the interaction of CV with the long loop regions might be responsible for the lowered binding constant value.

Next, we replaced the sequence of the loop regions of BCL2 iM and compared the fluorescence intensity of CV with specific replacements in the BCL2 iM to study whether CV is bound to the loop region. The replacement of the nucleotides of the first and third loops at the 5ʹ or 3ʹ end of the i-motif with the abasic nucleotides (BCL2 iM 5ʹL-A or BCL2 iM 3ʹL-A, Table S1) significantly decreased the fluorescence intensity of CV when (Fig. 3B) compared to that of the wild type (WT) BCL2 iM. However, the replacement with abasic oligonucleotides at the second loop of the central position (BCL2 iM CL-A, Table S1) still showed a similar level as that of the WT BCL2 iM. These results suggest that a single CV molecule bound to the region formed with the first and third loops through a sequence-specific manner.

Considering the binding manner studied by these assays, the tertiary structure of the loop regions was studied by molecular dynamics (MD) simulations. Because the tertiary structure of BCL2 iM has not been determined through NMR or X-ray crystallography, we utilised the solved structure of the human telomere i-motif as the initial model48. We used simulation data obtained during 100 ns simulation of three different runs and analysed the root mean square deviation (RMSD) calculated by fitting the coordinates of the C4’ atom in the core region of the C:C+ base pair and the first and third loop regions’ bases. RMSD data suggest that the core regions fluctuated less compared to the first and third loop regions (Fig. S7), because the loop regions should be more flexible than the core region. However, at the RMSD level, the first and third loop regions were in equilibrium, suggesting that the interactions form a preferred structure within the loop region. After 86 ns simulation, we sampled a snapshot of the simulated structure and found that there were inter-loop interactions between C12 in the first loop and G30 in the third loop (Fig. 4), forming a Watson–Crick-like base pair with two hydrogen bonds between them on the non-planar configuration. Other snapshots indicated interactions between C15 and G30 (Fig. S8). Therefore, it can be inferred that the first and third loops can interact through the cytosine bases in the first loop and G30 in the third loop. Furthermore, the MD simulation revealed that G14 in the first loop interacted with either C29 or C31 in the third loop. Moreover, G28 and G34 formed non-Watson–Crick base pairing (Fig. 4). These results suggest that the loop bases G14, G28, and G34 likely play a crucial role in creating the binding site for CV.

Figure 4
figure 4

Representative structure of BCL2 iM in the MD simulation. The panel on the right is the enlarged view of the loop-loop interactions. Representation of green, magenta, purple, cyan, and brown sticks show guanine, cytosine, protonated cytosine, and thymine bases, and the backbone, respectively. Blue lines represent the hydrogen bonds.

Full size image

To confirm the abovementioned possibility, we systematically mutated the CCCGC sequence between C11 and C15 in the first loop and the CGCCC sequence between C29 and C33 in the third loop of the BCL2 iM. Mutating each nucleotide individually will generate a large set of mutant sequences, which will pose difficulty in their analysis. Therefore, to decrease the number of mutated sequences, we chose to mutate C to T or G to A (Table S1) to generate a significant number of mutated sequences from which we could identify the loop nucleotides responsible for the specific binding sites of CV. Using CD spectra, we confirmed the folding of mutant sequences into an i-motif structure. Some of the representative CD spectra are shown in Fig. S1. Next, we checked the fluorescence response of CV with the mutated sequences (Fig. 3C, D). As expected, G14A mutation on the first loop significantly reduced fluorescence, indicating its crucial role in recognising CV by G14 (Fig. 3C). The magnitude of the emission intensity of CV with the 5′L-CCCAC mutant was tenfold lower than that of the WT, similar to that observed with the abasic sequences of the first loop (BCL2 iM 5ʹL-A). In contrast, mutation at C11, C12, and C13 on the first loop to thymine (5′L-TTTGC) drastically increased the fluorescence, revealing the highest intensity. The magnitude of the intensity of CV with the 5′L-TTTGC mutant was eightfold higher than that of the WT. The binding constant of CV with the 5′L-TTTGC mutant (2.6 × 105 M−1) was also higher than that with the WT BCL2 iM at 25 °C (8.0 × 104 M−1) (Fig. S6B). On the other hand, the 5′L-CTCAT mutant, which exhibited low fluorescence intensity, did not show significant increase in fluorescence intensity of CV (Fig. S6B). As the MD simulation data indicated that the cytosines could interact with G30 on the third loop, these interactions might sterically interfere with CV binding. The snapshot data of the MD simulation indicated that the 5′L-TTTGC mutant had less interaction between the first and third loops; thus, there was space for CV to easily access the binding site (Fig. S9). In the case of the mutation at the third loop, the fluorescence intensity was varied, but there was no clear sequence dependency on the magnitude of the fluorescence changes (Fig. 3D). Moreover, a ninefold decrease in the fluorescence response of CV due to the G34A mutation (Fig. S10) revealed the importance of G34, positioned as the nearest neighbour at the end of the C:C+ core, as a key nucleotide for creating a binding site of CV. Similarly, G28 can be considered as another key nucleotide, as evidenced from the MD simulation study. These results reveal that the specific bases of the first and third loops of BCL2 iM form inter-loop hydrogen bonding and CV possibly interacts with these hydrogen-bonded loop structures through stacking and van der Waals interactions leading to enhanced fluorescence intensity of CV. Therefore, the dynamicity of the BCL2 iM loops is a significant factor for CV fluorescence emission.

Dynamic function of BCL2 loops depending on the stability of i-motif with CV

To investigate how the stability of i-motif affects the fluorescence behaviour of CV, we measured the stability dependency of the BCL2 iM on the fluorescence property. The UV melting analysis of BCL2 iM was carried out at pH 5.0 in the same buffer condition as that in fluorescence assays (Fig. S11). The decrease in the absorbance at 295 nm in the presence of CV reveals the probable interaction between the loop region and CV. Both in the absence and presence of CV, the stability of BCL2 iM at 37 °C (−∆G°37) at pH 5.0 with CV was both 6.1 kcal mol−1 (Table 2), indicating that BCL2 iM stability remains unaffected by CV. The change in the enthalpy (∆H°) and entropy parameters (T∆S°) for the i-motif formation did not reveal a significant change in the absence and presence of CV. Moreover, the melting temperatures (Tm) of BCL2 iM were found to be 65.4 °C both in the absence and presence of CV, indicating no overall stabilisation due to CV binding. We also performed the UV-melting experiments on the 5′L-TTTGC mutant in the absence and presence of CV (Fig. S11). A significant decrease in the absorbance at 295 nm revealed that the probable interaction between the loop region of the mutant sequence and CV was much higher compared to that with the WT sequence. However, the value of −∆G°37 of the 5′L-TTTGC mutant was 4.7 kcal mol−1 (Table 2), revealing a decrease in the i-motif stability of 1.4 kcal mol−1, compared with that of the WT BCL2 iM. The Tm of the 5′L-TTTGC mutant was found to be 60.5 °C, which is 4.9 °C lower than that of the WT sequence. The decrease of 6.0 kcal mol−1 in enthalpy contribution [∆∆H° = (−∆H° for 5′L-TTTGC mutant) ‒ (−∆H° for WT BCL2 iM)] for the 5′L-TTTGC mutant compared with the WT sequence could be responsible for the destabilisation. This destabilisation may be due to reduced interaction between the first and third loops than that in the WT BCL2 iM, as shown in the MD simulation. However, in the presence of CV, the 5′L-TTTGC mutant showed significantly increased stability (−∆G°37 = 5.5 kcal mol−1), resulting in 0.8 kcal mol−1 stabilisation compared to that without CV (−∆G°37 = 4.7 kcal mol−1), which is also supported by a 1.1 °C increase in Tm. ∆∆H°L [∆∆H°L = (−∆H° for 5′L-TTTGC mutant with CV) ‒ (−∆H° for 5′L-TTTGC mutant without CV)] and [∆(T∆S°L) = (−T∆S° for 5′L-TTTGC mutant with CV) ‒ (−T∆S° for 5′L-TTTGC mutant without CV)] values were 8.3 kcal mol−1 and 7.5 kcal mol−1, respectively, indicating that the binding of CV with the 5′L-TTTGC mutant is enthalpically favoured. This favourable enthalpy contribution could arise from an increased stacking interaction between the phenyl groups of CV with the loop bases of the 5′L-TTTGC mutant. These findings suggest that in the case of the WT sequence, the stabilisation effect on i-motif structure by CV binding and 0.8 kcal mol−1 stabilisation in the 5′L-TTTGC mutant, were cancelled out by the destabilisation of the loop-loop interaction formed before CV binding. On the other hand, the −∆G°37 of the 5′L-CTCAT mutant, which exhibited low fluorescence intensity, was found to be 4.7 kcal mol−1 in the absence of CV and 4.6 kcal mol−1 in the presence of CV (Table 2). Both values were very similar to that of the 5′L-TTTGC mutant in the absence of CV. Although the addition of CV did not stabilise the overall i-motif, the binding of CV led to the reorientation of the loop bases, indicating the flexibility of the loop region to generate specific binding sites for small molecules for increase in the affinity. Thus, even if the stability of the whole i-motif structure was not changed, the dynamics of the i-motif structure might be varied upon binding CV to the BCL2 iM.

Table 2 Thermodynamic parameters and melting temperatures of BCL2 iMa.
Full size table

To understand the effect of CV binding on the dynamics of the i-motif structure, we tested different numbers of C-tracts of BCL2 iM at pH 5.0 (Fig. S12A). The detailed study of Hurley’s group on the WT BCL2 iM structure43 revealed that among the six consecutive cytosine runs (excluding C22 and C23), 3 C from run I (C5 to C7), 4 C from run III (C16 to C19), 3 C from run IV (C25 to C27), and 4 C from run VI (C35 to C38) participate in the formation of seven intercalated C:C+ base pairs, which we denote as C3C4C3C4 tracts. Extending this logic, we prepared BCL2 2323, 4545, and 5656 as artificial sequences to investigate the effect of C-tract on the loop region. It is well-established that the stability of the i-motif increases with an increasing number of C-tracts. Compared with the original sequence (WT BCL2 having C3C4C3C4 tracts), the BCL2 2323 sequence having C2C3C2C3 tracts exhibited reduced fluorescence intensity of CV, whereas BCL2 4545 (C4C5C4C5 tracts) and BCL2 5656 (C5C6C5C6 tracts) showed higher fluorescence intensities. The UV melting data indicated the stability increased depending on the C-tract number (Table 2), as expected49. However, all the melting profiles (Fig. S12B) indicated that the i-motifs were completely formed at 25 °C, which was the temperature at which fluorescence was measured. Thus, the core region stability might affect the loop regions dynamics, which regulate the loop function to accommodate CV for fluorescence emission. Our MD simulation revealed that G14, G28, and G34, which played key roles in creating binding sites for CV, were located at the nearest neighbour position of the end of the C:C+ core. The CV fluorescence, depending on the stability of the core, indicated that the dynamics of the core region propagated to the loop regions, resulting in the fluctuation of the binding site of CV followed by the different fluorescence emissions. These results suggest that the factors like solution properties, which alter the loop dynamics, can affect the fluorescence property of the CV-BCL2 iM system.

Characteristics of the CV-BCL2 iM complex under molecular crowding

Molecular crowding affects the physicochemical properties of the solution through excluded volume effect, alteration of water activity, dielectric constant, and medium viscosity. Molecular crowding can stabilise i-motif DNAs under various pH conditions including neutral pH due to the increase in the pKa value of cytosines9, 13, 50. However, limited information is available regarding the effect of molecular crowding on the loops of iMs. Therefore, to investigate the effect of molecular crowding with different solution properties on the dynamics of the loop region, the CV fluorescence with BCL2 iM was investigated using various crowders. Here, we used cosolutes of different molecular weights, such as ethylene glycol (EG), PEG200, PEG8000, and Ficoll70. These different crowders can be utilised to mimic different intracellular crowding conditions. Previous reports have demonstrated that the solution containing different crowders, namely Ficoll70, PEG200, and BSA, can effectively mimic local intracellular conditions such as the nucleus, nucleolus, and cytosol, respectively51. Small cosolutes, such as EG and PEG200, effectively decrease a solution’s water activity and dielectric constant, whereas large cosolutes, such as PEG8000 and Ficoll70 cause the excluded volume effect. The fluorescence intensities of CV with BCL2 iM remarkably decreased in 10 wt% EG and PEG200 solutions (Fig. 5) compared with that in the no cosolute condition. From UV melting assays (Fig. S13), the −∆G°37 values indicated that EG minimally destabilised BCL2 iM, but no effective stabilisation effect was observed for PEG200 (Table S2), which were also supported by the Tm values of the i-motif in those respective solutions. Modulation of hydration state through uptake of water molecules by EG destabilised the i-motif structure10, lowering the fluorescence intensity. Both EG and PEG200 reduce water activity, thereby inhibiting hydration around DNA duplexes in the presence of these crowders52. Therefore, the presence of a long and ordered loop region of BCL2 iM, which can serve as a scaffold for water molecules, may become destabilised in the presence of PEG200, balancing the core region stabilisation by the release of water molecules induced by the crowder10. Destabilisation of base pairings of loop regions in PEG200 was responsible for a larger decrease in the fluorescence intensity. Furthermore, Ficoll70 having only excluded volume effect showed a slight decrease in fluorescence signal. This result suggests that both the core and the loop structures might be stabilised by the excluded volume, but the energetic contribution between the core and the loop regions compensates for the effect on the stability of the i-motif and binding of CV. In the case of PEG8000, it was unexpected that the fluorescence signal decreased more than in EG- and PEG200. Previously, PEGs (of large molecular weights) were reported to directly interact with G4 to stabilise the structures53. Thus, it is possible that PEG8000 interacted with the guanines that acted as a structural platform for CV binding and repressed the CV binding on the loop regions. Except for PEG8000, the solution environment can indirectly and dynamically regulate the CV binding on BCL2 iM. Furthermore, as we observed the hypochromisity of absorbance upon the melting assays in the presence of CV, all crowding agents did not cause major structural changes of the BCL2 iM tetraplex (Fig. S13). Therefore, the crowding agents affected the structure and/or dynamics of loop regions. Hence, the CV binding in cells can be applied to detect and regulate the formation of BCL2 iM depending on intracellular conditions.

Figure 5
figure 5

Fluorescence responses of CV with BCL2 iM in the presence of different cosolutes.

Full size image

Effect of CV on BCL2 gene expression in cells

To confirm the utility of CV for BCL2 iM in cells, we investigated intracellular gene expression from the endogenous BCL2 gene in the MCF-7 human breast cancer cell line (see experimental section). Cancer cells were expected to have an intracellular pH indicating ~ 7.454, although some i-motif formations were identified13. We quantified the mRNAs transcribed from BCL2 using qRT-PCR. The expression levels of mRNA in the absence and presence of the ligand were normalised to the expression level of β-actin mRNA, which is a typical housekeeping gene. Interestingly, the addition of CV decreased the expression levels of BCL2 relative to that of β-actin in a dose-dependent manner (Fig. 6A). Furthermore, we verified that the expression of the reference gene β-actin remained unaffected when using 1 and 5 μM CV in MCF-7 cells. This observation indicates that the concentration ranges of CV used in this study did not exhibit significant toxicity to the MCF-7 cell lines. These results indicate that CV repressed the transcription of the BCL2 gene. Previously, IMC-48 that binds to BCL2 iM upregulated the transcription of the BCL2 gene16. This opposite effect might be due to the differential binding sites of the ligands to the loops of BCL2 iM. IMC-48 was reported to bind to the central loop of the i-motif, whereas in our case, CV was found to bind to the first and third loops of BCL2 iM. As reported by Hurley’s group, molecular recognition of the BCL2 iM by the transcriptional factor, heterogeneous nuclear ribonucleoprotein L-like (hnRNP LL), involves the first and third loops of the i-motif for transcriptional activation22. For this process, the binding of RNA recognition motifs (RRMs) of hnRNP LL requires two consensus sequences found in the first (CCCGC) and third (CGCCC) loops of BCL2 iM sequence. It has been suggested that BCL2 iM has an equilibrium between the hairpin and i-motif structures at neutral pH. Furthermore, hnRNP LL has been shown to preferentially bind to the hairpin form of BCL2 iM 44. As IMC-48 binds to the BCL2 iM central loop, the recognition process of hnRNP LL remains unaltered, and upregulates the BCL2 gene expression. To check whether there was a competition between hnRNP LL protein and CV in binding to BCL2 iM, we performed CD measurements (Fig. 6B). In the absence of CV, the ellipticity of ~ 284 nm band (at pH 6.5) of BCL2 iM gradually decreased with a blue shift in the peak position with increasing hnRNP LL concentration (up to 2 equiv of DNA), indicating partial unfolding of BCL2 iM with the addition of hnRNP LL (Fig. 6B left). On the other hand, in the presence of CV, the reduction of the ellipticity and the blue shift by hnRNP LL were slight especially by the addition of 0.2 and 1 equiv of hnRNP LL compared with those in the absence of CV (Fig. 6B right). The peaks in the regions longer than 300 nm in the presence of CV could be induced CD signals of CV upon binding to DNA. It was further observed that CV exhibited similar interactions with the first and third loops in the BCL2 iM, regardless of the presence of the hairpin structure. However, the complete reduction in fluorescence intensity of CV upon binding to BCL2 iM (Fig. S3) indicates that CV does not bind to the hairpin form. Therefore, the binding of CV to BCL2 iM interferes with the transition to the hairpin form, which is preferred by hnRNP LL. Therefore, the binding of CV to the consensus sequences of the first and third loops possibly hinders the recognition process by the transcription factor resulting in downregulation of gene expression (Fig. 6C).

Figure 6
figure 6

(A) Relative expression level of endogenous BCL2, HRAS, and KRAS mRNA against β-actin mRNA 24 h after the addition of CV measured by qRT-PCR. The qRT-PCR assay was conducted for different mRNAs extracted from three independent cell cultures in each case. The relative expression level was calculated to average these three results and normalized by the one of the results without CV. Statistical significance was evaluated using a Student’s t test. Asterisks, p < 0.05. (B) CD analysis showing that unfolding of BCL2 iM by hnRNP LL in the absence (left) and presence of CV (right) at pH 6.5. (C) Proposed scheme of regulation of gene expression in the absence or presence of CV.

Full size image

To check the specificity of CV on BCL2, we compared the expression levels of BCL2 with the HRAS and KRAS genes in the presence of CV. Insignificant change in the expression levels of the HRAS and KRAS genes in the presence of CV relative to the remarkable decrease in the expression level of the BCL2 gene with CV (Fig. 6A) ensures the specific effect of CV on BCL2. The gene expression of HRAS iM did not respond due to lack of the binding sites of CV within the interloop region, which was also confirmed by the fluorescence assay (Fig. 2). Although the KRAS mid iM does not significantly differ in the number of bases in the first (7 nucleotides) and third (7 nucleotides) loops17 from BCL2 iM (8 and 7 nucleotides in the first and third loops, respectively), it lacks guanine bases, which are proven to be essential for binding CV to BCL2 iM, justifying the relative expression levels of the genes in the presence of CV. Furthermore, the possibility of CV binding to BCL2 G4 in cells cannot be excluded. Our results indicate that CV considerably decreased the gene expression level of BCL2 than those of HRAS and KRAS, complying with the relative gene expression levels of the i-motifs in the cell-based assay experiment with CV (Fig. 6A). However, reduced expression of HRAS and KRAS suggests that CV interacts with the G4s of HRAS and KRAS. Thus, compared to KRAS and HRAS, CV preferentially linked to the i-motif of the BCL2 gene, although some downregulation of BCL2 gene expression may be due to CV binding to BCL2 G4. Thus, this study indicates the novel function of i-motifs for regulating gene expression that can be tuned by targeting the loop regions of i-motifs using specific small molecules.

Conclusion

In this study, we demonstrated that the loop region of i-motif can be used as a novel platform for the specific binding of small molecules revealing light-up fluorescence properties. CV was found to be a specific binder of BCL2 iM over other i-motifs, G4s, and non-tetraplex DNAs. It was observed to bind within the specific site created within the first and third loops of the i-motif, whereas CV did not bind to the hairpin form of the BCL2 iM sequence. Through the local stabilisation of the structure of these loops by CV, the gene expression of BCL2 in cancer cells was successfully repressed. Therefore, the CV-BCL2 iM system can be a promising tool for cancer theranostics. Furthermore, the dynamic behaviour of the binding of CV to the loop regions of BCL2 iM was regulated by crowding environment. This property would be beneficial to analyse or regulate the i-motif formation in different cancer cells, as the formation of i-motifs depends on intracellular conditions, such as cell cycles. Therefore, targeting the dynamics of loop regions to regulate i-motif formation would be a breakthrough approach to analyse and control the i-motif formations in different types of cells, including cancer cells with different malignancies.

Methods

Materials

All the DNA oligonucleotide sequences used in this investigation were purchased from Japan Bio Services Co., Ltd. and purified using high-performance liquid chromatography (HPLC). The DNA samples were dissolved in Milli-Q water and stored at −20 °C until use. The concentrations of the oligonucleotides were determined by measuring the absorbance at 260 nm at 90 °C using the extinction coefficients. Potassium dihydrogen phosphate (KH2PO4), potassium chloride (KCl), ethylene glycol (EG), and poly(ethylene glycol) 200 (PEG200) were purchased from Wako Pure Chemical Industries (Japan), poly(ethylene glycol) 8000 (PEG8000) and Ficoll70 were purchased from Sigma-Aldrich (USA) and dipotassium ethylenediaminetetraacetate (K2EDTA) was purchased from Dojindo Molecular Technologies (Japan), and all these chemicals were used as received. BBR, CV, ThT, and TO were purchased from Sigma-Aldrich (USA); PR and BG were purchased from Wako Pure Chemical Industries (Japan); MG and NR were purchased from TCI (Japan), and all these dye molecules were used without further purification. Stock solutions (2 mM) of the dyes were prepared in Milli-Q water and the required aliquots were taken from the stock solutions for experiments. Addition of the dye solutions to the buffer did not change the pH of the solutions during all experiments.

Protein expression and purification

The gene of truncated hnRNP LL was constructed for expression as a C-terminal his-tagged protein in E. coli by Eurofins Genomic Service (Tokyo, Japan). The truncated genes were subcloned into pET-21a (Novagen). The resulting plasmid was transfected into E. coli BL21(DE3). Transformants were cultured in the Overnight Express medium (Merck) including ampicillin at 37 °C for 20 h. The cells were harvested and lysed by sonication. His-tagged hnRNP LL was purified from the soluble fraction of the extract using His-trap HP (Cytiva) by using buffer containing 20 mM Tris–HCl (pH 7.2) and 500 mM NaCl with a gradient of imidazole. The purified fractions were further purified on a HiTrap Heparin HP column (Cytiva). His-tagged hnRNP LL was eluted with a gradient of NaCl. Fractions were concentrated by Amicon Ultra (Merck) and stored at 4 °C until use.

UV melting studies

UV melting profiles were obtained using a Shimadzu UV-1700 spectrophotometer with a thermoprogrammer. All the experiments were conducted in a buffer containing 10 mM KH2PO4, 1 mM K2EDTA, and 50 mM KCl with pH 5.0 at 25 °C. EG, PEG, and Ficoll were added to the buffer as required. The sample solutions were kept at 95 °C for 5 min, followed by a decrease in temperature to 0 °C at a rate of 0.5 °C min−1 to anneal. Thereafter, the samples were heated to 95 °C at a rate of 0.5 °C min−1 to melt the DNA motif. A constant stream of dry N2 gas was flushed to avoid water condensation on the exterior of the cuvette. The melting profiles were analysed by a fitting equation considering the two-state melting process including subtraction of temperature dependent baselines for the coil and helix forms55. No information was lost during the analysis of the normalised melting profiles when compared with analyses of the original ones. Tm values were determined from the ∆H° and ∆S° values of the melting profiles. The experiments were performed in triplicate.

Fluorescence studies

For scanning the fluorescence response of the dyes in the presence of DNA, 15 µM DNA was taken in a buffer solution containing 10 mM KH2PO4, 1 mM K2EDTA, and 50 mM KCl (pH 5.0 at 25 °C). EG, PEG, and Ficoll were added to the buffer as required. The DNA solutions were heated at 95 °C for 3 min, and then cooled down to 25 °C at the rate of −1.0 °C min−1 and incubated for 30 min. After this, 6 µM dye solution was added to each DNA solution. Then the fluorescence response was recorded on a Varioskan Flash microplate reader (Thermo Fisher Scientific, USA). The excitation and emission monitoring wavelengths for the dyes were CV: 580 nm and 620 nm, MG: 620 nm and 690 nm, BG: 610 nm and 670 nm, PR: 550 nm and 620 nm, NR: 535 and 610 nm, BBR: 350 nm and 530 nm, ThT: 425 nm and 490 nm, and TO: 488 nm and 535 nm. The experiments were performed in triplicate. For the dye-DNA binding constant calculation, the fluorescence data was fitted with a 1:1 non-linear curve-fitting equation33, 56:

$$\mathrm{F}=\mathrm{Fmin}+ \frac{\mathrm{Fmax}-\mathrm{Fmin}}{2[\mathrm{DNA}]}\times {\left(\left[\mathrm{DNA}\right]+\left[\mathrm{Dye}\right]+1/\mathrm{K}\right)-{(\left(\left[\mathrm{DNA}\right]+\left[\mathrm{Dye}\right]+1/\mathrm{K}\right)}^{2}-4[DNA][Dye])}^{0.5}$$

where F is the fluorescence intensity of the dye in the presence of DNA, Fmin is the fluorescence intensity of the dye in the absence of DNA, and Fmax is the maximum fluorescence intensity after titration with DNA. K is the dye-DNA binding constant.

Job’s plot

To determine the stoichiometry between CV and BCL2 iM, independent fluorescence intensities were obtained using various concentrations of CV and BCL2 iM, while the total concentrations of CV and BCL2 iM were fixed at 10 μM. The experiments were performed in triplicate.

Circular dichroism (CD) measurements

The CD spectra were measured on a JASCO J-1500 spectropolarimeter equipped with a temperature controller. All spectra were collected at 4 °C. A stream of dry N2 gas was constantly flushed in the cuvette-holding chamber to avoid water condensation on the cuvette exterior. The spectra were measured in the range of 200 to 350 nm in 0.1 cm path-length cuvettes, with a scan rate of 100 nm min−1. The composition of buffer was 10 mM KH2PO4, 1 mM K2EDTA, and 50 mM KCl (pH 5.0 at 25 °C).

For the hnRNP LL assay, 5 μM BCL2 iM was incubated with or without 50 µM CV and different concentration of hnRNP LL (0, 0.2, 1 and 2 equiv.) in 50 mM MES-KOH (pH 6.5), 100 mM KCl and 1 mM DTT at 25 °C for 5 min before CD spectra measurement. To correct baseline, CD spectra of BCL2 iM containing hnRNP LL with or without CV were subtracted by the spectra of the solution containing each concentration of hnRNP LL with or without CV in the buffer.

Molecular dynamics (MD) simulation

The structures of the BCL2 iM and its mutant sequences were prepared by modification of NMR structure for the i-motif DNA structure (PDB ID: 1EL2)48. The structures of these sequences were obtained by base substitution and DNA strand extension using Discovery Studio Visualizer57. The the simulation box size was set to ensure a minimum distance of 20 Å from the DNA to the wall of the box. The box was filled up by water molecules, potassium and chloride ions as a 0.05 M concentration of KCl solution. The optimisations by the steepest descent method were performed with 10,000 steps for solvent and 10,000 steps for the whole systems. The systems were gradually heated from 0 to 298.15 K for 100 ps and equilibrated by the constant condition of the particle number, volume, and temperature (NVT ensemble) for 1 ns. Equilibrations by the constant condition of the particle number, pressure, and temperature (NPT ensemble) was performed at 1 atm and 298.15 K for 1 ns. Finally, sampling simulations by the NPT ensemble were performed for 100 ns. Three of these simulations with random initial velocity vectors were run for each system using GROMACS Ver. 2020.6. Force fields of ff14SB58 and TIP3P59 employed general nucleotides and water molecules. The force field of protonated cytosine nucleotides was GAFF60 with atom charges determined by the RESP calculation61 of Gaussian0962.

Cell-based assay experiments

Human breast cancer cell line MCF-7 (ECACC 86012803) was purchased from the European Collection of Authenticated Cell Cultures (ECACC). Cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum in the absence and presence of ligand at concentrations of 1.0 and 5.0 µM; after 24 h, the cells were collected for extracting endogenous RNA. Total RNA was extracted using the Qiagen RNeasy Mini kit and cDNA was prepared according to the manufacturer’s protocol. Quantitative reverse transcription qRT-PCR analysis was performed using SYBR® Green Real-time PCR Master Mix (Toyobo) using the StepOne™ Real-Time PCR System (Applied Biosystems). The transcription conditions of BCL2, HRAS, KRAS and β-actin genes were: 60 s at 95 °C, 40 cycles at 95 °C for 15 s, and 55 °C for 60 s. The primer sets were as follows: bcl2 forward, 5′-ATCGCCCTGTGGATGACTGAGT-3′, reverse 5′-GCCAGGAGAAATCAAACAGAGGC-3′; hras forward, 5′-ACGCACTGTGGAATCTCGGCAG-3′, reverse 5′-TCACGCACCAACGTGTAGAAGG-3′; kras forward, 5′-CAGTAGACACAAAACAGGCTCAG-3′, reverse 5′-TGTCGGATCTCCCTCACCAATG-3′; β-actin forward, 5′-TTTAATAGTCATTCCAAATATGAGA-3′, reverse 5′-ACATAATTTACACGAAAGCA-3′. β-actin was used as a control for data normalisation. Experiments were performed in triplicate for three independent cultures of cells. The relative expression level was calculated to average these three results and normalized by the one of the results without CV. Statistical significance was evaluated using a Student’s t test, and results were considered statistically significant when P < 0.05.