Computational and Experimental Characterization of rDNA and rRNA G-Quadruplexes

Abstract

DNA G-quadruplexes in human telomeres and gene promoters are being extensively studied for their role in controlling the growth of cancer cells. G-quadruplexes have been unambiguously shown to exist both in vitro and in vivo, including in the guanine (G)-rich DNA genes encoding pre-ribosomal RNA (pre-rRNA), which are transcribed in the cell’s nucleolus. Recent studies strongly suggest that these DNA sequences (“rDNA”), as well as the transcribed rRNA, are a potential anticancer target through the inhibition of RNA polymerase I (Pol I) in ribosome biogenesis, but the structures of ribosomal G-quadruplexes at atomic resolution are unknown and very little biophysical characterization has been performed on them to date. In the present study, circular dichroism (CD) spectroscopy is used to show that two putative rDNA G-quadruplex sequences, NUC 19P and NUC 23P and their counterpart rRNAs, adopt predominantly parallel topologies, reminiscent of the analogous telomeric quadruplex structures. Based on this information, we modeled parallel topology atomistic structures of the putative ribosomal G-quadruplexes. We then validated and refined the modeled ribosomal G-quadruplex structures using all-atom molecular dynamics (MD) simulations with the CHARMM36 force field in the presence and absence of stabilizing K⁺. Motivated by preliminary MD simulations of the telomeric parallel G-quadruplex (TEL 24P) in which the K⁺ ion is expelled, we used updated CHARMM36 force field K⁺ parameters that were optimized targeting data from quantum mechanical calculations and the polarizable Drude model force field. In subsequent MD simulations with optimized CHARMM36 parameters, the K⁺ ions are predominantly in the G-quadruplex channel and the rDNA G-quadruplexes have more well-defined, predominantly parallel-topology structures as compared to rRNA. In addition, NUC 19P is more structured than NUC 23P, which contains extended loops. Results from this study sets the structural foundation for understanding G-quadruplex functions and the design of novel chemotherapeutics against these nucleolar targets and can be readily extended to other DNA and RNA G-quadruplexes.

Graphical Abstract

Modeled structure of NUC 19P putative rDNA G-quadruplex that was refined by MD simulations. The CD spectra of TEL 24 (black), NUC 19 (red), and NUC 23 (blue) in potassium (solid) and sodium (dotted) solutions.

Introduction

G-quadruplexes, which are formed by GC-rich regions of DNA and RNA, such as the human telomeric repeat and gene promoters^1,2, have been extensively studied for their cell-regulatory functions, as targets of anticancer therapies^3–5, and biosensor and nanostructure construction^6,7. DNA and RNA G-quadruplexes have been found in vivo extensively throughout mammalian genomes, including human, and their formation is dynamic in cells^2,8–12. G-quadruplexes consist of four Hoogsteen hydrogen-bonded guanine bases in a square planar arrangement and feature at least two G-tetrads π-stacked on top of one another^13,14 (Fig. 1). The G-tetrads are connected by nucleotide loops that can be organized in a variety of different orientations depending upon the glycosidic torsion angles of the guanine bases, giving rise to parallel, antiparallel, and mixed topologies^15–19. The G-tetrads form an ideal platform for the π-π stacking of polyaromatic small-molecule ligands that stabilize the G-quadruplex structure²⁰. Each guanine is oriented with the carbonyl oxygen (O6) towards the center of the tetrad, which produces a negatively polarized G-quadruplex channel that accommodates predominantly monovalent cations such as potassium or sodium^21,22. Removal of these ions destabilizes the G-quadruplex, which has also been shown to proceed by the unfolding of the secondary structure in MD simulations²³.

Figure 1:

Structures of ribosomal G-quadruplexes. Schematic representations of (A) parallel conformation. The rectangles represent the guanines that form the three parallel tetrads that form the G-quadruplex structure. The dark black lines represent the nucleic acid backbone. The (B,C) NUC 19P structure modeled and refined by MD simulations. The nucleotides’ heavy atoms are shown in a stick representation, and a gray tube lines its backbone. The ions are spheres. The guanine bases of each tetrad are translucent and space-filled.

Another regulatory role of G-quadruplex DNA may exist in the cell’s nucleolus, which is currently being pursued as a potential cellular target for anticancer therapeutics^24–30. The nucleolus plays a central role in the production of ribosomal RNA (rRNA) to sustain ribosome biogenesis during aberrant cell growth and proliferation. Ribosome biogenesis is tightly regulated by many cell signaling pathways that converge on the RNA polymerase I complex (Pol I)²⁶. Human ribosomal DNA (which we will henceforth refer to as “rDNA”) is GC-rich (~73% G in the 13,357-base-pair 45S pre-rRNA gene, NCBI reference NR 046235). The transcription of rDNA by Pol I is thought to be accompanied by the formation of G-quadruplex structures in the nontemplate strand. Since rDNA transcription is upregulated in transformed cells (as manifest by greatly enlarged nucleoli compared with normal cells)^24,25, inhibition of Pol I-mediated transcription by small molecules, such as G-quadruplex binders, represents an attractive therapeutic strategy^31–133. While indiscriminate damage to nuclear DNA results in genotoxicity and unwanted side effects, targeting the nucleolus may trigger cancer cell death more selectively while sparing normal cells^27,30.

Many chemotherapeutics derived from anthraquinones³⁴, fluorenones³⁵, acridines³⁶, and other polyaromatic chromophores show a high affinity for G-quadruplexes. Among these is quarfloxin, a fluoroquinoline derivative that accumulates in the nucleolus²⁷, as demonstrated in lung adenocarcinoma cells³⁷. Quarfloxin was the first G-quadruplex binding agent to enter clinical trials but left Phase II clinical trials due to bioavailability issues even though it was well tolerated by patients¹. It selectively inhibits rRNA synthesis by competing with nucleolin, a nucleolar phosphoprotein believed to bind to the G-quadruplex in the untranslated strand of rDNA, and triggers apoptosis³⁷. Recently, Xu et al. found that CX-5461, like other RNA pol 1 inhibitors, bind and stabilize G-quadruplex structures in vitro and selectively kill BRCA1/2 deficient cancer cells²⁸. It is currently undergoing Phase 1 clinical trials (NCT02719977).

Despite the potential to improve anticancer therapy and its successes so far, ribosomal G-quadruplex structures that could open a new direction in the development of G-quadruplex selective binders for rDNA and rRNA are largely unexplored. While search algorithms have proven useful for scanning the genome for putative G-quadruplex-forming sequences³⁸, they are statistically overrepresented and it is unknown whether they all translate into the formation of G-quadruplex structures in the cell. Thus, rigorous biophysical validation of their topologies is required¹.

To address this gap in our understanding of ribosomal G-quadruplexes and to set the foundation for biophysical studies of G-quadruplexes where the structures are not yet empirically characterized, we modeled, using all-atom MD simulations, two putative rDNA and rRNA G-quadruplex sequences, a 19-nt (NUC 19) and a 23-nt (NUC 23), from the human pre-rRNA 45S gene, as well as the telomeric DNA and RNA G-quadruplex (TEL 24) as a control. Addition of a “P”, an “A”, or a “M” to these sequence designations indicates that the structures are parallel, antiparallel, or mixed topology, respectively, but the sequences have not been changed (except for TEL 24P as described in Materials and Methods). We also introduce novel G-quadruplex-specific topological order parameters of center of mass base-base distances and torsion angles over the defined timescales to structurally characterize the fluctuations and distortions of each rDNA and rRNA G-quadruplex.

Results and Discussion

Although sequence analysis algorithms can be used to identify DNA sequences that might be prone to form G-quadruplexes, it is impossible to predict structural and topological features from them alone. Currently, no known experimentally resolved structure of a ribosomal G-quadruplex is available, and it is impractical to solve the crystallographic or NMR structures of every putative G-quadruplexes in the 45S gene. High-resolution information is required for the rational design of drugs that can be targeted at them. Low resolution biophysical techniques such as circular dichlorism (CD) can only provide an ensemble-averaged assessment of the G-Quadruplex topology; thus, we combined computational modeling with MD simulation to enhance our assessment of structural stability from our CD measurements.

Optimization of CHARMM36 Force Field for MD Simulations of G-Quadruplexes

There have been many MD simulations of G-quadruplexes with the additive CHARMM and AMBER force fields. In a notable study, Bian et al. showed using MD simulations combined with the enhanced-sampling metadynamics method that the trapping of a K⁺ ion in the channel stabilizes a triplex intermediate and the trapping of a second K⁺ ion stabilizes the fully folded G-quadruplex structure³⁹. However, several other studies refer to issues with ions in G-quadruplex channel^40–42. Fadrna et al. observed MD simulations of G-quadruplexes with both the AMBER (parm99, parm99-bsc0) and CHARMM27 to result in ion expulsion from the G-quadruplex channel⁴³. Lemkul and co-workers also observed ion expulsion in their CHARMM36 MD simulations and they instead performed Drude polarizable MD simulations that were able to keep the ions on the G-quadruplex channel. They attribute the success of the Drude polarizable MD simulations to the multibody water effects on the ion-base interactions^44,45. These and other similar studies suggest that the additive MD simulations force fields may be sufficient for inducing folding of G-quadruplexes and capturing the ions but insufficient for retaining them in their channels.

In our preliminary CHARMM36 MD simulations of the well-studied parallel telomeric G-quadruplex (TEL 24P), we performed 20 independent trajectories, and the monovalent ions escaped the G-quadruplex channel formed by the guanine-O6 atoms during the equilibration period for 12 trajectories and in less than 5 ns during the subsequent production run in the remaining 8 trajectories (Fig. S1). When the equilibration period was increased from 50 ps up to 500 ps, ejection of ions from the channel was still observed (Fig. S2). To determine whether the van der Waals radius of the potassium ion in the CHARMM force field was too large, we reduced the radius to that of sodium, which still did not prevent ejection of ions (Fig. S3A). Another potential cause of the instability of the ion coordination was the mutual electrostatic repulsion between the potassium ions in the channel. However, even when the charges between the potassium ions in the channel were removed, ion expulsion could not be eliminated (Fig. S3B).

The limitations of the ion parameters in the additive CHARMM⁴⁶ and AMBER⁴⁷ force fields and similar issues have previously been observed in MD simulations of G-quadruplexes^40–42. Fadrna et al. modified the AMBER force field parameters for the interaction between the K⁺ ion and the guanine O6 atom⁴³. While this fix can alleviate the issue, it can be also resolved through the development of polarizable force fields^48–52 and subsequent optimization of additive force fields, as we will show below. Gkionis et al. showed that polarization of the monovalent ions become more significant when multiple ions are present in the channel, and they suggest that the nonpolarizable nature of current additive force fields may limit their accuracy⁵³. Savelyev and MacKerell optimized the Na⁺ ion interaction with bases in CHARMM36 to improve consistency with quantum mechanical calculations⁵⁴. They performed similar optimizations with the K⁺ ion that were previously tested and we also use in our present study. For linguistic simplicity, we will subsequently refer to these pair-specific optimized LJ parameters as “NBFIX”.

To validate the optimized parameters, we performed 10 independent 100 ns MD simulations of TEL24P using the CHARMM36 force field with NBFIX and the AMBER parm99-bsc0 force fields (see Fig. 2 for representative trajectories). For CHARMM36 with NBFIX, both ions remained predominantly in the G-quadruplex channel with at most one ion leaving at a time but then returning within 5 ns for 9 trajectories and returning after about 7 ns in one trajectory. For the AMBER parm99-bsc0 force field, both ions remained in the channel throughout for 3 trajectories, with at most one ion leaving at a time but returning within 5 ns for 2 trajectories, and 5 trajectories where the ion left the channel and did not return. Using one of the several possible modifications of Fadrna et al.⁴³, Islam et al. performed a 1.5 μs MD simulation with the K⁺ ion remained positioned between the G-quartets⁵⁶.

Figure 2:

The distance of the two potassium ions from the TEL 24 P G-quadruplex channel with (A) CHARMM36 and our nonbonded correction and (B) AMBER parm99-bsc0. The red and blue lines refer to the distances of the two potassium ions relative to their ideal positions inside the G-quadruplex channel.

NUC 19 and NUC 23 Form Predominantly Parallel Topologies Regardless of Cation

In a very important study, Drygin et al. used a search algorithm to identify putative G-quadruplex forming sequences in the human genome in the nontemplate strand of each human rRNA gene. On the basis of (unreported) CD spectra recorded of the corresponding oligonucleotides, they were classified as parallel, antiparallel, and mixed topologies³⁷. Building upon this we used CD spectroscopy to study 2 of the reported 14 sequences, NUC 19 and NUC 23, along with the telomeric G-quadruplex, TEL 24, in the presence of K⁺ or Na⁺ ions (Fig. 3).

Figure 3:

CD spectra of the telomeric G-quadruplex, TEL 24 (black traces) and two putative rDNA G-quadruplexes, NUC 19 (red traces) and NUC 23 (blue traces). Solid and dashed traces correspond to G-quadruplexes formed in K⁺- and Na⁺-containing solution, respectively.

In complete agreement with previous studies, TEL 24 in K⁺ buffer exhibits a positive feature with a maximum at 285 nm and a shoulder around 270 nm, consistent with the mixed parallel/antiparallel (“3+1” hybrid) topology of TEL 24M that exists under these conditions^57,58. In Na⁺-containing buffer, this sequence shows positive bands at 240 and 290 nm and a negative band at 260 nm that are characteristic features of an antiparallel topology (TEL 24A)⁵⁹. For NUC 19 and NUC 23, a CD signature with characteristic pronounced positive and negative features at 260 nm and 240 nm, respectively, is obtained both in the presence of K⁺ and Na⁺, pointing to parallel topology structures. Furthermore, short loop lengths with a G₃NG₃ motif with 1-nt loop greatly favor the parallel topology G-quadruplexes chain-reversal loops^60–63. There are two such motifs in NUC 19 and one in NUC 23, which explains the strongly parallel topology CD profile.

Based on the CD spectra of the rDNA G-quadruplexes (Fig. 3), the parallel forms of both rDNA sequences, NUC 19P (major component) and NUC 23P, were investigated by MD simulations. Their starting structures were derived from the solid-state structure of the parallel telomeric G-quadruplex. The propeller-type structure was used as a scaffold and its TTA loops were replaced with appropriate loops adopted from experimental G-quadruplex coordinates available in the Protein Data Bank (PDB) and the Nucleic Acid Database (NDB). The models were then constructed by replacing the bases and making minimal changes to the experimental geometries as required to thread the correct sequence (Figs. 4A,B). The corresponding RNA G-quadruplex structures were also built by converting 2´-deoxyribose into ribose and “mutating” thymine (T) into uracil (U).

Figure 4:

Modeled structures of the (A) NUC 19P and (B) NUC 23P DNA G-quadruplexes. The G-tetrad core was based on the parallel telomeric G-quadruplex structure (ball-and-stick representation) and the loops (licorice representation) were based on existing G-quadruplex loop structures. Their sequences are below with the tetrad guanines underlined.

Modeled G-Quadruplexes have Stable Cores and Flexible Loops

To quantify the structures of the TEL 24 and the modeled NUC 19P and NUC 23P DNA and RNA G-quadruplexes, we performed 250 ns MD simulations with K⁺ ions using the CHARMM36 force field with NBFIX parameters. In every trajectory, the RMSD of the G-quadruplex tetrads were ~2 Å while the RMSD of the loops fluctuated more and equilibrated to ~6 Å (Fig. 5). The similarity of the RMSDs in the modeled DNA and RNA G-quadruplexes suggest that our modeling methodology produces stable structures that are suitable for further studies via MD simulations.

Figure 5:

G-quadruplex tetrad and loop RMSDs of TEL 24P (A,B), NUC 19P (C,D), and NUC 23P (E,F). The MD simulations for the G-quadruplexes were performed for DNA (left column) and RNA (right column) G-quadruplexes.

Structural Geometric Order Parameters for G-Quadruplexes

While G-quadruplexes seem like relatively simple structures, it is not trivial to quantify their structural fluctuations and distortions. For example, one could use a global RMSD but local information would be lost in their characterization. One could also simply count the number of hydrogen bonds or measure the glycosidic torsion angles, but then even very simple G-quadruplex structures become very complicated to understand. Given the square planar geometry of G-tetrads and their π-stacking, we instead chose a middle-ground by introducing local structural geometric order parameters that make it intuitive to quickly identify where the G-quadruplex structure deviates from an ideal structure.

The introduced measures to quantify the structure of the G-quadruplexes included two novel metrics 1) the center of mass base-to-base distances between two diagonal guanine pairs in a quartet (Fig. 6A) and 2) a torsion angle defined by four guanines in a tetrad (Fig. 6B). Our local metrics precisely identify the modes in which the structures are similar or deviate from an ideal G-quadruplex structure. In a well-defined G-quadruplex structure, we would expect native hydrogen bond distance between the heavy atoms to be ~3 Å, the base-to-base distances to be in close proximity (~9 Å), and the torsion angle to be zero if they are in the same plane. Based on deviations from these ideal values one can readily surmise where in the G-quadruplex structure is being distorted, specifically which bases and which tetrads.

Figure 6:

Two novel geometric order parameters to quantify G-quadruplex structure in the MD simulations. A) To quantify the proximity of the guanines in the tetrads, we measured the base-to-base distances of the center of masses of two diagonally positioned guanines. B) To quantify whether the guanines were in the same plane, we measured the torsion angle defined by the center of masses of the four guanines in a tetrad.

The base-to-base guanine tetrad distances (Fig. 7A,B) and tetrad torsion angles (Fig. 7C,D) were computed for both CHARMM36 and AMBER parm99-bsc0 force fields in a representative MD simulation where the ion left the channel but returned within 5 ns (Fig. 2). For both force fields, we observe similar probability distributions with sharp peaks at about 9 Å for the base-to-base guanine tetrad distances (Fig. 7A,B) and about 0 degrees for the tetrad torsion angles (Fig. 7C,D), suggesting that the structures were largely well-defined throughout the MD simulations using both force fields.

Figure 7:

Comparison of MD simulations of TEL 24P with CHARMM36 and our nonbonded correction (left column) and AMBER parm99-bsc0 (right column). (A,B) The probabilities of the (C,D) base-to-base guanine tetrad distances and (C,D) tetrad torsion angles.

Benchmarking G-Quadruplex Local Structural Geometric Order Parameters with TEL 24P

To directly compare our local metrics (native G-tetrad hydrogen bond distances, base-to-base distances, and torsional angles) to the global RMSD measure we performed all-atom CHARMM36 MD simulations of the parallel topology telomeric DNA and RNA G-quadruplexes (TEL 24P) in the presence and absence of potassium ions. The parallel topology TEL 24P DNA structure has been resolved using X-ray crystallography⁶⁴. However, since the experimentally resolved TEL 24P RNA G-quadruplex structure is not yet resolved, we trivially generated the RNA G-quadruplex structures by replacing the 2’ hydrogen with a hydroxyl group and replacing each thymine with a uracil. In the presence of the potassium ions, we performed the TEL 24P DNA and RNA G-quadruplex MD simulations for 250 ns. In the absence of the potassium ions, our MD simulations were run for 1.25 μs but only trajectories from the last 250 ns were used for analyses.

In each of those cases, we calculated the native G-tetrad hydrogen bond distances, diagonal base-to-base tetrad guanine distances (Fig. 6A), and tetrad guanine torsion angles (Fig. 6B) and generated histograms of their distributions (Fig. 8A,D,G,J; 9A,D,G,J; 10A,D,G,J). In excellent agreement with its RMSD distributions, the TEL 24P DNA and RNA in the presence of potassium have a single peak at about 3 Å for tetrad native hydrogen bond distances (Fig. 8A,G), 9 Å for all guanine pair distances (Fig. 9A,G), and all tetrad torsion angles also have a single peak at about 0 degrees (Fig. 10A,G). For the TEL 24P DNA in the absence of potassium, there is also a single peak for the guanine pair distances at about 9 Å (Fig. 9D), but the native hydrogen bond distance distribution is bimodal with a second peak at around 3.7 Å (Fig. 8D) and the torsion angle distribution for the bottom tetrad is peaked about +5 to +10 degrees away from the ideal 0 degrees peak that is observed for the top and middle tetrads (Fig. 10D), indicating that the bottom tetrad has slightly destabilized, as expected. For the TEL 24P RNA in the absence of potassium, native hydrogen bond distance peaks range from around 11–18 Å (Fig. 8J), the guanine pair distance distributions have single or multiple peaks with locations that range from around 4–40 Å (Fig. 9J), and the tetrad torsion angles all deviate significantly from 0 degrees, ranging from about −70 to −20 degrees (Fig. 10J). Clearly, the TEL 24P RNA structure has significantly destabilized, as expected, again in excellent agreement with the RMSD. Further, we are able to use these local geometric measures to pinpoint the location of the structural disorder that persists throughout the entire structure, which would not be possible with only RMSD measurements.

Figure 8:

The normalized frequency distributions of the G-tetrad native hydrogen bond distances of TEL 24P (left column), NUC 19P (center column), and NUC 23P (right column). The MD simulations for the G-quadruplexes were performed for DNA (A-F) and RNA (G-L) G-quadruplexes in the presence (A-C; G-I) and absence (D-F; J-L) of potassium ions in the G-quadruplex channel.

Figure 9:

The normalized frequency distributions of the base-to-base tetrad guanine distances of TEL24P (left column), NUC 19P (center column), and NUC 23P (right column). The MD simulations for the G-quadruplexes were performed for DNA (A-F) and RNA (G-L) G-quadruplexes in the presence (A-C; G-I) and absence (D-F; J-L) of potassium ions.

Figure 10:

The normalized frequency distributions of the tetrad guanine torsion angles of TEL 24P (left column), NUC 19P (center column), and NUC 23P (right column). The MD simulations for the G-quadruplexes were performed for DNA (A-F) and RNA (G-L) G-quadruplexes in the presence (A-C; G-I) and absence (D-F; J-L) of potassium ions in the G-quadruplex channel.

NUC 19P and NUC 23P DNA in Potassium have Well-Defined Parallel G-Quadruplex Structures

For each modeled parallel ribosomal DNA G-quadruplex structure, NUC 19P and NUC 23P, we performed all-atom CHARMM36 MD simulations in the presence and absence of potassium ions. Since cations in the G-quadruplex channel are required for the G-quadruplex to maintain its structure, the removal of the potassium ions is expected to induce the destabilization and eventual unfolding of the G-quadruplex.

Just as with TEL 24P, MD simulations of the NUC 19P and NUC 23P DNA G-quadruplexes in the presence of potassium ions were performed for 250 ns. Throughout the trajectories, the native hydrogen bond distances, base-to-base distances of each diagonal guanine pairs, and the torsion angle of each tetrad were monitored, and we observed narrow distributions with a single peak for both the distances and torsion angles, indicating stable G-quadruplex structures throughout the trajectories (Fig. 8B,C; 9B,C; 10B,C). The peak of the distribution for the native hydrogen bond distances was around 3 Å, the base-to-base distances was around 9 Å, and the one for the torsion angle was zero.

We again performed the corresponding MD simulations without potassium ions for 1.25 μs and performed analyses on the final 250 ns to directly compare to the MD simulations with the potassium ions. For the NUC 19P, we still observe a single peak for all three metrics (Fig. 8E; 10E; 11E), suggesting that the loss of potassium does not significantly impact its structure at these timescales. In contrast, the NUC 23P became much more unstructured without potassium ions. The native hydrogen bond distances range from 7–12 Å where the top tetrad had a bimodal distribution (Fig. 8F). The base-to-base distances consisted of unimodal and bimodal distributions with peaks ranging from 3–20 Å (Fig. 9F). Furthermore, the torsion angle peaks range from about −10 to +130 degrees (Fig. 10F). These metrics together suggest a largely unstructured G-quadruplex.

NUC 19P and NUC 23P RNA Are Less Stable Parallel G-Quadruplexes

The ribosomal RNA G-quadruplexes in potassium exhibited slightly higher fluctuations of the base-to-base distances and torsion angles in the top and bottom tetrads (Fig. 8H,I; Fig. 9H,I) as compared to their corresponding DNA G-quadruplexes (Fig. 8B,C; Fig. 9B,C). Specifically, for NUC 19P RNA in the presence of potassium, a single guanine distance pair in the top tetrad exhibited multiple peaks with the dominant peak around 22 Å (Fig. 9H), and the torsion angle had two peaks, one peaked at about 40 degrees (Fig. 10H). For NUC 23P RNA, while all distance pair peaks were ideal, all three tetrads deviated from 0 degrees by about 5 degrees (Fig. 8I).

In the absence of the potassium ions, both the distributions of the guanine pair distances and the tetrad torsion angles deviated significantly from ideal geometries (Fig. 9K,L; 10K,L), indicating that the structures were significantly destabilized. It is unclear to us why the RNA G-quadruplexes are more unstable than their DNA counterparts in our MD simulations, though it is likely that there is a deeper local minima for the quadruplex structure of DNA versus the RNA G-quadruplex in the absence of the ions. Additional studies are required to address this in greater detail.

Conclusions

The nucleolus is becoming recognized as a major new target for anticancer therapies that inhibit ribosome biogenesis in cancerous cells^24–27. The DNA genes encoding pre-ribosomal RNA (rDNA) are experimentally observed to form DNA G-quadruplexes, however, the lack of any experimentally resolved high resolution structures of G-rich 45S rDNA that are predicted to adopt G-quadruplexes is a major barrier for rational drug design studies. Recently, rRNA G-quadruplexes have also been identified in vitro¹¹ and in vivo in the nucleoli⁶⁵ and are likely to be important drug targets as well. To biophysically characterize putative ribosomal G-quadruplexes and generate well-defined structures for future drug design studies, we performed atomistic MD simulations of two modeled putative ribosomal G-quadruplexes, NUC 19P and NUC 23P, which had previously been predicted using bioinformatics methods to be prone to G-quadruplex formation³⁷. CD spectra recorded for both sequences show predominantly parallel topology structures, regardless of whether the cation is Na⁺ or K⁺ (Fig. 3).

To quantify their structures throughout the MD simulations, we introduced two novel local geometric measures: 1) the base-to-base distances of diagonal guanines in a tetrad and 2) the torsion angle of the four guanines in a tetrad (Fig. 6). MD simulations show that NUC19P and NUC23P maintain well-defined DNA G-quadruplex structures throughout the trajectories in the presence of potassium (Fig. 8,9). This stability is consistent with the CD spectra of those species (Fig. 3). Remarkably, NUC 19P, with shorter loops than NUC 23P, has a well-defined structure even in the absence of potassium. Previous studies on other G-quadruplexes have already shown that G-quadruplexes with shorter loops were more stable and that structures with two single-nucleotide loops were constrained to the parallel topology⁶⁶. Overall, the rDNA G-quadruplexes had more well-defined structures than their counterpart rRNA G-quadruplexes.

The results of these calculations suggest that the studied sequences fold into structurally unique and well-defined parallel topologies that are robust to changes in ionic conditions, in contrast to telomeric G-quadruplexes that can adopt antiparallel or mixed topology topologies, depending on whether potassium or sodium ions are present⁵⁸. The parallel topology of ribosomal DNA G-quadruplexes may be biologically relevant, and an appropriate target for the design of novel drugs to inhibit Pol I transcription in the nucleolus. While other parallel topologies, such as a “snapback” parallel G-quadruplex⁶⁷ that is indistinguishable from the standard parallel topology G-quadruplexes that we modeled may exist, small molecules can be designed to specifically targeting the standard parallel topology G-quadruplexes, including the loops linking the bases, and shift the equilibrium away from presumably minor species. Binding studies have been performed with acridine-based intercalators with their planar scaffold used to develop pharmacophores targeted at nucleolar ribosomal G-quadruplexes. The study demonstrated that these exogenous ligands not only bind selectively to parallel G-quadruplexes, but are able to convert antiparallel into parallel topologies even in the absence of stabilizing monocations⁶⁸. As such, our present study sets the basis for screening the topological space of additional rDNA and rRNA G-quadruplex parallel topologies and the docking and refinement of specific ligand interactions with G-quadruplexes.

Materials and Methods

Modeling of NUC 19P and NUC 23P DNA and RNA G-Quadruplexes

For each of the parallel topology G-quadruplex structures we modeled, the G-tetrads were based on the structures of the telomeric crystal structure of a parallel G-quadruplex (PDB: 1KF1; (TTAGGG)4)⁶⁹. We identify the telomeric parallel G-quadruplex in our current study as “TEL 24P”, but we note that the experimentally resolved structure we used in our MD simulations is missing the first two thymines such that it is 22 nucleotides in length. A database of parallel DNA G-quadruplexes from the PDB and NDB was compiled and the loops from each structure were extracted. The glycosidic torsion angles for both guanines connecting a given loop were computed as well as the center of mass base-base distance for each guanine. From these two calculations, the loops that were the best fit for the 1KF1 structure were aligned using VMD⁷⁰. Some of the bases were then mutated using the script mutateNA.pl from the MMTSB online toolset and excess bases were removed to match the DNA sequences NUC 19 (GGGGGTGTGGGGGGGAGGG) and NUC 23 (GGGTGGCGGGGGGGAGAGGGGGG). The underlined guanines are part of the G-tetrads in the G-quadruplex, and all other nucleotides in the loop regions. The nucleotides were then renumbered using the script convpdb.pl from the MMTSB online toolset⁷¹ to match the numbering of the putative sequence and re-aligned using VMD. In addition, the RNA structure was made from the DNA structure by mutating the thymines for uracils using mutateNa.pl and building the 2’-OH moieties based on internal coordinates.

CD Spectra of DNA G-Quadruplexes

Synthetic oligodeoxyribonucleotides corresponding to the sequences NUC 19, NUC 23, and the human telomeric repeat, d(TTAGGG)₄ (TEL 24), were purchased from Integrated DNA Technologies, Inc. (Coralville, IA), where they were synthesized using standard phosphoramidite chemistry and desalted. Without further purification, the sequences were dissolved in buffer (10 mM tetrabutylammonium phosphate, 1 mM EDTA, pH 7.4) supplemented with either 100 mM NaCl or 100 mM KCl to obtain the desired cation concentration. After adjusting the final DNA concentration to 10 μM (strand), the sequences were allowed to fold into the most stable intramolecular G-quadruplex structures by annealing in the appropriate buffers by heating the samples at 95 °C for 10 minutes, followed by gradual cooling to room temperature over 8 hours in a digital dry bath incubator. CD spectra were recorded on an AVIV Model 215 Circular Dichroism spectrometer in quartz cuvettes, which were maintained at 25 °C with a thermoelectrically controlled cell holder. CD signatures were monitored in the range of 200–350 nm at 1-nm resolution and an averaging time of 2 s. A blank spectrum was used for background subtraction from all spectra, and a moving average smoothing algorithm (10 periods) was applied.

MD Simulation Protocol

After the models for the DNA and RNA G-quadruplexes were created, we performed all-atom MD simulations using the NAMD 2.9 software suite⁷² with the CHARMM36⁴⁶ or AMBER parm99-bsc0⁷³ nucleic acid force fields and their associated ion parameters from Beglov and Roux for CHARMM⁷⁴ and Aqvist et al. for AMBER⁷⁵ to keep the force fields consistent. We used TIP3P waters within a rectangular cube with 10 Å padding in each direction and periodic boundary conditions with ion concentration set to 100 mM NaCl or KCl to match experimental solvent conditions. The simulations were run at 298 K with a 2 fs timestep. A minimization was run using a conjugate energy gradient and the system was equilibrated for 50 ps by raising the temperature from 0 K to 298 K. Each DNA and RNA model was run for a production time of 250 ns with potassium and sodium in between the tetrads to model the folded states and 1,250 ns without potassium or sodium between the tetrads to model the unfolded states. For the “unfolded” state, we analyzed only the last 250 ns of each trajectory to match the timescale of the folded state trajectory.

Synopsis.

Cancer is the uncontrolled growth of cells, and pharmacological studies attempt to find novel means of inhibiting this growth by using drugs that selectively inhibit a critical process in cancerous cells. A very promising direction is to develop small molecules that stabilize DNA and RNA G-quadruplex structures that generally act as roadblocks for proteins involved in cancer-related processes. Recent studies have pointed to a new direction by targeting the nucleolus, which is the site of ribosome biogenesis. We develop structural models of the putative rDNA and rRNA G-quadruplex structures from the nucleolus, which were validated using molecular dynamics simulations. The DNA structures are significantly more stable than the RNA structures, and the one with shorter sequence loops are more stable than the one with longer sequence loops. As a result, the shorter sequence DNA G-quadruplex may be the more biologically relevant structure for the rational design of new drugs that target the human nucleolus and potentially provide a viable treatment for cancer.

Abbreviations

MD

molecular dynamics

CD

circular dichroism

Pol I

polymerase I

pre-rRNA

pre-ribosomal RNA

rDNA

DNA sequences that encode rRNA

QM

quantum mechanical