Slow motions in A·T rich DNA sequence

In free B-DNA, slow (microsecond-to-millisecond) motions that involve equilibrium between Watson–Crick (WC) and Hoogsteen (HG) base-pairing expand the DNA dynamic repertoire that could mediate DNA–protein assemblies. R1ρ relaxation dispersion NMR methods are powerful tools to capture such slow conformational exchanges in solution using 13C/15 N labelled DNA. Here, these approaches were applied to a dodecamer containing a TTAAA element that was assumed to facilitate nucleosome formation. NMR data and inferred exchange parameters assign HG base pairs as the minor, transient conformers specifically observed in three successive A·T base pairs forming the TAA·TTA segment. The abundance of these HG A·T base pairs can be up to 1.2% which is high compared to what has previously been observed. Data analyses support a scenario in which the three adenines undergo non-simultaneous motions despite their spatial proximity, thus optimising the probability of having one HG base pair in the TAA·TTA segment. Finally, revisiting previous NMR data on H2 resonance linewidths on the basis of our results promotes the idea of there being a special propensity of A·T base pairs in TAA·TTA tracts to adopt HG pairing. In summary, this study provides an example of a DNA functional element submitted to slow conformational exchange. More generally, it strengthens the importance of the role of the DNA sequence in modulating its dynamics, over a nano- to milli-second time scale.

DNA-protein recognition processes occur through so-called direct and indirect readout of DNA by proteins. The formation of nucleoprotein complexes requires in particular, recognition of DNA chemical patterns specific to each base, and DNA structural and dynamic features that are sequence dependent. Deciphering the dynamics of DNA is not experimentally easy. Nuclear Magnetic Resonance (NMR) has long been, and remains a powerful technique for capturing picosecond dynamics at atomic resolution; more recent developments of relaxation dispersion experiments extend the timescale up to milliseconds and quantitatively investigate slow conformational exchange processes [1][2][3][4] .
Initially, the existence of slow dynamic movements in DNA emerged from early NMR data collected on adenines of TpA steps, by detecting an excess linewidth of adenine H2-resonance protons in biologically active DNA sequences 5,6 . Further investigations showed that this linewidth broadening was sensitive to the TpA tetranucleotide sequence context [6][7][8] . The observation of these resonance experiments was interpreted as being due to a slow exchange between two conformational states 7,9,10 , which could arise because of poor TpA stacking 9,10 . R 1ρ relaxation dispersion NMR experiments have led to major advances in this field. These experiments were used to reveal and characterize slow conformational equilibria between major and minor conformational states, assimilated to ground and excited states in analogy with spectroscopy. The use of double 13 C and 15 N labelled molecules allowed the detection and analysis of excited conformer populations of less than 1%. This approach initially provided insights on various topics essentially related to proteins such as folding, enzymatic catalysis, ligand binding and recognition 1,2,11 , but now also gives information about intrinsic slow motions of nucleic acids 3,[12][13][14][15][16] . Applied to B-DNA double helices, this methodology revealed an unexpected equilibrium between two schemes of base-pairing: standard Watson-Crick (WC) base pairs being able to transiently adopt the Hoogsteen (HG) configuration 15,[17][18][19][20] . Prior to these NMR studies, the first experimental evidence of HG pairing in DNA was obtained in 2002 using a DNA containing 6 A·T base pairs that all crystallized in the HG mode 21 . Both experimental 22 and theoretical 23 approaches showed that d(AT) n sequences were globally more stable in the classical WC B-DNA than in a fully HG double helix, at least in dilute aqueous solutions and in the absence of cofactors. However, earlier models 24 as well as more sophisticated MD structures 15,25,26 demonstrated that HG pairing can coexist with WC base pairs in the same DNA without generating prohibitive energy cost. Thus, the presence of transient HG base pairs in solution, as shown by NMR, is strongly supported by these previous reports.  www.nature.com/scientificreports/ Here, the H2 chemical shifts of A 7 and A 19 are in fact strongly high-field shifted ( Fig. 1a) in the unique TpA 7 ·TpA 19 complementary dinucleotide of the studied dodecamer (see the sequence in above "Introduction" section). Although these adenines share the same tetranucleotide context (TTA 7 A·TTA 19 A), their H2 chemical shifts are clearly different (Fig. 1a). Thus, the sensitivity of H2 chemical shifts to the sequence depends on both 5′ and 3′ nearest neighbours and not only on the 3′ base as previously postulated 10,40 .
A key point is that the H2 resonances of A 7 and A 19 are severely broadened compared to those of A 8 , A 9 and A 20 (Fig. 1a). These A 7 and A 19 H2 broadenings are detected below the Tm of 57 °C (Supplementary Fig. S2) and their amplitudes are maximal between 25 and 30 °C (Fig. 1b). Such behaviour of H2 adenine resonances resembles other TpA results that were interpreted as being the signature for conformational motion 5-9 occurring at the microsecond-millisecond range 41 . C2-H2 cross-peaks were then identified from a constant-time 1 H-13 C spectrum. Weak intensity and significant broadenings are observed for only A 19 and A 7 cross-peaks. 1D H2 inversion-recovery and 1D dispersion R1ρ relaxation spectra show that both H2 and C2 resonances of A 19 and A 7 are involved in these specific cross-peak particularities. So, the motions detected here influence the NMR behaviour of at least these two adenine atoms. The next step was to apply R 1ρ relaxation dispersion experiments to the 13 C/ 15 N labelled dodecamer, as a means to follow and describe slow conformational exchanges. The 1D selective R 1ρ spectrum of the A 9 resonance illustrates the good selectivity of this type of experiment, ensuring that the excitation of a given carbon does not affect its spectrum neighbours (Fig. 2).
On-resonance 13 C R 1ρ dispersion relaxation experiments: evidence of slow conformational exchanges. R 1ρ relaxation dispersion experiments 14 in the on-resonance version constituted a first approach for identifying those carbon atoms of nucleotides submitted to slow conformational exchange. The R 1ρ rates were carefully measured to obtain the best estimate of the exponential decreasing 13 C magnetization. Typical plots of mono-exponential decays of the type shown in Supplementary Fig. S3 illustrate the quality of the data. The R 1ρ profiles measured as a function of effective spin-lock field power were fitted using a two-state model and two variables, the relaxation rate R 2 and the exchange rate R ex (see "Materials and Methods"); 500 runs were performed per fit, to obtain R 2 and R ex standard deviations.
C2 atoms of adenines. The C2 atoms of A 7 , A 8 , A 9 , A 19 and A 20 (Fig. 3) were studied first. Evidence of a slow conformational exchange on A 7 and A 19 emerges from the profiles of their R 1ρ (= R 2 + R ex ) rates measured as a function of effective spin-lock field power (Fig. 3). These profiles and their fits show that C2-R ex is higher for A 19 than for A 7 ( Fig. 3 and Supplementary Table S2-1 for both R 2 and R ex values). The consistency between these results and those obtained on H2 adenine atoms (Figs. 1 and 2) argues in favour of a major effect of slow motion on H2 and C2 linewidths. The on-resonance data collected on A 8 , A 9 and A 20 C2 atoms reveal invariant R 1ρ values (Fig. 3) and thus null R ex values (Supplementary Table S2-1).
C6 or C8 atoms. The C6/C8 on-resonance NMR signals were resolved for over half of the dodecamer nucleotides, comprising A 9 , but severe spectral overlaps occurred for several residues, preventing in particular the distinction of A 7 from A 19 , and A 9 from A 20 . The fitting R 2 and R ex values for the profiles of R 1ρ rates as a function of effective spin-lock field power are given in Supplementary Table S2-2. Thus, the existence of a conformational exchange is attested for A 7 and/or A 19 , A 9 and/or A 20 , and A 8 (Fig. 4), with R ex values similar to the highest ones obtained on previous DNA slow motion studies performed in appropriate pH and temperature conditions 15,17,18 . Such events are excluded or at least much more questionable for the remaining nucleotides that display flat R 1ρ profiles (examples in Supplementary Fig. S4) and therefore null or low R ex values (< 2.3 Hz) (Supplementary Table S2-2).    On-resonance R 1ρ relaxation dispersion profiles of adenine C8 atoms. R 1ρ (= R 2 + R ex ) rates of the C8 atoms of adenines were plotted as a function of the effective spin-lock field power (ω1/2π). The experiments were performed at 25 °C. The two-state model fits (solid lines) were obtained using the protocol described in "Materials and Methods". (R 2 + R ex ) standard deviations were calculated from the 500 runs carried out for each fit. The averaged R ex values are specified in each panel. . Assuming that this property is true for all the nucleotides, the present relaxation dispersion experiments help to interpret the data subject to C8 atom resonance overlaps which concern in particular the nucleotide couples A 7 /A 19 and A 9 /A 20 (Fig. 4). Thus, both A 7 and A 19 likely contribute to C8 relaxation dispersion; the R ex values, clearly higher for C1′ of A 9 than for C1′ of A 20 , advocate for a major contribution of A 9 to C8-R ex .
In sum. Null R ex values show that there is no slow exchange on C·G base pairs, which is the norm at our pH conditions (pH 6.5) 42 . Also, the R ex values calculated for the thymines paired with the five adenines do not furnish any robust evidence for motions (Supplementary Table S2). In contrast, non-null R ex values appear on the five dodecamer adenines (Table 1). It is clear that A 7 and A 19 in T 6 pA 7 ·T 18 pA 19 are submitted to a slow conformational equilibrium, according to consistent, high R ex values from C2, C8 and C1′ on-resonance experiments and to H2 and C2 linewidth broadenings. Among the three neighbouring adenines, A 8 and A 9 exhibit signs of dynamic events although the behaviour of their H2 and C2 atoms differ from those of A 7 and A 19 ; the case of A 20 is much more disputable, given the low R ex value associated with its C1′ atom.

Figure 5.
On-resonance R 1ρ relaxation dispersion profiles of adenines C1′ atoms. R 1ρ (= R 2 + R ex ) rates of the C1′ atoms of adenines were plotted as a function of the effective spin-lock field power (ω1/2π). The experiments were performed at 25 °C. The two-state model fits (solid lines) were obtained using the protocol described in "Materials and Methods". (R 2 + R ex ) standard deviations were calculated from the 500 runs carried out for each fit. The averaged R ex values are specified in each panel. www.nature.com/scientificreports/ On-resonance experiments are commonly used to provide information about the presence or absence of slow motion but by themselves they are insufficiently accurate for a viable quantification of the exchange parameters. Consequently, more sophisticated off-resonance experiments were also carried out to extract significant information about the conformational exchange process.
Off-resonance R 1ρ dispersion relaxation experiments: characteristics of slow conformational exchange. Off-resonance dispersion relaxation experiments performed at a single magnetic field were used to obtain thermodynamic, kinetic and structural information about conformational exchange. This type of experiment was applied to the C1′ resonances of A 7 , A 8 , A 9 and A 19 ; the four adenines for which signs of exchange arise from on-resonance dispersion relaxation experiments. Additional investigations focused on T 6 , C 4 , G 15 and G 21 , nucleotides associated to very low or null R ex (Supplementary Table S2). C1′ resonances were chosen because they showed large signal to noise ratios and were subject to only a few overlaps so that data could be collected for most nucleotides of the dodecamer (Supplementary Tables S2, S3).
The off-resonance data are compatible with slow motions for A 7 , A 8 , and A 19 alone (Fig. 6). A 9 , as well as the other nucleotides are associated with flat (R 2 + R ex ) profiles (examples in Supplementary Fig. S5). The A 7 , A 8 , and A 19 (R 2 + R ex ) profiles were fitted using two distinct methods, an approach that enables to assess the robustness of the resulting parameters. We implemented first a classical analytic, algebraic method 15,43-45 , described as Method 1 in "Materials and Methods", performing 1000 runs of calculations for each fit. The same data were also analysed using a completely different method recently published 3 , called here Method 2, which is based on the numerical integration of the Bloch-McConnell equations. Basically, both methods rely on a two-state exchange model and three variables, the rate exchange (k ex ), the population of the minor conformer (p E ) and the difference between the chemical shifts of major and minor conformers (Δω). Examples of fits with Method 2 are given in Fig. 6.
Applied to A 7 , A 8 and A 19 data, the exchange parameters from either Methods 1 or 2 are remarkably consistent (Tables 2 and S3), in the range of those previously published 15,[17][18][19][20] . The rather modest standard deviations associated with the parameter values from Method 1 show the reliability of each run series. Nevertheless, some subtle differences are observed for the A 7 exchange obtained from the two methods. Although it is clear that this nucleotide is submitted to a slow exchange, the A 7 -Δω value and standard deviation (4.16 ± 1.06 ppm) obtained from Method 1 appear too high compared to published data 15,[17][18][19][20] . In addition, one expects a substantial p E value, similar to that of A 19 , given their comparable high R ex values that are primarily sensitive to the excited state population 3 . For these reasons, we prefer to give more weight to the A 7 -parameters from Method 2.
The data related to A 7 and A 19 strengthen the interpretation of H2 and C2 linewidth broadening proposed above from on-resonance results. Indeed, there is now a clear parallel between H2 and C2 broadenings (Figs. 1 and 2), R ex values (Fig. 3) and p E ( Table 2), all of these parameters being more accentuated for A 19 than for A 7 . www.nature.com/scientificreports/ We therefore postulate that H2 and C2 broadenings not only reveal slow motions, but are also correlated with the excited state population. We will see below that this conclusion is very pertinent in the reconsideration of the results of previous 1D NMR experiments. The case of A 9 is intriguing and deserves a short comment. The R ex profiles are either incompatible (offresonance experiments, Supplementary Fig. S5) or compatible (on-resonance experiments, Fig. 5) with a conformational exchange. This unexpected situation has already been encountered 46 and may correspond to a scenario in which the typical limits of slow exchange are no longer valid; in particular, A 9 exchange rate could be faster than those of A 7 , A 8 or A 19 .
The results above were obtained with NMR datasets from only C1′ sugar resonances. Because of numerous resonance overlaps, off-resonance data could only be collected for C8 and C2 atoms of A 19 , which allowed the constitution of a new extended dataset, involving sugar and base atoms. The fits of this dataset were performed assuming that C1′, C8 and C2 atoms are subjected to the same conformational exchange, with one unique value for k ex and one unique value for p E . The resulting values of k ex , p E and Δω are remarkably coherent with those obtained from C1′ resonances alone (Table 3), confirming in particular that the population of A 19 minor conformer reaches 1.2/1.1%. From a methodological point of view, this approach shows that an off-resonance dataset from only one atom type is sufficient to produce reliable exchange parameters, with the obvious condition that the atom under consideration is implicated in the conformational transition of interest.
The similarities between the two-state exchange characteristics of A 7 , A 8 and A 19 ( Table 2) could be compatible with simultaneous transitions of the three nucleotides between ground and excited states, implying the transient co-existence of three successive base pair minor conformers. Thus, we completed our analyses by testing this hypothesis of synchronous motions. A collective transition implies that k ex and p E are identical for the three nucleotides and these conditions were therefore integrated as restraints in the fit calculations. This model provides reasonable values of conformational exchange parameters ( Table 4) that, at first sight, are comparable  Table correspond to the population of the minor conformer (p E ), the rate exchange (k ex ) and the difference between the chemical shifts of major and minor conformers (Δω); they were inferred from individual fits of C1′ off-resonance relaxation dispersion data according to a classical, home-made protocol (Method 1, see "Materials and Methods") or an approach developed by Al Hashimi's group (Method 2). Experimental data were collected at 600 MHz, 25 °C and pH 6.5.     Table 5). In accordance with the p-values that show the statistical significance of the tests, this means without ambiguity that the model assuming individual motions of each adenine is more effective at best representing the off-resonance data. Thus, the interpretation of the off-resonance dispersion relaxation experiments draws a picture of a block of three specific adenines that undergo slow motions between two states, the excited one representing from 0.6 to 1.2% of the conformers. While the three adenines are clustered in the T 6 A 7 A 8 ·T 17 T 18 A 19 segment, they adopt a non-synchronous regime of motions.
Nature of the excited conformer. As mentioned above, both on-and off-resonance dispersion relaxation experiments and the inferred exchange parameters agree remarkably well with those obtained by Al Hashimi's group within B-DNA sequences 15,[17][18][19][20] . The asymmetry of the NMR data in A·T base pairs, in which evidence of slow dynamics is observable on adenines but not on their thymine partners, is also in line with these earlier results. It is now accepted that these NMR-based characteristics correspond to a dynamic equilibrium of particular base pairs between the canonical Watson-Crick (WC) conformation and an excited, short-lived state of low abundance, which is the Hoogsteen (HG) pairing 3,15,18,19 . HG base-pairing corresponds to N7 purine − N3 pyrimidine and N6/O6 purine − O4/N4 pyrimidine hydrogen bonds, which imply purine and pyrimidine nucleotides in syn-and anti-configurations, respectively. Thus, the WC → HG transition does not dramatically affect the pyrimidine but requires one major change, i.e. the slow motion of the purine around the glycosidic angle χ. Here, we can confidently attribute a WC ↔ HG equilibrium to the three successive A·T base pairs composing the T 6 A 7 A 8 ·T 17 T 18 A 19 segment. It should be underlined that the HG percentages calculated here for A 7 , A 8 and A 19 are among the highest measured populations 15,[17][18][19][20] .
In addition to the purine rotation, the WC → HG transition is accompanied by some changes in sugar pucker and backbone torsion angles, which likely optimize hydrogen-bonding and stacking with neighbours 19,27,42 . In particular, a study of either WC or HG A·T containing oligomers showed that 31 P signals are up-field shifted in and around HG A·T base pairs compared to their WC counterparts 42 . Such shifts are usually interpreted as being due to the presence of more BI conformers 47,48 .
Examining high-resolution X-ray structures of free DNA suggests that another backbone alteration could also occur on α/γ angles, as briefly previously mentioned 27 . Indeed, structures containing exclusively HG pairings (PDB codes 1GQU and 1RSB, of 2.5 and 2.2 Å of resolution, respectively) enclose TpA junctions adopting unusual α/γ:g+ /g− angles, instead of the canonical g-/g+ configuration. Unusual α/γ angles are infrequent in WC B-DNA X-ray structures 49,50 because they are energetically very costly to generate 50 . Nevertheless, as rare as they are, these atypical α/γ conformations are mainly encountered in ApA steps or A·T-rich contexts (Supplementary Table S4). So, the potential ability of A·T-rich patches to adopt unusual α/γ conformations could be a factor promoting the emergence of HG base pairs. Sequence effect on TpA steps. The results presented above demonstrate that adenine H2 resonances associated with linewidth broadenings reflect the existence of a slow conformational exchange, as previously postulated 7,9,10 . However, any adenine submitted to slow motions is not associated with such anomalies that exclusively arise on TpA adenines. Accordingly, H2 linewidth broadenings are observed in the dodecamer on only those adenines belonging to TpA, A 7 and A 19 (Fig. 1). This specific phenomenon may be related to the especially high-field shifted H2 resonance of the major conformer (Fig. 1) that likely maximizes Δω, and consequently R ex -recalling that a high R ex broadens the resonance.
This intimate connection between H2 linewidth broadening and slow motion allows the re-examination of earlier NMR data. A study reported measurements of H2 linewidth broadenings on 14 oligomers containing the 16 possible immediate sequence contexts of TpA steps (i.e. NTAN) 7,8 . This systematic approach showed that broadening of H2 occurs in diverse TpA environments, the maximal values corresponding to the central adenines of (Y/R)TAA tetramer fragments (Table 6).
However, considering the nearest neighbour is insufficient to explain the slight but significant disparity of H2 linewidth broadenings of the TpA adenines that are in the same tetrameric environment in CTTTAA ATTTAA AG 8 and GCTTA ATT AAGC 7 (Table 6). Similarly, H2 broadening (Fig. 1) and slow motion characteristics differ between A 7 and A 19 in TTA 7 AA·TTA 19 A, for instance the population of the excited state (p E ) ( Table 2). These differences suggest a subtle, long-range influence of the sequence on the DNA's ability to undergo WC ↔ HG transitions. At this stage, the only certitude is that slow DNA motions are modulated at dinucleotide and tetranucleotide levels, as previously established in solution for nano-second dynamics 37,47,51 .

Discussion
This NMR study focused on a dodecamer containing the TTAAA·TTTAA fragment that was assumed to be an element facilitating histone anchoring upon nucleosome formation [34][35][36][37] . Slow conformational exchanges in the dodecamer were revealed and characterized from R 1ρ relaxation dispersion experiments. These NMR approaches also allowed a re-examination of classical 1D and 1D-T1 inversion-recovery experiments and to demonstrate the relationship between H2 linewidth broadenings and slow motions. Given that linewidth is easily observable and quantifiable, it represents an interesting means of detecting nucleotide slow dynamics, even if it appears to be limited to TpA adenine H2 protons. As expected, most nucleotides of the dodecamer do not show consistent, conclusive signs of any slow motion. However, A 7 , A 8 , and A 19 in the TA 7 A 8 ·TTA 19 fragment are specifically subjected to conformational exchange in the milli-second time range. The corresponding exchange parameters perfectly match the signature of WC ↔ HG base pair equilibrium as published by Al-Hashimi's group 15,[17][18][19][20] . Thus, we can confidently postulate that the TA 7 A 8 ·TTA 19 fragment is composed of three base pairs that transiently flip toward the HG conformation.
Although the crystallographic form of the ATT AAT ·ATT AAT hexamer 52 attests for the co-existence of three or more successive HG base pairs, our analyses agree with non-simultaneous WC ↔ HG transitions of the three A·T base pairs, which does not exclude a certain degree of cooperation between the motions. For instance, that one HG base pair locally destabilizes the double helix 42 could favour a WC → HG flip of its closest neighbours. In any case, individual motions have the effect of magnifying the time during which one HG base pair is present in the TAA·TTA tract. Thus, according to the individual HG populations calculated here (~ 1% for each of the three HG base pairs), and given the non-simultaneous character of the motions highlighted by our fits, 3% of TAA·TTA fragment contains one HG A·T base pair.
This relatively high occurrence underlines the relevance of the question of an eventual function of transient HG base pairs in the nucleosome context, remembering that the TTAAA element likely helps histone anchoring. The first idea is that DNA wrapping could involve HG base pairs but our analyses failed to detect such pairing in high-resolution X-ray nucleosome structures, conversely to what was observed on several other types of DNA-protein complexes 27 . Another proposal relates to the exploitation of shape features specific to HG base pairs upon nucleosome formation. To our knowledge, there is only one structure ensemble based on NMR data that was collected on an oligomer containing one HG m 1 A·T base pair in the Am 1 ACC·GGTT environment 42 . This oligomer has as a main characteristic a major-groove kink. Indeed, analyses of X-ray structures of DNA bound to proteins or small molecules suggested that such major-groove curvature is induced by any HG base pair in any context 27,42 . This structural particularity cannot be considered as a favourable pre-organization for the nucleosome since TTAAA bound to the histone octamer presents a minor groove curvature 34,53 . In addition to the determination of the structure of our HG oligomer itself, another factor such as long-range consequence of HG base pair could be investigated. It has to be borne in mind that, at this point in time, there is no clear indication of how those HG base pairs present in the TAA element impact on nucleosome structure or dynamics. Indeed, it is now widely accepted that nucleosome assembly is primarily modulated by sequences that introduce structural variability for helical parameters such as roll or slide along the free WC B-DNA.
The percentage of HG conformers of the three A·T base pairs in T 6 A 7 A 8 ·T 17 T 18 A 19 is unusually high (~ 1%) for such minor, short-lived states that very rarely exceed 0.5% 15,[17][18][19][20] . Multiple examples of enhanced dynamics on TA·TA adenines in the same TAA·TTA context emerge when one considers studies reporting linewidth broadenings (Table 4) 7,8 . However, the structural foundation for this specific, sequence dependent, stabilisation of HG base pairs remains unclear and poorly documented. One can only formulate hypotheses, being aware that such speculations will require future investigation. A first point concerns atypical features observed in A·T containing sequences. Thus, the photo reactivities of TT steps in TTAA·TTAA leading to T^T dimers are clearly outsized compared to those of any other dinucleotides, comprising other TT steps 53 ; this particularity is added to (but cannot be totally explained by) the TTAA·TTAA marked positive rolls and low twists attested by both NMR 37,54 and simulation 55 data. Also, as highlighted in one of the above sections, A·T rich sequences show a specific propensity to adopt unusual α/γ backbone angle conformations (Supplementary Table S4). Such unusual structural www.nature.com/scientificreports/ elements could destabilize the WC form of A·T, favour the flipping-out of adenine from the double helix via the major groove 15,56 , which preludes WC ↔ HG conversions, and, in the case of α/γ angles, stabilize the HG base pair. Another suggestion emerges from the X-ray structure of the DNA bound to the MATα2 homeodomain 57 .
In this complex, the T 6 A 7 A 8 ·T 36 T 37 A 38 fragment contains one HG base pair, A 7 ·T 37 , in which A 7 engages two hydrogen bonds with T 37 and T 36 . A similar way to increase the stability of HG conformers may occur in any TAA·TTA fragment, in the present case via hydrogen bonds between A 7 A 8 and T 16 T 17 T 18 on the one hand and between A 19 and T 5 T 6 on the other. Indeed, the effect of sequence in WC ↔ HG transition whilst incontestable remains complex and thus so far is only partially elucidated as already noted 17 . Further investigations are clearly required to clarify in particular, the effect of the tetra or even hexanucleotide contexts on the HG population. This topic is in fact essential for capturing the multiple aspects of DNA double helix functional versatility. A lot of information about nanosecond dynamics keeping WC pairing intact is already available; these are primarily sensitive to the dinucleotide sequence while modulated at the tetramer level, as shown by experimental X-ray 49,58,59 or NMR data 47,48,51,54 and modelling 38,55,60 . NMR data in particular revealed that such rapid motions are especially enhanced in G·C rich elements 47,51,54 . Several examples demonstrated their importance in the structural adjustment of DNA to its protein partners, transcription factors [61][62][63] or other proteins 37,51,64 . Beyond what occurs on WC G·C rich sequences at short timescales, A·T rich elements could be more specialized in milli-second conformational transitions, with the possibility of generating regions of Hoogsteen base-pair hot-spots that could play a topological role in genomic DNA 27 .

Materials and methods
Samples and resonance assignments. Samples were purchased as single-stranded oligonucleotides 5′-CCG CTT AAA CGC -3′ and 5′-GCG TTT AAG CGG -3′ from Eurogentec (Belgium) for unlabelled DNAs or from Eurisotop (France) for fully 15  The full assignment of 1 H and 13 C resonances of the dodecamer were previously described 37,54 . To complete the carbon assignments and notably to obtain those of the quaternary carbons of bases, we performed additional experiments on a Bruker Avance spectrometer equipped with a 5-mm triple-resonance cryogenic probe, at 600 MHz frequency or 150 MHz for 13 C experiments: (i) constant time 1 H-13 C HSQC optimized separately for aromatic and aliphatic carbons, C2 of adenine, C8 of purines, C5 and C6 of pyrimidines and C1′ of any base and, (ii) for the quaternary carbons C4, C5, C6 of adenines, 3D TROSY relayed HCCH-COSY 65 according to Hansen et al. 14 .

R 1ρ relaxation dispersion experiments.
All experiments were performed on the same spectrometer as used for assignments, at pH 6.5 and 25 °C. As specified in the Results section, the linewidth broadening of H2-proton resonances is maximal at this temperature. 1D selective 13 C R 1ρ pulse was applied as previously described 14,16 . The spin-lock powers (ω) needed to be carefully controlled, and were calibrated accordingly 14 . On-and off-resonance experiments were performed on C2, C6/C8 and C1′ atoms of the 13 C/ 15 N-labeled DNA. On-resonance data were recorded at various ω, from ~ 100 to 3,500 Hz; off-resonance data were collected at various spin-lock offset frequencies (Ω) and at three or four different spin-lock powers (ω). Details of these experiments are given in Supplementary Table S1. Ten delays were used to determine T relax , the monoexponential decrease of the magnetization: 0, 4,8,12,12,16,20,24,32,32  Note that the duplication of two delays in each experiment allowed an evaluation of errors in the measurement of the peak heights. These errors were subsequently used in Monte Carlo simulations to determine R 1ρ uncertainties. The data corresponding to Hartmann-Hahn matching were omitted from the fits of on-and offresonance measurements as previously described 14 . The 1D dispersion data were processed using NMRPipe 66 .
Equilibrium parameters from R 1ρ relaxation dispersion experiments. Our approach assumed a two-state equilibrium between a ground state G and a minor, excited state E in which p G and p E , the G and E populations, are strongly asymmetric: pG ≫ pE. The equilibrium parameters that can be extracted from relaxation dispersion experiments are pG, pE, k ex the exchange rate, Δω GE the difference between the frequencies of G and E, and R ex , the quantity added to the relaxation rate due to exchange, which depends on p G , p E , k ex and Δω GE .
On-resonance experiments. R ex was obtained by fitting the profiles of data from on-resonance experiments carried out in function of spin-lock offset powers (ω). In these experiments, θ, the angle between the effective field and the z axis, has the particular value of π/2. The fit is based on Eq. (1), a version of Eq. (2) that is simplified by using θ = π/2 to represent the condition of on-resonance experiments. For each fit, 500 runs were performed using R 1ρ values. R 1 and R 2 are the intrinsic longitudinal and transverse relaxation rates, respectively, which are supposed to be identical for the two-states G and E; ω 1 is the spin-lock power.
In this case, restraints were applied on R 2 : R 2 ≥ 16 s −1 and R 2 ≤ 19 s −1 , according to the range inferred from on-resonance experiments; Ω E and Ω G are the resonance offsets for the excited and the ground states E and G, respectively; Δω GE = Ω E − Ω G ; ω 1 is the spin-lock power related to by the relation k ex = k G + k E is the exchange rate with k G = p E k ex and k E = p G k ex , k G and k E representing the forward and reverse rate constants, respectively. For each fit, 1000 runs were performed using R 1ρ values. The R 1ρ values from off-resonance experiments were systematically and randomly varied using standard deviations derived from the fits of the exponentially decreasing intensities of the peaks as function of the relaxation time. This protocol implemented using homebased Python and R scripts enables to derive the errors associated to each exchange parameter according to a strategy described previously 67 .
Method 2 used the "Bloch-McConnell Numerical Simulator" (BMNS) developed by Al-Hashimi's group 3 , which is exclusively based on a two-state exchange model. The calculations were performed with the same constraints on the R 2 relaxation rates than for the fits described above.