Insights into the Dynamics of Specific Telomeric Single-Stranded DNA Recognition by Pot1pN

Abstract

The N-terminal oligonucleotide/oligosaccharide-binding fold domain of the Schizosaccharomyces pombe protection of telomeres 1 (Pot1) protein, Pot1pN (residues 1–187 of full-length Pot1), specifically recognizes telomeric single-stranded DNA (ssDNA) via a complex series of molecular interactions that are punctuated by unusual internucleotide hydrogen bonds. While the structure of ssDNA-bound Pot1pN provides an initial model for understanding how the Pot1pN–ssDNA complex is assembled and how specific nucleotide recognition occurs, further refinement requires knowledge of the ssDNA-free state of Pot1pN and the dynamic changes that accompany the binding of ssDNA. Using NMR strategies, we found that ssDNA-free Pot1pN adopts a similar overall protein backbone topology as ssDNA-bound Pot1pN does. Although the backbone structure remained relatively unchanged, we observed unexpected differential dynamic changes within the ssDNA-binding pockets of Pot1pN upon binding of cognate ssDNA. These studies support a model in which conformational selection and induced fit play important roles in the recognition of ssDNA by Pot1pN. Furthermore, the studies presented here provide a more comprehensive understanding of how specific nucleotide recognition is achieved by the telomere-end protection family of essential proteins.

Introduction

Telomeres are the nucleoprotein complexes that cap the ends of eukaryotic linear chromosomes and are responsible for maintaining the overall stability of genomic DNA and sustaining continued cellular proliferation (reviewed by Verdun and Karlseder¹ and Aubert and Lansdorp²). Telomeres terminate in a conserved 3′ single-stranded overhang that, left uncapped, triggers damaging cellular responses, including inappropriate processing of the telomere (i.e., end-to-end chromosomal fusions and recombination events) caused by activation of the DNA-damage response pathway.^3–8 Such events are circumvented through the capping of the single-stranded telomere ends by a family of highly specialized proteins called the telomere-end protection (TEP) proteins (reviewed by Croy et al.⁹).

TEP proteins bind telomeric single-stranded DNA (ssDNA) with high affinity and specificity, and our current understanding of how these proteins function arises from a combined structural and biochemical effort examining both full-length TEP proteins and their isolated ssDNA-binding domains (DBDs).^10–15 Although functionally similar, the DBDs of TEP proteins display a surprising lack of sequence conservation.⁹ In contrast, structural studies, focused primarily on the ssDNA-bound states of these proteins, have identified a shared, highly conserved protein fold, the oligonucleotide/oligosaccharide-binding fold (OB fold). These structures reveal that ssDNA recognition by OB folds occurs primarily through interactions with the bases of bound oligonucleotides. The rich variety of amino acids present in TEP proteins in the context of a common structural framework has led to an assortment of chemically distinct ssDNA-binding interfaces.^{11,12,14,16,17} Despite this diversity, analysis of the ssDNA-bound structures of TEP proteins has revealed two primary conserved modes of interaction with the bound oligonucleotide: hydrogen-bonding and aromatic/hydrophobic base stacking interactions. Current models explain specific nucleotide recognition based on these bound structures, and rely most heavily on hydrogen-bonding interactions. However, biochemical data indicate that the specific nucleotide recognition can occur even in the absence of extensive hydrogen bonding between the protein and the nucleotide.^13,18 Furthermore, some structurally identified hydrogen bonds that appear to be important for specific nucleotide recognition, when mutated, show no apparent effect on binding affinity.¹¹ These results suggest that the specific recognition of telomeric ssDNA is more complex than indicated by simple interpretation of the interaction details available in these bound structures. Thus, the development of a more sophisticated understanding of specific recognition requires knowledge of the structural and dynamic changes that contribute to the thermodynamics of telomeric ssDNA binding.

We chose to study the protection of telomeres 1 (Pot1) proteins, which represent the largest subfamily of TEP proteins and are found throughout nature in plants,^5,19 fungi,^7,20 and mammals,²¹ including humans.^7,22 Pot1pN (residues 1–187 of Schizosaccharomyces pombe Pot1)⁷ is an autonomously folded DBD residing in the N-terminal region of full-length Pot1 that tightly binds a minimal oligonucleotide representing a single unit of the core repeating telomeric S. pombe sequence, d (GGTTAC) (specifically recognized nucleotides in boldface).¹⁰ The ssDNA-binding interface present in Pot1pN is formed from a single face of the OB fold β-barrel (consisting of β-strands β3 and β5) as well as Loop₁₂ (connecting β1 and β2) and Loop₄₅ (connecting β4 and β5) that together form the distinctive “taco-like” ssDNA-binding groove (Fig. 1). This ssDNA-binding interface constrains the inherently flexible oligonucleotide in an unusual, compact conformation punctuated by internucleotide hydrogen bonds connecting G1 to T3, G2 to T4, and A5 to the phosphodiester backbone of T3 and T4¹⁰ (Fig. 1). In addition to these internucleotide interactions, a complex network of hydrogen bonds and aromatic stacking interactions is formed between the protein and the bound oligonucleotide. While the Pot1pN/d(GGTTAC) complex (Pot1pN_B) structure provides an initial foundation for understanding the biochemically derived nucleotide specificity, it does not provide a complete picture of how the assembly of the complex occurs and its impact on specific nucleotide recognition.

Fig. 1.

(a) Ribbon (Pot1pN) and stick (ssDNA) representation of the high-resolution crystal structure of the Pot1pN/ssDNA complex¹¹ highlighting the compact conformation of the bound oligonucleotide, d(GGTTAC). Pot1pN is shown in a ribbon representation with the α-helical elements shown in purple, the β-strand elements shown in yellow, and connecting loops shown in magenta. Each connecting loop found within the OB fold is labeled according to the β-strands that it connects (e.g., the loop connecting β−1 and −2 is designated as Loop₁₂). d(GGTTAC) is shown in stick representation and colored according to standard atomic colors. (b–d) Stick representations detailing the molecular interactions that occur between individual residues in Pot1pN (blue sticks) and d(GGTTAC) (white sticks) in the three binding pockets—G1/G2, T3/T4, and A5/C6. Hydrogen-bonding interactions formed between Pot1pN and d(GGTTAC) are designated by black dashed lines. All figures were made using PyMOL version 1.0.²³

Since structural and dynamic changes play important roles in the process of protein/ligand assembly and the thermodynamics of specific recognition, we used solution NMR-based relaxation, chemical shift perturbation (CSP), and residual dipolar coupling (RDC) experiments to probe the structural and dynamic changes that occur upon binding of Pot1pN to its minimal cognate ssDNA substrate, d (GGTTAC). Chemical shift indexing (CSI) and RDC experiments of ssDNA-free Pot1pN (Pot1pN_F) and Pot1pN_B indicate that the backbone structure of the Pot1pN_F is highly similar to the OB fold present in the Pot1pN_B crystal structure. Furthermore, these results indicate that the binding of ssDNA does not induce large global conformational changes in the structure of Pot1pN in solution. Rather, the binding of d(GGTTAC) to Pot1pN induces highly localized and varied structural and dynamic changes within the amide backbones of residues found within the loops comprising the G1/G2-, T3/T4-, and A5/C6-binding pockets. Taken together with previously reported structural and biochemical work, these results allow for the development of a more sophisticated mechanistic model for describing how the assembly of Pot1pN_B contributes to the observed nucleotide sequence specificity.

Results

Backbone atom assignment of Pot1pN_F

To gain insight into Pot1pN_F in solution, we first assigned the backbone atoms (¹HN, ¹⁵N, ¹³CO, ¹³C^β, ¹³C^α) of Pot1pN_F using standard triple-resonance NMR experiments (Supplemental Table 1) and annotated the ¹⁵N heteronuclear single-quantum coherence (HSQC) spectrum of Pot1pN_F with completed assignments (Fig. 2). The construct of Pot1pN used in these experiments contains 189 observable residues, of which 155 (82%) have been completely assigned (in ¹HN, ¹⁵N, ¹³CO, and ¹³C^α) and 8 (4%) have been partially assigned (only in ¹³CO, ¹³C^α, and, where applicable, ¹³C^β). A total of five peaks could not be assigned in the ¹⁵N HSQC spectrum of Pot1pN_F (Fig. 2) due to their absence in the triple-resonance data and/or intervening proline residues.

Fig. 2.

¹⁵N HSQC spectrum of the Pot1pN_F collected on a Varian 500-MHz spectrophotometer equipped with a room temperature probe at 30 °C. Peaks representing the individual amino acids present in Pot1pN are labeled with their corresponding residue assignments. For clarity, selected regions of the ¹⁵N HSQC spectrum are enlarged [panels (a–e)] to display selected assignments.

We were able to almost fully recapitulate the assignment coverage observed in Pot1pN_B with the exception of Ser8, Gly102, Leu165, and Asn166 (Supplemental Table 1), thus allowing for direct comparisons of almost all the residues present in each form of Pot1pN. Furthermore, we were able to assign Trp65, which was not visible in the ¹⁵N HSQC spectrum of Pot1pN_B due to exchange broadening (Supplemental Table 1).²⁴ The addition of Trp65 provides a complete residue assignment of the ssDNA-binding interface present in Pot1pN, as defined by the high-resolution crystal structure of Pot1pN_B.¹¹ Finally, as observed in the assignment of Pot1pN_B, the remaining unassigned residues in Pot1pN_F cluster either to the N- and C-termini or to a region of the protein directly adjacent to the ssDNA-binding interface (Supplemental Table 1).¹¹

CSI, RDC measurements, and CSP mapping indicate minimal structural changes upon binding of ssDNA to Pot1pN

CSI analysis has been proven to be a reliable method^25,26 for the identification of secondary structures. We used CSI analysis of ¹³C^α atoms to probe for changes that occur in the secondary structure elements present in Pot1pN upon binding of ssDNA. We found that the secondary structural elements predicted by CSI for both Pot1pN_F and Pot1pN_B are completely consistent with those found in the high-resolution structure of Pot1pN_B (Supplemental Fig. 1). Moreover, differences between the ¹³C^α CSI values for Pot1pN_F and Pot1pN_B are quite modest, indicating that little perturbation of the overall secondary structure of Pot1pN occurs upon binding of ssDNA (Supplemental Fig. 1).

We also examined the perturbations in the backbone tertiary structure of Pot1pN that accompany the binding of d(GGTTAC) using RDC measurements. RDCs between ¹H and ¹⁵N (D_NH) atoms in weakly aligned liquid crystalline solutions provide structural restraints that correlate the overall orientation of individual bond vectors within a fixed molecular tensor axis system.²⁷ Partial alignments of both Pot1pN_F and Pot1pN_B were obtained using a filamentous Pf1 bacteriophage liquid crystal system,²⁸ resulting in D_NH values ranging from 5 to 10 Hz (Supplemental Fig. 2 and Table 2). Determination of a self-consistent alignment tensor (principal component axes S_xx, S_yy, and S_zz) and respective Euler rotations (α, β, and γ) for Pot1pNF and that for Pot1pN_B were done in an iterative fashion by removing D_NH values with >3 Hz error that were rejected during the analysis of 1 million Monte Carlo simulated data sets (26 of 154 for Pot1pN_B and 29 of 147 for Pot1pN_F) (data not shown). The majority of residues that did not fit were found to be localized to nonregular secondary structure regions of the N-terminal non-OB-fold elements and loops connecting the regular secondary structure elements in the OB fold, as determined by the high-resolution crystal structure of Pot1pN_B.¹¹

Spectra collected to determine D_NH values for Pot1pN_F in low NaCl (50 mM) resulted in peak broadening and poor spectral quality. These spectral issues were alleviated by increasing the overall NaCl concentration to 400 mM (data not shown). In contrast, Pot1pN_B exhibited excellent spectral quality at both salt concentrations, indicating that the free and bound states align via different mechanisms. Alignment by Pf1 bacteriophage occurs via electrostatic interactions,²⁸ and since the electrostatic characters of Pot1pN_F and Pot1pN_B are significantly different, it is not surprising that they have unique alignment tensors (Supplemental Table 3). Therefore, to directly compare D_NH values of Pot1pN_F and Pot1pN_B, we back-calculated D_NH values based on the atomic coordinates present in the Pot1pN_B crystal structure after rotation into each specific alignment tensor. Correlation plots of experimental versus calculated D_NH values for Pot1pN_F (Supplemental Fig. 3) and Pot1pN_B (Supplemental Fig. 3) show Q-factors of ~0.42, indicating that both Pot1pN_F and Pot1pN_B are in good agreement with the 2.4-Å Pot1pN_B crystal structure, with the potential exception of those residues located in nonstructured regions that were originally excluded from the analysis.^29–32

Finally, consistent with the minimal structural changes predicted by CSI and RDC analysis, CSP mapping of Pot1pN_F and Pot1pN_B indicates that the majority of residues (118 of 155) present in Pot1pN are unperturbed by the binding of cognate ssDNA (Fig. 3a and b). Rather, we observed perturbation of residues that are located within the ssDNA-binding interface as defined in the high-resolution crystal structure of Pot1pN_B.¹¹ Taken together, these three independent experiments provide ample evidence that the global backbone conformation of the OB-fold present in Pot1pN is preserved in the presence and in the absence of cognate ssDNA.

Fig. 3.

¹⁵N–¹H CSP analyses of Pot1pN_F and Pot1pN_B. (a) Differences in the composite ¹⁵N–¹H chemical shifts for Pot1pN_F and Pot1pN_B plotted as a function of residue number. The secondary structure elements of the high-resolution X-ray crystallographic structure of Pot1pN_B¹¹ are shown above for reference. (b) Surface (left) and ribbon (right) representation of the high-resolution X-ray crystallographic structure of Pot1pN_B¹¹ with the individual residues colored according to their respective chemical shift differences: >0.6 ppm in red; 0.6–0.3 ppm, orange; <.3–0.1 ppm, yellow; and <0.1 ppm, white. Nonassigned residues in either Pot1pN_F or Pot1pN_B are shown in gray. All figures were made using PyMOL version 1.0.²³

Binding of d(GGTTAC) alters the fast-timescale dynamic behavior of Loop₁₂ but not that of Loop₄₅

The dynamic behaviors of backbone amides in Pot1pN_F and Pot1pN_B on the picosecond–nanosecond timescale were assessed using standard two-dimensional ¹⁵N HSQC-based relaxation experiments [¹{H}–¹⁵N steady-state heteronuclear Overhauser enhancement (HetNOE), T₁, and T₂] collected at two field strengths (Fig. 4 and Supplemental Table 4). We were able to obtain R₁ (spin–lattice relaxation rate), R₂ (spin–spin relaxation rate), and HetNOE values for all the backbone amides present in Pot1pN_F except Ile54. In the case of Pot1pN_B, R₁, R₂, and HetNOE values were obtained for all backbone amides except Leu23, Thr38, Arg48 (HetNOE only), Ile54, Ser70 (HetNOE only), Tyr82 (R₂ only), Leu99 (R₂ only), Lys136 (HetNOE only), and Lys187 (HetNOE only). In both cases, these select amide peaks exhibited poor peak dispersion and/or intensity and were excluded from our dynamics analysis. As expected, the terminal amides (1–30 and 188–197) and the majority of those present in the loops connecting individual β-strands in Pot1pN_F and Pot1pN_B experience dynamic motions on the fast timescale that are unaffected by the binding of d(GGTTAC) and are consistent with unstructured protein elements (Fig. 4 and Supplemental Fig. 4).

Fig. 4.

Parameters determined from ¹⁵N relaxation experiments. HetNOE (upper panel), R₁ (middle panel), and R₂ (lower panel) are plotted as a function of residue number for Pot1pN_F (black) and Pot1pN_B (magenta) collected at 500 MHz at 30 °C. Associated errors were derived as follows: HetNOE errors represent the standard deviation from triplicate experiments, while R₁ and R₂ errors were derived from the fitting of 300 simulated Monte Carlo data sets that included the noise errors estimated from duplicate points.³³ The secondary structure elements of the high-resolution X-ray crystallographic structure of Pot1pN_B¹¹ are shown above for reference.

The majority of amides present in Pot1pN_F and Pot1pN_B were insensitive to changes in their respective picosecond–nanosecond dynamic motions upon binding of d(GGTTAC), with the notable exception of the ssDNA-binding Loop₁₂ (Fig. 4). For Loop₁₂, significant increases in the HetNOE ratios (Fig. 4) (average change of Loop₁₂ amides from 0.54 in Pot1pN_F to 0.73 in Pot1pN_B) indicate a decrease in the dynamic behavior of these amides in response to ssDNA binding, moving from a partially disordered state to a highly ordered structure (Fig. 4 and Supplemental Fig. 4). Surprisingly, we found that the backbone amides in Loop₄₅, which provides the other half of the ssDNA-binding interface (as described in Fig. 1a), undergo a modest change in the HetNOE ratios (average change from 0.61 in Pot1pN_F to 0.67 in Pot1pN_B) upon binding of d (GGTTAC) (Fig. 4 and Supplemental Fig. 4), suggesting that, unlike Loop₁₂, Loop₄₅ remains moderately flexible in the presence and in the absence of ssDNA (Fig. 4 and Supplemental Fig. 4). The dynamic characteristics of these amides in Loop₁₂ and Loop₄₅ in Pot1pN_B observed in our studies correlate well with their respective B-factor values reported in the Pot1pN_B crystal structure.¹¹ These results indicate that the two ssDNA contacting loops, Loop₁₂ and Loop₄₅, which form the “taco-like” ssDNA-binding groove in Pot1pN, undergo differential dynamic and structural changes upon binding of d(GGTTAC).

Model-free analyses of Pot1pN_F and Pot1pN_B

Model-free analyses of relaxation^34,35 for Pot1pN_F and Pot1pN_B were performed using R₁, R₂, and HetNOE values collected at 500 and 600 MHz (Supplemental Table 5). Fitting of the data using ModelFree³⁶ resulted in the majority of residues fitting to either model 1 or model 2 as described by Mandel et al.,³⁶ indicating that the observed relaxation properties of Pot1pN_F and Pot1pN_B could be explained by simple motional models (Supplemental Table 5). The calculated overall molecular correlation times for Pot1pN_F (12.5 ns) and Pot1pN_B (13.4 ns) are consistent with the masses of free and bound Pot1pN (Supplemental Table 6). We observed a dramatic increase in the S² (generalized order parameter) values for many of the backbone amides present in Loop₁₂ and a less dramatic increase in the S² values for those present in Loop₄₅ (Fig. 5). We also detected dynamic changes in elements of the OB fold that do not make direct interactions with the bound oligonucleotide: β1, which directly precedes L₁₂ (Val43, Asn44, Leu45, Phe46, Gly47, Ile48, Phe52, and Thr53), and the amides preceding and forming the C-terminal α-helix (Arg150, Leu151, Gly155, Asp156, Glu159, Gln160, and Ala162). Although these two regions do not directly interact with ssDNA, they do interact extensively with each other through predominantly hydrophobic interactions. Our relaxation data suggest that the loss of dynamic motion in Loop₁₂ upon binding of d(GGTTAC) leads to a simultaneous loss of dynamic motion in β1 that is then propagated into the C-terminal helix (Fig. 5).

Fig. 5.

ModelFree S² values for the backbone amides present in Pot1pN_F (black) and Pot1pN_B (magenta) derived from HetNOE, R₁, and R₂ experiments collected at 500 and 600 MHz at 30 °C. The upper panel contains S² values calculated for Pot1pN_F (black) and Pot1pN_B (magenta) plotted as a function of residue number. Associated errors in S² values were calculated by the fitting of 300 Monte Carlo simulated data sets with ModelFree.³⁶ The lower panel is a bar graph showing the per-residue differences between the Pot1pN_F and Pot1pN_BS² values plotted as a function of residue number. Residues falling within the average error of (shaded black box) the calculated S² values were found to have no difference in the overall structure (black bars). Residues outside the average error were found to be less structured either in the absence (blue bars) or in the presence (red bars) of d(GGTTAC).

¹⁵N CPMG relaxation dispersion indicates a significant decrease in the overall dynamic behavior of Pot1pN on the microsecond–millisecond timescale upon binding d(GGTTAC)

Dynamic behavior on the microsecond–millisecond timescale can be associated with the conformational exchange of backbone amides between an NMR observed and often highly populated (visible) state and an invisible and generally sparsely populated (invisible) state of the protein (reviewed by Korzhnev and Kay³⁷). Termed relaxation dispersion (R_ex), molecular motions on these timescales directly increase the apparent R₂ values of affected amides. We measured R_ex directly using a modified R₂ experiment in which R₂ values of backbone amides in Pot1pN_F and Pot1pN_B were measured for a fixed period, during which CPMG (Carr–Purcell–Meiboom–Gill) pulses were applied at variable time delays (τ_CPMG) (20 µs to 1 ms in our experiment).³⁸

Based on our R₂-CPMG analysis (Supplemental Fig. 5 and Table 7 show data and fits for selected amides), we found that in the absence of bound oligonucleotide, there is a total of 40 backbone amides that experience microsecond–millisecond dynamic motion, of which 28 (70%) were found to cluster to the G1/G2- and T3/T4-binding pockets, with the remaining 12 located in the non-OB-fold structural elements found at the N- and C-termini of Pot1pN (Fig. 6a and Supplemental Fig. 6). The binding of d(GGTTAC) to Pot1pN results in a significant decrease in the overall dynamic behavior of Pot1pN on the microsecond–millisecond timescale, as only 26 backbone amides display microsecond–millisecond dynamic motions (Fig. 6b and Supplemental Fig. 6). While data collected for Pot1pN_F and Pot1pN_B at low field (600 MHz) were not of sufficient quality to perform an in-depth analysis of the “invisible” states sampled by these dynamically active backbone residues, mapping of these sites onto the Pot1pNB structure allows us to propose how such dynamic changes can contribute to specific nucleotide recognition. Specifically, we found that all the backbone amides present in the G1/G2-binding pocket that showed microsecond–millisecond dynamic behavior in Pot1pNF do not show the same motions in Pot1pN_B. Additionally, we also observed a significant loss of dynamic motion in amides present within the T3/T4-binding pocket upon binding d(GGTTAC), specifically with respect to those found in Loop12 and Loop34. Finally, although the overall dynamic behavior of Pot1pNB was significantly decreased, we did observe a few amides that became dynamic upon binding d(GGTTAC) (Fig. 6b). Of these amides, Phe88 and Gln84 make contact with the bound oligonucleotide in the Pot1pNB crystal structure via aromatic stacking and hydrogen-bonding interactions, respectively. Taken together, these results suggest that the binding of d(GGTTAC) to Pot1pN directly impacts the overall dynamic behavior of backbone amides on the microsecond–millisecond timescale, with dynamic motion being quenched, especially in those loops that mediate specific nucleotide recognition.

Fig. 6.

Ribbon (Pot1pN) and stick (ssDNA) representation of the high-resolution crystal structure of the Pot1pN–ssDNA complex¹¹ highlighting the residues that undergo dynamic motion on the microsecond–millisecond timescale in both Pot1pN_F and Pot1pN_B. (a) Residues experiencing microsecond–millisecond dynamic motion in Pot1pN_F only are shown in navy, those in both Pot1pN_F and Pot1pN_B are shown in magenta, and those that showed no motion are shown in white. (b) Residues experiencing microsecond–millisecond dynamic motion in Pot1pN_B only are shown in navy, those in both Pot1pN_F and Pot1pN_B are shown in magenta, and those that showed no motion are shown in white. In both cases, residues shown in gray represent unassigned residues in both Pot1pN_F and Pot1pN_B. All figures were made using PyMOL version 1.0.²³

Discussion

TEP proteins are a biologically important family that specifically recognizes the 3′ ssDNA overhangs present at telomere ends and are critical to proper telomere function (reviewed by Croy et al.⁹). Our current understanding of the underlying mechanisms that drive specific telomeric ssDNA recognition has been restricted due to the limited data sets addressing the ssDNA-free form of the TEP family. To date, two studies of ssDNA-free TEP proteins have been reported, but their overall ability to provide further insight into the process of ssDNA recognition has been hindered by limited data in the case of Cdc13-DBD³⁹ and the presence of a heterogeneous TEBPα₃₅-d(T₄G4) complex in which an unbound protein co-crystallized with bound complexes.¹⁵ Therefore, we examined Pot1pNF by NMR using RDC experiments and CSI, which reveal that the overall backbone topology of the OB fold present in the Pot1pN bound state is essentially unchanged in the absence of bound oligonucleotide. These observations are further supported by our ¹H–¹⁵N CSP data, which show that the binding of ssDNA does not lead to widespread perturbation of residues throughout Pot1pN. Rather, we found that changes are limited to residues present within the ssDNA-binding interface as defined by the Pot1pNB crystal structure.¹¹ These results indicate that the binding of telomeric ssDNA to Pot1pN effects minimal change in the overall protein topology. In contrast, the ligand d(GGTTAC) is inherently disordered when free in solution²⁵ and undergoes a substantial structural rearrangement to form the compact configuration observed in the Pot1pN_B crystal structure.¹¹ Taken together, these studies highlight an emerging shared mechanism for the recognition of telomeric ssDNA within the TEP family in which a disordered oligonucleotide undergoes a large structural rearrangement to fold into a largely ordered protein surface.

The absence of large protein conformational changes upon binding of ssDNA suggests that perhaps more subtle changes in dynamic behavior contribute to the observed nucleotide specificity conferred by Pot1pN. Two nonexclusive mechanisms described by these subtle dynamic motions have been implicated in mediating specific nucleotide recognition in RNA–protein interactions: conformational selection and induced fit (reviewed by Williamson,⁴⁰ Leulliot and Varani,⁴¹ and Boehr and Wright⁴²). Conformational selection involves the transient exchange of thermodynamically preferred structural conformations in the absence of ligand. Such selection involves dynamic motions that are generally thought to occur on a slower (microsecond–second) timescale. The structuring of disordered protein and/or ligand regions that occurs in induced fit can alter motions that occur on a faster (picosecond–nanosecond) timescale (reviewed by Mittermaier and Kay⁴³). The use of NMR allows us to directly assess backbone motions occurring on both of these timescales, thus giving insight into the interaction mechanisms that are involved in the assembly of Pot1pN_B.

Formation of the G1/G2-binding pocket

The G1/G2-binding pocket is formed by the coordinated interactions of β4 and Loop₅¹¹ (Fig. 1b). In the absence and in the presence of ssDNA, the amide backbone of residues present within this binding pocket exhibits picosecond–nanosecond dynamic motions that are typical of well-ordered protein structural elements. Although we observed little change in picosecond–nanosecond motions, amides found within the G1/G2-binding pocket do experience microsecond–millisecond timescale motions in the absence of ssDNA that are quenched upon binding (Fig. 6a and b). Such dynamic attributes support a model in which the formation of the G1/G2-binding pocket occurs via conformational sampling, in which the final ssDNA-bound protein conformational state is selected from a dynamic ensemble of amide backbone conformations.⁴⁴ In addition to conformational sampling, the binding of ssDNA to Pot1pN is also accompanied by a large structuring of the oligonucleotide, indicating that induced-fit mechanisms may also be employed.

Structural and mutagenesis data support a model in which the specific recognition of G1 arises in part from the complex hydrogen-bond network formed between the binding pocket and the oligonucleotide.^10,11 Our dynamics data build upon these initial observations by suggesting that the well-ordered nature of the G1/G2-binding pocket in the absence of ssDNA is thermodynamically primed for optimum hydrogen-bonding interactions with guanosine at position 1. The presence of a guanosine-specific binding pocket also lends insight into previously published mutagenesis data that show the closely related inosine (which is identical with guanosine but lacks its 2′ exocyclic amine) is thermodynamically accommodated by Pot1pN, while cytosine is not.¹¹ In contrast, the specific recognition of guanosine at position 2 arises from hydrogen-bonding interactions formed with thymine at position 4 and not from those provided by the G1/G2-binding pocket.¹¹ Consistent with our dynamics studies, the guanosine-primed G1/G2-binding pocket may play an important role in the induced fitting of G2 by providing a well-ordered template that optimizes the positioning of G2 with hydrogen-bonding interactions with T4.

Assembly of the T3/T4-binding pocket

Unlike the G1/G2-binding pocket, which is well ordered in the presence and in the absence of ssDNA, picosecond–nanosecond motions show that backbone amides present in Loop12, which forms the centerpiece of the T3/T4-binding pocket, are significantly disordered in the absence of ssDNA and ordered upon binding (Fig. 4). This loop is well structured in Pot1pN_B, suggesting that formation of the T3/T4-binding pocket proceeds via a “mutually” induced-fit type of mechanism that requires the coordinated co-folding of both Loop12 and Loop34 with the unstructured oligonucleotide^40,45 (Fig. 4). The question then arises as to how the T3/T4-binding pocket and ssDNA oligonucleotide co-fold with one another. One possible mechanism is that the T3/T4-binding pocket and d(GGTTAC) fold into the conformation observed in the Pot1pN_B crystal structure in an unbiased fashion based on a random sampling of an infinite number of conformational states. However, the microsecond–millisecond dynamic behavior exhibited by the backbone amides present in Loop₁₂, β2, β3, and Loop₃₄ supports a more complex mechanism in which the mutually induced fitting of oligonucleotide and protein is augmented by conformational sampling.⁴⁴

Examination of the static crystal structure suggests that the specific recognition of T3 arises from hydrogen-bonding interactions offered solely by the T3/T4-binding pocket, while T4 arises from interactions formed with both G2 and the binding pocket.¹¹ Our picosecond–nanosecond dynamics data indicate that, unlike the G1/G2-binding pocket, the important interactions that dictate the specific recognition of thymine at positions 3 and 4 are not presented within the context of a rigid structural template. Rather, microsecond–millisecond motions support a model in which these interactions are presented in the context of a transiently sampled ensemble of alternate conformations that are formed in the absence of ssDNA. Mechanistically, the formation of the T3/T4-binding pocket is similar to the formation of the G1/G2-binding pocket, as a thermodynamically preferred conformation, sampled from the ensemble, guides the correct induced fitting of the protein and oligonucleotide. This template may also provide positional regulation of T4 that results in an optimum orientation for forming the thermodynamically important interactions with G2.

Formation of the A5/C6-binding pocket

The binding pocket for A5/C6 is formed from the interactions of Loop₁₂, β3, β4, Loop₄₅, and β5 (Fig. 1d). Analysis of the picosecond–nanosecond timescale dynamics data shows that backbone amides present in Loop₄₅ are partially disordered in Pot1pN_F and that, unlike Loop₁₂, Loop₄₅ undergoes little change in dynamics upon binding of d (GGTTAC) (Fig. 4). In addition, Loop₄₅ contains only a few backbone amides within the A5/C6-binding pocket that exhibit dynamic motion on the microsecond–millisecond timescale in the absence of ssDNA, Arg56, Arg113, Tyr115, and Gln120, all of which make contact with C6 (Figs. 1d and 6b). The binding of ssDNA leads to differential dynamic effects on the amide backbones of these residues. Arg56, Arg113, and Tyr115 undergo drastic dynamic losses, while the amide backbone of Gln120 remains dynamic. In addition, we found that the binding of d(GGTTAC) to Pot1pN induces a microsecond–millisecond dynamic behavior in the amide backbone of Gln84, which hydrogen bonds with C6.

Biochemical studies have shown a clear thermodynamic preference for adenosine at position 5 by Pot1pN, while preference for cytosine at position 6 is greatly reduced.¹⁰ Such thermodynamic characteristics are consistent with our dynamics data showing that Loop₄₅, which directly contacts C6 in d(GGTTAC) (Fig. 1d), remains partially disordered and flexible in the presence and in the absence of bound oligonucleotide (Fig. 4). These dynamic motions reveal a malleable binding pocket that is able to accommodate nucleotides other than cytosine at position 6 with a lessened thermodynamic consequence. In addition to the retention of dynamic motion on the picosecond–nanosecond timescale, we also found that the binding of d (GGTTAC) induces the amide backbones of residues that hydrogen bond to C6 to become or remain dynamically active on the microsecond–millisecond timescale (Fig. 6b). These dynamic properties support a model for the specific recognition of A5 that arises from interactions not offered by the A5/C6-binding pocket. Instead, it is likely that the specific recognition of A5 results from the internucleotide interactions with the phosphodiester backbones of T3 and T4. This model is supported by the crystal structure of Pot1pN bound to d(GGTTA) (which does not contain cytosine at position 6) that shows minimal perturbation in the overall positioning of A5 relative to the Pot1pN_B crystal structure.¹¹ Such a mechanism for recognition of A5 requires exquisite control over the final bound conformation of the T3/T4-binding pocket and bound oligonucleotide.

The conformational selection and induced-fit models that we predicted to be important in the specific recognition of ssDNA by Pot1pN have also been proposed to be important in the context of the interaction of the human spliceosomal protein U1A with its cognate substrate, the polyadenylated inhibition element from a 3′ untranslated region of the U1A pre-mRNA.⁴⁶ Specific recognition of mRNA by U1A is mediated in part by residues found within loop 3 (L₃). In the absence of mRNA, L₃ exhibits backbone dynamic motions that are consistent with a disordered loop undergoing conformational sampling. Upon binding mRNA, L₃ becomes structured, effectively quenching the conformational sampling observed in the mRNA-free state. In addition to Pot1pN and U1A, similar dynamic observations have been ascribed to the recognition of RNA by a mouse RNA-binding protein, Musashi1.⁴⁷ These studies together indicate that, like specific antigen–antibody interactions,^48–50 the specific recognition of single-stranded nucleic acids is a complex and diverse event that incorporates physical interaction elements found in both the induced-fit model and the conformational selection model. This structural and dynamic analysis highlights the ability of NMR to complement and augment the currently available high-resolution X-ray crystallography structures, thus providing a more comprehensive interpretation of specific recognition of ssDNA by this important class of proteins.

Materials and Methods

Chemicals, reagents, and proteins

All chemicals and reagents were obtained from Fisher Scientific (Pittsburg, PA) unless otherwise indicated. HPLC-purified d(GGTTAC) was commercially synthesized (Operon, Huntsville, AL) and reverse-phase purified over a semipreparative C18 column (Grace Vydac, Hesperia, CA) in 1% triethanolamine acetate/H₂O and eluted with a 5%–40% gradient of 1% triethanolamine acetate/19% H₂O/80% acetonitrile. Chromatography materials were purchased from GE Healthcare (Piscat-away, NJ). Pf1 bacteriophage²⁹ was a generous gift from Professor Arthur Pardi. Finally, the plasmid-encoding Pot1pN was graciously provided by Professor Thomas Cech.⁷

Expression and purification of Pot1pN

The 22-kDa (His)₆-Pot1pN DBD from the S. pombe Pot1 protein was expressed in Escherichia coli, purified, and stored according to the protocols established by Lei et al.¹⁰ with final purified yields of 25 mg/L. Uniform ¹³C–¹⁵N or ¹⁵N labeling was performed by expressing each protein in modified minimal M9 media.³⁹ Complexes of ¹⁵N-labeled Pot1pN and d(GGTTAC) at 1.25 mM in NMR sample buffer (50 mM KH₂PO₄, pH 6.15, 50 mM NaCl, 1 mM d-DTT, and 10% D₂O) contained an excess of 0.25 molar equivalents of d(GGTTAC) to ensure that Pot1pN was present in a homogeneous ssDNA-bound complex. Due to the aggregation tendencies of Pot1pN_F, NMR experiments were conducted at a final Pot1pN concentration of 350 µM. Finally, additional residues present in the N-terminal 6×His tag were identified with negative values to be consistent with the residue numbering scheme presented by Lei et al.¹¹

NMR backbone assignment of Pot1pN

Standard Varian BioPack pulse sequence-derived two-and three-dimensional sensitivity-enhanced, gradient-selected non-TROSY assignment experiments [¹⁵N and ¹³C HSQC, HNCA, HN(CO)CA, HNCACB, CBCA(CO) NH, HN(CA)CO, and HNCO]⁵¹ were acquired at 30 °C on a Varian Inova 500-MHz spectrometer equipped with a Nalorac HCN warm probe containing single-axis Z-gradients. Collected data sets were processed using NMRPipe,³³ and backbone resonance assignments were manually made using CCPNMR Analysis v1.5.⁵²

CSI analyses of Pot1pN_F and Pot1pN_B

Secondary structural elements of both forms of Pot1pN were determined using per-residue ¹³C^α CSI methods previously published.^25,26 Determination of CSI parameters for Pot1pNF and that for Pot1pN_B were calculated using CCPNMR Analysis v1.5 software package-supplied macros.⁵² CSI values were determined by comparing the ¹³C^α values obtained for both Pot1pNF and Pot1pN_B with those obtained for respective residues present in a random-coil pentapeptide sequence, GGXGG.⁵³

RDC measurements of Pot1pN_F and Pot1pN_B

Measurements of RDCs (D_NH) were determined using liquid crystal solutions containing 15 mg/mL of Pf1 bacteriophage in NMR buffer containing 400 mM NaCl.^28,54 One-bond ¹H–¹⁵N J-couplings in the absence (J_HN) and in the presence (J_NH+D_NH) of alignment media for Pot1pN_F and Pot1pN_B were collected at 30 °C on a Varian Inova 500-MHz spectrometer equipped with a Nalorac HCN probe containing single-axis Z-gradients. Data were collected in an interleaved fashion using a two-dimensional doublet-selective, sensitivity-enhanced ¹H–¹⁵N HSQC pulse sequence⁵⁴ and processed using NMRPipe.³³ Contributions of D_NH to the couplings observed in the aligned state were calculated by subtracting the aligned-state couplings from the nonaligned-state couplings using the CCPNMR Analysis v1.5 software package.⁵² Principal component axes S_xx, S_yy, and S_zz of the alignment tensor and the protein orientation with respect to the molecular frame (Euler angles α, β, and γ) for Pot1pN_B and Pot1pN_F were determined using the atomic coordinates present in the Pot1pN_B crystal structure using REDCAT.⁵⁵

¹⁵N relaxation experiments

Picosecond-nanosecond dynamic (HetNOE, R₁, and R₂⁵⁶) and microsecond-millisecond dynamic (R₂-CPMG experiments³⁸) experiments were performed at 30 °C at two independent field strengths (500 and 600 MHz for HetNOE, R₁, and R₂ experiments and 600 and 900 MHZ for R₂-CPMG experiments) on Varian Inova spectrometers equipped with a Varian HCN warm probe containing single-axis Z-gradients, except for R₂-CPMG experiments, which were collected at 600 MHz using a Varian cold probe containing single-axis Z-gradients. However, due to the poor sensitivity for the R₂-CPMG experiments collected at 600 MHz (data not shown), only 900-MHz data were analyzed.

HetNOE data were collected with and without a 3-s ¹H saturation period with a constant recycle delay of 5 s. ¹⁵N R₁ and R₂ experiments were recorded in a pseudo-randomized, interleaved fashion with a constant recovery delay of 4 s with variable T₁ relaxation delay periods of 0, 52.24, 104.48*, 208.96*, 417.92, 626.88, 835.84, 1253.76, 1606.78, and 1959 ms and variable T₂ relaxation delays of 0, 30.93*, 46.89, 61.86, 77.32, 92.78*, 123.64, 154.64, and 170.10 ms (the asterisk represents repeated data points implemented at randomized positions). Heat-compensated R₂-CPMG experiments were collected in a pseudo-randomized fashion with a constant T₂ relaxation delay of 40 ms, a recovery delay of 2.5 s, and pseudo-randomized application of 180° refocusing CPMG pulses at frequencies of 0, 50*, 100, 200, 250, 400, 500, 550, 700, 750*, 800, and 950 Hz. Resultant spectra were processed using NMRPipe,³³ and data were treated in an identical fashion using squared cosine-bell apodization functions, linear prediction to 2× points in the indirect dimension, and zero filled in both dimensions to 2× points. Data from the R₂-CPMG experiments were treated in a similar fashion, with the exception that Gaussian apodization functions were used. Peak volumes were determined by nlinLS,³⁴ with errors representing the Gaussian distribution of the root-mean-squared deviation of the peak volume to the noise threshold of the experiment.

Individual backbone amide HetNOE ratios were derived from peak volume data by taking the ratio of the non-¹H saturated peak volume to the ¹H saturated peak volume, with errors (1σ) derived from triplicate experiments collected at each magnetic field strength. R₁ and R₂ values were calculated using the peak analysis and quantification macros built into NMRPipe,³⁴ with errors derived from a series of exponential fits of 300 synthetic data sets generated by Monte Carlo simulation.³⁴ R₂-CPMG experiments were processed using macros developed by Professor Lewis Kay, and normalized peak volumes were used to calculate the effective R₂ (R_2eff) values (1/s) as a function of CPMG frequency for each residue according to Eq. (1):

$$R_{2\mathit{\text{eff}}}(v_{\mathit{\text{CPMG}}})=((-1/T_{2\text{delay}})(\text{ln}(\text{Normalized Peak Volume})))$$ (1)

All data fitting was done using a hyperbolic tangent CPMG function embedded in CPMGFit^†. Calculated errors for R_ex and k_ex were derived from jacknife simulation of the individual data sets.

Model-free analyses of Pot1pN_F and Pot1pN_B

Relaxation parameters obtained for Pot1pN_F and Pot1pN_B collected at 500 and 600 MHz were subjected to the Lipari–Szabo model-free analysis^34,35 using the Model-Free 4.15 and FAST-ModelFree software packages.^36,57 Estimations of the global isotropic correlation time (τ_c) for each form of Pot1pN were derived from the 1σ adjusted mean R₂/R₁ values collected at 500 MHz. Initial estimations of the axially symmetric diffusion tensor were estimated from the Pot1pN_B crystal structure¹¹ and the measured 1σ adjusted mean R₂/R₁ ratios⁵⁸ using pdbi-nertia and R2R1_diffusion^‡. An axially symmetric diffusion tensor was chosen to describe diffusion of Pot1pN_F and Pot1pN_B and refined using data sets that eliminated the terminal highly dynamic residues (1–30 and 187–197) as well as residues for which all three of the ¹⁵N relaxation parameters (R₁, R₂, and HetNOE) could not be accurately determined. Calculation of ModelFree parameters was done by simultaneously fitting all ¹⁵N relaxation data collected at 500 and 600 MHz to one of five motional models that were selected based on the methods reported by Mandel et al.³⁶ Errors reported in S² values were derived from 300 synthetic data sets generated by Monte Carlo simulation.

Abbreviations used

CSA

chemical shift anisotropy

CSI

chemical shift indexing

CSP

chemical shift perturbation

CPMG

Carr-Purcell-Miboom-Gill

DBD

ssDNA-binding domain

D_NH

residual dipolar coupling

HSQC

heteronuclear singlequantum coherence

J_NH

J-coupling

NMR

nuclear magnetic resonance

NOE

nuclear Overhauser effect

OB

oligonucleotide/oligosaccharide binding

Pot1

protection of telomeres 1

Pot1pN

residues 1–187 of full-length Pot1

Pot1pN_F

ssDNA-free form of Pot1pN

Pot1pN_B

Pot1pN/d (GGTTAC) complex

Q_F

Q-factor

R₁

spin–lattice relaxation

R₂

spin–spin relaxation

RDC

residual dipolar coupling

R_ex

relaxation dispersion

RMS

root mean squared difference

ssDNA

single-stranded DNA

HetNOE

¹H¹⁵N steady state heteronuclear NOE

S²

generalized order parameter

S_zzS_yyS_zz

principal component axes of molecular alignment tensor

τ_c

overall molecular correlation time

τ_e

local correlation time