Non-specific recognition is achieved in Pot1pC through the use of multiple binding modes

Summary

Pot1 is the protein responsible for binding to and protecting the 3’ single-stranded DNA (ssDNA) overhang at most eukaryotic telomeres. Here we present the crystal structure of one of the two OB-folds (Pot1pC) that make up the ssDNA-binding domain in S. pombe Pot1. Comparison with the homologous human domain reveals unexpected structural divergence in the mode of ligand binding that explains the differing ligand requirements between species. Despite the presence of apparently base-specific hydrogen bonds, Pot1pC is able to bind a wide range of ssDNA sequences with thermodynamic equivalence. To address how Pot1pC binds ssDNA with little to no specificity, multiple structures of Pot1pC bound to non-cognate ssDNA ligands were solved. These structures reveal that this promiscuity is implemented through new binding modes that thermodynamically compensate for base-substitutions through alternate stacking interactions and new H-bonding networks.

Introduction

The execution of many biological activities requires the specific binding of a protein to a double or single-stranded nucleic acid. Structural and biochemical studies in several model systems have revealed how this specificity can be achieved(vonHippel and Berg, 1986; Freemont et al., 1991). Canonically, sequence-specific discrimination is thought to occur largely through the recognition of a pattern of hydrogen-bond donor and acceptor atoms characteristic of nucleotide sequence. Less well characterized, however, are the equally important nucleic acid recognition events that require indiscriminate recognition, such as the binding exhibited by SSB(Raghunathan et al., 2000), non-specific nucleases(Li et al., 2003), DNA polymerase(Ollis et al., 1985), and PCNA(Krishna et al., 1994). Studies of this type of recognition suggest that binding is achieved by non-specific stacking/hydrophobic interactions and/or interactions with the phosphate backbone(Record et al., 1976). While these concepts were largely inferred from the study of double-stranded DNA-binding proteins, the same principles have been applied to single-stranded nucleic acid binding proteins(Messias and Sattler, 2004; Cléry et al., 2008).

The family of proteins that bind the 3’ single-stranded DNA (ssDNA) overhangs of telomeres provides an ideal system for the characterization of ssDNA recognition. Telomere-end protection (TEP) proteins bind the ssDNA overhang with high affinity to insulate the DNA from damage response elements and to regulate access by telomerase. Disruption of this interaction can lead to activation of DNA-damage-response machinery, which can cause chromosomal fusions and lead to senescence and cell death(Wu et al., 2006; Denchi and de Lange, 2007; de Lange, 2009; Rai et al., 2010). TEP proteins must therefore efficiently recognize the specific G-rich sequence present at telomeres. As nucleotide sequence is not conserved between species, orthologous TEP proteins use different mechanisms to recognize their specific overhang (Lewis and Wuttke, 2012). S. nova (formerly known as O. nova) and human telomeres are composed of perfect repeats of short signature motifs and their TEP proteins (TEBPα/β and hPOT1 respectively) use multiple OB-folds to bind ~2 repeats in a specific fashion(Horvath et al., 1998; Lei et al., 2004). Specificity is achieved in these systems through the orientation of the bases toward the protein to form base-specific H-bonds. In contrast to the uniformity of most telomeres, S. cerevisiae telomeres are overall GT-rich, but do not have a strictly conserved repeat (Forstemann et al., 2000). The S. cerevisiae TEP protein, Cdc13, meets this challenge through the use of a single OB-fold that specifically recognizes a d(GXGT) motif at the 5’ end of the minimal DNA required for high affinity binding. It also interacts with the remaining seven nucleotides required for high affinity binding, but does so non-specifically to accommodate the sequence heterogeneity typical of S. cerevisiae telomeres(Mitton-Fry et al., 2002; Anderson et al., 2003; Eldridge et al., 2006). While Cdc13 binds the entire 11-nt ligand in a bases-inward fashion, a reduced number of H-bonds may explain the lack of specificity for the 3’ portion of the sequence.

The fission yeast S. pombe also has degenerate telomeres that can be described by the sequence d(GGTTAC)(A/AC)_0–1(G)_0–7 (Trujillo et al., 2005; Leonardi et al., 2008). The ssDNA overhang is bound by the TEP protein SpPot1, which has a domain organization analogous to that of hPOT1(Theobald and Wuttke, 2004; Croy et al., 2006). This includes a DNA-binding domain (Pot1-DBD) that is composed of two OB-folds (Pot1pN and Pot1pC in S. pombe and hOB1 and hOB2 in humans) (Supplemental Fig. S1A). Pot1pN is structurally very similar to its human counterpart hOB1(Lei et al., 2004), but recent biochemical work suggests accommodation of degenerate telomeric sequence is conferred by the structurally uncharacterized domain Pot1pC(Altschuler et al., 2011). Comparison to hOB2 gives little insight into the mechanistic basis for this promiscuity due to a lack of sequence identity and differing biochemical features between these domains. To understand how Pot1pC achieves non-specific binding, we have solved the crystal structures of Pot1pC bound to its minimal cognate and a variety of non-cognate ligands. These structures reveal a binding surface distinct from that of hOB2 and a novel structural mechanism by which Pot1pC is able to accommodate heterogeneous ssDNA ligands. Furthermore, these mechanisms likely explain the anomalous lack of specificity in several other single-stranded nucleic acid binding proteins.

Results

Pot1pC+9mer structural overview

The Pot1pC domain was originally defined as encompassing amino acids 178–389 of the full-length Pot1 protein(Croy et al., 2009). This construct was active, but proved refractory to structural work. For these studies, we used a smaller, proteolytically defined construct spanning residues 198–339 and including a mutation (V199D) (Supplemental Fig. S1A), which was more amenable to structural characterization. This construct has a similar global structure, as determined by NMR (Supplemental Fig. S1B), and comparable ssDNA binding characteristics as 178–389 (Supplemental Fig. S1C). Moreover, 1–339 fully recapitulates binding by the original Pot1-DBD construct, 1–389 construct (data not shown). 198–339 V199D (hereafter referred to as Pot1pC) was copurified with its minimal 9-nt ssDNA ligand d(GGTTACGGT) and the crystal structure of the complex was solved using standard strategies to 1.7 Å resolution (Table 1 and Methods).

Table 1.

Data Collection and refinement statistics

	Pot1pC+9mer (4HIK)
Data collection
Space group		P212121
Cell dimensions
a, b, c (Å)		44.87, 57.39, 66.17
α, β, γ (°)		90, 90, 90
	Peak	Inflection	Remote
Wavelength	0.979862	0.9801	0.9428
Resolution (Å)	50.0-1.71 (1.74-1.71)	50.0-1.70 (1.73-1.70)	50.0-1.64 (1.67-1.64)
Rsym or Rmerge	.058 (.141)	.048 (.125)	.050 (.146)
I / σI	36.07 (11.73)	35.59 (11.56)	33.76 (10.28)
Completeness (%)	99.1 (94.5)	97.7 (67.0)	99.1 (93.8)
Redundancy	8.4 (7.0)	8.4 (6.7)	8.4 (6.9)
Refinement
Resolution (Å)		35.35-1.71
No. reflections		21007
Rwork / Rfree		19.17/21.93
No. atoms		1654
Protein		1172
Ligand/ion		185
Water		297
B-factors
Protein		16.744
Ligand/ion		20.393
Water		28.75
R.m.s deviations
Bond lengths (Å)		0.006
Bond angles (°)		1.06

	G2C (4HID)	T3A (4HIM)	T4A (4HIO)	A5T (4HJ5)
Data collection
Space group	P212121	P212121	P212121	P212121
Cell dimensions
a, b, c (Å)	40.96, 59.12,	45.02, 57.40,	41.34, 59.87,	40.87, 59.24,
	66.09	66.41	66.28	65.76
α, β, γ (°)	90, 90, 90	90, 90, 90	90, 90, 90	90, 90, 90
Resolution (Å)	50.0-1.8 (1.9-1.8)	50.0-1.75 (1.8-1.75)	50.0-1.76 (1.82-1.76)	50.0-2.05 (2.12-2.05)
Rsym or Rmerge	.072 (.321)	.034 (.147)	.057 (.366)	.058 (.384)
I / σI	14.46 (2.65)	26.15 (7.36)	15.59 (2.52)	12.61 (2.28)
Completeness (%)	95 (83.8)	98.8 (98.9)	98.5 (95.7)	93.0 (95.0)
Redundancy	3.3 (2.5)	3.0 (2.9)	3.3 (3.0)	2.4 (2.5)
Refinement
Resolution (Å)	30.1-1.8	31.3-1.75	30.3-1.75	29.95-2.04
No. reflections	14118	17732	16745	9833
Rwork / Rfree	20.5/21.8	22.14/25.16	21.46/23.6	22.60/26.09
No. atoms	1550	1530	1555	1444
Protein	1172	1172	1177	1167
Ligand/ion	182	186	186	184
Water	196	172	192	93
B-factors
Protein	24.273	25.643	19.705	37.004
Ligand/ion	33.477	33.49	27.893	48.707
Water	31.688	29.038	27.042	38.349
R.m.s deviations
Bond lengths (Å)	0.007	0.007	0.009	0.008
Bond angles (°)	0.797	0.813	0.877	1.1

	C6G (4HJ7)	G8C (4HJ8)	+1 5’ (4HJ9)	+2 5’ (4HJA)
Data collection
Space group	P212121	P212121	P212121	P212121
Cell dimensions
a, b, c (Å)	40.76, 58.18,	44.58, 57.55,	44.00, 57.47,	45.85, 56.78,
	65.63	66.58	66.15	66.26
α, β, γ (°)	90, 90, 90	90, 90, 90	90, 90, 90	90, 90, 90
Resolution (Å)	50.0-1.79 (1.85-1.79)	50.0-2.05 (2.12-2.05)	50.0-1.77 (1.83-1.77)	50.0-2.00 (2.07-2.00)
Rsym or Rmerge	.036 (.192)	.073 (.459)	.047 (.512)	.059 (.421)
I / σI	22.62 (5.51)	12.648 (2.33)	17.58 (1.85)	12.37 (2.26)
Completeness (%)	99.5 (99.5)	99.5 (98.9)	99.1 (95.5)	99.3 (93.4)
Redundancy	3.5 (3.4)	3.4 (3.2)	3.2 (2.9)	3.4 (3.1)
Refinement
Resolution (Å)	17.51-1.78	18.78-2.04	31.23-1.85	31.4-2.10
No. reflections	15326	11245	14821	12144
Rwork / Rfree	17.37/20.69	19.23/23.00	20.06/23.96	19.26/24.79
No. atoms	1544	1479	1512	1540
Protein	1172	1172	1185	1177
Ligand/ion	188	182	204	225
Water	184	125	123	138
B-factors
Protein	22.241	28.572	33.955	25.383
Ligand/ion	22.831	37.279	46.113	39.191
Water	30.668	33.278	39.876	29.053
R.m.s deviations
Bond lengths (Å)	0.007	0.007	0.007	0.008
Bond angles (°)	1.14	1.046	1.125	1.23

One crystal was used for each structure. Values in parentheses are for highest-resolution shell.

The unit cell dimensions fall into 2 groups: one group (9mer, T3A, G8C, +1 5’, and +2 5’) is ~45×57×66 A while the other (G2C, T4A, A5T, and C6G) is ~41×59×66 A. This appears to be a result of 2 types of crystal packing near the 5’ end of the ssDNA. Despite these differences, these two groups do not correlate with any of the structurally or biologically relevant features of this protein. Thus, we assume this to be a crystal-packing artifact with no bearing on our conclusions.

Pot1pC is an OB-fold that binds d(GGTTACGGT) (9mer) roughly across the canonical OB-fold binding face (Fig. 1A). This surface includes β-strands β1–β5 and is centered on strands β2 and β3(Theobald et al., 2003; Horvath, 2011). The OB-fold differs from most in that it contains an extended loop between β2 and β3 (L23) and a sixth β-strand that creates an additional loop (L56). Both of these features augment the DNA binding cleft to create a surface of 691 Ų that is buried upon 9mer binding(Voss and Gerstein, 2010). While most OB-folds bind ssDNA in an extended fashion, 9mer is bent ~90° between bases 5 and 6. This bend guides the DNA around L12, but the DNA does not make contact with the loop. This lack of contact creates an unfilled surface in the binding cleft that is atypical of OB-fold/ligand complexes (Fig. 1B).

Figure 1.

Pot1pC+9mer structure overview. (A) Crystal structure of Pot1pC bound to its cognate 9mer substrate d(GGTTACGGT). 9mer is depicted as lavender sticks and its electron density is contoured to 1.5σ. Pot1pC is colored N to C-terminus (blue to red). Pot1pC is an OB-fold, and 9mer lies across the canonical ligand-binding surface augmented by L23 and L56. (B) Surface representation of Pot1pC in which the DNA and surface atoms within 5 A are colored by element (C-green/violet, O-red, N-blue, P-orange, Se-yellow). This depiction highlights the chemical diversity of the binding interface that includes hydrophobic and polar contacts. Additionally, this depiction illustrates the wide binding pocket and exposed surface between L12 and nucleotides 2–4. Figures were created with MacPyMOL(Schrodinger, LLC, 2010). See also Supplemental Figure S1.

A detailed analysis shows that the interface consists largely of stacking interactions and base mediated H-bonds (Fig. 2). G1 is somewhat removed from the primary cleft, and is stacked onto Trp72. This base still participates in extensive interactions with the protein surface, forming five intermolecular H-bonds with both side-chain and main-chain atoms of Pot1pC (Fig. 2A). Bases 2–4 form an off-centered stack that is capped on the bottom by Ile107 and Ile70 and on top by Leu101. These bases form several direct and water-mediated H-bonds with side chains along strands β4 and β5 (Fig. 2B and C). A5 is removed from this stack and inserts into a deep pocket composed of Arg68 and Phe47. The exocyclic amine extends back to form an intramolecular H-bond with T4, and several water-mediated H-bonds are formed between the Watson-Crick face and the protein (Fig. 2C). Following A5, the DNA bends ~90° to form two intramolecular H-bonds involving C6 (Fig. 2D). This bend is likely induced by steric interactions with L23 and favorable intramolecular H-bonding interactions. There is an additional twist that orients the phosphate backbone towards the protein and several intermolecular H-bonds are formed with L23. Additionally, C6 forms a well-aligned stack with G7, Trp27, T9, and Tyr28. As with the other bases, G7 and T9 also form a variety of intermolecular H-bonds (Fig. 2E). G8 is flipped out of this stack, but is located in a pocket formed by G7 and Arg57. G8 forms an array of both intra- and intermolecular H-bonds with the phosphate backbone and L23, respectively (Fig. 2F). In total, there are five intramolecular and 27 intermolecular H-bonds formed by the DNA in addition to stacking and hydrophobic interactions.

Figure 2.

The Pot1pC+9mer interface consists of extensive H-bonding and stacking interactions. (A) The G1 binding pocket involves a stacking interaction with Trp72 and four direct intermolecular H-bonds. (B) Bases 2, 3, and 4 form an off-centered stack. G2 forms three direct H-bonds while T3 forms one direct and one water-mediated H-bond. (C) Leu101 forms the top of the T4 binding pocket while T4 forms two intermolecular and one intramolecular H-bond. A5 stacks between Arg68 and Phe47 and forms three water mediated H-bonds with Pot1pC. (D) An ~90° kink in the DNA orients the backbone towards L23 while bases 6 and 7 point towards the solvent filled portion of the binding pocket. Despite this orientation, C6 still forms two base-specific water-mediated intramolecular H-bonds. (E) Bases G7 and T9 stack between Trp27 and Tyr28. The Watson-Crick face of G7 is solvent exposed, but one direct and one water-mediated H-bond are formed along the Hoogsteen face. T9 forms two direct and one water-mediated H-bond with the protein backbone. (F) G8 is flipped out of the main binding pocket, but is packed between G7 and Arg57. Additionally, G8 forms an extensive array of both inter- and intramolecular H-bonds. See also Supplemental Figure S3.

The ssDNA interface is not conserved between hOB2 and Pot1pC

The closest structural match to Pot1pC is hOB2 of human POT1(Holm and Rosenstrom, 2010). Despite limited sequence conservation(Croy et al., 2006), the protein structures align with an RMSD of 1.9 A for 125 of the 139 α-carbons(Holm and Rosenström, 2010). As might be expected, the β-barrel cores align well, but, unexpectedly, there is a major displacement of L23 between hOB2 and Pot1pC. Part of L23 of Pot1pC is peeled away from the β-barrel core, whereas L23 of hOB2 packs against the β-barrel and fills the binding surface utilized by Pot1pC (Fig. 3A). This difference modulates the surface utilized for ligand binding dramatically such that hOB2 contacts only four nucleotides along a surface that is rotated around the “side” of the β-barrel(Lei et al., 2004). In contrast, Pot1pC utilizes a more extensive surface on the “front” of the β-barrel, which results in a stunning lack of ligand overlap between the two structures. In fact, the binding pocket for only one nucleotide is shared between the two domains (Fig. 3B). Even though the interfaces are quite different, the amino acids that make up this shared pocket are conserved.

Figure 3.

Alignment of hOB2(Lei et al., 2004) (yellow) and Pot1pC (green) highlight the unexpected ssDNA-binding mode. (A) The beta-barrel cores of the proteins align well, but L23 adopts an extended conformation in Pot1pC (purple) while L23 in hOB2 (red) lies in the canonical ssDNA-binding pocket. (B) The positioning of L23 in the human structure obscures the surface along which the majority of the DNA lies in the S. pombe structure. This causes the human DNA to bind along a unique binding surface with the position of only one base conserved between species.

These structural dissimilarities raise the question of whether the 9-nt sequence bound by Pot1pC can be extended in the 5’ direction to access the binding surface of hOB2 and vice versa. To address these questions, we solved the structures of Pot1pC bound to DNA sequences with one and two extra nucleotides on the 5’ end (data not shown). In both structures, the immediate 5’ base simply stacks onto G1 and extends into space, making no substantive contacts with the protein. Addition of a second 5’ nucleotide continues this stacking into space, but is poorly defined in the electron density suggesting significant mobility. In neither case is there evidence of the bases interacting with the surface utilized by hOB2. The amino acids that make up the G2 binding pocket in Pot1pC (Arg97 and Glu105) are conserved in the hOB2 structure (Supplemental Fig. S3), suggesting the possibility that an extended ssDNA substrate could be bound by hPot1. Previous work, however, has shown that human POT1 has a similar affinity for ssDNA ligands extended in the 3’ direction suggesting no additional interactions(Lei et al., 2004). Furthermore, other key DNA contact residues in Pot1pC (e.g., Trp27 and Tyr28) are not conserved in the hOB2 structure (Supplemental Fig. S3). These data, combined with the presence of the loop that occludes the binding surface, suggest that the two domains recognize ssDNA in distinct fashions.

Pot1pC accommodates non-cognate ligands through distinct structural mechanisms

Pot1pC shows little specificity for the sequence of its ssDNA ligand(Croy et al., 2009) (Table 2) and this is even more pronounced in the context of the full DNA-binding domain(Altschuler et al., 2011). This biochemical observation is at odds with the large number of apparently base-specific H-bonds in the Pot1pC+9mer complex. The intermolecular interface contains a total of 22 base-mediated H-bonds, with each base participating in at least two (Fig. 2). To determine how non cognate ligands are accommodated, we solved the structures of Pot1pC bound to six different ligands with individually substituted bases. One of these substitutions, d(GCTTACGGT) (G2C), significantly disrupts affinity (36-fold), but the others have little to no effect (< 3-fold) on affinity (Table 2). The specificity at position 2 could be rationalized by the presence of three direct H-bonds between the base and protein (Fig. 2b), but position 1 has four direct H-bonds (Fig. 2a) that confer no specificity (Croy et al., 2009). Thus, the cognate structure alone cannot be used to predict specificity at a given location. In the non-cognate complexes the overall protein topology is maintained. Surprisingly though, despite the similar binding affinities, there were a number of unanticipated structural changes to the interface in these non-cognate complexes. These changes range from slight local shifts to global rearrangements of the entire interface.

Table 2.

Biochemical and structural effects of complementary base substitutions

DNA ligand	KD (nM)a	Fold changeb	ΔH (kcal mol-1)a	TΔS (kcal mol-1)a	DNA RMSD (Å)c
9mer - GGTTACGGT	24	-	−2.9	−18	-
G2C - GCTTACGGT	855	36	−3.0	−22	3.05
T3A - GGATACGGT	37	1.5	−2.9	−18	0.81
T4A - GGTAACGGT	21	0.88	−2.0	−10	2.70
A5T – GGTTTCGGT	63	2.6	−3.0	−20	0.96
C6G - GGTTAGGGT	6	0.25	−2.6	−14	0.87
G8C - GGTTACGCT	22	0.92	−2.6	−16	1.00

^aApparent K_D, ΔH, and TΔS values are averaged from duplicate ITC experiments. Representative data is shown in Supplemental Figure S1B.

^bFold change is relative to the 9mer affinity

^cRMSD of the DNA is relative to 9mer and is calculated for all non-substituted nucleotides using MacPyMol(Schrodinger, LLC, 2010)

Local reorientation of a base

Substitutions at positions 3, 5, and 6 are all accommodated by slight shifts in the orientation of the base to compensate for lost H-bonds by forming new ones. These shifts do not substantially impact the positioning of the rest of the DNA, and RMSD values for unsubstituted nucleotides are all < 1 Å (Table 2). The protein backbone and sidechains are largely unaffected as well, with only slight changes in an area of L56 distal to the interface.

Of these ligands, T3A undergoes the largest local shift. One direct H-bond with Arg68 and one water-mediated H-bond are broken (Fig. 2B), but one new H-bond is formed with a slightly rotated His100. This shift also creates a better-aligned stack with G2 that may compensate for some of the lost H-bonding energy (Fig. 4B). The phosphate backbone adopts a slightly altered conformation, but this does not affect the positioning of any other intermolecular contacts (Fig. 4A).

Figure 4.

Complementary base substitution can be accommodated by local adjustments that break and reform H-bonds, but maintain the global binding interface and retain high affinity. (A) The phosphate backbone of T3A (yellow) is altered slightly compared to the original 9mer substrate (lavender), but the protein backbone is unaffected. (B) The adenine substitution at position 3 shifts the base such that two H-bonds are broken and one new H-bond is formed. (C) The A5T complex (orange and white) is globally similar to the cognate (green and lavender). (D) The substituted thymine stacks between Arg68 and Phe47, but two H-bonds are lost relative to the cognate binding mode. (E) The C6G complex (yellow and blue) is globally similar to the cognate (green and lavender). (F) The guanine at position 6 is in the same plane, but rotated 90 degrees relative to the cytosine. This orients the Watson-Crick face towards the protein, which forms a new H-bond with Tyr136. Another new intramolecular H-bond is formed with its own phosphate group while the intramolecular H-bond is maintained with T3. (G) The G8C substrate (blue) is accommodated by a local shift in L23 of Pot1pC (pink) relative to the cognate complex (green and lavender). (H) The cytosine at position 8 is rotated 90 degrees relative to the guanine in the original binding mode. This maintains the hydrophobic and stacking interactions, but creates a new network of hydrogen bonds that is unique from the original substrate (lavender). See also Supplemental Figure S2.

Substitution at position 5 causes very little change to the DNA or protein and many of the contacts are still maintained. The substitution breaks two water-mediated H-bonds and one intramolecular H-bond (Fig. 2C). One of the water-mediated H-bonds is recouped by repositioning of the water molecule, but the rest are lost. The smaller thymine base is, however, able to reach far enough into the pocket to maintain favorable stacking between Arg68 and Phe47 (Fig. 4D).

Substitution at position 6 also has little global effect on the interface (Fig. 4E), but the base is rotated dramatically. A 180° rotation around the 1’ glycosidic bond maintains the planarity and stack with G7. The flexibility of the ssDNA backbone allows additional slight rearrangements that effectively swap the Watson-Crick and Hoogsteen faces. This does little to conserve the donor/acceptor pattern, but it maintains one of the two water-mediated H-bonds from the cognate complex while forming one new water-mediated bond and one new direct H-bond to Tyr136 (Fig. 4F). As with bases 3 and 5, the substituted base maintains the same planarity, thereby preserving the stacking interactions present in the cognate complex.

Local reorientation of a base and the protein

Some base substitutions lead to local changes in both the DNA and protein conformation. Substitution from guanine to cytosine at position 8 is accommodated by a 180° rotation of the base similar to what is seen in C6G. In addition, the rotation of G8C is coupled with a rearrangement of L23 (Fig. 4G). In the cognate complex, G8 forms a complex network of hydrogen bonds that involves every potential donor and acceptor atom on the base (Fig. 2F). Substitution to cytosine disrupts all of these interactions, but rotation of the base creates an equally intricate network of H-bonds with almost entirely new inter- and intramolecular partners (Fig. 4H). The H-bond with Lys25 is maintained, but new water-mediated bonds are formed with the side-chain of Glu85 and backbone of Thr26. Interestingly, these residues are unaltered from the cognate complex, as if poised to interact with the ssDNA; it is only the reorientation of the base that is required for the formation of new bonds. Interaction with L23 is maintained, but only by a concerted reorientation of the loop. The new protein conformation allows for a novel H-bond with the side-chain of Ser55, and it repositions the side-chain of Arg57 to maintain the favorable packing interaction. While these adjustments are all proximal to the substituted base, this structure illustrates the role of the conformational plasticity of the protein in addition to the DNA in accommodating base changes.

Global reorientation of the DNA and protein

While some non-cognate ligands can be accommodated by local adjustments to the interface, others lead to global reorganization of the complex. In the structures of the G2C and T4A complexes, we found that Pot1pC is able to utilize a second binding mode. Pot1pC binds G2C and T4A in a new fashion in which the positions of six of the nine bases are altered (ligand RMSDs of 3.05 and 2.70 A respectively) as well as L23 of the protein (Fig. 5A and Supplemental Fig. S2A). The affinity for G2C is decreased 36-fold relative to 9mer, and this is reflected by the poorly defined electron density for bases 2 and 3 (Supplemental Fig. S2D). Unfortunately, this lack of density makes it difficult to determine why this complex adopts an alternate binding mode. T4A, however, is bound with an affinity equal to the cognate sequence, and the high quality structure allows for detailed analysis of this alternate mode.

Figure 5.

Pot1pC binds 9mer T4A in an alternate binding mode. (A) Alignment of Pot1pC+9mer (green and lavender) and pot1pC+9mer T4A (blue and gray) shows that L23 adopts a new conformation in the T4A binding mode, but the majority of the protein remains unchanged. The DNA maintains the same general contact surface, but the positioning of the majority of bases is altered to some extent. (B) The adenine base at position 4 flips down into the previously unfilled portion of the binding pocket. This new conformation stacks with Arg68, but results in steric clashes with bases T3 and C6 (illustrated by spheres representing the Van der Waals radii of relevant atoms). (C) The steric clash with T3 causes the base to shift and disrupt two H-bonds. This is compensated by an improved stacking interaction between G2 and His100. (D) The steric clash with C6 causes a shift that disrupts the network of H-bonds between the phosphate backbone and L23. L23, however, is able to adopt a new conformation that forms an equally extensive network of Hbonds. (E) The shifts at C6 and L23 disrupt the H-bond network involving G8, but the base is able to rotate over 90° to form a new network of H-bonds. This new network involves amino acids Thr26 and Glu85, which were previously not involved in binding. (F) The shift at position 6 propagates down to the stack involving G7, Trp27, T9, and Tyr28. These bases and amino acids, however, are able to rotate slightly to maintain the majority of stacking and H-bonding interactions seen previously.

A step-by-step examination of the T4A interface suggests how the protein is able to access this alternate binding mode. The repositioning of base 4 triggers the global reorganization. The larger adenine in the T4A complex is unable to fit in the binding pocket occupied by thymine in the cognate complex. As a result, the base rotates 90° around the phosphodiester backbone, flipping it out of the original binding pocket and into the largely unoccupied region above L12 (Fig. 5A). This reorientation creates a stacking interaction with Arg68 (Fig. 5B), but the H-bonds and stacking interactions present in the cognate complex are completely disrupted (Fig. 2C). This same reorientation occurs in the G2C complex (Supplemental Fig. S2B and C), but the reason is less clear due to the lower quality structure.

The new orientation of base 4 clashes with T3 and C6 (Fig. 5B), and their subsequent shifts are propagated across the interface. The clash with T3 causes the base to shift in the pocket in a manner similar to the T3A substitution (Fig. 4A and 5C). This breaks several H-bonds, but is accompanied by a rotation of the His100 sidechain to create a new stacking interaction. The end result is a well-aligned stack between G2, T3, and His100 that likely compensates for the energetics lost by T3 and T4. The steric clash with C6 also likely leads to a rearrangement that pushes the base down in the binding pocket and pulls the phosphate backbone away from L23 (Fig. 5D). The backbone is easily accommodated by the flexible L23, which simply adjusts to create a new H-bonding network. These movements, however, disrupt the G8 pocket, which forces the base to rotate 180° to create a new network of H-bonds (Fig. 5E). This orientation is different, even from the G8C complex, but the planarity of the base and its packing interactions with Arg57 are maintained. Finally, the downward shift of C6 in the binding pocket causes a concomitant shift of G7, Trp27, T9, and Tyr28 to maintain the stack seen in the cognate complex (Fig. 5F). The end result is a unique binding mode utilized by Pot1pC to bind a non-cognate ligand with an affinity equal to that of the cognate sequence.

Discussion

The structure of the Pot1pC/ssDNA complex supports a difference in domain organization between the human and S. pombe Pot1 proteins. Previous work illustrated that Pot1pN and Pot1pC could be separated and expressed individually(Baumann and Cech, 2001; Croy et al., 2009) while hOB1 and hOB2 appear to function exclusively as a tightly packed unit(Lei et al., 2004). These observations, in addition to structural predictions(Croy et al., 2006), suggested the presence of an expanded linker between Pot1pN and Pot1pC. The structure of Pot1pC, in combination with that of Pot1pN, shows that, indeed, there are 25 proteolytically labile amino acids between Pot1pN and Pot1pC (Supplemental Fig. S1A) compared to only five spacer residues between hOB1 and hOB2. Furthermore, if Pot1pN and Pot1pC were packed together in the same manner as hOB1 and hOB2, there would be a 22 Å gap between the 3’ end of the DNA bound to Pot1pN and the 5’ end of the DNA bound to Pot1pC. These observations suggest that Pot1pN and Pot1pC are flexibly tethered subdomains in contrast to the DNA-binding domain of hPot1 that functions as a tightly packed unit. These arrangements are likely evolved to accommodate the different telomeres in each species: hPot1 only needs to recognize an invariant repeat while SpPot1 must accommodate degenerate sequences and likely does so in part via domain-domain rearrangement. The degenerate telomeres found in S. pombe may also be accommodated by the alternate binding modes seen in the non-cognate structures. High-affinity binding of the ssDNA overhang is necessary for protection from the DNA damage response machinery(Wu et al., 2006; Denchi and de Lange, 2007) and it is therefore crucial for SpPot1 to bind a variety of sequences.

Previous work also pointed to a ligand length discrepancy between Pot1pC and hOB2. Pot1pC in isolation binds a minimal 9-nt ligand(Croy et al., 2009) and this binding activity is likely recapitulated in the context of the full DNA-binding domain, which binds a 15-nt ligand with high affinity (PotpN+6mer and Pot1pC+9mer)(Croy et al., 2006; Altschuler et al., 2011). While hOB2 has not been studied in isolation, it contacts only four bases when in tight association with hOB1(Lei et al., 2004). This discrepancy is explained here by the discovery of nearly orthogonal binding surfaces. The disparity between the hOB2 and Pot1pC binding surfaces is particularly surprising given the overall structural similarity between the two domains. Only L23 differs significantly, yet the reorientation of these 17 amino acids allows the protein to utilize a completely different binding interface. This observation is particularly relevant in light of the widespread popularity of modeling and prediction of protein structures and interactions. Often the general fold of the domain can be reliably predicted(Wu and Zhang, 2007; Kelley and Sternberg, 2009), but the details pertinent to the function are not necessarily captured in the prediction. In this case, structural predictions of Pot1pC correctly predict a global fold quite similar to the actual structure (best RMSD=1.3 A for 102/139 α-carbons), but they predict L23 to be in a compact conformation similar to that of hOB2. This conformation occludes the majority of the DNA-binding cleft and fails to capture any of the true features of the ssDNA interaction surface. This disparity substantiates the concern that models may not be accurate enough to capture important functional features, even when the majority of the structure is accurately predicted.

We next located the evolutionary point at which the transition was made from the S. pombe to human binding interface in order to determine the functional significance of the switch. Many residues that specifically contact the ssDNA in S. pombe, but not human, are in the highly variable L23 region and difficult to trace through evolution. We were, however, able to use Trp27 and Tyr28 as markers to suggest the presence of a binding surface similar to that of S. pombe. These residues form an aromatic cluster that anchors the 3’ end of the ssDNA, thus their presence should indicate a binding mode similar to that seen in S. pombe. We found that the WY motif is conserved throughout the Schizosaccharomyces genus and a single aromatic residue is conserved through other fungi and up to the sea anemone. The single aromatic, however, is lost in all species from sea urchins to humans (Supplemental Fig. S3). This pattern of volutionary conservation suggests that the extended S. pombe binding surface is likely not related to degenerate telomeres as many of the yeast species with a conserved aromatic residue have uninterrupted GGTTAG repeats. Rather, while the possibility of intermediate binding modes remains, the interface switch appears to be correlated with the evolution of more complex multicellular organisms. While speculative, it could suggest a role the interface in the regulation of telomerase activity.

One of the most surprising features of the Pot1pC/9mer structure is the abundance of apparently sequence-specific H-bonds. The presence of base-mediated H-bonds raises questions regarding our understanding of non-specific protein/nucleic acid interactions. Base-mediated H-bonds are frequently assigned roles in conferring specificity, yet Pot1pC can accommodate complementary base substitutions at most positions with no impact on binding affinity(Croy et al., 2009) (Table 2). Pot1pC takes advantage of several structural elements to accommodate sequence heterogeneity, including ligand flexibility, protein backbone flexibility, an enlarged binding cleft, and sidechain and water-mediated H-bonds. The flexibility of the ssDNA ligand allows for 180° rotation around the 1’ glycosidic bond as seen in the C6G (Fig. 4F) and G8C (Fig. 4H) complexes. Additionally, the flexibility allows bases to undergo more subtle shifts within binding pockets like that seen in the T3A complex (Fig. 4B). Malleability of the protein backbone - specifically L23 - also plays a role in sequence accommodation. This is aptly illustrated in the G8C complex, where L23 reorients to maintain contact with the substituted base (Fig. 4H). Surprisingly, even distal substitutions such as G2C and T4A cause chain reactions that are accommodated by movements in L23 several angstroms away from the site of substitution. The enlarged binding cleft further facilitates accommodation. The solvent accessible region above L12 allows base 4 to flip out of its binding pocket and into the unfilled space in the G2C (Supplemental Fig. S2) and T4A (Fig. 5A) structures. Finally, flexible sidechain and water-mediated H-bonds help accommodate base substitutions. Sidechain and water-mediated H-bonds allow for the reorganization of H-bonding networks to accommodate different ligands as illustrated by the T3A (Fig. 4B) and A5T (Fig. 4D) complexes, respectively. These new H-bonds are able to thermodynamically compensate for those lost upon base-substitution. Pot1pC relies upon assembling this variety of structural elements in varied ways to bind ssDNA promiscuously.

While Pot1pC accommodates a variety of sequences, many OB-fold containing proteins bind with high specificity suggesting these mechanisms of accommodation are not a universal feature of OB-folds(Theobald et al., 2003; Croy, 2006). There have, however, been a limited number of other studies that addressed the lack of specificity conferred by base-mediated H-bonds. Most of these studies illustrated the importance of a flexible ssDNA/RNA substrate. Two studies showed that ssDNA/RNA can be accommodated by nucleotide shuffling: the process by which a base is flipped out of its binding pocket and replaced by an adjacent base(Theobald and Schultz, 2003; Valley et al., 2012). While this process is capable of accommodating nucleotide substitution, we do not see anything comparable in the structures of Pot1pC. Other studies showed that bases can be rearranged in the binding cleft or rotated around the glycosidic bond to accommodate substitutions(Lu and Hall, 2011; Daubner et al., 2012). These rearrangements resemble those seen in the T4A and C6G/G8C complexes, respectively, and may represent a commonly utilized mechanism of accommodation. In addition to adjustments of the ssDNA/RNA, another study illustrated the ability of sidechains and water-mediated H-bonds to rearrange and accommodate nucleotide substitutions(Wang et al., 2009). These rearrangements are similar to those seen in the T3A and A5T complexes and again may represent a widely used mechanism of accommodation. What is unique about Pot1pC, however, is its ability to combine these features to create an unexpectedly different binding mode. In the G2C and T4A structures, the DNA is rearranged within the enlarged binding cleft, L23 is altered, and H-bonding networks are completely broken and reformed (Fig. 5 and Supplemental Fig. S2). Multiple mechanisms of accommodation are combined in thoroughly unexpected ways to create a globally altered binding mode.

The structures of Pot1pC give a more complete picture of how non-specific and quasi-specific proteins, such as RPA, SSB, U2AF and Cdc13, likely recognize a variety of nucleic acid sequences. While structural features such as the rigidity of the protein/nucleic acid, narrow binding clefts, and/or direct H-bonds are crucial for the specific recognition of nucleic acids, the degree of accommodation observed in this family of structures is remarkable. These structures suggests that the repertoire of strategies for achieving non-specific recognition is wider than previously imagined and should be considered when ascribing specificity to interactions observed in high-resolution structures.

Experimental Procedures

Protein Expression and Purification

Pot1pC was expressed and purified using essentially the same method as described for Pot1-DBD(Altschuler et al., 2011). Briefly, native and selenomethionine-substituted protein were expressed as an intein/chitin-binding domain fusion protein in BL21 (DE3) E. coli at 18 °C. Purification was performed using chitin beads (New England Biolabs) and the fusion protein was cleaved by incubation with 100 mM βME for 20 hours at 4 °C. Upon elution, at least 1.5-fold molar excess oligonucleotide (Integrated DNA Technologies) was added and the protein-ssDNA complex was purified using size-exclusion chromatography (GE Healthcare HiLoad Superdex 75 size exclusion column preequilibrated with buffer containing 20 mM Tris, 50 mM NaCl, 3 mM βME, pH 8.0) where all complexes eluted at the time consistent with the monomeric molecular weight. Free ssDNA was efficiently separated and fractions containing a 1:1 protein-ssDNA complex were pooled and concentrated to 10 mg/ml.

Free Pot1pC for ITC studies was expressed and cleaved as described above. Following elution from the chitin beads the protein was concentrated and injected onto the Superdex 75 equilibrated with 100 mM Tris pH 8, 100 mM KCl, 0.1% (w/v) deoxycholate, 3 mM βME, and 5% (v/v) glycerol. This buffer was found to stabilize the free protein more effectively than the buffer used for the complex. Upon elution from the size exclusion column, the protein was concentrated to 9.4 mg/ml (550 µM), snap frozen in liquid nitrogen and stored at −80 °C.

Crystallization

Crystals were grown using the hanging drop vapor diffusion method at 18 °C. Drops contained 1 µl mother liquor and 1 µL 1:1 protein-ssDNA complex (5–10 mg/ml). Crystallization conditions for each complex are listed in Supplemental Table S1. Crystals in conditions containing <30% (w/v) PEG were cryoprotected by sequentially transferring the crystal into reservoir solution supplemented with 5, 10, 15, and 20% (v/v) ethylene glycol. Crystals in conditions with ≥30% (w/v) PEG did not require an additional additive for successful cryoprotection. All crystals were flash frozen in liquid nitrogen.

Data Collection and Refinement

A selenomethionine labeled crystal was used for the cognate complex structure determination. This complex (pC+9mer) diffracted to 1.7 A at the ALS beamline 8.2.1. Multiwavelength Anomalous Dispersion (MAD) datasets were collected at 100 K at the wavelengths indicated in Table 1. Reflections were indexed, integrated, and scaled using HKL-2000(Otwinowski and Minor, 1997). Three out of four possible selenium sites were located and initial experimental phases were calculated using BnP(Weeks et al., 2002). Due to the high quality of the SeMet crystal, no native dataset was required. PHENIX Autobuild(Adams et al., 2010) was used to create an initial model of the protein. This left clear electron density for the ssDNA ligand, which was manually modeled using Coot(Emsley et al., 2010). Refinement was performed using PHENIX Refine and manual adjustments in Coot to arrive at the reported model (Fig. 1 and Table 1). The final model was evaluated using MolProbity(Chen et al., 2010), which showed good stereochemistry, with 97.1% favored and 0% outlier Ramachandran angles respectively. Ramachandran statistics for non-cognate complexes are in Supplemental Table S1.

All datasets for non-cognate complexes were collected in-house using either an Incoatec microfocus generator or a Rigaku RUH2R rotating anode generator interfaced to a Rigaku Raxis IV++ image plate detector (wavelength: 1.54 A temperature: 100 K). Data were indexed and scaled using HKL-2000. Rigid body refinement was performed using the cognate complex without the ssDNA as a starting model in PHENIX. The non-cognate ligand was then manually built using Coot and subsequent refinement was performed in the programs PHENIX and Coot.

Isothermal Titration Calorimetry

Free Pot1pC was thawed and dialyzed overnight at 4 °C in buffer containing 20 mM potassium phosphate pH 8.0, 50 mM NaCl, 3 mM βME. Oligonucleotides were resuspended in the same buffer. All experiments were performed on a MicroCal ITC 2000 (GE Healthcare) at 25 °C. 270 µl of 5 µM Pot1pC was added to the sample cell and 40 µl of 40 µM ssDNA was titrated in. Concentrations of protein and ssDNA were measured by absorbance at 280 and 260 nm, respectively, and calculated using extinction coefficients provided by ExPASy ProtParam and IDT respectively. Heat of dilution experiments showed no detectable heat evolved and thus were not subtracted from binding experiments. Data were integrated and fit by nonlinear least-squares fitting using Origin ITC Software (Microcal Software).

Highlights.

• Pot1pC binds ssDNA distinctly from hOB2 as a result of a dramatically shifted loop

• Despite binding non-specifically, the interface has H-bonds that appear base-specific

• Pot1pC accommodates nucleotide substitutions with alternate binding modes