Interactions of the Cold Shock Protein CspB fromBacillus subtiliswith Single-stranded DNA

The cold shock protein CspB fromBacillus subtilis binds T-based single-stranded DNA (ssDNA) with high affinity (Lopez, M. M., Yutani, K., and Makhatadze, G. I. (1999) J. Biol. Chem. 274, 33601–33608). In this paper we report the results of CspB interactions with non-homogeneous ssDNA templates containing continuous and non-continuous stretches of T bases. The analysis of CspB-ssDNA interactions was performed using fluorescence spectroscopy, analytical centrifugation and isothermal titration calorimetry. We show that (i) there is a strong correlation between the CspB affinity and stoichiometry and the T content in the oligonucleotide that is independent of which other bases are incorporated into the sequence of ssDNA; (ii) the binding properties of CspB to ssDNA templates with continuous or non-continuous stretches of T bases with similar T content is very similar, and (iii) the mechanism of interaction between CspB and the T-based non-homogeneous ssDNA is mainly through the bases (a stretch of three T bases located in the middle of the ssDNA templates makes the binding independent of the ionic strength). The biological relevance of these results to the role of CspB as an RNA chaperone is discussed.

The detailed mechanism by which living cells survive a cold shock remains controversial (1). It has been observed that in many bacteria cold shock is accompanied by an increase in the expression levels of a specific set of the so-called cold shock proteins (CSPs) 1 (2,3). In Bacillus subtilis the major CSP is a small acidic protein, CspB, whose induction increases dramatically when the temperature is down-shifted from 37°C to 15°C (4). The first evidence of any activity of CspB was demonstrated using gel retardation experiments, and it was shown that CspB binds to single-stranded DNA (ssDNA) containing the Y-box motif (ATTGG) (5,6). Later, however, it was shown that CspB binding is not limited only to this sequence (7,8). Several aromatic residues (Phe 15 , Phe 17 , Phe 27 , Phe 30 , Phe 38 , Trp 8 ) and several basic amino acid residues (His29, Lys 7 , Lys 13 , Lys 39 , Arg 56 ) have been proposed to be involved in the protein interaction with ssDNA (9). The three-dimensional structure of the protein shows that these amino acid residues are located in the same side of the protein molecule (5,6).
In our laboratory, using four different techniques (gel shift mobility assays, isothermal titration calorimetry (ITC), analytical ultracentrifugation, and fluorescence spectroscopy), it was shown that CspB binds preferentially to polypyrimidine ssDNA (8). CspB binding to poly(dC) and poly(dT) has different properties showing higher affinity for homogeneous T-based oligonucleotides with a binding site size of 6 -7 T bases (8). The analysis of such binding according to the Epstein model (10) showed that the CspB binding to T-based ssDNA template has high affinity (K a ϳ 3 ϫ 10 6 M Ϫ1 ) and moderate cooperativity ( ϳ 16). To further characterize the CspB binding to T-based ssDNA templates, two major questions must be answered. 1) How is CspB binding to continuous stretches of T bases affected by the presence of other bases in the oligonucleotide? 2) How is such a binding affected when the T bases form non-continuous stretches? To address these questions, we used fluorescence spectroscopy, analytical centrifugation, and isothermal titration calorimetry and measured the CspB binding properties to series of ssDNA (23-mers) in which the T bases were placed in the middle of the oligonucleotide (23CpTi, 23ApTi, and 23GpTi, where i ϭ 3, 5, 7, 11, 15, and 17), forming a continuous stretch but varying its length. The results of these experiments were compared with the effect of non-continuous stretches of T bases, the effect of location of the T bases within short ssDNA template and the effect of the total length of the oligonucleotide on the CspB binding. In addition, the nature of the mechanism responsible for the CspB binding to non-homogeneous ssDNA template was investigated by comparing the binding profiles at high (1 M) and low (0.1 M) NaCl concentrations.
Our results show that ssDNA binding properties of CspB are strongly correlated with the T-base content in the ssDNA templates. Indeed, ssDNA templates with similar T content have similar binding properties, independent of whether the T bases form continuous or non-continuous stretches. We also show that the forces involved in the CspB binding to most of the non-homogeneous T-based ssDNA template are not via electrostatic interactions of protein groups with the phosphate backbone because no difference in the binding profiles is observed under experimental conditions with different ionic strength. This implies that the interactions of CspB with non-homogeneous T-based ssDNA templates occur mainly through the bases. These observations combined with the sequence analysis of 5Ј-UTR of cold shock-inducible proteins suggests possible function of CspB as RNA chaperone.

MATERIALS AND METHODS
CspB and ssDNA Purification-CspB was overexpressed in Escherichia coli and purified as described previously (8,11). The protein concentration was measured spectrophotometrically using the extinction coefficient ⑀ 280 ϭ 5,690 M Ϫ1 cm Ϫ1 (8). The ssDNA oligonucleotides were obtained from Biosynthesis Inc. and purified as described previously (8). The ssDNA concentration was calculated considering the following extinction coefficients at 260 nm: 8400 M Ϫ1 cm Ϫ1 (for T), 12010 M Ϫ1 cm Ϫ1 (for G), 7050 M Ϫ1 cm Ϫ1 (for C), and 15,200 M Ϫ1 cm Ϫ1 (for A) (12).
Analytical Centrifugation Experiments-The analytical centrifugation experiments were performed in a Beckman XLA ultracentrifuge. The runs were performed at 4°C under two different speeds, 20,000 and 25,000 rpm, until equilibrium was achieved (usually Ͼ10 h). Absorbencies at 260 and 280 nm were recorded simultaneously for three cells. Each cell contained only ssDNA, CspB, or CspB-ssDNA complex. The ssDNA concentrations were 2 M for 23ApT3 and 23ApT11 and 2.5 M for 23CpT3, 23CpT7, and 23CpT11. The protein concentration was 20-fold higher than the ssDNA concentration. The CspB-ssDNA complex concentration was the same as in the other cells. To allow for the complex formation, the protein was incubated with the ssDNA at room temperature for 25 min prior to each run. The partial specific volume for the protein, ssDNA, and complex were 0.74, 0.55, and 0.65 cm 3 /g, respectively and calculated as described previously (8).
ITC-ITC experiments were performed using a VP-ITC (Microcal Inc. Northampton, MA) instrument. The protein and ssDNA were dialyzed simultaneously in 50 mM Tris, pH 7.5, 100 mM NaCl buffer. The protein concentration was usually between 10 and 15 M. Typical concentrations of the ssDNA were between 85 and 135 M. Series of several injections of 5 or 10 l each of solution containing ssDNA template into ITC cells containing CspB were made until no further heat effect was observed. The heat of dilution was measured by injecting identical amounts of ssDNA into the ITC cell containing only buffer. The heat of the reaction was obtained by integration of the peak area after each injection of ssDNA solution, using ORIGIN software provided by the manufacturer. ⌬H cal was calculated by summing the individual heats, corrected for the heats of dilution, and dividing by the total number of moles of CspB present in the ITC cell during the experiment. Detailed description of the ITC experiment has been reported previously (8).
Fluorescence Measurements-The CspB fluorescence intensity was measured using a FluoroMax spectrofluorometer with DM3000F software and thermostated cell holder connected to a circulating water bath. The equilibrium titrations were performed at low (0.3 M) protein concentration in 50 mM Tris-HCl, pH 7.5, with 100 mM NaCl (low ionic strength) or 1 M NaCl (high ionic strength). Tryptophan fluorescence was excited at 287 nm, and the emission was recorded at 349 nm. Stoichiometric titrations were performed at high protein concentration (14 M) in low ionic strength buffer. The excitation and emission wavelengths were 300 and 349 nm, respectively. All experiments were performed at 25°C, and the initial volume in the sealed quartz cell was 1.1 ml. Small aliquots of concentrated solutions of ssDNA were added into the cell containing solution of CspB. The solution in the cell was gently stirred during the titration and the intensity values were corrected by dilution; inner filter effect and blanks were subtracted.
The equilibrium titration profiles, under conditions when only one molecule of CspB was bound to the ssDNA template, were fitted to Equation 1 (8).
Q represents the quenching effect on the CspB Trp fluorescence after each addition of ssDNA, Q max is the maximal quenching obtained upon complete saturation of the protein with ssDNA, [CspB] tot and [ssDNA] tot are the total concentration of protein and ssDNA in solution, respectively, and K a is the equilibrium association constant for the CspB-ssDNA interaction.

RESULTS
Previously, using a combination of four different techniques, we showed that CspB seems to bind T-containing homogeneous ssDNA template with much higher affinity (K d ϭ 42 nM at 25°C) than C-, G-, or A-containing 23-mer homogeneous ssDNA templates (13). In this study we used hetero-ssDNA templates to investigate how the change in length of a T stretch flanked by other nucleotides (A, C, and G), the relative position of the T stretches within the template and continuous versus non-continuous stretches of T bases affect the thermodynamics of the CspB-ssDNA interactions. With few exceptions the ssDNA templates were 23 nucleotides long (Table I). CspB-ssDNA interactions were monitored using three different techniques: fluorescence spectroscopy, isothermal titration calorimetry, and analytical centrifugation.
Effect of 23-mer Non-homogeneous ssDNA Templates with Different Number of T Bases on the CspB Tryptophan Fluorescence-The changes in the protein intrinsic tryptophan fluorescence is a widely used to monitor protein-DNA interactions (e.g. Refs. 14 and 15). The cold shock protein CspB has only one Trp residue that is located in the putative nucleic acid binding face of the protein (9). The fluorescence intensity of this Trp is dramatically quenched upon interaction with T-based ssDNA templates (8). This quenching of CspB Trp fluorescence correlates with the degree of binding upon the CspB-ssDNA interactions (8). Fig. 1 (A-C) shows the quenching effects caused by nonhomogeneous ssDNA templates 23CpTi, 23GpTi, and 23ApTi (i ϭ 3, 5, 7, 11, 15, and 17; see Table I for sequences) on the CspB intrinsic Trp fluorescence. In all cases there is a shift in the titration profiles toward lower ssDNA concentrations as the number of T bases in ssDNA template increases. The difference in the binding competences, defined as [ssDNA] 0.5 , between XT3 and XT17, where X represents C, G, or A bases, is about 1 order of magnitude (from ϳ4 ϫ 10 Ϫ7 M to ϳ6 ϫ 10 Ϫ8 M, respectively). Comparing the three series of ssDNA templates (23CpTi, 23GpTi, and 23ApTi), it is clear that ssDNA templates with the same T content behave in very similar fashion in the 23CpTi and 23GpTi series (Fig. 1, A and B). In fact, there is a good correspondence between the binding competences for the ssDNA templates with the same number of T bases in both series (Fig. 2). The correlation coefficient is 0.98, indicating that the interaction is independent of the presence of C or G bases surrounding the stretches of T. This result strongly indicates that the CspB interactions with these ssDNA templates are probably independent of three-dimensional structure of ssDNA in agreement with previous results (13).
The titration profiles of CspB with the 23ApTi template series are different than the ones obtained with the 23CpTi or 23GpTi series (Fig. 1, compare C with A and B). The titration profiles for the ssDNA templates 23ApT3, 23ApT5, 23ApT7, and 23ApT11 overlap. This result can be explained considering that, in the 23ApTi series, some of the ssDNA templates have the ability to form double-stranded DNA due to the classical Watson-Crick base pairing, as will be further supported by the analytical centrifugation data (see below). However, ssDNA templates with larger number of T bases (23ApT15 and 23ApT17) do show an enhancement in binding, as seen by the decrease in the [ssDNA] 0.5 as the stretch of T bases becomes longer (Fig. 1C).
Taken together, these results indicate that there is a strong correlation between the [ssDNA] 0.5 and the T content in the template when the T bases are located in the middle of the ssDNA template and form a continuous stretch. Observed differences in [ssDNA] 0.5 for different templates can arise from the changes in the affinity of binding and/or in the stoichiometry of the interaction.
It must be noted that the ssDNA templates used in this study are not homogeneous and relatively short (most of them are 23-mer). Unfortunately, no rigorous mathematical model has been developed for the analysis of the titration profiles in which more than one molecule of protein binds to the short and heterogeneous matrix (14,16). To the best of our knowledge, the mathematical models available to date have been developed either for homogeneous and infinitely long linear templates (17), or for homogeneous and short lattices (10) or even homogeneous templates for which linearity between the signal change and the fractional saturation does not necessarily hold during the binding reaction, but not for short non-homogeneous templates (for reviews, see Refs. 14 and 16). We thus are not able to estimate the CspB/ssDNA binding affinities. We can, however, experimentally estimate the stoichiometry of the CspB binding to non-homogeneous ssDNA templates using analytical centrifugation and isothermal titration calorimetry.
Stoichiometry of Non-homogeneous ssDNA-CspB Complexes-For homogeneous ssDNA templates, it was shown that three CspB molecules bind to 23pT and only one CspB molecule binds to 23pC (8). However, the templates used in present study are not homogeneous, and thus different stoichiometry among templates with different T-content is expected.

5Ј-TTTTTTTTTTTTTTCCCCCCCTTTTTTTTTTTTTT
5Ј-TTTGGGGGGGGGGGGGGGGGTTT 23 6 ϩϩ 23pT8G 5Ј-TTTTGGGGGGGGGGGGGGGTTTT 23 8 ϩ 23pT12G 5Ј-TTTTTTGGGGGGGGGGGTTTTTT 23 12 Ϫ 23pT16G 5Ј-TTTTTTTTGGGGGGGTTTTTTTT 23 16 Ϫ 23pT18G 5Ј-TTTTTTTTTGGGGGTTTTTTTTT 23 18 Ϫ 23pT20G 5Ј-TTTTTTTTTTGGGTTTTTTTTTT 23 20 II. Under our experimental conditions, the average molecular masses for the protein and the 23CpTi ssDNA (8.0 Ϯ 0.4 and 7.1 Ϯ 0.2 kDa, respectively) are in good correspondence with the expected values for monomeric species (7.5 and 6.9 kDa, respectively) (Fig. 3A). Considering the apparent molecular masses of the CspB:23CpTi complexes and individual species (Table II), we find that two molecules of CspB bind to 23CpT3, 23CpT7, and 23CpT11. Although this result is clear for the 23CpT7 and 23CpT11 templates, a lower value was obtained for 23CpT3 (the ratio CspB:23CpT3 is 1.5). Nevertheless, isothermal titration calorimetry experiments (see below) support our conclusion that more than one molecule of CspB binds to 23CpT3. We can therefore conclude that the difference in binding competences observed in the equilibrium titration profiles from 23CpT3 to 23CpT11 (Fig. 1A) is due to an increase in the protein effective binding affinity and not due to a change in the stoichiometry of the interaction. It is notable that the increase in effective binding affinity directly correlates with the T-content in the ssDNA.
The calculated stoichiometries for 23ApT3 and 23ApT11 show that these oligonucleotides bind only one CspB molecule (Table II). This is not surprising for 23ApT3 because CspB does not bind to homogeneous A-based templates (8). So, if one molecule of CspB binds to the three T bases in 23ApT3, this will represent an "imperfect" binding because we have shown previously using 23pT template that the CspB binding site size is 6 -7 bases long (8). An imperfect binding should translate into a lower affinity of the protein for this ssDNA, as is observed experimentally (Fig. 1C). When only one molecule of protein binds to the ssDNA template, the interaction is accurately described by the classical binding equation (see Equation 1, under "Materials and Methods"), and it has been used to describe the thermodynamics of binding in different protein-DNA systems (8,18). From the fit of the titration profile of CspB with ssDNA templates to Equation 1, the calculated equilibrium association constant (K a ) for CspB interaction with 23ApT3 is (4.3 Ϯ 0.1) ϫ 10 6 M Ϫ1 , which is indeed lower than the effective affinity for 23pT ϳ5 ϫ 10 7 M Ϫ1 (8). In contrast, 23ApT11 has enough T bases to bind one molecule of CspB with the same affinity as 23pT. Experimentally, however, we observe that the normalized titration curve for 23ApT11 overlaps with that for 23ApT3 (Fig. 1C), indicating that 23ApT3 and 23AT11 have comparable affinities. The fit of the CspB:23ApT11 titration curve to Equation 1 gives an equilibrium constant K a of (4.4 Ϯ 0.1) ϫ 10 6 M Ϫ1 , identical to that of 23ApT3 but an order of magnitude lower than the effective affinity for 23pT. The reason why the CspB molecule bound to 23ApT11 presents lower affinity than expected seems to be related to the fact that this non-homogeneous template has the ability to form doublestranded DNA. Indeed, the measured molecular mass for the 23ApT11 template is higher than that predicted for the ssDNA (10.1 Ϯ 0.1 kDa versus 6.8 kDa, respectively, Table II), indicating probably the presence of a significant population of doublestranded DNA under our experimental conditions. CspB does not have detectable affinity to double-stranded DNA (5, 6). Thus, if at least part of the 23ApT11 forms double-stranded DNA via classical Watson-Crick A-T base pairing, the effective concentration of ssDNA will be overestimated and that would result in a lower apparent affinity, as is observed experimentally (Fig. 1C).
The stoichiometry of the CspB interaction with some of the oligonucleotides within the 23CpTi series was also studied using ITC. The protein solution (in the calorimetric cell) was titrated with the ssDNA (in the syringe) until no heat effects were observed. Under experimental conditions with high protein concentration (11-15 M CspB in solution), we observe that the titration profiles are very steep and rapidly level off (Fig. 3B), allowing determination of the stoichiometry of the interaction from the breaking point in the titration profile. The breaking point for the 23CpT15 template is ϳ 0.3, indicating that 3 molecules of CspB bind to 23CpT15. The titrations with 23CpT11 and 23CpT5 are shifted toward lower stoichiometry. The breaking point for the titrations of these two templates is at a ratio ssDNA/CspB ϳ 0.5, indicating that 2 molecules of CspB bind to 23CpT11 and 23CpT5 templates. Finally, the titration with 23CpT3 is not as sharp as the other ones; how-    (Fig. 3B). This result suggests that more than one molecule of protein binds to the 23CpT3 template and supports our conclusion based on the analytical centrifugation experiments that probably two CspB molecules bind to 23CpT3 template. The enthalpies of the interaction per binding site for these oligonucleotides were Ϫ105 Ϯ 9 kJ/mol, comparable to the values obtained for 23pT, Ϫ119 Ϯ 6 kJ/mol (8). Similarity of the enthalpies of interactions with different T-based ssDNA templates suggests similar structural features for the CspB-ssDNA complexes. In summary, the stoichiometries for the CspB interaction with the 23CpTi oligonucleotides determined by analytical centrifugation and ITC are in good agreement. The results show that 23CpT3, 23CpT5, 23CpT7, and 23CpT11 bind two CspB molecules. Knowing that the CspB binding site size is 6 -7 T bases (8), our results suggest that not only the T bases but also the C bases are somehow involved in the CspB binding to those ssDNA templates. In addition, the ITC experiments show that there are changes in the total number of protein molecules bound to 23-mer ssDNA templates (two or three CspB molecules bind to 23-mer ssDNA templates depending on the number of T bases in the oligonucleotide). (Table II, Figs. 1 (A-C) and 3B). To evaluate how the different content of T bases forming non-continuous stretches of T might affect the CspB-ssDNA interaction, the ATTGG sequence was incorporated in four 23-mer oligonucleotides: 23CT 2 , 23CT 4 , 23CT 8 , and 23CT 12 (see Table I for sequence). The reason we chose this sequence to be incorporated into the ssDNA was that initially it was considered that CspB binds specifically to ATTGG sequence (19). Later, however, it has been shown that the CspB binding to ssDNA is not limited to this sequence (7,8). Nevertheless, it constitutes a good model to study the CspB interaction with 23-mer ssDNA with noncontinuous stretches of T bases. It is important to note that the T bases in these non-continuous stretches are only one or two bases apart. The normalized equilibrium titrations with these oligonucleotides are shown in Fig. 4. The binding curves for the CspB titration with 23CT 12 (with a total of 12 T) and 23CpT11 (with a total 11 T) overlap. The same is observed for the CspB titrations with the other pairs of templates: 23CT 8 (8 T bases) and 23CpT7 (7 T bases), 23CT 4 (4 T bases), and 23CpT5 (5 T bases) and 23CT 2 (2 T bases) and 23CpT3 (3 T bases). These results suggest that the [ssDNA] 0.5 for these oligonucleotides containing non-continuous stretches of T bases separated by one or two bases is very comparable to the values obtained for ssDNA with similar T content forming continuous stretches of T. The importance of the potential implications of these findings is obvious. First, it is not the local structure of the ssDNA that is important for CspB binding. Second, not all bases in the CspB binding site are equally important for the interaction. The latter is particularly significant because non-continuous stretches of T bases are more often found in DNA, as compared with long and continuous stretches of T bases.

Properties of the CspB Binding to 23-mer ssDNA with Noncontinuous Stretch of T Bases-The experiments described in the previous sections have shown that CspB interactions with ssDNA templates vary with the number of T bases forming a continuous stretch
CspB Binding to ssDNA Templates Containing Two Contin- uous Stretches of T Bases-The next step was to study the CspB binding to non-homogeneous oligonucleotides of different length with stretches of T in which the T bases were more than two bases apart and not in the middle of the template but situated at the ends. A constant stretch of seven C bases was always in the middle of these templates, which are flanked by different number of T bases ( Table I). The reason C bases, as opposed to A or G, were selected to be in the middle of the oligonucleotides was to avoid possible double-stranded DNA formation (in the presence of A bases), and we chose seven because it appears that the size of the CspB binding site on T-based ssDNA template is 6 -7 nucleotides long (8). Fig. 5A shows the titration profiles of CspB with T18C7T18, T14C7T14, and T9C7T9 under stoichiometric conditions. Since titrations were performed under stoichiometric conditions, we observe, from the breaking point in the titrations, that about 6 nucleotides are covered by a molecule of CspB in agreement with our previous observations (8). These results indicate that, when the T content in the ssDNA template is high (72% or more) and the continuous stretches of T bases are on both ends of the ssDNA, these templates behaved as if they were homogeneous. Fig. 5B shows the normalized equilibrium titrations of CspB with T18C7T18, T14C7T14, T9C7T9, T7C7T7, and T5C7T5. It is clear that the [ssDNA] 0.5 for these ssDNA templates increases as the length of the ssDNA decreases, being more dramatic for T5C7T5. A similar effect was observed by Kowalczykowski et al. (20). They studied the binding of gene 32 protein to poly(rA) as a function of the oligonucleotide length (Ͼ400, 120, and 56 nucleotides). A rigorous binding analysis of the titration profiles showed that the affinity and cooperativity was the same for all three oligonucleotides, although more protein was needed to saturate the 56-mer poly(rA) than to saturate the 120-mer or Ͼ400 mer. This effect has been also observed for other proteins interacting with shorter oligonucleotides, i.e. the mitochondrial Y-box protein RBP16 interacting with guide RNAs. RBP16 has a cold shock domain and interacts with guide RNAs through an oligo(U) tail (21). This interaction required higher [RNA] 0.5 (defined as the molar excess required to achieve 50% inhibition of the RBP16 binding to 32 P-labeled guide RNA gA6) as the length of the oligo(U) decreased from 40 to 4 nucleotides (22).
Importance of the Relative Position of the T Bases in 23-mer ssDNA Templates-The effect of relative position of the T bases within ssDNA template was investigated by comparing the CspB binding to the 23GpTi series (the T bases are in the middle) and the 23pTiG series (in which the T bases are on both ends; see Table I for sequence) keeping constant the total length of the ssDNA template at 23 bases. Fig. 6 shows the normalized profiles for the titrations of CspB with the 23pTiG oligonucleotides. The binding curves for the 23pT20G, 23pT18G, and 23pT16G overlap. The CspB binding competence for these ssDNA templates is about 5.8 ϫ 10 Ϫ8 M, which is very close to the CspB binding competence for 23pT, 6.1 ϫ 10 Ϫ8 M (8). This result again indicates that when the T content in non-homogeneous ssDNA templates is high (70% in 23pT16G or more) and the T bases are on both ends of the ssDNA templates they behave as if they were homogeneous. We can also conclude, from this and previous results with the TiC7Ti series, that as long as the stretch in the middle is short enough (30% or less of the total length of the template) the oligonucleotide will behave as homogeneous 23pT independent of which bases are in the middle, C or G (Figs. 5B and 6A, respectively). As the T content in the ssDNA templates decreases (e.g. 23pT12G, 23pT8G, and 23pT6G series), there is a corresponding increase in the [ssDNA] 0.5 (Fig. 6A). To understand whether this effect is due to the difference in the number of T bases or due to the different position of these bases within the oligonucleotides, we plotted the [ssDNA] 0.5 for the 23pTiG and 23GpTi series as a function of the T content (percentage of T bases) in the ssDNA (Fig. 6B). The dependence of the CspB binding competences on the T content in the 23pTiG and 23GpTi templates is very comparable for both series of ssDNA, except for the 23pT6G template in which "end effects" or possible aggregation of the G bases may play a role. This similar behavior suggests that it is the T content in the ssDNA templates, and not its relative position, which defines the properties of CspB binding to T-based ssDNA, although some "end effects" cannot be completely ruled out. The importance of the "end effects" was clearly demonstrated for the interaction of the Y-box protein RBP16 from Trypanosoma brucei, which contains a cold shock domain motif, with 34-mer oligonucleotides (22). The 34-mer oligonucleotides had a patch of four adjacent uridylates, forming the binding site, and the rest of the oligonucleotide was oligo(dC). The tetraU patch was moved from 5Ј to 3Ј at certain intervals and the binding of RPB16 to these different templates was compared. It was shown that binding is weaker when tetraU patch was located close to the 5Ј or 3Ј ends of the template. Furthermore, Pelletier et al. (22) showed that the affinity also depends on whether tetraU patch is located at 5Ј or 3Ј end.
Effect of High Ionic Strength on the CspB Binding to Nonhomogeneous ssDNA-It has been shown that CspB binding to 23pC is strongly dependent on the salt concentration and could be practically abolished in 1 M NaCl, whereas the CspB binding to 23pT is independent of salt concentration (8). These results were interpreted to mean that there are two different modes of CspB interactions with polypyrimidine ssDNA templates. In mode I, the interactions of CspB with poly(dC) occurs predom- inantly through the phosphate backbone and thus is largely electrostatic and strongly dependent on the ionic strength. In mode II, the interactions of CspB with poly(dT) occurs predominantly through van der Waals and stacking interactions with the bases and are thus independent of the ionic strength. We thus investigated which of these two modes of interactions was involved in the CspB binding to non-homogeneous ssDNA templates. This was done by comparing the CspB binding to different oligonucleotides (with continuous and non-continuous stretches of T bases) measured in low (0.1 M) and high (1 M) NaCl concentrations. It was anticipated that if the CspB binding to C bases is independent of the presence of T bases, then under 1 M NaCl concentration the CspB binding to the C bases in the ssDNA would be dramatically impaired, which would translate in a lower quenching effect on the CspB Trp fluorescence upon binding. Qualitative results of the analyses presented in Table I show that the presence of 1 M NaCl did not affect the titration profiles of CspB with 23CpTi (i ϭ 3-23 forming a continuous stretch of T bases). These results suggest that CspB interacts directly with the T bases, mainly through hydrogen bonding, van der Waals interactions, and hydrophobic effects, but not through electrostatic interactions of protein groups with the phosphate backbone. How many T bases are required in order to have a salt-independent binding? To answer this question, the interactions of CspB with two other ssDNA, 23CpT1 and 23CpT2, with just one or two T bases (Table I) at low and high salt concentrations were studied. It was found that for these two oligonucleotides there is an effect of 1 M NaCl on the titration profiles. The decrease, however, is not as pronounced as in the case of 23pC, where 1 M NaCl completely abolishes CspB binding (8). We thus can conclude that the switch from mode I to mode II of CspB-ssDNA interactions requires presence of at least 3 T bases in the template.
The only other templates, which also exhibit the dependence of the binding on the ionic strength, are 23CT 2 and 23pT6G (Table I). In the case of 23CT 2 , such a behavior is understandable because this template contains only two T bases. The 23pT6G template contains six T bases arranged into two triplets. According to the observations made on the other ssDNA templates, the presence of three T bases should render saltindependent quenching effect on CspB-ssDNA interactions. However, these T bases are located at the 5Ј and 3Ј ends of the 23pT6G template, and the end effects might be important. Indeed, the effect on 1 M NaCl on CspB titration is significantly reduced in 23pT8G and completely abolished in 23pT12G (Table I). We thus propose that the switch from mode I to mode II binding for CspB ssDNA requires at least 3 T bases located within the ssDNA template or 4 T bases if located at the 5Ј or 3Ј ends.
These results suggest that the mechanism of the CspB interaction with most of the studied non-homogeneous T-based ssDNA templates is through the bases and does not involve electrostatic interactions of protein groups with the phosphate backbone. The large enthalpies of binding to the non-homogeneous T-based ssDNA templates (Ϫ105 Ϯ 9 kJ/mol) also support this conclusion because binding to the 23pC (which is strongly salt-dependent) is accompanied by 4 times lower enthalpy, Ϫ25 Ϯ 2 kJ/mol (8). Moreover, in terms of the binding enthalpy, the non-homogeneous T-based oligonucleotides behaved as the homogeneous 23pT, indicating that hydrogen bonding, van der Waals interactions, and hydrophobic effects are the major forces responsible for CspB binding to ssDNA templates (8). DISCUSSION In this study it has been shown that the ability of the cold shock protein CspB from B. subtilis to bind ssDNA templates correlates with the content of T bases present in the template. Moreover, the binding competences of CspB for continuous stretches of T are independent of the surrounding bases (Fig.  2), suggesting that the local structure of the ssDNA does not play significant role in the interactions. Comparable binding properties are observed when CspB binds continuous or noncontinuous stretches enriched in T bases (Figs. 4, 5, and Fig.  6B). This result suggests that the presence of T is not equally important at all position of the binding site consisting of 6 -7 bases. Our findings also strongly suggest that, in the CspB interaction with non-homogeneous ssDNA templates, it is the protein binding to the T bases which drives the interaction. Just three T bases are enough to switch the mechanism of interaction from CspB interacting with ssDNA phosphate backbone (like in 23pC) to CspB interacting directly with the bases (like in 23pT and most of the other studied ssDNA series) as the CspB binding becomes independent of the ionic strength (Table I). Our recent studies have shown that binding of CspB to ssDNA-template based on another pyrimidine nucleotide U is much more similar to the binding characteristics of 23pT FIG. 7. Possible secondary structure formation in the 5-UTR of cspB. The secondary structure was generated using RNAFold software based on the algorithm described by Zuker et al. (34). Cold shock boxes identified previously by Graumann et al. (7) are underlined.
If the biological function of CspB is indeed related to its high affinity for stretches of ssDNA enriched in T bases, there are two T-rich regions identified in DNA. One occurs at factorindependent transcription termination signals (see Refs. 23-27 for reviews). The other is frequently located downstream from the promoter sequences, as part of the sequences contained within 5Ј-UTR of cold shock proteins (9). Recently, we have shown that the CspB binding parameters for 23pU are quite comparable to the ones for 23pT (13) (although the former has somewhat lower affinity, it might be related to the fact that we used d(pU) instead of r(pU)). There are two cold shock boxes enriched in U bases within the cspB 5Ј-UTR (Fig. 7), which are highly conserved in another member of the cold shock family of proteins from B. subtilis, the cspC 5Ј-UTR (7). We have performed CspB binding studies with each of these motifs using ssDNA templates with U bases substituted by T and found that CspB binds with high affinity, as it is expected from their high T content. 2 Thus, our results support the idea that indeed these motifs could have a regulatory role in vivo as suggested by Graumann et al. (7). Since the CspB binding affinity increases as the temperature decreases (8), the protein binding to the cold shock Box1 and Box2 is more favored at low temperatures. Protein binding to such regions would prevent formation of secondary structure of the 5Ј-UTR (Fig. 7); hence, CspB would facilitate translation at low temperatures. This 5Ј-UTR region has been shown to be important in stabilizing the mRNA for the major cold shock protein CspA from E. coli (28). Comparison of the 5Ј-UTRs among the cold shock proteins in E. coli clearly shows that the proteins that are cold-inducible (CspA, CspB, CspG, and CspI) have very long 5Ј-UTR (Ͼ100 bases) and have tendency to form similar secondary structures at low temperatures (Fig. 7). The same appears to be true for 5Ј-UTR of CspB and CspC from B. subtilis. The average Gibbs energy of secondary structure formation for the cold shock-inducible 5Ј-UTR calculated according to SantaLucia (29) is Ϫ(260 Ϯ 50) kJ/mol. However, those proteins that are not cold-inducible (CspC, CspD, CspE) have short 5Ј-UTRs (40 -90 bases), have relatively low secondary structure stability, Ϫ(120 Ϯ 50) kJ/ mol, and thus low tendency to fold at low temperatures. This correlation suggests that, upon cold shock, the increase in CspB induction might be essential for keeping its 5Ј-UTR region unstructured. Comparison of the 5Ј-UTR regions among cold shock proteins in E. coli has showed that there is a well conserved 11-base sequence (cold shock box) (28). However, the sequence of the cold shock box in the 5Ј-UTR of cold-inducible proteins in E. coli (UGACGUACAGA) is very different from the cold shock boxes found in the same region in B. subtilis (AUUAUUUUUGUUC) that is very rich in U bases (Fig. 7). It is conceivable that cold shock proteins have diverse functions. The ability to facilitate transcription antitermination has been suggested for CspA from E. coli (30), whereas B. subtilis CspB might be preventing mRNA folding and thus facilitating translation at low temperatures. This possible functional difference is supported by the observation that cold-inducible proteins from different bacteria may recognize different sequences, as we have already demonstrated for CspB from B. subtilis and CspA from E. coli (8,13).
Our results suggest a model for the role of the cold shock protein CspB from B. subtilis. At low temperatures and low concentration of CspB, the nascent mRNA will fold, thus interfering with translation. It is important to note that cold shock boxes are located within two putative stem structures of the 5Ј-UTR (Fig. 7). In the presence of CspB, protein binding to the cold shock boxes 1 and 2 can be the driving force to prevent the mRNA to adopt stem structure. Thus, CspB binding will effectively prevent secondary structure formation and thus will facilitate translation. Probably, two CspB molecules will bind with high affinity per each of two cold shock boxes, as these boxes are about 14 nucleotides long and, although not completely homogeneous, are highly enriched in U bases (Fig. 7). This would suggest that cold shock proteins possess RNAchaperone activity (31), a possible function of cold shock proteins that is in accord with the earlier observations. It has been shown that both CspA (32,33) and CspB (7) bind RNA and increase its susceptibility to the ribonuclease digestion, thus indicating that CspA/CspB binding prevents the formation of secondary structure of RNA, i.e. acting as an RNA chaperone.