The structure of a highly-conserved picocyanobacterial protein reveals a Tudor domain with an RNA-binding function

Abstract

Cyanobacteria of the Prochlorococcus and marine Synechococcus genera are the most abundant photosynthetic microbes in the ocean. Intriguingly, the genomes of these bacteria are strongly divergent even within each genus, both in gene content and at the amino acid level of the encoded proteins. One striking exception to this is a 62-amino-acid protein, termed Prochlorococcus/Synechococcus hyper-conserved protein (PSHCP). PSHCP is not only found in all sequenced Prochlorococcus and marine Synechococcus genomes, but it is also nearly 100% identical in its amino acid sequence across all sampled genomes. Such universal distribution and sequence conservation suggest an essential cellular role of PSHCP in these bacteria. However, its function is unknown. Here, we used NMR spectroscopy to determine its structure, finding that 53 of the 62 amino acids in PSHCP form a Tudor domain, whereas the remainder of the protein is disordered. NMR titration experiments revealed that PSHCP has only a weak affinity for DNA, but an 18.5-fold higher affinity for tRNA, hinting at an involvement of PSHCP in translation. Isothermal titration calorimetry experiments further revealed that PSHCP also binds single-stranded, double-stranded, and hairpin RNAs. These results provide the first insight into the structure and function of PSHCP, suggesting that PSHCP appears to be an RNA-binding protein that can recognize a broad array of RNA molecules.

Introduction

With the mean annual average of 3.6 × 10²⁷ cells, cyanobacteria from the genus Prochlorococcus (1) and Synechococcus (2) numerically dominate microbial communities of the global ocean (3). Because of such abundance, these photosynthesizing bacteria are extremely important players in the global carbon cycle (3). The small genomes and limited gene content of individual Prochlorococcus and Synechococcus cells (1.6–2.9 Mb, encoding ∼1700–3100 genes (4)) are counterbalanced by a diverse collective gene pool, which in the case of Prochlorococcus is estimated to contain 80,000 genes (5, 6). The genomes of Prochlorococcus and Synechococcus are also surprisingly divergent: in pairwise comparisons, the genome-wide average amino acid identity (7) of the encoded proteins is often below 60% even within each genus (8). Maintenance of such remarkable sequence and gene content divergence remains poorly understood (6). However, one protein-coding gene is a curious exception, with its product showing almost 100% amino acid sequence conservation across all currently available Prochlorococcus and marine Synechococcus genomes (supporting Datasets S1 and S2) (9, 10). This gene encodes a 62-amino-acid protein of unknown function, dubbed PSHCP² for Prochlorococcus/Synechococcus Hyper-Conserved Protein (9, 10). Although at the time of its discovery the gene encoding PSHCP was limited to the cyanobacterial clade consisting of Prochlorococcus and marine Synechococcus, sequence similarity searches now identify the PSHCP gene in genomes of freshwater Synechococcus (11); several Cyanobium spp. from both freshwater and marine environments; in cyanobacterial sponge symbiont “Candidatus Synechococcus spongarium” (12); in three species of Paulinella, which is a photosynthetic protist with a chromatophore thought to be derived from marine Synechococcus spp. (13); and in metagenomically-assembled cyanobacteria from globally distributed marine brackish and freshwater environments. Within cyanobacteria, all characterized PSHCP-containing organisms form a clade (14) within the order Synechococcales (15). Hence, the PSHCP protein remains extremely conserved in a specific subgroup of cyanobacteria (supporting Datasets S1 and S2) but is undetectable outside of this subgroup.

The remarkable conservation of the PSHCP protein and retention of this gene even within extremely reduced chromatophore genomes (13) suggest that this protein may encode an important housekeeping function. In three examined strains of Prochlorococcus and marine Synechococcus, the gene is expressed, and its protein product is abundant (10). A large proportion of positively-charged amino acids (16%) (9), a predicted pI of 11.3 (9), and the protein's association with the ribosomal protein L2 in pulldown assays (10) prompted the hypotheses of PSHCP involvement in binding of either DNA or RNA and its possible association with a ribosome. Yet, the amino acid sequence of PSHCP shows no significant sequence similarity to any protein domain with a known function.

To gain more insight into the possible function of PSHCP, we determined the structure of the PSHCP protein using nuclear magnetic resonance (NMR) spectroscopy, and we investigated the protein's interactions with nucleic acids in vitro.

Results

Structure of PSHCP

We began by screening different constructs of PSHCP for their behavior in structure determination. An initial round of construct screening revealed that residues 57–62 could be removed without altering the overall stability of the protein. This is in good agreement with disorder prediction using the IUPred server (16), which suggests that residues 57–62 are disordered. PSHCP_1–56 could be easily concentrated above 1 mm making it ideal for structure determination. Secondary structure probabilities as calculated using TALOS-N (17) from ¹H,¹³C,¹⁵N chemical shifts show that the first three residues of PSHCP_1–56 are disordered and that the remainder of the protein contains five β-strands (Fig. S1A). To further investigate the dynamics of PSHCP_1–56, we performed a {¹H}-¹⁵N heteronuclear nuclear Overhauser effect (NOE) experiment. In the {¹H}-¹⁵N heteronuclear NOE experiment, a peak intensity ratio closer to one indicates little motion of the N–H bond on the picosecond to nanosecond time scale, whereas a smaller peak intensity ratio corresponds to more motion on this time scale and thus the residues that are disordered. The {¹H}-¹⁵N heteronuclear NOE data recorded on PSHCP_1–56 is in good agreement with the secondary structure probabilities confirming that the structured core of the protein corresponds to residues 4–56 (Fig. S1B). For the structure calculation of PSHCP_1–56, 1195 NOE-derived distance constraints and 44 dihedral angle constraints were used (Table 1). Excluding the first three amino acids, this resulted in 22.1 distance constraints per amino acid. The 20 lowest energy structures align with an average pairwise backbone RMSD of 0.72 ± 0.15 Å for residues 4–29 and 35–56 (Fig. 1A). Residues 30–34 form a flexible loop between β-strands two and three, and thus they were excluded from the RMSD calculation. The overall fold of PSHCP contains five β-strands, which form a single antiparallel β-sheet with an overall barrel-like architecture (Fig. 1, B and C).

Table 1.

Structural statistics of PSHCP_1–56

	PSHCP1–56
NMR distance and dihedral constraints
Distance constraints
Total NOE	1195
Intraresidue	282
Inter-residue	913
Sequential (|i − j| = 1)	311
Medium range (1〈|i − j|〉 = 4)	148
Long range (|i − j| = 5)	454
Dihedral angle constraints
ϕ	22
ψ	22
Structure statistics
Violations (mean ± S.D.)
Distance constraints (Å)	0.036 ± 0.004
Dihedral angle constraints (°)	0.028 ± 0.069
Deviations from idealized geometry (mean ± S.D.)
Bond lengths (Å)	0.015 ± 0.001
Bond angles (°)	1.724 ± 0.054
Impropers (°)	1.731 ± 0.158
Ramachandran statistics
Most favored	82.2%
Additionally allowed	16.3%
Generously allowed	1.1%
Disallowed	0.4%
Average pairwise root mean square deviation (Å)
Heavy (4–29, 35–56)	1.32 ± 0.190
Backbone (4–29, 35–56)	0.72 ± 0.150

Figure 1.

Structure of PSHCP. A, ribbon diagram of PSHCP_1–56 showing the overlay of the 20 lowest energy structures. B, topology diagram of PSHCP generated using the pro-origami webserver (48). C, cartoon representation of PSHCP shown in two different orientations. Each β-strand in the structure is labeled and numbered. The N and C termini are also labeled.

To determine whether PSHCP is part of a conserved domain family, we used the Dali server (18) to search the Protein Data Bank (PDB) for proteins with structural similarity to PSHCP. PSHCP aligns well to a number of Tudor domains, including the Tudor domain from SMN (Fig. S2) (19). The highest structural similarity is to the Tudor domain from the human TDRD3 protein: the 53 amino acids of PSHCP (residues 4–56) align to it with a backbone RMSD of 1.5 Å (Fig. 2, A–C). Tudor domains are small domains, ∼60 amino acids long, containing four or five β-strands in a barrel-like fold. Most known Tudor domains are found in eukaryotes (98.97% of the InterPro records for Tudor domain), where they are often present in 2–3 adjacent copies within large multidomain proteins (20). In contrast, full-length PSHCP consists entirely of a single Tudor domain (residues 4–56).

Figure 2.

PSHCP contains a Tudor domain. A, cartoon representation of PSHCP. B, cartoon representation of the Tudor domain from human Tudor domain-containing protein 3 (PDB code 3PMT). C, overlay of the structures from A and B. D, close-up view of the aromatic cage from the Tudor domain of PHF1 (PDB code 4HCZ) bound to trimethylated lysine. E, close-up view showing the residues in PSHCP that are located in the position of the aromatic cage residues in typical Tudor domains. D and E, all residues that make up the aromatic cage are shown as stick representations and labeled.

PSHCP is unlikely to be involved in recognition of methylated proteins

The best-characterized functions of Tudor domains are to bind methylated lysine and arginine residues (20, 21). This frequently results in the targeting of Tudor domain–containing proteins to histones to regulate DNA damage responses and chromatin remodeling. The interaction of Tudor domains with methylated residues occurs through a conserved aromatic cage (Fig. 2D) (20, 22). This aromatic cage is composed of four aromatic residues, which are individually located in β-strands one through four of the Tudor domain. However, PSHCP contains only one aromatic residue in any of the positions corresponding to the aromatic cage residues and is therefore unlikely to bind methylated amino acids (Fig. 2E).

PSHCP has a weak affinity for DNA

Tudor domains have also been shown to regulate DNA function by directly binding DNA (23, 24). In PSHCP, 10 of 62 amino acids are positively charged (Fig. 3A). The majority of these residues cluster in one of two locations on the surface of PSHCP, generating two distinct positively-charged patches that may support DNA binding (Fig. 3B). To determine whether PSHCP is able to bind to DNA, we incubated ¹⁵N-labeled PSHCP_1–56 with different concentrations of a double-stranded GC-rich DNA sequence (dsGCDNA) that has been previously shown to bind the double-interdigitated Tudor domain of retinoblastoma-binding protein 1 (RBBP1) (23). When titrating dsGCDNA into PSHCP, seven peaks shifted in the PSHCP_1–56 ¹H-¹⁵N HSQC spectrum, demonstrating that PSHCP_1–56 can bind to DNA (Fig. 4A). As only seven peaks shifted upon DNA binding, this suggests that PSHCP_1–56 does not undergo large structural changes upon DNA binding. The peaks for Leu-14, Glu-15, Ser-16, Gly-20, and Val-22 all shifted upon DNA binding. These five residues all sit in the loop between β-strands 1 and 2 suggesting that this region is important for DNA binding.

Figure 3.

Electrostatics of PSHCP. A, amino acid sequence of the PSHCP from P. marinus strain CCMP1375. All positively charged residues are highlighted in yellow. B, electrostatic surface representation of PSHCP shown in two orientations. The electrostatic surface was generated using APBS (42), with blue and red representing positively-charged and negatively-charged surfaces, respectively.

Figure 4.

PSHCP weakly binds to dsGCDNA. A, ¹H-¹⁵N HSQC spectra of 50 μm PSHCP_1–56 in the presence of increasing concentrations of dsGCDNA. The ¹H-¹⁵N HSQC spectrum of PSHCP in the absence of dsGCDNA is shown in blue. The inset shows the chemical shift perturbation for Glu-15 with the arrow pointing in the direction of the shift with increasing DNA concentrations. B, CSP data for Glu-15 (cyan), Gly-20 (orange), Val-22 (black), and Thr-46 (green) as a function of DNA concentration. All data points are shown as circles. The fits to the data are shown as lines of the same color as the data points.

To determine the dissociation constant (K_d) for PSHCP_1–56–DNA binding, we monitored chemical shift perturbations (CSPs) at the four residues that undergo the largest CSPs in PSHCP (Glu-15, Gly-20, Val-22, and Thr-46) over a concentration range of 50 to 500 μm dsGCDNA (Fig. 4B and Table 2). The calculated K_d values ranged from 121.3 to 173.3 μm with an average K_d of 140.6 μm. These data demonstrate that PSHCP has a weak affinity for dsDNA, which is similar to the previously measured affinities of Tudor domains for DNA. In comparison, the double-interdigitated Tudor domain of the histone-binding protein RBBP1 has a K_d of 84 μm for the same DNA sequence (23).

Table 2.

DNA-binding data

Residue	Calculated Kd (μm)	R2
Glu-15	121.3	0.99837
Gly-20	138.4	0.99680
Val-22	173.3	0.97596
Thr-46	129.2	0.99165

PSHCP binds tRNA with a low micromolar affinity

Although most Tudor domains that have been studied are involved in the regulation of DNA, some Tudor domains are located in RNA-binding proteins and are involved in the regulation of RNA metabolism (21, 25, 26). Specifically, the knotted Tudor domain from Saccharomyces cerevisiae Esa1 has been shown to bind oligo-RNAs (27), and more recently, the Tudor domain of Escherichia coli ProQ was shown to interact with the small noncoding RNA SraB (28). All of this supports a broader role for Tudor domains in the regulation of nucleic acids, including through direct binding of RNA.

Intriguingly, the PSHCP gene in completely sequenced genomes of marine Synechococcus and Prochlorococcus marinus spp. is flanked on the 5′ end by a gene encoding a tryptophan tRNA and on the 3′ end by genes encoding an aspartic acid tRNA and the glutamyl-tRNA synthetase (Fig. S3) (9, 10). Additionally, in P. marinus strain MED4 and MIT9313 and in Synechococcus sp. WH8102, the PSHCP gene is sometimes co-transcribed with the Trp–tRNA gene (10). Because genes in bacteria are often clustered and co-transcribed based on their joint functionality in a process or a pathway, we hypothesized that PSHCP may interact with tRNA. To test this, we titrated a mixture of tRNA isolated from E. coli into ¹⁵N-labeled PSHCP_1–56 and monitored CSPs in a series of ¹H-¹⁵N HSQC spectra (Fig. 5A). At a 1.6:1 mixture of tRNA and PSHCP_1–56 almost every peak in the ¹H-¹⁵N HSQC had broadened to the point where it was no longer visible (Fig. S4). Only five peaks did not completely broaden during the tRNA titration, and these peaks corresponded to residues at the termini of PSHCP (residues 1–4 and 56). Because of the substantial broadening of the peaks in the ¹H-¹⁵N HSQC, we were only able to monitor CSPs for residues 2–4, which still had visible peaks after completion of the titration. Calculated K_d values for PSHCP_1–56–tRNA binding ranged from 2.9 to 10.4 μm with an average K_d of 7.6 μm (Fig. 5B and Table 3). As such, PSHCP displays an 18.5-fold higher affinity for tRNA than dsDNA, suggesting a potential role for PSHCP in protein translation.

Figure 5.

PSHCP binding to E. coli tRNA. A, ¹H-¹⁵N HSQC spectra of 50 μm PSHCP_1–56 with increasing concentrations of E. coli tRNA. The inset shows the chemical shift perturbation for Leu-3 with the arrow showing the direction of the shift with increasing tRNA concentrations. B, CSP data for Glu-2 (cyan), Leu-3 (orange), and Asp-4 (gray) as a function of tRNA concentration. C, ¹H-¹⁵N HSQC spectra of 50 μm PSHCP_1–62 with increasing concentrations of E. coli tRNA.

Table 3.

tRNA-binding data

Residue	Calculated Kd (μm)	R2
Glu-2	10.4	0.99510
Leu-3	2.9	0.97705
Asp-4	9.4	0.98226

To determine whether the C-terminal tail of PSHCP (residues 57–62) plays a role in tRNA binding, we repeated the HSQC titration experiment with ¹⁵N-labeled PSHCP_1–62. The addition of tRNA also led to significant line broadening of the PSHCP_1–62 ¹H-¹⁵N HSQC spectra (Fig. 5C). Although peak broadening was similar between PSHCP_1–56 and PSHCP_1–62, clear peak shifts at the N terminus were no longer observed. As such, we were unable to determine a K_d value from these spectra.

Multiple PSHCP molecules bind to a single tRNA

To determine a K_d value for both PSHCP_1–56 and PSHCP_1–62, we utilized isothermal titration calorimetry (ITC) (Fig. 6). K_d values for PSHCP_1–56 and PSHCP_1–62 binding to tRNA were 1.32 and 1.39 μm, respectively, suggesting that residues 57–62 do not play an important role in tRNA binding (Table 4). In addition, the stoichiometry (N) measured for these binding experiments was 0.22 and 0.17 for 1–56 and 1–62, respectively. This suggests that multiple PSHCP molecules are able to bind to a single tRNA. An N around 0.2 suggests that five PSHCP molecules may bind to a single tRNA molecule. This would result in a complex of ∼61 kDa for PSHCP_1–62 binding to a single tRNA. This is consistent with the line broadening that was observed in the ¹H-¹⁵N HSQC spectra during the titration with tRNA as line broadening can occur as a result of increasing molecular weight.

Figure 6.

Residues 57–62 do not alter tRNA binding. A, ITC thermogram (top) and binding isotherm (bottom) for tRNA titrated into PSHCP_1–56 in buffer containing 50 mm NaCl. B, ITC thermogram (top) and binding isotherm (bottom) for tRNA titrated into PSHCP_1–62 in buffer containing 50 mm NaCl.

Table 4.

ITC data

Sample	Kd (μm)	N	ΔH (×104 cal/mol)	ΔS (cal/mol/K)
50 mm NaCl
PSHCP1–56 + tRNA	1.32 ± 0.012	0.219 ± 0.142	−5.11 ± 1.13	−145 ± 38.2
PSHCP1–62 + tRNA	1.39 ± 1.08	0.172 ± 0.020	−3.60 ± 2.21	−94 ± 72.9
25 mm NaCl 75 mm KCl
PSHCP1–56 + tRNA	3.93 ± 1.21	0.212 ± 0.077	−2.61 ± 0.699	−62.7 ± 24.5
PSHCP1–62 + tRNA	3.80 ± 0.673	0.164 ± 0.023	−3.64 ± 0.544	−97.8 ± 18.7
PSHCP1–62 + tRNA competition	3.20 ± 0.072	0.162 ± 0.050	−2.87 ± 0.305	−71.25 ± 10.2
PSHCP1–62 + ssRNA	6.74 ± 0.289	0.634 ± 0.182	−2.01 ± 0.322	−43.8 ± 10.7
PSHCP1–62 + dsRNA	4.21 ± 0.656	0.331 ± 0.141	−5.32 ± 3.45	−154 ± 116
PSHCP1–62 + hpRNA	5.93 ± 0.397	0.192 ± 0.046	−4.74 ± 0.479	−135 ± 16.3

To determine whether the line broadening that we observed in the ¹H-¹⁵N HSQC spectra was due to the increased molecular weight from the formation of the PSHCP–tRNA complex, we purified ²H, ¹⁵N PSHCP_1–62. We titrated tRNA into the ²H ¹⁵N PSHCP_1–62 and recorded a series of ¹H-¹⁵N TROSY spectra. If the line broadening in Fig. 5 was due to the increased molecular weight from the formation of the PSHCP–tRNA complex, then the ¹H-¹⁵N TROSY spectra should have less line broadening and allow for visualization of the peaks upon complex formation. Upon titrating tRNA into ²H, ¹⁵N PSHCP_1–62, we observed less line broadening in the ¹H-¹⁵N TROSY spectra, and clear peak shifts are now visible (Fig. 7A). This result is therefore in good agreement with a model where multiple PSHCP molecules bind to a single tRNA.

Figure 7.

Mapping the tRNA-binding interface on PSHCP. A, ¹H-¹⁵N TROSY spectra of 30 μm ²H-¹⁵N PSHCP_1–62 with increasing concentrations (0.5–100 μm) of E. coli tRNA. The inset shows the chemical shift perturbation for Glu-15 with the arrow pointing in the direction of the shift with increasing tRNA concentrations. B, CSP data for PSHCP_1–62 in the presence of 100 μm tRNA. Peak assignments that could not be transferred from the PSHCP_1–56 assignment were given no CSP value. C, cartoon representation of PSHCP showing the five residues with the highest CSPs as stick representations. Each of the five amino acids is labeled along with the N and C termini.

Residues involved in tRNA binding

To determine which regions of PSHCP bind to tRNA, CSPs were monitored in the ¹H-¹⁵N TROSY spectra for PSHCP_1–62 in the presence of 100 μm tRNA (Fig. 7B). The five largest CSPs observed were for residues Glu-15, Ser-16, Gly-20, Arg-23, and Asn-50 (Fig. 7C). Residues Glu-15, Ser-16, and Gly-20, all of which sit in the loop between β-strands 1 and 2, also shifted upon DNA binding suggesting that PSHCP utilizes similar interaction surfaces for both DNA and tRNA binding.

Because our NMR experiments were performed at 50 mm NaCl, we wanted to determine whether PSHCP binds tRNA at higher salt concentrations that may more closely mimic physiological salt concentrations. Intracellular NaCl concentrations have not been measured for marine Prochlorococcus and Synechococcus, which live in environments with high NaCl concentrations (450–700 mm NaCl). However, measurements on a variety of other cyanobacteria species revealed that these species have intracellular Na⁺ concentrations ranging from 10 to 80 mm when extracellular Na⁺ concentrations are 300–500 mm (29). In addition, higher K⁺ concentrations or other compatible solutes like sucrose are frequently utilized to prevent hypertonic stress (30). As such, we monitored tRNA binding to PSHCP in a buffer containing a mixture of 25 mm NaCl and 75 mm KCl using ITC (Fig. 8, A and B). The K_d value of PSHCP for tRNA was slightly reduced in this buffer to 3.93 and 3.80 μm for PSHCP_1–56 and PSHCP_1–62, respectively (Table 4).

Figure 8.

Binding site mutations on PSHCP. ITC was performed titrating tRNA into different PSHCP samples in buffer containing 25 mm NaCl and 75 mm KCl. ITC thermogram (top) and binding isotherm (bottom) for tRNA titrated into PSHCP_1–56 (A) and tRNA titrated into PSHCP_1–62 (B). C, tRNA titrated into PSHCP_1–62 E15R23AA. D, tRNA titrated into PSHCP_1–62 E15K30AA. E, tRNA titrated into PSHCP_1–62 S16K30AA. F, tRNA titrated into buffer alone.

PSHCP contains two positively-charged patches on its surface (Fig. 3). Our CSP data suggest that the surface surrounding Arg-23, including β-strand 2 and the loop in between β-strands 1 and 2, is the main interaction site for tRNA. To further test this, we chose to mutate Glu-15, Ser-16, and Arg-23 to alanine as these residues had the largest CSPs during the tRNA titration (Fig. 7). We also chose to mutate Lys-30 to alanine as this residue is located in the other positively-charged patch located on the opposite side of PSHCP from Arg-23. We generated three double mutants in PSHCP_1–62, E15R23AA, E15K30AA, and S16K30AA, and performed ITC experiments in buffer containing 25 mm NaCl and 75 mm KCl (Fig. 8, C–E). The exothermic peaks that were previously observed for PSHCP binding to tRNA were not visible in the titrations with any of the mutant proteins suggesting that none of these proteins bind tRNA. Small endothermic peaks are visible, but these peaks are also observed when tRNA is titrated into buffer alone demonstrating that these peaks are the result of tRNA dilution and not a binding interaction (Fig. 8F). Taken together, our CSP and mutagenesis data show that Glu-15, Ser-16, and Arg-23 are all important for tRNA binding.

PSHCP has a broad specificity for RNA

To determine whether PSHCP binding is specific for tRNA, we initially performed competition binding experiments using ITC. In these experiments, PSHCP_1–62 was incubated with an equimolar concentration of dsGCDNA in buffer containing 25 mm NaCl and 75 mm KCl for 30 min. ITC was then performed with tRNA being titrated into the mixture of PSHCP_1–62 and dsGCDNA. In these experiments a K_d of 3.20 μm and an N of 0.162 was observed, which is consistent with the interaction of PSHCP_1–62 for tRNA (Fig. 9A). This demonstrates that dsGCDNA does not compete with tRNA for PSHCP binding.

Figure 9.

RNA-binding preferences of PSHCP. ITC experiments with different RNAs being titrated into PSHCP_1–62 in buffer containing 25 mm NaCl and 75 mm KCl. A, ITC thermogram (top) and binding isotherm (bottom) for tRNA titrated into a 1:1 mixture of PSHCP_1–62 and dsGCDNA. B, ssRNA titrated into PSHCP_1–62. C, dsRNA titrated into different PSHCP_1–62. D, hpRNA titrated into PSHCP_1–62. E, mixture of ATP, GTP, UTP, and CTP titrated into PSHCP_1–62.

To determine which structural features of the tRNA are recognized by PSHCP, we generated three different RNA fragments. These fragments were single-stranded RNA (ssRNA), double-stranded RNA (dsRNA), and an RNA hairpin (hpRNA). These sequences were derived from different regions of the aspartic acid tRNA sequence that is adjacent to the PSHCP gene. PSHCP was able to bind to all three of these RNAs with K_d values ranging from 4.21 to 6.74 μm suggesting that PSHCP has a broad specificity for RNA (Fig. 9, B–D, and Table 4). In addition, the N for dsRNA and hpRNA was similar to the N for tRNA, suggesting that multiple PSHCP molecules can bind to a single dsRNA or hpRNA. In contrast, the N for PSHCP_1–62 binding to ssRNA was 0.634 suggesting that less PSHCP molecules bind to a single ssRNA. Because PSHCP binds to dsRNA, hpRNA, and ssRNA with similar affinities, this suggests that PSHCP is likely recognizing the phosphodiester backbone of RNA and not any specific sequence or structure. To further test this, we performed ITC with a mixture of ATP, GTP, UTP, and CTP (Fig. 9E). We found that PSHCP was unable to bind to these individual nucleotides. Taken together, our data demonstrate that PSHCP has a broad specificity for RNA.

Discussion

We used NMR spectroscopy to determine the structure of the highly conserved 62-amino acid cyanobacterial protein PSHCP. PSHCP residues 4–56 contain a Tudor domain, with five β-strands arranged in a barrel-like architecture. The remainder of the protein is disordered. Two positively-charged patches on the surface of PSHCP hinted that it may bind directly to nucleic acids. We found that PSHCP binds to a variety of RNAs with low micromolar affinities. This suggests that PSHCP may likely be involved in RNA regulation. We also found that residues 57–62 did not enhance the binding affinity of PSHCP for tRNA, and it remains to be seen what role this disordered tail may have in PSHCP function.

The majority of the Tudor domains that have been studied are eukaryotic Tudor domains that regulate histone function by interacting with methylated lysine residues (20). However, there are several examples of Tudor domains that bind directly to nucleic acids instead of methylated proteins. These nucleic acid–binding Tudor domains are broadly distributed from bacteria to eukaryotes, suggesting that the ancestral function of Tudor domains may be nucleic acid binding rather than protein binding. Structures of such nucleic acid–binding Tudor domains have been determined for the DNA damage-response protein 53BP1 from Homo sapiens (31), the histone-binding protein RBBP1 from H. sapiens (23), the methyltransferase Esa1 from S. cerevisiae (27), and the RNA-binding protein ProQ from E. coli (28). The Tudor domain of PSHCP shares structural similarity with the core region of these four nucleic acid–binding Tudor domains (Fig. S5), but there is a surprising divergence in the mechanism of nucleic acid binding between these different Tudor domains, as detailed below.

Both 53BP1 and RBBP1 contain two adjacent Tudor domains (23, 31). 53BP1 contains two sequential Tudor domains that are separated in primary sequence and do not share any secondary structure (Fig. S5A). In contrast, RBBP1 forms an interdigitated Tudor domain where some of the secondary structures of the two Tudor domains alternate in the primary sequence (Fig. S5B). In addition, the two Tudor domains of RBBP1 share two β-strands generating a larger rigid scaffold for DNA binding. DNA binding to either 53BP1 or RBBP1 relies on residues in both Tudor domains (31). RBBP1 contains a positively-charged pocket in between the two Tudor domains that allows for DNA binding (23). This is a distinct mode of nucleic acid binding formed by the overall interdigitated structure of the two Tudor domains. PSHCP aligns well to the individual Tudor domains from these structures but does not contain an extended surface like that of 53BP1 or RBBP1 for nucleic acid binding.

Esa1 contains a knotted Tudor domain due to the presence of two additional β-strands (β0 and β6) that lie just before and just after the classical Tudor domain (Fig. S5C) (27). PSHCP aligns well to the short version of the Esa1 Tudor domain, which lacks these additional β-strands. The short version of Esa1 is incapable of binding to RNA, whereas the knotted Tudor domains bind to poly(U) RNA with a 21.6 μm K_d value in buffer containing 10 mm NaCl (27). This binding interaction has the highest affinity of any of the interactions tested. It is surprising then that PSHCP is able to bind RNA with a low micromolar affinity even though it is most similar in structure to the short version of Esa1, which is incapable of binding RNA.

The E. coli ProQ protein contains two domains, an N-terminal FinO-like domain and a C-terminal Tudor domain (28). Both of the domains interact with small RNAs to facilitate binding with low nanomolar affinities (32). The primary RNA interaction region on ProQ appears to be through the FinO domain (28). However, two regions on the ProQ Tudor domain were also shown to interact with RNA. One of these regions spans β-strands 2 and 3. This region has some overlap with the primary RNA-binding surface on PSHCP, which also includes β-strand 2. Although these Tudor domains have some common RNA-binding surfaces, the locations of the positively-charged amino acids in these regions are in different β-strands. In PSHCP, Arg-23 is located on β-strand 2, and in ProQ Arg-214 is located on β-strand 3. The binding affinity of the ProQ Tudor domain for RNA has not been experimentally determined, and therefore it is unclear whether the ProQ Tudor domain has a similar affinity for RNA as PSHCP.

All of the discussed “nucleic acid–binding” Tudor domains exist as part of multidomain proteins that range in size from 232 to 1257 amino acids. The interaction of these proteins with nucleic acids often requires more than one domain for efficient binding. Because PSHCP is only 62 amino acids, it is rather remarkable that it has such a strong binding affinity for a variety of RNA molecules in such a small number of amino acids. It will be interesting to determine exactly what role the PSHCP–RNA interaction plays in a subgroup of Synechococcales and why it contains such a broad specificity for RNA. Does this broad specificity allow PSHCP to bind to a large number of different RNA targets in cells or is there a specific target that has not been identified? In pulldown assays, PSHCP was observed to be associated with the ribosomal protein L2 (10). Taken together with its RNA-binding ability reported here, it is tempting to speculate that PSHCP is part of the ribosome complex or perhaps is involved in protein translation. Such functions could explain the extraordinary amino acid conservation of the protein, as ribosomal and translational machinery proteins are among the most conserved. Further studies are needed to evaluate these hypotheses.

Experimental procedures

Protein expression and purification

PSHCP_1–56 and PSHCP_1–62 were cloned into the ligation-independent cloning vector (1B), which was a gift from Scott Gradia (Addgene, 29653). Plasmids containing PSHCP were transformed into BL21 (DE3) star cells (Invitrogen, C601003). PSHCP mutants were generated using Q5 site-directed mutagenesis kit (New England Biolabs, E0554S). Cells were grown in Terrific Broth (Fisher, BP9728-2) to an optical density at 600 nm of ∼3.0. For unlabeled protein the expression was induced with 1 mm isopropyl thio-β-d-galactoside (IPTG; Amresco, 97061), and cultures were grown for 18 h at 18 °C. For uniformly-labeled ¹⁵N or ¹⁵N and ¹³C, protein cells were pelleted (3000 × g, 20 min) and resuspended in M9 minimal media containing 3 g/liter [¹⁵N]ammonium chloride (Cambridge Isotope Laboratories, NLM-467-25) with 10 g/liter glucose or 10 g/liter [¹³C]glucose (Cambridge Isotope Laboratories, CLM-1396-25). Expression was induced using 1 mm IPTG (Amresco, 97061), and cultures were grown for 18 h at 18 °C. Cells were harvested and stored at −80 °C. Cell pellets were thawed and resuspended in 50 mm Tris, pH 8.0, 500 mm NaCl, 5 mm MgCl₂, 0.1% (v/v) Triton X-100 buffer containing protease inhibitors (Roche Applied Science, 11836170001). Cells were lysed using a French press (Thermo Electron, FA-032), and lysates were cleared by centrifugation for 50 min at 40,000 × g at 4 °C. Cleared lysates were incubated with TALON resin (Clontech, 635504), which was pre-equilibrated with 50 mm Tris, pH 8.0, 500 mm NaCl. Protein was eluted in buffer containing 200 mm imidazole. TEV protease was used to cleave off the hexahistidine tag. Cleaved protein was flowed over a desalting column to remove imidazole, and a second purification using TALON resin was used to remove the TEV protease and any uncleaved protein. A final purification step was carried out using a HiLoad Superdex 75 PG column (GE Healthcare, 28-9893-33) equilibrated in 20 mm sodium phosphate, pH 6.5, and 50 mm NaCl or 25 mm NaCl, 75 mm KCl for nucleotide-binding experiments or 20 mm sodium phosphate, pH 6.5, 200 mm NaCl, and 0.2 mm tris(2-carboxyethyl)phosphine (Amresco, K831–26) for structural studies.

Deuterated protein expression and purification

PSHCP_1–62 (1B) was transformed into BL21 RILP cells (Agilent, 230280). For the production of deuterated, uniformly ¹⁵N-labeled protein, we followed the protocol outlined in Tugarinov et al. (33). Briefly, cells were grown in the presence of [¹⁵N]ammonium chloride (Cambridge Isotope Laboratories, NLM-467-25), deuterated glucose (Cambridge Isotope Laboratories, DLM-2062-10), and deuterium oxide (Cambridge Isotope Laboratories, DLM-4-1000). Protein expression was induced with 100 μm IPTG, and cultures were grown for 18 h at 18 °C. Protein was purified as under “Protein expression and purification” in 20 mm sodium phosphate, pH 6.5, and 50 mm NaCl.

RNA synthesis

RNA was produced by in vitro transcription (IVT) using the HiScribe T7 high-yield RNA synthesis kit (New England BioLabs, E2040S). DNA templates for the IVT reaction were produced by annealing complementary DNA oligos (IDT) for the T7 polymerase promoter sequence (TAATACGACTCACTATAGGG) and the appropriate RNA template. Oligos were resuspended in Annealing Buffer (10 mm Tris, pH 8.0, 50 mm NaCl) and mixed in equimolar amounts, heated at 94 °C for 2 min, then cooled to room temperature to achieve annealing of the two oligos. IVT reaction products were purified using the Monarch RNA Cleanup Kit (New England Biolabs, T2050S). RNA products were analyzed by A₂₆₀/A₂₈₀ for purity using a NanoDrop instrument. The RNAs produced were ssRNA (GGGUUGAACUGGUUA), dsRNA (GGGUAACCAGUUCAA annealed to ssRNA), and hpRNA (GGGCGCCUGUCACGGCG).

NMR spectroscopy and structure determination

All NMR experiments for assignment and structure determination were performed at 298 K on a 700 MHz Bruker Avance III spectrometer. The sequence-specific backbone assignment was determined using 2D ¹H-¹⁵N HSQC, 3D HNCA, 3D HN(CO)CA, 3D HNCO, 3D HN(CA)CO, 3D CBCA(CO)NH, 3D HNCACB, and 3D HBHA(CBCACO)NH experiments. Aliphatic side-chain assignments were determined using 3D (H)CCH-TOCSY, 3D HC(C)H-COSY, 3D HC(C)H-TOCSY, 3D H(CCCO)NH, and 3D (H)CC(CO)NH experiments. Amide side-chain assignments were determined using a 3D ¹⁵N-resolved NOESY. ¹H chemical shifts were externally referenced to 0 ppm methyl resonance of 2,2-dimethyl-2-silapentane-5-sulfonate, whereas ¹³C and ¹⁵N chemical shifts were indirectly referenced according to the IUPAC recommendations (34). All NMR spectra were processed using Topspin 3.5 (Bruker). Processed spectra were analyzed using CARA. Backbone and side chain assignments were deposited in the BMRB under accession 30559.

For structure calculation, a 2D ¹H-¹H NOESY, 3D ¹⁵N-resolved NOESY, and 3D ¹³C-resolved NOESY were recorded and processed using Topspin 3.5 (Bruker). UNIO was used for iterative automated NOE peak picking and NOE assignment by ATNOS/CANDID (35, 36) and structure calculation with CYANA version 2.1 (37, 38). The 20 lowest energy structures were used for water refinement in CNS version 1.3 with the RECOORD scripts (39). The quality of the final ensemble was verified using NMR-Procheck, and the RMSD of these structures was determined using MolMol (40, 41). The final structure was deposited in the PDB under code 6NNB. All structure figures were generated using the PyMOL Molecular Graphics System, version 1.8 (Schrödinger, LLC). APBS was used to generate the electrostatic surface representation of PSHCP (42).

To determine the intensity ratio of peaks from the {¹H}-¹⁵N heteronuclear NOE experiment, peak intensity values were determined using CARA (http://cara.nmr.ch/). Experimental uncertainties for the {¹H}-¹⁵N heteronuclear NOE experiment were determined by measuring the root mean square noise of background regions in the unsaturated and saturated spectra using NMRPipe (43), as described previously (44).

NMR titration experiments

dsGCDNA (5′-CCG CGC GCG CGG-3′) was synthesized by IDT. DNA was resuspended in 20 mm sodium phosphate, pH 6.5, 50 mm NaCl, heated at 100 °C for 5 min, and cooled at room temperature to allow for the DNA to anneal. tRNA from E. coli (Sigma R1753) was resuspended in 20 mm sodium phosphate, pH 6.5, 50 mm NaCl. The concentration of the E. coli tRNA was determined by running a range of concentrations of the tRNA on an agarose gel and comparing them to a standard concentration. ¹⁵N-Labeled PSHCP was mixed with nucleic acid samples to a final protein concentration of 50 μm. The DNA and tRNA concentrations were varied based on the overall titration. ¹H-¹⁵N HSQC spectra were recorded on a Bruker Avance 600 MHz spectrometer using a 1.7-mm probe. For the TROSY titrations, ²H,¹⁵N-labeled PSHCP_1–62 was mixed with tRNA, and ¹H-¹⁵N TROSY spectra were collected on a 700 MHz Bruker Avance III spectrometer. For these titrations, PSHCP_1–62 was at a final concentration of 30 μm, and tRNA was at a final concentration of 0.5, 25, 50, or 100 μm. CSP values were determined using δΔ = √((δ¹H)² + 0.14(δ¹⁵N)²). CSP data were fit using MATLAB (MathWorks) to f(x) = CSP_max·(K_d + x + P_tot) − √((K_d + x + P_tot)[caret]2 − 4·(x·P_tot)))/(2·(P_tot)), where P_tot is the total concentration of protein (50 μm).

Isothermal titration calorimetry

ITC was performed using a VP-ITC (Microcal). All proteins and tRNA were prepared in 20 mm sodium phosphate, pH 6.5, 50 mm NaCl, or 20 mm sodium phosphate, pH 6.5, 25 mm NaCl, 75 mm KCl. Approximately 200 μm E. coli tRNA (Sigma R1753) was titrated into 20 μm PSHCP (either PSHCP_1–62 or PSHCP_1–56) at 24 °C. tRNA concentration was determined by running a tRNA titration on a 1% agarose gel. tRNA samples were then compared with samples of known concentration. ssRNA, dsRNA, and hpRNA were produced in-house (see under “RNA synthesis”) in the same buffer. These RNAs, at 200 μm, were titrated into PSHCP_1–62 at 20 μm except for the ssRNA, which was at 400 μm.

For ITC competition assays, 20 μm dsGCDNA was mixed with 20 μm PSHCP_1–62 for 30 min. Subsequently, 200 μm tRNA was titrated into the mixture of PSHCP_1–62 and dsGCDNA. Data were analyzed using Origin (Microcal) to produce both ITC thermograms and binding isotherms.

Survey of PSHCP taxonomic distribution and amino acid conservation

The amino acid sequence of the PSHCP gene from P. marinus str. MIT 9312 (RefSeq ID WP_002807701.1) was used as a query in a BLASTP search of the nr database via the NCBI web site (accessed on September 12, 2018; BLAST version 2.8.0+ (45); E-value cutoff 10⁻⁴). Taxonomic distribution was examined via Taxonomy Report provided with the BLAST search results. Matches with sequences not identical to the query were retrieved and aligned in ClustalX version 2.1 (46). Additionally, the same query sequence was used in a BLASTP search of the 703 genomes of Prochlorococcus and marine Synechococcus available via IMG/ProPortal (47). This database includes both isolates and single cell genomes, the latter ones with an average completion of 60% (47). The BLASTP 2.6.0+ search was carried out via IMG/MER website at https://img.jgi.doe.gov/on September 11, 2018, with E-value cutoff of 10⁻⁵. The obtained 463 PSHCP homologs were retrieved and aligned in ClustalX version 2.1 (46). Two poor quality sequences (defined as those containing at least one X) and one partial sequence at the beginning of a short contig in an incomplete genome were removed from the alignments. Both alignments are available as supporting Datasets S1 and S2.

Author contributions

K. M. B. and R. D. data curation; K. M. B. and R. D. formal analysis; K. M. B., R. D., M. P., O. Z., and M. J. R. investigation; K. M. B., M. P., O. Z., and M. J. R. writing-review and editing; O. Z. and M. J. R. conceptualization; O. Z. and M. J. R. writing-original draft; M. J. R. supervision.