Recognition of an α-helical hairpin in P22 large terminase by a synthetic antibody fragment

The crystal structure of Fab4 bound to the epitope of the bacteriophage P22 large terminase subunit, together with a 1.15 Å resolution crystal structure of the unliganded Fab4, which is the highest resolution ever achieved for a Fab, elucidate the principles governing the recognition of this novel helical epitope.

Abstract

The genome-packaging motor of tailed bacteriophages and herpesviruses is a multisubunit protein complex formed by several copies of a large (TerL) and a small (TerS) terminase subunit. The motor assembles transiently at the portal protein vertex of an empty precursor capsid to power the energy-dependent packaging of viral DNA. Both the ATPase and nuclease activities associated with genome packaging reside in TerL. Structural studies of TerL from bacteriophage P22 have been hindered by the conformational flexibility of this enzyme and its susceptibility to proteolysis. Here, an unbiased, synthetic phage-display Fab library was screened and a panel of high-affinity Fabs against P22 TerL were identified. This led to the discovery of a recombinant antibody fragment, Fab4, that binds a 33-amino-acid α-helical hairpin at the N-terminus of TerL with an equilibrium dissociation constant K _d of 71.5 nM. A 1.51 Å resolution crystal structure of Fab4 bound to the TerL epitope (TLE) together with a 1.15 Å resolution crystal structure of the unliganded Fab4, which is the highest resolution ever achieved for a Fab, elucidate the principles governing the recognition of this novel helical epitope. TLE adopts two different conformations in the asymmetric unit and buries as much as 1250 Ų of solvent-accessible surface in Fab4. TLE recognition is primarily mediated by conformational changes in the third complementarity-determining region of the Fab4 heavy chain (CDR H3) that take place upon epitope binding. It is demonstrated that TLE can be introduced genetically at the N-terminus of a target protein, where it retains high-affinity binding to Fab4.

1. Introduction  

Enterobacteria phage P22 is a member of the Podoviradae family of short-tailed bacterial viruses that infects Salmonella typhimurium (Teschke & Parent, 2010 ▸). P22 packages its ∼43 kb genome into an empty precursor capsid (or procapsid) using a ‘headful packing’ mechanism, a packaging strategy whereby the amount of DNA packaged inside the virion is determined by the interior volume of the mature particle (Casjens & Weigele, 2005 ▸; Catalano, 2005 ▸). Like many bacterial viruses and herpesviruses, P22 encodes two terminase subunits, known as the large (TerL) and small (TerS) terminases, whereas herpesviruses also have a third subunit (Heming et al., 2014 ▸). The packaging reaction requires the assembly of a genome-packaging motor (Bhardwaj et al., 2014 ▸; Sun et al., 2010 ▸; Casjens, 2011 ▸) that transiently forms onto the unique vertex of the icosahedral procapsid occupied by the portal protein (Chen et al., 2011 ▸). This ring-shaped dodecameric protein provides a conduit for DNA entry into the capsid (Dedeo et al., 2019 ▸) while changing conformation during genome packaging (Olia et al., 2011 ▸; Lokareddy et al., 2017 ▸). Both TerL and TerS are essential for genome packaging, although the exact molecular mechanisms by which these two subunits assemble into and function as a molecular complex are poorly understood. In P22, TerL and TerS form a complex that can be purified from infected cells (Poteete & Botstein, 1979 ▸) or assembled in vitro from recombinant factors (McNulty et al., 2015 ▸).

We have previously characterized TerL (499 amino acids, 57.6 kDa; Roy & Cingolani, 2012 ▸) and TerS (162 amino acids, 18.6 kDa; Roy et al., 2011 ▸, 2012 ▸) from phage P22. TerL, the packaging ATPase, binds directly to the procapsid conformation of the P22 portal protein (Lokareddy et al., 2017 ▸). It contains an N-terminal ATPase domain (Sun et al., 2008 ▸; Zhao et al., 2013 ▸) with ATP-binding Walker A and B motifs, and a C-terminal RNAse H-fold nuclease (Smits et al., 2009 ▸; Roy & Cingolani, 2012 ▸). The nuclease domain of TerL cleaves the concatemeric P22 genome at different stages of the packaging reaction (Wu et al., 2002 ▸). At the beginning of packaging, TerL makes sequence-specific cuts in the pac region known as ‘series-initiation cleavage’ to generate a DNA end that is inserted into the procapsid unidirectionally. At the end of packaging, TerL cleaves the DNA off the nascent virion, which is then sealed to prevent DNA leakage by three tail accessory factors: gp4 (Olia et al., 2006 ▸), gp10 (Olia et al., 2007 ▸) and gp26 (Bhardwaj et al., 2007 ▸, 2016 ▸). TerS also plays various roles in the genome-packaging reaction. As a specific DNA-recognition subunit, it binds to packaging-initiation sites (pac; Jackson et al., 1978 ▸; Wu et al., 2002 ▸) in the P22 genome, promoting their recognition by TerL; it also stimulates the ATPase activity associated with genome packaging (Roy et al., 2012 ▸). These two activities are likely to be coupled, as the stimulation of the ATPase activity is enhanced by viral DNA (Roy et al., 2012 ▸). A C-terminal basic moiety in P22 TerS (residues 140–162) named the LBD (large terminase binding domain) mediates association with TerL and viral DNA (Roy et al., 2012 ▸; McNulty et al., 2015 ▸). Although the LBD binds TerL in a 1:1 stoichiometry in vitro, the TerL–TerS complex purified from bacteria contains a substoichiometric number of TerL subunits compared with the nonameric TerS (McNulty et al., 2015 ▸).

Synthetic antibodies developed using phage-displayed antibody technology are a powerful tool in biology that complement and expand the repertoire of natural antibodies. Synthetic Fabs can specifically target functional states of a protein, often trapping discrete protein conformations (Fellouse et al., 2007 ▸; Paduch et al., 2013 ▸), thereby enabling structural studies. This paper describes the identification and biophysical characterization of a novel synthetic Fab that binds an α-helical hairpin at the N-terminus of phage P22 TerL.

2. Materials and methods  

2.1. Phage display  

The synthetic antibody library built using the 4D5 Fab scaffold, selection criteria and hit characterization have been described previously (Fellouse et al., 2007 ▸; Paduch et al., 2013 ▸). A biotinylated AviTag-TerL from phage P22 (see below) was subjected to phage display. After three rounds of library sorting, phages specific to TerL were identified by competitive phage ELISA, which provided an estimate of the affinity and the conformational specificity of each clone. The Fabs, with variable regions in CDR-H3, listed in Table 1 ▸ from ten phagemids were PCR-amplified and cloned into SalI/PaeI-linearized pSFV4 vector using a Gibson Assembly Cloning Kit (NEB).

Table 1. List of anti-TerL Fabs generated in this study.

+, association detected by size-exclusion chromatography (SEC); −, no association; s.c., the Fab stuck to the column; n.t., not tested.

Fab	Variable region in CDR-H3	ELISA	SEC	Native GEL
1	YSKGWVYVIHSWWYVYAF	1.338	++	++
2	VKEYWNYVYMFYYYSWWGF	1.655	—	—
3	WEYYYSDRYSYWEPHSGM	1.004	—	—
4	YSWPWVSYKPYYGLHFSAM	0.962	+++	+++
5	SGGWDVSWLYSSWFHSGI	1.036	—	—
6	SYWQYWLFSYTYPGL	0.988	+++	+++
7	GSEPGPFQMWGYVWYMAF	1.400	—	—
8	SPWLYNWYSSAL	1.394	—	—
9	GGYESYIMYYWYWSYKAAI	1.276	s.c.	n.t.
10	SESYSSWWVSWWYYGWAL	1.165	+++	+++

2.2. Biochemical techniques  

The expression and purification of P22 TerL (plasmid pET30b-TerL) have been described previously (Roy & Cingolani, 2012 ▸; Lokareddy et al., 2017 ▸; McNulty et al., 2015 ▸). An AviTag (MGLNDIFEAQKIEWHEGSS) was introduced at the N-terminus of 6His-TerL using site-directed mutagenesis (plasmid pET30b-AviTag-6His-TerL). AviTag-6His-TerL was expressed and purified like 6His-TerL. Purified AviTag-6His-TerL was biotinylated using the BirA-500 kit (Avidity). The efficiency of biotinylation was verified by binding to Streptavidin Sepharose resin (GE Healthcare). A peptide spanning P22 TerL residues 1–34 (referred to as TerL_pep) was synthesized by Peptide2go and purified to 90% homogeneity for crystallization. Recombinant Fabs were expressed as described previously (Bartesaghi et al., 2013 ▸). Briefly, Fab variants were subcloned into the expression vector (plasmid pSFV4-Fab4). The protein was expressed using the Escherichia coli BL21 cell line, with the cells being grown in 2×YT medium, and expression was induced at an OD₆₀₀ of ∼0.6. Induction proceeded for 5 h, at which point the cell pellets were harvested. Fabs were purified as described previously using ProteinG-A1 resin for single-step purification (Bailey et al., 2014 ▸). To form Fab–TerL complexes, a twofold molar excess of Fab was added to TerL and the mixture was purified on a Superdex 200 16/60 gel-filtration column (GE Healthcare) in ITC buffer (20 mM Tris–HCl pH 8.0, 150 mM NaCl, 1 mM MgCl₂, 5% glycerol). The gel-filtration column was calibrated with molecular-weight markers as described previously (Lokareddy et al., 2013 ▸). Fab4 and the Fab4–TerL complex were concentrated to ∼15 mg ml⁻¹ using a 30 000 molecular-weight cutoff ultrafiltration spin column (Vivaspin 20, Sartorius). TerS-LBD was expressed and purified as a fusion protein with MBP, as described previously (Roy et al., 2011 ▸). The P22 TerL minimal epitope (residues 1–23) was introduced by PCR using asymmetric megaprimers between the N-terminal 6His-MBP affinity tag and the Mycobacterium tuberculosis necrotizing toxin (TNT) gene in the plasmid pML1995 described previously (Sun et al., 2015). This modified plasmid (pML3977), encoding both 6His-MBP-TLE-TNT (69 kDa) and the antitoxin IFT (19.8 kDa), was expressed in E. coli BL21 (DE3) LOBSTR cells, which were grown in LB medium to an OD₆₀₀ of ∼0.6 and induced with 0.5 mM isopropyl β-d-1-thiogalactopyranoside for 3 h. The cells were pelleted by centrifugation (16 000g, 30 min, 4°C), resuspended in 20 mM Tris–HCl pH 8.0, 200 mM NaCl, 1 mM phenylmethyl­sulfonyl fluoride (PMSF), sonicated and centrifuged to recover soluble proteins (30 000g, 30 min, 4°C). The 6His-MBP-TLE-TNT–IFT complex was purified from the supernatant using amylose resin. The 6His-MBP tag was cleaved using TEV protease and cleared by binding to nickel–agarose resin. IFT was removed from the TLE-TNT–IFT complex by boiling at 70°C for 10 min. The TLE-TNT (25.3 kDa) was further polished via size-exclusion chromatography (SEC) using a Superdex 200 10/300 GL column (GE Healthcare) equilibrated with 20 mM Tris pH 8.0, 150 mM NaCl. TLE-TNT was incubated with a twofold molar excess of Fab4 and the complex was purified on a Superdex 200 10/300 GL column. Native gel electrophoresis was performed on a 1.5% agarose gel as described previously (Nardozzi et al., 2010 ▸; Mitrousis et al., 2008 ▸). In this assay, 20 µg P22 TerL was mixed with 10–20 µg Fab fragments or MBP-tagged LBD and the mixture was separated on a 1.5% agarose gel at room temperature for 1 h. After electrophoresis, the gel was fixed in gel-fixing solution [25%(v/v) 2-propanol, 10%(v/v) acetic acid] for 20 min and then equilibrated with 95%(v/v) ethanol for 2 h. The gels were then dried, stained for 10 min in 0.4%(w/v) Coomassie Brilliant Blue R250 in gel-fixing solution and destained in gel-fixing solution until the background was clear.

2.3. Biophysical analysis in solution  

Analytical ultracentrifugation sedimentation-velocity analysis of Fab4 was carried out in a Beckman XL-A Analytical Ultracentrifuge at the Sidney Kimmel Cancer Center X-ray Crystallography and Molecular Interaction Facility. Gel-filtration-purified Fab4–TerL complex dissolved at 0.5 mg ml⁻¹ in 20 mM Tris pH 8.0, 150 mM NaCl, 2.5% glycerol, 1 mM MgCl₂, 1 mM β-mercaptoethanol, 0.1 mM PMSF was spun at 40 000 rev min⁻¹ at 6°C. Absorbance values at 280 nm were fitted to a continuous sedimentation-coefficient [c(s)] distribution model in SEDFIT (Schuck, 2000 ▸). ITC experiments were carried out at 25°C using a nano-ITC calorimeter (TA Instruments). For ITC analysis, both Fab4 and P22 TerL were dialyzed overnight against ITC buffer (20 mM Tris–HCl pH 8.0, 150 mM NaCl, 5% glycerol, 1 mM MgCl₂, 1 mM β-mercaptoethanol, 0.1 mM PMSF) at 4°C. Fab4 (160 µM) was injected in 2.0 µl increments into a calorimetric cell containing 195 µl TerL (22 µM). The spacing between injections was 180 s. Titrations were performed in triplicate, and data were analyzed using the NanoAnalyze data-analysis software (TA Instruments). Heats of dilution were determined from control experiments carried out by injecting Fab4 against ITC buffer and were subtracted from enthalpies obtained by titrating Fab4 against TerL. Curve fitting was performed in the NanoAnalyze data-analysis software using a single set of binding sites model. The concentration of Fab4 and TerL used for ITC was accurately determined using the Lowry protein assay and spectrophotometric determination using the theoretical extinction coefficient.

2.4. Crystallographic methods  

All crystallization droplets were set up using the hanging-drop vapor-diffusion method by mixing 2 µl purified protein at ∼15 mg ml⁻¹ with an equal volume of crystallization solution. Crystals of Fab4–TerL_(1–23) were obtained from a gel-filtration-purified complex of Fab4 bound to full-length TerL using 100 mM KCl, 25 mM MgCl₂, 50 mM sodium cacodylate trihydrate pH 6.0, 15%(v/v) 2-propanol as the precipitant. Crystals of Fab4–TerL_pep were obtained in the presence of 0.2 M succinic acid pH 7.0, 20%(w/v) PEG 3350. Unliganded Fab4 was crystallized in the presence of 0.1 M Tris pH 8.5, 25%(w/v) polyethylene glycol 3350. Crystals were harvested in nylon cryo-loops, cryoprotected with 27% ethylene glycol and flash-cooled in liquid nitrogen. Complete diffraction data were collected on beamline 9-2 at Stanford Synchrotron Radiation Lightsource (SSRL) and beamline 23-ID-D at the Advanced Photon Source (APS) using a Dectris PILATUS 6M detector (Table 2 ▸). The structure of Fab4–TerL was solved by molecular replacement (MR) using the heavy and light chains of a recombinant Fab against the HIV-1 integrase catalytic core (PDB entry 5eu7; Galilee et al., 2016 ▸) as a search model using Phaser (McCoy et al., 2007 ▸). The initial MR solution was refined using phenix.refine (Adams et al., 2002 ▸), and the variable loop and TerL residues (1–23) were built manually using Coot (Emsley et al., 2010 ▸). The model was then subjected to additional cycles of positional, real-space and TLS B-factor refinement using phenix.refine (Adams et al., 2002 ▸). Final re-refinement using PDB-REDO (Joosten et al., 2014 ▸) usually yielded the best R _cryst/R _free and stereochemistry. The final model including Fab4 and TerL_(1–23) was refined to an R _cryst and R _free of 18.8% and 23.9%, respectively, at 2.40 Å resolution. The structures of Fab4–TerL_pep and unliganded Fab4 were solved by MR using the model of Fab4 and were refined as described above to an R _cryst of 16.7% and an R _free of 20.3% at 1.51 Å resolution and to an R _cryst of 15.7% and an R _free of 17.0% at 1.15 Å resolution, respectively (Table 3 ▸). There are five cis-prolines in Fab4: three in the heavy chain (Pro321, Pro383 and Pro385) and two in the light chain (Pro8 and Pro143). These cis-prolines have outstanding density in the 1.51 and 1.15 Å resolution structures but are difficult to refine in the 2.4 Å resolution structure. All models have excellent geometry, with >99.5% of residues in the most favored regions of the Ramachandran plot, and root-mean-square deviations (r.m.s.d.s) on bond lengths and angles of 0.010 Å and 1.34° for Fab4–TerL_(1–23), 0.012 Å and 1.09° for Fab4–TerL_pep and 0.009 Å and 1.27° for unliganded Fab4, respectively.

Table 2. Data collection and processing.

Values in parentheses are for the outer shell.

	Fab4–TerL(1–23)	Fab4–TerLpep	Fab4
Diffraction source	Beamline 9-2, SSRL	Beamline 9-2, SSRL	Beamline 9-2, SSRL
Wavelength (Å)	0.980	0.978	0.978
Temperature (K)	100	100	100
Detector	PILATUS 6M PAD	PILATUS 6M PAD	PILATUS 6M PAD
Crystal-to-detector distance (mm)	350	250	200
Rotation range per image (°)	0.5	0.5	0.5
Total rotation range (°)	250	225	250
Exposure time per image (s)	3	2	3
Space group	P21	P21	P21
a, b, c (Å)	77.7, 139.3, 163.7	74.1, 86.3, 86.1	64.7, 65.8, 107.4
α, β, γ (°)	90.0, 98.2, 90.0	90.0, 97.7, 90.0	90.0, 99.8, 90.0
Mosaicity (°)	0.25	0.20	0.30
Resolution range (Å)	50–2.49 (2.49–2.40)	51–1.51 (1.56–1.51)	15–1.15 (1.19–1.15)
Total No. of reflections	3100030	3417037	5789033
No. of unique reflections	127623	161937	293620
Completeness (%)	94.0 (94.2)	96.0 (94.3)	93.8 (59.8)
Multiplicity	2.5 (2.3)	2.9 (2.7)	4.7 (3.0)
〈I/σ(I)〉	11.3 (1.6)	32.7 (1.9)	50.4 (1.9)
R merge (%)	11.8 (63.9)	9.2 (71.3)	5.4 (77.9)
R r.i.m. (%)	8.5 (55.5)	4.2 (57.3)	2.7 (55.9)
CC1/2 (outer shell)	0.508	0.456	0.511
Overall B factor from Wilson plot (Å2)	32.5	20.4	15.9

Table 3. Structure solution and refinement.

Values in parentheses are for the outer shell.

	Fab4–TerL(1–23)	Fab4–TerLpep	Fab4
PDB code	6vi1	6xmi	6vi2
Resolution range (Å)	15–2.40	15–1.51	15–1.15
Completeness (%)	94.0 (94.2)	94.5 (95.5)	93.8 (59.8)
No. of reflections, working set	124119	159136	291451
No. of reflections, test set	1985	1994	1996
Final R cryst (%)	18.8	16.7	15.7
Final R free (%)	23.9	20.3	17.0
No. of non-H atoms
Protein	21330	7262	6785
Water	836	807	1189
Total	22166	8069	7974
R.m.s. deviations
Bonds (Å)	0.010	0.012	0.009
Angles (°)	1.34	1.09	1.27
Average B factors (Å2)
Protein (Fab4)	52.2	36.4	25.7
Ligand (TerL)	48.5	43.1	n.a.
Water	42.9	43.8	33.9
Ramachandran plot
Most favored (%)	96.2	97.9	97.0
Allowed (%)	3.6	2.1	1.8
MolProbity score	1.7	1.1	1.3
Clashscore	8.6	3.0	4.4

2.5. Structure analysis and modeling  

All ribbon diagrams and surface representations were prepared using PyMOL (version 2.0; Schrödinger; https://pymol.org/2/). Nonlinear Poisson–Boltzmann electrostatic calculations were performed using APBS Tools (Dolinsky et al., 2004 ▸). Secondary-structure superimpositions were carried out in Coot (Emsley et al., 2010 ▸). The free energy of assembly dissociation (ΔG _diss) was calculated by PISA (Krissinel & Henrick, 2007 ▸), and intramolecular contacts were measured using PDBsum (Laskowski et al., 1993 ▸). A 3D model of P22 TerL was calculated using Phyre2 (Kelley et al., 2015 ▸), and docking used ZDOCK (Pierce et al., 2011 ▸).

3. Results and discussion  

3.1. Isolation of recombinant Fabs that bind P22 TerL  

P22 TerL is a two-domain enzyme of 499 amino acids that contains an N-terminal ATPase domain (residues 1–285) connected to a C-terminal nuclease domain (residues 294–482) by a protease-sensitive loop (GIPTMGSG). The purified enzyme is poorly stable in solution and prone to degradation (Roy & Cingolani, 2012 ▸). Our previous attempt to crystallize the full-length protein yielded crystals of the C-terminal nuclease domain alone, which we solved to 2.02 Å resolution. Similarly, the in vitro assembled TerS–TerL complex is heterogeneous (Roy et al., 2012 ▸) and is unsuitable for high-resolution structural analysis. In an attempt to identify a crystallization chaperone for TerL, we screened an unbiased, chemically diverse synthetic phage Fab library against TerL. Phagemid hits were characterized using an in vitro binding assay, which led to the identification of ten putative binders for P22 TerL (Table 1 ▸). We cloned, expressed and purified all ten recombinant Fabs and screened them for binding to the purified TerL using size-exclusion chromatography (SEC). Four Fabs, Fab1, Fab4, Fab6 and Fab10, markedly shifted TerL migration on a Superose 6 column (Table 1 ▸), although we were not able to make large quantities of Fab10 for biochemical studies. We further validated the association of Fabs with TerL by native gel electrophoresis. Although Fab1, Fab4 and Fab6 all shifted TerL mobility on an agarose gel (Fig. 1 ▸), Fab4 gave the most quantitative mobility shift. Thus, we focused on Fab4, which is well expressed in bacteria, easy to purify in milligram quantities and highly soluble, as expected for a crystallization chaperone.

Figure 1.

Identification of recombinant Fabs specific for P22 TerL. Native gel electrophoresis on agarose showing the binding of purified Fab1, Fab4 and Fab6 to 200 pmol P22 TerL. A 1× molar ratio is equal to 200 pmol of Fab and a 2× molar ratio is equal to 400 pmol.

3.2. Biophysical characterization of Fab4 binding to TerL  

To investigate the binding stoichiometry of Fab4 and TerL, we added a threefold molar excess of Fab4 to TerL and subjected the mixture to SEC. The two proteins eluted as a major peak of ∼100 kDa (peak 1), preceded by a smaller peak of ∼300 kDa that eluted more rapidly (peak 2; Fig. 2 ▸ a). Next, we subjected three fractions of the Fab4–TerL complex (fraction Nos. 68 and 72 from peak 1 and fraction No. 34 from peak 2) to analytical ultracentrifugation (AUC) sedimentation-velocity analysis. Fig. 2 ▸(b) shows a typical sedimentation profile of Fab4–TerL fractions obtained in 150 mM sodium chloride, 2.5% glycerol at 10°C (Table 4 ▸). In a concentration range between 1 and 10 µM, samples from peak 1 (fractions Nos. 68 and 72) migrated as homogeneous species with an apparent sedimentation coefficient (s*) of ∼3.4 S (absolute sedimentation coefficient S _20,w = ∼3.8 S), corresponding to a mass of 109.9 kDa, which is unambiguously consistent with one copy of TerL bound to Fab4 (expected molecular mass of ∼108.1 kDa). The frictional ratio estimated based on the sedimentation data was f/f ₀ = 2.2, which is suggestive of an elongated molecular assembly. In contrast, fraction No. 34 corresponding to peak 2 (Figs. 2 ▸ a and 2 ▸ b), which also had stoichiometric bands for Fab4 and TerL on the gel, appeared to be polydisperse, possibly indicative of a soluble aggregate. We used nano isothermal titration calorimetry (nano-ITC) to quantify the binding affinity of Fab4 for TerL. We measured the heat released upon the titration of increasing amounts of 160 µM Fab4 into a cell containing purified 22 µM TerL at 25°C (Fig. 3 ▸ a). We observed an exothermic reaction with ΔG = −9.87 kcal mol⁻¹, which saturated within 17–18 injections. Binding data were fitted using a one independent binding site model, yielding an equilibrium dissociation constant K _d of ∼71.5 nM and an n value of ∼0.992, which is also consistent with a 1:1 interaction. Interestingly, the association of Fab4 with TerL (Fig. 3 ▸ b) has a negative enthalpy (ΔH = −10.97 kcal mol⁻¹), indicating the formation of favorable ionic and hydrogen bonds as well as van der Waals interactions. The negative entropy (ΔS = −3.50 kcal mol⁻¹) suggests that Fab4 binding to TerL leads to a reduction in the overall conformational entropy of the complex, which is possibly explained by a reduction in the mobility and/or the flexibility of either protein upon complex formation. Thus, ITC analysis at 25°C revealed that Fab4 binds P22 TerL in a 1:1 stoichiometry with a K _d of ∼71.5 nM. Attempts to repeat the same titration at a higher temperature (for example 30–37°C) were unsuccessful owing to the tendency of TerL to aggregate and come out of solution. This problem was alleviated, but not eliminated, at 25°C, suggesting that the K _d determined at this temperature may slightly underestimate the equilibrium binding affinity of Fab4 for TerL.

Figure 2.

Stoichiometry of Fab4 binding to P22 TerL. (a) Highlighted in light blue is a typical elution profile of the Fab4–TerL complex analyzed on a Superdex 200 (16/60) gel-filtration column (the green trace in the background is for TerL alone). Fractions from the main eluted peak were analyzed by SDS–PAGE (bottom gel). Excess Fab4 is not visible as an individual peak as the antibody alone is ‘sticky’ and binds to the Superdex matrix nonspecifically, eluting as a smaller-than-expected species at around 108 ml (Hornsby et al., 2015 ▸). (b) Sedimentation-velocity profiles of three fractions of Fab4–TerL eluted from SEC (fraction No. 34 from peak 2 and fraction Nos. 68 and 72 from peak 1). All samples were measured in 20 mM Tris–HCl pH 8.0, 150 mM NaCl, 3 mM DTT, 5% glycerol, 1 mM MgCl₂ at 6°C. Top panel: raw absorbance at 280 nm plotted as a function of the radial position. Data at intervals of 20 min are shown as dots for sedimentation at 40 000 rev min⁻¹. Middle panel: the residuals between the fitted curve and the raw data. Bottom panel: the fitted distribution of the apparent sedimentation coefficient (s*) calculated for Fab4–TerL in fraction Nos. 68 and 72 is about 3.4 S (∼80% sample), corresponding to an estimated molecular mass of ∼109 kDa (Table 2 ▸).

Table 4. List of analytical ultracentrifugation parameters for the Fab4–TerL complex.

	Fraction No. 34	Fraction No. 68	Fraction No. 72
Protein concentration (mg ml−1)	0.5	0.5	0.5
Apparent sedimentation coefficient s (S)	n.a.	3.399	3.409
Absolute sedimentation coefficient s 20,w (S)	n.a.	3.858	3.870
Frictional ratio f/f 0	n.a.	2.20	2.17
Abundance (%)	n.a.	81.18	77.42
Molecular mass (kDa)	n.a.	109.86	108.65
Oligomeric state (Fab4:TerL)	Polydisperse	1:1	1:1

Figure 3.

Calorimetric analysis of the interaction of Fab4 with P22 TerL. (a) Titration of 160 µM Fab4 (in the syringe) into a cell containing 22 µM TerL. Top panel: raw heats of injection. Bottom panel: integrated, buffer-subtracted binding enthalpy plotted as a function of the Fab4:TerL molar ratio. (b) Histogram showing the overall variation of enthalpy (ΔH), entropy (TΔS) and Gibbs energy (ΔG) upon the titration of TerL against Fab4.

3.3. Structure of Fab4 bound to the TerL N-terminal α-helical hairpin  

To shed light on the structure of P22 TerL, we crystallized the TerL–Fab4 complex that eluted from SEC in peak 1 (Fig. 2 ▸ a) and obtained large plate-like crystals after a month that diffracted X-rays to 2.4 Å resolution (Table 2 ▸). Crystallographic analysis revealed a large unit cell containing six copies of Fab4 arranged in a monoclinic asymmetric unit that has about 50% solvent content but not enough room for TerL. Interestingly, all six Fabs displayed strong and continuous electron density in the antigen-binding site (Fig. 4 ▸ a). Despite the modest resolution (∼2.4 Å), the electron density in the antigen-binding site was sufficiently clear to allow the unambiguous tracing of 23 amino acids organized as an α-helical hairpin (Fig. 4 ▸ a). A BLAST search revealed that the sequence of this epitope matches perfectly with residues 1–23 of P22 TerL (Fig. 4 ▸ b). Thus, we fortuitously crystallized Fab4 bound to a cleavage product of TerL that consists of an N-terminal α-helical hairpin. This epitope contains two α-helices: α₁, spanning residues 3–9, and α₂, which encompasses residues 12–23. As for the structure of the P22 TerL nuclease domain (Roy & Cingolani, 2012 ▸), our attempts to crystallize the full-length TerL yielded a proteolytic fragment of TerL, further confirming the extreme conformational flexibility of this protein, which has eluded crystallization efforts for a decade. The final structure of Fab4–TerL_(1–23) was refined to an R _cryst and R _free of 18.8% and 23.9%, respectively, at 2.4 Å resolution (Table 3 ▸).

Figure 4.

Structure of Fab4 bound to TerL. (a) A 2.4 Å resolution F _o − F _c electron-density difference map visible in the antigen-binding site of Fab4 co-crystallized with TerL (Table 3 ▸). The difference map (in gray) is displayed at 2.25σ above the background and is overlaid on residues 1–23 of the refined TerL_(1–23) model. The CDR-H3 is also shown in yellow. (b) Amino-acid sequence of P22 TerL residues 1–40. The region of TerL built into the electron density displayed in (a) is underlined. The amino-acid sequence of TerL_pep is shown in red. ‘H’ stands for residues in helical conformation in TerL_pep (black residues are always helical, while gray residues in helix α₂ are helical only in complex A; see Fig. 5 ▸ b).

Inspecting the Fab4–TerL_(1–23) interface, it became clear that the third complementarity-determining region of the Fab4 heavy chain (CDR H3) extends past TerL residue 23 in helix α₂ (Fig. 4 ▸ a), suggesting that Fab4 recognizes a longer epitope in TerL than just the first 23 residues that are visible in the crystal structure. In support of this idea, a secondary-structure prediction of the TerL N-terminus suggested that helix α₂ (residues 11–23) seen in the crystal structure could extend for an additional ten amino acids C-terminal to residue 23 (Fig. 4 ▸ b). To test this hypothesis, we synthesized a peptide spanning residues 1–34 of TerL (TerL_pep; Fig. 4 ▸ b) that we co-crystallized with Fab4. Large crystals were obtained in only two days that yielded complete diffraction data to 1.51 Å resolution. We solved this structure by MR using the previously determined model of Fab4 and found two Fab4 molecules in the asymmetric unit arranged in an antiparallel fashion (Table 3 ▸). The electron density for the TerL epitopes was exceptionally well resolved (Fig. 5 ▸ a), which allowed us to build an unambiguous atomic model. In one of the two asymmetric unit assemblies (complex A), TerL helix α₂ extends up to residue 33, making intimate contacts with the Fab4 CDR-H3 (Fig. 5 ▸ b). In the other assembly (complex B), helix α₂ ends at residue 23, while residues 24–30 adopt a random-coil conformation that is stabilized by the constant domain (C_H) of Fab4 from complex A (Fig. 5 ▸ c). In complex B, helix α₂ melts at the beginning of a stretch of four glutamates (23-Glu-Glu-Glu-Glu-26) which are fully helical in complex A. The average B factor of TerL residues 1–23 is 40.6 and 35.9 Ų in complexes A and B, respectively, and increases to 53.3 and 60.1 Ų for the remainder of the C-terminal residues. The final crystallo­graphic model of Fab4–TerL_pep was refined to an R _cryst and R _free of 16.7% and 20.3%, respectively, at 1.51 Å resolution (Table 3 ▸).

Figure 5.

1.51 Å resolution crystal structure of Fab4 bound to TerL_pep. The structure revealed two Fab4–TerL_pep complexes (referred to as A and B) assembled in an antiparallel fashion in a monoclinic asymmetric unit. (a) Refined 2F _o − F _c electron-density map for TerL_pep residues 1–8 displayed at 1.6σ above the background. The refined model is overlaid on the density. (b) Ribbon diagram of complex A, with Fab4 colored cyan (light chain) and yellow (heavy chain), and TerL_pep, which is visible between residues 1 and 33, in red. (c) Ribbon diagram of complex B, with Fab4 colored cyan (light chain) and yellow (heavy chain), and TerL_pep, which is visible between residues 1 and 30, in green. The region 24–30 of TerL in complex B adopts a random-coil conformation. C_H/C_L and V_H/V_L are the constant and variable domains of the heavy and light chains, respectively.

3.4. Intimate recognition of TerL by Fab4  

TerL_pep helices α₁ and α₂ expose an acidic bonding surface for Fab4 (the calculated isoelectric point of TerL_pep is 3.7). In the extended conformation of helix α₂ seen in complex A (Fig. 5 ▸ b) the hairpin projects 21 residues towards the epitope-binding site of Fab4, including nine Glu residues and three Asp residues, which make 11 hydrogen bonds, one salt bridge and 101 van der Waals contacts within a cutoff distance of 4 Å (Fig. 6 ▸ a). Residues 1–23 in TerL make the majority of bonds with Fab4. Only two residues C-terminal to Glu23, namely Glu26 and Arg29, make contacts with Fab4, which may explain why this region of helix α₂ adopts a random-coil conformation in complex B (Fig. 5 ▸ c). The TerL α-helical hairpin is stabilized intramolecularly by seven hydrophobic residues (Met1, Ile6, Leu7, Leu10, Leu18, Leu19 and Leu22) that form a hydrophobic core. PISA (Krissinel & Henrick, 2007 ▸) estimates a solvation free-energy gain upon the formation of the Fab4–TerL_pep interface of ΔG = −7.30 kcal mol⁻¹, which is slightly lower than the value experimentally observed by ITC (ΔG = −9.87 kcal mol⁻¹). We attribute this difference to entropic effects such as conformational changes or solvent hydration that depend on the geometry and dynamics of the Fab4–TerL_pep interface (Krissinel, 2011 ▸) and are not captured by in silico analysis of interface properties. The Fab4–TerL_pep binding interface buries a total solvent-accessible surface of 1250 Ų upon assembly formation, mostly involving the heavy chain. For comparison, Fab–protein complexes bury on average 777 ± 135 Ų of surface area (Ye et al., 2008 ▸). Thus, the large surface complementarity between Fab4 and TerL_pep observed in the structure is more similar to those found in protein–protein binding interfaces than in Fab–peptide complexes.

Figure 6.

Schematic of the recognition of the TerL N-terminal hairpin by Fab4. (a) Schematic summary of the interactions made by Fab4 with TerL_pep. 11 hydrogen bonds (shown as blue dashed lines) and one salt bridge (red dashed line) are displayed. 19 nonbonded interactions are shown as circle schematics. Residues from the Fab4 heavy and light chains are colored yellow and cyan, respectively. TerL residues making intramolecular contacts between helices α₁ and α₂ are colored green. (b) Superimposition of Fab4 bound to TerL_pep and unliganded Fab4, showing conformational changes in the CDR-H3 loop, which is colored yellow in the bound state and light orange in the unliganded conformation.

To decipher how Fab4 remodels in response to TerL binding, we also solved a high-resolution structure of the unliganded Fab4, which we refined to an R _cryst and R _free of 15.7% and 17.0% at 1.15 Å resolution (Table 3 ▸). To our knowledge, this is the highest resolution ever achieved for a Fab, and sheds light on the atomic details of this synthetic Fab. The 1.15 Å resolution structure contains two Fabs arranged in the asymmetric unit that had excellent electron density for the CDR loops. Superposition of the Fab4 structure in the antigen-bound and unliganded states results in an r.m.s.d. of only 0.503 Å (Fig. 6 ▸ b). Deviations are located mainly in CDR-H3, between heavy-chain residues H_323–333. In the unliganded state, the CDR-H3 loop swings away from the light chain, whereas in the TerL-bound conformation of complex A (Fig. 5 ▸ b) it collapses onto helix α₂. Overall, there is a 5 Å displacement in the main-chain position of CDR-H3 in the two states and, notably, Tyr328 and Tyr329 swing 180° towards the epitope in the bound state, with a total displacement of ∼13 Å compared with the unliganded Fab. Since the CDR-H3 loop is well defined in both electron-density maps, the conformational changes described here are directly induced by the binding of TerL_pep to Fab4 (Fig. 6 ▸ b). In support of this idea, Tyr328 and Tyr329 have no discernible side-chain electron density in complex B of Fab4–TerL_pep (Fig. 5 ▸ c), where helix α2 is too short to make direct contacts with these residues in the CDR-H3 loop. The closing of the CDR-H3 loop towards the antigen is consistent with the negative variation in entropy upon complex formation calculated from the ITC data (Fig. 2 ▸ b), pointing to a reduction in the conformational entropy of Fab4 upon TerL recognition.

3.5. Protein engineering with the TerL epitope (TLE)  

We took a protein-engineering approach in order to determine whether the TerL epitope could be introduced at the N-terminus of a target protein and retain high-affinity binding to Fab4. We fused the minimal P22 TLE spanning residues 1–23 to the M. tuberculosis necrotizing toxin (TNT), an exotoxin that we have previously determined crystallo­graphically in complex with the antitoxin IFT (Sun et al., 2015 ▸). Firstly, we cloned TLE at the N-terminus of TNT, which was expressed as a 6His-MBP-tagged fusion (69 kDa) in the presence of the antitoxin IFT (19.8 kDa) to avoid the cytotoxic effect of TNT, which is a potent NAD⁺/NADP⁺ glycohydrolase (Tak et al., 2019 ▸; Fig. 7 ▸ a, lane 1). We then cleaved off 6His-MBP using Tobacco etch virus (TEV) protease (Fig. 7 ▸ a, lane 2), incubated the mixture with nickel–agarose resin to capture the 6His-MBP tag (Fig. 7 ▸ a, lane 3) and, finally, boiled off IFT to isolate TLE-TNT (25.3 kDa; Fig. 7 ▸ a, lane 4). The toxin was then incubated with a twofold molar excess of Fab4 and analyzed by SEC to assess whether Fab4 retains activity towards TLE fused to an exogenous protein. Remarkably, Fab4 associated stoichiometrically with TLE-TNT, shifting the migration of this protein by 3 ml (Fig. 7 ▸ b). The TLE-TNT–Fab4 complex (fractions a–g in Fig. 7 ▸ b) was visualized by SDS–PAGE under nonreducing and reducing (Fig. 7 ▸ c) conditions, confirming the presence of Fab4 in the peak fractions. Thus, a minimal TLE encompassing just 23 residues retains high-affinity binding to Fab4 when fused to the N-terminus of TNT.

Figure 7.

Fab4 binds TLE-tagged TNT. (a) Purification of TLE-TNT. A complex of 6His-MBP-TLE-TNT (69 kDa) and IFT (20 kDa) purified on amylose beads (lane 1) was subjected to TEV protease, cleaving 6His-MBP (44 kDa) from TLE-TNT (25 kDa) (lane 2). The mixture after the recapture of 6His-MBP over nickel–agarose resin (lane 3) was boiled off at 70°C to remove IFT and obtain TLE-TNT (lane 4). (b) SEC profile on a Superdex 200 10/300 column of TLE-TNT alone and in complex with Fab4, showing a 3 ml shift in the complex elution. (c) SDS–PAGE showing a 1:1 stoichiometric complex of TLE-TNT and Fab4 purified on a Superdex 200 10/30 column. The TLE-TNT–Fab4 complex fractions (labeled a–g) were visualized under both nonreducing and reducing conditions.

3.6. Modeling the full-length structure of P22 TerL  

The folds of the TerL ATPase and nuclease domains are conserved among tailed bacteriophages and herpesviruses, despite low sequence similarity. The relative orientation of these two domains varies dramatically in different crystal structures owing to the flexibility of the linker that connects them (Zhao et al., 2013 ▸), which is protease-sensitive in P22 TerL (Roy & Cingolani, 2012 ▸). To generate an accurate model of the full-length TerL from P22, we took advantage of two lines of evidence. Firstly, the atomic structures of both the C-terminal nuclease domain (residues 289–482; PDB entry 4dkw; Roy & Cingolani, 2012 ▸) and the N-terminal α-helical hairpin (residues 1–33) of P22 TerL have been determined. Residues 32–288 in the ATPase domain represent the only structurally uncharacterized part of TerL, which is less than half of the 499 residues in TerL. Secondly, the ATPase domain is conserved in other viral TerLs of known structure. A database search reveals that P22 TerL is 13% identical to the TerL subunits (Figs. 8 ▸ a–8 ▸ d) from the thermophilic bacteriophages D6E (a member of the Myoviridae; Xu et al., 2017 ▸) and P74-26 (a member of the Siphoviridae; Hilbert et al., 2015 ▸), 9% identical to the TerL from Sf6 (a member of the Podoviridae; Zhao et al., 2013 ▸) and 8% identical to gp17 from T4 (a member of the Myoviridae; Sun et al., 2008 ▸). Lower sequence identity is also detectable to the N-terminal domain of the DNA-packaging ATPase from bacteriophage phi29 (Mao et al., 2016 ▸). As a starting point to model the unknown residues of P22 TerL, we focused on the central β-sheet of the ATPase domain, which consists of eight β-strands sandwiched by α-helices (Figs. 8 ▸ a–8 ▸ d). In all TerLs, the β-sheet starts with an α₁–β₂–α₂ motif (shown in yellow in Figs. 8 ▸ a–8 ▸ d), whereby helix α₁ connects to the second strand (β₂) of the β-sheet that continues into helix α₂ and from there to the fifth strand (β₅) of the β-sheet. The nucleotide is held between helix α₁ and helix α₂, which harbors a classical phosphate-binding loop (P-loop). Interestingly, the first helix of the α₁–β₂–α₂ motif is amino-terminal in the TerLs of D6E (Fig. 8 ▸ a), P74-26 (Fig. 8 ▸ b) and Sf6 (Fig. 8 ▸ c), while it contains a 138-amino-acid extension in T4 gp17 (shown in red in Fig. 8 ▸ d) that has previously been hypothesized to function like the transmission of a car (Sun et al., 2008 ▸). With this in mind, we generated an atomic model of P22 TerL residues 40–499 using Phyre2 (Kelley et al., 2015 ▸), which is shown in Fig. 9 ▸(a). Next, we docked TerL_pep from complex A against the predicted ATPase core either by letting the docking software probe the entire predicted structure of TerL (40–499) or by restricting the docking area to a region within ∼30 Å of residue 40 of TerL. This distance mimics the length of a seven-amino-acid linker between residues 34 and 40, assuming ∼4 Å per amino acid (Ainavarapu et al., 2007 ▸). In either case, automated docking positioned the acidic cradle of TerL_pep recognized by Fab4 against a patch of basic residues at the interface between the TerL ATPase and nuclease domains, which includes the C-terminal β-hairpin involved in portal protein binding (residues 480–497; McNulty et al., 2015 ▸; Lokareddy et al., 2017 ▸; Fig. 9 ▸ a). This model predicts an intramolecular association between TerL_pep and the C-terminal β-hairpin that ‘locks’ the protein into a closed, possibly less active conformation, as previously suggested for the large terminase from bacterio­phage D6E (Xu et al., 2017 ▸). This model explains the slow turnover of P22 TerL measured in vitro and the need for TerS to stimulate the weak ATPase activity associated with genome packaging. Because both TerS and Fab4 bind the N-terminus of TerL between residues 1 and 58 (McNulty et al., 2015 ▸), we asked whether the two proteins make simultaneous or mutually exclusive interactions with TerL. A native gel electrophoresis assay confirmed that TerS-LBD binds TerL (McNulty et al., 2015 ▸; Fig. 9 ▸ b, lane 4), and this complex is super-shifted by the addition of an equimolar quantity of Fab4 (Fig. 9 ▸ b, lane 5), which is indicative of a trimeric complex. Thus, Fab4 and TerS harbor distinct binding sites on TerL, as explained by at least two models of association. TerS could bind C-terminal to Fab4, between residues 34 and 58, although this region (colored yellow in Figs. 8 ▸ and 9 ▸ a) is partially buried inside the ATPase core. Alternatively, TerS-LBD could bind the helical surface of TerL helices α₁ and α₂ (Fig. 5 ▸ c), which is solvent-exposed when Fab4 is bound. Future structural studies will clarify the interaction between TerL and TerS and the interplay with Fab4.

Figure 8.

The conserved topology of the TerL ATPase domain. Ribbon diagrams of the ATPase domain of TerL from (a) D6E (PDB entry 5oe8), (b) P74-26 (PDB entry 4zni), (c) Sf6 (PDB entry 4idh) and (d) T4 (PDB entry 3cpe). In all panels, the eight-stranded β-sheet is colored gray, with the α₁–β₂–α₂ motif in yellow. The long N-terminal insertion domain found in T4 is colored red. An N-terminal 16-amino-acid insertion that contains a short α-helix (7-­SDKFFELL-14) is also present at the N-terminus of P74-26 TerL but is not shown in (b).

Figure 9.

A complete 3D model of the full-length TerL of phage P22. The model of TerL spanning residues 40–499 was generated using Phyre2 (Kelley et al., 2015 ▸). The ATPase and nuclease domains are colored gray and cyan, respectively. The α₁–β₂–α₂ motif in the ATPase domain is colored yellow as in Fig. 6 ▸. The N-terminal α-helical hairpin TerL_pep is colored red and was docked onto the structure using ZDOCK (Pierce et al., 2011 ▸). (b) Native gel electrophoresis on agarose showing the binding of purified Fab4 (lane 1), TerL (lane 2) and TerS-LBD (lane 3). LBD binds TerL stoichiometrically (lane 4) and the addition of Fab4 yields a super-shift (lane 5), while Fab4 and LBD do not bind to each other.

4. Conclusions  

In this paper, we present the identification of a synthetic Fab that recognizes a 33-amino-acid α-helical hairpin at the N-terminus of P22 TerL. High-resolution crystal structures of the unliganded Fab4 and of Fab4 bound to TerL_pep revealed the detailed molecular recognition of this helical epitope. Furthermore, we show that a minimal epitope of TerL encompassing residues 1–23 can be genetically introduced at the N-terminus of a target protein, TNT, and retain high-affinity binding to Fab4. Although Fab4 did not help in obtaining crystals of the full-length TerL from P22, which is unstable and short-lived (Roy & Cingolani, 2012 ▸), future studies will need to determine whether Fab4 can be used as a tool for protein engineering and structural studies.

Supplementary Material

PDB reference: Fab4, unliganded, 6vi2

PDB reference: bound to P22 TerL_(1–23), 6vi1

PDB reference: bound to P22 TerL_pep, 6xmi

Funding Statement

This work was funded by National Institutes of Health grant R01 GM100888 to Gino Cingolani. National Cancer Institute grants P30 CA56036, S10 OD017987, and OD023479.